Proc. Natl. Acad. Sci. USA
Vol. 95, pp. 7121–7126, June 1998
On the actions that one nerve cell can have on another:
Distinguishing ‘‘drivers’’ from ‘‘modulators’’
S. MURRAY SHERMAN*†AND R. W. GUILLERY‡
*Department of Neurobiology, State University of New York, Stony Brook, New York 11794-5230; and‡Department of Anatomy, University of Wisconsin School
of Medicine, 1300 University Avenue, Madison, WI 53706
Communicated by Francis Crick, The Salk Institute for Biological Studies, San Diego, CA, April 8, 1998 (received for review January 29, 1998)
postsynaptic effect can vary greatly. In sensory systems,
inputs from ‘‘drivers’’ can be differentiated from those of
‘‘modulators.’’ The driver can be identified as the transmitter
of receptive field properties; the modulator can be identified
as altering the probability of certain aspects of that trans-
mission. Where receptive fields are not available, the distinc-
tion is more difficult and currently is undefined. We use the
visual pathways, particularly the thalamic geniculate relay for
which much relevant evidence is available, to explore ways in
which drivers can be distinguished from modulators. The
extent to which the distinction may apply first to other parts
of the thalamus and then, possibly, to other parts of the brain
is considered. We suggest the following distinctions: Cross-
correlograms from driver inputs have sharper peaks than
those from modulators; there are likely to be few drivers but
many modulators for any one cell; and drivers are likely to act
only through ionotropic receptors having a fast postsynaptic
effect whereas modulators also are likely to activate metabo-
tropic receptors having a slow and prolonged postsynaptic
When one nerve cell acts on another, its
When one nerve cell signals to another, the effect can vary
greatly. One distinction recognized currently is that between
‘‘drivers’’ and ‘‘modulators’’ (1–4). The former carry the
latter can alter the effectiveness of the drive without contrib-
uting significantly to the general pattern of the message.
However, it is often difficult to define whether afferents are
drivers or modulators; it also is difficult to define precisely
what is meant by either term. We here explore the distinction
between drivers and modulators and add a third category:
‘‘disrupters.’’ We base our analysis on the thalamus, where
these categories are likely to prove useful, and we explore the
extent to which they also may apply to other parts of the brain.
In many sensory relays, drivers can be recognized because
they transmit information from peripheral receptors to the
brain. For the major sensory relays of the thalamus, we
proposed (5) that afferents can be assigned, on the basis of
their fine structure and function, to one of two classes (Fig. 1).
the receptive field properties of the postsynaptic cells, or else
modulators, which change the pattern of transmission but do
not alter the receptive field properties significantly. The clear-
est examples of drivers are visual, auditory, or somatosensory
axons that innervate the lateral geniculate, the medial genic-
ulate, or ventrobasal nucleus, respectively. These all have
similar fine structural characteristics and synaptic relation-
ships, and we argued (5) that other comparable axons inner-
vating ventral lateral, ventral anterior, or the anterior thalamic
their relay nuclei (Fig. 1).
Further, two distinct fiber types going from the cerebral
cortex to the thalamus can be recognized. All thalamic nuclei
appear to receive afferents from cells in cortical layer 6, and
we shall argue that these are modulators. For nuclei like the
lateral or medial geniculate or the ventrobasal nucleus, which
5), this is their only cortical innervation. Other thalamic nuclei,
like the pulvinar, the mediodorsal nucleus, or the posterior
thalamic group (labeled HO for ‘‘higher order’’ relays in Fig.
terminals whose fine structural and synaptic relationships
resemble the primary afferents from the retina or the medial
to be the drivers. That is, they define the receptive field
pathways, then, drivers carry the receptive field properties and
modulators do not. In nuclei like the mediodorsal nucleus, and
in many other thalamic nuclei, receptive fields are not defined,
and there is as yet no criterion for distinguishing a driver from
Beyond the thalamus, the problem is more difficult and of
broader significance. For instance, Crick and Koch (1) recently
used a classification of ‘‘drivers’’ and ‘‘modulators’’ in a
discussion of corticocortical pathways, and part of their argu-
ment concerned the distinction between drivers and modula-
tors. They showed that the functional significance of a set of
connections depends on whether a component is a driver or a
modulator. They propose connectional ground rules for iden-
tifying a pathway as a driver or modulator, but at present, these
are ad hoc rules. We know of no critical evidence to identify
a corticocortical pathway as one or the other in the way that
receptive field properties can be used to identify afferents in
some thalamic relays.
