Many scientists currently assume that a pattern of neural activity codes for something in the world – maybe a feature of an object, an action plan, a word or an idea. For example, when we see something red, some neurons in the visual cortex fire, and others remain silent, and this signature pattern indicates the colour of what we are seeing.
Researchers at the University of Oxford, working with colleagues from Princeton University, have shown that neurons can rapidly change their selectivity, depending on many contextual factors. For example, if we are looking for something red, some neurons in the frontal lobe fire selectively for red items, but if we are looking for something green, those same neurons might fire selectively for green items.
To date, this adaptive coding has been observed but never modelled precisely. For a neuron to be adaptive, it would need to change its synaptic input weights. And for this neuron to be useful, the output synaptic weights may also need to change. Associate Professor Sanjay Manohar and team previously used a simple rule called 'Hebbian plasticity' to simulate flexibly coding neurons, and showed that these can perform several useful tasks. In their previous paper they made a prediction about how these flexible codes might reveal themselves in a real brain (Manohar et al. 2019).
In this latest research, published in PNAS, they looked for data that might be able to test these predictions. The researchers identified some suitable, previously collected data from the prefrontal cortex of monkeys who had to remember coloured items. Signatures of plasticity were present in those data.
The team found that when they simulated plasticity rules, they produced distinctive signatures in the pattern of neural activity, that changed after seeing different kinds of information. For example, if we see something red twice, the same neurons would be active each time. But if we see something else, such as blue, in between the two red items, then this red-to-red similarity breaks down.
The great power of neurons to remember things may come from the fact they have thousands of arms (axons and dendrites), each of which can store an independent piece of information. If we simulate the idea that each of these arms can rapidly store information, this produces a distinctive signature of when the neuron will become active.