University of Southern California Dissertations and Theses

Sensory information processing by retinothalamic neural circuits

(USC Thesis Other)

Sensory information processing by retinothalamic neural circuits

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

SENSORY INFORMATION PROCESSING
BY RETINOTHALAMIC NEURAL CIRCUITS

by
Xin Wang

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(NEUROSCIENCE)

May 2010

Copyright 2010 Xin Wang
ii

To my family
iii

ACKNOWLEDGEMENTS
I owe my deepest gratitude to my advisors, Judith A. Hirsch and Friedrich T. Sommer, for
years of support and guidance on my research. I am as much indebted to the other members of
my thesis committee, Bartlett W. Mel, Alapakkam P. Sampath and Bosco Tjan, from whom I have
been receiving constant inspirations and encouragement. I also thank my collaborators, Luis M.
Martinez Otero, Kilian Koepsell and W. Martin Usrey for discussion regarding my studies.
The body of research presented in this dissertation would not be completed without
assistance from the following of my colleagues: Cristina Soto Sanchez, Qingbo Wang, Vishal
Vaingankar and Yichun Wei with in vivo experiment; Jackie M. Provost, Brittany Gary, Mary
Bathen, Shelly X. Xing and Morgan Gerstmar with anatomical reconstruction and data analysis.
iv

TABLE OF CONTENTS
Dedication ii
Acknowledgements iii
List of Figures v
Abstract vii
Chapter 1 Introduction 1
Chapter 2 Processing of visual information by thalamic inhibitory neurons 11
Chapter 2 Introduction 11
Chapter 2 Results 14
Chapter 2 Discussion 32
Chapter 2 Experimental procedures 40
Chapter 2 Supplemental materials 46
Chapter 3 Thalamic control of dual modes of firing during naturalistic viewing 49
Chapter 3 Introduction 49
Chapter 3 Results 51
Chapter 3 Discussion 69
Chapter 3 Experimental procedures 74
Chapter 3 Supplemental materials 82
Chapter 4 Recoding of sensory information across the retinothalamic synapse 90
Chapter 4 Introduction 90
Chapter 4 Results 94
Chapter 4 Discussion 112
Chapter 4 Experimental procedures 119
Chapter 4 Supplemental materials 125
Chapter 5 Conclusion 134
References 137

v

LIST OF FIGURES
Figure 1. Push-pull responses of an OFF-center relay cell and ON-center interneuron. 15
Figure 2. Receptive fields of relay cells and interneurons and prediction of neural responses
by linear-nonlinear models.
17
Figure 3. Quantitative comparison of postsynaptic currents recorded from all cells. 21
Figure 4. Voltage dependence of postsynaptic potentials recorded from relay cells and
interneurons.
23
Figure 5. Visual modulation of synaptic inputs to relay cells and interneurons. 25
Figure 6. Rates of unitary synaptic events recorded from relay cells and interneurons. 27
Figure 7. Spatial distribution of relay cells and interneurons. 29
Figure 8. Synaptic transmission of information by different forms of inhibitory inputs. 31
Figure 9. Morphology of interneurons. 47
Figure 10. Diagram of a hypothetical circuit for the interneuron’s receptive field. 48
Figure 11. Intracellular responses of a thalamic interneuron. 48
Figure 12. Spatially opponent excitation (push) and inhibition (pull) in the relay cell’s
receptive field.
52
Figure 13. Intracellular responses of a relay cell to repeated presentations of natural movies. 55
Figure 14. Receptive fields of synaptic excitation (push), inhibition (pull) and spikes
reconstructed with natural movies.
60
Figure 15. Pull from the center of the receptive field precedes bursts but not tonic spikes
evoked by natural movies.
64
Figure 16. Model prediction of the occurrence of burst and tonic spikes based on feedforward
synaptic inhibition.
65
vi

Figure 17. Retinogeniculate synaptic inputs trigger putative T-currents that evoke bursts. 67
Figure 18. Extraction of neural events and components from voltage-damped recordings. 73
Figure 19. Explanation and controls for receptive field reconstruction. 78
Figure 20. Control for the construction of the inhibitory receptive field. 84
Figure 21. Burst, tonic and long-ISI tonic spike-triggered-average of membrane current and
voltage.
85
Figure 22. Tradeoff between strength and duration of the suppression that primes T-
currents.
86
Figure 23. Postsynaptic versus presynaptic mechanisms for burst generation. 88
Figure 24. Identification of the joint relevant subspace across the retinothalamic synapse. 93
Figure 25. Spatiotemporal analysis of the joint retinothalamic feature space. 96
Figure 26. Change of single-spike code across the retinothalamic synapse. 99
Figure 27. Change of pair-wise correlation code across the retinothalamic synapse. 102
Figure 28. Transformation of retinal correlation code into thalamic single-spike code may
explain the changes in coding efficiency across the retinothalamic synapse.
104
Figure 29. “Paired-spike enhancement” as a model of the retinothalamic transmission. 107
Figure 30. Performance of the retinothalamic transmission models. 109
Figure 31. Identification of retinal spike trains from single ganglion cells. 125
Figure 32. Sequential identification of the joint relevant subspace using the information-
theoretic spike-triggered average and covariance (iSTAC) analysis.
127
Figure 33. “Missing” retinal spikes – a control study. 130
Figure 34. Pair-wise synergy in the joint retinothalamic feature space. 132
vii

ABSTRACT
In the mammalian visual system, sensory information captured by the retina is routed
through the lateral geniculate nucleus (LGN) of the thalamus before reaching the cerebral cortex.
The lateral geniculate circuits thus operate as a gateway for visual information flowing from the
sensory periphery to the central nervous system. Traditionally, the LGN has been regarded as a
passive relay station rather than an active processing center because the receptive fields of
geniculate neurons closely resemble those of retinal ganglion cells. This simplistic view, however,
is at odds with the complexity of the thalamic neural circuits – the LGN houses a substantial
population of inhibitory neurons that can modulate the activity of the thalamocortical projecting
neurons. How do the excitatory and inhibitory neurons in the LGN form circuits to process
visual information from the retina to the cortex? In this dissertation, three studies that address
this question will be presented. Using in vivo patch recordings, these studies directly probe the
functional components of the thalamic neural circuit in the whole animal; automated visual
stimulation, quantitative analysis and computer simulation are then used to reveal the principles
underlying the spatio-temporal operation of retinothalamic transmission. Specifically, it will be
demonstrated that inhibitory neurons of the LGN process retinal inputs in a distinct manner and
actively transform the temporal patterns of neural activities from the retina to the cortex.
Moreover, it will be shown that the biophysics of the retinothalamic synapse accounts for a
recoding of the visual information represented in the thalamus. Taken together, this body of
viii

work creates a better understanding of neural mechanisms of the visual thalamus and their
functional consequences for sensory information processing.
1

CHAPTER 1
INTRODUCTION
The lateral geniculate nucleus was historically viewed as a passive relay station, channeling
retinal inputs to visual cortex without sophisticated processing. The main experimental
observation that fostered this view was the qualitative similarity between the spatial organization
of retinal and geniculate receptive fields (Hubel, 1960). This idea was later supported by the
finding that many relay cells inherit center-surround receptive fields from one or very few retinal
inputs (Cleland et al., 1971; Cleland and Lee, 1985; Mastronarde, 1987a, b; Usrey et al., 1998,
1999). However, the LGN contains about a third as many inhibitory interneurons as relay cells
(Fitzpatrick et al., 1984; Guillery, 1966; LeVay and Ferster, 1979; Lin et al., 1977). These form
local inhibitory circuits that are capable of modifying patterns of activity from the retina to the
cortex (Ahlsen et al., 1985; Blitz and Regehr, 2005; Bloomfield and Sherman, 1988; Hamos et al.,
1985; Kim et al., 1997; McIlwain and Creutzfeldt, 1967; Sillito and Kemp, 1983; Singer et al., 1972).
Thus, the thalamic neural circuits seem to process sensory information actively.

Inhibitory circuits of the lateral geniculate nucleus
Two anatomically distinct sources of intrathalamic inhibition impinge on thalamocortical
projecting relay cells. One source is feedforward inhibition mediated by inhibitory interneurons
within the principle layers of the geniculate; these cells receive substantial synaptic inputs from
retinal ganglion cells (Erisir et al., 1998; Guillery, 1969a; Hamos et al., 1985; Montero, 1991; Van
2

Horn et al., 2000). The other source of inhibition stems from cells in the perigeniculate nucleus,
the visual sector of the thalamic reticular nucleus, a GABAergic structure immediately dorsal to
the geniculate (Cucchiaro et al., 1991; Kim et al., 1997; Uhlrich and Cucchiaro, 1992; Uhlrich et al.,
1991). These cells provide feedback inhibition; they are driven by relay cell axon collaterals and
corticothalamic axons but do not receive retinal input (Laties and Sprague, 1966; Sanderson and
Pearson, 1977; Uhlrich and Cucchiaro, 1992). Different dendritic compartments of relay cells are
targeted by the two types of inhibition: feedforward inhibition synapses on proximal dendrites,
feedback distal (Guillery, 1969a; Hamos et al., 1985; Montero, 1987; Van Horn et al., 2000; Wilson
et al., 1984). Both sources of inhibition can have strong impact on their targets; shocks to the
optic tract evoke large disynaptic IPSPs (Ahlsen et al., 1985; Bloomfield and Sherman, 1988) and
activation of perigeniculate cells also evokes strong IPSPs in relay cells (Kim et al., 1997).

Synaptic mechanisms of intrageniculate interneurons
In terms of synaptic integration of the inhibitory circuitry, intrageniculate interneurons are
unique in many ways. First, they receive massive retinal inputs (Guillery, 1969a; Montero, 1991)
which are mediated by both ionotropic and metabotropic glutamate receptors (Govindaiah and
Cox, 2006). Second, intrageniculate interneurons inhibit their targets by means of two distinct
synaptic mechanisms; they make either axodendritic synapses with local projecting axons (Hamos
et al., 1985; Montero, 1987), or dendrodendritic synapses on relay cell dendrites. The
dendrodendritic synapses are building blocks of the structures known as “triads” (Montero, 1987;
Wilson et al., 1984), where the dendritic compartments of the interneuron and the relay cell that
3

form the inhibitory synapse are both postsynaptic to a retinal afferent. Third, about a quarter of
total synapses on an interneuron are inhibitory (Hamos et al., 1985; Montero, 1991), suggesting
substantial connections among local interneurons (feedback inhibition from the perigeniculate
does not target interneurons (Cucchiaro et al., 1991)). Last, in contrast to relay neurons,
intrageniculate interneurons have electrotonically extended dendrites (Bloomfield et al., 1987;
Bloomfield and Sherman, 1989). How is the uniqueness of synaptic integration of the geniculate
interneuron manifested in its activity, and what is the function of this unique activity during
vision? These are the questions CHAPTER 2 and CHAPTER 3 will directly address.

Receptive fields of intrageniculate inhibitory neurons
The behavior of a sensory neuron is usually characterized by its “receptive field”, a region of
the receptor-topic map from which the neuron receives input. The structure of the receptive field
determines the “feature” that the neuron extracts from the stimulus. The two types of inhibitory
neurons of the visual thalamus have distinct receptive fields. Like those of relay cells, the
receptive fields of intrageniculate interneurons have been reported to have center-surround
spatial organizations (Humphrey and Weller, 1988b), likely inherited from the retinal drive. The
receptive fields of perigeniculate neurons, on the other hand, have been described as large and
amorphous, and these cells are usually binocular and have spatially overlapping responses to both
contrast polarities (Sanderson et al., 1969; Uhlrich et al., 1991). These two types of inhibitory
receptive fields probably serve different functional purposes, and can be distinguished
experimentally with little ambiguity.
4

Synaptic structure of the receptive fields of relay cells
Since the responses of thalamic relay cells are shaped by both the retinal afferents and local
interneurons, their receptive fields thus reflect contributions from retinal as well as local
inhibitory receptive fields. The excitatory component of the thalamic receptive field is largely
built by inputs from retinal ganglion cells, as shown by quantitative receptive field mapping
studies (Cai et al., 1997; Shapley and Hochstein, 1975; Usrey et al., 1999). Further, studies using
cross-correlation analysis suggest that the retinogeniculate connection is highly specific, with one
relay cell receiving inputs from very few ganglion cells (Cleland et al., 1971; Hamos et al., 1987;
Levick et al., 1972; Mastronarde, 1987a; Usrey et al., 1999); often one dominant retinal input can
account for most of the postsynaptic thalamic spikes (Levine and Cleland, 2001; Sincich et al.,
2007; Usrey et al., 1999).
The synaptic physiology of the thalamic receptive field is poorly researched, even though
there is a good amount of anatomical knowledge about the thalamic inhibitory circuits. A few
studies demonstrate that intrageniculate interneurons have center-surround receptive fields
(Dubin and Cleland, 1977; Humphrey and Weller, 1988a, b; Sherman and Friedlander, 1988).
Additional work suggests that the distribution of excitation and inhibition within the receptive
fields of relay cells might have a spatially opponent organization (Mastronarde, 1987a; McIlwain
and Creutzfeldt, 1967; Sillito and Kemp, 1983). A simple explanation of this observation is a local
pairing of relay cells and interneurons with highly overlapped opponent receptive fields, but this
is not compatible with the fact that there are many fewer interneurons than relay cells. Moreover,
5

spatially opponent receptive fields do not necessarily come with strict temporal opponency. On
the contrary, with the emergence of the “lagged” profile (Humphrey and Weller, 1988a, b;
Mastronarde, 1987a, b), temporal diversity of receptive fields is much greater in the geniculate
than retina, which could be of functional significance in building direction selectivity (Wolfe and
Palmer, 1998). In CHAPTER 3, a comparative study will be presented that quantitatively
describe the spatiotemporal structures of the excitation versus inhibition of the thalamic receptive
fields.

Dual firing modes of thalamic relay cells
The activity pattern of thalamic relay neurons depends strongly on the level of membrane
polarization. When the membrane rests at depolarized levels, relay cells produce tonic trains of
action potentials; when the membrane is hyperpolarized they fire rapid bursts. The bursts are
initiated by low-threshold (T-type) calcium channels that open transiently at voltages below spike
threshold and remain inactivated until exposed to substantial hyperpolarization (Jahnsen and
Llinas, 1984a). Early recordings made from unanaesthetized animals suggested that the tonic
mode of firing was associated with wakefulness and the burst mode with drowsiness or sleep
(Fourment et al., 1984; Livingstone and Hubel, 1981; Steriade et al., 1993). Recent work, however,
suggests that firing modes are not strictly linked to behavioral state; instead, they play a functional
role in sensory processing (Denning and Reinagel, 2005; Guido et al., 1992; Lesica and Stanley,
2004; Ramcharan et al., 2000; Reinagel and Reid, 2000; Swadlow and Gusev, 2001). Bursts,
though not common, occur routinely in awake animals and can be evoked by sensory stimuli
6

(Guido et al., 1992; Ramcharan et al., 2000; Swadlow and Gusev, 2001; Weyand et al., 2001). The
signaling of sensory information by bursts is very important for downstream processing. For
example, the temporal pattern of spike trains determines the amount and type of information that
can be conveyed about the stimulus (Denning and Reinagel, 2005; Liu et al., 2001; Reinagel et al.,
1999). Further, bursts activate the cortex more effectively than slower trains of spikes (Swadlow
and Gusev, 2001; Swadlow et al., 2002), likely because they evoke synaptic potentials that
summate in time (Usrey et al., 2000) and also because they occur after long silences that permit
recovery from synaptic depression (Swadlow and Gusev, 2001; Swadlow et al., 2002).

Exploration of synaptic mechanisms with intracellular recording in vivo
Intracellular studies of the thalamus that allow direct analysis of synaptic processing have
only been carried out intensively in in vitro preparations. These studies have shown that the dual
firing modes are controlled by membrane polarization because of the gating properties of the T-
type calcium channels (Jahnsen and Llinas, 1984b; McCormick and Feeser, 1990) and that the
firing modes mostly correspond to behavioral state of the animal (Livingstone and Hubel, 1981).
In these experiments, however, the global patterns of activity during normal vision could not be
applied to the preparation. On the other hand, extracellular studies in vivo have shown that
visually evoked bursts are most likely to occur after prolonged exposure to non-preferred stimuli
(Alitto et al., 2005; Denning and Reinagel, 2005; Lesica and Stanley, 2004), leading to the
hypothesis that prolonged suppressive stimuli could evoke hyperpolarizations strong enough to
revive the T-type calcium channels that trigger thalamic bursting. At this point, though
7

supported by most extracellular studies in vivo and intracellular studies in vitro, this hypothesis
cannot be directly tested unless intracellular recording is made from relay cells in vivo; this is the
very approach used by the studies described in CHAPTERS 2, 3 and 4.

Using natural stimuli
There is heated debate about how best to use stimuli with natural statistics to explore in
sensory physiology. Natural stimuli lack the parameterized, and thus well controlled, statistics
that make the quantitative characterization of neural responses theoretically and practically
tractable. Thus, some argue that these stimuli should not be used as a tool to characterize neural
responses but rather as a test to evaluate the characterization (Rust and Movshon, 2005).
However, natural stimuli simulate the ecological conditions under which sensory organs and
brains have evolved; hence, they might evoke important biological responses that artificial stimuli
would fail to elicit (Felsen and Dan, 2005).
Burst firing of geniculate relay cells is such a case. The dynamics of T-type calcium channels
make strong and prolonged hyperpolarization a requirement for de-inactivation (Huguenard and
McCormick, 1992; Jahnsen and Llinas, 1984c). Since natural statistics feature a concentration of
energy at the low-frequency end of the power spectrum, both in space and in time (Dong and
Atick, 1995), they might prime bursts routinely. The results of two extracellular studies suggest
that this is the case; stimuli with naturalistic statistics evoke burst firing more frequently than do
white noise stimuli (Denning and Reinagel, 2005; Lesica and Stanley, 2004). This establishes the
rationale for using natural stimuli in studying visually evoked thalamic bursts.
8

Retinogeniculate transmission of action potentials
The processing of sensory information, from the periphery to the brain, reflects the nature of
the computation performed by neural circuits. To understand how neural circuits transform
information represented by neural activities (i.e. spike trains) from one stage of sensory
processing to the next, it is necessary to record simultaneously from synaptically connected
neurons. Such recordings were first made with extracellular electrodes; (Hubel, 1960)
demonstrated that “S-potentials”, synaptic potentials evoked by retinal ganglion cells (Kaplan and
Shapley, 1984), can be recovered along with thalamic spikes. Later, simultaneous recording from
retina and LGN also allowed recording of spiking activities of synaptically connected ganglion
cells and relay cells (Cleland et al., 1971). More recent studies conducted with both techniques
and cross-correlation analysis have made it clear that retinothalamic connectivity is highly
specific. Pre- (retinal) and postsynaptic (thalamic) receptive fields show great spatial resemblance
(Usrey et al., 1999); most relay cells are dominated by inputs from a single ganglion cell (Levine
and Cleland, 2001; Sincich et al., 2007) and receive a few minor inputs with low contribution
(Hamos et al., 1987; Usrey et al., 1999); the dominant input accounts for the majority of the
postsynaptic spikes (Sincich et al., 2007; Usrey et al., 1999).
Given this simplicity of retinothalamic connectivity, it seems reasonable to approximate
retinothalamic transmission by simple statistical rules. The first step taken in this direction was
to study the efficacy of retinothalamic spike transmission as a function of presynaptic inter-spike-
intervals; this led to the discovery of the “paired-spike enhancement” effect (Levine et al., 1996;
9

Sincich et al., 2007; Usrey et al., 1998). However, it remained to be tested if this simple statistical
principle suffices to describe retinothalamic transmission of spike trains. In CHAPTER 4,
models based on such statistical rules of retinothalamic transmission will be fitted and evaluated.

Linear-nonlinear framework and spike-triggered stimulus statistics
Quantitative models of sensory neural processing are widely used, because they provide not
only a concise description of neurophysiological data but also a means to predict neural responses
to novel stimuli. Therefore, an ideal model of a sensory neuron is one that predicts its responses
to a wide variety of stimulus conditions. In the early visual system, models based on one or more
linear filters and subsequent nonlinearities point in a promising direction towards the ideal
(Carandini et al., 2005). This linear-nonlinear (LN) model framework has proved successful in
different stages of visual processing: retina (Chichilnisky, 2001), LGN (Lesica and Stanley, 2004;
Mante et al., 2005), V1 (David et al., 2004; Felsen et al., 2005; Rust et al., 2005; Sharpee et al., 2006;
Touryan et al., 2005) and V4 (David et al., 2006). Given well-behaving statistics of the stimulus,
the parameters of the LN model can be fitted easily with spike-triggered stimulus statistics, e.g.
mean, covariance, etc (Schwartz et al., 2006). Further, information theoretic approaches have
recently been developed for LN model fitting (Pillow et al., 2005; Sharpee et al., 2004).
All previous studies using this framework, however, focused only on single stages of visual
processing in isolation. With simultaneously recorded data from two sequential stages, it is
possible to resolve the underlying synaptic mechanisms by means of modeling in the same LN
framework – retinothalamic transmission is a useful model system for such attempts. In
10

CHAPTER 4, a model that describes neural responses in both the retina and the LGN will be
fitted and tested.
11

CHAPTER 2
PROCESSING OF VISUAL INFORMATION
BY THALAMIC INHIBITORY NEURONS
Synapses made by local interneurons dominate the thalamic circuits that process sensory
signals traveling from the eye to cortex. The anatomical and physiological differences between
these interneurons and the relay cells that project downstream are vast. To explore how these
variations might contribute to visual processing, we compared intracellular recordings obtained
from both types of cells in vivo, for the first time. Macroscopically, the receptive fields were
similar, built of a concentric center and surround in which dark and bright stimuli evoked
responses of the reverse sign. Microscopically, the visually-evoked synaptic potentials had
opposite profiles. Optimal stimuli drove trains of single EPSPs in relay cells but graded
depolarizations in interneurons. By contrast, suppressive stimuli elicited graded
hyperpolarizations in relay cells but unitary IPSPs in interneurons. Operating in concert, these
complementary patterns of response regulate the excitability of relay cells while preserving
information encoded in the timing of retinal spikes.

CHAPTER 2 INTRODUCTION
Inhibitory neurons dominate the intrinsic circuits of the lateral geniculate nucleus of the
thalamus. Relay cells, the thalamic neurons that project to cortex, rarely form local contacts
(Bickford et al., 2008); rather, most intranuclear connections come from local interneurons
12

(Montero, 1991; Van Horn et al., 2000). Even the earliest physiological recordings from relay
cells, conducted a half century ago, emphasized that visual responses in the lateral geniculate are
subject to far stronger suppression than in the retina. The inhibition is powerful; it can determine
whether relay cells fire tonically or in bursts (Denning and Reinagel, 2005; Lesica and Stanley,
2004; Wang et al., 2007), sharpen selectivity for stimulus features (Hubel and Wiesel, 1961;
Ruksenas et al., 2000) and otherwise gate the transmission of visually-evoked activity downstream
(Sillito and Kemp, 1983). Studies using fixed tissue (Famiglietti and Peters, 1972; Guillery, 1969b;
Hamos et al., 1985; Montero, 1986, 1987; Sherman, 2004) or brain slices (Acuna-Goycolea et al.,
2008; Cox and Sherman, 2000; Cox et al., 1998; Govindaiah and Cox, 2004; Govindaiah and Cox,
2006; Pape and McCormick, 1995) have provided enormous insight to the building blocks
(synapses, ion channels, receptors) of thalamic inhibitory circuits. Yet, there is scant knowledge
of how these local networks operate as a whole during sensory processing. By combining whole-
cell recording in vivo, automated visual stimulation and computational modeling, we have
compared the synaptic responses that visual stimuli evoke from identified relay cells and
interneurons.
Relay cells have receptive fields built from two concentric subregions, a center and surround
(Hubel and Wiesel, 1961) which have push-pull responses to stimuli of the opposite contrast — in
regions where bright stimuli excite, dark stimuli inhibit (Martinez et al., 2005; Wang et al., 2007).
This pattern of synaptic response can be explained by a simple feedforward circuit. Specifically,
retinal ganglion cells evoke excitation directly, via monosynaptic input, and elicit inhibition
indirectly, through local interneurons (Hamos et al., 1985; Lindstrom, 1982; Montero, 1991).
13

Consistent with this scheme, our results show that interneurons have receptive fields with a
center-surround organization and even have a push-pull arrangement of excitation and inhibition
within subregions. Moreover, a widely used computational model (Schwartz et al., 2006) that
succeeds in predicting how retinal and thalamic neurons fire in response to novel visual stimuli
(Carandini et al., 2005; Dan et al., 1996) explained the intracellular responses of relay cells and
interneurons equally well. Given the similarity of the receptive fields of both types of cells, one
might assume that all thalamic receptive fields are built the same way. There is steadily
accumulating evidence, however, that relay cells and interneurons have profoundly different
anatomical and physiological features. For example, ganglion cells form synapses on the proximal
dendrites of relay cells (Hamos et al., 1987; Sherman and Guillery, 1998) but innervate the distal
compartments of interneurons (Hamos et al., 1985; Sherman, 2004); this remote input is
propagated towards the soma by active currents (Acuna-Goycolea et al., 2008). Further, relay
cells communicate with other neurons by means of conventional axonal contacts whereas both
the dendrites and axons of local interneurons form synapses with target neurons (Famiglietti and
Peters, 1972; Guillery, 1969b; Hamos et al., 1985; Montero, 1986, 1987).
Our results show that the intrinsic differences between cell types are matched by
commensurately distinct responses to visual stimulation. For relay cells, preferred stimuli evoked
a series of large, unitary excitatory postsynaptic potentials (EPSPs) whereas non-preferred stimuli
elicited graded inhibition. The picture for interneurons was the inverse. In these cells, visually-
evoked excitation was smooth while inhibitory responses were built by jagged trains of unitary
inhibitory postsynaptic potentials. The rates and receptive field structures of both types of
14

synaptic events were similar to those of retinal ganglion cells, suggesting a feedforward origin.
The disparity between the two classes of synaptic responses was so stereotyped that a simple index
devised to measure their shapes yielded a bimodal distribution. Further, results from past
ultrastructural studies (Coomes et al., 2002; Famiglietti and Peters, 1972; Godwin et al., 1996a;
Guillery, 1969b; Hamos et al., 1985; Montero, 1986, 1987; Pasik et al., 1973; Sherman, 2004)
combined with recent discoveries in vitro (Acuna-Goycolea et al., 2008; Cox and Sherman, 2000;
Govindaiah and Cox, 2004; Govindaiah and Cox, 2006) can, in principle, account for shapes of
these different signals. In sum, even though relay cells and interneurons have the same type of
receptive field, they adopt very different strategies to process retinal information. Since the
structure of inhibitory circuits in the lateral geniculate nucleus is largely conserved across
modalities and species, it is likely that our results illustrate a basic feature of thalamic processing.