The distinction between drivers and modulators may prove
useful for some pathways other than thalamic and cortical
circuits. A first step has to be a rigorous definition. We start
with the sensory relay nuclei of the thalamus, particularly the
lateral geniculate. Here, we have information about afferents,
receptive fields, and relay cells. The thalamus may be unique
from the point of view of the classification that we are
exploring, and, further, circuits concerned with sensory relays
may have special properties not shared by other thalamic
nuclei. However, our analysis of the geniculate aims not so
much to show well defined rules for the proposed classification
but, rather, to explore observations that may be relevant in
other thalamic nuclei and other brain centers. There is cur-
rently a lack of critical factual observations for establishing
criteria to distinguish between drivers and modulators, either
just for the thalamus or for a broad range of brain centers.
Afferents to the Lateral Geniculate Nucleus
It is almost a tautology to define retinal afferents to the lateral
geniculate nucleus as drivers. We start with this assertion and
The publication costs of this article were defrayed in part by page charge
payment. This article must therefore be hereby marked ‘‘advertisement’’ in
accordance with 18 U.S.C. §1734 solely to indicate this fact.
© 1998 by The National Academy of Sciences 0027-8424?98?957121-6$2.00?0
PNAS is available online at http:??www.pnas.org.
†To whom reprint requests should be addressed. e-mail: ssherman@
treat all of the other afferents as modulators so that we can
explore what distinguishes the drivers (5). The retinal afferents
pass their receptive field properties to geniculate cells with
only minor changes. These afferents have relatively large axon
terminals that make synaptic contacts on dendrites close to the
cell body of relay cells. They generally form multiple synaptic
zones, often in glomeruli, which are complex synaptic zones
related to the innervation of some relay cells. The retinal
ionotropic receptors. One can argue that some or all of these
features will characterize other drivers in the thalamus, and
possibly also in other parts of the brain, but they are not a
reliable identifier of drivers, in the thalamus or elsewhere.
situation is no clearer. They include glutamatergic afferents from
cortex, cholinergic, serotonergic, and noradrenergic cells from
the brainstem, and histaminergic cells from the hypothalamus.
Other modulatory afferents come from local GABAergic cells
in the adjacent thalamic reticular nucleus. Synaptic terminals
from modulator inputs are relatively small and generally have
single active zones. Modulator inputs from the cortex make
are found primarily proximally, adjacent to retinal inputs, and
those from GABAergic sources make contacts with all parts of
the dendritic arbor (8–11). Modulators outnumber drivers sig-
nificantly. Earlier estimates of ?15–20% for the proportion of
synaptic junctions in the lateral geniculate nucleus of the cat
entering the thalamus contact the inhibitory thalamic reticular
cells, the retinal afferents do not send any branches into the
A first step stab at distinguishing drivers from modulators in
the thalamus, then, suggests that drivers have relatively few,
large axon terminals making contacts close to the cell body
whereas modulators have many more, smaller terminals dom-
inating on peripheral dendrites. Further, drivers may activate
only ionotropic receptors whereas modulators also activate
metabotropic receptors. However, these distinctions are not
diagnostic, and if the classification is to be useful in other brain
regions, further criteria are needed.
Receptive Fields and Cross-Correlograms
In the awake or the lightly anesthetized animal, retinal inputs
dominate geniculate relay cell responses in terms of their recep-
tive field properties. The center?surround receptive field of a
that of afferents from visual cortex or thalamic reticular nucleus.
It is probably unlike those from the brainstem or the hypothal-
receptive fields. Thus, in terms of receptive field properties,
retinal afferents dominate the output of a geniculate relay cell
when it is transmitting visual information, and all of the other
afferents modulate the input?output relationships to control
quantitative features of the relay (4, 5).
