How does your brain represent the feelings and thoughts that are a part of conscious experience? Even the simplest aspects of this question are still a matter of heated debate, reflecting science’s continuing uncertainty about “the neural code.” The fact is that we still don’t have a clear picture of the ways in which neurons transmit information. Here’s a quick guide to current theories, beginning with well-established theories and moving into ideas that are considered more theoretical.
The canonical model: firing rate
Clearly, neurons encode some information in the rate of their firing. Although individual neurons are relatively unreliable and noisy devices, average firing rates across hundreds or thousands of neurons provides a more reliable spatio-temporal code for conveying information. Some recent evidence, reviewed below, suggests that this simple conception of the neural code is in need of elaboration.
Firing rate coherence
Our connection to the external world occurs only through the thalamus, through which all sensory signals (except olfaction) must pass in order to gain access to the neocortex. Although our sensory systems are exquisitely sensitive (9 photons or less are sufficient for inducing a conscious visual experience, and observers can report indentations of their skin only 1-3 microns deep), connections between thalamus and neocortex are surprisingly weak (by some estimates, 30 times weaker than intracortical connections).
Recently, Bruno & Sakman demonstrated that this incredible sensitivity may be enabled by the synchronous firing of neurons in the thalamus. According to this model, the relatively weak thalamocortical connections can be “amplified” by the coordinated and synchronous firing of populations of thalamic neurons.
Synchronous firing also seems important for motor functioning in healthy subjects, and there are some indications that reduced synchrony, but not reduced firing rate, may underlie some symptoms of Parkinson’s disease.
Precise Spike Timing: Phase relationships in Neural Firing
Two populations of neurons may fire at the same rate; these two populations may also show phase synchrony, where the peak activity in one population occurs at nearly the same time as that of another population. Yet there might be additional information hidden in the temporal relationship of these populations. Such a mechanism would be far more sophisticated than firing rate and neuronal synchrony alone: information might be conveyed in a “relational code” between the spikes comprising the activity in each population.
There is emerging evidence that even such tiny temporal differences might have an important computational role in neural networks. For example, synaptic efficacy may be particularly malleable when action potentials fire with a particular temporal relationship – known as “spike timing dependent plasticity,” or STDP.
Others have argued that behavioral responses can often be initiated too quickly for “firing rate” to be a reliable mechanism of generating these responses. According to this argument, “time-to-first-spike” might be a faster and more reliable carrier of information – and indeed, recent simulations indicate it may be 10-20 msec faster than firing-rate codes.
N:M Nested Oscillations: phase-relationships in firing rate coherence
In their in-press TICS article, Jensen & Colgin review evidence from direct, intracranial electrode recordings in humans (perhaps the holy grail of current recording methods, allowing for an unsurpassed combination of spatial and temporal precision). In one such study, the amount of activity occuring at relatively high frequencieis (30-150Hz, aka the gamma rhythm) was systematically modulated by a much slower frequency peak in the spectrum of neuronal oscillations (5-8 Hz, aka the theta rhythm) but that the same relationship did not exist with other slow oscillations. This “cross-frequency coupling” (also known as “n:m phase synchrony“) was observed across a wide swath of cortex, and across a wide variety of tasks, suggesting it may have a central role in neural communication.
This claim has been met with much resistance in the neuroscience community, partially because it’s unclear how such precise “multiplexed” signals might emerge from real neural networks, with all their inherent noise and apparent randomness. As an implicit response to this question, Jensen & Colgin point to a neural network model in which slow and fast GABAergic feedback signals give rise to concurrent theta and gamma waves.
What computational role might these multiplexed, cross-frequency phase couplings have? The authors indicate that if the faster oscillation (gamma) functionally “divides” the slower oscillation into multiple time slots, then each of those slots might convey different information. This idea has given rise to a model for the physiological basis of working memory capacity (at least, the estimate of 7 +/- 2, now known to be a little high), for Sternberg scanning, and for phase precession in the hippocampus during spatial navigation.
This is a highly selective list of what I feel are the most likely candidates for helping us to understand the neural code. Certainly there are others I’ve left out – including Hameroff & Penrose’s claims about neuro-quantum computation, and other recent proposals involving computation with soliton waves.