Modeling reveals substantial dendritic crosstalk in SACs despite significant voltage attenuation.

A) Reconstructed morphology of a mouse SAC, with dendrites color-coded based on voltage attenuation towards the recording site (indicated in red). The bottom panel displays example voltage responses to a 100 ms-long current pulse injected at one of four positions denoted by circles. The black and red traces represent the potentials at the injection and recording sites, respectively.

B) Analysis of signal attenuation in SAC dendrites relative to the recording site marked in panel A, as a function of distance from the soma (negative values indicate positions on the opposite side of the dendritic tree).

C) Comparison of peak depolarization resulting from synaptic stimulation of the recorded branch (labeled as ’single dendrite’; stimulated area shaded in grey) with the impact of driving synapses on other dendritic sites excluding the target branch (right, light grey). The combined activation of all inputs is shown for reference (dotted)

Evolutionary algorithm-based model enhances directional voltage responses in SACs, reproducing key features of the space-time model.

A-B) Schematic representation of the bipolar-SAC circuit model. B, left, demonstrates the spatial components of the bipolar RF center (grey) and surround (dashed) components. The spatiotemporal RF components were convolved with horizontally moving bar stimuli (B, center) to generate the inputs for the multicompartmental SAC model (B, right). Two distinct bipolar groups, each with a unique RF formulation, innervated the proximal and more distal SAC dendrites. B, right, simulated SAC outputs are color coded by their DSI levels. The degree of postsynaptic direction selectivity was measured within 30 µm from the horizontal axis (these outputs are highlighted with black strokes).

C) The evolutionary algorithm training process involved iterative selection and mutation steps. Each generation included candidate solutions for bipolar RF templates (top row) that were integrated into SAC dendrites (middle row) and ranked based on the directionality and amplitude of calcium signals (bottom row). The best solutions underwent mutation and were propagated to the next generation.

D) Example response dynamics of the proximal (blue) and distal (orange) BCs (top), representative voltage (middle), and calcium (bottom) signals recorded from a SAC dendrite (location as in Figure 1). Dots represent peak response amplitudes in inward (grey) and outward (black) stimulation directions. The model was trained on five velocities (top, units: mm/s).

E) Mean (±SD) directional tuning achieved by the model (solid circles, n = 15). Open circles represent the optimal direction selectivity index (DSI) in a bipolar-SAC model with an identical formulation of proximal and distal BCs. In this scenario, direction selectivity is mediated by voltage filtering in SAC dendrites.

Impact of BC RF components on DS performance.

A) Representative responses of bipolar dynamics (top); voltage (middle), and calcium (bottom) in SAC dendrites (stimulus speed = 0.5mm/s) in a model where the proximal and distal bipolar RF formulations differed in a single parameter: response lag (’delay’), rise/decay kinetics, or the spatial extent of RF components. The original, unconstrained configuration and a model with identical RFs are included for comparison.

B-C) Mean (±SD) values of direction selectivity index (B) and the distribution of RF parameters in the proximal/distal presynaptic groups (C) for each scenario (n = 10). Dotted lines in B indicate the mean values of the full and identical RFs modes. Rise/decay kinetics are presented on a logarithmic scale. The spatial extent of the center and surround RF components is expressed as the full width at half maximum (FWHM). ∗∗∗∗ p<10-6 vs. the full model. #### p<10-6 vs. the identical RF model (ANOVA followed by Tukey’s test).

Direction selectivity in the space- time wiring model is independent of dendritic isolation.

A) Representative voltage profiles and dendritic calcium signals in the original model (left), SAC with reduced inter-dendritic interactions due to elevated internal resistance in the perisomatic area (middle), and a “hyperconnected” SAC model with low signal attenuation in the dendritic tree (right).

B) Similar to panel A, but for models evolved to enhance DS signals in a single stimulated branch. The cell morphology is color-coded based on voltage attenuation from a distal release site of the stimulated dendrite. Insets show voltage and calcium signals recorded on the opposite side of the cell.

C) Summary of directional tuning observed with different levels of compartmentalization, suggesting a minimal impact of isolation on SAC performance. ∗ p = 0.01 (ANOVA followed by Tukey’s test).

Recording of glutamatergic drive to SACs during full-field motion

A) Two-photon image of one example field of view (FOV) displaying the average iGluSnFR fluorescence (left) and the processed dF/F signals (right). Floxed iGluSnFR expression was induced using AAVs in ChAT-Cre mice.

B) Responses from the regions of interest (ROIs) indicated in panel A to horizontally moving bars (speed = 0.25 mm/s).