Defining drivers as transmitters of receptive field properties
takes us a step further, but it will not serve to distinguish drivers
from modulators in nuclei other than the first order sensory or
some of the higher order thalamic nuclei. First and higher order
nuclei are defined in Fig. 1; the former receive primary (driving)
afferents from ascending pathways whereas the latter receive
primary, driving afferents from pyramidal cells in layer 5 of the
cortex. Thus, in the first order relays for visual or somatosensory
pathways, receptive fields of relay cells come from the ascending
primary afferents (retinal or lemniscal fibers). These receptive
field properties are resistant to cortical lesions or cooling (re-
viewed in refs. 4 and 5). However, in the pulvinar and posterior
thalamic group, which are higher order relays, the relay cell
receptive fields are lost when the afferents from the cortex are
for sensory relays, as noted above, receptive field properties
cannot be used in nuclei like the mediodorsal nucleus or in other
regions where receptive fields have not been shown. It is neces-
thalamic nuclei are shown: a first order (FO) relay on the left and a
higher order (HO) relay on the right. A first order relay receives its
driver inputs on proximal dendrites from subcortical sources via
ascending pathways whereas a higher order relay receives its driver
inputs from cells in cortical layer 5 (see ref. 40). The first order relay
sends a driver input to layer 4 of cortical area A (thick line), and that
same cortical area sends a modulator input (thin line with small
terminals onto distal dendrites of the thalamic relay cell) from layer 6
back to the same first order thalamic nucleus. Cortical area A in turn
sends a driver input from layer 5 to the higher order thalamic relay.
This higher order relay sends its thalamocortical axons (shown as
drivers, on the assumption that all thalamocortical inputs to layer 4 are
drivers, although empirical data are lacking) to cortical area B and
receives a modulator input back from layer 6 of cortical area B. Note
that there are two paths by which cortical area A can influence area
B. One is the transthalamic path, shown by small arrows and drawn as
thick lines indicative of a driver pathway. The other is the direct
corticocortical pathway (a ‘‘feed-forward’’ pathway, as defined by ref.
39), and this is shown by small arrows and as alternating thin and thick
lines to indicate that we do not know whether this (or any other
corticocortical pathway) represents a driver or a modulator input. As
one specific example of such circuitry, we indicate the lateral genic-
ulate nucleus (LGN) as a first order relay innervating area 17 and the
pulvinar as a higher order relay receiving driver input from layer 5 of
area 17 (41) and, in turn, innervating area 18.
Schema to illustrate cortical and thalamic pathways. Two
7122Neurobiology: Sherman and Guillery Proc. Natl. Acad. Sci. USA 95 (1998)
sary to look at other features in the geniculate relay to arrive at
more useful generalizations.
The linkage between retinal afferents and geniculate relay
cells involves more than the structure of their receptive fields.
in relay cells and their retinal afferents, most retinal action
potentials are followed by a single action potential in the relay
cell with a fixed latency (13–15). A cross-correlogram between
the spikes in the retinal axon and the geniculate relay cell
shows the close link of the relay cell and its retinal afferents.
Such a cross-correlogram has a relatively narrow peak with a
latency of several milliseconds and a relatively low, flat
baseline (Fig. 2 A and C; refs. 16 and 17 provide a further
discussion of cross-correlograms).
Here, then, is a relationship based on individual action
potentials that can be applied to relays where receptive fields
cannot be defined. Where, as in the transmission of receptive
fields, critical temporal relationships are a key function of the
driver, any transmission not producing a sharp cross-
correlogram (i.e., a narrow peak arising from a flat baseline)
would lose information along the time axis. An additional
feature for the identification of drivers is the shape of the
cross-correlogram This should be sharp for a driver, providing
One problem about the cross-correlogram arises when a cell
receives input from more than one major source: for example,
axons with adjacent but nonoverlapping receptive fields con-
verging to innervate a single postsynaptic cell as in the
geniculocortical path (18, 19). If the synaptic influences sum
linearly, the postsynaptic receptive field will reflect all of the
postsynaptic cell in relation to action potentials from any of its
inputs or only if several of its inputs are firing concurrently. If
the afferents fired independently of each other, the cross-
correlogram based on one of these afferents could still be fairly
sharp, but the peak would be smaller and the baseline would
be higher and noisier because of the firing of the other
afferents. A cross-correlogram can identify the driver as long
as the number of convergent, independently firing afferents
does not prevent the baseline from obscuring the peak that
each alone would produce. Clearly, convergence among inde-
pendently firing drivers must be limited if a diagnostic, sharp
cross-correlogram is to be produced. Large numbers of con-
vergent inputs could produce a sharp cross-correlogram only
if their firing were highly correlated. This idea is well shown in
geniculocortical connections: Cross-correlograms for genicu-
late and cortical cells indicate a driver input (Fig. 2D) but are
sharper when the relevant geniculate cells fire in synchrony
two neurons, one presynaptic to the other. The cross-correlograms represent the firing of the postsynaptic cells relative to a spike at time zero for
the presynaptic cell. (A) Retinogeniculate cross-correlogram based on spontaneous activity in both the retinal and geniculate neurons. Note the
narrow peak rising out of a flat, low baseline that marks this as a driver connection. Redrawn from Fig. 3A of ref. 14, with permission of the publisher.
