Spike timing, synchronization and information processing on the sensory side of the central nervous system
Introduction
“Noise in the nervous system must in general be treated as a tentative hypothesis since apparent noise may have any of several uses” T.H. Bullock (1970).
Early studies of spike trains in various sensory systems strongly suggested that in sensory fibers (1) the emission of spikes was not deterministic but had a component of stochasticity and (2) within the dynamic range of the neurons (between the sensory threshold and the saturation), there was in general a nice monotonic relationship between the intensity of the response, measured by the average spike count over an appropriate time window (or the average firing rate in spikes/s) under repeated trials and the intensity of the applied stimulus.
It has appeared in the last decades, however, that these rules might be insufficient to account for the coding of information in the sensory periphery of the central nervous system (CNS). Given that spikes propagating along a given fiber are separated in time by definite interspike intervals (ISI), researchers soon thought that these ISIs were good candidates to take part in the coding of information in the brain, alongside rate. The idea that codes that utilize spike timing can make more efficient use of the capacity of neural connections than those relying simply on the average rate of firing can be traced to MacKay and McCulloch (1952). But the proposition that, given two trains of impulses of the same over-all frequency, messages of different informational content may be signaled during each measuring unit of time by the (time) positional differences of the impulses was already present in the early works of pioneers such as Lorente de No (1939) and Pitts and McCulloch (1947). The first experimental observation of the effect of changing the pattern of stimulations, on the contraction of different crustacean muscles, was performed as early as 1950 (Wiersma and Adams, 1950). The possibility of a temporal code, using ISIs to carry sensory information, was also transparent in the 1967 works of Mountcastle (1967). How temporal coding was important seemed uncertain to him, but in this article, he gives one example of sensory information transfer (the frequency of cutaneous flutter-vibration in mechanoreceptive fibers) that can hardly be explained on any ground other than that the CNS can measure intervals.
In parallel with these developments, it became progressively clear that sensory data are generally not transmitted along single lines, but rather through pipes, mobilizing populations of neurons working in parallel. Because ISIs may play a role in coding, not only at the level of a single line, but across neurons in these pipes, the possibility that these populations of neurons do not work independently but in a temporally coordinated fashion (in synergy, so to speak), deserved to be carefully tested.
In the past decade, the paradigm of ‘synchronicity’, that is, the idea that simultaneous inputs (to within an appropriate time window, much shorter than the mean ISI) were much more apt to trigger a spike in a target cell than randomly occurring inputs thus received a renewed attention. In fact, the idea was not new. A coincidence code was theoretically proposed as early as 1948 (Jeffress, 1948) and signaled as probably operating in some central auditory neurons in the cat as early as 1966, following an experiment by Rose et al. (1966). More recently, the possibility that coincidence codes were effective not only at the level of whole neurons, but also at the level of separate dendritic branches has also been considered (Bair, 1999).
I shall review here the evidence for and against temporal coding at the level of the sensory side of the nervous system, between the receptors (of which the transduction mechanisms will not be considered) and the centers of the CNS that are thought to make the interface with higher nervous function, such as the generation of perception, or action. One of the main issues at stake is the organization of the ensemble coding—how these neuron assemblies are temporally coordinated—as opposed to population coding—in the case where each neuron of the assembly simply contributes to the population in proportion of its firing rate—(Theunissen and Miller, 1995). Another appropriate opposition is between temporal coding and temporal encoding. Temporal coding of a signal is characterized by a one-to-one correspondence between the time of occurrence of a sensory event and the time of occurrence of the corresponding neural influx, whereas temporal encoding of a signal corresponds to situations in which information about static or dynamic signals is encoded in the temporal pattern of action potentials (AP) without the AP being tied to changes in the signal itself. As we shall see, the present knowledge makes a sharp contradistinction between the temporal coding of externally timed events and the temporal encoding of other characters. While the first makes a large use of very fast, precise channels (using frequency channels of 100 Hz or higher), the second type of processing seems to limit itself to carrying channels of much lower frequencies, bound by the so-called gamma range (35–70 Hz), implying a temporal resolution not over 15–25 ms (Mainen and Sejnowski, 1995, Nowak et al., 1997, Buracas and Albright, 1999, Reich et al., 2000). While I will acknowledge this dichotomy, I would like to suggest the possibility that temporal encryption at a precise scale still escapes our present means of investigation.
