Download Intracellular and computational evidence for a

Current Opinion in Neurobiology, in press, 2011.
Special issue “Networks, Circuits and Computation”, edited by Feldman D., Feller M. and Dayan P.
Intracellular and computational evidence for a dominant role
of internal network activity in cortical computations
Alain Destexhe
Unité de Neurosciences, Information et Complexité (UNIC),
CNRS, 91198 Gif-sur-Yvette, France
Correspondence: Tel: 33-1-6982-3435, Fax: 33-1-6982-3427; Email: [email protected]
Keywords: Spontaneous activity, Intrinsic activity, Spike-triggered average,
Conductances, Network state, State-dependent computations, Modeling
Abstract
The mammalian cerebral cortex is characterized by intense spontaneous activity, depending on brain
region, age and behavioral state. Classically, the cortex is considered as being driven by the senses, a
paradigm which corresponds well to experiments in quiescent or deeply anesthetized states. In awake
animals, however, the spontaneous activity cannot be considered as “background noise”, but is of comparable – or even higher – amplitude than evoked sensory responses. Recent evidence suggests that this
internal activity is not only dominant, but it shares many properties with the responses to natural sensory
inputs, suggesting that the spontaneous activity is not independent of the sensory input. Such evidences
are reviewed here, with an emphasis on intracellular and computational aspects. Statistical measures,
such as the spike-triggered average of synaptic conductances, show that the impact of internal network
state on spiking activity is major in awake animals. Thus, cortical activity cannot be considered as being
driven by the senses, but sensory inputs rather seem to modulate and modify the internal dynamics of
cerebral cortex. This view offers an attractive interpretation not only of dreaming activity (absence of
sensory input), but also of several mental disorders.
1
The awake and conscious brain of adult mammals is characterized by ample spontaneous activity. Intracellular recordings of cortical neurons in awake adult cats [1, 2, 3], monkey [4] or mice [5] show that the
neurons are always active and rarely exhibit periods of quiescence (reviewed in [6]). Indeed, the resting
membrane potential of cortical neurons typically cannot be observed in vivo, except in some cases of deep
anesthesia or under the action of drugs [7]. It was shown that in the active regime, cortical neurons are
subject to large amounts of fluctuations, often called “synaptic noise”. This activity is major, as its total conductance can be several-fold larger than the resting membrane conductance, a situation called the
“high-conductance state”, which may have many important consequences on the integrative properties of
cortical neurons (reviewed in [8, 9]).
In awake subjects, the electroencephalogram (EEG) is typically of low amplitude, fast frequency and is very
irregular, a pattern which is called “activated state” or “desynchronized EEG”. Multiple unit recordings
in the aroused brain display irregular firing with very low levels of synchrony, which contrasts with the
synchronized and slow oscillatory activities seen during slow-wave sleep [6, 10, 11, 12, 13]. Because it is
during this apparently noisy regime that the main computational tasks are performed, understanding this
type of stochastic network state is crucial [14].
This strong spontaneous activity is classically considered as “noise” independent of the input signal. However, experimental and modeling evidence suggest that it is significantly structured and is different from
independent additive noise [15]. The present article reviews such evidences, with an emphasis on intracellular and modeling results, as well as their combination.
The intrinsic activity of the brain
The first proposal that neurons are not passive relays being driven by external inputs dates back to the early
20th century with the Belgian electrophysiologist Frederic Bremer [16, 17]. He proposed that neurons
generate intrinsic and self-sustained activity under the form of intrinsic oscillatory properties. The current
thinking at the time was that oscillatory activity arises from circulating waves of activity, a theory called the
“circus movement theory”. Bremer was an opponent to this theory, and he proposed instead that neurons
can display intrinsically-generated oscillatory activity, and that such oscillators synchronize into population
oscillations, two concepts which are well known today. The presence of such intrinsic properties was later
demonstrated and characterized, in invertebrate preparations of pattern generators [18, 19], and in various
parts of the central nervous system [20].
Looking into the morphological details, it becomes clear that the brain is not wired to be driven by sensory
inputs. In cerebral cortex, the synapses arising from thalamocortical fibers constitute a small minority (a
few percent), even in Layer IV, the main recipient of the thalamic input. The vast majority of synapses arise
from cortico-cortical input, either from local axon collaterals or from long-range cortico-cortical fibers [21].
