Fig. 5: A normalization model using shunting inhibition explains luminance gain control in medulla neurons.
From: Neural pathways and computations that achieve stable contrast processing tuned to natural scenes
a Sketch of the proposed circuit implementing normalization (luminance gain control) via spatial pooling and shunting inhibition. The Tm neuron’s main LMC input is normalized by the input of a wide-field neuron via shunting inhibition. Tm1 circuit with the main L2 presynaptic input and a linear normalization factor. b Simulated L2 (blue) and Tm1 (magenta) traces based on the normalization model in (a). c Amplitude of contrast responses (F1 amplitude) across luminance from the simulated Tm1 responses (magenta, solid lines) and experimental data from Tm1 (magenta, dashed) and L2 (blue, dashed) neurons. Experimental data from Figs. 1, 2. d Tm9 circuit with the main L3 presynaptic input and a non-linear normalization factor. e Same as (b) for L3 (green) and Tm9 (teal). f Same as (c) for Tm9 and L3. Simulated Tm9 responses (teal, solid lines) and experimental data from Tm9 (teal, dashed) and L3 (green, dashed). g Imaging the glutamate input onto Tm1 and Tm9 dendrites using iGluSnFR. h Ternary white-noise stripes to extract spatio-temporal receptive fields (STRF) via reverse correlation analysis. Left: STRF of Tm1 dendritic glutamate (magenta frame) and Tm1 calcium at the axon terminals (dark magenta frame). Color axis depicts positive (ON)—negative (OFF) correlation with the stimulus. Middle: Spatial filters of STRFs. Right: Quantification of the FWHM of the spatial filters. i Same as (h) for Tm9 neurons. j iGluSnFR responses of a single Tm1 (magenta) and Tm9 (teal) neurons to drifting 1 Hz gratings of constant 100% Michelson contrast and changing luminances. k Mean normalized contrast responses (F1 amplitude) of Tm1 dendritic glutamate (magenta) Tm9 dendritic glutamate (teal) signals. Tm1 and Tm9 calcium signals at the axon terminals (dark teal) for the same stimuli are shown for comparison (from Fig. 2). l Slopes of the contrast responses depicting the log-luminance dependence for each genotype. **p < 0.01, ***p < 0.001, two-sided Student’s t-test. P values: Tm1»GCaMP6f, Tm1»iGluSnFR = 7.2124e-05, Tm9»GCaMP6f, Tm9»iGluSnFR = 0.0056. Error bars represent ±SEM. Means are calculated across cells for (h, i) and across flies for (k, l). Model values in (c, f) come from 1000 simulated traces, using sinusoidal stimuli with a random phase to simulate different spatial locations. Sample sizes are given as # cells for (h, i) and # flies(cells) for (c, f, k, l). For (c, f), L2 exp n = 19(141), Tm1 exp n = 11(148), L3 exp n = 8(150), Tm9 exp n = 9(111). For (h, i), Tm1»iGluSnFR n = 23 cells, Tm1»GCaMP6f n = 151 cells, Tm9»iGluSnFR n = 39 cells, Tm9»GCaMP6f n = 53 cells. For (j–l), Tm1»GCaMP6f: n = 11(148), Tm1»iGluSnFR: n = 10(78), Tm9»GCaMP6f: n = 9(111), Tm9»iGluSnFR: n = 7(23).
