Fig. 3: Gabor filter model links orientation tuning and surround suppression to broadband responses.

a Schematic of a Gabor filter bank to model neural responses. Gabor filters of different orientations and spatial frequencies were used to jointly predict the measured responses to the presented broadband stimuli (see Supplementary Figs. S3 and S4 for details). Model predictions were either based on the responses of a central Gabor alone (“tuning only” model; bottom left, yellow box) or by additionally accounting for the impact of surround suppression (“suppression” model; bottom right, red box). b Mean response amplitude towards narrow bandwidth stimuli, depending on the orientation tuning of V1 neurons (n = 11 sessions in 4 mice). Neurons in each session were binned based on their orientation tuning to a set of full-contrast sinewave gratings. Yellow circles and lines show the best-fitted response of the tuning-only model. Red circles and lines show the best fit of the suppression model. (c) Same as (b) for broad orientation bandwidth stimuli. Both the tuning only and the suppression model were fit to narrow and broad orientation bandwidth stimuli at the same time. d Mean response modulation of orientation-tuned cells by expanding the orientation bandwidth (derived from responses in b and c). Yellow circles and lines show the predicted response modulation from the tuning-only model. Red circles and lines show predicted response modulation from the suppression model. The horizontal black dotted line at 0 indicates no bandwidth modulation. e Same as panel d) but for recruitment modulation of orientation-tuned cells by expanding the orientation bandwidth. The ORI recruitment modulation index was calculated as the difference between the numbers of responsive neurons to narrow and broad orientation bandwidth stimuli divided by their sum. Yellow and red circles indicate fits of the tuning only and surround suppression model, respectively. Box plots indicate the median (horizontal line), interquartile range (box bounds: 25th–75th percentiles), and whiskers (1.5 × interquartile range). Stars in panels (d and e) mark significant differences against 0 based on an LME model (two-sided test). The significance threshold for each test was α = 0.01 after Bonferroni correction for five individual tests. Panel a was created in BioRender. Balla, E. https://BioRender.com/g05l789 (2025).