Cortical Feedback Controls the Frequency and Synchrony of
Oscillations in the Visual Thalamus
Thierry Bal,
1
Damien Debay,
1
and Alain Destexhe
1,2
1
Unite´ de Neurosciences Inte´gratives et Computationnelles, Centre National de la Recherche Scientifique, Unite´ Propre de
Recherche 2191, Institut de Neurobiologie A. Fessard, 91 198, Gif-sur-Yvette Cedex, France, and
2
Department of
Physiology, Laval University, Que´bec G1K 7P4, Canada
Thalamic circuits have an intrinsic capacity to generate state-
dependent oscillations of different frequency and degrees of
synchrony, but little is known of how synchronized oscillation is
controlled in the intact brain or what function it may serve. The
influence of cortical feedback was examined using slice prepa-
rations of the visual thalamus and computational models. Corti-
cal feedback was mimicked by stimulating corticothalamic ax-
ons, triggered by the activity of relay neurons. This artificially
coupled network had the capacity to self-organize and to gen-
erate qualitatively different rhythmical activities according to the
strength of corticothalamic feedback stimuli. Weak feedback
(one to three shocks at 100–150 Hz) phase-locked the sponta-
neous spindle oscillations (6–10 Hz) in geniculate and peri-
geniculate nuclei. However, strong feedback (four to eight
shocks at 100–150 Hz) led to a more synchronized oscillation,
slower in frequency (2–4 Hz) and dependent on GABA
B
recep-
tors. This increase in synchrony was essentially attributable to a
redistribution of the timing of action potential generation in lateral
geniculate nucleus cells, resulting in an increased output of relay
cells toward the cortex. Corticothalamic feedback is thus capa-
ble of inducing highly synchronous slow oscillations in physio-
logically intact thalamic circuits. This modulation may have im-
plications for a better understanding of the descending control of
thalamic nuclei by the cortex, and the genesis of pathological
rhythmical activity, such as absence seizures.
Key words: corticothalamic; spike and wave; absence seizure;
GABA
B
; spindle waves; thalamus; thalamic reticular nucleus;
closed loop system
Thalamic circuits can display different types of oscillation charac-
terized by their frequencies and levels of synchrony. The most
common rhythmical activity seen in intact thalamic circuits is the
7–14 Hz spindle rhythm, which consists of periodically recurring
waxing and waning oscillations (Andersen and Andersson, 1968;
Steriade and Descheˆnes, 1984; von Krosigk et al., 1993; Contreras
et al., 1996). In slice preparations from the ferret, 6–10 Hz spindle
oscillation can be self-generated provided that the circuitry linking
the dorsal lateral geniculate nucleus (LGNd) and the perigenicu-
late nucleus (PGN) is conserved intact. Blockade of GABA
A
receptors by application of bicuculline transforms the spontaneous
spindling pattern to a slower and more synchronized oscillation at
2–4 Hz (von Krosigk et al., 1993; Bal et al., 1995a,b). This slower
rhythm is strikingly similar to the typical 3 Hz frequency of
absence seizures in humans, in which the thalamus is thought to be
a key player (Gloor and Fariello, 1988).
The genesis of these autonomous rhythms is now well under-
stood in terms of the thalamic recurrent circuits and intrinsic
cellular properties involved (Steriade et al., 1993). However, our
understanding of thalamic rhythmic generation remains limited
concerning its control by external sources, such as the feedforward
retinal projection or the feedback projection from cortex. We know
that cortical feedback provides an extremely dense projection to
the thalamus (Guillery, 1969; Liu et al., 1995; Erisir et al., 1997a,b;
Liu and Jones, 1999), which may serve to control faster rhythms of
thalamic oscillation, in the gamma frequency range (20–60 Hz)
reported in vivo during sensory processing (Ghose and Freeman,
1992; Sillito et al., 1994; Neuenschwander and Singer, 1996). Cor-
ticothalamic feedback is also essential in coordinating widespread,
coherent, sleep-related synchronized oscillation of different tha-
lamic nuclei (Contreras et al., 1996). More recently, computational
models have predicted that corticothalamic feedback could control
the transitions between different ranges of thalamic oscillation,
defined by their frequency and synchrony (Destexhe, 1998).
