Retinal boutons and neurons in the SC were well tuned to motion direction and orientation.

A. AAV-SyGCaMP6f was delivered to RGCs via intravitreal injection (left). Example average image of retinal boutons in the SC (right). B. Visual stimuli were presented on three monitors covering the visual field from −135° to 135° azimuth and −42° to 42° elevation. Grating stimuli covered all screens. Stimulus directions—specifically “forward”/temporo-nasal and “backward”/naso-temporal—were described from the perspective of the left eye, because RFs of all recorded units were in the left hemifield. C. AAV-GCaMP6f is delivered to SC neurons via injection (right). Example average image of neurons in the SC (left). D. Single-trial fluorescence traces (gray) and mean fitted traces (black) of example boutons (marked in A) in response to sinusoidal gratings drifting in 12 directions. Gray shading indicates the 2 s stimulus presentations. E–F. Direction (E) and orientation (F) tuning curves of example boutons (same as in D), showing single-trial response amplitudes (gray dots), mean amplitudes (black dots), and preferred direction/orientation (red triangles). F – forward, D – down, B – backward, U – up, V – vertical, H – horizontal. G–I. As in D–F, for example SC neurons (marked in C). J–M. Direction (J) and orientation (K) tuning curves of all DS or OS (top) and non-selective (bottom) retinal boutons recorded across all sessions, sorted by their preferred direction of motion or orientation. Each tuning curve was scaled to a range between 0 and 1. Triangles indicate tuning curves of example boutons from A. Corresponding data for SC neurons in L and M. N–Q. Mean distribution (± SEM) of preferred directions (N) and orientations (O) across all sessions of retinal boutons (red). Datasets with fewer than 20 selective units were excluded. Corresponding data for SC neurons (teal) in P and Q. For comparison, mean distributions of retinal boutons are overlaid (red; same data as in N, O). R. Preferred direction versus orientation for retinal boutons selective to both features. Example boutons from A marked in color. S. Direction versus orientation selectivity for all tuned retinal boutons. Example boutons from A marked in color. Insets show histograms of selectivity types. T, U. As in R and S, for SC neurons.

RF locations deviated from retinotopy by about one quarter of their diameter.

A. Example average image of retinal boutons in the SC. B. ON (top) and OFF (bottom) receptive subfields of 3 example retinal boutons (marked in A). Outlines (center ± 1 SD of fitted 2D Gaussian) are shown in black. Examples 1 and 2 had significant ON and OFF subfields, while example 3 had only an OFF subfield (dashed outline of the ON subfield is shown for reference). C. RF outlines for all retinal boutons recorded in session from A. Example boutons marked in color. D–F. Same as in A–C, for an example recording of SC neurons (same FOV as in Figure 1C). G. Relationship between position in the FOV and RF azimuth (left) or elevation (right) for retinal boutons from the session in A. The colored background gradients represent the best linear fit between FOV and RF positions. H. Distances between mapped and retinotopic (predicted from FOV position) RF locations for retinal boutons (red) and SC neurons (teal). Retinal boutons had slightly smaller distances than SC neurons (medians (triangles): 2.6° and 2.9° for boutons and SC neurons; p < 1e-6, Wilcoxon rank-sum test). I. Same as in G, for SC neurons and example session from D. J. RF diameters (2 SDs of the fitted Gaussian) across retinal boutons (red) and SC neurons (teal). RF sizes were not significantly different between boutons and SC neurons (medians (triangles): 11.3° and 11.1° for boutons and SC neurons; p = 0.11, Wilcoxon rank-sum test).

Topography of directions and orientations was consistent with topography originally identified in the retina.