In the thalamus, the synapses coming from afferent sensory fibers are also less numerous compared to the
synapses arising from cortical axons [22]. So here also, the cortex is the main afferent to the thalamus,
which is difficult to reconcile with the idea of the thalamus being a simple “relay” of sensory information
en route to cortex. It rather seems that the thalamocortical system is wired to favor internal processing.
At a point of view of brain dynamics, it is important to note that the brain is not silent but displays considerable spontaneous activity, independent of the sensory input. For example, during slow-wave sleep or
rapid-eye movement (REM) sleep, the brain is as active as during wakefulness and despite the fact that
sensory inputs are not processed [6]. Strikingly, there is little electrophysiological difference between the
activity of neurons and local field potentials between REM sleep, the “Up states” of slow-wave sleep and
2
wakefulness [13, 23, 24].
Based on such observations, Llinas & Paré [23] proposed that most the activity of the adult brain is intrinsically generated, and is only influenced by the senses rather than being driven by it. With no sensory input,
the intrinsic activity is left alone, which corresponds to REM sleep with dreaming experience. Thus, the
awake brain activity is seen as a dream modulated by the senses [23].
Following this seminal paper, several studies provided strong support to this view. First, it was shown
that the spontaneous activity of the brain is not simply “noise” but is much more structured. For instance,
in the visual cortex of ferrets, it was demonstrated that the spontaneous activity – largely absent in very
young animals – becomes progressively more intense and structured with age [25]. Moreover, in the adult,
the spontaneous activity was only slightly modified by the visual input. Interestingly, analyzing the spike
patterns produced in response to natural images, with those produced by spontaneous activity revealed
no particular resemblance in young animals (Fig. 1A), but they were strikingly similar in adults [25, 26]
(Fig. 1B). Similar observations were made in other structures, such as the auditory and somatosensory
cortex of rats [27], where spontaneous spike patterns were found to be very similar to evoked responses.
In the primary visual cortex of anesthetized cats using voltage-sensitive dye imaging, the visual responses
were not only of comparable amplitude as the spontaneous activity, but the spatiotemporal activity patterns
were also very similar [28].
Thus, it seems that one cannot distinguish between spontaneous activity and the activity evoked by natural
stimuli, which shows that most of this activity is internally generated, and that the net effect of sensory
input is small. In other words, these results suggest that sensory-evoked activity represents a modulation of
ongoing cortical spontaneous activity [25], very similar in spirit to the Llinas & Paré [23] proposal.
Is there quiescence in the absence of input?
Although such findings offer a nice perspective to explain population recordings, they are not consistent
with all of the available experimental data. In particular, in the primary visual cortex (V1), a large number
of studies have demonstrated clear visual responses and selectivity of neurons to features such as orientation, direction and contract [29]. Not much spontaneous activity seems to be present in such single-cell
experiments (see also [30]). Very low levels of spontaneous activity were also reported in experiments
using patch electrodes in vivo, in motor and somatosensory cortex [31, 32].
These seemingly contradicting observations can be reconciled based on three observations. First, that
the level of spontaneous activity is highly dependent on the state of the animal. Some of the above experiments were done under anesthesia, which may considerably limit spontaneous activity [7]. Because
intense spontaneous activity is responsible for setting a “high-conductance state” in cortical neurons [8, 9],
some anesthetized states can artificially augment neuronal responsiveness by setting the membrane of cortical neurons into lower conductance states, making the membrane more excitable. With some anesthetics,
however, it is possible to obtain conductance states very close to that of wakefulness [8].
A second factor, already mentioned above, is that the level of spontaneous activity considerably depends
on the age of the animal [25]. Newborn or very young ferrets display very modest amount of spontaneous
activity, while it is much more prominent in the adult, where spontaneous activity is more intense than
visual responses [25]. Unfortunately, the growth of spontaneous activity with age is paralleled with axon
myelinization in cerebral cortex. Since myelinization makes patch recordings to be increasingly difficult
with age, it is possible that the relatively young age of the animals explains the low levels of spontaneous
3
activity observed in some of the patch-recording experiments.
A third important parameter is that spontaneous activity may be specific to each layer of cerebral cortex.
