Surround motion modulates the encoding properties of primate retinal ganglion cells

SUMMARY

Antagonistic interactions between center and surround regions of the receptive field are widely observed across sensory systems. In the early visual system, these interactions contribute to important computations such as edge detection. However, less is known about how center-surround interactions depend on the spatiotemporal properties of the visual input. Here, we show that surround motion strongly modulates the response properties of two understudied primate ganglion cell types. Broad thorny cell responses are strongest when motion in the center and surround is uncorrelated, similar to object-motion-sensitive cells found in other species. A different pattern is observed in On smooth monostratified cells: surround activation is suppressive for static stimuli and facilitatory for motion. These effects of surround activation diverge significantly from classical center-surround models and more closely resemble how surround motion affects responses in primate visual cortex.

Graphical abstract

In brief

Appleby et al. investigate how motion in the receptive-field surround influences the responses of two primate retinal ganglion cell types. They find that broad thorny cells exhibit object-motion sensitivity, while On smooth monostratified cells display motion-dependent facilitation—diverging from classical models and echoing surround modulation seen in visual cortex.

INTRODUCTION

The receptive field provides a concise summary of the stimulus selectivity of a sensory neuron, and much of what we know about early sensory systems has been learned by measuring receptive-field properties.^1,2 In the early visual system, spatial receptive fields are classically composed of center and surround regions with distinct stimulus selectivity—the same stimulus falling in the center versus surround typically elicits responses with opposite polarity.^2–4 These differences in stimulus preference between center and surround subserve important general computations such as edge detection, wavelength selectivity, and efficient encoding of natural scenes.^5–10

The canonical center-surround receptive field is described as a linear combination of responses to two opposing Gaussian spatial weighting functions—a relatively compact one describing the center, and a broader one describing the surround. In this framework, an important function of the surround is to decorrelate spatial inputs. However, receptive-field surrounds of different retinal ganglion cell types can be generated by distinct circuitry, and, correspondingly, surrounds can differ in their functional properties.¹¹ Cell-type-specific suppression of non-primate ganglion cells to saccade-like image shifts provides an example of specialized surround function.^12–14 Object-motion-sensitive cells provide another example. These cells respond strongest when motion in the center and surround is uncorrelated, thus enhancing the cell’s response to local motion.^15–19 It is unclear whether primate ganglion cell surrounds share this diversity of motion-related functions, in large part because the majority of primate cell types have not been characterized functionally.

To better understand the contributions of the receptive-field surround to early visual processing in primates, we measured how surround activation by moving and non-moving stimuli affected the response properties of several understudied cell types in the macaque monkey retina—specifically broad thorny and smooth monostratified ganglion cells. Both cells exhibit classic center-surround antagonism for stationary stimuli. However, motion in the receptive-field surround produced distinct effects on the response properties of these cells. Broad thorny cells showed sensitivity to differential motion between the center and surround regions of the receptive field, similar to the object-motion selectivity observed in other vertebrate retinas.^15–19 Unlike the broad thorny cells, motion in the receptive-field surround increased (facilitated) responses of On smooth monostratified cells to motion and non-motion stimuli confined to the receptive-field center. Thus, the effect of the receptive-field surround on responses of primate ganglion cells depends on the retinal pathway and the statistics of incoming stimuli.

RESULTS

Our goal was to determine the impact of center-surround interactions on motion processing in several understudied primate ganglion cell types—particularly On smooth monostratified and broad thorny cells. First, we show that center-surround interactions cause broad thorny cells, but not On smooth monostratified cells, to respond more strongly to local motion than to global motion. Second, we show that luminance and motion in the receptive-field surround have distinct effects on the responsiveness of On smooth monostratified cells—changes in luminance produce classical receptive-field antagonism while motion produces non-classical response facilitation. Third, we show that the enhancement of On smooth monostratified cell responses produced by surround motion persists after motion ends.

Cell identification and selective activation of center and surround

On and Off parasol, On and Off smooth monostratified, and broad thorny ganglion cells were identified based on their characteristic responses to light steps, receptive-field size, and dendritic morphology (Figures 1A and 1B).^20,21 We focused on how center-surround interactions influence responses to motion in On smooth monostratified and broad thorny cells, and we were particularly interested in cases in which such center-surround interactions differed from those observed in better-studied cells such as On and Off parasol cells.

Figure 1. Ganglion cells show classical surround suppression to stationary spots.

(A) Spike responses to a 250-ms light step from On parasol, Off parasol, On smooth monostratified, and broad thorny ganglion cells measured with cell-attached recording. The duration of the light increment is highlighted in yellow. Error bars indicate mean ± SEM.

