Neurobiology: Understanding the consequences

20 June 2002



			Nature 417, 803 - 806 (2002); doi:10.1038/417803a

Neurobiology: Understanding the consequences

THOMAS J. CAREW

Thomas J. Carew is in the Department of Neurobiology and Behavior, 2205 McGaugh Hall, and the Center for the Neurobiology of Learning and Memory, University of California at Irvine, California 92697-4550, USA.
e-mail: tcarew@uci.edu

We learn in several ways, one of which involves forming an association between an action and its consequence. Studies of a marine mollusc shed light on how this creature forms a connection between biting and food.

The ability to assess the consequences of one's actions is fundamental to survival: an animal must learn an effective hunting strategy if it is to eat, and to elude predators if it is to live to see another day. Writing in Science, Brembs, Lorenzetti and colleagues¹ describe their studies of the neural basis of this type of learning in the marine mollusc Aplysia, which could serve as a useful model for understanding more complicated organisms.

This general class of learning is known in the trade as 'operant' or 'instrumental' conditioning². It was first brought into the laboratory over a century ago by Thorndike³, who studied the ability of cats to learn to escape from a 'puzzle box'. At about the same time, another form of learning was discovered by Pavlov⁴, who described how animals form an association between a neutral (also termed a 'conditioned') stimulus, such as a ringing tone, and an 'unconditioned' stimulus that has inherent meaning, such as food. This is known as classical or pavlovian conditioning. Together these two types of learning provide most of the tools that all animals need to negotiate their environment successfully, enabling them to associate their behaviour with its consequences and to learn predictive relationships about the world.

In modern neuroscience, the analysis of learning mechanisms is a thriving enterprise. But although classical and operant conditioning are both important, the mechanisms of classical conditioning have received far more attention. Why should this be? The reason lies, at least in part, in how researchers approach the problem. The basic goal of a neurobiological analysis of associative learning is twofold: first, to identify the site of the association in the brain; and second, to characterize the neural mechanisms involved in forming the association. For classical conditioning, this strategy is relatively straightforward. One would first identify the pathways of neurons that respond to the conditioned and unconditioned stimuli. The points at which the two pathways converge would be good candidates for the sites at which an association between the stimuli is formed.

But matters are potentially more complicated in the case of operant conditioning. Here, an association is made not between two stimuli but between an action and its consequence, such as a benefit (or 'reward', in learning parlance) or punishment. So the site of association is not intuitively obvious. For example, it could occur where information about the reward converges with the brain region that initiates the behaviour, which could be more difficult to locate than the regions that respond to conditioned and unconditioned stimuli.

Brembs et al.¹ have overcome these difficulties by studying the operant conditioning of feeding in Aplysia (Fig. 1). The authors attacked the problem at several levels, from the behaviour of the whole animal down to the electrophysiological properties of single neurons. They focused on feeding in Aplysia because this is known to be capable of operant conditioning⁵ and, more importantly, because the neural circuitry underlying feeding has been well characterized.

Figure 1 Learning model — the marine mollusc Aplysia. The new paper by Brembs et al.¹ gets to grips with how Aplysia learns to associate biting with a food reward, a type of 'operant' conditioning.
High resolution image (47k)

First, Brembs et al. looked at the electrical activity of the oesophageal nerve in whole animals, and found that it increased when the animals ingested food. Presumably, this activity signals the presence of a reward — food. Next, Brembs et al. 'trained' the animals to associate spontaneous biting (whether or not food was ingested) with a reward by stimulating the oesophageal nerve, mimicking the usual reward signal, during biting. The result was that the molluscs made significantly more spontaneous bites than controls, both immediately and 24 hours after training. In other words, the operant response (biting) can be reinforced by a food-related reward signal (stimulation of the oesophageal nerve); moreover, the memory of the association between biting and reward can persist for at least 24 hours.

The authors then turned their attention to where this memory might be stored. Here, the previous detailed characterization of the neural circuitry underlying feeding behaviour was a big help. In the central nervous system, a particular group of neurons — the buccal ganglia — controls biting, and the authors studied one of these, called B51, because it is essential in generating the correct programme of neuronal activity. By recording the electrical activity of B51 neurons in buccal ganglia that had been surgically isolated from Aplysia, Brembs et al. showed that the burst threshold was lower and the input resistance higher in neurons from trained animals than in controls. The changes in these properties together increase the likelihood that B51 will become active, and hence improve its ability to generate ingestion-related neural programmes.

So it seems that operant conditioning can alter certain properties of the B51 neuron. But it was not clear whether B51 is a genuine site of convergence between bite behaviour and reward — that is, if it is where the association is formed and stored — or whether it is simply affected by that site. To find out, Brembs and colleagues isolated and cultured B51 neurons and examined whether similar changes in properties could be induced by mimicking the operant conditioning procedure at the single-neuron level. They paired the activation of B51 (that would generate a bite) with brief pulses of dopamine, a neuromodulator that probably serves as the reward signal in this system⁶ and many others⁷. They found a significant reduction in burst threshold and increase in input resistance in B51. This did not occur when dopamine application and B51 activation were unpaired.

How do these observations relate to the learned association between biting and reward? The idea is this. B51 activity is required for an animal to attempt to take a bite of food. If that attempt is successful — if food ends up in the mouth — the oesophageal nerve is stimulated. This causes the release of dopamine in the buccal ganglia, increasing the excitability of B51. Given that direct activation of B51 can elicit ingestion-related activity in isolated ganglia⁸, one can envisage that a direct consequence of enhancing B51 excitability would be a greater likelihood of further biting.

So, collectively the data show that B51 is directly affected by operant conditioning, implying that it is at least one important site of memory storage. Moreover, an exciting aspect of the study is that it provides a neuronal mechanism for operant conditioning that can explain not only its associative features, but also how that associative component might control future biting behaviour.

But the significance of the paper extends further. Now that an associative site for operant conditioning has been found, it will be possible to examine the underlying cellular and molecular mechanisms in detail — though this is no easy chore, to be sure. Moreover, Aplysia shows classical conditioning⁹, and much is known about the molecular and cellular mechanisms underlying this form of learning^10-12. So it might be possible to compare classical and operant conditioning in Aplysia in mechanistic terms. If they have features in common, an exciting principle might emerge: evolution may have come up with a neural 'associative cassette' that can be used in either type of conditioning, depending on the neural circuit in which it is embedded. Of course, this is pure speculation, but the work by Brembs and colleagues will be instrumental in exploring this intriguing possibility.