Superficial layers display very sparse firing, while deep layers have more profuse spontaneous activity [33].
Whole-cell recordings are usually made in superficial layers, which may also explain the low level of
spontaneous activity observed using this technique.
Taken together, these results suggest that a fully active network with prominent spontaneous activity is
totally relevant to cortical information processing. In such conditions, the neurons are presumably in highconductance states [8], as indicated by conductance measurements from intracellular recordings in awake
cats [34].
Intracellular and computational evidence for dominant internal dynamics
Another set of evidence is provided by model-based analyses of intracellular recordings in vivo. Figure 2A
shows examples of intracellular recordings of cat V1 neurons during spontaneous activity (SA) and the
presentation of natural images (NI) [35] (original data from [36]). To measure statistical similarity, the
frequency scaling exponent was computed from the power spectrum of the signals (Fig. 2B), yielding
exponent values for different stimulus conditions. Remarkably, the exponents were very close between NI
and SA conditions (Fig. 2C), but not for other stimulus conditions (not shown). This analysis shows that
the spontaneous activity of V1 has similar subthreshold statistics as during natural images, a result with is
in full agreement with findings on awake animals [25].
In an attempt to quantify the conductance variations underlying such effects, an efficient method is to compute the spike-triggered average (STA) of the synaptic conductances during different conditions. In models
receiving in vivo–like background synaptic inputs at excitatory and inhibitory synapses, the conductance
STAs revealed a specific pattern, with an increase of excitatory conductance and a decrease of inhibitory
conductance (Fig. 3A). This pattern was also replicated in dynamic-clamp experiments in vitro [37, 38]
(Fig. 3B).
A specific procedure was designed to estimate STAs from intracellular recordings in vivo, where the conductances are not known (unlike models and dynamic-clamp experiments where the conductances are preset). The estimation procedure was in two steps. First, the total conductances and their variances are
estimated using a method called “VmD method” [39]. Second, a maximum likelihood procedure is used
to estimate the STA of conductances [37]. These procedures were tested in models and in dynamic-clamp
experiments in vitro [37, 39]. Application to intracellular recordings in awake and naturally sleeping cats
revealed the same pattern of conductance STA with a decrease of inhibition [34] (Fig. 3C).
This pattern of conductance with excitation increase and inhibition decrease is opposite to what was observed in several sensory systems. A feed-forward drive (such as sensory inputs) would predict an increase
of excitation closely associated to an increase of inhibition. This is indeed what was observed in many
instances of evoked responses during sensory processing [40, 41, 42, 43, 44]. Figure 3D shows an example of such concerted increase of excitation and inhibition in rat barrel cortex [43]. Note, however, that
one study reported a considerable cell-to-cell diversity in the conductance variations evoked by synaptic
inputs; among them, some of the cells displayed a mirror image for excitatory and inhibitory conductance
variations (see [41]).
Computational models can be used to further show that this pattern of conductance STA corresponds to
spontaneous network activity. Sparsely connected networks of excitatory and inhibitory neurons can dis4
play “asynchronous irregular” states [45], which noisy-like properties are very close to recordings in awake
animals [46]. One such state is illustrated in Fig. 4A-B (model from [47]). Calculating conductances in
this model revealed conductance STAs with increase of excitation and decrease of inhibition (Fig. 4C).
These results show that the same pattern of opposite conductance variations is seen in different cases, when
the neuron is driven by stochastic synaptic activity in models (Fig. 3A) or dynamic-clamp experiments
(Fig. 3B), or in the awake animal (Fig. 3C). This pattern is also seen in models of self-generated network
activity (Fig. 4C). However, a different pattern of concerted conductance variations is seen in many sensory
systems (Fig. 3D).
Proposed scheme to account for the different experiments
To account for the disparity of the above results, a simple model of a neuron receiving two input sources was
simulated. First an “intrinsic activity” consisting of stochastic release at excitatory and inhibitory synapses,
and second, an “external input” consisting of a controlled stimulation of an independent set of excitatory
and inhibitory synapses (see scheme in Fig. 5). For weak external inputs, the activity is dominated by
intrinsic activity and one recovers the pattern of opposite conductance variations (Fig. 5A). For medium
input strength, an additional component appears in the conductance STAs (Fig. 5B). For strong inputs,
the latter component dominates, and the STAs consist of concerted variations of excitation and inhibition
(Fig. 5C).