Here we test this hypothesis using a new ferret slice preparation
preserving the optic tract and the optic radiation bundle in which
corticofugal fibers can be stimulated, while maintaining intact the
endogenous genesis of spindles (see Fig. 2A). The sequence of
synaptic and voltage-gated cellular events following activation of
cortical feedback was studied with intracellular current-clamp re-
cordings of perigeniculate and LGN relay neurons. Computational
models and multiple extracellular recordings were used to assess
the level of local recruitment and synchronized activity in the
network. From the experimental results and computational models
we propose principles of functional network organization able to
explain the cortical control of thalamic oscillations.
MATERIALS AND METHODS
Slice preparation. Adult ferrets, 3- to 15-months-old (n 5 24) (Marshall
Europe, Lyon, France; Dostes, St. Creac, France), were anesthetized with
sodium pentobarbital (45 mg/kg). LGNd slices (350 mm) were prepared in
a solution (see below) in which NaCl was replaced with sucrose while
maintaining an osmolarity of 307 mOsm (adapted from Aghajanian and
Rasmussen, 1989). After preparation, slices were placed in an interface-
style recording chamber (Fine Science Tools, Heidelberg, Germany). The
bathing medium contained (in mM): NaCl, 124; KCl, 2.5; MgSO
4
, 1.2;
NaH
2
PO
4
, 1.25; CaCl
2
, 2; NaHCO
3
, 26; dextrose, 10, and was aerated with
95% O
2
and 5% CO
2
to a final pH of 7.4. Bath temperature was main-
tained at 34.5–35.5°C. LGNd slices were cut in a plane parallel to the most
proximal extent of the optic tract (see Fig. 2A). This new procedure
preserved a portion of the optic radiation containing corticothalamic axons
and their synaptic connections to thalamic cells, as well as several milli-
meters of the optic tract. Extracellular multiunit recordings from LGN
laminae revealed periodic spontaneous spindle waves that were indistin-
guishable from those obtained from sagittal slices used previously (von
Krosigk et al., 1993; Bal et al., 1995a). The presence of this network
activity indicates that synaptic connections between perigeniculate and
thalamocortical cells were functionally intact.
Received March 16, 2000; revised June 26, 2000; accepted July 14, 2000.
This work was supported by the Centre National de la Recherche Scientifique, the
Fondation Franc¸aise pour la Recherche sur l’E
´
pilepsie, the Institut Electricite´ Sante´ of
France, and by the Medical Research Council of Canada (MT-13724). We acknowl-
edge the outstanding help of Gerard Sadoc for data acquisition and signal analysis, and
K. Grant, Y. Fregnac, and B. S. Gutkin for in depth discussion and comments on this
manuscript. We also thank Alan Carleton and David Desmaisons for their helpful
input.
Correspondence should be addressed to Dr. T. Bal, Unite´ de Neurosciences Inte´-
gratives et Computationnelles, Centre National de la Recherche Scientifique Unite´
Propre de Recherche 2191, Institut de Neurobiologie A. Fessard, 1 Avenue de la
Terrasse, 91 198, Gif-sur-Yvette Cedex, France. E-mail: Thierry.Bal@iaf.cnrs-gif.fr.
Copyright © 2000 Society for Neuroscience 0270-6474/00/207478-11$15.00/0
The Journal of Neuroscience, October 1, 2000, 20(19):7478–7488

Electrophysiology. Extracellular recordings were obtained with low-
resistance (,5MV) tungsten microelectrodes (Frederick Haer, Bowdoin-
ham, ME). Intracellular recording electrodes were made on a Sutter
Instruments P-87 micropipette puller from medium-walled glass (WPI,
1B100F) and beveled on a Sutter Instruments beveler (BV-10M). Micropi-
pettes were filled with 1.2 M K acetate and had resistances of 90–120 MV
after beveling. Cells were included in the present study if they exhibited a
stable resting membrane potential for at least 10 min (typically 30–180
min), were able to generate bursts of overshooting action potentials, and
exhibited apparent input resistances of at least 30 MV (on average 80.4 6
32.8 MV; n 5 15). LGN cells had an average resting membrane potential
of 62.2 6 3.8 mV (n 5 16).
Cells were identified in extracellular recordings according to their
location, the duration of their action potentials, and the temporal structure
of their action potential bursts: in PGN cells, but not in thalamocortical
cells, the frequency of action potential generation within each burst in-
creased, then decreased in frequency in an “accelerando-decelerando”
pattern (Domich et al., 1986; Hu et al., 1989; Bal et al., 1995b).
