Venue: Centre Broca
Naoya Takahashi (IINS)
Romain Caze
CNRS – IEMN (Lille).
Where: Broca Auditorium
Title: On the computational capacity of neurons
Abstract
Nonlinear dendritic operations are thought to enhance neurons’ computational capacity. In contrast to the supralinear dendritic integration, generated by dendritic spikes in pyramidal neurons, interneurons have been shown to exhibit sublinear dendrites. Our theoretical work has shown that sublinear dendrites permit the implementation of linearly-nonseparable function that cannot be implemented by linear operations. Such sublinear linearly-nonseparable computations are predicated on the neuron being selectively sensitive to synaptic signals scattered across multiple electrotonically distant sites (e.g. different dendritic branches).
We set out to test this prediction experimentally using glutamate uncaging to mimic synaptic activation of dendrites and whole-cell patch clamp recordings to monitor subthreshold and suprathreshold somatic activity. Consistent with the theoretical predictions we found that clustered synaptic-like activation of single dendrites evoked compound excotatory potential that were smaller than if a similar number of synaptic-like stimuli were distributed across two different dendrites. Moreover, we showed that this subthreshold observation generated an output spiking probability that was higher for distributed inputs. These results are consistent with a “scatter sensitive” synaptic input-output transformation. We were also to directly demonstrate that a neuron could implement a linearly non-separable computation of its synaptic inputs, provided we adjusted the resting membrane potential appropriately. These data were confirmed with biophysical model simulations which also allowed us to generalize the findings to presynaptic activity patterns involving many more synapses. Because our findings require only sublinear summation of synaptic potentials, we can generalize that all neurons exhibiting sublinear summation will exhibit enhance computational abilities. How these specific neuronal computations can be leveraged for circuit computations under specific behavioral contexts will merit future theoretical and experimental investigation.