These simulations suggest a coherent picture to explain all the above results. If the neuron’s spiking is
mostly dominated by intrinsic network activity, then the conductance STAs display opposite variations of
excitatory and inhibitory conductances. This is the case for neurons receiving random inputs (either in models or in vitro; see Fig 3A-B) or in a self-generated spontaneous activity state in a network (Fig. 4C), as well
as in awake animals [34] (Fig. 3C). Conversely, when the external input is strong compared to spontaneous
activity, the conductance STA consists of a concerted increase of excitatory and inhibitory conductances
(Fig. 5C), similar to many observations during sensory responses [40, 41, 42, 43] (see Fig. 3D).
Interestingly, the fact that the same opposite pattern is found in awake animals, suggests that the spiking in
the awake cat is also determined mainly by spontaneous activity, while the direct effect of afferent activity
is not visible. This is in full agreement with the results reviewed in the first part of the paper, where the
statistical similarity between spontaneous and evoked activity also suggests that the direct effect of afferent
activity is either negligible or dominated by the “intrinsic” activity of the network.
Conclusions
In summary, the intracellular and computational results reviewed here collectively suggest that the spiketriggered conductance patterns is a very powerful way to determine whether a system is dominated by its
internal activity, or is driven by afferent activity, as summarized in Fig. 5. This approach makes a number
of predictions, which are briefly discussed here.
First, it suggests that in many sensory systems, the measurements reporting a concerted conductance increase can be explained either because the level of spontaneous activity was very low (which may be in
part because of anesthesia), or because the input was unusually strong. The fact that a large diversity of
conductance combinations was observed in some cases [41], including concerted and opposite conductance
variations, suggests that part of these observations is due to different network states with different levels
5
of spontaneous activity in the system. Such conductance measurements should be corroborated with an
analysis of the level of spontaneous activity to verify what component of the conductance variations are
really due to the input.
Second, the finding that most spikes in awake cat association cortex are due to spontaneous activity and not
to afferent volleys indeed suggests that most of the computing of this brain area is done internally, while
the processing of afferent inputs is limited. Whether this is specific to that particular area, or constitutes a
general principle of brain computation, should be addressed by future work.
Finally, the results reviewed here are in total agreement with the original proposal of Llinas and Paré [23]
that most of the brain activity is internal, intrinsic and ongoing, while the effect of sensory input is a
modulation of this internal activity. It is also in agreement with the similarity between activity patterns
during natural viewing and spontaneous patterns [25, 35] (see Figs. 1 and 2). The fact that most of the
spikes in the awake cat are due to spontaneous activity also supports these findings.
Note that the fact that the sensory drive triggers internal dynamics is not necessarily inconsistent with the
low levels of spontaneous activity reported in some experiments (such as [30]). Because of the dense
recurrent connectivity in cortex, such low-level spontaneous activity can still be associated with highly
active subthreshold membrane potential dynamics, as indeed found intracellularly in awake animals [5].
Previous computational approaches have modeled the interaction between spontaneous activity and sensory inputs based on the data available at the time [48, 49, 50]. It was also proposed that the spontaneous
activity of cerebral cortex exhibits several distinct states, similar to a finite-state machine [51]. No electrophysiological evidence for such multiple states have been provided; it rather seems that spontaneous
activity consists of a continuously changing state, modulated by the sensory input. Spontaneous activity
may also represent a form of prediction of the sensory input (reviewed in [15]). Modeling such interactions
represent considerable challenges for computational studies.
Thus, the proposal that “wakefulness is a dream modulated by the senses” [23] still holds, and seems in
agreement with electrophysiological and conductance measurements. This view offers original ways to
interpret some pathologies which may be due to disorders of intrinsic activity, which could give rise to
various symptoms of altered processing, such as day dreaming, inattentional blindness, mind wandering or
schizophrenia. Further work should carefully integrate and measure spontaneous activity, and view it not
as “noise”, but as an integral component of brain computations.