The optic radiation (OR) was stimulated at a distance of 400–800 mm
from the PGN, using bipolar tungsten electrodes similar to those used for
extracellular recordings, spaced 200–450 mm apart, and oriented perpen-
dicular to the corticothalamic fiber bundles (see Fig. 2A). Tips were
electroplated with gold in a 2% HAuCL
4
solution. Stimulations ranged
from 10 to 40 mA (0.1 msec duration). Feedback stimuli of corticothalamic
fibers were triggered by the activity of thalamic relay cells using a custom
data acquisition software (Acquis1; developed by G. Sadoc, Unite´ de
Neurosciences Inte´gratives et Computationnelles, Centre National de la
Recherche Scientifique Gif-sur-Yvette, Agence Nationale pour la Valori-
sation des Applications de la Recherche Biological). The software detected
the LGN discharges in intracellular (Axoclamp-2B amplifier; Digidata
1200 analog-to-digital converter; Axon Instruments, Foster City, CA) and
multiunit recordings by a voltage threshold, and set the latency at which a
command was sent to an OR-stimulating unit (A360; WPI). In some
intracellular recordings, the voltage threshold was set below the peak of the
low-threshold calcium spike. A minimum interstimulus interval of 100
msec was set after first spike detection to avoid overstimulation triggered
by the recurrence of spikes within the burst itself. The results presented
here, using the feedback paradigm, were obtained in 21 slices taken from
17 animals.
Electrical stimulation of the optic radiation resulted in orthodromic
activation of corticofugal axons and generated mixed IPSPs and EPSPs,
recorded intracellularly in thalamocortical relay cells. Antidromic invasion
was observed only exceptionally (2 of 40 LGN cells), and those cases were
discarded from the present analysis. Antidromic spikes were recognized by
their short and stable latency (0.67 6 0.1 msec from artifact to peak; n 5
50 events) and the lack of underlying EPSP, whereas monosynaptic corti-
cothalamic EPSPs, recorded in the same cell, had a longer and more
variable latency (2.74 6 0.29 msec; n 5 50 events) consistent with the
latency of EPSPs mediated monosynaptically in LGN principal cells by
slowly conducting corticogeniculate fibers described previously in vivo
(Tsumoto et al., 1978; Ahlsen et al., 1982). The effect of antidromic
activation was also tested using the model (data not shown), which indi-
cated that the antidromic activation of a minority of LGN or PGN cells had
no detectable effect on network behavior. Thus, whereas the contribution
of antidromic activation of LGN relay cells to the response of LGN and
PGN cells cannot be completely ruled out, it certainly remains small
compared to the contribution of the orthodromic activation of corticotha-
lamic axons.
To block GABA
B
responses, the antagonist CGP 35348 (gift of Ciba-
Geigy) was delivered locally with the pressure-pulse technique in which an
air puff (3 psi extruded volume of 2–20 pl) (Picospritzer; General Valve,
Fairfield, NJ; 10–100 msec). The drug was applied either to the surface or
in the depth of the slice within 50–100 mm of the entry point of the
recording electrode.
Data analysis and statistics. Analysis was performed using Acquis1, a
custom software. The level of cell recruitment and synchronized activity in
the network was best visualized by half rectifying the multiunit signal, and
then smoothing it by a moving average technique to enhance the detection
of cells coactive within a given window. Smoothed integration was per-
formed with a 10 msec time constant, except for the responses illustrated
in Figure 6A (20 msec). Autocorrelation functions were applied to the
reconstructed local field potential and to the intracellular current-clamp
recordings, after removal of action potentials and stimulation artifacts by a
software routine (Bringuier et al., 1997). Measurements of the period of
oscillatory activity were derived from the abscissa of the first peak in the
normalized autocorrelation functions calculated on at least 80, and up to
500 cycles of oscillation. Nonparametric Wilcoxon matched paired tests
were applied to the oscillation period values observed in the control case
and in the presence of a cortical feedback (with a significance level of 0.05).
Models. Computational models of thalamic neurons were designed based
on previous studies (Destexhe et al., 1996; Destexhe, 1998). LGN and PGN
neurons were modeled by single compartment representations including
various intrinsic voltage- and calcium-dependent currents, such as I
T
, I
h
,
I
Na
, and I
K
in LGN cells and I
T
, I
Na
, and I
K
in PGN. These intrinsic
currents were represented by Hodgkin–Huxley-type models. In addition,
I
h
contained an upregulation by intracellular calcium as described previ-
ously (Destexhe et al., 1996). LGN and PGN neurons generated bursts of
action potentials with a strength and voltage dependence similar to that
observed experimentally.