CHAPTER 2 RESULTS
To compare the visual response properties of interneurons with relay cells, we made whole-
cell recordings from 119 cells in 22 adult female cats, 1.5 - 4.5 kg. We recorded from the main
layers A, A1, C, of the lateral geniculate. Of the cells that we were able to fill with dye and thus
categorize using standard anatomical criteria (Friedlander et al., 1981; Humphrey and Weller,
1988b; Sherman and Friedlander, 1988), 27 were relay cells and 9 were interneurons. These and
remaining cells were also classified by physiological criteria we discuss below.

15

Figure 1. Push-pull responses of an OFF-center relay cell and ON-center interneuron. (A)
Anatomical reconstruction of an OFF-center relay cell drawn above the averaged responses of the
membrane voltage to dark and bright disks flashed in center (B) annuli flashed in the surround
(C) of the receptive field. The icons at left depict stimulus shape and contrast while the gray line
under the traces marks the stimulus duration. Diagram of a hypothetical circuit for the relay cell’s
receptive field (D). Ganglion cells, bottom, connect with thalamic cells, top. Neurons are drawn
as their receptive fields with OFF subregions blue and ON subregions red. For excitatory cells, the
center is solid whereas the surround is unshaded and delimited by a solid line. For inhibitory
cells, the center is hatched and the surround unshaded; both types of subregions are bordered by
dashed lines. Connections (boutons) are color coded to match the stimulus preference of the
presynaptic neuron and plus and minus symbols indicate the sign of the input. Responses for an
ON-center interneuron (E) evoked from the center (F) and surround (G) of the receptive field;
conventions as in (B, C). Records from relay cells are in black and from interneurons in blue.
The reconstruction is of dendrites only; labeled axon was too pale to trace continuously.

16

The push-pull distribution of excitation and inhibition in the interneuron’s receptive field
The first and most basic question we asked was whether or not the receptive fields of
interneurons resembled those of relay cells. Relay cells have receptive fields made of a center and
surround in which stimuli of the reverse contrast evoke responses of the opposite sign (Wang et
al., 2007), Figure 1A. Dark disks flashed in the center of an OFF X relay cell, Figure 1A, evoked
an excitatory response, Figure 1B, top, whereas bright ones were hyperpolarizing, Figure 1B, as
illustrated by traces of the membrane voltage that are averages of repeated trials of the stimulus;
icons depicting the stimulus are shown to the left. The equivalent situation held for responses to
annuli flashed in the surround, Figure 1C. The excitation, or push, is almost certainly fed
forward from retinal ganglion cells of the same center sign (Bullier and Norton, 1979; Cleland et
al., 1971; Levick et al., 1972; Mastronarde, 1987b; Usrey et al., 1999). A simple explanation for the
pull is that it comes from interneurons that also have receptive fields with a center-surround
structure but have the opposite preference for stimulus contrast, Figure 1D (see (Wang et al.,
2007) for evidence showing that the hyperpolarization results from synaptic inhibition rather
than the withdrawal of excitatory drive).
17

Figure 2. Receptive fields of relay cells and interneurons and prediction of neural responses by
linear-nonlinear models. Spatial receptive fields (peak frame) (A) and temporal response (peak
pixel) (B) of two relay cells and two interneurons computed from the intracellular response.
Stimulus size is indicated in the figure, color conventions are shown in the legend below. Scatter
plots of the actual intracellular response against that obtained using a linear filter made from the
spatiotemporal receptive field show how the nonlinear (red curve) component of the model was
fit (C). Comparison of the actual and predicted responses (D); the actual response was
normalized so that the mean was zero and the variance was unity. Performance of the model (E),
quantified by explained variance, for populations of relay cells and interneurons; color codes for
the different classes of cells are provided in the legend.

18

Consistent with this scheme, we found that interneurons had receptive fields with a center-
surround structure, as illustrated by the averaged response of an ON interneuron, Figure 1E,
Figure 9B. Moreover, there was a push-pull arrangement of responses in each subregion: bright
stimuli in the ON or dark stimuli in OFF surround were depolarizing whereas the reverse was
true for stimuli of the opposite contrast, Figures 1F, G. This result not only met our prediction
about the qualitative shape of the interneurons' receptive field structure, but went further to show
that these cells also have push-pull responses. The excitatory response of this interneuron rose
more slowly than that of the relay cell; however this does not indicate a trend (see Figure 2B).
Also note that the time-course of thalamic responses spans a broad continuum from brief to
prolonged (Wolfe and Palmer, 1998).

Quantitative comparison of the receptive fields of relay cells and interneurons
Discs and annuli drive the center and surround strongly; they are useful for exploring the
basic layout of the receptive field and for isolating spatially opponent excitation and inhibition.
The next step in our analysis was to compare the synaptic responses of relay cells and
interneurons quantitatively. For this analysis we used Gaussian white noise, a series of grayscale
checkerboards, each made of randomly arrayed pixels whose luminance is drawn from a Gaussian
distribution (Schwartz et al., 2006).
Past studies of relay cells have shown that the spatiotemporal receptive field predicts neural
behavior quite well (Carandini et al., 2005) when included in a simple linear-nonlinear model
19

(Carandini et al., 2005; Schwartz et al., 2006). We asked if the same were true for interneurons.
From the synaptic responses to the Gaussian noise, we used reverse correlation to generate
spatiotemporal receptive fields. We illustrate the spatial peaks of the receptive field with contour
plots, Figure 2A, and the time course for the peak pixel, Figure 2B, for 4 cells (an ON and OFF
relay cell and an ON and OFF interneuron). The plots for all types of cells were similar in shape
and time. Note that these 2D stimuli, which lack spatial coherence and were not of high contrast,
do not drive the surround well; thus the fields in Figure 2B show mainly the centers.
Next we used the spatial and temporal filters recovered from the response-triggered average
to build a standard linear-nonlinear cascade model. The model comprised two stages. The first
component was a linear filter that represents the receptive field (i.e. the response-triggered
average of the stimulus) for which we substituted averages of the membrane current, binned at
the stimulus update rate, instead of the conventional spikes. The second component was a
nonlinearity that maps the output of the linear filter to the strength of response (Carandini et al.,
2005; Schwartz et al., 2006). The fits of the nonlinearity are shown in Figure 2C; we used a
sigmoidal function (though a linear fit performed almost as well, not shown). Past studies of relay
cells have shown that simple linear-nonlinear models are able to predict responses to novel
stimuli quite well (Carandini et al., 2005). We streamed a novel noise sequence (i.e. one different
from that used to estimate the linear filter and nonlinearity) through the model and plotted the
result against the actual response to the novel stimulus, Figure 2D. The results of our analysis
show that the simplistic model predicted the responses of all cells to the same extent, as measured
by explained variance, Figure 2E. Thus, the synaptic receptive fields of both relay cells and
20

interneurons appear to be similar in their shapes and explanatory power. Note that the explained
variance is less than that achieved by past studies for spikes. This is likely due to the fact the
stimulus sequence we used to map the field was long (Wang et al., 2007), which precluded the
collection of multiple repeats that can easily be obtained with extracellular methods and also
because intracellular signals have far more complicated, effectively “noisier”, shapes than spikes
(Olshausen and Field, 2005). Also, our model was simple, as the intracellular technique precluded
lengthy characterization of other aspects of the neural response (e.g. contrast gain control) that
can furnish additional parameters to achieve better fits (Mante et al., 2008).

Correlating two classes of intracellular waveform with relay cells and interneurons
So far, we have described averaged responses to visual stimuli. Our next step was to analyze
the intracellular records at a finer grain in order to learn how receptive fields are built. Taken
together, our results show that relay cells and interneurons process feedforward drive in
stereotypically different ways, as follows. We investigated the detailed structure of neural
responses by recording the membrane current during the presentation of Gaussian noise or
natural movies. We chose to use voltage-clamp rather than current-clamp mode in order to
reduce the influence of intrinsic membrane conductances on the shapes of synaptic input (Wang
et al., 2007). The signals we recorded from most cells (roughly 75% of our sample) were
dominated by trains of prominent excitatory postsynaptic currents (EPSCs), Figure 3A and see
(Koepsell et al., 2009; Wang et al., 2007). We were able to stain 27 of the 89 neurons from which
these recordings were made; all were relay cells including X, Y and W subclasses (Koepsell et al.,
21

2009; Wang et al., 2007). This observation was not surprising; the retinal boutons that contact
relay cells are large and proximal (Hamos et al., 1987; Sherman and Guillery, 1998) and produce
commensurately prominent EPSPs (Blitz and Regehr, 2005) (cortical neurons also project to relay
cells, but their small boutons target distal dendrites (Sherman and Guillery, 1998) and generate
events that are typically invisible unless the membrane resistance is increased with drugs
(Granseth and Lindstrom, 2003).

Figure 3. Quantitative comparison of postsynaptic currents recorded from all cells. Examples of
membrane currents characteristic of relay cells in black (A) and interneurons in blue (B), each
trace is from a different cell. Normalized deflection indices plotted against time scale for the
whole dataset (C); darker thicker curves are from neurons illustrated in A, B. Histogram plotting
the distribution of the first principal component of the deflection indices (D).

22

The recordings obtained from the remaining cells were strikingly different. The most salient
components were brief, net hyperpolarizing currents that were often preceded by depolarizing
transients from which the occasional spike escaped, Figure 3B. We labeled 9 of the 30 cells with
this physiological profile; all were interneurons. Thus, there was a striking qualitative distinction
between the two sorts of waveforms; one was characterized by unitary depolarizing currents and
the other by unitary hyperpolarizing currents.
To quantify how different the two waveforms were, we devised a metric called the "deflection
index". It measures the asymmetry of the direction of the membrane current over multiple
timescales. The value of the index is positive when the membrane trajectory is hyperpolarizing
and negative when membrane travels in a depolarizing direction. Plots made by measuring the
value of the index from short to long intervals for single cells confirm the impression made by eye,
Figure 3C. The curves for the (labeled and putative) interneurons peaked at a positive value and
those for the (labeled and putative) relay cells peaked at a negative value. Further, we used
principal component analysis to characterize the structure of the curves plotted in Figure 3C. A
histogram of results shows that the data divided into two separate modes, Figure 3D.
Importantly, all labeled relay cells displayed one profile and all labeled interneurons the other.

Voltage dependence of intracellular currents recorded from relay cells and interneurons
It seemed unlikely that the disparity between waveforms merely reflected levels of membrane
polarization rather than differences in the complement of synaptic inputs. As can easily be seen
23

in Figures 1 and 3, recordings were made at holding levels above the reversal for inhibition and
just below the threshold for firing in order to visualize excitatory and inhibitory input (e.g. see
Figure 12). Further, one class of response did not switch to the other, not even during the
unfortunate instances when inhibitory input became unphysiologically pronounced as a damaged
neuron depolarized, lost the ability to spike and expired. As well, we usually recorded both types
of response in a single animal (19 out of 22), sometimes one immediately after the next,
suggesting the difference in response does not correlate with different animals or physiological
state.

Figure 4. Voltage dependence of postsynaptic potentials recorded from relay cells and
interneurons. Spontaneous inputs to a relay cell (A) and interneuron (B) recorded during
different amounts of current injection, as indicated at left. Amplitudes of PSPs as a function of
current injection, normalized to PSP amplitude at rest for 3 relay cells and 3 interneurons; error
bars indicate the standard deviation; darker lines indicate responses shown in A and B (C).
Records from relay cells are in black and from interneurons in blue.

To assess the voltage dependence of the intracellular waveforms systematically, we made
recordings from single cells while we injected different amounts of depolarizing and
hyperpolarizing current through the pipette. As anticipated for EPSPs, the inputs recorded from
24

the majority population (labeled and putative relay cells) grew larger as the membrane was made
more hyperpolarized, Figure 4A and see (Lo and Sherman, 1990). By contrast the unitary events
recorded from the minority population (labeled and putative interneurons) reversed sign when
the membrane was made negative to the threshold for firing, as expected for events dominated by
IPSPs, Figure 4B. Averaged event amplitude with respect to the sign and strength of current
injected (3 examples for each cell type) are plotted in Figure 4C, providing a graphical illustration
of the voltage dependence illustrated by the raw traces. These results support the conclusion that
the disparity between the shapes of the unitary events reflects different patterns of synaptic input,
one dominated by excitation and the other by inhibition. They also show that changes in resting
level do not convert one class of waveform to the other.
From the analyses in Figures 3 and 4, we conclude that relay cells and interneurons can be
classified based on their synaptic responses. Thus, for the remainder, we refer to cells whose
intracellular waveforms are dominated by EPSCs as relay cells and those dominated by IPSCs as
interneurons.
Are there other physiological characteristics that distinguish relay cells from interneurons?
Previous work in vitro showed that interneurons often have thinner action potentials than relay
cells (Pape and McCormick, 1995) but that the distributions overlap (Pape and McCormick,
1995). We did not attempt to measure spike width at half height since the high frequency
components of the intracellular signals were sometimes filtered. Likewise the general shape of
bursts fired at anode break seemed different for feline relay cells and interneurons (Pape and
McCormick, 1995) but too variable for quantitative comparisons across class.
25

Figure 5. Visual modulation of synaptic inputs to relay cells and interneurons. Responses to
disks of the preferred and anti-preferred contrast flashed in the centers of the receptive field of an
OFF-center relay cell (A, black) and interneuron (B, blue). The dendritic arbors of each cell are
drawn above responses to two individual presentations of the stimulus (darker colors) and the
average for all trials (lighter color); gray bars indicate stimulus duration.

Visual modulation of synaptic inputs in relay cells and interneurons
How do the different patterns of synaptic input illustrated in Figures 3 and 4 sum to create
the push-pull responses that were depicted as averages in Figure 1? To address this question we
analyzed individual responses to flashed dark and bright disks. For relay cells, the push, or
excitation to a stimulus of the preferred contrast, was made from trains of EPSCs. This pattern of
behavior is illustrated by responses of an OFF cell to dark disks; Figure 5A, top, shows two
sample trials of the stimulus in black with the average of all trials in a lighter shade. By contrast,
26

the pull, inhibition to a stimulus of the non-preferred contrast, evoked by bright disks was graded,
Figure 5A, bottom. The situation for the interneurons was inverted. The push was quite smooth,
it was often impossible to visualize depolarizing synaptic events in the single trials, Figure 5B, top.
On the other hand, the pull was made by the accumulated contribution of rapid trains of
inhibitory events, Figure 5B, bottom. For an impression of responses recorded in current-clamp
mode, see Figure 11, which shows individual responses recorded from the cell illustrated in
Figure 1.
The receptive fields of all the unitary excitatory and inhibitory events had a center-surround
structure, as if driven by retinal afferents. We asked if the unitary events shared another feature
with ganglion cells, sustained fast rates (Bullier and Norton, 1979; Frishman and Levine, 1983;
Kara et al., 2002; Ruksenas et al., 2000). Cortical neurons in layer 6 also project to the thalamus
but their maintained rates are slow (Gilbert, 1977), at least in the anesthetized animal. Thus, we
detected unitary events before, during and after bright and dark disks were flashed in the center of
the receptive fields (see EXPERIMENTAL PROCEDURES). Sample traces of membrane
currents recorded from a single relay cell and interneuron are shown above histograms of event
counts (EPSCs for relay cells and IPSCs for interneurons) from five different cells of each type,
Figure 6. Responses to the disk of the preferred contrast are shown on the left, Figure 6A, C, and
to the non-preferred contrast on the right Figure 6B, D. The range in sustained event rates
(measured during the second half of the stimulus interval) for excitatory stimuli was 34.2 to 76.9
(54.5 ± 17.3, mean ± standard deviation) event/s for interneurons and 40.3 to 184.3 (112.2 ± 65.5)
27

event/s for relay cells. The range for suppressive stimuli was 0.8 to 9.2 (4.1 ± 3.8) event/s for
interneurons and 0.0 to 38.2 (8.2 ± 16.8) event/s for relay cells.

Figure 6. Rates of unitary synaptic events recorded from relay cells and interneurons.
Responses to disks of the preferred (A) and anti-preferred (B) contrast flashed in the center of the
receptive field of a relay cell (a single trial in black over the variance across trials in gray) shown
atop histograms of EPSC rates for the same (black) and 4 additional relay cells (different colors);
bin size is 5ms. Inset shows a segment of the recording at an expanded time scale and doubled
gain to reveal differently sized EPSCs. Companion responses and plots of event rate for 5
interneurons (C, D).

The event rates for some relay cells (e.g. black traces, Figure 6 A, B) were faster than those for
others and all interneurons. These higher rates might represent convergent retinal input (Hamos
et al., 1987; Levick et al., 1972; Mastronarde, 1992; Usrey et al., 1999). Accordingly, an expanded
28

segment of the trace in Figure 6B, inset reveals large and small EPSCs, presumably generated by
more than one ganglion cell. The results of this analysis suggest that the unitary events recorded
from interneurons and relay cells can keep pace with either single or multiple retinal inputs.

Spatial distribution of receptive fields of relay cells and interneurons
Anatomical studies shows that relay cells and interneurons populate the full extent of the
geniculate (Fitzpatrick et al., 1984; Montero, 1991; Van Horn et al., 2000). Thus, one would
expect that the two types of responses should be recorded at all retinotopic positions. We
assessed the distribution of the two types of cell with respect to the location of the receptive field
in visual space. The range of occurrence for relay cells, Figure 7A, B, and interneurons, Figure
7C, D, were similar, as predicted. The entries on the polar plots, Figure 7A, C, and histograms,
Figure 7B, D, cover most of the lateral geniculate. The inferior and superior locations that seem
under sampled are represented by only narrow crescents of the nucleus (Sanderson, 1971). Thus,
the two types of processing we describe are a ubiquitous feature of all layers of the lateral
geniculate nucleus.
29

Figure 7. Spatial distribution of relay cells and interneurons. Spatial distribution of the
receptive fields of relay cells (A, B) and interneurons (C, D) shown as polar plots (A, C) and
frequency histograms (B, D). Different types of relay cells illustrated are with different colors, as
indicated in the legend.

Transmission of information using different forms of synaptic inhibition
Thalamic inhibition ultimately affects vision by means of action potentials sent from relay
cells to cortex. Thus, we asked how the different forms of excitation and inhibition (jagged versus
smooth) might be adapted to convey visual information to the cortex by using simulations with a
“leaky integrate-and-fire” relay cell and analyzing the results using information theory (see
EXPERIMENTAL PROCEDURES). For simplicity, the simulated neuron received excitatory
30

input from one retinal cell and one interneuron; more of fewer presynaptic cells could be
simulated by changing input rate (Figure 8A). To focus on the influence of inhibition, the EPSC
rate was held constant (Figure 8B) while the IPSC (Figure 8C) rate was modulated in time and
thus conveyed information (note that the statistics of natural scenes suggest that simultaneous
push and pull commonly occur (Simoncelli and Olshausen, 2001)). We then ran a series of
simulations in which the shape of the inhibition ranged from jagged, i.e. few large IPSCs (Figure
8D, E), to smooth, i.e. more numerous small IPSCs (Figure 8F, G), while keeping the total
amount of inhibitory postsynaptic current constant (see EXPERIMENTAL PROCEDURES).
The two forms of inhibition had very different effects on the relay cell’s firing pattern, as
illustrated by two curves that plot the relay cell’s firing rate as a function the strength of inhibition
for relatively jagged (closed circles) and smooth (open circles) profiles, Figure 8H. The firing rate
proved far more sensitive to smooth inhibition; the gain (i.e. the slope of that curve) is greater
than that of the other.
31

Figure 8. Synaptic transmission of information by different forms of inhibitory inputs. (A) A
schematic drawing of a leaky integrate-and-fire neuron that receives excitatory and inhibitory
input. (B, C) Modeled firing of the two presynaptic neurons at 1× smoothness of inhibition.
Spike rasters for 50 trials with instantaneous firing rates (thick curves) overlaid. (D) Modeled
membrane potential of one trial (black curve) superimposed on 9 additional trials (gray curves)
for 1× smoothness of inhibition; dashed line marks threshold. (E) The raster plot and peri-
stimulus time histogram of the relay cell’s spike trains at 1× smoothness of inhibition. (F, G)
Results of the simulation as in (D, E) with 50× smoothness of inhibition. (H) Comparison of the
input-output relationships at 1× versus 50× smoothness of inhibition. The relay cell’s firing rate is
plotted against the strength of the inhibitory input. (I) Rate of information conveyed by the relay
cell’s firing as a function of smoothness of inhibitory inputs; error bars show standard deviation.

32

Moreover, we quantified the amount of information conveyed by the relay cell’s spikes at
different levels of smoothness (Figure 8I). The results showed that smooth inhibition optimizes
the rate of information transmitted from the interneurons to the cortex.

CHAPTER 2 DISCUSSION
Our study provides the first intracellular analysis of how local interneurons in the lateral
geniculate nucleus of the thalamus encode visual stimuli. Studies of relay cells had shown that
ON and OFF stimuli evoke responses of the opposite sign when flashed in the center or in the
surround of the receptive field (Wang et al., 2007) and see (McIlwain and Creutzfeldt, 1967;
Singer et al., 1972). This push-pull arrangement of synaptic responses is most easily explained by
direct feedforward excitation from the retina and indirect feedforward inhibition routed through
local interneurons, Figure 1D. Consistent with this scheme, we found that local interneurons had
center-surround receptive fields. Moreover, the receptive fields of the relay cells and interneurons
were not only qualitatively but also quantitatively similar; a standard computational model was
equally successful in predicting visual responses of both types of cell. Despite this similarity, the
transmembrane currents recorded from the two types of cells during visual stimulation had
almost inverted shapes. Excitatory stimuli evoked rapid sequences of unitary excitatory events in
relay cells but relatively smooth and graded depolarizations in interneurons. By contrast,
suppressive stimuli elicited smooth and graded inhibition in relay cells but rapid trains of unitary
inhibitory events in interneurons. The rates and receptive fields of both types of unitary events
appeared to be inherited, directly or indirectly, from retinal ganglion cells. Thus, the high-
33

frequency components of afferent activity were retained in the excitatory responses of relay cells
but low-pass filtered in the inhibitory responses; and vice versa for interneurons. Collectively, the
results show that relay cells and interneurons in the main layers of the lateral geniculate nucleus
use different patterns of synaptic integration to build similar retinal receptive fields. Further,
simulations using a simple model suggest that the form of synaptic inhibition which the relay cell
ultimately receives is well suited for relaying information from interneurons to cortex.

Physiological differences between relay cells and interneurons
We have found that membrane currents recorded from the lateral geniculate divided into two
different types of waveforms, one associated with relay cells and the other with interneurons. At
first, we wondered if this dichotomy might reflect location of the recording site (i.e. the soma
versus the dendrite). This was not the case, reconstruction of the electrode track confirmed that
all patches were proximal, made at the soma or adjacent to it. We also wondered if the dichotomy
reflected the presence or absence of triads. Triads are structures in which the synaptically coupled
dendrites of an interneuron and a relay cell (usually X, rarely Y or W) or another interneuron
each receive input from a single retinal bouton (Datskovskaia et al., 2001; Famiglietti and Peters,
1972; Guillery, 1969b; Hamos et al., 1985; Szentagothai et al., 1966). However, the membrane
currents recorded from all relay cells, whether or not they had anatomical features that indicated
the presence of triads, were similar (Koepsell et al., 2009; Wang et al., 2007), as is consistent with
work by others (Lo et al., 1991). Rather, the disparities in waveform correlated with relay cells
versus interneurons.
34

Receptive field structure of interneurons
Earlier studies of the thalamus reached the general consensus that interneurons within the
main layers of the geniculate have receptive fields with a center-surround structure, but provided
only descriptive accounts based on extracellular records (Dubin and Cleland, 1977; Humphrey
and Weller, 1988b; Mastronarde, 1992; Sherman and Friedlander, 1988; Wilson, 1989). Our
analyses move beyond these early studies in two essential ways. First, we were able to explore the
inhibitory as well as excitatory contributions to the receptive field because we recorded
intracellularly. Our results show that interneurons, like the relay cells, have a push-pull
arrangement of excitation and inhibition within the center and the surround (i.e. where a bright
stimulus excites, a dark stimulus inhibits and vice versa). Push-pull in a single neuron extends the
dynamic range of operation and also speeds responses to reversals in stimulus polarity (Hirsch,
2003). These effects might be amplified when presynaptic cells also have push-pull responses.
For example, interneurons would alternately inhibit or disinhibit their targets as luminance
contrast changes from the non-preferred to the preferred. The presence of push-pull responses in
both excitatory and inhibitory cells is also seen in retina (Nelson, 1982) and in cortical layer 4
(Hirsch, 2003; Martinez et al., 2005); hence, it appears to be a basic principle for constructing
neural circuits in the early visual pathway. Second, we were able to combine intracellular analysis
with current methods of receptive field mapping and statistical analysis. By adapting a standard
linear-nonlinear model developed to analyze spike trains for use with intracellular records, we
found that the spatiotemporal receptive fields of relay cells and interneurons were equally useful
35

in predicting responses to novel stimuli. Of course we have only begun to explore the most basic
features of the receptive fields of relay cells and interneurons. It is possible, for example, that the
two types of cells will adapt differently to contrast and/or luminance (Bonin et al., 2006; Mante et
al., 2008) (our stimulus was confined to a fixed contrast and so our results do not address such
questions). As well, it is worth remembering that interneurons receive up to seven times more
retinal synapses than do relay cells (Montero, 1991; Van Horn, 2000) and so might have larger
receptive fields; our current sample is too small to determine if this is the case.
Last, it is important to note that our sample might exlude interneurons located in or
bordering the zones between the main layers of the lateral geniculate (none of the neurons labeled
during in this study were located there). These cells have different intrinsic membrane properties
(Sanchez-Vives et al., 1996), appear to receive input from relay cells (Bickford et al., 2008;
Montero, 1989) rather than retina (Montero, 1989) and might be binocular. In fact, interlaminar
neurons might more closely resemble cells in the perigeniculate nucleus of the thalamic reticular
formation more than neurons in the main layers of the geniculate (Montero, 1989; Sanchez-Vives
et al., 1996).