Acknowledgments
Thanks to Thierry Bal, Olivier Marre, Cyril Monier, Zuzanna Piwkowska, Igor Timofeev and Yves Frégnac
for sharing experimental data, and all members of the UNIC for continuing support. This work has been
supported by the CNRS, the Agence Nationale de la Recherche (ANR HR-Cortex) and the European Community (FET grants FACETS FP6-015879, BRAINSCALES FP7-269921).
References
[1] Woody CD, Gruen E. (1978) Characterization of electrophysiological properties of intracellularly recorded
neurons in the neocortex of awake cats: a comparison of the response to injected current in spike overshoot and
undershoot neurons. Brain Res 158, 343-357.
6
[2] * Baranyi A, Szente MB, Woody CD. (1993) Electrophysiological characterization of different types of neurons
recorded in vivo in the motor cortex of the cat. II. Membrane parameters, action potentials, current-induced
voltage responses and electrotonic structures. J Neurophysiol 69, 1865-1879.
This is the first detailed electrophysiological characterization of neurons recorded intracellularly in awake
animals.
[3] * Steriade M, Timofeev I, Grenier F. (2001) Natural waking and sleep states: a view from inside neocortical
neurons. J. Neurophysiol. 85, 1969-1985.
First intracellular recording during natural sleep, showing that REM and awake states are almost indistinguishable as viewed from “inside”.
[4] Matsumura M, Cope T and Fetz EE. (1988) Sustained excitatory synaptic input to motor cortex neurons in
awake animals revealed by intracellular recording of membrane potentials. Exp Brain Res. 70, 463-469.
[5] Poulet JF and Petersen CC. (2008) Internal brain state regulates membrane potential synchrony in barrel cortex
of behaving mice. Nature 454, 881-885.
[6] Steriade, M. (2003) Neuronal Substrates of Sleep and Epilepsy (Cambridge University Press, Cambridge UK).
[7] * Paré D, Shink E, Gaudreau H, Destexhe A, Lang EJ. (1998) Impact of spontaneous synaptic activity on the
resting properties of cat neocortical neurons in vivo. J Neurophysiol. 79, 1450-1460.
First characterization of the impact of spontaneous network activity on the membrane properties of neurons.
Intracellular recordings were made before and after suppression of network activity, which revealed the “true”
resting state of the neurons. It was found that spontaneous activity has a major impact on the conductance state
of the membrane.
[8] Destexhe A, Rudolph M and Paré D. (2003) The high-conductance state of neocortical neurons in vivo. Nature
Reviews Neurosci 4, 739-751.
[9] Destexhe, A. (2007) High-conductance state. Scholarpedia 2: 1341.
http://www.scholarpedia.org/article/High-Conductance State
[10] Hubel D (1959) Single-unit activity in striate cortex of unrestrained cats. J. Physiol. 147, 226-238.
[11] Evarts EV (1964) Temporal patterns of discharge of pyramidal tract neurons during sleep and waking in the
monkey. J. Neurophysiol. 27, 152-171.
[12] Steriade M, Deschênes M and Oakson G. (1974) Inhibitory processes and interneuronal apparatus in motor
cortex during sleep and waking. I. Background firing and responsiveness of pyramidal tract neurons and interneurons. J. Neurophysiol. 37, 1065-1092.
[13] Destexhe A, Contreras D and Steriade M. (1999) Spatiotemporal analysis of local field potentials and unit
discharges in cat cerebral cortex during natural wake and sleep states. J. Neurosci. 19: 4595-4608.
[14] Destexhe A and Contreras D. (2006) Neuronal computations with stochastic network states. Science 314, 85-90.
[15] Ringach DL. (2009) Spontaneous and driven cortical activity: implications for computation. Curr. Opin. Neurobiol. 19, 439-444.
[16] ** Bremer, F. (1938) L’activité électrique de l’écorce cérébrale. Actualités Scientifiques et Industrielles 658,
3-46.
Article (in French) where the author was the first to propose that rhythmical brain activity is due to intrinsic
properties of neurons (intrinsic oscillations in this case) and that these intrinsically active neurons synchronize
into population oscillations, two concepts which are well known today.
7
[17] Bremer F. (1958) Cerebral and cerebellar potentials. Physiol. Reviews 38, 357-388.
[18] Selverston, A.I. (Ed.) (1985) Model Neural Networks and Behavior. New York: Plenum Press.