A–C. Cartoons illustrating the topography of direction preferences (A) of DSGCs and orientation preferences (B, C) of OSGCs for the left eye. Direction preferences align with two axes defining longitudes (A), while orientation preferences align with two axes defining longitudes similar to those for direction (B) and two further axes defining latitudes (C). Dotted rectangle within left visual field marks the part of the visual field that contains RFs of our data (shown in D–G). D–G. Direction (D) and orientation (E) preferences of retinal boutons across visual space (RF positions based on mapped RFs, not retinotopy). Arrows in D point into directions defined by the DSGCs longitudes (A). Colored lines in E follow orientations of longitudes (B) and latitudes (C) for OSGCs. Numbers indicate unit count in each polar histogram. Corresponding data for SC neurons in F and G. RF positions for SC neurons were based on retinotopic fits (see Figure 2). H–K. Classification of all DS (H) and OS (I) boutons according to the closest predicted vector from the geometric model. Each dot shows one recording session (dot size reflects total number of boutons); black line indicates median proportion across sessions. Naming follows previous conventions, where DSGC classes were named after the behavior that elicits their responses, e.g., “advance cells” for DSGCs preferring movement in naso-temporal direction. Arrows and lines at the top are example vectors from a central location in D and E. Corresponding data for SC neurons in J and K. L–O. Angular differences between preferences and closest predicted vectors for direction (L) and orientation (M) across all tuned retinal boutons. The median difference (triangle) of the original data is compared to the median (2.5–97.5 percentile intervals) expected when RF locations were randomly permuted (dark gray) or when preferences were randomly drawn from a uniform distribution (light gray). Direction: 11.6° original, 13.9°–15.4° permuted, 21.0°–24.0° uniform; orientation: 7.5° original, 8.8°–10.1° permuted, 14.1°–16.9° uniform. Corresponding data for SC neurons in N and O. Direction: 12.2° original, 13.3°–14.5° permuted, 21.0°–24.0° uniform; orientation: 10.1° original, 9.9°–11.2° permuted, 13.2°–15.7° uniform.

Local populations showed low consistency of direction and orientation preferences.

A–D. Direction (A) and orientation (B) preferences for all selective retinal boutons, plotted at their mapped RF locations in visual space. Corresponding data for SC neurons in C and D. For neurons, RF locations are based on the retinotopic fits (Figure 2). E–H. Mean direction (E) and orientation (F) preferences obtained by averaging across boutons within circular patches of 10° diameter in visual space (illustrated by circle in the lower-right corner). Transparency indicates patches with fewer than 20 boutons. Corresponding data for SC neurons in G and H. I–L. Consistency of direction (I) and orientation (J) preferences within 10° patches, quantified as the inverse circular variance of preferences (same pooling as in E and F) for retinal boutons. Corresponding data for SC neurons in K and L. M–P. Distributions of consistency for direction (M) and orientation (N) across patches (red; at least five units per patch) compared to null distribution obtained by permuting RF locations across boutons (black line: median; gray band: 2.5th–97.5th percentile interval). Corresponding data for SC neurons in O and P. RF locations for SC neurons are based on retinotopic fit (Figure 2). Q–T. Mean consistency in direction (Q) and orientation (R) preferences across all patches per recording for retinal boutons (black triangles), compared to corresponding null distribution for each recording (gray; dot: median; vertical line: 2.5th–97.5th percentile interval). Recordings with significant consistency values are marked by a star. Corresponding data for SC neurons in S and T.

Pairwise similarity of direction and orientation tuning was largely independent of lateral distance.