(B) Corticogeniculate cross-correlogram based on spontaneous activity in both the layer 6 cell in area 17 and geniculate neuron. Glutamate was
applied to the cortex to enhance the spontaneous firing of the afferent cell. Between the vertical, dashed lines, it is possible to discern a very gradual,
prolonged, and small peak arising from a noisy, high baseline that marks this as a modulator connection. Redrawn from Fig. 2A of ref. 42, with
permission of the publisher. (C and D) Cross-correlograms taken from the same laboratory by using identical techniques for easier comparison.
Both are based on visually driven activity and involve a ‘‘shuffle correction’’ (43), and they are normalized against the firing level of the afferent,
which is why some bins fall below zero. Both represent driver inputs and include another retinogeniculate pair (C) plus a geniculocortical pair (D).
Note the difference in vertical scale, indicating that the retinal input accounts for more postsynaptic spikes in the geniculate cell (C) than does
the geniculate input to the layer 4 cell of the striate cortex (D). Note also that the time represented by these cross-correlograms is much briefer
than that for A and B. Nonetheless, both cross-correlograms have narrow peaks rising from a flat, low baseline, marking them as driver inputs.
Data kindly provided by the authors for replotting. C is redrawn from data of Usrey et al.(15), and D is redrawn from Fig. 2 of Reid and Alonso
Cross-correlograms displaying the difference between drivers and modulators. Each is based on simultaneous recordings in cats from
Neurobiology: Sherman and GuilleryProc. Natl. Acad. Sci. USA 95 (1998)7123
Geniculate cells and other thalamic sensory relay cells have
an unusual property that makes them into poor exemplars for
a general definition of drivers versus modulators. The lateral
geniculate relay appears to be the only relay, from retinal
receptor to higher cortical visual areas, that produces no
significant spatial change in receptive field properties. Genic-
ulate receptive fields are essentially like those of retinal
ganglion cells. Where new receptive field structures are syn-
thesized, as at the geniculocortical synapse, we must expect a
more complex grouping of afferents, and a cross-correlogram
with a sharp peak rising from a flat baseline, like that in Fig.
2 A, C, and D, may not always be apparent. Nonetheless, the
relatively sharp cross-correlogram seen for the geniculocorti-
cal synapse (Fig. 2D) identifies this as a possible driver
identifiable at a synapse beyond the confines of the thalamus.
A cross-correlogram obtained from a concurrent recording
of a layer 6 cortical cell and its target geniculate cell (Fig. 2B)
shows a small peak with a broad foundation on a high baseline.
The difference between this and Fig. 2 A, C, and D is critical
for the distinction between modulators and drivers. It even-
tually may be necessary to quantify the difference, but at
present, there is too little relevant evidence for this.
The difference between driver and modulator cross-
correlograms relates to the fact that there are likely to be many
modulators but few drivers. The modulators may show signif-
icant convergence, and the effect of any one modulator,
possibly one of thousands, may be miniscule. A driver, if it is
to drive, must produce a distinct, measurable effect. The
quantitative relationships of putative drivers to their postsyn-
aptic neurons almost certainly will prove important, and the
extent to which any one driver in any other relay can actually
produce a cross-correlogram as sharp as that in Fig. 2 A and
Because modulator inputs may often far outnumber driver
is sometimes used to gauge its importance in signal transfer,
may instead often indicate modulatory influences. That is, one
cannot simply ascribe the dominant (i.e., driver) input on the
basis of large numbers. These inputs do not act like a demo-
cratic assembly, and understanding the functioning of circuits
requires identifying drivers and modulators by criteria other
than density of inputs.
If other afferents to geniculate cells (e.g., from brainstem or
local GABAergic cells) are all to be regarded as modulators,
late axons outnumber geniculate relay cells by a factor of
10–100 (21). Because corticogeniculate axons branch richly
and are likely to be contacting many geniculate cells, the
amount of convergence is likely to be much greater. Except for
the special and physiologically implausible case of correlated
firing in all corticogeniculate axons (e.g., by electrical stimu-
lation), this convergence would lead to cross-correlograms
with small peaks embedded in a noisy baseline (cf. Fig. 2B).