[19] Getting, P.A. (1989) Emerging principles governing the operation of neuronal networks. Annual Rev. Neurosci.
12, 185-204.
[20] Llinás RR (1988) The intrinsic electrophysiological properties of mammalian neurons: a new insight into CNS
function. Science 242, 1654-1664.
[21] Braitenberg, V. and Schüz, A. (1998) Cortex: statistics and geometry of neuronal connectivity (2nd edition).
Springer-Verlag, Berlin.
[22] Sherman, S. M. and Guillery, R. W. (2001) Exploring the Thalamus. Academic Press, New York.
[23] ** Llinas, R.R. and Paré, D. (1991) Of dreaming and wakefulness. Neuroscience 44, 521-535.
This paper is a remarkably forward-looking essay proposing for the first time that brain activity is always ongoing due to thalamocortical loops, and that this ongoing activity is associated to conscious processes. Dreams
occur when this ongoing activity is left alone, with no sensory input. The authors propose that wakefulness is
functionally identical to dreaming activity, with the sole difference that sensory inputs are present, so wakefulness is seen as a “dream modulated by the senses”.
[24] Destexhe A., Hughes SW, Rudolph M and Crunelli V. (2007) Are corticothalamic ‘up’ states fragments of
wakefulness? Trends Neurosci. 30: 334-342.
[25] ** Fiser J, Chiu C and Weliky M (2004) Small modulation of ongoing cortical dynamics by sensory input
during natural vision. Nature 431, 573–578.
An important paper showing that, in awake ferrets, the spike responses of neuron populations are strikingly
similar in spontaneous activity and when the animal views natural scenes. However, this is only true for the
adult, as the two activities are markedly different in young animals.
[26] Berkes P, Orban G, Lengyel M and Fiser J. Spontaneous cortical activity reveals hallmarks of an optimal internal
model of the environment. Science 331, 83-87.
[27] Luczak A, Barthó P and Harris KD. (2009) Spontaneous events outline the realm of possible sensory responses
in neocortical populations. Neuron 62, 413-425.
[28] Kenet, T., Bibitchkov, D., Tsodyks, M., Grinvald, A. and Arieli, A. (2003) Spontaneously emerging cortical
representations of visual attributes, Nature 425, 954-956.
[29] Hubel, D. and Wiesel, T.N. (1962) Receptive fields, binocular interaction and functional architecture in the cat’s
visual cortex. J. Physiol. 160, 106-154.
[30] Niell CM and Stryker MP. (2010) Modulation of visual responses by behavioral state in mouse visual cortex.
Neuron 65, 472-479.
[31] Waters J, Helmchen F. (2006) Background synaptic activity is sparse in neocortex. J. Neurosci. 26, 8267-8277.
[32] Wolfe J, Houweling AR and Brecht M. (2010) Sparse and powerful cortical spikes. Curr. Opin. Neurobiol. 20,
306-312.
[33] * Sakata S and Harris KD. (2009) Laminar structure of spontaneous and sensory-evoked population activity in
auditory cortex. Neuron 64, 404-418.
Nice study showing different multi-electrode recording configurations revealing fundamental differences between superficial layers, characterized by very sparse firing, and deep layers displaying more sustained and
strong spontaneous activity.
8
[34] Rudolph M, Pospischil M, Timofeev I and Destexhe A. (2007) Inhibition determines membrane potential dynamics and controls action potential generation in awake and sleeping cat cortex. J. Neurosci. 27, 5280-5290.
[35] El Boustani, S., Marre, O., Béhuret, S., Baudot, P., Yger, P., Bal, T., Destexhe, A. and Frégnac, Y. (2009)
Network-state modulation of power-law frequency scaling in visual cortical neurons. PLoS Computational
Biology 5, e1000519.
[36] Baudot P, Levy M, Monier C, Chavane F, René A and Frégnac Y. (2004) Time-coding, low noise Vm attractors,
and trial-to-trial spiking reproducibility during natural scene viewing in v1 cortex. Soc. Neurosci. Abstracts
948: 12.
[37] * Pospischil M, Piwkowska Z, Rudolph M, Bal T and Destexhe A. (2007) Calculating event-triggered average
synaptic conductances from the membrane potential. J Neurophysiol 97, 2544-2552.