Patterns of synaptic input that build receptive fields
The push-pull structure of the relay cell’s receptive field can be explained by feedforward
excitatory input from the retina and indirect feedforward input from local interneurons that have
overlapping receptive fields and the opposite preference for stimulus contrast, Figure 1D. Before
we began this study, it seemed likely that the same wiring diagram would apply equally well to
36

interneurons. This idea was based, in part, on our past work in cortical layer 4 (the main target of
the thalamus), which showed that intralaminar excitatory and inhibitory cells had similar patterns
of connectivity and synaptic integration (Hirsch et al., 2003). While our present results show that
relay cells and local interneurons have similarly shaped receptive fields, the way in which the two
types of cells process their synaptic inputs is not the same. How might these differences arise and
what are the corresponding implications for retinothalamic connectivity?
The push, or main excitatory drive to relay cells, is almost certainly made of unitary synaptic
inputs that derive from retina. That is, these inputs not only have the prominent size and
stereotyped shapes of retinal input (Blitz and Regehr, 2003; Chen and Regehr, 2000; Koepsell et al.,
2009; Wang et al., 2007) but also preserve the fast rates and receptive field structure of ganglion
cells (Bullier and Norton, 1979; Frishman and Levine, 1983; Kara et al., 2002; Ruksenas et al.,
2000). Why, then, should unitary retinal inputs to interneurons be difficult to detect in vivo?
Biophysical models and intracellular studies in vitro provide clues. Cable models of interneurons
show that many retinal inputs are electrotonically remote and thus, if the dendritic membrane
were passive, would attenuate greatly before reaching the soma (Bloomfield and Sherman, 1989;
Sherman, 2004). However, recent work shows that retinal input activates L-type calcium currents
that propagate feedforward excitation over long lengths of dendrite (Acuna-Goycolea et al., 2008).
Thus, retinal inputs to interneurons could be masked by the intrinsic conductances they evoke,
see (Acuna-Goycolea et al., 2008). Moreover, summed input from multiple retinal inputs
(Acuna-Goycolea et al., 2008; Montero, 1991; Van Horn et al., 2000) and metabotropic
components of synapses between ganglion cells and interneurons (Godwin et al., 1996b;
37

Govindaiah and Cox, 2004; Govindaiah and Cox, 2006) could further smooth the time course of
feedforward drive. Last, local collaterals of relay cells might supplement retinal inputs to relay
cells (Bickford et al., 2008) and interneurons (Bickford et al., 2008; Cox et al., 2003; Lorincz et al.,
2009) by providing disynaptic feedforward excitation.
For relay cells, we hypothesize that the pull, or inhibitory response to stimuli of the opposite
contrast, derives from local interneurons whose pooled input (Hamos et al., 1985) averages to
generate a graded signal. Local interneurons also have strong pull responses, but these are built of
a series of single hyperpolarizing deflections that occur at the fast maintained rates typical of
ganglion cells (Bullier and Norton, 1979; Frishman and Levine, 1983; Kara et al., 2002; Ruksenas
et al., 2000). A brief depolarizing notch often precedes each deflection, indicating the presence of
a leading excitatory component. We can only speculate about the circuitry that generates this
pattern of response and how it might differ from mechanisms that build the pull in relay cells.
Here we outline one simple idea that takes the rate and shape of the deflections into account and
is consistent with ultrastructural evidence that shows that interneurons not only receive
substantial input from the retina (Montero, 1991; Van Horn et al., 2000) but also form
dendrodendritic synapses with each other (Coomes et al., 2002; Famiglietti, 1970; Pasik et al.,
1976). We illustrate this idea using an ON interneuron whose push is generated by presynaptic
ON ganglion cells in the retina, as above. Imagine that a dendrite of the ON interneuron is also
contacted by an OFF ganglion cell which simultaneously synapses with a dendrite of a nearby
OFF interneuron, Figure 10. Further, the ON interneuron is postsynaptic to the dendrite of that
OFF interneuron, which releases transmitter every time the OFF ganglion cell fires. For the ON
38

interneuron, each retinal spike would generate a monosynaptic EPSP that is rapidly overtaken by
a disynaptic IPSP fed forward from the OFF interneuron, producing an event like that we observe.
Of course other patterns of connectivity might shape the events that build the pull in
interneurons. The notches that precede IPSCs might come from strong intrinsic repolarizing
currents (Person and Perkel, 2005), dendritic spikelets (Contreras et al., 1993) or electrically
coupled cells (Deleuze and Huguenard, 2006; Landisman et al., 2002). There are also alternatives
to a dendrodentric origin for the IPSPs. These signals might be generated by axonal input from
multiple interneurons or, though unlikely (Acuna-Goycolea et al., 2008; Govindaiah and Cox,
2006), by a subset of sign-inverting retinal synapses onto interneurons (by analogy to the
connection between photoreceptors and ON bipolar cells). It is doubtful that the pull comes from
the perigeniculate nucleus of the reticular formation. Reticular neurons provide weak if any input
to local interneurons (Cucchiaro et al., 1991; Wang et al., 2001) and, in any event, do not have
receptive fields with a center-surround structure (Dubin and Cleland, 1977; Sanderson et al., 1971;
Uhlrich et al., 1991).

Different patterns of synaptic inhibition influence the transmission of information to the cortex.
Relay cells fire action potentials that lock to retinal input with millisecond fidelity (Dubin and
Cleland, 1977; Koepsell et al., 2009; Sincich et al., 2007; Usrey et al., 1999); presumably, such tight
coupling between ganglion and relay cells is based on the large and discrete (jagged) shapes of
retinogeniculate EPSP. This temporal precision is important because spike timing is critical for
39

encoding sensory information (Butts et al., 2007; Koepsell et al., 2009; Liu et al., 2001; Reinagel
and Reid, 2000) and for activating cortical targets (Temereanca et al., 2008; Usrey et al., 1998).
The contribution of local interneurons to cortical processing is indirect since only relay cells
project outside of the thalamus. We asked if one or the other form of inhibition we have observed,
smooth or jagged, might better permit the relay cell to transmit information conveyed by the
suppressive input. By combining simulations with information theory, we found that the smooth
form of inhibition, paired with the jagged excitation, optimized the amount of information
transmitted from inhibitory input to the relay cell’s spike train.
How could the local microcircuit generate this smooth, or low-passed, profile? The push
signal in interneurons is probably blurred by the prolonged, regenerative currents (Acuna-
Goycolea et al., 2008) engaged by retinal input, as described above. This form of synaptic
integration might serve to decouple the timing of a ganglion cell’s input from an interneuron’s
output. Past work provides support for this idea. Cross-correlations made from spike trains of
simultaneously recorded ganglion cells and putative interneurons are broader than those made
from ganglion cells and relay cells (Dubin and Cleland, 1977). Thus, convergent inhibitory inputs
to a relay cell would arrive asynchronously and average to form a smooth signal; additional
dendrodendritic input, if present, might be graded. Last, why should the pull signal in
interneurons retain the high-frequency component of retinal spike trains? Perhaps the reason is
to disinhibit relay cells on the timescale of single EPSPs and so preserve the temporal structure of
retinal input. In the future, we hope to test temporal relationships between relay cells and
interneurons by making multicellular recordings from the thalamus.
40

CHAPTER 2 EXPERIMENTAL PROCEDURES
Preparation
Adult female cats (1.5 - 3.5 kg) were prepared as described earlier (Wang et al., 2007);
anesthesia induced with propofol and sufentanil (20 mg/kg + 1.5 μg/kg, i.v.) and maintained with
propofol and sufentanil (5 μg/kg/hr + 1.5 μg/kg/hr, i.v.). All procedures were in accordance with
the guidelines of the National Institute of Health and the Institutional Animal Care and Use
Committee of and the University of Southern California.

Stimulation
Discs and annuli and natural movies were displayed at 19 - 50 frames per second on a
computer monitor (refresh rate 144 - 160 Hz) by means of a stimulus generator (Vsg2/5 or
ViSaGe, Cambridge Research Design, Ltd., Cambridge, UK) as described earlier (Wang et al.,
2007). We also used 2D Gaussian white noise at 33% contrast with a spatial resolution of 0.5 or 1
degree (luminance values below 0 and above 2×mean were truncated); one stimulus trial typically
included 16384 frames, updated at 48 Hz with a video refresh of 144 Hz.

Recordings
Whole-cell recordings with dye filled pipettes were made with standard techniques (Hirsch et
al., 2003) except that we often used electrodes with resistances >20 MΩ to improve chances of
41

recording from small cells. Signals were recorded with an Axopatch 200A amplifier (Axon
Instruments, Inc., Union City, CA, now Molecular Devices, Sunnyvale, CA), digitized at 10 - 20
kHz (Power1401 data acquisition system, Cambridge Electronic Design, Ltd., Cambridge, UK)
and stored for further analysis. It was often impractical to assign absolute resting voltage as the
ratio of access to seal resistance led to a voltage division in the neural signal (Hirsch et al., 1998).
Unless otherwise noted all recordings were made above the reversal potential for inhibition and
below the threshold for firing. The integrity of the recordings was monitored by responses to
current injection.

Anatomical analysis
Following histological processing (Hirsch et al., 2003) cells were identified as interneurons
(Guillery type III cells (Guillery, 1966)) using standard criteria (Bloomfield and Sherman, 1989;
Friedlander et al., 1981; Guillery, 1966; Humphrey and Weller, 1988b; Pape and McCormick,
1995; Sherman and Friedlander, 1988) such as complicated and often thin dendrites, appendages
on distal processes and small somas (see Figure 9). Different classes of relay cells were also
distinguished on the basis of various anatomical characteristics, such as somal size, shape of the
dendritic arbor and the presence of grape-like appendages on primary dendrites (Friedlander et
al., 1981; Humphrey and Weller, 1988b). Some cells were reconstructed in 3D using a
Neurolucida System (MicroBrightfield, Colchester, VT).

42

Spatiotemporal receptive fields and linear-nonlinear models
Standard methods of reverse-correlation (Schwartz et al., 2006) were used to compute the
spatiotemporal receptive fields except that the continuous membrane current (from which action
currents were removed (Wang et al., 2007)) was substituted for traditional, discrete spike times, as
follows. First the stimulus was rewritten as a 2D matrix, , of size where was the
number of time bins and the number of pixels in the receptive field. The receptive field was
then T

T
, where was the continuous response signal of size 1 . If the
stimulus is white noise, as in our experiments, its autocorrelation will be identity and the
receptive field can be computed simply by reverse correlation T
.
We computed the spatiotemporal receptive fields from first 15/16 (15360 frames) of the
Gaussian white noise sequence and used these as the linear component of the model (the
remaining 1/16 (1024 frames) was reserved to assess the performance of the model). The time bin,
or temporal resolution (20.8 ms) was set by the rate of stimulus update (48 Hz). The static
nonlinearity function was estimated by fitting (least-mean-square) the intracellular response to
the output of the spatiotemporal receptive field. The shape of the nonlinearity for interneurons
and relay cells was captured by a sigmoid function , which takes into account
slight saturation and thresholding of the response. When a linear function was substituted for the
sigmoid, the prediction of the model was only slightly worse, less than 3%; thus this choice of
parameterization did not appreciably influence the performance of the model. Finally, the
performance of the model was assessed by cross-validation (using the reserved 1/16 of the data)
43

and quantified as the explained variance in the response that the model predicted (since the data
did not contain multiple trials of the same stimulus, the percentage of explained variance was
calculated with respect to the total variance in the signal).

Deflection index
In order to capture asymmetric structures in the direction, or sign, of the membrane
trajectory across various time scales, we devised an index that reflects the dominance of inward
versus outward deflections in the intracellular signal. First, we differentiated the membrane
current signals (with action currents removed, see (Wang et al., 2007)) to yield a set of
processed signals ; by differentiators of multiple time scales .
; 1

We then formed distributions of the differentiated signals and computed the deflection
indices as the skewness of the differentiated signals.

Here and are the second and third central moments of the distribution of
; . To characterize the asymmetry of deflection across all time scales, we normalized the
deflection indices and performed a principal component analysis on the normalized index as a
function of time scales for the whole population of 119 neurons.

44

Event sorting and counting
Intracellular, voltage-damp recordings were filtered digitally (Gaussian filter, 0.5ms
bandwidth) and then differentiated twice. Potential neural events, spikes and unitary synaptic
currents, were detected as concave local minima (zero crossing of the first derivative with a
negative second derivative). Neural events were clustered using commercial software (Spike2,
Cambridge Electronic Design, Ltd., Cambridge, UK).

Leaky integrate-and-fire model
To compare the influences different types (i.e. jagged versus smooth) of synaptic inputs have
on the output spike train, we built a leaky integrate-and-fire model to simulate the transmission
of sensory information conveyed by inhibitory inputs. The evolution of the membrane potential
(dimensionless) was governed by the following differential equation

where and were are the excitatory and inhibitory postsynaptic current, and

is the leak current. When crosses a fixed threshold 1 , an output spike is
generated and then it is instantaneously reset to 0 . The reversal potential for the leak
conductance is also set to 0 .
The synaptic currents in the model and derived from two presynaptic spike trains,
and .

45

where and are kernels of synaptic transformation; and are neural response
functions for the excitatory and inhibitory presynaptic spike trains.

In practice, synaptic inputs were treated as instantaneous events:

and

and the time constant of the membrane was 25 ms. Thus, EPSPs and IPSPs,
both, had an instantaneous rise and had and exponential decay of 25 ms. The size of single EPSP
was 0.8, as for earlier models (Carandini et al., 2007). The input spike trains, and , were
Poisson point processes with rate functions and .
In order to explore how well different forms of inhibition convey information, the rate of
excitatory input was held constant,
= 30 spike/s. We generated by filtering a stimulus
(Gaussian white noise) by a realistic (difference-of-Gamma) biphasic receptive field (Cai et al.,
1997) and then using an exponential nonlinearity. The mean firing rate of inhibition was
controlled by the scale of the nonlinearity. Thus, we were able to compare simulations in which
different forms of inhibition, ranging from jagged to smooth interacted with the same excitatory
waveform. For each case of smoothness, the mean firing rate of inhibition was set to
and the size IPSCs set to
/ in order to balance the total amount of excitation
and inhibition. Finally we evaluated the influence of the various forms of inhibitory input on the
relay cells’ spike train by plotting firing rate against the strength of the inhibitory input
46

(angular brackets indicate average over trials). Since the rate of excitatory input rate was
held constant, all changes information encoded in the relay cell’s spike train resulted from
changes in the pattern of inhibition.
The information rate of the relay cells’ spike train was assessed using the following formula
(Brenner et al., 2000).
log

where is the time-dependent firing rate of the output spike train with a mean firing rate of .
Estimations of information rate were obtained from 250 simulated trials, with each trial lasting 30
seconds.

CHAPTER 2 SUPPLEMENTAL MATERIALS
We included three supplemental figures in this section. Figure 9 illustrates anatomical
features used to classify inhibitory neurons. Figure 10 is a schematic diagram of a circuit that
might explain the synaptic responses we recorded from interneurons; it pertains to the
DISCUSSION. Figure 11 illustrates single-trial responses of the interneuron shown in Figure 1.
47

Figure 9. Morphology of interneurons. (A) Histogram comparing the soma size for labled
interneurons and relay cells; distributions are similar to those reported previously (Fitzpatrick et
al., 1984). (B) Micrographs of distal processes (branch order as indicated) for 5 interneurons
show characteristic appendages grouped as clusters or standing alone on slender stalks; arrow
heads point to examples. Each row shows images from a different cell.

48

Figure 10. Diagram of a hypothetical circuit for the interneuron’s receptive field. Retinal
ganglion cells, bottom, connect with thalamic interneurons, top. Neurons are drawn as their
receptive fields with OFF subregions blue and ON subregions red. For excitatory cells, the center
is solid whereas the surround is unshaded and delimited by a solid line. For inhibitory cells, the
center is hatched and the surround unshaded; both types of subregions are bordered by dashed
lines. Connections (boutons) are color coded to match the stimulus preference of the presynaptic
neuron and plus and minus symbols indicate the sign of the input.

Figure 11. Intracellular responses of a thalamic interneuron. Responses evoked by stimuli of the
preferred (A) or non-preferred (B) contrast; same cell as in Figure 1. Three individual trials are
illustrated above the averaged response to 10 trials.

49

CHAPTER 3
THALAMIC CONTROL OF DUAL MODES OF FIRING
DURING NATURALISTIC VIEWING
Thalamic relay cells transmit information from retina to cortex by firing either rapid bursts or
tonic trains of spikes. Bursts occur when the membrane voltage is low, as during sleep, because
they depend on channels that cannot respond to excitatory input unless primed by strong
hyperpolarization. Cells fire tonically when depolarized, as during waking. Thus, mode of firing
is usually associated with behavioral state. Growing evidence, however, suggests that sensory
processing involves both burst and tonic spikes. To ask if visually evoked synaptic responses
induce each type of firing, we recorded intracellular responses to natural movies from relay cells
and developed methods to map the receptive fields of the excitation and inhibition the images
evoked. In addition to tonic spikes, the movies routinely elicited lasting inhibition from the
center of the receptive field that permitted bursts to fire. Therefore naturally evoked patterns of
synaptic input engage dual modes of firing.

CHAPTER 3 INTRODUCTION
Thalamic relay cells determine how input from the eye is transmitted to cortex. The pattern
of activity that these neurons send downstream depends strongly on the level of membrane
polarization. When the membrane rests at depolarized levels, relay cells produce tonic trains of
action potentials but when the membrane is hyperpolarized they fire rapid bursts. The bursts are
50

initiated by calcium channels that open transiently at voltages below spike threshold and remain
inactivated until exposed to substantial hyperpolarization (Jahnsen and Llinas, 1984a). Early
recordings made from unanaesthetized animals suggested that the tonic mode of firing was
associated with wakefulness and the burst mode with drowsiness or sleep (Fourment et al., 1984;
Livingstone and Hubel, 1981; Steriade et al., 1993). Recent work, however, indicates that firing
mode is not strictly linked to behavioral state (Denning and Reinagel, 2005; Guido et al., 1992;
Lesica and Stanley, 2004; Ramcharan et al., 2000; Reinagel and Reid, 2000; Swadlow and Gusev,
2001; Wolfart et al., 2005). Bursts, though not common, occur routinely in awake animals and
can be evoked by sensory stimuli (Guido et al., 1992; Ramcharan et al., 2000; Swadlow and Gusev,
2001; Weyand et al., 2001). The possibility that bursts contribute to normal sensory function is
important. For example, the temporal pattern of spike trains determines the amount and type of
information that can be encoded about the stimulus (Denning and Reinagel, 2005; Liu et al., 2001;
Reinagel et al., 1999). Further, bursts activate the cortex more effectively than slower trains of
spikes (Swadlow and Gusev, 2001; Swadlow et al., 2002), likely because they evoke synaptic
potentials that summate in time (Usrey et al., 2000) and also because they occur after long silences
that permit recovery from synaptic depression (Swadlow and Gusev, 2001; Swadlow et al., 2002).
Extracellular studies have shown that visually evoked bursts are most likely to occur after
prolonged exposure to non-preferred stimuli (Alitto et al., 2005; Denning and Reinagel, 2005;
Lesica and Stanley, 2004). These findings suggest lasting and suppressive stimuli somehow evoke
hyperpolarizations strong enough to revive the calcium channels that trigger thalamic bursting
(Alitto et al., 2005; Denning and Reinagel, 2005; Lesica and Stanley, 2004). To explore the
51

intracellular mechanisms that might prime bursts during vision, we made whole-cell recordings
in vivo. To lay the foundation for our study we explored the synaptic basis of the thalamic
receptive field. Tests with simple visual patterns revealed that relay cells have receptive fields in
which stimuli of the reverse contrast evoke synaptic responses of opposite sign, an arrangement
called push-pull. Since both the excitatory and inhibitory contributions to the receptive field have
the center-surround structure characteristic of ganglion cells, we concluded that feedforward
circuits give rise to both push and pull. Further, we observed that lasting stimuli of the non-
preferred contrast evoked strong inhibition that enabled bursts. This result recalled ecological
viewing conditions: in nature, light levels within the receptive field can remain steady for long
durations (Denning and Reinagel, 2005; Dong and Atick, 1995). Thus it seemed reasonable to
suppose that synaptic responses to time varying natural images (movies) could drive the
membrane between tonic and burst modes. Our recordings confirmed this prediction. We next
devised methods to map the spatiotemporal organization of the excitation and inhibition the
movies evoked to learn how response pattern relates to receptive field structure and, by inference,
underlying circuitry. Our results lead to a simple conclusion: retinogeniculate (feedforward)
inhibition driven from the center of the receptive field is sufficient to prime bursts for all types of
relay cells.

CHAPTER 3 RESULTS
To explore how visually evoked synaptic input influences firing pattern we made whole cell
recordings from 42 relay cells in 12 adult cats. We recorded from all layers (A, A1, C) of the
52

lateral geniculate and were able to classify 20 cells by anatomical criteria (Friedlander et al., 1981)
including X cells (n = 9), Y cells (n = 8), W cells (n = 3).

Figure 12. Spatially opponent excitation (push) and inhibition (pull) in the relay cell’s receptive
field. (A) Anatomical reconstruction of an ON center Y cell in layer A1; scale bar 100 μm. (B)
Averaged responses to ten trials of bright (red) and dark (blue) disks and annuli in the receptive
field center and surround; disk size 1°, annulus size 2°, 20°; horizontal bars under the traces mark
stimulus duration; scale bars are 10 mV and 200 ms in this and the following panel. (C) Voltage
dependence of the hyperpolarization evoked by suppressive stimuli; response recorded at different
holding currents: -0.2 nA (dark gray line), 0 nA (black line), -0.2 nA, -0.4 nA (light gray line). (D)
Wiring diagram for feedforward, push-pull responses mediated by local interneurons, left and for
feedback inhibition from the perigeniculate nucleus, right. Cells are represented as their receptive
fields; blue indicates OFF subregions, red ON subregions and minus signs label interneurons;
solid and dashed lines indicate excitatory and inhibitory connections respectively. Drawings of
overlapping ON and OFF receptive fields in the retina and the lateral geniculate nucleus are offset
in the figure for the purpose of illustration.

53

Synaptic structure of thalamic receptive fields
Extracellular recordings have shown that thalamic relay cells, like retinal ganglion cells, have
circular receptive fields made of two concentric subregions that have the opposite preference for
stimulus contrast; bright stimuli falling within ON subregions evoke firing as do dark stimuli
shown within OFF subregions (Wiesel, 1959). We made whole-cell recordings of responses to
classical stimuli, disks and annuli, to map the dominant patterns of excitatory and inhibitory
input to the thalamic receptive fields. Our intracellular analyses showed that within each
subregion, stimuli of the opposite contrast evoke responses of the opposite sign – a push-pull
profile. For example, a bright disk flashed in the ON center of a Y cell, Figure 12A, evoked a
depolarization (push), Figure 12B, top left, whereas a dark disk of the same size and position
evoked a strong hyperpolarization (pull), Figure 12B, bottom left. Rebound responses of the
opposite sign followed the withdrawal of the stimulus; traces are averages of ten trials. A similar
pattern was elicited by annuli presented to the surround, Figure 12B, right. To determine if the
pull (stimulus evoked hyperpolarization) was caused by synaptic inhibition versus withdrawal of
excitatory drive, we made recordings when the membrane potential was lowered by means of
current injection. The amplitude of the response diminished with progressive hyperpolarization,
Figure 12C, (n = 4 cells). These results suggest that the pull was dominated by GABAergic
currents, which reverse below the resting membrane potential (Crunelli et al., 1988). Note that a
hyperpolarization resulting from withdrawal of excitation would have had the opposite voltage
dependence; it would have increased in amplitude when the membrane was made more negative.
54

A simple circuit that might explain the responses we recorded is illustrated in Figure 12D,
left. The excitation, or push, is created by ganglion cells whose receptive fields have the same
position and center sign as the relay cell. The inhibition, or pull, is generated by interneurons
supplied by ganglion cells whose receptive fields share the same position but opposite sign as the
relay cell. There is, in fact, evidence that interneurons in the main layers of the lateral geniculate
nucleus have center-surround receptive fields (Humphrey and Weller, 1988b; Sherman and
Friedlander, 1988). By contrast, the inhibition we recorded was unlikely to be structured by input
from the overlying perigeniculate nucleus, which receives input from relay cells en route to cortex
and feeds back inhibition to the lateral geniculate. Receptive fields in the perigeniculate are
usually large, irregularly shaped and have overlapping ON and OFF responses (Dubin and
Cleland, 1977; Uhlrich et al., 1991; Wörgötter et al., 1998), Figure 12D, right.

55

Figure 13. Intracellular responses of a relay cell to repeated presentations of natural movies.
(A) Anatomical reconstruction of an OFF center X cell in layer A; scale bar 100 μm. (B) Insets
show examples of movie frames that introduced marked changes in luminance in the receptive
field center (blue ellipse); arrows point to the time at which each frame appeared. (C) Receptive
field of the retinal input reconstructed from responses to repeated presentations of the movie;
white ellipse is the 1.5σ contour from a 2D Gaussian fit. (D) Clips of responses to three
presentations of the movie recorded in current-clamp mode; scale bars 200 ms and 10 mV. (E)
Responses to the same segment of the movie recorded in voltage-clamp mode, which damped
intrinsic conductances. (F) Separation of components of the raw data, as illustrated for the lowest
trace in E. The black trace shows spikes, the red traces shows templates fit to each EPSC and the
blue trace show the residual (spike and EPSC subtracted) currents, which were mainly
hyperpolarizing but also contained occasional putative T currents (asterisk), scale bars 200 ms and
100 pA. (G) Responses to defined suppressive stimuli. Individual responses to a bright disk
flashed in the center of the receptive field of an OFF cell are shown in black above the average of
20 trials in gray. Bursts followed withdrawal of the stimulus, red asterisks. Horizontal bars mark
stimulus duration, the stimulus size was 1° and the scale bars are 100 pA and 100 ms.

56

Tonic and burst firing mode in response to naturalistic stimulation
In order to understand how the structure of the receptive field influences neural output under
ecological viewing conditions, we recorded responses evoked by various natural movies as
illustrated for an OFF center X cell, Figure 13A. Insets at top, Figure 13B, show frames of a
movie of windblown tree branches in which the luminance of the patch that fell within the center
of the receptive field, Figure 13C, was variously brighter or darker. Intracellular responses to
repeated clips of the natural movie show that the stimulus evoked a broad but reproducible range
57

of behaviors that included both tonic and burst firing. Responses recorded in current clamp,
Figure 13D, are shown above those recorded in voltage-clamp mode, Figure 13E. Voltage clamp
mode mildly damped the membrane currents. Thus, strong depolarizing currents that evoked
bursts and spikes remained robust but small fluctuations in the shapes of individual synaptic
events were reduced such that excitatory post-synaptic currents (EPSCs) could be reliably
detected and analyzed. When the frames of the movie are compared to the recordings, one sees
that darker patterns that fell over the center of the receptive field evoked high rates of synaptic
input and tonic firing whereas brighter patterns led to hyperpolarizing responses. Moreover, the
transition from lasting brightness to darkness was accompanied by bursts of spikes on top of slow
depolarizing waveform, Figs. 2D, E. This sequence, in which bursts followed prolonged exposure
(of the receptive-field center) to stimuli of the non-preferred to the preferred sign, was common
in our sample and has been reported in extracellular studies (Alitto et al., 2005; Lesica et al., 2006).
Similar results were found when the movies were shown through a central aperture within the
receptive field center (n = 2, not shown), suggesting that most of the response we observed was
mediated from the center versus the surround of the receptive field. Further, disks of the non-
preferred contrast flashed in the center of the receptive field also evoked inhibition that primed
bursts. (Figure 13G).