Tonic and Burst Modes in Thalamic Relay Cells
Although a very sharp cross-correlogram can identify a tha-
lamic driver, this cannot be the complete story. Thalamic relay
cells exhibit two different response modes, ‘‘tonic’’ and
‘‘burst,’’ that necessarily affect the appearance of these cross-
correlograms. Which is present depends on the activation state
of a voltage-dependent Ca2?conductance: When the conduc-
tance is inactive, the cell fires in tonic mode, and when it is
active, the cell fires in burst mode. Details of this Ca2?
conductance can be found elsewhere (5, 22, 23). Switching
between states requires maintaining the appropriate mem-
brane potential for roughly ?100 msec, faster voltage fluctu-
ations being ineffective. Tonic firing results in a steady stream
of unitary action potentials. However, the Ca2?conductance
leads to a large, spike-like depolarization that produces a burst
of conventional action potentials, and these bursts are sepa-
rated by silent periods. This firing pattern characterizes burst
Action potentials in the tonic firing mode result directly
from excitatory postsynaptic potentials, but during bursting,
they result from the Ca2?spike and thus are linked indirectly
to the excitatory postsynaptic potential, which would affect the
cross-correlograms. During tonic firing, an action potential in
the retinal afferent may evoke an action potential in the relay
cell with a tight one-to-one coupling, resulting in a cross-
correlogram with a peak only a few milliseconds across (Fig.
2 A and C). During burst firing, an action potential in the
retinal afferent may activate a Ca2?spike, and the resultant
burst of several action potentials lasts for ?20 msec. Thus, the
coupling between input and output action potentials is no
longer one-to-one. There are as yet no published retinogenicu-
late cross-correlograms for burst firing of the relay cell.
However, we can expect that the peak in the cross-correlogram
would be broader during burst than tonic firing, but it would
still be quite sharp compared with that produced by modula-
We thus further define a driver as an input that produces a
narrow peak in the cross-correlogram against a low, flat
baseline. The width of the peak at half-height would be less
than a few milliseconds during tonic firing and perhaps 25–50
msec during burst firing. This may be regarded as the signature
of a driver, and, as we shall see, inputs to relay cells from
modulators would be expected to produce broader and less
clear peaks in their cross-correlograms. For instance, the
modulator (corticogeniculate) cross-correlogram of Fig. 2B
has a sharp peak (though less sharp than for the drivers of Fig.
2 A, C, and D), but it rises out of a broad, noisy bulge and not
a flat baseline. It is plausible that this broad peak in Fig. 2B
represents activation from the cortex of metabotropic recep-
tors and the narrower peak rising from it reflects activation of
Latencies and Duration of Postsynaptic Responses
There is another factor besides the morphology and cross-
correlogram that is probably important for distinguishing
drivers from modulators. Postsynaptic potentials from drivers
postsynaptic action potential must be tightly coupled in time to
aptic potential would produce too much latency variability for
an evoked action potential and thus would create too much
temporal summation for closely spaced presynaptic action
potentials to have individual postsynaptic signatures. The main
temporal determinant of postsynaptic potentials for relay cells
is their postsynaptic receptors. These come in two general
forms: ionotropic and metabotropic (24–28). Ionotropic re-
ceptors are linked directly to ion channels, and their activation
produces a fast postsynaptic conductance change resulting in
response meets the requirements of a driver and supports a
sharp cross-correlogram. In contrast, metabotropic receptors
are linked indirectly to postsynaptic ion channels via complex
second messenger pathways, and these pathways eventually
produce a postsynaptic potential with a gradual rise and a
prolonged response (hundreds of milliseconds to seconds)§.
This adds considerable temporal scatter to the latency distri-
bution of the postsynaptic action potentials so that cross-
§One can divide ionotropic glutamate receptors into N-methyl-D-
aspartate and non-N-methyl-D-aspartate types. The former has a
slower time course than the latter, but both respond much faster than
do metabotropic glutamate receptors. Some metabotropic receptors
[e.g., those involved in the synapse from rods to on-bipolar cells (29,
30)] respond rapidly, but this seems rare.
7124 Neurobiology: Sherman and GuilleryProc. Natl. Acad. Sci. USA 95 (1998)
correlograms, even without any convergence of inputs, would
have peaks much broader than defined above for drivers. The
slow, prolonged changes in membrane potential produced by
activation of metabotropic receptors are ideal for modulation
because these sustained changes affect excitability of the relay
cell and serve to control activation of the many voltage-
dependent conductances. Also, activation of second messen-
ger pathways via metabotropic receptors can produce other
long-term changes in the relay cell.
