Ephaptic coupling improves the neural population code

Neurons share an extracellular medium, so every spike's current sets up a local field that feeds back onto the membrane potential of nearby cells, an interaction known as ephaptic coupling. The local field is usually studied by imposing it externally, for example by injecting sinusoidal currents through an extracellular electrode, which entrains and synchronizes firing. However, an endogenous field, self-generated by the network's own spiking, is a different phenomenon whose functional role stays unclear, because it cannot be removed without silencing the neurons that produce it. We built matched spiking networks with and without a self-consistent endogenous field, and a rate-matched control separating the field from firing rate. The results show that the origin of the field determines the effect. A field the network generated for itself decorrelated the population, expanded its dimensionality, and improved decoding, whereas the same field imposed from outside did the opposite, synchronizing the population and reducing coding performance. The benefit held across a spatial and a temporal coding task and survived rate-matching. A closed-form theory connects the decorrelation to the field acting as a spatial filter on the slow component that neurons share. To look for this in cortex, we analyzed multi-patch recordings, showing that a neuron's spikes left a distance-dependent trace in unconnected neighbors which we reproduced by the same field model. The self-generated field is therefore not a by-product but a mechanism that, without wiring, can improve the population code and reshape the geometry of neural population dynamics.