Mapping receptive fields from responses to natural movies
In order to understand the responses to natural movies in terms of the contributions from
different regions of visual space, we developed means to extract the receptive fields of the inputs,
58

excitatory and inhibitory synaptic currents, and the outputs, spikes. The initial step was to
separate the different components of the response. Unitary events, namely the excitatory synaptic
inputs and spikes, were sorted on the basis of peak amplitude and maximum slope. The
inhibitory part of the response could not be obtained in the same fashion because individual
inhibitory post-synaptic currents (IPSCs) were not visible; rather they pooled to form graded
currents. Thus, our approach was to extract the inhibitory component by removing the excitatory
events from the intracellular signal. First, we constructed a record in which each spike, Figure
13F, top, and EPSC, Figure 13F, middle, was represented by a template of the average event in
each class. We then subtracted those simulated waveforms from the raw record (see
EXPERIMENTAL PROCEDURES). The residual signal, Figure 13F, bottom, consisted almost
entirely of the hyperpolarizing responses, albeit with a small contribution from unclamped
inward currents like those that drove bursts, asterisk. These inward bumps under the bursts most
likely resulted from low-threshold calcium currents, T-currents (Jahnsen and Llinas, 1984a),
possibly with modest contributions from glutamate receptors (Blitz and Regehr, 2003) or rapid
trains of retinal input (Sincich et al., 2007) masked by intrinsic conductances. As a control for the
above method of isolating inhibition, we subtracted records made at membrane levels negative to
the holding levels and found similar results (n = 2, see SUPPLEMENTAL MATERIALS, Figure
20).
The next step was to map the receptive fields of the different components of the responses to
movies. To accomplish this task the intrinsic spatial and temporal correlations in the movies
59

(Dong and Atick, 1995) had to be taken into account. The approach we used was to obtain the
receptive fields by finding the spatiotemporal kernel that best predicted the actual neural response
(see EXPERIMENTAL PROCEDURES). Although the method yields only the linear component
of the response, the predictions fit the data quite well, as expected from previous analyses of
thalamic spike trains (Dan et al., 1996; Mante et al., 2005). Further, we checked the
reconstructions made from the natural stimuli against receptive fields mapped conventionally by
sparse or dense noise and found similar patterns of push and pull for all stimulus conditions (see
EXPERIMENTAL PROCEDURES). The responses to the movies were dominated by excitation
and inhibition evoked from the center of the receptive field, as illustrated for morphologically
identified X, Y and W relay cells of both center signs. Anatomical reconstructions, Figure 14A,
are shown above the receptive fields for push (EPSCs), Figure 14B, top; for pull (inhibition),
Figure 14B, middle; and for spikes, Figure 14B, bottom, white ovals mark 2D Gaussian fits of
the center. The temporal pattern of response for each component is shown in Figure 14C.
60

Figure 14. Receptive fields of synaptic excitation (push), inhibition (pull) and spikes
reconstructed with natural movies. (A) Anatomical reconstruction of six relay cells; anatomical
class and center polarities as labeled; scale bar 100 μm. (B) Receptive fields of the push (top), the
pull (middle) and spikes (bottom); white ellipses are 1.5σ contours of 2D Gaussian fits of the
centers. (C) Time course of the responses. (D) Scatter plot of the overlap index against the ratio
of sizes of the push and pull with histograms of the distributions of the overlap index (top) and the
ratio of sizes are (right) next to graphical depictions of the two measures; crosses outline the mean
± SD. (E) Elliptic eccentricities of the receptive field centers of push, pull and spikes; horizontal
bars indicate the mean ± SEM. Values for X cells are in red, Y cells in blue, W cells in green and
remaining unlabeled cells in black.

61

It seemed logical to think that the excitatory postsynaptic events came from retina; there is
general consensus that feedforward input shapes the relay cell's receptive field (Cleland et al., 1971;
Levick et al., 1972; Usrey et al., 1999). By contrast, intracellular studies in vitro show that
corticogeniculate EPSCs inputs are difficult to detect (Granseth and Lindstrom, 2003), as is
consistent with their origin on distal dendrites (Sherman and Guillery, 1996). We further
reasoned that if the shape of the pull evoked by the movies resembled that of the push, then the
pull was likely to arise from feedforward retinal input relayed via local interneurons rather than
feedback from the perigeniculate (see Figure 12D). Thus, we quantified the similarity between
the push and pull using several measures, as below.
First we adapted Schiller’s overlap index (Schiller et al., 1976) (see EXPERIMENTAL
PROCEDURES) to measure the extent of overlap between the push and pull in the central
subregion of the receptive field. The index gives a value of 1 for cospatial subregions and of 0 for
subregions that lie side by side. The score for the entire population was high, 0.720 ± 0.022 (n =
41), indicating a high degree of overlap (Martinez et al., 2005). Scores within different classes of
cells were similar, Figure 14D, top histogram: X cells (0.740 ± 0.025, n = 9); Y cells (0.761 ± 0.033,
n = 8); W cells (0.787 ± 0.025, n = 3); unlabeled cells (0.686 ± 0.038, n = 21). Next, we asked
whether the pull and push were similar in size by calculating the ratio of their center areas. The
value of the pull-push ratio was 1.00 ± 0.04 (n = 41) for the whole sample, Figure 14D, right
histogram. The ratios within classes were 0.87 ± 0.08 (X cells, n = 9), 1.01 ± 0.06 (Y cells, n = 8),
0.93 ± 0.09 (W cells, n = 3) and 1.05 ± 0.07 (unlabeled cells, n = 21). A plot of the overlap index
62

against the pull-push ratio illustrates that the push and pull usually overlapped and were of
comparable area; the colored crosses are centered on the population mean and mark the standard
deviation, Figure 14D. Last, we asked if the shape of the pull was as round as that of retinal input
by using an index of elliptical eccentricity for which 0 indicates circular symmetry and values that
approach 1 indicate progressively ovoid shapes. The values for X cells were 0.61 ± 0.05 for push
and 0.65 ± 0.03 for pull (n = 9); for Y cells, 0.54 ± 0.05 for push and 0.63 ± 0.05 for pull (n = 8);
for W cells, 0.72 ± 0.08 for push and 0.83 ± 0.02 for pull (n = 3) and for unlabeled cells, 0.60 ±
0.03 for push and 0.61 ± 0.04 for pull (n = 21); values for the entire sample were 0.60 ± 0.02 (n =
41) for push and 0.64 ± 0.02 (n = 41) for pull. Although the degree of elongation of the receptive
fields varied somewhat, as reported for retinal ganglion cells (Shou et al., 1986), the push and pull
had similar geometries (see EXPERIMENTAL PROCEDURES), Figure 14E. In sum, all three
indices are consistent with a feedforward origin of the pull.
The overlap indices between push and pull that we measured here (and see (Martinez et al.,
2005)) are consistent with predictions based on the anatomical layout of ON and OFF ganglion
cells in the retinal (Wässle et al., 1981) and the spatial extent of ganglion cells' receptive fields
(Peichl and Wässle, 1983). Specifically, the mean value we calculate from our data, 0.72, and the
value we calculate based on studies of the retinal mosaic 0.70 are similar (see SUPPLEMENTAL
MATERIALS). Of course other features could influence the relative overlap between push and
pull such as convergence and divergence in the retinogeniculate connectivity (Hamos et al., 1987).

63

Feedforward inhibition primes bursts
Visual inspection of our records had suggested that long periods of inhibition primed bursts
during naturalistic viewing, Figure 13. To quantify the relationship between visually evoked
hyperpolarization and firing mode, we made spike triggered averages (STAs) of the neural
responses that preceded tonic spikes (tSTA) or the first spike in a burst (bSTA) (Lu et al., 1992).
There were stereotyped differences between the averages computed for tonic and burst spikes as
exemplified by two cells, one for which we recorded the membrane voltage, Figure 15A and the
other for which we recorded the membrane current, Figure 15B. Specifically, there was a
sustained hyperpolarization, Figure 15A, or outward current, Figure 15B, before burst spikes
whereas a brief depolarization, Figure 15A, or inward current, Figure 15B, preceded tonic spikes.
This difference was easily seen when spikes and EPSCs were subtracted from the averaged
currents that preceded either burst or tonic events and the two resulting signals were overlaid,
Figure 16C. A further analysis of the difference between the shape of the pull that precedes tonic
spikes with long interspike intervals and bursts is provided in the SUPPLEMENTAL
MATERIALS, see Figure 21.
64

Figure 15. Pull from the center of the receptive field precedes bursts but not tonic spikes evoked
by natural movies. (A, B) Burst (bSTA) and tonic (tSTA) spike-triggered-averages of membrane
current (I
m
) and voltage (V
m
) of two relay cells (black traces); sigmoid fit of bSTA and single
exponential fit of tSTA (gray bands); scale bars are 100 ms and 2 mV in A, 100 ms and 25 pA in B.
(C) Time courses of the hyperpolarization, or outward currents, and depolarization, or inward
currents, preceding burst versus tonic spikes; values for membrane currents are shown in white
and membrane voltage in black for different cell types, as indicated.

We quantified the membrane deflections preceding bursts and tonic spikes by fitting the time
courses of the spike triggered averages (see EXPERIMENTAL PROCEDURES). The times to
reach half maximal amplitude for the inhibition that preceded bursts were 250.0 ± 28.0 ms (n = 12)
for membrane voltage and 248.7 ± 18.7 ms (n = 22) for membrane current. The times to reach
half maximal amplitude for the excitation preceding tonic spikes was 39.3 ± 10.6 ms (n = 12), for
membrane voltage and 36.7 ± 8.2 ms (n = 22) for membrane current. There were minor
differences among X, Y and W cells, Figure 15C, but the general pattern of response was the same.
65

Figure 16. Model prediction of the occurrence of burst and tonic spikes based on feedforward
synaptic inhibition. (A) Receptive field of synaptic inhibition reconstructed for the cell in Figure
15B. (B) Predicted response of the “pull” by the receptive field (red) overlaid with the actual
“pull” response (black); scale bar is 1 s. (C) bSTA (red) and tSTA (black) of the “pull” for the cell
in Figure 15B; gray shades a 500 ms time window before the occurrence of spikes and scale bars
are 100 ms and 10 pA. (D) Scatter plot of the second versus first principle components of the
“pull” waveform in a 500 ms window before tonic (black dots) and burst (red dots) spikes. Green
arrow represents the vector linking the tSTA to the bSTA. (E) Distribution of tonic (black) and
burst (red) events along the tSTA-bSTA axis; black ticks mark the different threshold levels used
to generate ROC curves in F and the blue line shows the threshold level used to predict event
identity in G. (F) Receiver operating characteristic (ROC) curves of the model's performance
(black) on another movie and for bootstrap resampling (gray); dots represent different threshold
levels; gray lines mark the 95% bootstrap confidence interval and the filled blue circle marks the
threshold level used in G. (G) Prediction of spike type (tonic or burst) based on the inhibitory
waveforms taken from responses to another movie; thick red lines represent bursts, scale bar is 1 s.
Note that the model produces serial bursts because it does not include a refractory period.

To illustrate further the link between feedforward inhibition and the generation of bursts, we
first ensured that the spatio-temporal receptive field constructed from the neural signal (Figure
16A) was able to predict the visually-evoked pull (Figure 16B). Next, we built a simple model to
66

show how the pull signal recorded in response to one movie can predict the occurrence of bursts
evoked in response to a different movie. To construct the model, we extracted the waveform of
the pull response in a 500 ms time window before each spike, tonic or burst (Figure 16C), and
used principal component analysis to capture the structure of the pull (Figure 16D). The
waveforms preceding tonic and burst spikes formed two displaced distributions in the space
spanned by the first two principle components, with the tSTA and the bSTA as centers-of-mass of
the distributions (Figure 16D). To classify the different shapes of the pull that precedes tonic
versus burst spikes, we projected the data onto a line (green arrow in Figure 16D, E) that
connected the tSTA and the bSTA and then selected a simple criterion with a fixed arbitrary
threshold on this axis (blue line in Figure 16E). The model predicted the occurrence of the bursts
quite well (Figure 16F, G), as follows. First, receiver operating characteristic (ROC) analysis
(black curve in Figure 16F) shows that the model correctly classified most bursts and gave few
false positives for a wide range of thresholds. Second, we generated a new ROC curve (gray curve
in Figure 16F) made by testing the model with a surrogate dataset produced with a bootstrap
procedure (1000 iterations) that randomized the temporal relationship between the pull and the
spike trains. The 95% confidence intervals for the surrogate data set lie far from the original ROC
curve; showing that the model's performance is statistically significant.
Taken together, our results suggest that bursts are primed by feedforward inhibition,
presumably mediated via local interneurons. That is, bursts are preceded by lasting periods of
67

hyperpolarizing input and the receptive fields reconstructed from these hyperpolarizations have
the shape of retinal input.

Figure 17. Retinogeniculate synaptic inputs trigger putative T-currents that evoke bursts. (A)
Crosscorrelogram of putative T-currents with EPSCs (EPSC at t = 0). The shift predictor has been
subtracted from the raw correlogram; bin width, 1.0 ms; n = 41. (B) Crosscorrelogram of the first
spikes in bursts with the putative T-currents (T-current at t = 0); conventions and sample size as
in A. (C) Conditional probability densities of the interval between a burst and most recent retinal
EPSC (EPSC at t = 0); time bin, 0.5 ms. Within-trial (solid line) and cross-trial (dashed line)
probability densities intercept at about 8 ms, arrow; the contribution value, i.e. the shaded area, is
0.80 in this case; n = 41. (D) Contribution values calculated for retinal inputs to putative T-
currents and putative T-currents to bursts for intervals of 0 - 12 ms and -5 - 10 ms respectively.
(E) Examples of retinal inputs that trigger T-currents that initiate bursts in turn; scale bars 50 pA
and 20 ms.
68

Feedforward excitation initiates thalamic bursts
There are two simple possibilities that could account for the mechanism that triggers bursts,
intrinsic regenerative currents that are activated by prolonged hyperpolarization (e.g. the anode
break response) or synaptic excitation. We examined possible synaptic contributions by
computing cross-correlations between retinal synaptic events and putative T-currents, Figure
17A, and between putative T- currents and the first spike of the burst, Figure 17B. We
performed this analysis on the population rather than single cells because the mean burst rate was
low, 0.32 Hz, compared to 11.4 Hz for all spikes (recall that the putative T- currents were obtained
from the trace that remained after spikes and EPSCs were extracted from the raw data, Figure 13F,
asterisk). The stimulus dependent correlations in the response were removed by subtracting the
shift predictor (Perkel et al., 1967). The correlograms were standardized by dividing the event
count within each bin by the total event count and the binwidth, which is approximately
equivalent to calculating the conditional probability density of two types of event, either p(t; T-
current|EPSC), Figure 17A, or p(t; burst|T-current), Figure 17B. The analysis suggests that
bursts were initiated by T-currents that, themselves, were evoked by retinal input. That is, the tall
peak between 0 and 12 ms in the correlogram for T-current to EPSC, Figure 17A, indicated that
T-currents were most likely to occur just after retinal input. There was similar relationship
between the onset of a T-current and the first spike of a burst, Figure 17B.
69

To tie together the relationship between retinal inputs and bursts, we computed the
correlogram between each burst and the EPSC that preceded it most directly, Figure 17C. In this
figure, we plot both the raw correlogram (solid line) and the shift predictor (dashed line). This
plot reveals a close temporal relationship between the triggering EPSC and the first burst spike, 0 -
8 ms. Most importantly, the plot addresses the question of how many of the bursts, and T-
currents, were initiated by retinal drive. In other words, it illustrates the "contribution" (Levick et
al., 1972), or estimate of percentage of cases in which a burst was initiated by retinal input; the
maximum possible value is 1.0. For our sample, the contribution was high, 0.8, as measured by
integrating the 0 - 8 ms peak of the correlogram, shaded region. Similarly high values for the
contribution of EPSCs to T currents and for T currents to bursts were found in all populations of
cells, Figure 17D. Insets below illustrate examples of the sequence of EPSC, putative T-current
and burst for an X, a Y and a W cell, Figure 17E.

CHAPTER 3 DISCUSSION
We asked if synaptic inputs evoked by natural visual stimuli are able to induce burst as well as
tonic modes of firing in relay cells. First, we investigated the synaptic structure of the relay cell's
receptive field with simple stimuli and found a push-pull arrangement in both center and
surround; that is, excitation from the retina was complemented by strong inhibition evoked by
stimuli of the opposite sign. Second, we found that responses to natural movies were composed
of two main patterns of response, either trains of excitatory synaptic events that elicited tonic
spikes or inhibitory periods followed by bursts. Third, we reconstructed the receptive fields of
70

synaptic inputs evoked by the movies and observed that the inhibition that preceded bursting
came mainly from the central subregion. The shape of the reconstructed inhibitory field matched
the shape of excitatory field reconstructed from retinal inputs, suggesting that it is made by
feedforward circuits (Figure 12D). Finally, the bursts themselves were initiated by direct retinal
input rather than intrinsic membrane properties. Thus, natural stimuli evoke sequences of
inhibition and excitation that engage dual modes of activity in the thalamus.
A great deal of evidence suggests that retinal (feedforward) input rather than cortical
(feedback) determines the shape of the thalamic receptive fields. For example, the receptive fields
of relay cells have the same outline as those of their strongest retinal inputs (Cleland et al., 1971;
Levick et al., 1972; Usrey et al., 1999) and change only modestly after cortical feedback is removed
(Cudeiro and Sillito, 1996). The two sources of inhibitory input to the lateral geniculate nucleus,
local intranuclear interneurons versus neurons in the overlying perigeniculate nucleus, have very
different receptive fields. Specifically, there is evidence that local interneurons have discrete
center-surround receptive fields (Dubin and Cleland, 1977; Humphrey and Weller, 1988b;
Sherman and Friedlander, 1988) whereas most neurons in the perigeniculate have large
amorphous receptive fields in which bright and dark stimuli are excitatory throughout (Uhlrich et
al., 1991; Wörgötter et al., 1998), Figure 12D.
Because we recorded intracellularly, we were able to measure the inhibitory as well as the
excitatory contributions to the relay cell's receptive field. We found strong push-pull in both the
center and surround, consistent with earlier pharmacological (Sillito and Kemp, 1983) and
intracellular (Martinez et al., 2005; McIlwain and Creutzfeldt, 1967; Singer and Creutzfeldt, 1970)
71

studies. Large annular shapes evoked the strongest response from the surround whereas stimuli
like movies or noise favored the center. The receptive fields mapped from noise or movies were
similar, even though the reconstructions of the receptive field captured only the linear component
of the response, which, as expected (Dan et al., 1996; Mante et al., 2005), accounted for more than
half of the variance of the responses we recorded.
The dominant pull in the relay cell's receptive field appears to be fed forward from retina via
local interneurons because its spatial structure resembled that of retinal receptive fields rather
than the receptive fields of cells in the perigeniculate nucleus. This inhibition, when recruited by
natural movies, was able to drive the membrane from tonic to burst mode. Burst triggered
averages showed that the inhibitory epochs that preceded bursts typically lasted for hundreds of
milliseconds, long enough to deinactivate T-type channels (Jahnsen and Llinas, 1984a;
McCormick and Huguenard, 1992; Soltesz et al., 1989). By contrast, tonic spikes were preceded
by brief depolarizations. For almost all bursts we recorded, the underlying slow depolarizations
(the putative T-currents) were initiated by excitatory synaptic events. Thus, natural movies
typically evoke bursts that are primed by prolonged inhibition but initiated by retinal drive.
Although bursts usually occurred at the end of a hyperpolarizing interval, we occasionally
observed cases in which they were produced by a single excitatory event that arrived in the midst
of prolonged inhibition. As well, we found that many putative T-currents were not large enough
to drive the membrane across the threshold for sodium spikes or led to only one sodium spike.
This variability in the appearance and action of T-currents recalls a recent study in vitro that
showed that the effects of these currents change depending on the level of background synaptic
72

activity in both hyperpolarized and depolarized regimes (Wolfart et al., 2005). Apparently,
synaptic interactions with T-channels have diverse effects on firing, with bursting the most
stereotyped case.
Our study illustrates the importance of the receptive field in influencing firing mode.
Naturalistic stimuli generate bursts in response to marked changes in luminance within the center
of the receptive field. Thus, the pull within the receptive field is not only able to suppress firing
but also to heighten the response to changes in stimulus polarity. This was common to all types
of relay cells in all layers of the lateral geniculate. Thus, our results are consistent with
extracellular (Alitto et al., 2005; Denning and Reinagel, 2005) and theoretical studies (Lesica and
Stanley, 2004) that conclude that bursts signal transitions from periods of lasting brightness or
darkness.
Somewhat similar interactions between inhibition and firing mode might operate at the level
of retina since ganglion cells have T-type channels (Lee et al., 2003) and receptive fields with
push-pull (Wiesel, 1959). In fact, based on extracellular analyses of the thalamus, it had been
proposed that every spike in a thalamic burst is generated by a corresponding presynaptic retinal
input (Sincich et al., 2007). Intracellular recordings do not support the view that thalamic bursts
depend exclusively on presynaptic mechanisms, however. The frequency of visually evoked bursts
increases when the membrane voltage of the postsynaptic relay cell is lowered by current injection
(see Figure 20 and (Lu et al., 1992)). Moreover, this increase significantly exceeds the number of
bursts that the retina generates (see Figure 23 for complete analysis).
73

Figure 18. Extraction of neural events and components from voltage-damped recordings. (A)
Raw recording of membrane current, scale bar 50 pA and 200 ms in this and other panels. (B)
The same signal smoothed and differentiated; all local minima were used as candidate neural
events. Events ultimately classified as EPSCs are labeled with red and those classified as spikes
with green in this and remaining panels. (C) Scatter plot of the candidate events plotted as peak
value (slope in the raw signal) against area under the peak (magnitude in the raw signal). (D)
Trains of spike times including bursts (magenta). (E) EPSC times. (F) Illustration of spikes
(action currents) isolated from the raw signal. (G) Isolated action currents. (H) Fitting the
template to model EPSCs. (I) Modeled EPSCs as the EPSC train convolved with the template. (J)
Residual currents remaining after subtraction of the action currents and modeled EPSCs; blue line
indicates 1/10 spike height. (K) Times of putative T-currents; defined as events that crossed and
remained below the blue line for longer than 10 ms in H.

Although our experiments were performed in anesthetized animals, our recordings were
made when the membrane rested in tonic mode. In that regime, natural movies evoked bursts
infrequently but predictably, reminiscent of patterns observed in awake animals (Guido and
Weyand, 1995; Ramcharan et al., 2000; Swadlow and Gusev, 2001; Weyand et al., 2001). This
74

sparseness of occurrence does not mean that bursts are unimportant. Bursts might be used to
provide a history of changes in the visual environment and to signal rare events that (perhaps
because they are rare) are especially important to attend (Alitto et al., 2005; Denning and Reinagel,
2005; Lesica and Stanley, 2004; Lesica et al., 2006). Thus, it seems that evolutionary pressures
favored the use of the same membrane channels in both sleep and in wakefulness.

CHAPTER 3 EXPERIMENTAL PROCEDURES
Extraction of synaptic currents in the intracellular membrane current
Preparation, stimulation and recordings. Anesthetized adult cats (1.5 - 3.5 kg) were prepared
as described earlier (Hirsch et al., 1998) except that anesthesia induced with propofol and
sufentanil (20 mg/kg + 1.5 μg/kg, i.v.) and maintained with propofol and sufentanil (5 μg/kg/hr +
1.5 μg/kg/hr, i.v.). All procedures were in accordance with the guidelines of the National Institute
of Health and the Institutional Animal Care and Use Committee of and the University of
Southern California. Whole cell recordings with dye filled pipettes were made with standard
techniques (Hirsch et al., 1998) (Axopatch 200A amplifier, Axon Instruments, Inc., Union City,
CA), digitized at 10 kHz (Power1401 data acquisition system, Cambridge Electronic Design, Ltd.,
Cambridge, UK) and stored for further analysis. It was often impractical to assign absolute
resting voltage as the ratio of access to seal resistance led to a voltage division in the neural signal
(Martinez et al., 2005). Following histology cells were identified as X, Y or W using criteria
outlined in (Friedlander et al., 1981; Humphrey and Weller, 1988b).
75

The stimuli, natural movies, sparse (Hirsch et al., 1998) and 2D Gaussian noise, and discs and
annuli were displayed at 19 - 50 frames per second on a computer monitor (refresh rate 128 - 160
Hz) by means of a stimulus generator (Vsg2/5 or ViSaGe, Cambridge Research Design, Ltd.,
Cambridge, UK). In some instances the movies were shown through an aperture in a uniform
mask set to mean luminance.
EPSCs and spikes. Intracellular voltage damped signals, Figure 18A, were digitally filtered
(Gaussian filter, 0.5ms bandwidth) and then differentiated twice to identify spikes and EPSCs,
Figure 18B. Potential spike and synaptic events were characterized as concave local minima (zero
crossing of the first derivative with a negative second derivative). Event times for spikes
and EPSCs
were identified from the set of potential events by plotting amplitude (the area
under the peak of the first derivative) against slope (the peak value of the first derivative). That is,
the spikes and EPSCs formed distinct clusters in this plot and could be separated from each other
and from noise by means of a threshold criterion, Figure 18C. Sometimes EPSCs split into
multiple clusters, indicating the presence of different inputs; in these instances we combined all
the clusters of EPSCs that were separable from noise into one group for further analysis.
IPSCs. IPSCs could not be resolved individually and were visible only as slow outward
currents. Our strategy to isolate these currents was to subtract spikes and EPSCs from the total
membrane current. We eliminated spikes by removing the data points from the raw signal at
each within a window of -0.5 - 1.5 ms and then closing the gaps by cubic interpolation,
Figure 18F. EPSCs were less simple to remove since their size varied and their long time courses
76

lead to frequent overlap. Thus, we modeled individual EPSCs by a linear-rise-exponential-decay
function:

; ,

,

,, ,
0

e
1 e

The EPSC template was fitted in two steps: First the linear rise of the template was fitted to
the EPSC onset (-1.5 to 2.5 ms window around each
) to determine the amplitude ,Figure
18H, upper trace. Second, the exponential decay was fitted to the EPSC-triggered-average of
membrane current that followed the onset of the EPSC, Figure 18H, lower trace. The current
that remained after removing spikes, Figure 18G, and EPSCs, Figure 18I, was a slow outward
current, Figure 18J, containing mainly IPSCs but including other rare features such as the
putative T-currents. This current was used to map the pull.
For two cells, we cross-checked this method of extracting the pull signal by subtracting a
recording made at a hyperpolarized voltage from one made at a depolarized voltage
(hyperpolarization reduces the amplitude of synaptic inhibition). The receptive fields obtained
from both methods were similar. For almost all cells, as a separate control, we compared the
receptive fields reconstructed from the movies with those made by averaging responses to the
noise stimuli; the fields were always similar.
Putative T-currents. First, the signal was high pass filtered at 1.0 Hz to remove slow
artifactual drift in the signal. Putative T-currents were detected in the pull currents by using a
threshold criterion; inward deflections were counted as putative T-currents if their amplitude
77

exceeded one tenth average action current height for more than 10 ms, Figure 18J. The onsets
were defined as the time the signal crossed threshold, Figure 18K.