First method ever proposed to calculate the spike-triggered average of synaptic conductances from the sole
measurement of the membrane potential. The method assumes that spontaneous cortical activity is well described by stochastic processes, and is tested using dynamic-clamp experiments in vitro.
[38] Piwkowska Z, Pospischil M, Brette R, Sliwa J, Rudolph-Lilith M, Bal T and Destexhe A. (2008) Characterizing
synaptic conductance fluctuations in cortical neurons and their influence on spike generation. J. Neurosci. Meth.
169, 302-322.
[39] Rudolph M, Piwkowska Z, Badoual M, Bal T, Destexhe A. (2004) A method to estimate synaptic conductances
from membrane potential fluctuations. J Neurophysiol 91, 2884-2896.
[40] Borg-Graham LJ, Monier C, Frégnac Y. (1998) Visual input evokes transient and strong shunting inhibition in
visual cortical neurons. Nature 393, 369-373.
[41] * Monier C, Chavane F, Baudot P, Graham LJ, Frégnac Y. (2003) Orientation and direction selectivity of synaptic inputs in visual cortical neurons: a diversity of combinations produces spike tuning. Neuron 37, 663-680.
First intracellular study revealing a considerable diversity of combinations of excitatory and inhibitory conductances underlying visual responses.
[42] Wehr M, Zador AM. (2003) Balanced inhibition underlies tuning and sharpens spike timing in auditory cortex.
Nature 426, 442-446.
[43] Wilent W, Contreras D. (2005) Dynamics of excitation and inhibition underlying stimulus selectivity in rat
somatosensory cortex. Nature Neurosci 8, 1364-1370.
[44] Monier C, Fournier J and Frégnac Y. (2008) In vitro and in vivo measures of evoked excitatory and inhibitory
conductance dynamics in sensory cortices. J. Neurosci. Meth. 169, 323-365.
[45] Brunel, N. (2000) Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons. J.
Comp. Neurosci. 8, 183-208.
[46] El Boustani, S., Pospischil, M.,Rudolph-Lilith, M. and Destexhe, A. (2007) Activated cortical states: experiments, analyses and models. J. Physiol. Paris 101, 99-109.
[47] Vogels TP and Abbott LF. (2005) Signal propagation and logic gating in networks of integrate-and-fire neurons.
J. Neurosci. 25, 10786-10795.
[48] * Buonomano DV and Merzenich MM. (1995) Temporal information transformed into a spatial code by a network with realistic properties. Science 267, 1028-1030.
Model proposing that the internal activity of the network may serve some forms of computations, such as converting temporal and spatial information. This is a precursor to what will be later known as “liquid computing”.
9
[49] Dehaene S and Changeux JP. (2005) Ongoing spontaneous activity controls access to consciousness: a neuronal
model for inattentional blindness. PLoS Biol. 3, e141.
[50] Marre O, Yger P, Davison AP and Frégnac Y. (2009) Reliable recall of spontaneous activity patterns in cortical
networks. J. Neurosci. 29, 14596-14606.
[51] Rutishauser U and Douglas RJ. (2009) State-dependent computation using coupled recurrent networks. Neural
Computation 21, 478-509.
[52] Destexhe A, Rudolph M, Fellous JM and Sejnowski TJ. (2001) Fluctuating synaptic conductances recreate
in-vivo–like activity in neocortical neurons. Neuroscience 107, 13-24.
10
A
B
Figure 1: Similarity of network activity patterns during spontaneous activity and visual responses. A.
Frequency of occurrence of activity patterns under spontaneous activity (SA) versus presentation of a movie
(M) in a young animal. B. Same plot for an adult animal. Each dot represents one of the 65,536 possible
binary activity patterns detected in the multi-electrode recording. The color code indicates the number of
spikes. Black line shows equality. The left panels show examples of neural activity patterns on the 16
electrodes in representative SA and movie trials for the same animals. Figure modified from [26].
