Is it the neuron or the wiring? Swapping spiking for rate neurons in a whole-brain model of the fly

Whole-brain models of the fly built on the FlyWire connectome use one neuron model, the leaky integrate-and-fire (LIF) model, in which a neuron fires an all-or-none spike whenever its voltage crosses a threshold. When such a model fails a behaviour, there is no standard way to tell whether the neuron model or the wiring is to blame. In my earlier LIF model, odour learning worked but sugar never recruited the dopamine neurons that signal reward. Here I replaced every LIF neuron with a rate neuron, which has no spikes and outputs a firing rate that rises and falls smoothly with its input. I calibrated the rate model to the LIF model, kept the wiring unchanged and preregistered every test. The rate model predicted the LIF model's responses to held-out odours. Sugar still did not reach the reward neurons, even after fitted gains let the Fox interneurons drive them, which is consistent with a sparse route from sugar-sensing neurons to Fox. Odour learning and the accumulation of a second memory survived the swap. The rate model also relearned in every cycle at the mushroom body, but it failed the preregistered relearning test, because one training block lowered the response of the approach output neurons by 13% in the rate model against 40% in the LIF model, even though the rate model's synapses changed more. In this model, whether a memory forms belongs to the wiring, the settings shared by both models and the plasticity rule, while how strongly it first shows depends on the neuron model.