Adaptation

Adaptation (also known as accommodation or neural fatigue) is the process that causes neurons to decrease their responsiveness over time, as a function of how active they have been. It is driven biologically by several different adaptation channels that operate over different lengths of time, most of which affect potassium (K) channel conductances.

The functional benefit of adaptation is to produce a novelty filter dynamic, where novel inputs hitting “fresh”, non-adapted neurons generate a stronger response relative to ongoing inputs that have already been processed, and are thus driving adapted neurons (Benda, 2021). You can explore this behavior in the individual neuron simulation, by toggling the state of various adaptation channels.

In addition to the various K channels, synaptic depression (Abbott et al., 1997) has been proposed as an important mechanism of synapse-specific adaptation, which would potentially be significantly more selective than neuron-level adaptation mechanisms. However, studies in awake, non-anesthetized animals have shown that spontaneous levels of ongoing neural activity are sufficient to effectively put most synapses in a state of tonic depression, thus limiting the functional contributions of this mechanism (Boudreau & Ferster, 2005, Borst, 2010, Buchholz et al., 2023).

Borst (2010) does a particularly thorough review of the available literature and concludes that synaptic depression is unlikely to play a significant role in neural computation in awake animals. He raises a number of points that also apply generally to the distinction between studies done on slices or cultures of neurons (in vitro) versus the awake, intact animal, which we have termed in activo, as distinct from in vivo which includes anesthetized animals. In general, as Borst (2010) notes, the behavior of the awake brain tends to be more linear overall, with less short-term plasticity and less bursting.

Another issue with synaptic depression is that it is largely driven by the spiking of the presynaptic neuron, and has little dependence on the activity state of the postsynaptic neuron (Neher & Sakaba, 2008). Therefore, it doesn’t really provide any more specificity than an adaptation mechanism operating on that presynaptic neuron as a whole. Computationally, this is convenient because it is much more expensive to simulate mechanisms at the synapse vs. those operating at the neuron level.