FIGURE 4. The effect of STDP on -LN spike timing.

a, Polar plot of Kenyon cell spike phase in the calyx (somata) (KC_C) and in the -lobe (KC_L), and -LN spike phase (from dendrites in the -lobe) relative to the LFP (in the calyx). All measurements from experiments. Green and red lines indicate extrema of the STDP curve (see panel c). b, Schematic of temporal relationships between LFP, Kenyon cell spike time, -LN spike time and the STDP rule. The Kenyon cell mean spike time in the calyx (KC_C) is about /2 after the LFP peak, and near the LFP trough in the -lobe (KC_L), owing to propagation delay. -LN mean spike time in the -lobe is at the LFP trough ( rad). The STDP curve is represented in colour gradients. The predicted effect of STDP on -LN spike time is schematized underneath. If the -LN spike occurs early, STDP should depress late Kenyon cell inputs (in this oscillation cycle), delaying this -LN spike at the next opportunity. The converse applies if the -LN spike occurs late. c, The STDP fit (two exponentials flanking a linear segment, see text) overlaid on experimental data. d, Simulations of STDP on -LN spike time (rasters) (model -LN excited by 10 model Kenyon cells during one LFP cycle). First trial at top. Three conditions are illustrated: left panel, low input weights (mean, 1.8 mV; range, 3 mV), causing late -LN-spike times (dt > 0); when STDP is turned on, potentiation shifts -LN spikes to earlier times. Right panel, high input weights (mean, 9 mV; range, 3 mV); when STDP is turned on, depression delays -LN-spike times. Middle panel, intermediate weights, causing no change. Histograms show the distribution of -LN spike times before (red) and after (black) STDP (1,000 runs per condition). e, Evolution of KC–-LN weights (upper panel), -LN mean spike phase (middle panel) and number of spikes per response (lower panel) over 50 trials following onset of STDP (at trial 1). AP, action potential. Each curve is an average of 200 simulations (11 different input distribution means). Asterisks indicate a corresponding condition in the three plots. f, Schematic of experimental design. LFP cycles (1...k) during which the recorded -LN-produced action potentials are selected. Peaks of LFP are used to trigger a sequence of current (I_m) pulses (a, b, c) into the -LN (one such sequence per oscillation cycle). The a-b-c sequence lasts less than one oscillation cycle, and is repeated for all selected cycles, over several trials. g, Example of protocol described in panel f, such that the -LN spike is phase-delayed to 3/2 (arrows). (Interrupted segments of -LN potential trace are bridge-balance artefacts.) h, Phase plot of spikes in one -LN before and after pairing (10 trials each), as in panel g. Fifteen pairing trials (estimated dt = 17 ms). i, Distributions of pairing-induced mean phase shifts, measured in 20 separate experiments (6 -LNs; mean  s.e.m. = -0.74  0.4 ms versus 2.0  0.4 ms).