The building up of the anatomical circuits in the brain in the course of ontogeny and its fine tuning in the maturation of CNS networks constitute in my view a strong indication that timing considerations are important in information processing in the CNS, which should not be neglected.
In cortical pyramidal cells of cats, Destexhe and Paré (1999) cite a density of synapses of 10–20 GABAergic synapses (per 100 μ2 of membrane) in axon initial segment, 8–12 GABAergic synapses and 55–65 AMPA synapses in dendrites. It is generally estimated that activation of 80–150 synapses on basal but >400 synapses at distal dendrites is necessary to trigger an AP at the soma, in absence of active propagation of EPSPs along the dendrites. Thus the distinction between proximal and distal synapses may be critical or relatively indifferent, depending on the functional conditions of the neurons. Particularly important, however, is the observation that while excitatory synapses are scattered all along proximal and distal dendrites, there are practically none of them on the soma of nervous cells, where synaptic contacts are mostly inhibitory. The functional meaning of this observation is not yet apparently understood.
In the presence of dendritic Na+ and K+ currents triggered by an intense background activity (or by appropriate neuromodulation, or by the coincidence of an EPSP with a back propagating potential), the effect on the soma of proximal or distal inputs becomes similar. Active dendrites therefore diminish the variability of the responses due to the location of inputs, as Cook and Johston (1997) have suggested.
Moreover, it is now relatively well accepted that the mature CNS is not the raw product of genomic construction but the result of a process of fine-tuning. In development, the CNS system uses a strategy of swarming and pruning, whereby a primitive super-abundance of neurons and synaptic contacts is followed by a quasi-Darwinian selection, finally leaving in place-optimized networks. It is certainly not the place here to give a detailed account of recent advances in cellular burgeoning and degeneration during development (see for instance Hamburger, 1975, Shatz, 1996), but it is important to keep these effects in mind, as a strong indication of a particularly precise temporal adjustment of the CNS in the course of development, maturation and final tuning.
Synchronization implies that excitation impinging on a given neuron through several afferents should arrive within a narrow window of time. As an example, it has been estimated that in the case of coincidence detector cells of the nucleus laminaris of the owl, excitation from afferent spikes coming from the nucleus magnocellularis spreads over the terminal axo-dendritic arborization over a span of time of about 250 μs (Gerstner et al., 1996). Verifying to what extent time constraints in the spreading of inputs on a single neuron apply also in other neurons in the CNS is important, for instance in the visual system, in which a looser time tuning might be of less consequence than in a system devoted to interaural delay detection. This would suggest a functional role of synchrony of synaptic excitation for single neurons and populations of neurons alike.
Anatomically, the local circuitry of a single neuron may seem quite complex, yet its temporal compactness remains remarkable. For example, Freund et al. (1989) studied a magnocellular thalamo-cortical neuron of a monkey. They found that this neuron had about 3200 boutons in three different cortical clumps, with a total elongation of the terminal arborization of about 1.8 mm (0.8 mm from clump center to clump center). Each bouton made 2±1 synapses. Each cortical target neuron received 8–10 boutons; therefore the number of target cells of a single magnocellular thalamo-cortical neuron was about 400. The interval between successive boutons was typically about 15 μm, which means that the interbouton delay was about 5 μs (assuming a value of 0.3 m/s for the conduction velocity, see below). The same typical architecture was found to apply to parvocellular thalamo-cortical cells as well, except that in this case, the terminal arborization was clearly concentrated in layer 4Cβ, the total spread of the arborization was narrower (only about 400 μm) and the thalamo-cortical divergence smaller, leading to an even narrower spread of time in the excitations impinging from a parvocellular neuron onto its cortical targets.
