Fig. 2: Head-fixed category learning depends on plasticity in early visual areas.

a, Head-fixed conditioning setup. A head-fixed mouse (m) is placed on an air-suspended Styrofoam ball, facing a computer monitor (v). Two lick spouts (r) in front of the mouse supply water rewards and record licks. b, The fraction of left/right choices of seven mice (data of stages II to VI). The grid (top left) shows an example stimulus space. Solid lines indicate fitted, individually learned category boundaries, while dashed lines indicate trained boundaries. c, Illustrations (bottom) showing systematic displacement of the stimulus position by repositioning the monitor. Per-stimulus category performance (top), as in b, but for different stimulus positions (averaged across training sessions and mice). The default stimulus position used during preceding category training was at 26° azimuth. For display purposes, the grids are flipped such that the stimulus-to-category mapping is similar across animals (top left, ‘lick-left’; bottom right, ‘lick-right’). d, Sigmoid fit of the fraction of left choices (mean ± s.e.m. across mice; n = 5), as a function of the stimulus’ distance to the category boundary for different stimulus positions. e, Fraction of correct trials for default (26°) and shifted (0°, −26°) stimulus positions (mean ± s.e.m. across mice, n = 5; gray lines represent individual mice; two-sided Kruskal–Wallis test, H₍₂₎= 6.1, P = 0.046; post hoc one-sided WMPSR test, −26° versus 26°: W = 0, P = 0.031; 0° versus 26°: W = 0, P = 0.031; n = 5 mice). f, As in e, for categorization steepness (of the sigmoid fit; two-sided Kruskal–Wallis test, H₍₂₎= 6.0, P = 0.049; post hoc one-sided WMPSR test, −26° versus 26°: W = 15, P = 0.031; n = 5 mice). g, As in e, for the boundary angle difference between trained and individually learned category boundaries (two-sided Kruskal–Wallis test, H₍₂₎= 9.1, P = 0.011; post hoc one-sided WMPSR test, −26° versus 26°: W = 15, P = 0.031; −26° versus 0°: W = 15, P = 0.031; n = 5 mice). *P < 0.05.