Tonic- and burst-spike-triggered-average
Identification of tonic and burst spikes. Bursts were defined by criteria used in extracellular
studies as these standards hold up well for intracellular recordings (Ramcharan et al., 2000).
Specifically, bursts were defined as events that comprised two or more spikes spaced less than 4
ms apart and in which the first spike of the rapid train occurred no less than 100 ms after the most
recent spike. The start of each burst was taken to be the timing of its first spike.
Spike-triggered-average of membrane current and voltage. The cross-correlogram of
membrane current/voltage to burst and tonic spikes was generated by event-triggered-averaging.
In recordings with a sufficient number of events the cross-correlogram to burst events were fitted
well by a sigmoid:
1exp /

and the cross-correlogram to tonic events were fitted well with a single exponential:
·2
/
/
where /
was used to quantify the time course in both cases. For many cells, the cross-
correlograms for bursts were very noisy because bursts occurred at low frequency (mean, 0.32 Hz,
n = 41). Our criterion for including cross-correlograms for further analysis was that the
correlation coefficient with the fitting function satisfied 0.75 .
78

Figure 19. Explanation and controls for receptive field reconstruction. (A) Fraction of explained
variance by the fitted (blue) and the predicted response (red) as a function of , the Lagrange
multiplier for regularization; prediction is made with a different movie. (B) The kernel that best
predicted the receptive field was chosen as the reconstructed receptive field, gray arrow in A. (C)
Predicted response (red) by the optimized kernel overlaid with the actual response (black); scale
bar is 500 ms. (D, E) Side-by-side comparisons of receptive fields reconstructed from responses to
movies (left) and receptive fields mapped with sparse noise (right). White ellipses are
1.5σ contours of 2D Gaussian fits of the centers; yellow ellipses replicate the white ones on the
left.

Reconstruction of spatio-temporal receptive field
Spatiotemporal receptive fields were estimated from responses to natural movies by
optimizing a linear convolution model for predicting responses from the stimulus. Firing rates of
EPSCs and spikes were estimated by a temporal Gaussian filter, 25 ms half-width. Since the pull
signal was analog versus digital, like spikes or EPSCs, we estimated the pull by averaging the
79

membrane currents in each frame of the movie. Spontaneous (stimulus independent) response
levels were estimated by average responses over the entire recording.
The receptive fields were estimated as the linear convolution kernels that minimized the
quadratic error between predicted and measured responses. For optimization we used a
regularized gradient method (Machens et al., 2004). The kernel, stimulus and response were
discretized in space and time. The spatial bin was determined according to the size of the
receptive field; bin size, range 0.125° - 1.0° and the temporal bin was set to the duration of a movie
frame; bin size, range 20.0 - 52.6 ms. The total duration of the kernel was set to 600 or 800 ms.
The three-dimensional kernel was reshaped into a one-dimensional vector , and similarly, the
stimulus into a matrix, . The optimization function is then given as a quadratic form:

1
2
· · 2
· ·
with the first term the prediction error and the second a regularization term for avoiding over-
fitting (because of limited amount of data, usually has more columns than rows). The
regularizer smoothed the kernel by taking differences of immediate neighbors in all three
dimensions (two spatial and one temporal).
We used a faster conjugate gradient algorithm than that used in earlier methods (Machens et
al., 2004) to find the minimum. The minimization procedure was initialized with a random
kernel and a Lagrange multiplier for regularization that was set to a large value. After each
optimization step using the conjugate gradient algorithm, was reduced in a stepwise fashion
with a power-law decay (Figure 19A). The decay parameter for was made small enough to
80

prevent trapping in local minima. The annealing process ended when reached an empirical
value that we determined from a case with sufficient data for cross-validation (Figure 19A, gray
arrow); this value of was that one that minimized prediction error of the response to a different
movie stimulus. Convergence was verified by using different initializations. Centers of
reconstructed receptive fields from natural movies by means of this approach agree with “sparse
noise” maps reasonably well (Figure 19D, E).
Reconstructed receptive fields were normalized to absolute peak value. To obtain the spatial
component of the receptive field we integrated the time frames within the temporal impulse
response peak of the center pixel. Temporal receptive fields were estimated by spatial integration
of the spatiotemporal receptive field weighted by the Gaussian fit of the spatial receptive field
center. The spatial receptive fields in the figures were plotted as cubic-smoothed contour-plots
(MatLab, The MathWorks, Inc., Natick, MA).

Quantification of receptive field properties
Gaussian fit of the center. The spatial centers of receptive fields were fitted with Gaussians
and the fit parameters used to represent receptive field properties. Each receptive field was first
rectified to its center polarity and fitted by a two-dimensional Gaussian with amplitude , spatial
widths and , rotation angle and center position and (least-mean-square fit using the
Nelder-Mead simplex algorithm, function fminsearch, MatLab).
81

Elliptic eccentricity. We chose
and defined the elliptic eccentricity as

/
0, 1 as measure for the degree of elongation of the receptive field.
Overlap Index. To quantify the extent to which two spatial receptive fields overlap, Schiller's
overlap index (Schiller et al., 1976) was extended to two dimensions: for two receptive fields with
the Gaussian parameters ,

,

,
,

,

and ,

,

,
,

,

, the two-
dimensional overlap index was defined by
Ω

where

, 1, 2 or 2, 1 are the widths of the two
Gaussians along the direction linking their centers, with

and the
Euclidian distance between the centers.
Receptive field size. To compare sizes of two receptive fields, we use the ratio of the areas
within the 1/e contour:

Cross-correlation and estimation of contribution
The cross-correlograms of two discrete event trains were plotted as conditional probability
density functions approximated by frequency histograms of conditional occurrence of one event
at various temporal distances from the other. To compensate for stimulus-driven correlations we
generated shift predictors by forming cross-correlograms of shuffled trials (Perkel et al., 1967).
82

The contribution of events A in triggering events B was defined as the fraction of all events B
preceded by positively correlated events A (cross-correlation larger than shift-predictor, i.e. inside
the peak of correlogram) (Levick et al., 1972; Usrey et al., 1999).

CHAPTER 3 SUPPLEMENTAL MATERIALS
Calculation of ON-OFF overlap index based on spatial statistics of retinal mosaic
We predicted the overlap between push and pull in the relay cell's receptive field based on
current knowledge about the retinal mosaic (the anatomical layout of the somas and dendrites of
different classes of ganglion cell) and the overlap between receptive fields of neighboring ganglion
cells. We defined the geometric parameters of the retinal mosaic as follows: r, the radius of the
ganglion cell's dendritic field; D, the mean distance between nearest neighbors of same sign; d, the
mean distance between nearest neighbors of the opposite sign and R the radius of the receptive
field center. It was reasonable to set r ≈ D because “the distance to the nearest neighbor for both
on- and off-centre cells was about 90 μm, which agrees well with the dendritic field radius of 100
μm” (Wässle et al., 1981). We assumed that d ≈ 0.5D ≈ 0.5r because values for the mean distances
between the nearest ON-ON neighbors is 90 μm, between nearest OFF-OFF neighbors is 85 μm
and between nearest ON-OFF neighbors is 43 μm (Wässle et al., 1981). Further, we set R = 1.4r
because receptive field centers are, on average, larger than dendritic field of a ganglion cell by a
factor of 1.4 (Peichl and Wässle, 1983). Thus were able to derive the mean overlap index as (2R -
d) / (2R + d) = (2×1.4r – 0.5r) / (2×1.4r + 0.5r) = 0.70. This value agrees very well with our
measurement, 0.72. Of course other features that we did not take into account could influence
83

the relative overlap between push and pull such as convergence and divergence in the
retinogeniculate connection (Hamos et al., 1987; Usrey et al., 1999).

Control for the construction of the inhibitory receptive field
Figure 20 illustrates a control for the method of isolating the pull that we commonly used
(see EXPERIMENTAL PROCEDURES); it compares this method with one in which the pull was
obtained from records made at different membrane levels. Figures 20A-C show responses of an
ON-center Y cell to the repeated presentation of movie clip; the recordings were made in current
clamp. The first 20 trials were recorded when the cell was slightly depolarized and remaining 20
trials were recorded when the cell was hyperpolarized by the injection of current.
Hyperpolarizing the membrane reduced the rate of spontaneous firing and increased the
occurrence of bursts (Figure 20B) but did not alter the pattern of retinal input (Figure 20A). It
also reduced the amplitude of the large fluctuations in the membrane current (Figure 20C). The
"difference signal" made by subtracting the depolarized from the hyperpolarized response (it is
proportional to the change of total membrane conductance) (Figure 20D) matched the “pull
current” extracted with our usual procedure (Figure 20E). The similarity between these two
waveforms validates the method we commonly used extract the “pull” signal (see
EXPERIMENTAL PROCEDURES). The receptive fields for the pull that we reconstructed from
each type of pull signal are similar. (Figure 20F, G).
84

Figure 20. Control for the construction of the inhibitory receptive field. (A) Raster plot of
retinogeniculate EPSPs evoked by a 5-second movie clip. The first 20 trials were recorded when
the cell was depolarized (unshaded), and then hyperpolarized by current injection (shaded);
vertical gray line indicates the onset of the stimulus. Estimated input rate (EXPERIMENTAL
PROCEDURES) is plotted below the rasters, depolarized (black line) and hyperpolarized (red
line). (B) Raster plot of thalamic spikes in which bursts are indicated with red markers. Estimated
firing rate is shown below the rasters. (C) Averaged membrane potential of trials recorded when
the cell was depolarized (black) and hyperpolarized (red). (D) Waveform generated by
subtracting the hyperpolarized from the depolarized record; this "difference signal" is proportional
to the total membrane conductance. (E) The “pull current” obtained from a voltage-damped
recording of the response to the same clip (EXPERIMENTAL PROCEDURES). (F) Receptive
fields reconstructed with the “pull current” (left) and the difference signal (right) from responses
of the same cell to a longer movie. White ellipses are 1.5σ contour of the best Gaussian fit to the
center; the yellow ellipse in the right plot duplicates the white ellipse on the left. (G) Receptive
fields reconstructed for another cell using “pull current” (left) and the difference signal (right);
conventions as in F.
85

Figure 21. Burst, tonic and long-ISI tonic spike-triggered-average of membrane current and
voltage. (A) Burst (red), tonic (black) and long-ISI tonic (blue) spike-triggered-averages of the
membrane current for two cells, the left example is the from the same cell used for Figure 16A.
Scale bars are 100 ms and 25 pA. (B) Burst (red), tonic (black) and long-ISI tonic (blue) spike-
triggered-averages of the membrane voltage for two cells, the left example is from the same cell
used for Figure 16B. Scale bars are 100 ms and 2 mV. Ripples in the STAs reflect intrinsic
oscillations in retinal inputs. (C) The difference index of the burst (pale red) and tonic (gray)
STAs compared with the long-ISI tonic STAs. Data shown were obtained from the same sample
used in Figure 16; error bars represent SEM.

Strength and duration of inhibition before bursts and tonic spikes
To determine if the pull that preceded bursts was stronger than that evoked during tonic
firing, we compared STAs made with different types of reference events — the first spike in a
burst (bSTA), a tonic spike preceded by an ISI > 100 ms (ltSTA) and all tonic spikes (tSTA). The
averages reveal that the degree of suppression before long ISI spikes was far weaker and briefer
86

than that before bursts, as seen in (Figure 21A, B) where the three averages are overlaid for
examples from both current-clamp and voltage-damped recordings. To quantify the disparity
among the three types of STA, we calculated the normalized energy in the difference between two
given STAs made for the 500 ms time window before each spike. The value of this "difference
index" is 0 when both STAs are same and becomes increasingly positive as the shapes of the STAs
diverge. The values for the index for ltSTA - bSTA and ltSTA - tSTA pairs are markedly different
(Figure 21C). Thus, the visually driven synaptic inhibition that primes bursts is significantly
stronger and more prolonged than that which simply prevents firing, consistent with earlier
extracellular studies (Alitto et al., 2005; Sincich et al., 2007).

Figure 22. Tradeoff between strength and duration of the suppression that primes T-currents.
(A) Two vectors used to study the strength-duration tradeoff of the suppression that primes T-
currents; gray lines represent zero level. (B) Distribution of the T-current-triggered ensemble of
membrane currents in the 2D subspace spanned by the two vectors shown in A; the red line is the
least-mean-square linear fit, Pearson’s r value is marked above the data.

Tradeoff between strength and duration of the suppression that primes T-currents
The kinetics of T currents as measured in vitro with precisely controlled voltage steps show a
tradeoff between the duration and the amplitude of the hyperpolarization required to deinactivate
T channels (Jahnsen and Llinas, 1984a). Although it is not possible to make such controlled
87

measures from responses to natural movies recorded in vivo, we were able to demonstrate the
essential feature of the tradeoff from our records. We extracted the waveforms preceding each
putative T-current and explored the structure of this ensemble in a 2D subspace defined by two
orthogonal vectors, v
1
and v
2
(Figure 22A). The coordinates along v
1
are the integration of
membrane currents in a 100 ms window before each T-current; higher values correspond to
stronger outward currents overall. Coordinates along v
2
are differences between the membrane
currents in the first and second half of the time window; higher values mean greater
concentration of outward currents in the most recent 50 ms. Given the amplitude-duration
tradeoff of the inhibition that primes T-channels, it follows that if a T-current were generated by
an overall weaker outward current (a smaller value along the v
1
axis), then this outward current
should be strongest just before the T current. Similarly, if the total outward current were strong,
then there would be no requirement for concentrated strength near T-current. Hence, there
should be a negative correlation between coordinates along v
1
and v
2
. Figure 22B shows such an
example, r = -0.429 (Pearson’s r value); for the same cell; the 95% confidence interval of
bootstrapped r value was [-0.101, 0.115], showing that the negative correlation was significant.
We repeated the analysis for our complete sample; the negative correlation was significant for 33
out of 41 cells and there were no significant positive correlations.

88

Figure 23. Postsynaptic versus presynaptic mechanisms for burst generation. (A) Raster plots of
presynaptic (EPSPs, top) and postsynaptic (spikes, bottom) event trains evoked by a natural movie
(data excerpted from Figure 20A, B). The first 20 trials were recorded when the relay cell was
depolarized and the remaining 20 trials (shaded by gray) were recorded when the cell was
hyperpolarized. Red ticks represent burst firing as classified using the 100 ms, 4 ms criterion; the
dark gray box marks the 60 ms time window used for analysis. (B) Quantification of the
occurrence of short (< 4 ms) ISIs in presynaptic and postsynaptic event trains (EPSP and spike)
within the 60 ms time window used for analysis; error bars represent SEM. Dotted outline above
the leftmost EPSP bar indicates the additional short ISIs in the EPSP superset.

Postsynaptic versus presynaptic mechanisms for burst generation
Our work suggests that the generation of bursts depends on inhibition, that is, on a
postsynaptic mechanism, Others have suggested, however, that thalamic bursts are generated
exclusively by presynaptic patterns of retinal activity (Sincich et al., 2007).
To estimate the relative contribution of pre- and postsynaptic mechanisms to bursting, we
quantified the frequency of occurrence of short (< 4 ms) ISIs in the presynaptic (retinogeniculate
89

EPSPs) and postsynaptic (action potentials) event trains obtained from recordings made when a
relay cell's membrane was either depolarized or hyperpolarized with injected current (see Figure
20 A, B). We reasoned that if every spike in burst were initiated presynaptically, by a retinal input,
then the sequence of EPSPs should contain at least as many short ISIs as the thalamic spike train.
In contrast, if there were many more thalamic spikes with short ISIs than EPSPs, then a
postsynaptic mechanism would be more plausible.
Figure 23A illustrates a clip excerpted from the data in Figure 20A, B, during which the
stimulus changed from the non-preferred to the preferred contrast. The transition evoked an
increase in the rates of EPSPs and of thalamic spikes. However the number of bursts evoked in the
hyperpolarized recording was greater than in the depolarized trace. We quantified ISI rate for
EPSPs and spikes within a 60 ms window centered on the transition, as plotted in Figure 23B.
Consistent with the postsynaptic scenario, the number of spikes with short ISIs greatly exceeded
the number of EPSPs with short ISIs (p < 0.01, Student’s t-test) when the relay cell was
hyperpolarized. To rule out the possibility that this result was incorrect because we missed EPSPs
hidden under spikes, we generated a superset of the retinal inputs by adding an EPSP whenever
there was a spike (dotted line in histogram in Figure 23B); even for this condition the difference
between pre- and postsynaptic event trains was significant (p < 0.05). Note also that there was no
significant difference between the number of EPSPs and spikes with short ISIs in the depolarized
record. Therefore, our data do not support an exclusively presynaptic mechanism for generating
each spike in a thalamic burst.
90

CHAPTER 4
RECODING OF SENSORY INFORMATION
ACROSS THE RETINOTHALAMIC SYNAPSE
The neural code that represents the world is transformed at each stage of a sensory pathway.
These transformations enable downstream neurons to recode information they receive from
earlier stages. Using the retinothalamic synapse as a model system, we developed a theoretical
framework to identify stimulus features that are inherited, gained or lost across stages.
Specifically, we observed that thalamic spikes encode novel, emergent, temporal features not
conveyed by single retinal spikes. Furthermore, we found that thalamic spikes are not only more
informative than retinal ones, as expected, but also more independent. Next we asked how
thalamic spikes gain sensitivity to the emergent features. Explicitly, we found that the emergent
features are encoded by retinal spike pairs and then recoded into independent thalamic spikes.
Finally we built a model of synaptic transmission that reproduced our observations. Thus, our
results established a link between synaptic biophysics and the recoding of sensory information.

CHAPTER 4 INTRODUCTION
The world we see is represented time and again in the hierarchical stages of visual system
(Van Essen et al., 1992). Each of these representations is defined by the activities of neuronal
populations. How neurons encode sensory information and the way in which neural strategies
for coding change from one stage to the next are central problems in systems neuroscience. We
91

addressed these questions using the connection between the retina and lateral geniculate nucleus
(LGN) of the thalamus as a model system.
We chose the retinogeniculate synapse to study for several reasons. First, it is the earliest site
in the visual pathway where sensory information is transmitted across a synapse between spiking
neurons. Second, the synapse is experimentally accessible. Retinal ganglion cells (RGCs) and
relay cells of the LGN form highly specific connections (Usrey et al., 1999) through a strong
excitatory synapse (Blitz and Regehr, 2003; Guillery, 1969a; Scharfman et al., 1990) on proximal
dendrites (Hamos et al., 1987), such that the pre- and postsynaptic spike trains can be
simultaneously recorded by extracellular (Bishop et al., 1958, 1962; Cleland et al., 1971; Hubel and
Wiesel, 1961; Kaplan and Shapley, 1984; Mastronarde, 1987b; Sincich et al., 2007; Sincich et al.,
2009) or intracellular techniques (Koepsell et al., 2009; Wang et al., 2007). Third, neural
responses in the early visual pathway are somewhat simple; retinal and thalamic responses can be
well characterized with standard linear-nonlinear models (Carandini et al., 2005). Last, the
statistical rules of transmission across the retinothalamic synapse (Rathbun et al., 2007; Sincich et
al., 2007; Usrey et al., 1998) have been described in detail and so prescribe a simple mechanistic
model (Carandini et al., 2007) for studying changes in neural coding.
Our approach combined experiments with theory. We recorded from connected retinal and
thalamic neurons by means of “cell-attached” or “whole-cell” patch recordings from the LGN and
then used information theory to compare pre- and postsynaptic spike trains. Just as information
theory has been used to analyze the communication of messages in telephone lines, it can be used
to understand neural processing (Barlow, 1961; Bialek et al., 1991; Borst and Theunissen, 1999;
92

Mackay and McCulloch, 1952; Rieke et al., 1997). This framework has been essential for
understanding how neurons encode sensory information at single stages in the visual system
(Koepsell et al., 2009; Reinagel et al., 1999; Reinagel and Reid, 2000; Sincich et al., 2009).
Here we asked how information is communicated from one stage in the hierarchy to the next.
The first step was to identify differences between the features that pre- and postsynaptic spikes
encoded. For this purpose, we developed a novel “joint encoding model” that was able to
subsume stimulus features encoded by the retinal or by the thalamic neuron. Using the model,
we were able to separate thalamic features that were merely inherited from retinal spikes from
those that emerged, or were “gained” by thalamic spikes in the course of synaptic transmission.
These results allowed us to compare strategies of neural coding in retina and thalamus.
Previous work had shown that information can be encoded by correlations (or patterns) in the
spike train, in addition to single independent spikes (Brenner et al., 2000). Thus, we measured
information encoded by single as well as pairs of retinal spikes. We found that the pairs of retinal
spikes convey significantly more information than independent single spikes. When we analyzed
thalamic spike trains, we found that each thalamic spike encoded more information than a retinal
spike (Sincich et al., 2009) and, moreover, encoded the information more independently. Could
the feature that is “gained” by individual thalamic spikes originate from correlations in the retinal
spike train (given that the single retinal spikes do not encode this feature)? Indeed, we discovered
that pair-wise correlations in the retinal spike train conveyed information about the “gained”
feature, information that is recoded into independent spikes in the thalamus. Last, we built a
mechanistic model (using temporal summation of retina inputs (Carandini et al., 2007; Usrey et
93

al., 1998)) that qualitatively explained how the thalamic neuron recodes the information about the
“gained feature” from retinal spike pairs to independent thalamic spikes.
In sum, our work demonstrates how a correlation code in an early stage of a sensory pathway
(retina) can be transformed into a more efficient (i.e. independent-spike code) in the next stage
(thalamus) by means of a simple biophysical mechanism.

Figure 24. Identification of the joint relevant subspace across the retinothalamic synapse. (A)
“Cell-attached” recording from the LGN and the detected retinal and thalamic spike trains. (B)
Sorting of the retinal and thalamic spikes. (C) Schematic view of the joint relevant subspace
(feature space) of the retinal and thalamic neurons. (D) The default linear-nonlinear-Poisson
encoding model utilizing the joint relevant subspace. (E) Identified joint feature space for an X
cell pair; the pre- and postsynaptic relevant subspaces are both one-dimensional. The raw, RGC-
spike-triggered and LGN-spike-triggered stimulus ensembles are illustrated as scatter plots and
marginal histograms; the ovals and smooth lines represent the Gaussian approximations.

94

CHAPTER 4 RESULTS
Our sample includes 26 neurons in LGN in 12 adult cats; all records were obtained by using
patch recordings in the “cell-attached” or “whole-cell” mode (Wang et al., 2007). Figure 24A
shows an example of a “cell-attached” recording from an X cell in the geniculate. The trace
includes two classes of stereotyped events, action potentials and “S-potentials”, extracellularly
recorded retinogeniculate EPSPs (Bishop et al., 1962; Hubel and Wiesel, 1961; Kaplan and
Shapley, 1984). By using a custom method to sort these events (Wang et al., 2007), we were able
to recover the spike trains of the thalamic cell and the presynaptic RGC (Figure 24B).

The “joint encoding model”
To characterize the different strategies the retina and the LGN use to encode visual
information, we developed a novel computational framework which we call “joint encoding
model” as it describes pre-and postsynaptic neurons at the same time. The “joint encoding model”
is essentially a linear-nonlinear (LN) model. The linear components of this model are filters that
map the high-dimensional spatio-temporal stimulus into the low-dimensional subspace to which
a neuron is sensitive (in essence, the receptive fields). This subspace is usually referred to as the
“relevant subspace” or the “feature space” (Aguera y Arcas and Fairhall, 2003; Bialek and de
Ruyter van Steveninck, 2005; Sharpee et al., 2004). In the “joint encoding model”, the linear filters
subsume the “relevant subspaces” of the pre- (RGC) and postsynaptic (LGN) neuron. We
schematize the pre- and postsynaptic “relevant subspaces” as vectors (solid blue) and
95

(solid red) respectively in Figure 24C ( , 1,, and , 1,, form basis sets of the
pre- and postsynaptic subspaces). The “joint” relevant subspace is the plane spanned by the pre-
and postsynaptic subspaces, Figure 24C; it has higher dimensions than either the pre- or the
postsynaptic feature spaces unless they are identical. Two additional subspaces are important for
characterizing how features are transformed across the synapse, the orthogonal complements of
the pre- and postsynaptic subspaces. These subspaces represent the features that are “gained” and
“lost” after synaptic transmission and are illustrated by (dashed red) and (dashed blue)
in Figure 24C.
The “joint” model can be built with any linear filters that form a basis of the “joint relevant
subspace”. In practice, we used the presynaptic (i.e. ) and the “gained” (i.e. ) feature
(Figure 24D). We obtained these filters from the simultaneously recorded retinal and thalamic
spike trains by using conventional spike-triggered average/covariance (Schwartz et al., 2006) in
addition to information-theoretic approaches (Pillow and Simoncelli, 2006), see
SUPPLEMENTAL MATERIALS for details. The pre- and postsynaptic “relevant subspaces”
were basically one-dimensional and did not coincide, thus they formed a two-dimensional “joint
relevant subspace”. Figure 24E depicts the distributions of the pre- and postsynaptic spike-
triggered stimulus ensembles in the joint-encoding model for an X cell pair. The arrows mark the
presynaptic, postsynaptic, “gained” and “lost” features of the subspace. Note that the
distributions of the pre- and postsynaptic spike-triggered stimulus ensembles are significantly
96

different from each other (blue versus red), suggesting that the feature sensitivity within the “joint
relevant subspace” is altered across the retinothalamic synapse.

Figure 25. Spatiotemporal analysis of the joint retinothalamic feature space. (A) Two types of
Gaussian white noise used in this study. (B) The presynaptic and “gained” stimulus features
(receptive fields) for an X cell; side and bottom plots show the factorized spatial and temporal
profiles. (C) Comparison of spatial and temporal components between the direct and indirect
features. (D, E) Cumulative energy of the direct and indirect features over space and time; the
spatial or temporal scales of the direct features are normalized to 1. (F, G) Population comparison
(26 RGC-LGN pairs) of the spatial and temporal scales of direct and indirect features, yellow line
and shade show the mean and standard deviation of the logarithm of “gained”-presynaptic ratio.
The two spatial types of stimulus are charted together: circles (1D) and squares (2D).

97

New temporal, but not spatial, features emerge across the retinothalamic synapse.
Next we quantified the difference between the visual features encoded by retinal versus
thalamic spikes. We used two types of spatio-temporal Gaussian noise patterns as stimuli (Figure
25A), the commonly used 2D “checkerboard” and a 1D “target pattern” which is centered on the
receptive field (see EXPERIMENTAL PROCEDURES). The presynaptic and “gained” features
of the X-cell pair in Figure 24A are illustrated in Figure 25B (the stimulus was the target pattern).
As for all retinothalamic cell pairs, both the presynaptic and the “gained” features were space-time
separable (Jones and Palmer, 1987), for the most part; the separated spatial and temporal
components (lines plotted in Figure 25B) accounted for 89 ± 7% (for the presynaptic) and 72 ±
14% (for the “gained”) of total variance (n = 26).
The change in the representation of the stimulus between retina and thalamus was mainly
temporal. Qualitatively, the presynaptic and the “gained” features resembled each other in space
but not time (Figure 25B). Quantitatively, the spatial features were very similar whereas the
temporal ones were nearly orthogonal, as measured by Pearson’s (Figure 25C). The features
differed not only in shape but also in size, or scale (Figure 25D, E show data for the population,
n=26). The spatial and temporal scales (see EXPERIMENTAL PROCEDURES) of the
presynaptic and “gained” features are depicted as plots of cumulative energy over time,
normalized to fix the presynaptic scale at one. Across the population, the two features had nearly
identical spatial (Figure 25D) but significantly different temporal scales (Figure 25E). Note that
the cumulative energy for the “gained” feature extended farther back in time for the “gained” than
98

for presynaptic features, which suggests that thalamic spikes are more sensitive to the history of
the stimulus than are retinal spikes. In addition, the variance in temporal scale was much higher
for the “gained” than for than the presynaptic feature (Figure 25G). Thus, there appeared to be
greater temporal diversity in thalamic versus retinal features. We found no difference between
features encoded by X and Y relay cells.

