Modelling a network of neurons

In order to find out whether specific behavioural responses could be produced using interneurons with such unspecific characteristics, computing techniques were used to construct a model of the network involved in leech local bending (Lockery et al., 1989). Software represented three levels of units, with one level equivalent to sensory neurons, one equivalent to motor neurons, and the third interposed between them like the interneu-rons (Fig. 8.4). The 18 interneuron units connected only with sensory and motor neurons and not with each other, which is the same pattern as in the leech. The strengths of connections between interneurons and sensory or motor neurons were variable.

In an experiment, a model network was gradually trained to reproduce the behaviour of a leech segmental ganglion, so that activation of a particular sensory neuron would generate an appropriate local bending-avoid-ance response. The way in which this was done was first to programme the computer to make connections between sensory neurons and interneurons and between interneurons and motor neurons. The pattern of the connections was set randomly, except that interneurons were arranged as left-right pairs, and all the connections were initially weak. The performance of the circuit was then tested to find out how closely its behaviour matched that of the leech. It would be very unlikely to perform well in this first test. Next, the computer altered the strengths of some of the connections, and the behaviour of the circuit was again tested. If the performance of the circuit had been improved, the same connections were again strengthened, but if it had deteriorated, these connections were weakened and others were strengthened. By repeatedly altering connection strengths and then testing the circuit, the computer gradually improved the circuit in a stepwise manner until the input-output relationship became the same as that observed in a leech. This technique of training a model network of neural elements is called back propagation. Successful training of the network required between 10000 and a 100000 repetitions of the training cycle.

Eighteen different experiments were performed. Each one produced a network that reproduced the behaviour of a leech but was different from the other 17 networks. All the networks contained model interneurons that resembled the leech interneurons in the lack of specificity in their connections, with more than 95 per cent receiving inputs from two or more P cells and 88 per cent making output connections to at least seven of the eight motor neurons. This means that it is possible to adopt a strategy of using diffusely connected elements in a network to link inputs and outputs in specific ways. Indeed, because the computer program always constructed a diffuse network rather than one in which individual units were dedicated to restricted functions, the strategy must be a good one.

In a leech segmental ganglion, therefore, each of the interneurons involved in dorsal bending makes many connections that are inappropriate to that response. Stimulation of a particular area of skin always causes the same bending response, which means that the particular balance of the strengths of connections within the ganglion must offset the apparently inappropriate nature of many of them. Although computer modelling generated a number of different circuits for controlling local bending, it is likely that the network of neurons in one leech is very similar to that in another. It would be interesting to know what factors in evolution have selected this particular network. Careful use of computer modelling techniques, such as this study on the leech, is an invaluable tool for neuroethology because it allows an experimenter to test whether a neuronal network operates in the way expected. An alternative approach to reconstructing circuits is to isolate individual neurons and allow them to make synaptic connections in culture (Box 8.1).

Box 8.1. Reconstructing circuits of live neurons

A very direct test of whether a circuit of neurons is sufficient to generate an activity is to isolate the circuit concerned from the nervous system. In the lobster stomatogastric ganglion, which contains a very small number of neurons this has been done by killing some neurons, so that only those involved in a particular circuit are left. Another technique was adopted by Syed, Bulloch & Lukowiak (1990) in a study of the pond snail, Lymnaea. This is a diving, air-breathing animal; it has a lung, and comes to the surface to exchange the air it contains by rhythmically opening and closing the aperture, called the pneu-mostome (a). Experiments using intracellular recording revealed a simple circuit of three interconnected interneurons that are active during the ventilatory rhythm (b). Is this circuit sufficient to generate the ventilatory rhythm, or are additional neurons required? In order to test this, cell bodies of the three neurons were carefully dissected from the central nervous system and placed in a dish containing tissue culture medium. Quite quickly, the isolated cell bodies began to sprout new processes and, within a day, the three neurons established synaptic connections with each other, in the same pattern as in the intact brain. Brief electrical excitation of one of the neurons caused several cycles of rhythmical activity very similar to that found in an intact brain. None of the neurons was capable of generating rhythmical activity on its own and, although I.P3.1 and V.D4 (which connect with motor neurons) made a simple circuit in which each inhibited the other, rhythmical activity was not produced unless the giant, dopamine-containing cell (Rpe.D1) was also present.

Pneumostome

Inspiration

Pneumostome

Inspiration

Essentials of Human Physiology

Essentials of Human Physiology

This ebook provides an introductory explanation of the workings of the human body, with an effort to draw connections between the body systems and explain their interdependencies. A framework for the book is homeostasis and how the body maintains balance within each system. This is intended as a first introduction to physiology for a college-level course.

Get My Free Ebook


Post a comment