To evaluate more closely the spread of time over neuronal target synapses, a detailed model (of real transcallosal fibers) was performed by Jean-Christophe Houzel and colleagues in Lausanne (Houzel et al., 1994, Innocenti et al., 1994). It is remarkable that all boutons located in a given cluster were activated within a very small window of time, of the order of 250 μs, and that, in the case where the afferent axon had an early branching, leading to several distinct clouds of synapses onto one of its target neurons, the synchrony of activation between the various clouds was almost as precise as the synchrony inside of a given cloud (not more than 400 μs). Although these figures are subject to some uncertainty, depending upon the exact value of conduction velocity in the thinner branches of axonic arborization (in this work the validity of the Rushton–Waxman linear law between velocity and diameter had been assumed), the submillisecond spreading in activation times within and across synaptic clusters of a given axonic arborization should be considered as a robust result. Thus, the fine structure and localization of neural connections can be used as a support for a role of fine timing, even though anatomy alone, of course, does not tell much about the functionality of these properties.
Since, however, the very idea of temporal coding is thwarted by the observation of the stochasticity of neuronal responses, one should first examine (Section 2) the question of fluctuations, or deviations at successive trials of actual spike trains both in spike times and in spike count from the mean represented by the poststimulus time histogram (PSTH), before addressing directly temporal correlations in sensory neural discharges.
Section snippets
Response variability in sensory systems
Fluctuations over successive trials of actual spike trains both in spike times and in spike count are usually considered as noise. This denomination is unfortunate, in that it suggests that fluctuations in the responses might be just random, as in a Poisson process, where spiking has a constant probability of occurrence independently of other spikes and where, in consequence, the number of spikes in a given time window follows a Poisson distribution (Tuckwell, 1988). A series of works have
Characterization of sensory systems from the point of view of temporal correlations
Not only synchronous inputs but also time-coordinated inputs have important effects on signal transmission by neurons. In the visual system, when two spikes from a single ganglion-cell axon arrive within 30 ms of each other, the second spike is much more likely than the first to produce a geniculate spike, an effect that Usrey et al. (1998) have termed cell paired-spike enhancement. This effect is demonstrated in Fig. 18, where pairs of retinal spikes that occurred 10.0±0.4 ms apart and which
Measuring information in sensory spike trains
The first work using Shannon's theory of information was that of MacKay and McCulloch (1952), who theoretically determined the total entropy of a spike train in the CNS and brought to light the possibility that a temporal code might be much more effective than a rate code. Later, Werner and Mountcastle (1965) were among the first researchers to apply the theory to real spike trains, in the somato-sensory system this time. They found that a single mechanoreceptive fiber was able to encode about
Synchrony, oscillations and the binding hypothesis
I shall be brief on this subject, which has been widely covered in many publications.
In the visual system, the presence of oscillations has been claimed in the retina, in the LGN and in the primary and higher visual cortices alike. But, whereas synchronized oscillations are found to be prevalent in V1 by some authors (reviewed by Singer and Gray, 1995), their prevalence was disputed by others (see for instance Nowak et al., 1999, Lampl et al., 1999). In the olfactory system, the relevance of
Conclusions
In most modalities, not only processing but also reshaping of sensory information certainly occurs at different stages, particularly at the periphery, in the thalamus and in the neocortex. The processing of information related to externally timed or time varying signals is performed at least in the two first stages, through specialized networks usually consisting of magnocellular cells, specially adapted to the treatment of sensory data with a great fidelity and very little variance of the
Acknowledgements
I would like to express my gratitude to A. Duvel, J.C. Lecas, B.J. Richmond and S. Sara for their critical reading of earlier versions of the manuscript. This work has been supported by the Centre National de la Recherche Scientifique (CNRS), Paris.
References (179)
Spike timing in the mammalian visual system
Curr. Opin. Neurobiol.
(1999)- et al.
A network capable of reading the temporal codes in insect olfactive discrimination
C. R. Acad. Sci. Paris III
(1999) - et al.
Mechanisms of concerted firing among retinal ganglion cells
Neuron
(1998) - et al.
Gauging sensory representations in the brain
TINS
(1999) - et al.
Efficient discrimination of temporal patterns by motion-sensitive neurons in primate visual cortex
Neuron
(1998) Temporal response properties of neurons in the auditory pathway
Curr. Opin. Neurobiol.
(1999)- et al.
Synaptic interactions mediating synchrony and oscillations in primate sensorimotor cortex
J. Physiol. Paris
(2000) - et al.
On the significance of temporally structured activity in the dorsal lateral geniculte nucleus (LGN)
Prog. Neurobiol.
(1997) - et al.
Rate coding versus temporal order coding: a theoretical approach
Biosystems
(1998) - et al.
Low response variability in simultaneously recorded retinal, thalamic, and cortical neurons
Neuron
(2000)
Synchronous membrane potential fluctuations in neurons of the visual cortex
Neuron
Time structure and stimulus dependence of precisely replicating patterns present in monkey cortical neuronal spike trains
Brain Res.
The effect of correlated variability on the accuracy of a population code
Neural Comput.
Role of cortical neurons: integrator or coincidence detector?
Isr. J. Med. Sci.
Corticonics
Cortical activity flips among quasi-stationary states
Proc. Natl. Acad. Sci. USA
Precisely correlated firing in cells of the lateral geniculate nucleus
Nature
Functional connectivity between simple cells and complex cells in cat striate cortex
Nat. Neurosci.
Stimulus dependence of two-state fluctuations of membrane potential in cat visual cortex
Nat. Neurosci.
Dynamics of ongoing activity: explanation of the large variability in evoked cortical responses
Science
Towards a theory of early visual processing
Neural Comput.
Cellular mechanisms contributing to response variability of cortical neurons in vivo
J. Neurosci.
Dynamic spike threshold reveals a mechanism for synaptic coincidence detection in cortical neurons in vivo
Proc. Natl. Acad. Sci. USA
Responses of neurons in primary and inferior temporal visual cortices to natural scenes
Proc. R. Soc. Lond. B Biol. Sci.
Correlated firing in macaque visual area MT: time scales and relationship to behavior
J. Neurosci.
Precise spatiotemporal repeating patterns in monkey primary and supplementary motor areas occur at chance levels
J. Neurophysiol.
The role of synchrony and oscillations in the motor output
Exp. Brain Res.
Threee factors limiting the reliable detection of light by the retinal ganglion cells of the cat
J. Physiol. (Lond.)
Responses to single quanta of light in retinal ganglion cells of the cat
Vision Res. (Suppl.)
Modulations corticales de la structure temporelle fine des trains d'impulsions dans le corps genouillé latéral dorsal du chat
C. R. Acad. Sci. Paris
Synaptic background activity influences spatiotemporal inegration in single pyramidal cells
Proc. Natl. Acad. Sci. USA
The structure and precision of retinal spike trains
Proc. Natl. Acad. Sci. USA
Reading a neural code
Science
Visual orientation and spatial frequency discrimination: a comparison of single neurons and behavior
J. Neurophysiol.
Synergy in a neural code
Neural Comput.
Synaptic origin of rhytmic visually evoked activity in kitten area 17 neurons
NeuroReport
The analysis of visual motion: a comparison of neuronal and psychophysical performance
J. Neurosci.
A relationship between behavioral choice and the visual responses of neurons in macaque MT
Vis. Neurosci.
Slow covariations in neural resting potentials can lead to artefactually fast cross correlations in their spike trains
J. Neurophysiol.
The reliability of neurons
J. Gen. Physiol.
As if time mattered: temporal strategies for neural coding of sensory information
Commun. Cognition-Artif. Intell. (CC-AI)
Temporal coding of periodicity pitch in the auditory system: an overview
Neural Plast.
Neuronal correlates of the pitch of complex tones, I. Pitch and pitch salience
J. Neurophysiol.
Synchronizaton of visual responses between the cortex, lateral geniculate nucleus, and retina in the anesthetized cat
J. Neurosci.
Active dendrites reduce location-dependent variability of synaptic input trains
J. Neurophysiol.
Efficient coding of natural scenes in the lateral geniculate nucleus: experimental test of a computational theory
J. Neurosci.
Coding of visual information by precisely correlated spikes in the lateral geniculate nucleus
Nat. Neurosci.
Cited by (137)
Modern approaches of signal processing for bidirectional neural interfaces
2021, Somatosensory Feedback for NeuroprostheticsSynchronization in finite/fixed time of fully complex-valued dynamical networks via nonseparation approach
2020, Journal of the Franklin InstituteFinite-time synchronization of complex-valued neural networks with reaction-diffusion terms: an adaptive intermittent control approach
2024, Neural Computing and Applications