|Moshe Abeles (2011), Scholarpedia, 6(7):1505.
|revision #91110 [link to/cite this article]
The concept of cell assembly was coined by the Canadian neuropsychologist D. O. Hebb (Hebb 1949) to describe a network of neurons that is being activated repeatedly during a certain mental process and in this way the excitatory synaptic connections among its members are being strengthened. In Hebb’s thinking the synaptic strengthening depends on the order of activation, and thus there is a time structure to the activation of cell assembly. Thus, the activity of a cell assembly is characterized by the spatiotemporal structure of the activity of its members. After being formed, activation of the first stage will revive the entire spatiotemporal signature of the cell assembly. Figure 1 (Figure 10 in the original manuscript) illustrates his thinking.
Present-day evolution of the concept
Nowadays the concept of cell assembly is used loosely to describe a group of neurons that perform a given action or represent a given percept or concept in the brain. Typically, one thinks of the group as having strong internal synaptic interactions which set them apart from other groups of neurons. Different users may use this concept with more or less permissive definitions. The examples below may illustrate the boundaries of this concept.
A motoneuron pool (i.e. all neurons whose axons connect to the same muscle) share clear common action, yet one would hesitate to call such a pool cell assembly. A fixed point attractor neural network is clearly a cell assembly. It may serve as a content addressable associative memory, where partial activation of its member will build up to full activation of all its members. Whether a multi layer perceptron is a cell assembly is debatable. On the one hand it may play a clear perceptual role of extracting combinations of simple features to generate a more complex mental representation. On the other hand, the interactions are strictly feed forward, so it may not be justified to treat them as mutual interactions among the assembly’s members.
In the examples above one seems to treat differently excitatory and inhibitory interactions. Neurons within a motoneuron pool do affect each other through mutual inhibition by Renshaw cells, yet we tend not to call them a cell assembly. In the fixed point attractor, the active neurons inhibit the rest of the network, yet we would not call the entire network a cell assembly, but only those neurons that excite each other. Note, that this distinction falls apart in some representations of the original ANN of Hopfield (Hopfield 1982) where the activity of a neuron is marked by +1 if active and -1 if not. There, when an attractor is active the neurons which are quiet (-1 state) excite the neurons which are active, and those which are active (+1 state) inhibit those which are quiet. In the cerebral cortex, neurons have plenty of mutual (direct or indirect) synaptic interactions. One would tend to call a mini-column in the primary visual cortex a cell assembly, yet in most physiological thinking the unit’s orientation selectivity is formed by feed forward connections (but see Ben-Yishay et al. 1995 for a different view).
From the examples above one gets the impression that we expect to see strong mutual excitatory connections among the members of the cell assembly. Yet a network with only inhibitory connections (such as in the reticular nucleus of the thalamus) may also show fixed-point attractors (Golomb & Rinzel 1993).
The bottom line is that there is no clear agreed-upon definition for the term "cell assembly", and one has to understand from the context what is meant.
Measuring activity in cell assemblies has to rely on recording the activity of many neurons in parallel. This might be achieved by multiple micro electrode recordings, by Ca imaging or by voltage sensitive dye imaging. A review dealing with methods for recording, analyzing, and simulating cell assemblies may be found in Gerstein and Kirkland (2001)
- Hebb D.O. The Organization of Behavior: a Neuropsychological Theory. New York: Wiley, 1949.
- Hopfiels J.J. Neural networks and physical systems with emergent collective computational abilities. PNAS 79:2554-2558, 1982.
- Ben-Yishai R., Bar-Or R.L and Sompolinsky H. Theory of orientation tuning in visual cortex. PNAS 92: 3844-3848, 1995.
- Golomb D. and Rinzel J. Dynamics of globally coupled inhibitory neurons with heterogeneity. Physical Review E. 48:4810-4814, 1993.
- Gerstein G.L. and Kirkland K.L. Neural assemblies: technical issues, analysis, and modeling. Neural Netw. 14:589-598, 2001.