[ print ]

Computational Neuroscientist; Francis Crick Professor, the Salk Institute; Coauthor, The Computational Brain
I have changed my mind about cortical neurons and now think that they are far more capable than we ever imagined.

How is it that insects manage to get by on many fewer neurons than we have? A fly brain has a few hundred thousand neurons, compared to the few hundred billion in our brains, a million times more neurons. Flies are quite successful in their niche. They can see, find food, mate, and create the next generation of flies. The traditional view is that unique neurons evolved in the brain of the fly to perform specific tasks, in contrast to the mammalian strategy of creating many more neurons of the same type, working together in a collective fashion. This view was bolstered when it became possible to record from single cortical neurons, which responded to sensory stimuli with highly variable spike trains from trial to trial. Reliability could be achieved only by averaging the responses of many neurons.

Theoretical analysis of neural signals in large networks assumed statistical randomness in the responses of neurons. These theories used the average firing rates of neurons as the primary statistical variable. Individual spikes and the times when they occurred were not relevant in these theories. In contrast, the timing of single spikes in flies has been shown to carry specific information about sensory stimuli important for guiding the behavior of flies, and in mammals the timing of spikes in the peripheral auditory system carried information about the spatial locations of sound sources. However, cortical neurons did not seem to care about the timing of spikes.

I have changed my mind about cortical neurons and now think that they are far more capable than we ever imagined. Two important experimental results pointed me in this direction. First, if you repeatedly inject the same fluctuating current into a neuron in a cortical slice, to mimic the inputs that occur in an intact piece of tissue, the spike times are highly reproducible from trial to trial. This shows that cortical neurons are capable of initiating spikes with millisecond precision.  Second, if you arrange for a single synapse to be stimulated a few milliseconds just before or just after a spike in the neuron, the synaptic strength will increase or decrease, respectively. This tells us that the machinery in the cortex is every bit as capable as a fly brain, but what is it being used for?

The cerebral cortex is constantly being bombarded by sensory inputs and has to sort though the myriad of signals for those that are the most important and to respond selectively to them. The cortex also needs to organize the signals being generated internally, in the absence of sensory inputs. The hypothesis that I have been pursuing over the last decade is that spike timing in cortical neurons is used internally as a way of controlling the flow of communication between neurons. This is a different from the traditional view that spike times code sensory information, as occurs in the periphery. Rather, spike timing and the synchronous firing of large numbers of cortical neurons may be used to enhance the salience of sensory inputs, as occurs during focal attention, and to decide what information is worth saving for future use. According to this view, the firing rates of neurons are used as an internal representation of the world but the timing of spikes is used to regulate the communication of signals between cortical areas.

The way that neuroscientists perform experiments is biased by their theoretical views. If cortical neurons use rate coding you only need to record, and report, their average firing rates. But to find out if spike timing is important new experiments need to be designed and new types of analysis need to be performed on the data. Neuroscientists have begun to pursue these new experiments and we should know before too long where they will lead us.