(B) Spots of multiple sizes (radius, 40–720 μm) delivered for 500 ms to the receptive-field center of each cell type. Spots were either contrast increments or decrements from a background, with contrast polarity matching the response polarity of each cell type. Broad thorny ganglion cell responses were measured for both contrast increments and decrements.

(C) Fluorescence images of each cell’s morphology generated on a confocal microscope. Scale bar on each image represents 100 μm.

Many of our experiments required independent stimulation of the receptive-field center and surround. Ganglion cells in the macaque retina exhibit considerable variability in receptive-field size—both across cell types and within a single type due to changes in eccentricity.²² Therefore, for each cell, we estimated the size at which responses became dominated by the receptive-field surround. This was achieved by presenting annuli with progressively increasing inner diameters to identify the point at which responses shifted from center to surround dominated—evident either as a cessation of spiking or a reversal in response polarity (Figure S1; see STAR Methods).^3,23–28 Center-targeted stimuli were restricted to circular spots smaller than this critical diameter, while surround-targeted stimuli were confined to annuli with an inner diameter larger than this value. A spatial gap between the center and surround stimuli further ensured minimal overlap and effective separation.

The On smooth monostratified and broad thorny cells show distinct sensitivities to center-vs-surround motion

When objects move independently of an observer within a visual scene, they generate patterns of input on the retina that differ from those produced by global motion, such as that caused by an observer’s eye movements. To probe sensitivity to object motion, we used two visual stimuli previously employed to identify this computation in non-primate retinal ganglion cells. The first stimulus mimicked key features of motion experienced during fixational eye movements.^15,16 This stimulus consisted of randomly moving gratings presented in three conditions: (1) motion restricted to the receptive-field center (center-only condition), (2) identical motion trajectories by gratings in both the center and surrounding receptive-field region (correlated condition), and (3) independent motion trajectories by the center and surround gratings (uncorrelated condition; Figure 2A, left; Video S1; see STAR Methods). The correlated and uncorrelated conditions simulate global and local motion and the center-only condition provides a reference.^15,16 Across ganglion cell types, the center-only condition reliably evoked strong responses, with peak firing rates exceeding 100 spikes s⁻¹. Figure 2B shows the average spiking responses across repeated stimulus presentations for each condition, highlighting differences in motion sensitivity.

Figure 2. Broad thorny and On smooth monostratified cells show distinct motion sensitivities.

(A) Jittering grating (left) and moving texture stimuli (right) used to characterize object versus background motion sensitivity (grating bar width, 50–75 μm; texture standard deviation, 15–30 μm). Single frames of each stimulus condition are depicted at the top of each column, along with a diagram indicating the relative trajectory of motion in the center and surround receptive-field regions (top). For example, in the uncorrelated condition, the center and surround gratings follow distinct trajectories, and this is indicated by blue and red arrows in the center and surround.

(B) Spike rate for the stimulus in (A) for four cell types: On parasol (top row), Off parasol (second row), On smooth monostratified (third row), and the broad thorny (bottom row). Each response corresponds to the stimulus condition indicated in the aligned top panel. Spike-rate responses to the moving texture stimulus are shown starting at motion onset, two seconds after the stationary texture was presented.

(C) Motion preferences were quantified by comparing the average spike rate during the period of motion with the surround (R) with those for the center-only stimulus (R_center). The normalized difference between these responses (y axis) measures the spiking preference for each full-field condition relative to its center-only condition. Error bars indicate mean ± SEM.

All four ganglion cell types examined in this study exhibited surround suppression in response to stationary spots (Figure 1). Based on the classical receptive-field model, one would expect that motion in the receptive-field surround would suppress responses relative to a condition in which motion is limited to the center (i.e., the center-only condition). Furthermore, because the correlated and uncorrelated motion conditions use gratings with identical properties (e.g., bar width, motion amplitude) except for the correlation between center and surround, the degree of suppression should be similar across those two conditions. We therefore used the center-only condition as the baseline for comparison. Specifically, we quantified the effect of surround motion by calculating the normalized difference between the responses in each surround motion condition and the center-only condition.

