Fig. 2: EM reconstruction of T4 dendrites across the eye reveals a stereotypical pattern of arborization.

a, Schematic of the Drosophila visual system, highlighting Mi1 and T4 cells in a column. b, EM reconstruction (FAFB dataset⁶) of four Mi1 cells arborizing in medulla layers M1, M5 and M9–M10. Scale bar, 10 μm. c, Medulla columns identified by the centres of mass of Mi1 cells in M10. Magenta dots, dorsal rim area (DRA) columns⁵⁶; middle band, equatorial region with seven (yellow) or eight (tan) photoreceptors in corresponding lamina cartridges (Extended Data Fig. 2b); black dots, central meridian dividing columns into approximately equal halves; empty circles, medulla columns lacking R7 and R8 inputs, presumably with no corresponding ommatidia⁵⁷. Scale bar, 10 μm. All boxed keys are defined in Extended Data Fig. 1d. d, Medulla columns mapped onto a 2D regular grid (‘Med | Reg’) with orthogonal +h and +v axes defined by equatorial region and central meridian. The +p and +q axes are shown for consistency with previous work^31,58. e, Dendritic arbors of 176 T4b cells in M10. The two highlighted examples are shown in g. Scale bar, 10 μm. f, An example T4b dendrite (T4b 139). Bolded branch (top) colour-coded by Strahler number (bottom). Each branch is represented as a vector, and the dendrite’s anatomical PD is defined as the vector sum of all Strahler number 2 and 3 branches. g, Example T4b and T4d dendrites, with PD and width indicated. Branches coloured by Strahler number (>3 in black). The seven circles represent the home column and six nearest neighbours. Scale bar, 1 μm. h, PDs mapped to regular grid using 19 neighbouring columns (Methods). Tail, centre and head of each PD vector indicated as in g. i, Distribution of angles between T4 PDs and the +v axis. j, PD amplitude distributions in regular grid units for T4b and T4d neurons (two-sided Wilcoxon rank test, P = 2.2 × 10⁻¹⁶). k, PD amplitudes normalized by respective hexagon length units (defined in inset; two-sided Wilcoxon rank test, P = 0.015). The scale bars for i–k span from zero to the height of the uniform distribution.