Specific stimulus features can be conveyed by thalamic and retinal spikes with more or with less
efficiency.
So far we have described the differences between visual features that are represented on either
side of the retinothalamic synapse. Next we explored how efficiently different features are
encoded by quantifying how much information each retinal and thalamic spike transmitted. By
fitting the data with a linear-nonlinear-Poisson encoding model, we established a lower bound on
the amount of information conveyed by single spikes (see EXPERIMENTAL PROCEDURES).
Then, using the “joint-encoding model”, we measured the total information about the entire
“relevant subspace” that was encoded by single spikes for all RGC-LGN pairs (Figure 26A).
Consistent with a recent study (Sincich et al., 2009), single spikes in the thalamus were more
informative than those in the retina.
99

Figure 26. Change of single-spike code across the retinothalamic synapse. (A) Population data
(n = 26) of single-spike information across the retinothalamic synapse. (B) The RGC-LGN joint
feature space for the example X cell pair in Figure 24E; each direction within the plane represents
a distinct feature. Only the temporal components are illustrated because the spatial component is
nearly unaltered. (C) The informational structure of the joint feature space. The “two-lobed”
polar plots are the single-spike information for particular features and the colored circles mark the
single-spike information for the joint feature space. (D) Marginal distributions of the raw, RGC-
spike-triggered and LGN-spike-triggered stimulus ensembles along the presynaptic ( ), “gained”
( ), postsynaptic ( ) and “lost” features ( ). (E) The presynaptic, “gained”, postsynaptic and
“lost” temporal features. (F) The pre- and postsynaptic single-spike information about the
presynaptic, “gained”, postsynaptic and “lost” features, horizontal lines being the information for
the joint feature space.

100

We then asked how much information retinal and thalamic spikes convey about specific
features in the “joint relevant subspace”. We demonstrate the result of the analysis with an RGC-
LGN X cell pair that we will continue to use as an example in the remaining figures. The two-
dimensional “joint subspace” for the example pair is depicted in Figure 26B (same as in Figure
24E), where each direction represents a distinct stimulus feature as illustrated by eight example
features (Only the temporal components are plotted since the spatial were nearly identical). The
efficiency with which the pre- and the postsynaptic spikes encoded these stimulus features is
displayed in Figure 26C as a polar plot of the amount of information each spike contained.
Although, on average, each thalamic spike was more informative about the whole subspace than
the average retinal spike (thick circles in pale blue and red), the relative information about specific
features (see EXPERIMENTAL PROCEDURES) differed (solid curves in saturated blue and red).
For some features (unshaded sectors in Figure 26C), the postsynaptic spike was more informative,
whereas for other features (shaded sector in Figure 26C), the presynaptic spike was more
informative. Finally, we compared the amount of information encoded by retinal and thalamic
spikes for the four characteristic features; presynaptic, postsynaptic, “gained” and “lost” (Figure
26E). Marginal distributions of the raw, pre- and postsynaptic spike-triggered stimulus
ensembles along these features are plotted in Figure 26D; here differences from the distribution
of the raw stimulus ensemble (black) indicate the content of the information conveyed by spike
rate. The information that the pre- and postsynaptic spikes convey about each of these four
features is quantified in Figure 26F. The average thalamic action potential was more informative
101

about the presynaptic feature than the average retinal spike, suggesting a thalamic “refinement” of
the retinal code. In addition, the thalamic spike encoded information about the “gained” feature
whereas the retinal spike did not, revealing a transformation in feature selectivity across the
retinothalamic synapse.
Even though our analysis suggests that single spikes in the thalamus convey more
information than those in the retina, this improved efficiency did not necessarily apply to each
feature encoded. Rather, some features were encoded more efficiently in the thalamus than in the
retina and others less. We next explore the mechanisms by which the thalamic spikes become
more informative about certain features, such as the “gained” feature.

Thalamic spikes encode information more independently than retinal spikes.
Thus far, we have described how efficiently individual spikes on either side of the
retinothalamic synapse encode visual information. We now move forward to quantify
information that is encoded in temporal patterns of spikes (Brenner et al., 2000). Patterns of
correlated spikes, such as spike pairs, can convey stimulus information that is not encoded by
single spikes. To explore the possibility of a correlation code in the retina or the LGN, we
measured the specific information about the “joint relevant subspace” transmitted by pairs of
spikes with various inter-spike-times. We also measured the corresponding “synergy” (see
EXPERIMENTAL PROCEDURES), a metric that addresses how independently each spike in a
pair encodes information. For this type of analysis, a “spike pair” is defined as two spikes with a
specific temporal offset, regardless of other spikes occurring before, after or in between (Brenner
102

et al., 2000). This temporal offset is named “inter-spike-time” in contrast to the term “inter-
spike-interval”, which is used to name the delay between two adjacent successive spikes.

Figure 27. Change of pair-wise correlation code across the retinothalamic synapse. (A)
Percentage of pair-wise synergy computed over the joint feature space for the X cell pair. (B-D)
Population data (n = 26) of the synergy in RGC and LGN spike trains. Population data are plotted
as mean (thick line) and standard deviation (shade) of the percentage of synergy as a function of
inter-spike-time.

Synergy is zero if the spikes encode information independently, whereas positive and negative
synergy indicates cooperativity or redundancy of the pair code, respectively. The percentage of
synergy as a function of inter-spike-time for the sample cell pair is shown in Figure 27A.
Presynaptic spikes with short inter-spike-times showed significantly positive synergy about the
“joint relevant subspace”, whereas the postsynaptic synergy did not significantly depart from zero.
This suggests that the thalamic spike pairs encode information about the joint feature space
independently whereas the retinal ones are cooperative. The population data (Figure 27B-D)
showed the same trend (n = 26). To compare pre- and postsynaptic synergy directly, we plotted
these quantities against each other for all cell pairs; specifically, the thalamic synergy (ordinate) is
103

plotted against the retinal synergy (abscissa) (Figure 27B). Thus, the data for each RGC-LGN cell
pair is represented by a trajectory in the graph, with each point along the trajectory corresponding
to a certain inter-spike-time (color-coded in Figure 27B). If the synergy values were the same for
the retinal and thalamic spike trains, the trajectories would fall along the diagonal of the plot.
However, the trajectories fell below the diagonal and clustered along the horizontal axis,
suggesting that the postsynaptic synergy is essentially zero (i.e. thalamic spikes are independent)
while the presynaptic varies from positive to zero (i.e. retinal spikes are synergistic). Statistics for
the population are presented in Figure 27C and 27D. The pre- and postsynaptic synergy as
functions of inter-spike-time for the population is plotted (gray lines) with the population mean
and standard deviation illustrated on top (in color) (Figure 27C). The retinal spike trains
exhibited significant positive synergy for short inter-spike-times, whereas the thalamic spike
trains had near-zero synergy for all inter-spike-times. These results suggest that spikes in the
thalamus encode the “joint relevant subspace” more independently than do those in the retina.
104

Figure 28. Transformation of retinal correlation code into thalamic single-spike code may
explain the changes in coding efficiency across the retinothalamic synapse. (A) Population data
(n = 26) of the pair-wise synergy for three specific features: presynaptic, “gained” and irrelevant
(control). (B) Pair-wise information content of the three features for the example X cell pair;
horizontal gray lines mark twice the amount of single-spike information. (C) Marginal
distributions of the raw, RGC-spike-triggered and LGN-spike-triggered stimulus ensembles along
the three features. (D) Marginal distributions of the RGC-spike-pair-triggered stimulus
ensembles along the three features.

The retinothalamic synapse transforms a spike-pair code to a single-spike code.
The previous analysis showed that synergistic encoding of information by correlation
between spikes is more prominent in retinal than in thalamic spike trains. This observation
suggests that there is a channel through which information about the “gained” feature is
transmitted across the retinothalamic synapse. Specifically, we tested the hypothesis that the
105

thalamus is able to extract the information about the “gained” feature from correlations between
retinal spikes, even though such information is not conveyed by single retinal spikes (Figure 26C).
To test this hypothesis, we estimated how much information (see EXPERIMENTAL
PROCEDURES) pair-wise correlations between retinal spikes encoded about specific features
(Figure 28A). For most of the cell pairs, we found significant positive synergy for both the
presynaptic and the “gained” feature. In particular, the synergy for the “gained” feature was
significantly positive at certain inter-spike-times (Figure 28A, middle) whereas the synergy for an
“irrelevant” feature (as a control, the “irrelevant” feature is a randomly generated feature forced to
be orthogonal to the “relevant subspace”) was not significant (Figure 28A, right). The
information content of single and paired retinal spikes is plotted in Figure 28B. Retinal spike
pairs with certain short inter-spike-times convey significant amounts of information about the
“gained” feature. Marginal distributions of the raw, retinal and thalamic spike-triggered stimulus
ensembles are plotted in Figure 28C (similar to Figure 24E and Figure 26D), whereas those for
the retinal spike-pair-triggered ensembles are plotted in Figure 28D. For some inter-spike-times
(from 0 to about 50 ms), the distribution of the retinal spike-pair-triggered ensemble projected on
the “gained” feature differed significantly from the distribution of the raw ensemble projected on
the gained feature, even though the distribution of the retinal (single) spike-triggered ensemble
projected on the gained feature is identical to the distribution of the raw stimulus triggered
ensemble projected onto the gained feature. These results suggest that information about the
106

“gained” feature is encoded in pair-wise correlations of retinal spikes, but not independently by
single retinal spikes.

“Paired spike enhancement” as a mechanism for retinothalamic recoding
What synaptic processes might explain how information available in pair-wise correlations
between retinal spikes could be read out and recoded as single spikes in the thalamus? A survey
of the literature pointed to one candidate in particular – “paired spike enhancement”. This term
describes the experimental observation that short inter-spike-intervals in the presynaptic spike
train increase the “efficacy” of synaptic transmission (Usrey et al., 1998), presumably as the result
of temporal summation of retinogeniculate EPSPs (Carandini et al., 2007). The “efficacy” is
defined as the fraction of presynaptic spikes that are relayed by the postsynaptic neuron.
107

Figure 29. “Paired-spike enhancement” as a model of the retinothalamic transmission. (A)
Conceptual simplification of the retinothalamic transmission as binary “editing” of the retinal
spike train and the corresponding model construct. (B) For the example X cell pair, the “efficacy”
plotted as a function of inter-spike-interval. (C) Rasters of 50 trials of the RGC spike train data
superimposed by the actual and modeled LGN spike trains. (D) Instant firing rates of the RGC
and the actual and modeled LGN cell. A Gaussian window of 5 ms was used to estimate the firing
rates. (E) Reproduction of Figure 24E for the “null” and inter-spike-interval models. (F)
Reproduction of Figure 26C for the two models. (G) Reproduction of Figure 27A, bottom for
the two models.

108

We built mechanistic models to determine if paired spike enhancement could account for
“recoding” at the retinothalamic synapse. To build the models, we assumed that transmission
across the retinothalamic synapse falls into three conceptual categories (Levine and Cleland,
2001): (1) retinal spikes that fail to be relayed, (2) retinal spikes that are successfully relayed and
(3) “anonymous” thalamic spikes without a retinal trigger. We disregarded the third case (see
controls in SUPPLEMENTAL MATERIALS) because there are very few “anonymous” spikes
(Sincich et al., 2007); these are produced during bursts (Wang et al., 2007), which occur only
rarely during vision (Guido and Weyand, 1995; Usrey et al., 1999)). Thus, we were able to model
transmission as a binary selection (i.e. relayed or not) of the retinal spike train (Figure 29A, see
EXPERIMENTAL PROCEDURES for details).

109

Figure 30. Performance of the retinothalamic transmission models. (A) Reproduction of Figure
26A for the “null” and inter-spike-interval models. (B-D) Model-data comparison of the LGN-
single-spike information content for the two models; the information contents are estimated with
regard to the joint feature space (B), the presynaptic feature (C) and the “gained” feature (D). (E)
Reproduction of Figure 27B for the “null” and inter-spike-interval models. (F) Reproduction of
Figure 27D for the two models.

110

111

We used two models to simulate synaptic transmission, the “null” model and the inter-spike-
interval model. In the “null” model, the “relayed” retinal spikes are randomly selected with a
probability matched to the mean “efficacy” in the recording. By contrast, the inter-spike-interval
model used the “efficacy” measured as a function of preceding inter-spike-interval to determine
the probability to relay a spike. For example, the “efficacy” of the second spike of the inter-spike-
interval as a function of the inter-spike-interval is plotted in Figure 29B for the sample cell pair.
As expected, the inter-spike-interval model matched the instantaneous firing rate to the data
slightly better than the “null” model (example in Figure 29C, D). Surprisingly, the inter-spike-
interval model, despite its simplicity, was able to reproduce three of our major observations. By
contrast, the “null” model predicted none of these. The observations are: (1) The thalamic spike
became more selective to the presynaptic feature than the retinal spike and developed selectivity
to the “gained” feature (Figure 29E, F compare to Figure 24E and Figure 26C). (2) Thalamic
spikes were more informative than the retinal impulses (Figure 29F compare to Figure 26C). (3)
Thalamic spikes with short inter-spike-times encoded information more independently than
retinal action potentials (Figure 29G compare to Figure 28A, bottom).
The observation that the inter-spike-interval model alone was able to describe the
experimental findings held for the entire population. Only the inter-spike-interval model was
able to reproduce the increase in the information content of each postsynaptic spike (Figure 30A)
(although, the increase in information the model achieved was smaller than that observed
112

empirically) (Figure 30B). The inter-spike-interval model also predicted the enhanced selectivity
to the presynaptic feature precisely (Figure 30C) and captured the emergent selectivity to the
“gained” feature in large measure (Figure 30D) (The quantitative disparity between modeled and
real results is likely due to the simplicity of the model, see DISCUSSION.) Last, the inter-spike-
interval model correctly predicted that the synergy of thalamic spike pairs is near zero for all
inter-spike-times while the “null” model did not (Figure 30E, F). Thus, our results suggest that
“paired spike enhancement” can explain recoding across the retinothalamic synapse.

CHAPTER 4 DISCUSSION
Here we addressed the basic question of how the representation of a stimulus changes across
a synapse by recording from connected retinal and thalamic neurons during vision. First we used
a novel computational approach that allowed us to identify visual features that thalamic spikes
detected but that presynaptic retinal spikes did not. We then asked how the thalamus gained
selectivity to these emergent features by assessing information encoded by temporal correlations
of retinal spikes rather than single action potentials. This analysis revealed that the “gained”, or
emergent features were conveyed by pair-wise correlations within retinal spike trains. Finally, by
building computational models of synaptic transmission, we demonstrated that the mechanism of
paired spike facilitation (or enhancement) could explain the transformation from a pair-wise to a
single spike code. Our work provides a first example of the transformation of a correlation code
113

into an independent code across a synapse through which sensory information flows from the
periphery to the brain.

Resolving visual features encoded by synaptically connected neurons
Classical linear-nonlinear models provide a means to identify the particular features of the
stimulus that a single neuron encodes (Carandini et al., 2005). Two linear-nonlinear models can
be used to describe separately the features encoded by two different neurons (Sincich et al., 2009).
However, if one wishes to understand how information about various features propagates from
one cell to the next – which features are inherited, which are gained and which are lost – it
becomes necessary to have a single model that encompasses the responses of both neurons. Thus,
we built a linear-nonlinear “joint encoding model” that subsumes the features (relevant subspaces)
that the pre- and postsynaptic spikes encode (see EXPERIMENTAL PROCEDURES). Using this
framework, we showed that thalamic neurons inherited selectivity for the two dimensional spatial
features encoded by their retinal inputs, but gained sensitivity to different temporal features (this
finding is consistent with recent work that compared temporal features in retina and thalamus,
(Sincich et al., 2009)). We next explore the broader theoretical implications of this result before
discussing specific roles in visual processing.

Comparing the efficiency with which spikes encode information in the retina and thalamus
We compared how efficiently thalamic and retinal spikes encoded visual information in
general as well as with respect to specific features. For the general case, we used the standard
114

approach of measuring the mutual information between single or paired spikes and the stimulus
(Brenner et al., 2000). The results of this analysis showed that thalamic spikes encode visual
information both more efficiently and more independently than do their retinal inputs. That is,
each thalamic action potential transmitted more information about the stimulus than each retinal
impulse, consistent with previous work (Sincich et al., 2009). Further, each thalamic spike in a
pair encoded information independently from the other whereas a pair of retinal spikes encoded
more information than did the two separate spikes.
Moreover, our “joint encoding model” allowed us to use information theory to explore the
differences in the efficiency of the spike code with respect to individual features across the
synapse. The difference in efficiency was most dramatic for the “gained” features, those that
emerged in the geniculate. Thalamic neurons transmitted a substantial amount of information
about these features whereas their retinal inputs conveyed almost none. Otherwise, remaining
features were encoded with either lesser or greater efficiency in thalamus than retina.

From a pair-wise correlation code to a rate code
How do retinal firing patterns transmit information about the emergent (“gained”) features
that single thalamic spikes encode? Since retinal spikes encoded the relevant features
synergistically, we hypothesized that information about the “gained” feature might be embedded
in the structure of presynaptic firing patterns. Our analyses provided support for this hypothesis;
information about the features gained in the thalamus was encoded by pair-wise correlations
115

between retinal impulses. We will address possible mechanisms for this transformation in a later
section.
In addition, the results of our analysis address the theoretical limit that the “data processing
inequality” (Cover and Thomas, 1991) imposes on the amount of information that can be
transmitted across a synapse. It had been suggested that this limit is set by the total amount of
information the presynaptic neuron transmits using single spikes (Sincich et al., 2009). However,
if information in correlations within the presynaptic spike train were recoded postsynaptically, as
independent spikes, then the rate of information transmitted by those postsynaptic impulses
could exceed that coded solely by independent presynaptic spikes without violating the “data
processing inequality”.

Experimental caveats
There are three main techniques used to record simultaneously from synaptically coupled
neurons in retina and thalamus, each with advantages and disadvantages. One approach involves
placing separate extracellular electrodes in retina (or optic nerve) and thalamus and using cross-
correlation analysis to identify coupled pairs (e.g. (Usrey et al., 1998)). We adopted the two
remaining methods, each of which uses a single electrode to record both retinal inputs and
thalamic spikes. One, an extracellular technique, takes advantage of the fact that retinogeniculate
EPSPs are often so large that they can be sensed near the neural surface as waveforms called “S-
potentials” (Bishop et al., 1958, 1962; Hubel and Wiesel, 1961; Kaplan and Shapley, 1984; Sincich
et al., 2007; Sincich et al., 2009). The other method, intracellular recording (Wang et al., 2007),
116

offers the chance to visualize additional inputs that might be too small to generate “S-potentials”.
The single electrode techniques offer some advantages over the dual electrode approach (Sincich
et al., 2007). For example, they allow one to focus on cells with a clear-cut dominant input for
analysis (see SUPPLEMENTAL MATERIALS for details about sorting the train of dominant
inputs from the signal) and do not rely on statistical inference to determine connectivity.
However, single electrode recordings also share a disadvantage. Since both spikes and EPSPs are
mixed in one signal, the former have the potential to occlude the latter. Hence, some EPSPs can
be hidden and the spikes these evoked appear, falsely, to arise from an anonymous source. To
assess concerns how the “masked” EPSPs could influence our result, we compared results using
the retinal train we detected to one in which an EPSP was added to match each anonymous spike
(see SUPPLEMENTAL MATERIALS). Both sets of results were the same.

Mechanistic models of retinothalamic processing
We used mechanistic models to explore physiological processes that might convert
presynaptic correlation codes to postsynaptic single spike codes. One simple model that
simulated “paired spike enhancement” (Usrey et al., 1998) (as the interval between two successive
inputs grows shorter, the probability that the later input will evoke a postsynaptic spike increases)
reproduced our main results. Specifically, the model captured the conversion of a correlation
code in retina to an independent code in thalamus for the “gained” feature, preserved the features
that thalamus inherited from the retina and accounted for the overall increase in the efficiency of
thalamic spikes. Thus, a single, well documented mechanism can account for our main results.
117

Still, the amount of information about the “gained” feature the model predicted was less than
the actual amount. This gap was not corrected by considering higher order statistics of the retinal
spike train, for example by taking not just one, but two, previous inter-spike-intervals into
account. Nonetheless, it seems likely that the performance of the model would be improved by
including additional biological components. Possible additions include cortical feedback,
undetected retinal inputs, or contributions from thalamic inhibitory neurons. The first two
options seem the least plausible of the three. Current evidence does not support a strong role for
cortical feedback; ablating visual cortex does not seem to influence size of thalamic receptive
fields (Cudeiro and Sillito, 1996), though it does influence inter-spike-interval distribution
(Wörgötter et al., 1998). As well, since the gained feature is temporal not spatial, undetected
input from ganglion cells probably cannot explain the difference between the simulated and
actual results. The reasoning for this assumption is as follows. If the “gained” feature did not
result from retinothalamic processing per se, but reflected input from an undetected ganglion cell
different from the dominant input, then we would also expect to find a spatial feature to emerge
in thalamus. This is because neighboring ganglion cells have receptive field that are displaced
from each other (Peichl and Wässle, 1983; Wässle et al., 1981) (A case in which we identified
spike trains from two neighboring ganglion cells supports this assumption, see SUPPLEMENTAL
MATERIALS). By contrast, previous theoretical studies support a role for intrathalamic
inhibition. These suggest that intrathalamic inhibition influences the temporal precision of
thalamic responses (Butts et al., 2007) and contribute to mechanisms that encode visual
information (Babadi et al., 2007).
118

Functional roles for the temporal features that emerge in the thalamus
Numerous studies have shown that time courses of thalamic responses are varied and often
reach peak amplitude long after retinal responses have begun to decline. This widespread
distribution of spike latency with respect to the stimulus is thought to accomplish different
functional tasks. For example, a leading model of direction selectivity in cortex depends on
convergent input from relay cells with staggered timings (Wolfe and Palmer, 1998). Our results
suggest that the recoding of visual features by means of “paired spike enhancement” at the
retinogeniculate synapse increases temporal diversity in thalamus.

Recoding of sensory representations
The representation of sensory information in neural activities is continuously transformed by
the neural circuits. To our knowledge, our work provides a first example in which information
about a sensory feature is encoded by temporal correlations in the spike train of a presynaptic
neuron and then recoded by the postsynaptic neuron, with enhanced efficiency, as independent
spikes. Such a transition from a correlation to a rate code at the retinothalamic synapse might be
iterated downstream in the cortex, continuing to increase the efficiency and independence of the
neural code. There is evidence that similar schemes of recoding exists in other sensory systems as
well. For example, studies of the somatosensory system (Ahissar and Arieli, 2001; Arabzadeh et
al., 2006) suggest that temporal patterns of firing in the periphery might be transformed into a
rate code at later stages of processing. Our new approach of using a combined pre- and
119

postsynaptic encoding model to study the transformation of neural spike trains should be useful
to explore how synapses recode information in diverse regions of the brain.

CHAPTER 4 EXPERIMENTAL PROCEDURES
Recording
In vivo patch recordings, in cell-attached or whole-cell mode, were made from adult cats
anesthetized with propofol and sufenta (Hirsch et al., 1998; Martinez et al., 2005; Wang et al.,
2007) and were digitized at 10 or 25 kHz.

Stimulation
Visual stimuli were displayed on a cathode-ray tube monitor with a monochrome phosphor
(x’ = 0.42, y’ = 0.53, CIE 1931) placed at 915 mm from the eyes; the luminance ranged from 0 to
110 cd/m
2
and the video refresh rate was 144 Hz. The stimuli were Gaussian white noise with
mean and standard deviation values of 55 and 18.33 cd/m
2
; luminance values outside the dynamic
range of the monitor were truncated. The stimulus update rate was 72 or 48 Hz and typical
sequences lasted approximately 10 minutes. For some cases, 50 or 60 repetitions of a different
stimulus sequence (10 to 20 seconds in duration) that had the same statistics as original sequence
was used for subsequent cross-validation of models. Spatial arrangement of the stimulus was
either a checker-board (2D) or a target-pattern (1D) of concentric rings centered on the receptive
120

field (Figure 25A). Alternating sinusoidal gratings (Hochstein and Shapley, 1976) were used to
classify neurons as X or Y cells.

Joint relevant subspace
We used spike-triggered average and covariance (STA/STC) analysis (Schwartz et al., 2006) as
well as the information-theoretic STA and STC (iSTAC) analysis (Pillow and Simoncelli, 2006) to
identify the “relevant subspace” of spike responses (details in SUPPLEMENTAL MATERIALS).
For the RGC-LGN data we acquired, the relevant subspace for both sides of the synapse was
essentially one-dimensional and STA could be used to recover the relevant feature space
efficiently. A temporal window of 250 ms was used for stimulus history in all cases and
appropriate spatial windows were chosen for each case.
To extend the relevant subspace of a single neuron to the “joint relevant subspace” of a pair of
synaptically connected neurons, we applied the analysis to the data sequentially. First, we
identified the relevant subspace of the presynaptic neuron (RGC) and then extended that
subspace by adding relevant filters for the postsynaptic neuron that were orthogonalized to the
presynaptic subspace. Alternatively, the joint relevant subspace could be obtained by extending
the relevant subspace of the postsynaptic neuron with orthogonalized relevant filters of the
presynaptic neuron.

121

Spatio-temporal feature analysis
The spatial and temporal components of the recovered relevant filters were obtained by least-
squared-difference factorization. The spatial and temporal scales of visual features were
computed as the average radii weighted by filter power. Specifically, the temporal scale is

where is the temporal filter. And likewise, the 1D spatial scale is

where is the spatial filter. For the 2D checkerboard stimulus, the spatial scale is

where

Information-theoretic analysis
Single-spike information with respect to a certain subspace was estimated using the
following formula.
spike; |spike log
|spike
where · is the stimulus projection on subspace . The prior was Gaussian
distributed because the stimulus was Gaussian white noise. In practice, we approximated the
122

posterior |spike as a Gaussian with the mean as the STA and the covariance as the STC. (We
illustrate both the non-parametric estimations of distributions (histograms) and the Gaussian
approximations throughout the figures.) Under such approximations, the formula is simplified to
spike; 1
2ln2
tr
ln
| |
| |

T

where ,
and , are the mean and covariance of the prior and posterior, and the
dimensionality of the relevant subspace . Given that the prior is strictly Gaussian, we have

; equality holds when the posterior is also strictly Gaussian. For our RGC-LGN recordings,
the Gaussian posterior approximation fit the data well. The accuracy of the estimation of
information was limited by the amount of data available (i.e. how densely the distributions were
sampled). To address this concern, we assessed the accuracy of our estimations by progressively
reducing the amount of data used and also gave a standard deviation of the estimation based on
the inverse-square-root law of the deviation as a function of the amount of data (Brenner et al.,
2000).
We estimated the spike-pair information content by regarding pairs of spikes as compound
events. The event “pair at ” is equivalent to the coincidence of a “spike at ” and another
“spike at ”. The spike-pair information with respect to a certain subspace was estimated
as
pair ; ,
|spike ,
log
,
|spike ,
,

123

where · and · . Gaussian approximations were also used to
estimate efficiently the spike-pair information. We then used the concept of “synergy” (Brenner
et al., 2000) to address the independence in pair-wise spike encoding of the feature space .
Syn

pair ; 2 spike;
Syn
% Syn
2
spike; 100
Analyses of spike pairs were done at the temporal resolution in accordance with the stimulus
update rate.