Retinal inputs activate only ionotropic receptors in relay
cells (31). All of the other inputs, from cortex, brainstem, and
local GABAergic cells, activate both ionotropic and metabo-
tropic receptors, indicating that the receptor type may be an
important property distinguishing drivers from modulators.
Also, although retinal inputs to relay cells activate only iono-
tropic receptors, these same axons often innervate presynaptic
dendritic terminals of interneurons, where they activate
metabotropic receptors (32). It thus is likely that individual
axons or pathways cannot be classified as drivers or modula-
tors. Rather, the classification must refer to the synaptic
contacts established by particular axons at particular sites. For
be a modulator for an interneuron. Further, interneurons in
the thalamus have two distinguishable parts—the postsynaptic
somadendritic surface and the dendritic appendages that
to differing functions of the interneuron (33). Therefore, the
action of a retinal afferent on an interneuron may depend on
where on the interneuron the synapse is made.
The importance of synaptic latency and duration in the
distinction between drivers and modulators may not apply to
functions or peripheral portions of olfactory, gustatory, or
nociceptive pathways). Only if the precise temporal patterning
of action potentials proves important to the neural code (e.g.,
ref. 34) would one expect to be able to recognize driver inputs
on the criteria we have suggested. The issue is tantalizingly
Key Differences Between Drivers and Modulators
We can now summarize our definition for drivers versus
modulators innervating relay cells of the lateral geniculate
nucleus. The key feature is the nature of cross-correlograms
(see Fig. 2). For driver inputs, they have a sharp peak rising out
of a low, flat baseline. For modulators, they would have a
broad, gradual peak, if any, set against a noisy baseline. The
sharp cross-correlogram conforms to the need of the relay cell
to convey faithfully the signal of its driver input to the cortex,
and a lack of such cross-correlograms for the modulator inputs
means that the signals they carry are not relayed faithfully to
products of a variety of circuit and cellular properties that
characterize the contacts made by these inputs. Thus, retino-
geniculate inputs show little convergence and make relatively
few contacts, which are near cell bodies, and they activate only
ionotropic receptors. Modulating inputs make more contacts,
which are generally further from the cell body and can show
considerable convergence; they all activate metabotropic re-
ceptors in addition to ionotropic receptors.
There are several characteristic features of drivers in the
lateral geniculate nucleus that distinguish them from modu-
lators. These characteristics include their fine structural ap-
pearance, their synaptic relationships, their degree of conver-
gence on relay cells, their relatively small number relative to
the modulators, their absence of connections to the thalamic
reticular nucleus, their transmitter and receptor characteris-
tics, and the nature of the cross-correlogram they produce
when stimulated. Many of these properties are seen in all first
order thalamic nuclei where the drivers can be identified.
Other properties remain to be studied in these other nuclei,
and some, such as the cross-correlograms, have not been
defined empirically for modulators in any other thalamic
nucleus. This summary for the lateral geniculate nucleus leads
to a number of clear questions about other thalamic nuclei,
where current knowledge of their functional organization does
not make clear which is the driver, although in some, the
morphological evidence has provided a useful clue. In extend-
ing the definition of drivers and modulators to other cerebral
centers, such as the cortex, the characteristics we have iden-
tified may prove to be relevant.
The Sleeping Thalamus
There are two different functional states for thalamic nuclei.
One involves the active, dynamic relay of driver activity to the
cortex and characterizes the waking state. The other involves
cells no longer respond to driver inputs and are seen often
during slow wave sleep. The switch from the waking to the
sleeping thalamus effectively disengages the drivers.
Cells of the thalamic reticular nucleus operate this switch, a
switch that seems to depend on the same sort of voltage-
dependent Ca2?conductance and bursting described above for
relay cells, although the bursts in the reticular cells last longer.
This leads to rhythmic bursting both in reticular cells and relay
cells, and modeling studies suggest that the interconnections
between reticular and relay cells and among reticular cells
serve to create and maintain synchrony (for details of this
behavior, see refs. 23, 35–37).
During this synchronized bursting, input from the thalamic
reticular nucleus dominates relay cells, and excitatory postsyn-
aptic potentials generated by driver inputs are insufficient to
break the stranglehold of reticular inputs on thalamic relay cell
responses, which is why effective thalamic relay functions are
blocked during slow wave sleep. Note that the relay is disen-
gaged not by silencing relay cells but, rather, by forcing these
cells to burst rhythmically and independently of driver input.
Thus, instead of silence, the cortex receives a positive signal
that the relay is disrupted. Silence alone would be ambiguous;
the absence of a visual stimulus could not be distinguished
from the absence of an effective relay of the stimulus. The
rhythmic bursting, by signaling the ‘‘no-relay’’ alternative,
avoids this ambiguity.
If we apply our cross-correlogram criterion, reticular input
to relay cells looks like a driver input during slow wave sleep
because there would be a fairly sharp peak with little baseline
in the cross-correlogram during the synchronized, rhythmic
bursting. One could argue that the thalamic relay cells are
responding to a ‘‘message’’ sent by the reticular cells. However,
this form of driving occurs because of the special relationships
that produce the highly correlated firing of the synchronized
reticular cells. Without such a correlation, there would be no
driving. It seems that the action of these afferents has to be
distinguished from drivers considered earlier but that it cannot
be regarded as modulation. We suggest that this action be
treated as a ‘‘disrupting’’ action that is distinct from driving
and modulation and that may be special to the thalamic
reticular nucleus and its thalamic connections.
The distinction between drivers and modulators (and disrupt-
ers) is important for understanding thalamic relays, and it may
prove particularly critical for defining the functional organi-
zation of thalamic nuclei where this cannot be studied in terms
of readily defined receptive field properties. Possibly, the
distinction can be applied much more broadly to the cerebral
cortex, as suggested by Crick and Koch (1), and possibly to
other cerebral centers as well. One would expect thalamocor-
Neurobiology: Sherman and Guillery Proc. Natl. Acad. Sci. USA 95 (1998)7125
tical axons, especially those going to cortical layer 4 and
possibly all of them, to be drivers, whether from a sensory relay
nucleus or a higher order nucleus like the pulvinar. The
classification of corticocortical pathways largely is untested in
terms of the criteria proposed here.
Current views of corticocortical communication (see refs.
38, 39) could be influenced significantly by experiments that
demonstrated which connections are drivers and which are
modulators. One distinct and intriguing possibility is that the
major source of a functional drive for corticocortical commu-
nication actually goes through the thalamus and derives from
cells in layer 5 of one cortical area, which then provide a driver
input to relay cells in a higher order thalamic nucleus such as
the pulvinar (see Fig. 1). These thalamic cells then, in turn,
send their axons as drivers to layer 4 of another cortical area.
An extreme corollary might be that most direct corticocortical
pathways serve as modulators, which would mean that infor-
mation flowing from one cortical area to another, by passing
through a thalamic relay, would be subject to the same control
of information flow as exists for information coming into
cortex from subcortical sources. The thalamus thus serves as
a ‘‘gate’’ not only in the control of information to particular
cortical areas about sensory events but also in the control of
information passed to other cortical areas from the descending
outputs emanating from layer 5.
If the categorization of inputs as driver or modulator (or
possibly as disrupter) is to have an agreed significance, or if
one is to determine whether disrupters are unique to the
thalamus or also can be identified in other parts of the brain,
then it becomes important that experimental criteria for
tried to provide an introduction to the problems that need to
be addressed if a classification that has wide applicability is to
be used. Possibly, it will prove that there are too many problem
service outside the thalamus. The observations remain to be
We thank Paul Adams, Ed Bartlett, Sherry Feig, Lew Haberly, Clay
Reid, Phil Smith, Dan Uhlrich, and Martin Usrey for many helpful
comments on an earlier draft of this manuscript. Sherry Feig also
produced Fig. 1. R.W.G. has been supported by U.S. Public Health
Service Grant EY11494, and S.M.S. has been supported by U.S. Public
Health Service Grant EY03038.
Crick, F. & Koch, C. (1998) Nature (London) 391, 245–250.
Feig, S. & Harting, J. K. (1998) J. Comp. Neurol. 395, 281–295.
Rockland, K. S. (1997) J. Comp. Neurol. 390, 515–536.
Sherman, S. M. & Koch, C. (1998) in The Synaptic Organization
of the Brain, ed. Shepherd, G. M. (Oxford Univ. Press, London)
Vol. 4, 289–328.
Sherman, S. M. & Guillery, R. W. (1996) J. Neurophysiol. 76,
Bender, D. B. (1983) Brain Res. 279, 258–261.
Diamond, M. E., Armstrong-James, M., Budway, M. J. & Ebner,
F. F. (1992) J. Comp. Neurol. 319, 66–84.
8. Erisir, A., Van Horn, S. C., Bickford, M. E. & Sherman, S. M.
(1997) J. Comp. Neurol. 377, 535–549.
Guillery, R. W. (1969) Z. Zellforsch. Mikrosk. Anat. 96, 39–48.
Guillery, R. W. (1969) Z. Zellforsch. Mikrosk. Anat. 96, 1–38.
Wilson, J. R., Friedlander, M. J. & Sherman, S. M. (1984) Proc.
Roy. Soc. London Ser. B. 221, 411–436.
Van Horn, S. C., Erisir, A. & Sherman, S. M. (1997) Soc.
Neurosci. Abstr. 23, 171.
Cleland, B. G., Dubin, M. W. & Levick, W. R. (1971) J. Physiol.
(London) 217, 473–496.
Mastronarde, D. N. (1987) J. Neurophysiol. 57, 357–380.
Usrey, W. M., Reppas, J. B. & Reid, R. C. (1998) Soc. Neurosci.
Abstr. 23, 170.
Nelson, J. I., Salin, P. A., Munk, M. H. J., Arzi, M. & Bullier, J.
(1992) Visual Neurosci. 9, 21–37.
Fetz, E., Toyama, K. & Smith. W. (1991) in Cerebral Cortex, eds.
Peters, A. & Jones, E. G. (Plenum, New York), Vol. 9, 1–47.
Ferster, D., Chung, S. & Wheat, H. (1996) Nature (London) 380,
Hubel, D. H. & Wiesel, T. N. (1962) J. Physiol. 160, 106–154.
Alonso, J. M., Usrey, W. M. & Reid, R. C. (1996) Nature
(London) 383, 815–819.
Sherman, S. M. & Koch, C. (1986) Exp. Brain Res. 63, 1–20.
Sherman, S. M. (1995) Visual Neurosci. 13, 205–213.
Steriade, M. & Llin?s, R. (1988) Physiol. Rev. 68, 649–742.
Conn, P. J. & Pin, J. P. (1997) Annu. Rev. Pharmacol. Toxicol. 37,
Godwin, D. W., Vaughan, J. W. & Sherman, S. M. (1996)
J. Neurophysiol. 76, 1800–1816.
Godwin, D. W., Van Horn, S. C., Erisir, A., Sesma, M., Romano,
C. & Sherman, S. M. (1996) J. Neurosci. 16, 8181–8192.
Pin, J. P. & Duvoisin, R. (1995) Neuropharmacology 34, 1–26.
Recasens, M. & Vignes, M. (1995) Ann. N.Y. Acad. Sci. 757,
Shiells, R. (1994) Curr. Biol. 4, 917–918.
Vardi, N. & Morigiwa, K. (1997) Visual Neurosci. 14, 789–794.
McCormick, D. A. & Von Krosigk, M. (1992) Proc. Natl. Acad.
Sci.USA 89, 2774–2778.
Bloomfield, S. A. & Sherman, S. M. (1989) Proc. Natl. Acad.
Sci.USA 86, 3911–3914.
Mainen, Z. F. & Sejnowski, T. J. (1995) Science 268, 1503–1506.
Golomb, D., Wang, X.-J. & Rinzel, J. (1994) J. Neurophysiol. 72,
Lytton, W. W., Destexhe, A. & Sejnowski, T. J. (1996) Neurosci.
Steriade, M., McCormick, D. A. & Sejnowski, T. J. (1993) Science
Bullier, J., Schall, J. D. & Morel, A. (1996) Behav. Brain Res. 76,
Felleman, D. J. & Van Essen, D. C. (1991) Cereb. Cortex 1, 1–47.
Guillery, R. W. (1995) J. Anat. 187, 583–592.
Tsumoto, T., Creutzfeldt, O. D. & Legendy, C.R. (1978) Exp.
Brain Res. 32, 345–364.
Perkel, D. H., Gerstein, G. L. & Moore, G. P. (1967) Biophys.J.
Reid, R. C. & Alonso, J. M. (1996) Curr. Opin. Neurobiol. 6,
7126 Neurobiology: Sherman and GuilleryProc. Natl. Acad. Sci. USA 95 (1998)