The Problem of Neural Coding

Spikes as the language of the brain

It is common practice in many laboratories to display the electrical activity from one or more neurons in an animal while it looks at, hears, and reacts to its environment. This neural activity appears as a sequence of very brief events, the action potentials, separated by gaps of variable duration. The intervals between spikes can be as long as a few tenths of a second or shorter than a hundredth of a second. The spikes and the intervals between them codify the neuron’s message. If we wish to decipher the neural code, we need to know how to interpret these messages.

It is well established that the only type of message that a neuron can send to another neuron, in a different part of the brain, must be represented in the sequences of spikes that are transmitted along its axon. The time-scale for neural computations involved in perception, thought, and action is too short to allow gene expression, protein synthesis, and chemical cascades to play a part in carrying information. Spikes are the only items in the alphabet, but unlike letters, they are of one kind only. Spikes are all-or-none events: size does not matter. As Rieke and colleagues put it, spike sequences are the language for which the brain is listening, the language the brain uses for its internal musings, and the language it speaks as it talks to the outside world [1].

The problem of neuronal coding

The question is how to read this sequence of spikes emitted by neurons as a function of time. What are the properties of the neural spike trains that provide the possibility to carry information or take part in information processing? What is the information contained in such a spatio-temporal pattern of pulses? What is the code used by the neurons to transmit that information? How might other neurons decode the signal? The above questions point to the problem of neuronal coding.

At present, a definite answer to these questions is not known. In the theory of neuronal information processing, there are two main hypotheses with respect to where in the spike train the neural information is encoded: whether in the neural firing rate or in the precise timing of the spikes. They are discussed in the following.

The rate coding hypothesis

Until recently, the most popular hypothesis was that most of, if not all, the relevant information was contained in the mean firing rate of the neuron, defined by the number of spikes that occur in a time window, divided by the length of the window. The concept of mean firing rates has been successfully applied during the last 80 years. In a seminal contribution of more than 75 years ago, Adrian showed that the firing rate of stretch receptor neurons is related to the force being applied to the muscles [2]. In [3], Adrian and his colleagues showed that each fiber originating from skin receptors responded to a particular type of stimulus (pressure, temperature or damage) and the frequency of the impulses was dependent upon the intensity of the stimulus.

The definition of the rate has been applied to the discovery of the properties of many types of neurons in the sensory, motor, and central nervous system, by searching for those stimuli that make neurons fire maximally. Further support for the frequency code theory came from the work by Eccles [4], based on spinal cord neuronal recordings. Eccles observed that the spike frequency of a peripheral nerve correlated with the intensity of the applied stimulus, and with the intensity of sensation as well. Later it was shown that all sensory fibers which terminate in the spinal cord express stimulus intensity dependent discharge rates [5]. Similarly, the frequency of action potentials recorded from neurons in the motor cortex [6], throughout the cortico-spinal tract [7] and motor units [8], correlated with the tone of the target muscle.

Why rate coding has been criticized

Since Adrian’s studies, the rate coding hypothesis has been dominant in the neural computational field. Rate coding explains how the presentation and intensity of the stimulus can influence neural activity, but this coding neglects the temporal organization of spike trains. For this reason, the firing rate concept has been criticized and it is the subject of an ongoing debate, stimulated also by experiments showing the importance of the temporal dimension in neural information processing.

The temporal coding hypothesis

Recently, more and more experimental evidence suggested that a straightforward firing rate concept based on temporal averaging may be too simplistic to describe brain activity. Experiments showed that temporal patterns matter. Indeed, for any rate there are an infinite number of possible temporal distributions of spikes. Vice versa, it has been experimentally shown that different stimuli or tasks can elicit different patterns of activity that have the same firing rate. In principle, taking into account the temporal structure of a spike train would expand the alphabet that the brain uses to encode information.

Recent observations on the behavior of cortical visual neurons demonstrated a precision in brain function timings higher than would be predicted from frequency coding. Humans, for example, can recognize and respond to visual scenes in less than 400 ms [9]. Recognition and reaction involve several processing steps from the retinal input to the finger movement at the output. If at each processing step neurons had to wait in order to perform a temporal average, the reaction time would be much longer. This result suggests the existence of computational processes based on the precise timing of spikes in neuronal ensembles. In the last decade, the focus of attention in experimental and computational neuroscience has therefore shifted towards the exploration of how the timing of single spikes is used by the nervous system.

There is evidence of precise temporal correlations between pulses of different neurons [10] [11] and stimulus dependent synchronization of the activity in neuronal populations [12] [13] [14]. These results suggest that the exact timing of spikes should play an important role. In the following, different types of rate and temporal codes are briefly introduced, each related to a strategy based on a different notion of rate or spike timing.

Domande frequenti

What is the neural coding problem?

It is the question of how neurons represent and transmit information through their electrical activity. Since the only signal a neuron sends along its axon is a sequence of spikes, deciphering the neural code means understanding which features of those spike trains carry information about the environment and our mental states.

What is the difference between rate coding and temporal coding?

Rate coding assumes that information is contained in the mean firing rate, the number of spikes in a time window divided by its length. Temporal coding holds that the precise timing of individual spikes also carries information, since different stimuli can produce different temporal patterns even at the same firing rate.

Why is rate coding alone considered insufficient?

Because temporal averaging would be too slow to explain rapid behavior: humans can recognize and respond to visual scenes in under 400 ms. If every processing step waited to average spikes over time, reaction times would be far longer, which points to computation based on precise spike timing.

Are spikes all the same?

Yes. Spikes are all-or-none events, so their size does not matter, and they are of one kind only. The information lies in their sequence and timing, which is why spike trains are described as the language the brain uses to listen, to reason internally, and to communicate with the outside world.