A–D. Examples of direction (A) and orientation (B) preferences of individual retinal boutons in one FOV. Each bouton is plotted at its imaging location; non-responsive boutons are shown in dark gray and non-selective boutons in light gray. Corresponding data for SC neurons in C and D. Neurons from four imaging planes are shown. E–H. Pairwise difference in direction (E) and orientation (F) preferences as a function of lateral distance in the SC (depth ignored) for all pairs of simultaneously recorded retinal boutons (dots colored by density). Mean difference as a function of distanc e (red) is compared to the null distribution obtained by permuting bouton locations (black line: median; gray band: 2.5th–97.5th percentile interval). For comparison, mean pairwise difference of SC neurons is overlaid (teal; same data as in G, H). Corresponding data for SC neurons in G and H. Insets show zoom-in. I–L. Mean pairwise difference in direction (I) and orientation (J) preference for each recording (gray lines, at least 15 selective units each) as a function of distance, normalized to the null distribution of that recording. Mean (horizontal black line) and ±3 SEMs (gray shade) of the null distribution are shown for comparison. Example recordings from A and B are shown as dotted lines; mean across all recordings depicted as bold black line. Corresponding data for SC neurons in K and L. M–P. Mean pairwise difference in direction (M) and orientation (N) preference for each recording (at least 15 selective units each) as a function of FOV size (length of diagonal). Example from A and B shown in red. Corresponding data for SC neurons in O and P.

Electrophysiological recordings across SC depth confirmed the lack of functional clustering.

A. Visual stimuli were presented on a single monitor covering −95° to 0° azimuth and −27° to 51° elevation. Drifting grating stimuli covered the full screen. B. Neuropixels probes were used to record single units from sSC (light gray) and dSC (dark gray). C. Single-trial (gray) and mean (black) spike rate traces of example units in response to sinusoidal gratings drifting in 12 directions. Grating presentations (2 s each) are indicated by gray shading. D, E. Direction (D) and orientation (E) tuning curves for example units in C, showing single-trial response amplitudes (gray dots), mean amplitudes (black dots), and preferred directions and orientations if significantly tuned (red triangles). F, G. Mean distribution (± SEM) of preferred directions (F) and orientations (G) across all recording sessions. Datasets with fewer than 10 selective units were excluded. H. Preferred direction of all recorded units, as a function of depth in SC for each recording session (multiple sessions per animal). I. Direction selectivity of all units as a function of depth in SC. J, K. Same as in H and I, for orientation tuning. For comparison, direction selectivity profile (green) from I is overlaid in K. L. Mean of pairwise direction difference of each recording (dots), compared to corresponding null distribution (gray line: median; light-gray box: 2.5th–97.5th percentile interval). M. Pairwise difference in direction preference as a function of depth separation in SC for all pairs of simultaneously recorded units (dots colored by density). Mean difference as a function of separation (teal) is compared to null distribution obtained by permuting unit depths (black line: median; gray band: 2.5th–97.5th percentile interval). N, O. Same as in L and M, for orientation tuning.

Retinal topography was progressively weakened from superficial to deep SC.

A. ON and OFF RFs of four example units (same as in Figure 6C). RF outlines (center ± 1 SD of fitted 2D Gaussian) shown in black. B. RF outlines of all units recorded in one session, including examples in A (colored). C. Depth distribution of units with RFs from B. Depths were aligned to the border between sSC and dSC. D, E. Direction (D) and orientation (E) preferences of SC units across visual space, plotted at their mapped RF locations (not retinotopic). Arrows in D point along directions defined by DSGC longitudes (see Figure 3A). Colored lines in E indicate orientations of longitudinal and latitudinal fields for OSGCs (see Figure 3B,C). Numbers indicate population size in each polar histogram. The depicted region of visual space in E includes two poles of the spherical axes: one for horizontal longitudes (cyan circle) and one for vertical latitudes (red circle). F–I. Classification of all DS units in sSC (F) and dSC (G) according to closest predicted direction vector. Each dot shows one recording session (dot size reflects total number of units); black line indicates median proportion across sessions. Arrows at the top illustrate example predicted vectors from a central location in D. Corresponding data for OS units in H and I. J–M. Angular differences between measured direction preferences and the closest predicted vectors in sSC (J) and dSC (K). The median difference (triangle) is compared to the distribution obtained by permuting RF locations (dark gray: median; band: 2.5th–97.5th percentile interval) and to a uniform preference model (light gray: median; band: 2.5th–97.5th percentile interval). sSC: 18.0° original, 15.2°–20.0° permuted, 18.3°–27.0° uniform; dSC: 23.2° original, 19.2°–25.6° permuted, 19.2°–27.9° uniform. Corresponding data for OS units in L and M. sSC: 12.4° original, 9.9°–13.9° permuted, 10.7°–17.7° uniform; dSC: 14.8° original, 12.8°–17.4° permuted, 12.0°– 17.9° uniform.

Direction and orientation tuning of SC neurons were consistent across stimulus types.

A–C. Single-trial (gray) and mean fitted (black) calcium traces of four example neurons in response to drifting gratings (A; 12 directions, 2 s each), moving bars (B; 8 directions, 4 s each), and static gratings (C; 4 orientations × 3 spatial phases, 1 s each). Stimulus epochs are indicated by gray shading. We recorded 2,283 SC neurons in response to moving bars and 2,314 neurons in response to static gratings, both within 6 sessions across 5 mice. As for drifting gratings, the kernel fits matched the data very well (mean ± SD explained variance: 0.33 ± 0.21 for bars, 0.30 ± 0.20 for static gratings). Across SC neurons, responsiveness dropped from 50% for drifting gratings to 31% (697/2,283) for moving bars and 41% (945/2,314) for static gratings. D–H. Direction (D, E) and orientation (F–H) tuning curves for the example neurons in A–C, showing single-trial response amplitudes (gray dots), mean amplitudes (black dots), and preferred direction or orientation (red triangles) when tuning was significant. Tuning is shown for drifting gratings (D, F), moving bars (E, G), and static gratings (H). 70% (488/697) of neurons responsive to moving bars were selective for direction (compared to 66% for drifting gratings), while 63% (437/697) were selective for orientation (compared to 61% for drifting gratings). In contrast, only 32% (307/945) of neurons responding to static gratings were selective for orientation. I–L. Direct comparisons of preferred directions (I) and preferred orientations (J–L) obtained from different stimulus classes. Colored dots highlight the example neurons from A–C. SC neurons that were significantly tuned in response to several visual stimuli exhibited highly coherent preferences for direction and orientation with an average difference of 5.0° in preferred direction for drifting gratings and moving bars, and average differences of up to 7.0° in preferred orientation (drifting gratings vs bars: 5.3°; drifting vs static gratings: 7.0°; bars vs static gratings: 1.5°). M, N. Mean distribution (± SEM) of direction (M) and orientation (N) selectivity measured with drifting gratins, across all sessions with at least 20 selective units. Triangles mark the median (across sessions) of the median selectivity index.

Proximity to monitor edges did not systematically affect measured tuning preferences.

A–D. Direction (A) and orientation (B) preferences versus distance from the RF center to the nearest vertical (left/peripheral) monitor edge, for all selectively tuned retinal boutons with mapped RFs. Corresponding data for SC neurons in C and D. E–H. Circular SD (red) of preferred directions (E) and orientations (F) for retinal boutons as a function of RF distance to nearest vertical monitor edge, compared to a null distribution obtained by permuting edge distances (black line: median; gray band: 2.5th–97.5th percentile interval). Corresponding data for SC neurons in G and H. If the monitor edge had biased tuning preferences towards a specific direction or orientation, we would expect that preferences of units closer to the monitor edge are more similar to each other than of units farther away from the edge, which was not the case in our data. I–P. As in A–H, but using distance from the RF to the nearest horizontal (top) monitor edge.

Topography for DS-only and OS-only units is similar to overall population.

A–D. Direction (A) and orientation (B) preferences of DS-only and OS-only retinal boutons across visual space (RFs mapped, not retinotopic). Arrows in A indicate the predicted DSGC longitude directions; colored lines in B indicate predicted OSGC longitudes and latitudes (see Error! Reference source not found.A–C). Numbers indicate population size in each polar histogram. Corresponding data for SC neurons in C and D. E–H. Classification of all DS-only (E) and OS-only (F) retinal boutons according to the closest predicted direction or orientation vector. Each dot shows one recording session (dot size reflects total number of boutons); black line indicates median proportion across sessions. Arrows and lines at top show example predicted vectors from a central location in A and B. Corresponding data for SC neurons in G and H. I–L. Angular differences between measured preferences and the closest predicted directions (I) or orientations (J) for DS-only and OS-only retinal boutons. Median differences (triangles) are compared to null distributions obtained by permuting RF locations (dark gray; median and 2.5–97.5 percentile intervals) or by drawing preferences from a uniform distribution (light gray; median and 2.5–97.5 percentile intervals). Direction: 9.6° original, 10.9°–12.5° permuted, 20.3°–24.4° uniform; orientation: 7.7° original, 7.9°–10.8° permuted, 12.2°–17.1° uniform. Corresponding data for SC neurons in K and L. Direction: 11.1° original, 11.7°–14.0° permuted, 19.7°– 25.2° uniform; orientation: 9.1° original, 7.4°–10.2° permuted, 11.5°–17.2° uniform. Direction preferences of retinal boutons matched model predictions significantly better than those of SC neurons (p = 0.0108, Wilcoxon rank sum test). Orientation preferences matched model predictions to equal degrees in retinal boutons and SC neurons (p = 0.49, Wilcoxon rank sum test).

Local consistency is not trivially explained by the number of sampled units.

A–D. Mean consistency of direction (A) and orientation (B) preferences within each 10° patch, plotted versus the number of retinal boutons in that patch (red dots). The background grayscale indicates the density of units from a null distribution obtained by permuting RF locations across boutons. Corresponding data for SC neurons in C and D.

Weak clustering persisted within strongly selective units, with RF-based distances, and under different stimulus conditions.

A–D. Pairwise differences in direction preference for DS-only units (A) and orientation preference for OS-only units (B) versus lateral distance in SC (depth ignored) for retinal boutons. For comparison, mean pairwise differences for all tuned boutons (pink) are overlaid (same data as in Figure 5E,F). Corresponding data for SC neurons in C and D. E–H. Pairwise difference in direction (E) and orientation (F) preferences versus distance between mapped RFs for all pairs of simultaneously recorded retinal boutons. For comparison, mean pairwise differences for SC neurons (teal) are overlaid (same data as in G, H). Corresponding data for SC neurons in G and H. Only retinal boutons with RFs closer than 2.2 visual degrees had direction and orientation preferences that were more similar than expected from the null distribution. Tuning differences for SC neurons were much closer to the null distribution and were only significantly smaller for neurons with RF distances of up to 5 and 11 visual degrees for direction and orientation preferences. I. Pairwise difference in direction preferences versus lateral distance in SC (depth ignored) for all pairs of simultaneously recorded SC neurons, with preferences measured from responses to moving bars. Mean pairwise tuning differences (85°) were close to those expected from uniform distributions. J, K. Same as in I, for orientation preferences measured from moving bars (J) and static gratings (K). Mean pairwise tuning differences were close to those expected from uniform distributions (37° for moving bars, 41° for static gratings). In all panels, dots are colored by density, and mean difference as a function of distance is compared to null distribution obtained by permuting neural or RF locations (black line: median; gray band: 2.5th–97.5th percentile interval).

Determining position of SC along Neuropixels probe and tuning properties in electrophysiology data.

A, B. Evoked local field potential (LFP, A) and current-source-density (CSD, B) profiles from an example recording in response to the most effective square of the visual noise stimulus. The surface of SC is defined at 25% of the peak amplitude in the evoke d LFP. The bottom of sSC (border between SGS and SO) is defined as the zero-crossing below the largest positive CSD peak. C. Spike amplitudes at spike times of all single units in the same recording as in A and B, plotted against their depths. D. Preferred direction versus orientation for SC units selective for both features. E. Direction versus orientation selectivity for all tuned SC units.