Cell assembly

A hypothetical concept referring to * phasic sets of coactive neurons that are assumed to encode * internal representations and perform computations over representations.

In 1949, Hebb published The organization of behavior, later to become the most influential book in the history of modern neuroscience ("classic). 'In this book', he wrote, 'I have tried... to bridge the gap between neurophysiology and psychology'. In essence, Hebb's monograph was about how the brain "perceives and represents the world. It has yielded important insights into brain function, as well as two major concepts. Typical of Hebb's integrative view of the brain, these concepts related to two "levels: the "synaptic and the "system. At the synaptic level, Hebb coined a postulate of use-dependent synaptic "plasticity (see "algorithm). At the system level, he proposed the existence of neuronal assemblies as vehicles for perception, "attention, "association, memory, and thought. Hebb (1949) envisaged that in the brain

. stimulation will lead to the slow development of 'cell assembly', a diffuse structure comprising cells in the cortex and diencephalon . capable of acting briefly as a closed system, delivering facilitation to other such systems . A series of such events constitutes a 'phase sequence'—the thought process. Each assembly action may be aroused by a preceding assembly, by a sensory event, or—normally—by both. The central facilitation from one of these activities on the next is the prototype of'attention'.

Although with time Hebb's synaptic postulate may have gained more popularity (despite being regarded by Hebb himself as less original; Milner 1986), it is the 'cell assembly' that was at the heart of his seminal book. In the past 50 years or so, the concept of'cell assembly' has remained viable in both experimental and theoretical research on perception, learning, and memory (e.g. Palm 1982; Crick 1984a; Dudai et al. 1987; von der Malsburg 1987; Gerstein et al. 1989; Singer et al. 1990; Nicolelis et al. 1997; Sakurai 1998).

The platonic cell assembly has the following attributes: (a) it encodes internal representations, in a spatiotemporal code; (b) a representation is distributed over many units in the set; (c) each unit may be a member of several assemblies; (d) the units in the assembly become coactive, and hence actualize the assembly and what it represents, in brief time-locked phases;1 and (e) the assembly is plastic, meaning that the representations could change over time, either in response to input or by endogenous rearrangements. That the cell assembly uses a distributed, alias ensemble, alias population code means that in big-enough assemblies, no single neuron is essential to any percept or memory; put in other terms, the assembly denies the existence of single-cell "homunculi.

Hebb's assemblies did not emerge out of the blue. As is the case with other great ideas, this one as well stood on the shoulders of giants.2 The possibility that sensori-motor information is processed by populations of neurons was raised much earlier (Young 1802). Sherrington, the great advocate of the cellular view of brain function, assumed that individual neurons do not have the representational complexity to account for higher properties of the nervous system (Sherrington 1941).Hebb was a student of Lashley, who attempted in vain to localize memory traces to specific brain regions, and reached the conclusion that the "engram is widely distributed (Lashley 1929). At about the same time, de No (1938), himself relying on earlier observations, singled out the role of neuronal loops and recurrent circuits in information processing in the nervous system. This was contrary to contemporary naive switchboard "metaphor, which described the brain in terms of many yet rather simple (sensory) input-(motor) output connectors. Hebb took the aforementioned ideas further. He formulated a comprehensive conceptual framework of brain function in which populations of neurons represent information about the world. As representations (and hence memories) are distributed over many nodes, localized lesions could fail to abolish memory. Furthermore, assemblies according to Hebb are dynamic entities. They form, "develop (first in the immature and later in the mature brain), associate, and disengage. This calls for synaptic plasticity; Hebb's famous synaptic postulate, mentioned above, was his solution to the mechanism of use-dependent modifications in local nodes in the assembly. The first attempts to model neuronal assemblies on a digital computer were carried out a few years after Hebb's book was published (Rochester et al. 1956). Since then, various mathematical "models have been suggested for the representation of information in neuronal assemblies, based on the principle of ensemble encoding; some of these models incorporate Hebbian local modification rules.

Do cell assemblies exist in real life? Many attempts have been made to identify them and observe their action. These attempts have involved combinations of cellular physiology, neuroanatomy, "functional neu-roimaging, and behaviour. The "zeitgeist is that population coding, which is in line with the assembly notion, plays a part in higher brain function (Jones 1972; Lee et al. 1988; Singer et al. 1990; Hurlbert and Derrington 1993; Tanaka 1993; Arieli et al. 1995; Goldman-Rakic 1996; Nicolelis et al. 1997; Sakurai 1998; Stopfer and Laurent 1999; Tsodyks et al. 1999). Time-locked phasic activity of large neuronal populations is also detected (e.g. "binding). Yet the inference that these are cell assemblies that encode and control behaviour requires more than that. One must establish a necessary, causal, and sufficient link between the coactive populations and specific instances of perception, memory, and behaviour ("criterion). Taking the devil's advocate stand, one could claim, for example, that what is construed as time-locked phasic population coding is manifestation of a "homeostatic device, or a process permissive for coding but not the code itself, or a step on the road to another type of coding, even on the road to the evasive homunculus.

Suppose cell assemblies do indeed encode internal representations; in that case, how big are the assemblies? The minimal number of neurons that is needed to encode and transmit physiologically meaningful information reliably in the cortex was proposed to be only < 100 (Shadlen and Newsome 1998). Similarly, it has been estimated that the number of neurons required to collectively encode a meaningful aspect of a visual scene is closer to 102-103 than to 104-106 (Crick and Koch 1998). The lower limit attest at most to 'mini-assemblies'. But a postulated assembly representing a complex scene could bind together many such mini-assemblies, each encoding an attribute of the representation. In such a case, the overall number of coactive neurons could reach »106. Each coactivation phase of the hypothetical assembly is expected to be in the millisecond to second range. Limits on the coactivation time can be deduced from the observation that 40-150 ms are sufficient to complete complex perceptual and cognitive tasks (Thorpe et al. 1996; Van Turennout et al. 1998).

The cell assembly illustrates a fruitful concept in the neurosciences that had preceded by three to four scientific generations the development of experimental tools to prove or refute it. Such tools now become available: sophisticated cellular physiology, functional neuroimaging, behavioural protocols, advanced data analysis, and, preferably, all combined. The hunt for the assembly "reduces the search for the engram from gross neuroanatomical cartography to the analysis of functional circuits and their interconnections. This by itself is instrumental in advancing our understanding of memory.

Selected associations: Cerebral cortex, Engram, Homunculus, Internal representation, Model

'During this period, the system can be said to become locked in a quasi-stable state of an energy minimum. 2For a fascinating history of this aphorism, commonly attributed to Newton, including some lessons on the intricacies of 'culture and 'collective memory, see Merton (1993).

Was this article helpful?

0 0

Post a comment