Object-motion-sensitive cells exhibit distinctive responses to correlated vs. uncorrelated center-surround motion stimuli.^15,16 These cells are suppressed when center and surround motion are correlated, and they respond most strongly when the surround moves independently of the center. In our recordings, broad thorny ganglion cells showed this signature response—they fired most vigorously when center and surround motion trajectories differed (i.e., local motion), consistent with object-motion-sensitive cells described in other species (Figures 2B and 2C; Figure S2; p value, correlated 3.9 × 10⁻³; n = 9 cells; Wilcoxon signed-rank test here and below). This differed from On smooth monostratified cells, which responded robustly to motion in the surround regardless of whether it was correlated or uncorrelated with the center. On smooth monostratified responses were stronger for motion in both center and surround compared to center-only motion (p value, correlated 6.1 × 10⁻⁵, uncorrelated 1.1 × 10⁻³, n = 15 cells). Off parasol cells were modestly but significantly suppressed by surround motion and this suppression was greater for correlated motion (p value, correlated 1.0 × 10⁻², uncorrelated 2.7 × 10⁻², correlated/uncorrelated 8.1 × 10⁻³; n = 13 cells). On parasol cells showed a significant increase in spiking to the correlated surround motion, but the effect was not significant for uncorrelated motion (p value, correlated 9.8 × 10⁻⁴, uncorrelated 8.3 × 10⁻²; n = 11 cells).

The second stimulus consisted of textures moving in a single direction and was adapted from a paradigm originally used to identify object-motion-sensitive ganglion cell types in the mouse retina.¹⁹ The stimulus consisted of three conditions: (1) a condition in which motion occurred only in the receptive-field center, (2) a condition in which motions in the center and surround were synchronous, and (3) a condition in which the direction of texture motion in the center was orthogonal to the direction of the surround texture (Figure 2A, right; Video S2). As with the jittering grating stimulus, the moving texture stimulus evoked instantaneous firing rates of at least 100 spikes s⁻¹ in each cell type. On average, broad thorny cell responses were larger for differential motion relative to center-only and synchronous motion, but this difference was not statistically significant (Figures 2B and 2C; p = 0.375; n = 4 cells). The moving textures in the receptive-field surround increased the spike rate of On smooth monostratified cells relative to the center-only motion condition (Figures 2B and 2C; p value, correlated 7.8 × 10⁻³, uncorrelated 1.6 × 10⁻²; n = 8 cells).

In rodents, rabbits, and salamanders, object-motion-sensitive ganglion cells inherit their stimulus selectivity primarily from their excitatory synaptic inputs.^16–19 To investigate the circuit mechanisms that enable broad thorny ganglion cells to detect such correlations, we performed whole-cell voltage-clamp recordings to isolate excitatory and inhibitory synaptic inputs during presentation of the jittering grating stimulus. The excitatory responses were weakest for the correlated stimulus condition that simulated global motion and strongest for the uncorrelated condition that simulated local motion (p = 3.1 × 10⁻²; n = 6 cells; Figures 3A–3C). This excitatory input pattern closely matched the spike output pattern in these cells (Figure 2). In contrast, inhibitory synaptic currents remained relatively constant across stimulus conditions, suggesting that the surround-driven sensitivity to local motion arises from presynaptic modulation of excitatory inputs rather than changes in postsynaptic inhibition (Figures 3D–3F).

Figure 3. Sensitivity to center-surround motion correlations in broad thorny cells is present in their excitatory synaptic inputs.

(A) Excitatory synaptic currents recorded for the three conditions of the jittering grating stimulus in a broad thorny ganglion cell. In this example, the trajectory of the grating over the receptive-field center was identical for the three conditions to aid in comparing the effects of surround stimulation. Uncorrelated motion between center and surround produced the largest inward (negative) currents, while correlated motion elicited the smallest currents. Dashed line indicates the leak current.

(B) Zoomed region for correlated and uncorrelated excitatory currents indicated by the shading in (A). Uncorrelated motion (blue) between center and surround produced larger inward deflections than correlated motion (green).

(C) Excitatory response of broad thorny ganglion cells (p correlated, 0.11; p uncorrelated, 0.031; p comparison, 0.016; n = 5 cells).

(D) Inhibitory synaptic currents for the broad thorny cell in (A) for the same stimulus conditions. Inhibitory currents were similar in magnitude across the three surround motion conditions. Dashed line indicates the leak current.

(E) Zoomed region for inhibitory correlated and uncorrelated motion currents indicated by the shading in (D). Uncorrelated inhibitory currents (blue) were similar in magnitude to correlated currents (green).

(F) Inhibitory response of broad thorny ganglion cells (p correlated, 0.16; p uncorrelated, 0.047; p comparison, 0.11; n = 5 cells).

In summary, these experiments revealed the distinct effects of surround motion on the response properties of four primate ganglion cell types. Surround motion had minimal impact on the responsiveness of On and Off parasol cells but significantly affected broad thorny and On smooth monostratified cells. Broad thorny cells were particularly sensitive to the degree of motion correlation between the center and surround regions of the receptive field—a hallmark of object-motion-sensitive responses previously observed in the salamander, rabbit, and mouse retina.^15–19 In contrast, the spike responses of On smooth monostratified cells were enhanced by surround motion regardless of its correlation with center motion. Thus, the influence of surround stimulation for On smooth monostratified cells depends on the stimulus properties—the surround exerts classical suppression in response to luminance changes but facilitates responses to motion (Figures 1 and 2).

