Is there a signal in the noise?

The irregular pattern of neural discharge has captured the interest and imagination of neuroscientists since Lord Adrian [1]. To comprehend the nature of information transmission in cerebral cortex, we must determine whether constellations of action potentials their intervals, coincidences, and precise timing relationships convey information or whether they arise capriciously. Do specific conditions cause the neuron to spike in one millisecond and not another, or are such variations the random instantiations of a desired discharge rate? Do synchronized spikes from two neurons encode the precise occurrence of some event, or do such spikes arise merely as manifestations of shared connectivity, bearing no special significance for information transmission? Whether the irregular spike discharge from cortical neurons reflects noise or information clearly depends on the rules governing the conversion of synaptic input to spike output. In our recent review for Current Opinion in Neurobiology [2], entitled 'Noise, neural codes and cortical organization', we suggested that cortical neurons fire irregularly because they are inundated with synaptic input [2]. Therefore, to achieve a graded, dynamic range of spike rates, the cortical neuron might balance excitation and inhibition. The statistics of this process would lead to an irregular interspike interval, effectively randomizing the spike output. If this is the case, then the precise pattern of spikes from a neuron is no more likely to convey specific information than the pattern of tea leaves at the bottom of a cup. One can only infer whether the neural response, hke tea, is stronger or weaker. Likewise the cerebral cortex must be organized to transmit and detect rate changes through noisy elements, necessitating redundancy and weak correlation. A contrasting viewpoint is presented in the preceding paper (see pp 239-247, in this issue), in which William Softky makes an interesting case for precise temporal signaling capacities in cortical neurons. Softky suggests that present knowledge of synaptic integration does not preclude a precise temporal coding scheme, and he has constructed a hand-tuned neuronal model that could implement such a scheme. Softky's model incorporates active dendritic conductances balanced by strong inhibition, together permitting only precise coincidences of synaptic excitation to propagate from the dendrite to the axon hillock. Were this the case, the neuron could signal the occurrence of certain combinations of presynaptic events with a temporal fidehty on the order of a millisecond or less (i.e. well under the average interspike interval). An action potential arises only from synaptic activity in the preceding millisecond, whereas inputs that fail to produce a spike in the subsequent millisecond are effectively erased. This notion appears fanciful to us because it extrapolates so far beyond existing data, but it remains a logical possibility, and we agree that it would permit the neuron to propagate a coincidence code. In fact, Softky's hand-tuned model is an excellent example of the sort of mechanism required of any scheme in which downstream neurons attach significance to synchronous spikes [3-7].