Model of retinothalamic transmission
With the assumptions that each thalamic spike is causally associated with a single retinal spike
and that the latency (less than 0.5 ms as measured empirically) between the two spikes is
negligible, one can simplify the retinothalamic transmission of spike trains as a binary parsing
process (i.e. determining whether each retinal spike is “relayed” or not) of the retinal spike train.
Control studies that we performed suggested that both of these assumptions are reasonable (see
SUPPLEMENTAL MATERIALS). Thus, we built models of retinothalamic transmission as
“binary parsing” processes (Figure 29A), as follows. Let the retinal spike train be denoted ,
and the binary “relayed” state . The resulting simplified model of retinothalamic spike
transmission is then a map of to . With the further assumption that does not depend
on its own history ,
, , the output of the model is then an instance of a stochastic
variable with the binary-transfer probability distribution
124

| ,
,
which forms the deterministic component of the model. The “relayed” state of a retinal spike,
however, is dependent on its own history ,
, . (Note that the “relayed” probability
1 | ,
, is closely related to cross-correlation analysis because its average over the
sub-index is also known as the “efficacy” (Usrey et al., 1998)). Assuming time-invariance, the
transfer probability distribution can be written as conditioned on inter-spike-intervals
|
,

, .
We constructed two versions of the model. The first, the “null” model, contained a transfer-
probability distribution without dependence on spike history: 1 LGN
RGC
and 0 1 LGN
RGC
( RGC
and LGN
being the average firing rates of the pre- and postsynaptic neurons);
this essentially “down-scales” retinal firing rate to match the overall “efficacy”. The second, the
inter-spike-interval model, used a transfer-probability distribution conditioned only on the inter-
spike-interval that immediately precedes the retinal spike, |
, which was empirically
estimated as the “efficacy” dependent on the preceding inter-spike-interval (Figure 29B). We
also evaluated second-order inter-spike-interval models |
,

for a control
study (results not shown). Both models were cross-validated.

125

CHAPTER 4 SUPPLEMENTAL MATERIALS
Detection of spike trains from single retinal ganglion cells

Figure 31. Identification of retinal spike trains from single ganglion cells. (A) Whole-cell
recording from a LGN relay neuron where inputs from two distinct retinal ganglion cells are
resolved. (B) Sorting of the two retinal and thalamic spikes. (C) Spatial receptive fields of the two
retinal inputs and the thalamic output. (D) Inter-spike-interval distribution of the two retinal
spike trains and the combined train. Red arrow marks the absolute refractory period (about 2 ms)
and the black marks the time window (less than 0.5 ms) in which the overlap of the waveforms
severely undermines the sorting of the two events. (E) Population data (n = 26) of the inter-
spike-interval distributions of the detected retinal spike trains used in this study.

In this study, we used S-potential or whole-cell recordings from postsynaptic thalamic
neurons to extract the pre- and postsynaptic spike trains (Wang et al., 2007). Even though the
output of most thalamic relay cell is dominated by a single retinal input, the typical convergence
ration from retina to thalamus is 6-to-1 (Usrey et al., 1999). Since this paper focuses on
information encoded by temporal correlations within spike trains of single RGCs, it was necessary
to accurately isolate the spike trains of the dominant RGC spike trains from our records.
126

Otherwise, information that we attributed to temporal correlations might include a component
from correlations among convergent RGCs.
Thus we took several measures to ensure that the retinal spike trains we used were from single
RGCs. First, we only used data with S-potentials or postsynaptic events that formed a distinct
cluster from noise (example in Figure 24B). Further, we generated histograms of inter-spike-
intervals (ISIs) to confirm the presence of an absolute refractory period, as indicated by (nearly)
empty bins below about 2 ms (population data in Figure 31E). In addition, we considered the
possibility that the “masking effect” might create artifactual refractoriness (when two events occur
very close to each other in time, their waveforms overlap and merge so that the sorting algorithm
detects only one or even no events). We excluded this possibility by detecting spikes from two
converging ganglion cells with displaced receptive fields (Figure 31A, C). The postsynaptic
events from each cell formed distinct clusters (Figure 31B). Histograms of ISIs made from the
events in each cluster revealed absolute refractoriness below 2 ms (Figure 31D, blue and green)
whereas the plot made from the combined retinal spike trains did not (Figure 31D, gray). Rather,
the “masking effect” created an empty bin at less than 0.5 ms, an interval far shorter than the
biological refractory period.

127

Figure 32. Sequential identification of the joint relevant subspace using the information-
theoretic spike-triggered average and covariance (iSTAC) analysis. (A) The presynaptic and
“gained” spatio-temporal features identified using the iSTAC analysis. (B) Information content
estimated during the two-step sequential application of iSTAC. The first step (blue solid)
identified a one-dimensional presynaptic subspace and the second (red dashed) a one-
dimensional “gained” subspace. (C) The spatial and temporal filters identified using a two-step
space-time-constrained iSTAC analysis.

Sequential spike-triggered analysis of the retinal and thalamic spike train
In the main text of this paper, we have explained the idea of the “joint relevant subspace” as a
framework to compare the pre- and postsynaptic encoding models with linear frontends. Here
we describe how we identified the “joint relevant subspace” by means of sequential application of
a system identification algorithm that maximizes mutual information. We used information-
theoretic spike-triggered average and covariance (iSTAC) analysis (Pillow and Simoncelli, 2006)
of responses to Gaussian white noise. For an example “joint relevant subspace”, Figure 25A, we
first identified features about which the retinal spikes are significantly informative (Figure 32B,
solid blue). In this case, we identified only one significant retinal (presynaptic) feature. We then
128

fixed this feature and identified an additional “gained” feature about which the thalamic spikes
were significantly informative (Figure 32B, dashed red). The analysis identified no further
features about which the retinal or thalamic spikes were informative. Therefore, the “joint
relevant subspace” for this case is essentially two-dimensional. Likewise, we applied the “space-
time constrained” iSTAC analysis (Pillow and Simoncelli, 2006) to the retinal and thalamic spike
trains (Figure 32C). Consistent with our factorization analyses of the spatio-temporal structure
of the “joint relevant subspace” (Figure 25), there was only one significant spatial filter but two
significant temporal filters. Last, we performed conventional STA and STC analyses. The
features identified using STA were almost identical to those identified using iSTAC. Hence,
simple spike-triggered average can be used to efficiently recover the two-dimensional “joint
relevant subspace” with the retinothalamic data in the cat.

“Missing” retinal spikes – a control study
Pre- and postsynaptic events are combined in single channel when recorded using the S-
potential or intracellular recording techniques. Thus, one faces the problem of sorting events
from a mixed signal in which retinal inputs might be masked by postsynaptic spikes. Four
examples of results obtained with our event sorting algorithm for one X cell pair are illustrated in
Figure 33A. The first example shows an individual retinal event. The second and third cases
depict instances in which the algorithm found retinal events that preceded thalamic spikes with
varied but short (less than 1 ms) latencies. When, as for the fourth example, the algorithm failed
129

to identify a retinal event at the foot a spike, it was possible that the retinal input was masked by a
very rapidly evoked thalamic action potential. Consistent with this possibility, we found that
typical synaptic latencies (as measured by cross-correlation analysis (Figure 33B) on the “naively”
detected spike trains) are very short, less than 1 ms.
An earlier study in macaque provided a solution for this masking problem based on the
assumptions that the spike waveforms can be replaced as a template and that the event waveforms
sum linearly (Sincich et al., 2007). The conclusion of that analysis was that almost all thalamic
spikes are preceded by a retinal action potential. Earlier work that we have done reached a similar
conclusion, with the exception of spikes generated during thalamic bursts (Wang et al., 2007).
Nevertheless, since bursts are rare, the assumption that each thalamic spike is causally related to a
retinal action potential is reasonable in most cases.
Thus, it was necessary to determine if the masking effect introduced artifacts. Therefore, we
generated retinal spike trains to which potentially “missing” retinal spikes were restored, as
follows. We used two empirically identified constants (Figure 33B): the refractory period

and the “masking” latency
. The former, 2 ms, was the biological absolute refractory period
(Figure 33B, the dip before the correlation peak). The latter, 0.25 ms, was the synaptic latency
below which the algorithm failed to detect retinal spikes. We restored the retinal spike train by
replacing the “missing” retinal spikes per the following rule: a retinal event was added at

for each thalamic spike if there was no retinal event within the interval

,
.
130

Figure 33. “Missing” retinal spikes – a control study. (A) Four incidences of possible occurrence
of retinal inputs during a “cell-attached” recording. The first case demonstrates a single retinal
input and the second and third show retinal inputs detected right before thalamic spikes, while in
the last case the event sorting algorithm did not report a retinal input. (B) Conventional cross-
correlation analysis of the retinal to the thalamic spike train. Vertical dashed lines mark two
important empirical time values based on which “missing” retinal inputs were recovered for the
models in Figures 29, 30. (C) A reproduction of Figure 29B using the “naively” detected spike
trains. (D-I) Reproductions of Figures 24E, 26C, 27A, 28B, 28A (top) and 28C using the
modified retinal spike trains in which the “missing” spikes were recovered.

131

We used the restored retinal spike train for two purposes. First, we used it to build a
retinothalamic transmission model based on inter-spike-interval (ISI). The “efficacy” as a
function of inter-spike-interval measured using the restored (Figure 29B) versus “naive” (Figure
33C) retinal spike trains were highly similar.
Second, we substituted the restored for the “naive” retinal spike trains in all relevant analyses.
All the effects we described in the main text held for analyses using the restored spike trains. (1)
the “joint relevant subspace” is two-dimensional and the emergent thalamic selectivity to the
“gained” feature is retained (Figure 33D); (2) single thalamic spikes are more informative than
single retinal spikes, although the information about a specific feature might increase or decrease
across the synapse (Figure 33E, F); (3) retinal spikes encode the “joint relevant subspace” with
significant positive synergy at short inter-spike-times (Figure 33G-I).

Pair-wise synergy in the “joint retinothalamic feature space”
We have shown that, information about the “gained” feature is implicitly encoded by retinal
spike correlations and is then recoded, by means of temporal summation, into an explicit single
spike code. It might seem that all information is recoded into single spikes in the LGN, leaving no
room for thalamic spike pairs to serve as potential coding elements (Figure 27D). However, this
is not the case, as we explain with the following argument. Even though a retinal spike is less
informative than a thalamic spike about the whole feature space, it can be more informative about
certain specific features (Figure 26C). Likewise, the synergy may not be zero for certain features
132

even though it is zero about the whole relevant subspace. Figure 34 shows the pre- and
postsynaptic synergy of the example X cell pair in Figure 27A. It is readily apparent that thalamic
spike pairs with various inter-spike-times can encode additional information about certain
features. In particular, the selectivity to the “lost” feature can be recovered if thalamic spike pairs
are regarded as elements of coding. Therefore, if downstream (cortical) neurons also shift
selectivity in the feature space, thalamic pair-wise correlations may indeed be exploited despite a
zero synergy.

Figure 34. Pair-wise synergy in the joint retinothalamic feature space. (A, B) Retinal and
thalamic pair-wise synergy within the joint feature space. Left column: twice single-spike
information in gray and spike-pair information as colored curves; right column: positive synergy
lobes in bright color and negative in dark.

133

For the reason stated above, the overall mutual information and synergy of a spike train
might not be helpful to understand the recoding across a synapse. This is because the change of
information and the synergy about certain feature subspace can greatly vary if there is a shift of
the relevant feature space across the synapse.
134

CHAPTER 5
CONCLUSION
Behaviorally relevant information is encoded in the brain by neural spike trains. In all
sensory systems, neural circuits continuously transform the neural code from one stage to the
next. In the mammalian visual system, the neural code of the retina is relayed by the lateral
geniculate nucleus of the thalamus, where the sensory information is recoded before reaching the
visual cortex. I present three studies in this dissertation which explore how the local neural
circuits of the LGN carry out the computation necessary to make visual signals ready for cortical
processing.
The characterization of a neural circuit should build upon the understanding of the behavior
of its components. In the thalamus, however, knowledge is scant about how local inhibitory
neurons process sensory information. Using the technique of in vivo patch recording, I was able
to compare the inputs (synaptic potentials) and outputs (spikes) of relay cells (excitatory) and
local inhibitory neurons in the visual thalamus. Retinal axons supply the major source of input to
both types of neurons; thus, it was not surprising to find that both have receptive fields derived
from retinal afferents. However, my work showed that these responses are built by remarkably
distinct patterns of synaptic input. Relay cells receive fast, strong, unitary excitatory inputs
specialized in preserving fine temporal precision. By contrast, inhibitory inputs are made of slow
and graded signals that, as I showed with an information-theoretic analysis, optimize the
transmission of information.
135

I then asked how the excitatory and inhibitory components of the thalamic network interact
to shape the output of thalamic relay cells sent to cortex. Thalamic relay cells operate in two
modes, firing tonic trains of single spikes or bursts. Bursts had been believed to correspond to
behavioral state of the animal rather than play a role in sensory processing. However, recent work
has challenged this view and suggested that bursts might be used to transmit information about
selected features of the stimulus. By combining experimental (whole-cell recording in vivo) and
theoretical analyses, I demonstrated that thalamic bursts reliably occur during naturalistic
viewing. Further, I showed that local inhibition is necessary and sufficient to account for the
occurrences of visually-evoked bursts. This work reveals that thalamic inhibitory circuits play an
important role in determining the pattern of activity transmitted to cortex during vision.
To understand further the principles of retinothalamic computation, I explored how the
neural representation (or code) is transformed from the retina to the cortex. Specifically, I asked
whether and how information conveyed by temporal spike correlations in the periphery can be
used by downstream neurons in the brain. I developed a novel theoretical framework to
characterize the transformation of sensory features across a neural synapse. That is, I identified
the visual features to which sensitivity is inherited, gained or lost across the retinothalamic
synapse. With these features, I was able to compare the efficiency of the neural code in thalamus
versus retina. My results suggested that the brain actively improves coding efficiency by
transforming correlated responses into independent ones during the flow of information across
the retinothalamic synapse. Finally, I built a biophysically realistic model of the retinothalamic
synapse that reproduced my experimental observations. This work not only gives a first example
136

of how a synapse recodes information conveyed by temporal spike correlations into independent
spikes, but also establishes the link between synaptic biophysics and the recoding of sensory
information.
Altogether, these three studies put up a general picture of the computation performed by the
feedforward components of the thalamic neural circuits. Local inhibitory neurons of the
thalamus receive retinal inputs as do the excitatory relay cells, but they integrate synaptic inputs
in a distinct way that optimizes information transmission. A direct consequence of such
inhibition is the control of firing modes of the thalamic output, which generates rich temporal
activity patterns that encode information about the visual scene. While local inhibition
introduces distinct firing patterns, the excitatory retinothalamic synapse transforms temporal
patterns (i.e. correlational structures of the spike train) of retinal input into independent spikes in
thalamic output. Combining these neural mechanisms, the thalamus actively increases the
efficiency of the neural code from the retina to the cortex.
137

REFERENCES
Acuna-Goycolea, C., Brenowitz, S.D., and Regehr, W.G. (2008). Active dendritic conductances
dynamically regulate GABA release from thalamic interneurons. Neuron 57, 420-431.
Aguera y Arcas, B., and Fairhall, A.L. (2003). What causes a neuron to spike? Neural
Computation 15, 1789-1807.
Ahissar, E., and Arieli, A. (2001). Figuring space by time. Neuron 32, 185-201.
Ahlsen, G., Lindstrom, S., and Lo, F.S. (1985). Interaction between inhibitory pathways to
principal cells in the lateral geniculate nucleus of the cat. Exp. Brain Res. 58, 134-143.
Alitto, H.J., Weyand, T.G., and Usrey, W.M. (2005). Distinct properties of stimulus-evoked bursts
in the lateral geniculate nucleus. J. Neurosci. 25, 514-523.
Arabzadeh, E., Panzeri, S., and Diamond, M.E. (2006). Deciphering the spike train of a sensory
neuron: counts and temporal patterns in the rat whisker pathway. J. Neurosci. 26, 9216-9226.
Babadi, B., Casti, A., Xiao, J., and Paninski, L. (2007). Visual response in LGN neurons beyond the
monosynaptic retino-geniculate transmission. Annual Meeting Society for Neuroscience 2007.
Barlow, H., B (1961). The coding of sensory messages. In Current Problems in Animal Behaviour
W. Thorpe, H, and O. Zangwill, L, eds. (Cambridge: Cambridge University Press), pp. 331-
360
Bialek, W., and de Ruyter van Steveninck, R. (2005). Features and dimensions: Motion estimation
in fly vision. arXiv:q-bio/0505003v1.
Bialek, W., Rieke, F., de Ruyter van Steveninck, R.R., and Warland, D. (1991). Reading a neural
code. Science 252, 1854-1857.
Bickford, M.E., Wei, H., Eisenback, M.A., Chomsung, R.D., Slusarczyk, A.S., and Dankowsi, A.B.
(2008). Synaptic organization of thalamocortical axon collaterals in the perigeniculate nucleus
and dorsal lateral geniculate nucleus. J. Comp. Neurol. 508, 264-285.
Bishop, P.O., Burke, W., and Davis, R. (1958). Synapse discharge by single fibre in mammalian
visual system. Nature 182, 728-730.
Bishop, P.O., Burke, W., and Davis, R. (1962). The interpretation of the extracellular response of
single lateral geniculate cells. J. Physiol. 162, 451-472.
138

Blitz, D.M., and Regehr, W.G. (2003). Retinogeniculate synaptic properties controlling spike
number and timing in relay neurons. J. Neurophysiol. 90, 2438-2450.
Blitz, D.M., and Regehr, W.G. (2005). Timing and specificity of feed-forward inhibition within
the LGN. Neuron 45, 917-928.
Bloomfield, S.A., Hamos, J.E., and Sherman, S.M. (1987). Passive cable properties and
morphological correlates of neurones in the lateral geniculate nucleus of the cat. J. Physiol.
383, 653-692.
Bloomfield, S.A., and Sherman, S.M. (1988). Postsynaptic potentials recorded in neurons of the
cat's lateral geniculate nucleus following electrical stimulation of the optic chiasm. J.
Neurophysiol. 60, 1924-1945.
Bloomfield, S.A., and Sherman, S.M. (1989). Dendritic current flow in relay cells and
interneurons of the cat's lateral geniculate nucleus. Proc. Natl. Acad. Sci. 86, 3911-3914.
Bonin, V., Mante, V., and Carandini, M. (2006). The statistical computation underlying contrast
gain control. J. Neurosci. 26, 6346-6353.
Borst, A., and Theunissen, F.E. (1999). Information theory and neural coding. Nat. Neurosci. 2,
947-957.
Brenner, N., Strong, S.P., Koberle, R., Bialek, W., and de Ruyter van Steveninck, R.R. (2000).
Synergy in a neural code. Neural Computation 12, 1531-1552.
Bullier, J., and Norton, T.T. (1979). Comparison of receptive-field properties of X and Y ganglion
cells with X and Y lateral geniculate cells in the cat. J. Neurophysiol. 42, 274-291.
Butts, D.A., Weng, C., Jin, J., Yeh, C.I., Lesica, N.A., Alonso, J.M., and Stanley, G.B. (2007).
Temporal precision in the neural code and the timescales of natural vision. Nature 449, 92-95.
Cai, D., DeAngelis, G.C., and Freeman, R.D. (1997). Spatiotemporal receptive field organization
in the lateral geniculate nucleus of cats and kittens. J. Neurophysiol. 78, 1045-1061.
Carandini, M., Demb, J.B., Mante, V., Tolhurst, D.J., Dan, Y., Olshausen, B.A., Gallant, J.L., and
Rust, N.C. (2005). Do we know what the early visual system does? J. Neurosci. 25, 10577-
10597.
Carandini, M., Horton, J.C., and Sincich, L.C. (2007). Thalamic filtering of retinal spike trains by
postsynaptic summation. Journal of Vision 7, 20 21-11.
139

Chen, C., and Regehr, W.G. (2000). Developmental remodeling of the retinogeniculate synapse.
Neuron 28, 955-966.
Chichilnisky, E.J. (2001). A simple white noise analysis of neuronal light responses. Network 12,
199-213.
Cleland, B.G., Dubin, M.W., and Levick, W.R. (1971). Simultaneous recording of input and
output of lateral geniculate neurones. Nature - New Biology 231, 191-192.
Cleland, B.G., and Lee, B.B. (1985). A comparison of visual responses of cat lateral geniculate
nucleus neurones with those of ganglion cells afferent to them. J. Physiol. 369, 249-268.
Contreras, D., Curro Dossi, R., and Steriade, M. (1993). Electrophysiological properties of cat
reticular thalamic neurones in vivo. J. Physiol. 470, 273-294.
Coomes, D.L., Bickford, M.E., and Schofield, B.R. (2002). GABAergic circuitry in the dorsal
division of the cat medial geniculate nucleus. J. Comp. Neurol. 453, 45-56.
Cover, T.M., and Thomas, J.A. (1991). Elements of Information Theory (Wiley-Interscience).
Cox, C.L., Reichova, I., and Sherman, S.M. (2003). Functional synaptic contacts by intranuclear
axon collaterals of thalamic relay neurons. J. Neurosci. 23, 7642-7646.
Cox, C.L., and Sherman, S.M. (2000). Control of dendritic outputs of inhibitory interneurons in
the lateral geniculate nucleus. Neuron 27, 597-610.
Cox, C.L., Zhou, Q., and Sherman, S.M. (1998). Glutamate locally activates dendritic outputs of
thalamic interneurons. Nature 394, 478-482.
Crunelli, V., Haby, M., Jassik-Gerschenfeld, D., Leresche, N., and Pirchio, M. (1988). Cl- - and
K+-dependent inhibitory postsynaptic potentials evoked by interneurones of the rat lateral
geniculate nucleus. J. Physiol. 399, 153-176.
Cucchiaro, J.B., Uhlrich, D.J., and Sherman, S.M. (1991). Electron-microscopic analysis of
synaptic input from the perigeniculate nucleus to the A-laminae of the lateral geniculate
nucleus in cats. J. Comp. Neurol. 310, 316-336.
Cudeiro, J., and Sillito, A.M. (1996). Spatial frequency tuning of orientation-discontinuity-
sensitive corticofugal feedback to the cat lateral geniculate nucleus. J. Physiol. 490, 481-492.
Dan, Y., Atick, J.J., and Reid, R.C. (1996). Efficient coding of natural scenes in the lateral
geniculate nucleus: experimental test of a computational theory. J. Neurosci. 16, 3351-3362.
140

Datskovskaia, A., Carden, W.B., and Bickford, M.E. (2001). Y retinal terminals contact
interneurons in the cat dorsal lateral geniculate nucleus. J. Comp. Neurol. 430, 85-100.
David, S.V., Hayden, B.Y., and Gallant, J.L. (2006). Spectral receptive field properties explain
shape selectivity in area V4. J. Neurophysiol. 96, 3492-3505.
David, S.V., Vinje, W.E., and Gallant, J.L. (2004). Natural stimulus statistics alter the receptive
field structure of V1 neurons. J. Neurosci. 24, 6991-7006.
Deleuze, C., and Huguenard, J.R. (2006). Distinct electrical and chemical connectivity maps in the
thalamic reticular nucleus: Potential roles in synchronization and sensation. J. Neurosci. 26,
8633-8645.
Denning, K.S., and Reinagel, P. (2005). Visual control of burst priming in the anesthetized lateral
geniculate nucleus. J. Neurosci. 25, 3531-3538.
Dong, D.W., and Atick, J.J. (1995). Statistics of natural time-varying images. Network: Comput.
Neural Syst. 6, 345-358
Dubin, M.W., and Cleland, B.G. (1977). Organization of visual inputs to interneurons of lateral
geniculate nucleus of the cat. J. Neurophysiol. 40, 410-427.
Erisir, A., Van Horn, S.C., and Sherman, S.M. (1998). Distribution of synapses in the lateral
geniculate nucleus of the cat: differences between laminae A and A1 and between relay cells
and interneurons. J. Comp. Neurol. 390, 247-255.
Famiglietti, E.V., Jr. (1970). Dendro-dendritic synapses in the lateral geniculate nucleus of the cat.
Brain Res. 20, 181-191.
Famiglietti, E.V., Jr., and Peters, A. (1972). The synaptic glomerulus and the intrinsic neuron in
the dorsal lateral geniculate nucleus of the cat. J. Comp. Neurol. 144, 285-334.
Felsen, G., and Dan, Y. (2005). A natural approach to studying vision. Nat. Neurosci. 8, 1643-
1646.
Felsen, G., Touryan, J., Han, F., and Dan, Y. (2005). Cortical sensitivity to visual features in
natural scenes. PLoS Biology 3, e342.
Fitzpatrick, D., Penny, G.R., and Schmechel, D.E. (1984). Glutamic acid decarboxylase-
immunoreactive neurons and terminals in the lateral geniculate nucleus of the cat. J. Neurosci.
4, 1809-1829.
141

Fourment, A., Hirsch, J.C., Marc, M.E., and Guidet, C. (1984). Modulation of postsynaptic
activities of thalamic lateral geniculate neurons by spontaneous changes in number of retinal
inputs in chronic cats. 1. Input-output relations. Neuroscience 12, 453.
Friedlander, M.J., Lin, C.S., Stanford, L.R., and Sherman, S.M. (1981). Morphology of functionally
identified neurons in lateral geniculate nucleus of the cat. J. Neurophysiol. 46, 80-129.
Frishman, L.J., and Levine, M.W. (1983). Statistics of the maintained discharge of cat retinal
ganglion cells. J. Physiol. 339, 475-494.
Gilbert, C.D. (1977). Laminar differences in receptive field properties of cells in cat primary visual
cortex. J. Physiol. 268, 391-421.
Godwin, D.W., Van Horn, S.C., Erisir, A., Sesma, M., Romano, C., and Sherman, S.M. (1996a).
Ultrastructural localization suggests that retinal and cortical inputs access different
metabotropic glutamate receptors in the lateral geniculate nucleus. J. Neurosci. 16, 8181-8192.
Godwin, D.W., Vaughan, J.W., and Sherman, S.M. (1996b). Metabotropic glutamate receptors
switch visual response mode of lateral geniculate nucleus cells from burst to tonic. J.
Neurophysiol. 76, 1800-1816.
Govindaiah, and Cox, C.L. (2004). Synaptic activation of metabotropic glutamate receptors
regulates dendritic outputs of thalamic interneurons. Neuron 41, 611-623.
Govindaiah, G., and Cox, C.L. (2006). Metabotropic glutamate receptors differentially regulate
GABAergic inhibition in thalamus. J. Neurosci. 26, 13443-13453.
Granseth, B., and Lindstrom, S. (2003). Unitary EPSCs of corticogeniculate fibers in the rat dorsal
lateral geniculate nucleus In vitro. J. Neurophysiol. 89, 2952-2960.
Guido, W., Lu, S.M., and Sherman, S.M. (1992). Relative contributions of burst and tonic
responses to the receptive field properties of lateral geniculate neurons in the cat. J.
Neurophysiol. 68, 2199-2211.
Guido, W., and Weyand, T. (1995). Burst responses in thalamic relay cells of the awake behaving
cat. J. Neurophysiol. 74, 1782-1786.
Guillery, R.W. (1966). A study of Golgi preparations from the dorsal lateral geniculate nucleus of
the adult cat. J. Comp. Neurol. 128, 21-50.
Guillery, R.W. (1969a). The organization of synaptic interconnections in the laminae of the dorsal
lateral geniculate nucleus of the cat. Zeitschrift für Zellforschung und Mikroskopische
Anatomie 96, 1-38.
142