C) RF mapping using the filtered back projection technique. Top: Changes in fluorescence from three example ROIs in response to bars flashed at 32 different spatial positions and five orientations. Bottom: Reconstructed spatial RFs. The yellow square represents the estimated extent of the two-photon FOV shown in panel A. The black curves represent the x and y RF profiles measured at the center of mass (indicated by white dotted lines).

Diversity of glutamatergic responses to motion in ON- and OFF-SAC populations.

A, C) Left: the average glutamate signals in functional clusters determined from ROI responses to motion (speed = 0.5 mm/s, color-coded by cluster identity). Shaded areas mark the standard deviation. The dotted line indicates the time of peak response of cluster C1. Right, heatmaps of the responses from individual ROIs.

B, D) Mean (±SD) waveform characteristics measured from individual ROIs in each cluster. TI, transiency index. Clusters are sorted based on their transiency.

The onset of motion responses depends on the extent of the receptive field and not on response lag measured with static stimuli

A-B) Correlation between the onset of motion responses and the static response lag (A), and RF width (B) color coded by functional cluster identity. Clusters with the longest lags have wide RFs and earliest responses to motion.

C) Illustration of the interaction between two mechanisms contributing to the time of response onset. Cells with prolonged visual processing delay will respond later to the presentation of a static stimulus (black). When moving bars are presented, response timing depends on the processing time lag, the size of the receptive field and stimulus velocity.

Estimated RF properties from presynaptic responses to motion.

A) Center-surround RF model was trained using an evolutionary algorithm to match experimentally recorded waveforms in ON- SACs. Experimental data is color-coded as in Figure 6. The output of the seven models is shown in black.

B) Comparison between experimentally recorded responses to 4- second-long full- field flashes and the predictions of the models (black).

C) Mean (±SD) RF properties measured from each of the models

(n = 10 repeats for each cluster) are shown in black. The corresponding parameters determined experimentally are shown in grey. Delay and rise time were measured from flash responses, and the FWHM of the center was analyzed from RF maps. The right panel illustrates the predicted spatial extent of the center (grey) and surround (dotted black) receptive field components in each functional cluster.

Directional tuning in bipolar–SAC models with experimentally recorded excitatory waveforms segregated into proximal-distal regions.

A) Schematic representation of the evolutionary algorithm employed to maximize DS by utilizing deconvolved waveforms from experimentally recorded clusters as the input.

B) Top: Direction selectivity achieved by the models when various combinations of input waveforms are targeted towards the proximal and distal SAC dendrites. Squares and dots represent cases where the waveforms are identical for all bipolar cells. Bottom: Representative calcium signals obtained from the best (left) and worst (right) combinations of input waveforms.

Optimal DS with experimentally recorded excitatory waveforms.

A) Investigation of directional performance in a multicompartmental bipolar- SAC model innervated by functional clusters experimental.

B) Illustration of an example solution superimposed on SAC morphology. The distribution of BC inputs was symmetrical along the soma-dendritic axis based on their distance from the soma (Left: small circles and grey annuli). Deconvolved waveforms from one of the experimentally recorded clusters were applied to all synapses within each bin. Output synapses are indicated with large circles and color coded by the DSI. The degree of postsynaptic direction selectivity was measured within 30 µm from the horizontal axis (these outputs are highlighted with black strokes).

C) The spatial distribution of functional clusters producing optimal directional signals across 15 model runs. Input color coding in B and C as in Figure 6.

D) Top: overlay of the velocity tuning dynamics for ON-C2 and ON-C6, representing the most commonly observed proximal and distal clusters. Bottom: calcium signals in a SAC dendrite generated by one of the evolved models. Dots indicate peak response amplitudes in inward (grey) and outward (black) stimulation directions.

E) Directional tuning as a function of stimulus velocity, with the solid curve representing the mean (±SD) results obtained from the model. Dotted, velocity tuning calculated from simulations with SAC innervation by a single functional cluster (see Figure 9). ‘Space-time wiring’ improves directional selectivity over a wide range of stimulation velocities.

F-G) as in D-E, but for OFF-SACs.

Dynamics of glutamate release onto SACs are sluggish and more sustained compared to the optimal excitatory drive produced in synthetic model.

Waveform parameters, measured for synthetic RFs driving optimal direction selectivity (squares) and for experimentally determined functional clusters (filled circles - ON, open circles – OFF). Color coding as in Figure 6. Clusters identified as best contributors for the Hessenstein-Reichardt correlator (proximal: ON- C2 and OFF-C1; distal: ON-C6 and OFF-C5) are marked with arrows.