A
Spontaneous activity (SA)
Natural images (NI)
25 mV
400 ms
C
0
3.5
Amplitude
10
NI
DG
GEM
DN
Scaling Exponent for SA
B
10
10
50
100
200
Frequency (Hz)
300
3
2.5
2
2
2.5
3
Scaling Exponent for NI
3.5
Figure 2: Similar statistical properties of spontaneous activity and responses to natural images in cat visual
cortex in vivo. A. Examples of intracellular responses during spontaneous activity (SA; top) and presentation of natural images (NI, bottom). B. Power spectra estimated for different types of inputs (DG - drifting
gratings; GEM = gratings with eye movements; DN = dense noise). C. Frequency scaling exponent calculated for different cells during natural images, and represented against that of spontaneous activity. Figure
modified from [35].
11
A
B
Computational model
(nS)
Intracellular recording in vitro
(dynamic-clamp)
ge (t)
ge (t)
gi (t)
gi (t)
V(t)
(nS)
Total conductance
Total conductance
40
100
gi (t)
50
20
ge (t)
40
30
gi (t)
30
20
ge (t)
10
10
40
0
30
20
10
0
Time preceding spike (ms)
Time preceding spike (ms)
C
i(t)
D
Intracellular recording in vivo
(awake cat)
(nS)
Intracellular recording in vivo
(anesthetized rat)
Total conductance
Total conductance
30
20
gi (t)
gi (t)
ge (t)
gi (t)
10
ge (t)
40
30
20
ge (t)
10
0
Time preceding spike (ms)
Figure 3: Spike-triggered averages of synaptic conductances in vivo, in vitro and in computo. A. Model
neuron receiving stochastic excitatory conductances mimicking high-conductance states. The conductance
STA shows an opposite pattern of increase of ge (t) and decrease of gi (t) prior to the spike. B. Same
paradigm in a dynamic-clamp experiment in vitro. C. Maximum likelihood estimation of conductance STAs
from intracellular recordings in awake cats. Most of the cells displayed the same pattern with ge increase
and gi decrease. D. Estimates of synaptic conductances following whisker stimulation in anesthetized
rats in vivo. In this case, both ge and gi show a concerted increase prior to spikes. Panels A,C modified
from [34]; B modified from [37]; D modified from [43].
12
A
Integrate-and-fire neurons
B
Spike raster
inh
4000
2000
exc
Cell number
3000
1000
40 mV
0
C
Conductance (nS)
1s
0
1000
2000
3000
4000
5000
Time (ms)
200
Total conductance
gi (t)
ge (t)
100
0
40
30
20
10
Time preceding spike (ms)
Figure 4: Spike-triggered averages of synaptic conductances in self-generated irregular network states. A.
Self-generated asynchronous irregular network state in a network of 4,000 excitatory and 1,000 inhibitory
integrate-and-fire neurons. B. Raster of spiking activity in this model. C. Conductance STA in a representative neuron of the network. The pattern of ge increase and gi decrease was similar to Fig. 3A-C. Figure
modified from El Boustani et al., 2007.
13
Vm
Intrinsic activity
A
External input
B
Weak input
C
Medium input
Strong input
40 mV
500 ms
Conductance (nS)
30
20
Total conductance
gi (t)
ge (t)
10
0
100
80
60
40
20
Time preceding spike (ms)
0
50
100
40
80
30
60
20
40
10
20
0
100
80
60
40
20
Time preceding spike (ms)
0
0
100
Overlay (5 x higher
time resolution)
80
60
40
20
0
Time preceding spike (ms)
Figure 5: Proposed scheme to account for the different experiments. Top scheme: a model neuron was
simulated, receiving two types of inputs, intrinsic activity under the form of stochastic release events at
excitatory and inhibitory synapses, and external inputs at an additional set of excitatory and inhibitory
synapses (with feedforward inhibition). Excitatory and inhibitory connections are indicated by solid and
dotted lines, respectively. STAs were calculated from all spikes. A. Weak input strength (gmax = 12.5 nS
and 25 nS for ge and gi , respectively). B. Medium input strength (gmax = 25 nS and 50 nS for ge and gi ).
C. Strong input strength (gmax = 50 nS and 100 nS for ge and gi ). The inset shows the concerted increase
of excitatory and inhibitory conductances at 5 times higher temporal resolution (compare with Fig. 3D). In
each case, the upper plot shows an example of the Vm activity, and the bottom plots show the conductance
STAs. Parameters for intrinsic activity: ge0 = 7.6 nS, gi0 = 20.7 nS, σe = 2 nS, σi = 5 nS (model from [52]).
14