Guillery, R.W. (1969b). A quantitative study of synaptic interconnections in the dorsal lateral
geniculate nucleus of the cat. Zeitschrift für Zellforschung und Mikroskopische Anatomie 96,
39-48.
Hamos, J.E., Van Horn, S.C., Raczkowski, D., and Sherman, S.M. (1987). Synaptic circuits
involving an individual retinogeniculate axon in the cat. J. Comp. Neurol. 259, 165-192.
Hamos, J.E., Van Horn, S.C., Raczkowski, D., Uhlrich, D.J., and Sherman, S.M. (1985). Synaptic
connectivity of a local circuit neurone in lateral geniculate nucleus of the cat. Nature 317, 618-
621.
Hirsch, J.A. (2003). Synaptic physiology and receptive field structure in the early visual pathway
of the cat. Cereb Cor. 13, 63-69.
Hirsch, J.A., Alonso, J.M., Reid, R.C., and Martinez, L.M. (1998). Synaptic integration in striate
cortical simple cells. J. Neurosci. 18, 9517-9528.
Hirsch, J.A., Martinez, L.M., Pillai, C., Alonso, J.M., Wang, Q., and Sommer, F.T. (2003).
Functionally distinct inhibitory neurons at the first stage of visual cortical processing. Nat.
Neurosci. 6, 1300-1308.
Hochstein, S., and Shapley, R.M. (1976). Quantitative analysis of retinal ganglion cell
classifications. Journal of Physiology 262, 237-264.
Hubel, D.H. (1960). Single unit activity in lateral geniculate body and optic tract of unrestrained
cats. J. Physiol. 150, 91-104.
Hubel, D.H., and Wiesel, T.N. (1961). Integrative action in the cat's lateral geniculate body. J.
Physiol. 155, 385-398.
Huguenard, J.R., and McCormick, D.A. (1992). Simulation of the currents involved in rhythmic
oscillations in thalamic relay neurons. J. Neurophysiol. 68, 1373-1383.
Humphrey, A.L., and Weller, R.E. (1988a). Functionally distinct groups of X-cells in the lateral
geniculate nucleus of the cat. J. Comp. Neurol. 268, 429-447.
Humphrey, A.L., and Weller, R.E. (1988b). Structural correlates of functionally distinct X-cells in
the lateral geniculate nucleus of the cat. J. Comp. Neurol. 268, 448-468.
Jahnsen, H., and Llinas, R. (1984a). Electrophysiological properties of guinea-pig thalamic
neurones: an in vitro study. J. Physiol. 349, 205-226.
143

Jahnsen, H., and Llinas, R. (1984b). Ionic basis for the electro-responsiveness and oscillatory
properties of guinea-pig thalamic neurones in vitro. J. Physiol. 349, 227-247.
Jahnsen, H., and Llinas, R. (1984c). Voltage-dependent burst-to-tonic switching of thalamic cell
activity: an in vitro study. Archives Italiennes de Biologie 122, 73-82.
Jones, J.P., and Palmer, L.A. (1987). The two-dimensional spatial structure of simple receptive
fields in cat striate cortex. J. Neurophysiol. 58, 1187-1211.
Kaplan, E., and Shapley, R. (1984). The origin of the S (slow) potential in the mammalian lateral
geniculate nucleus. Exp. Brain Res. 55, 111-116.
Kara, P., Pezaris, J.S., Yurgenson, S., and Reid, R.C. (2002). The spatial receptive field of thalamic
inputs to single cortical simple cells revealed by the interaction of visual and electrical
stimulation. Proc. Natl. Acad. Sci. 99, 16261-16266.
Kim, U., Sanchez-Vives, M.V., and McCormick, D.A. (1997). Functional dynamics of GABAergic
inhibition in the thalamus. Science 278, 130-134.
Koepsell, K., Wang, X., Vaingankar, V., Wei, Y., Wang, Q., Rathbun, D.L., Usrey, W.M., Hirsch, J.,
and Sommer, F.T. (2009). Retinal oscillations carry visual information to cortex. Front. Sys.
Neurosci. 3.
Landisman, C.E., Long, M.A., Beierlein, M., Deans, M.R., Paul, D.L., and Connors, B.W. (2002).
Electrical synapses in the thalamic reticular nucleus. J. Neurosci. 22, 1002-1009.
Laties, A.M., and Sprague, J.M. (1966). The projection of optic fibers to the visual centers in the
cat. J. Comp. Neurol. 127, 35-70.
Lee, S.C., Hayashida, Y., and Ishida, A.T. (2003). Availability of low-threshold Ca2+ current in
retinal ganglion cells. J. Neurophysiol. 90, 3888-3901.
Lesica, N.A., and Stanley, G.B. (2004). Encoding of natural scene movies by tonic and burst spikes
in the lateral geniculate nucleus. J. Neurosci. 24, 10731-10740.
Lesica, N.A., Weng, C., Jin, J., Yeh, C.I., Alonso, J.M., and Stanley, G.B. (2006). Dynamic
encoding of natural luminance sequences by LGN bursts. PLoS Biology 4, e209.
LeVay, S., and Ferster, D. (1979). Proportion of interneurons in the cat's lateral geniculate nucleus.
Brain Res. 164, 304-308.
Levick, W.R., Cleland, B.G., and Dubin, M.W. (1972). Lateral geniculate neurons of cat: retinal
inputs and physiology. Investigative Ophthalmology 11, 302-311.
144

Levine, M.W., and Cleland, B.G. (2001). An analysis of the effect of retinal ganglion cell impulses
upon the firing probability of neurons in the dorsal lateral geniculate nucleus of the cat. Brain
Res. 902, 244-254.
Levine, M.W., Cleland, B.G., Mukherjee, P., and Kaplan, E. (1996). Tailoring of variability in the
lateral geniculate nucleus of the cat. Biol. Cybern. 75, 219-227.
Lin, C.S., Kratz, K.E., and Sherman, S.M. (1977). Percentage of relay cells in the cat's lateral
geniculate nucleus. Brain Res. 131, 167-173.
Lindstrom, S. (1982). Synaptic organization of inhibitory pathways to principal cells in the lateral
geniculate nucleus of the cat. Brain Res. 234, 447-453.
Liu, R.C., Tzonev, S., Rebrik, S., and Miller, K.D. (2001). Variability and information in a neural
code of the cat lateral geniculate nucleus. J. Neurophysiol. 86, 2789-2806.
Livingstone, M.S., and Hubel, D.H. (1981). Effects of sleep and arousal on the processing of visual
information in the cat. Nature 291, 554-561.
Lo, F.S., Lu, S.M., and Sherman, S.M. (1991). Intracellular and extracellular in vivo recording of
different response modes for relay cells of the cat's lateral geniculate nucleus. Exp. Brain Res.
83, 317-328.
Lo, F.S., and Sherman, S.M. (1990). In vivo recording of postsynaptic potentials and low threshold
spikes in W cells of the cat's lateral geniculate nucleus. Exp. Brain Res. 81, 438-442.
Lorincz, M.L., Kekesi, K.A., Juhasz, G., Crunelli, V., and Hughes, S.W. (2009). Temporal framing
of thalamic relay-mode firing by phasic inhibition during the alpha rhythm. Neuron 63, 683-
696.
Lu, S.M., Guido, W., and Sherman, S.M. (1992). Effects of membrane voltage on receptive field
properties of lateral geniculate neurons in the cat: contributions of the low-threshold Ca2+
conductance. J. Neurophysiol. 68, 2185-2198.
Machens, C.K., Wehr, M.S., and Zador, A.M. (2004). Linearity of cortical receptive fields
measured with natural sounds. J. Neurosci. 24, 1089-1100.
Mackay, D.M., and McCulloch, W.S. (1952). The limiting information capacity of a neuronal link
Bulletin of Mathematical Biology 14, 127-135.
Mante, V., Bonin, V., and Carandini, M. (2008). Functional mechanisms shaping lateral
geniculate responses to artificial and natural stimuli. Neuron 58, 625-638.
145

Mante, V., Frazor, R.A., Bonin, V., Geisler, W.S., and Carandini, M. (2005). Independence of
luminance and contrast in natural scenes and in the early visual system. Nat. Neurosci. 8,
1690-1697.
Martinez, L.M., Wang, Q., Reid, R.C., Pillai, C., Alonso, J.M., Sommer, F.T., and Hirsch, J.A.
(2005). Receptive field structure varies with layer in the primary visual cortex. Nat. Neurosci.
8, 372-379.
Mastronarde, D.N. (1987a). Two classes of single-input X-cells in cat lateral geniculate nucleus. I.
Receptive-field properties and classification of cells. J. Neurophysiol. 57, 357-380.
Mastronarde, D.N. (1987b). Two classes of single-input X-cells in cat lateral geniculate nucleus. II.
Retinal inputs and the generation of receptive-field properties. J. Neurophysiol. 57, 381-413.
Mastronarde, D.N. (1992). Nonlagged relay cells and interneurons in the cat lateral geniculate
nucleus: receptive-field properties and retinal inputs. Vis. Neurosci, 8, 407-441.
McCormick, D.A., and Feeser, H.R. (1990). Functional implications of burst firing and single
spike activity in lateral geniculate relay neurons. Neuroscience 39, 103-113.
McCormick, D.A., and Huguenard, J.R. (1992). A model of the electrophysiological properties of
thalamocortical relay neurons. J. Neurophysiol. 68, 1384-1400.
McIlwain, J.T., and Creutzfeldt, O.D. (1967). Microelectrode study of synaptic excitation and
inhibition in the lateral geniculate nucleus of the cat. J. Neurophysiol. 30, 1-31.
Montero, V.M. (1986). Localization of gamma-aminobutyric acid (GABA) in type 3 cells and
demonstration of their source to F2 terminals in the cat lateral geniculate nucleus: a Golgi-
electron-microscopic GABA-immunocytochemical study. J. Comp. Neurol. 254, 228-245.
Montero, V.M. (1987). Ultrastructural identification of synaptic terminals from the axon of type 3
interneurons in the cat lateral geniculate nucleus. J. Comp. Neurol. 264, 268-283.
Montero, V.M. (1989). The GABA-immunoreactive neurons in the interlaminar regions of the cat
lateral geniculate nucleus: light and electron microscopic observations. Exp. Brain Res. 75,
497-512.
Montero, V.M. (1991). A quantitative study of synaptic contacts on interneurons and relay cells of
the cat lateral geniculate nucleus. Exp. Brain Res. 86, 257-270.
Nelson, R. (1982). AII amacrine cells quicken time course of rod signals in the cat retina. J.
Neurophysiol. 47, 928-947.
146

Olshausen, B.A., and Field, D.J. (2005). How close are we to understanding V1? Neural
Computation 17, 1665-1699.
Pape, H.C., and McCormick, D.A. (1995). Electrophysiological and pharmacological properties of
interneurons in the cat dorsal lateral geniculate nucleus. Neuroscience 68, 1105-1125.
Pasik, P., Pasik, T., and Hámori, J. (1976). Synapses between interneurons in the lateral geniculate
nucleus of monkeys. Exp. Brain Res. 25, 1-13.
Pasik, P., Pasik, T., Hamori, J., and Szentagothai, J. (1973). Golgi type II interneurons in the
neuronal circuit of the monkey lateral geniculate nucleus. Exp. Brain Res. 17, 18-34.
Peichl, L., and Wässle, H. (1983). The structural correlate of the receptive field centre of alpha
ganglion cells in the cat retina. J. Physiol. 341, 309-324.
Perkel, D.H., Gerstein, G.L., and Moore, G.P. (1967). Neuronal spike trains and stochastic point
processes. II. Simultaneous spike trains. Biophysical Journal 7, 419-440.
Person, A.L., and Perkel, D.J. (2005). Unitary IPSPs drive precise thalamic spiking in a circuit
required for learning. Neuron 46, 129-140.
Pillow, J.W., Paninski, L., Uzzell, V.J., Simoncelli, E.P., and Chichilnisky, E.J. (2005). Prediction
and decoding of retinal ganglion cell responses with a probabilistic spiking model. J. Neurosci.
25, 11003-11013.
Pillow, J.W., and Simoncelli, E.P. (2006). Dimensionality reduction in neural models: an
information-theoretic generalization of spike-triggered average and covariance analysis.
Journal of Vision 6, 414-428.
Ramcharan, E.J., Gnadt, J.W., and Sherman, S.M. (2000). Burst and tonic firing in thalamic cells
of unanesthetized, behaving monkeys. Visual Neuroscience 17, 55-62.
Rathbun, D.L., Alitto, H.J., Weyand, T.G., and Usrey, W.M. (2007). Interspike interval analysis of
retinal ganglion cell receptive fields. J. Neurophysiol. 98, 911-919.
Reinagel, P., Godwin, D., Sherman, S.M., and Koch, C. (1999). Encoding of visual information by
LGN bursts. J. Neurophysiol. 81, 2558-2569.
Reinagel, P., and Reid, R.C. (2000). Temporal coding of visual information in the thalamus. J.
Neurosci. 20, 5392-5400.
Rieke, F., Warland, D., de Ruyter van Steveninck, R.R., and Bialek, W. (1997). Spikes: exploring
the neural code (MIT Press).
147

Ruksenas, O., Fjeld, I.T., and Heggelund, P. (2000). Spatial summation and center-surround
antagonism in the receptive field of single units in the dorsal lateral geniculate nucleus of cat:
comparison with retinal input. Vis. Neurosci, 17, 855-870.
Rust, N.C., and Movshon, J.A. (2005). In praise of artifice. Nat. Neurosci. 8, 1647-1650.
Rust, N.C., Schwartz, O., Movshon, J.A., and Simoncelli, E.P. (2005). Spatiotemporal elements of
macaque v1 receptive fields. Neuron 46, 945-956.
Sanchez-Vives, M.V., Bal, T., Kim, U., von Krosigk, M., and McCormick, D.A. (1996). Are the
interlaminar zones of the ferret dorsal lateral geniculate nucleus actually part of the
perigeniculate nucleus? J. Neurosci. 16, 5923-5941.
Sanderson, K.J. (1971). The projection of the visual field to the lateral geniculate and medial
interlaminar nuclei in the cat. J. Comp. Neurol. 143, 101-108.
Sanderson, K.J., Bishop, P.O., and Darian-Smith, I. (1971). The properties of the binocular
receptive fields of lateral geniculate neurons. Exp. Brain Res. 13, 178-207.
Sanderson, K.J., Darian-Smith, I., and Bishop, P.O. (1969). Binocular corresponding receptive
fields of single units in the cat dorsal lateral geniculate nucleus. Vision Res. 9, 1297-1303.
Sanderson, K.J., and Pearson, L.J. (1977). Retinal projections in the native cat, Dasyurus
viverrinus. J. Comp. Neurol. 174, 347-357.
Scharfman, H.E., Lu, S.M., Guido, W., Adams, P.R., and Sherman, S.M. (1990). N-methyl-D-
aspartate receptors contribute to excitatory postsynaptic potentials of cat lateral geniculate
neurons recorded in thalamic slices. Proceedings of the National Academy of Sciences, USA
87, 4548-4552.
Schiller, P.H., Finlay, B.L., and Volman, S.F. (1976). Quantitative studies of single-cell properties
in monkey striate cortex. I. Spatiotemporal organization of receptive fields. J. Neurophysiol.
39, 1288-1319.
Schwartz, O., Pillow, J.W., Rust, N.C., and Simoncelli, E.P. (2006). Spike-triggered neural
characterization. Journal of Vision 6, 484-507.
Shapley, R., and Hochstein, S. (1975). Visual spatial summation in two classes of geniculate cells.
Nature 256, 411-413.
Sharpee, T., Rust, N.C., and Bialek, W. (2004). Analyzing neural responses to natural signals:
maximally informative dimensions. Neural Computation 16, 223-250.
148

Sharpee, T.O., Sugihara, H., Kurgansky, A.V., Rebrik, S.P., Stryker, M.P., and Miller, K.D. (2006).
Adaptive filtering enhances information transmission in visual cortex. Nature 439, 936-942.
Sherman, S.M. (2004). Interneurons and triadic circuitry of the thalamus. Trends in Neuroscience
27, 670-675.
Sherman, S.M., and Friedlander, M.J. (1988). Identification of X versus Y properties for
interneurons in the A-laminae of the cat's lateral geniculate nucleus. Exp. Brain Res. 73, 384-
392.
Sherman, S.M., and Guillery, R.W. (1996). Functional organization of thalamocortical relays. J.
Neurophysiol. 76, 1367-1395.
Sherman, S.M., and Guillery, R.W. (1998). On the actions that one nerve cell can have on another:
distinguishing "drivers" from "modulators". Proc. Natl. Acad. Sci. 95, 7121-7126.
Shou, T., Ruan, D., and Zhou, Y. (1986). The orientation bias of LGN neurons shows topographic
relation to area centralis in the cat retina. Experimental Brain Research 64, 233-236.
Sillito, A.M., and Kemp, J.A. (1983). The influence of GABAergic inhibitory processes on the
receptive field structure of X and Y cells in cat dorsal lateral geniculate nucleus (dLGN). Brain
Res. 277, 63-77.
Simoncelli, E.P., and Olshausen, B.A. (2001). Natural image statistics and neural representation.
Annu. Rev. Neurosci. 24, 1193-1216.
Sincich, L.C., Adams, D.L., Economides, J.R., and Horton, J.C. (2007). Transmission of spike
trains at the retinogeniculate synapse. J. Neurosci. 27, 2683-2692.
Sincich, L.C., Horton, J.C., and Sharpee, T.O. (2009). Preserving information in neural
transmission. J. Neurosci. 29, 6207-6216.
Singer, W., and Creutzfeldt, O.D. (1970). Reciprocal lateral inhibition of on- and off-center
neurones in the lateral geniculate body of the cat. Exp. Brain Res. 10, 311-330.
Singer, W., Poppel, E., and Creutzfeldt, O. (1972). Inhibitory interaction in the cat's lateral
geniculate nucleus. Exp. Brain Res. 14, 210-226.
Soltesz, I., Lightowler, S., Leresche, N., and Crunelli, V. (1989). Optic tract stimulation evokes
GABAA but not GABAB IPSPs in the rat ventral lateral geniculate nucleus. Brain Res. 479,
49-55.
149

Steriade, M., McCormick, D.A., and Sejnowski, T.J. (1993). Thalamocortical oscillations in the
sleeping and aroused brain. Science 262, 679-685.
Swadlow, H.A., and Gusev, A.G. (2001). The impact of 'bursting' thalamic impulses at a
neocortical synapse. Nat. Neurosci. 4, 402-408.
Swadlow, H.A., Gusev, A.G., and Bezdudnaya, T. (2002). Activation of a cortical column by a
thalamocortical impulse. J. Neurosci. 22, 7766-7773.
Szentagothai, J., Hamori, J., and Tombol, T. (1966). Degeneration and electron microscope
analysis of the synaptic glomeruli in the lateral geniculate body. Exp. Brain Res. 2, 283-301.
Temereanca, S., Brown, E.N., and Simons, D.J. (2008). Rapid Changes in Thalamic Firing
Synchrony during Repetitive Whisker Stimulation. J. Neurosci. 28, 11153-11164.
Touryan, J., Felsen, G., and Dan, Y. (2005). Spatial structure of complex cell receptive fields
measured with natural images. Neuron 45, 781-791.
Uhlrich, D.J., and Cucchiaro, J.B. (1992). GABAergic circuits in the lateral geniculate nucleus of
the cat. Progress in Brain Research 90, 171-192.
Uhlrich, D.J., Cucchiaro, J.B., Humphrey, A.L., and Sherman, S.M. (1991). Morphology and
axonal projection patterns of individual neurons in the cat perigeniculate nucleus. J.
Neurophysiol. 65, 1528-1541.
Usrey, W.M., Alonso, J.M., and Reid, R.C. (2000). Synaptic interactions between thalamic inputs
to simple cells in cat visual cortex. J. Neurosci. 20, 5461-5467.
Usrey, W.M., Reppas, J.B., and Reid, R.C. (1998). Paired-spike interactions and synaptic efficacy
of retinal inputs to the thalamus. Nature 395, 384-387.
Usrey, W.M., Reppas, J.B., and Reid, R.C. (1999). Specificity and strength of retinogeniculate
connections. J. Neurophysiol. 82, 3527-3540.
Van Essen, D.C., Anderson, C.H., and Felleman, D.J. (1992). Information processing in the
primate visual system: an integrated systems perspective. Science 255, 419-423.
Van Horn, S.C., Erisir, A., and Sherman, S.M. (2000). Relative distribution of synapses in the A-
laminae of the lateral geniculate nucleus of the cat. J. Comp. Neurol. 416, 509-520.
Wang, S., Bickford, M.E., Van Horn, S.C., Erisir, A., Godwin, D.W., and Sherman, S.M. (2001).
Synaptic targets of thalamic reticular nucleus terminals in the visual thalamus of the cat. J.
Comp. Neurol. 440, 321-341.
150

Wang, X., Wei, Y., Vaingankar, V., Wang, Q., Koepsell, K., Sommer, F.T., and Hirsch, J.A. (2007).
Feedforward excitation and inhibition evoke dual modes of firing in the cat's visual thalamus
during naturalistic viewing. Neuron 55, 465-478.
Wässle, H., Boycott, B.B., and Illing, R.B. (1981). Morphology and mosaic of on- and off-beta cells
in the cat retina and some functional considerations. Proc. R. Soc. Lond. B Biol. Sci. 212, 177-
195.
Weyand, T.G., Boudreaux, M., and Guido, W. (2001). Burst and tonic response modes in
thalamic neurons during sleep and wakefulness. J. Neurophysiol. 85, 1107-1118.
Wiesel, T.N. (1959). Recording inhibtion and excitation in the cat's retinal ganglion cells with
intracellular electrodes. Nature 183, 264-265.
Wilson, J. (1989). Synaptic organization of individual neurons in the macaque lateral geniculate
nucleus. J. Neurosci. 9, 2931-2953.
Wilson, J.R., Friedlander, M.J., and Sherman, S.M. (1984). Fine structural morphology of
identified X- and Y-cells in the cat's lateral geniculate nucleus. Proc. R. Soc. Lond. B Biol. Sci.
221, 411-436.
Wolfart, J., Debay, D., Le Masson, G., Destexhe, A., and Bal, T. (2005). Synaptic background
activity controls spike transfer from thalamus to cortex. Nat. Neurosci. 8, 1760-1767.
Wolfe, J., and Palmer, L.A. (1998). Temporal diversity in the lateral geniculate nucleus of cat. Vis.
Neurosci, 15, 653-675.
Wörgötter, F., Nelle, E., Li, B., and Funke, K. (1998). The influence of corticofugal feedback on
the temporal structure of visual responses of cat thalamic relay cells. J. Physiol. 509 (Pt 3),
797-815.

Abstract (if available)

Abstract In the mammalian visual system, sensory information captured by the retina is routed through the lateral geniculate nucleus (LGN) of the thalamus before reaching the cerebral cortex. The lateral geniculate circuits thus operate as a gateway for visual information flowing from the sensory periphery to the central nervous system. Traditionally, the LGN has been regarded as a passive relay station rather than an active processing center because the receptive fields of geniculate neurons closely resemble those of retinal ganglion cells. This simplistic view, however, is at odds with the complexity of the thalamic neural circuits - the LGN houses a substantial population of inhibitory neurons that can modulate the activity of the thalamocortical projecting neurons. How do the excitatory and inhibitory neurons in the LGN form circuits to process visual information from the retina to the cortex? In this dissertation, three studies that address this question will be presented. Using in vivo patch recordings, these studies directly probe the functional components of the thalamic neural circuit in the whole animal

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Visual information processing by the feedforward and feedback inputs to the lateral geniculate nucleus

PDF

Exploring sensory responses in the different subdivisions of the visual thalamus

PDF

Spatial and temporal precision of inhibitory and excitatory neurons in the murine dorsal lateral geniculate nucleus

PDF

Neural circuits underlying the modulation and impact of defensive behaviors

PDF

Sensitivity and dynamic range of rod pathways in the mammalian retina

PDF

Contextual modulation of sensory processing via the pulvinar nucleus

PDF

Synaptic mechanisms for basic auditory processing

PDF

Functional magnetic resonance imaging characterization of peripheral form vision

PDF

Neural circuits control and modulate innate defensive behaviors

PDF

Functional properties of the superficial cortical interneurons

PDF

Dendritic computation and plasticity in neuromorphic circuits

PDF

Synaptic circuits for information processing along the central auditory pathway

PDF

Integration in neural circuits controlling ingestive behaviors

PDF

Signal processing by the mammalian retina near absolute visual threshold

PDF

From sensory processing to behavior control: functions of the inferior colliculus

PDF

Functional synaptic circuits in primary visual cortex

PDF

Cortical and subcortical responses to electrical stimulation of rat retina

PDF

Astrocyte-mediated plasticity and repair in CMOS neuromorphic circuits

PDF

Improving stimulation strategies for epiretinal prostheses

PDF

Synaptic mechanism underlying development and function of neural circuits in rat primary auditory cortex

Asset Metadata

Creator Wang, Xin (author)

Core Title Sensory information processing by retinothalamic neural circuits

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Neuroscience

Publication Date 03/03/2010

Defense Date 01/21/2010

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag GABAergic,information theory,inhibition,interneuron,lateral geniculate nucleus,neural coding,OAI-PMH Harvest,receptive field,retina,thalamus

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Hirsch, Judith A. (committee chair), Mel, Bartlett W. (committee member), Sampath, Alapakkam P. (committee member), Sommer, Friedrich T. (committee member), Tjan, Bosco S. (committee member)

Creator Email caseus.viridis@gmail.com,xinw@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m2862

Unique identifier UC1230829

Identifier etd-Wang-3481 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-292270 (legacy record id),usctheses-m2862 (legacy record id)

Legacy Identifier etd-Wang-3481.pdf

Dmrecord 292270

Document Type Dissertation

Rights Wang, Xin

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu