Binding by Neural Synchrony

Neuronal systems have to solve immensely complex combinatorial problems and require efficient binding mechanisms in order to generate representations of perceptual objects and movements. In the context of cognitive functions, combinatorial problems arise from the fact that perceptual objects are defined by a unique constellation of features, the diversity of possible constellations being virtually unlimited (cf. BINDING PROBLEM). Combinatorial problems of similar magnitude have to be solved for the acquisition and execution of motor acts. Although the elementary components of motor acts, the movements of individual muscle fibers, are limited in number, the spatial and temporal diversity of movements that can be composed by combining the elementary components in ever-changing constellations is again virtually infinite. In order to establish neuronal representations of perceptual objects and movements, the manifold relations among elementary sensory features and movement components have to be encoded in neural responses. This requires binding mechanisms that can cope efficiently with combinatorial complexity. Brains have acquired an extraordinary competence to solve such combinatorial problems, and it appears that this competence is a result of the evolution of the CEREBRAL CORTEX.

In the primary visual cortex of mammals, relations among the responses of retinal ganglion cells are evaluated and represented by having the output of selected arrays of ganglion cells converge in diverse combinations onto individual cortical neurons. Distributed signals are bound together by selective convergence of feed forward connections (Hubel and Wiesel 1962). This strategy is iterated in prestriate cortical areas in order to generate neurons that detect and represent more complex constellations of features including whole perceptual objects.

However, this strategy of binding features together by recombining input connections in ever-changing variations and representing relations explicitly by responses of specialized cells results in a combinatorial explosion of the number of required binding units. It has been proposed, therefore, that the cerebral cortex uses a second, complementary strategy, commonly called assembly coding, that permits utilization of the same set of neurons for the representation of different relations (Hebb 1949). Here, a particular constellation of features is represented by the joint and coordinated activity of a dynamically associated ensemble of cells, each of which represents explicitly only one of the more elementary features that characterize a particular perceptual object. Different objects can then be represented by recombining neurons tuned to more elementary features in various constellations (assemblies). For assembly coding, two constraints need to be met. First, a selection mechanism is required that permits dynamic, context dependent association of neurons into distinct, functionally coherent assemblies. Second, grouped responses must get labeled so that they can be distinguished by subsequent processing stages as components of one coherent representation and do not get confounded with other unrelated responses. Tagging responses as related is equivalent with raising their salience jointly and selectively, because this assures that they are processed and evaluated together at the subsequent processing stage. This can be achieved in three ways. First, nongrouped responses can be inhibited; second, the amplitude of the selected responses can be enhanced; and third, the selected cells can be made to discharge in precise temporal synchrony. All three mechanisms enhance the relative impact of the grouped responses at the next higher processing level. Selecting responses by modulating discharge rates is common in labeled line coding where a particular cell always signals the same content. However, this strategy may not always be suited for the distinction of assemblies because it introduces ambiguities, reduces processing speed, and causes superposition problems (von der Malsburg 1981; Singer et al. 1997). Ambiguities could arise because discharge rates of feature-selective cells vary over a wide range as a function of the match between stimulus and receptive field properties; these modulations of response amplitude would not be distinguishable from those signalling the relatedness of responses. Processing speed would be reduced because rate coded assemblies need to be maintained for some time in order to be distinguishable. Finally, superposition problems arise, because rate coded assemblies cannot overlap in time within the same processing stage. If they did, it would be impossible to distinguish which of the enhanced responses belong to which assembly. Simultaneous maintenance of different assemblies over perceptual time scales is required, however, to represent composite objects.

Both the ambiguities and the temporal constraints can be overcome if the selection and labeling of responses is achieved through synchronization of individual discharges (Gray et al. 1989; Singer and Gray 1995). Synchronicity can be adjusted independently of rates, and so the signature of relatedness, if expressed through synchronization, is independent of rate fluctuations. Moreover, synchronization enhances only the salience of those discharges that are precisely synchronized and generate coincident synaptic potentials in target cells at the subsequent processing stage. Hence the selected event is the individual spike or a brief burst of spikes. Thus, the rate at which different assemblies can follow one another within the same neuronal network without getting confounded is much higher than with rate coding. It is only limited by the duration of the interval over which synaptic potentials summate effectively. If this interval is in the range of 10 or 20 ms, several different assemblies can alternate within preceptually relevant time windows.

If synchronization serves as a selection and binding mechanism, neurons must be sensitive to coincident input. Moreover, synchronization must occur rapidly and show a relation to perceptual phenomena.

Although the issue of coincidence detection is still controversial (König, Engel, and Singer 1996; Shadlen and Newsome 1994), evidence is increasing that neurons can evaluate temporal relations with precision among incoming activity (see e.g., Carr 1993). That cortical networks can handle temporally structured activity with high precision and low dispersion follows from the abundant evidence on the oscillatory patterning and precise synchronization of neuronal responses in the -frequency range (Singer and Gray 1995; König, Engel, and Singer 1996). Synchronization at such high frequencies is only possible if integration time constants are short. Precise synchronization over large distances is usually associated with an oscillatory patterning of responses in the - and -frequency range, suggesting a causal relation (König, Engel, and Singer 1995). This oscillatory patterning is associated with strong inhibitory interactions (Traub et al. 1996), raising the possibility that the oscillations contribute to the shortening of integration time constants.

Simulations with spiking neurons reveal that networks of appropriately coupled units can undergo very rapid transitions from uncorrelated to synchronized states (Deppisch et al. 1993; Gerstner 1996). Rapid transitions from independent to synchronized firing are also observed in natural networks. In visual centers, it is not uncommon that neurons engage in synchronous activity, often with additional oscillatory patterning, at the very same time they increase their discharge rate in response to the light stimulus (Neuenschwander and Singer 1996; Gray et al. 1992). One mechanism is coordinated spontaneous activity that acts like a dynamic filter and causes a virtually instantaneous synchronization of the very first discharges of responses Fries et al. 1997b). The spatio-temporal patterns of these spontaneous fluctuations of excitability reflect the architecture and the actual functional state of intracortical association connections. Thus, grouping by synchronization can be extremely fast and still occur as a function of both the prewired associational dispositions and the current functional state of the cortical network.

Evidence indicates that the probability and strength of response synchronization reflects elementary Gestalt criteria such as continuity, proximity, similarity in the orientation domain, colinearity, and common fate (Gray et al. 1989; Engel, König, and Singer 1991; Engel et al. 1991; Freiwald, Kreiter, and Singer 1995; Kreiter and Singer 1996). Most importantly, the magnitude of synchronization exceeds that expected from stimulus-induced rate covariations of responses, indicating that it results from internal coordination of spike timing. Moreover, synchronization probability does not simply reflect anatomical connectivity but changes in a context-dependent way (Gray et al. 1989; Engel, König, and Singer 1991; Freiwald, Kreiter, and Singer 1995; Kreiter and Singer 1996), indicating that it is the result of a dynamic and context-dependent selection and grouping process. Most of the early experiments on response synchronization have been performed in anesthetized animals, but more recent evidence from cats and monkeys indicates that highly precise, internally generated synchrony occurs also in the awake brain, exhibits similar sensitivity to context (Kreiter and Singer 1996; Fries et al. 1997a; Gray and Viana Di Prisco 1997), and is especially pronounced when the EEG is desynchronized (Munk et al. 1996) and the animals are attentive (Roelfsema et al. 1997). Direct relations between response synchronization and perception have been found in cats who suffered from strabismic amblyopia, a developmental impairment of vision associated with suppression of the amblyopic eye, reduced visual acuity, and disturbed perceptual grouping (crowding) in this eye. Quite unexpectedly, the discharge rates of individual neurons in the primary visual cortex fail to reflect these deficits (see Roelfsema et al. 1994 for references). The only significant correlate of amblyopia is the drastically reduced ability of neurons driven by the amblyopic eye to synchronize their responses (Roelfsema et al. 1994), and this accounts well for the perceptual deficits: by reducing the salience of responses, disturbed synchronization could be responsible for the suppression of signals from the amblyopic eye, and by impairing binding, it could reduce visual acuity and cause crowding.

Another close correlation between response synchronization and perception has been found in experiments on binocular rivalry (Fries et al. 1997a). A highly significant correlation exists between changes in the strength of response synchronization in primary visual cortex and the outcome of rivalry. Cells mediating responses of the eye that won in interocular competition increased the synchronicity of their responses upon presentation of the rival stimulus to the other, losing eye, while the reverse was true for cells driven by the eye that became suppressed.

These results support the hypothesis that precise temporal relations between the discharges of spatially distributed neurons matter in cortical processing and that synchronization may be exploited to jointly raise the salience of the responses selected for further processing, that is, for the dynamic binding of distributed responses into coherent assemblies.

The example of rivalry also illustrates how synchronization and rate modulation depend on each other. The signals from the suppressed eye failed to induce tracking EYE MOVEMENTS, indicating that the vigorous but poorly synchronized responses in primary visual areas eventually failed to drive the neurons responsible for the execution of eye movements. Thus, changes of synchronicity result in changes of response amplitudes at subsequent processing stages. This convertibility provides the option to use both coding strategies in parallel in order to encode complementary information.

References

Carr, C. E. (1993). Processing of temporal information in the brain. Annu. Rev. Neurosci. 16:223-243.

Deppisch, J., H.-U. Bauer, T. B. Schillen, P. König, K. Pawelzik, and T. Geisel. (1993). Alternating oscillatory and stochastic states in a network of spiking neurons. Network 4:243-257.

Engel, A. K., P. König, and W. Singer. (1991). Direct physiological evidence for scene segmentation by temporal coding. Proc. Natl. Acad. Sci. USA 88:9136-9140.

Engel, A. K., A. K. Kreiter, P. König, and W. Singer. (1991). Synchronization of oscillatory neuronal responses between striate and extrastriate visual cortical areas of the cat. Proc. Natl. Acad. Sci. USA 88:6048-6052.

Freiwald, W. A., A. K. Kreiter, and W. Singer. (1995). Stimulus dependent intercolumnar synchronization of single unit responses in cat area 17. Neuroreport 6:2348-2352.

Fries, P., P. R. Roelfsema, A. K. Engel, P. König, and W. Singer. (1997a). Synchronization of oscillatory responses in visual cortex correlates with perception in interocular rivalry. Proc. Natl. Acad. Sci. USA 94:12699-12704.

Fries, P., P. R. Roelfsema, W. Singer, and A. K. Engel. (1997b). Correlated variations of response latencies due to synchronous subthreshold membrane potential fluctuations in cat striate cortex. Soc. Neurosci. Abstr. 23: 1266.

Gerstner, W. (1996). Rapid phase locking in systems of pulse-coupled oscillators with delays. Phys. Rev. Lett. 76:1755-1758.

Gray, C. M., A. K. Engel, P. König, and W. Singer. (1992). Synchronization of oscillatory neuronal responses in cat striate cortex -- temporal properties. Visual Neurosci. 8:337-347.

Gray, C. M., P. König, A. K. Engel, and W. Singer. (1989). Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties. Nature 338:334-337.

Gray, C. M., and G. Viana Di Prisco. (1997). Stimulus-dependent neuronal oscillations and local synchronization in striate cortex of the alert cat. J. Neurosci. 17:3239-3253.

Hebb, D. O. (1949). The Organization of Behavior. Wiley.

Hubel, D. H., and T. N. Wiesel. (1962). Receptive fields, binocular interaction and functional architecture in the cat's visual cortex. J. Physiol. (Lond.) 160:106-154.

König, P., A. K. Engel, and W. Singer. (1995). Relation between oscillatory activity and long-range synchronization in cat visual cortex. Proc. Natl. Acad. Sci. USA 92:290-294.

König, P., A. K. Engel, and W. Singer. (1996). Integrator or coincidence detector? The role of the cortical neuron revisited. Trends Neurosci. 19:130-137.

Kreiter, A. K., and W. Singer. (1996). Stimulus-dependent synchronization of neuronal responses in the visual cortex of awake macaque monkey. J. Neurosci. 16:2381-2396.

Munk, M. H. J., P. R. Roelfsema, P. König, A. K. Engel, and W. Singer. (1996). Role of reticular activation in the modulation of intracortical synchronization. Science 272:271-274.

Neuenschwander, S., and W. Singer. (1996). Long-range synchronization of oscillatory light responses in the cat retina and lateral geniculate nucleus. Nature 379:728-733.

Roelfsema, P. R., A. K. Engel, P. König, and W. Singer. (1997). Visuomotor integration is associated with zero time-lag synchronization among cortical areas. Nature 385:157-161.

Roelfsema, P. R., P. König, A. K. Engel, R. Sireteanu, and W. Singer. (1994). Reduced synchronization in the visual cortex of cats with strabismic amblyopia. Eur. J. Neurosci. 6:1645-1655.

Shadlen, M. N., and W. T. Newsome. (1994). Noise, neural codes and cortical organization. Curr. Opin. Neurobiol. 4:569-579.

Singer, W., A. K. Engel, A. K. Kreiter, M. H. J. Munk, S. Neuenschwander, and P. R. Roelfsema. (1997). Neuronal assemblies: necessity, signature and detectability. Trends Cognitive Sci. 1(7):252-261.

Singer, W., and C. M. Gray. (1995). Visual feature integration and the temporal correlation hypothesis. Annu. Rev. Neurosci. 18:555-586.

Traub, R. D., M. A. Whittington, I. M. Stanford, and J. G. R. Jefferys. (1996). A mechanism for generation of long-range synchronous fast oscillations in the cortex. Nature 383:621-624.

von der Malsburg, C. (1981). The correlation theory of brain function. Internal Report 81-2 Max-Planck-Institute for Biophysical Chemistry. Reprinted 1994, in E. Domany, J. L. van Hemmen and K. Schulten, Eds., Models of Neural Networks II. Springer-Verlag, pp. 95-119.

Binding by Neural Synchrony

See also

References

Further Readings