Postsynaptic currents mediated by glutamate (AMPA and NMDA re-
ceptors) and GABA (GABA
A
and GABA
B
receptors) were simulated
using kinetic models of postsynaptic receptors (Destexhe et al., 1998b).
The synaptic interactions modeled were LGN3 PGN (AMPA receptors),
PGN 3 PGN (GABA
A
receptors), and PGN 3 LGN (GABA
A
and
GABA
B
receptors; Fig. 1A,B), as found experimentally (von Krosigk et al.,
1993). Each cell type established connections within a local area of 10% of
the size of the network, in a topographically organized manner (Destexhe
et al., 1996). Corticothalamic feedback was mediated by AMPA receptors
on both LGN and PGN cells. NMDA receptors were also incorporated
in some simulations (conductance of 25% of that of AMPA receptors),
but they did not affect the present results (data not shown). mGluR
receptors have been described in LGN and reticularis neurons (von
Krosigk and McCormick, 1993; Cox and Sherman, 1999), but were not
incorporated here.
Feedback simulations were designed similarly to the experiments re-
ported here. In a network consisting of two one-dimensional rows of 100
LGN and 100 PGN cells, the suprathreshold activation of a single LGN cell
was chosen as trigger (Fig. 1A). When this LGN cell fired, the first spike
was used to trigger a burst of presynaptic stimulation of corticothalamic
synapses, after a delay of 10–50 msec. As in experiments, the number and
strength of stimuli were varied.
RESULTS
We first describe with the model the paradigm and the hypothesis
tested here, namely that corticothalamic feedback can control the
type of oscillation displayed by the thalamus. We then investigate
this theoretical prediction experimentally, and establish the char-
acteristics of the control of thalamic oscillations by cortical feed-
back. Finally, based on these experimental data, we return to the
model to analyze the network mechanisms underlying the cortical
control of thalamic oscillations.
Models predict that corticothalamic feedback can
control thalamic oscillations
A thalamocortical network model was introduced previously to
model ;3 Hz spike-and-wave seizures based on the biophysical
properties of neurons and synapses in thalamocortical circuits
(Destexhe, 1998). This model formulated one main prediction, that
corticothalamic feedback can force the intact thalamus from con-
trol spindles (8–12 Hz) to a different oscillatory mode, slower (;3
Hz) and more synchronized. To test this prediction in thalamic
slices, one must reconstitute the thalamus-cortex-thalamus loop.
We have thus elaborated a paradigm that consists of forming an
artificial feedback loop between the activity of the LGN neurons
and the stimulation of corticothalamic fibers (Fig. 1A). This can
later be tested experimentally.
We first simulated this paradigm using a model network of 100
PGN and 100 LGN cells interconnected via AMPA, GABA
A
, and
GABA
B
receptors (Fig. 1A). Cells were modeled by a single-
compartment incorporating calcium- and voltage-dependent cur-
rents (Fig. 1B) as in previous models (Destexhe et al., 1996). The
spike activity of one LGN cell was used to trigger the stimulation
of corticothalamic EPSPs across the entire network. A burst of
action potential in the trigger LGN cell started a high-frequency
(100 Hz) burst of AMPA-mediated corticothalamic EPSPs in LGN
and PGN neurons (see Materials and Methods). The strength of
the feedback stimulation was adjusted by controlling the number of
corticothalamic EPSPs (number of shocks).
In the case of mild feedback (one to four shocks at 100 Hz for the
conductance settings given in Materials and Methods), the pattern
of LGN and PGN discharge was typical of spindle oscillations (Fig.
1C; one shock): individual LGN cells showed subharmonic bursting
activity, and were not tightly synchronized. In this case, the pres-
ence of the feedback did not disrupt the pattern of spindle oscil-
lations, but only slightly increased the synchrony of LGN cells (see
below). This is consistent with previous models showing that mild
corticothalamic feedback can control the onset and distribution of
spindling activity, but does not change its cellular features (Des-
texhe et al., 1998a).
A radically different picture was obtained for stronger feedback
stimulation. When the number of stimuli was increased to five shocks
or more, the pattern of bursting changed qualitatively, and the
Bal et al. • Control of Thalamic Oscillations by Cortical Feedback J. Neurosci., October 1, 2000, 20(19):7478–7488 7479