Spatial correlations between moving objects modulates ganglion cell spike outputs

The classic jittering grating stimuli used in Figure 2 distinguish cells that are sensitive to local versus global motion. This distinction has been helpful in identifying the roles these cells may play in vision. However, local and global motion represent two extreme forms of motion correlations, and testing sensitivity only to these extremes may not fully capture how the spatial scale of motion correlations impact responses. Such spatial scales will be important in relating motion preferences to natural vision (see discussion). Hence, we measured how a cell’s responses change as we systematically altered the spatial scale of motion correlations.

We measured tuning to the spatial scale of motion correlations using a stimulus composed of small moving squares, each following a trajectory with two components: an independent motion component unique to each square and a correlated motion component shared across squares. Initially, each square’s trajectory was drawn independently from a normal distribution. Correlations were then introduced based on the distance between squares and a correlation length (λ), which defined how strongly motion was correlated as a function of spatial separation (see STAR Methods).

We presented stimuli with a range of discrete correlation distances (Figures 4A and 4B) and measured the mean spike rate for each cell at each distance, yielding a tuning curve that quantified sensitivity to the spatial extent of motion correlations (Figure 4C). Different ganglion cell types exhibited distinct tuning profiles: On parasol and On smooth monostratified cells showed little sensitivity to correlation length, whereas broad thorny cells and several other less-studied types displayed strong preferences for motion correlations occurring over short spatial scales.

Figure 4. Spatial correlations modulate ganglion cell responses to motion.

(A) Stimulus used to measure the effects of spatial correlations on ganglion cell spike responses. An equal number of black and white squares (width, 40–50 μm; density, 200 squares mm⁻²; correlation time, 200 ms) was presented to the retina. The coherence or correlation in the motion between the squares depended on the correlation length. At lower correlation lengths, motion was correlated locally and motion coherence between squares increased with increasing correlation length, resulting in more global motion.

(B) Average spike rates for an On smooth monostratified (top) and a broad thorny ganglion cell (bottom). Traces correspond to the correlation lengths shown in (A). The On smooth monostratified cell spike response was similar for the three correlation lengths, but the broad thorny cell responded with a higher spike rate for the shortest correlation length (100 μm) than for the longer correlation lengths.

(C) Average responses to the correlated motion stimulus for eight primate retinal ganglion cell types. Bold traces correspond to On smooth and broad thorny cell types. Included cell types are the On smooth monostratified (n = 11), On narrow thorny (n = 1), On parasol (n = 8), Off parasol (n = 11), broad thorny (n = 6), Off smooth monostratified (n = 5), Off narrow thorny (n = 1), and recursive bistratified (n = 2). Points are mean ± SEM.

Figures 2 and 4 show that broad thorny cells share several characteristics with the object-motion-sensitive cells identified in other species, most notably a strong preference for local over global motion. The On smooth monostratified cells showed a facilitatory interaction between motion in the surround and responses elicited in the receptive-field center. This differs from the same cells’ receptive fields for luminance, which show classic center/surround antagonism. Hence, we turned to additional experiments that further characterize how surround motion influences On smooth monostratified responses.

Motion in the receptive-field surround modulates ganglion cell response properties

The experiments presented above revealed nuances in the effects of the receptive-field surround on the response properties of On smooth monostratified ganglion cells. Stimulation of the surround using large spots decreased the spike rate of these cells relative to spots that primarily stimulated the receptive-field center, which is consistent with classical surround antagonism (Figure 1). However, stimulation of the surround with jittering gratings increased the spike outputs of these cells relative to when the grating was not present in the surround (Figure 2). These seemingly paradoxical effects of the surround were also observed to a smaller degree in On parasol cells but were absent in the other cell types tested (Figure 2).

The stimuli that elicited distinct response patterns—flashed spots and jittering gratings—differed in both spatial and temporal structure. Because smooth monostratified and parasol ganglion cells encode motion in their spike output, we designed a stimulus to probe the spatiotemporal features that enhance their responsiveness.^29–31 Specifically, we tested whether the pairwise correlations characteristic of translational motion increased spiking. A flickering spot was presented over the receptive-field center, with contrast values drawn randomly from a Gaussian distribution on each stimulus frame (mean, 0.0; standard deviation, 0.1–0.3). In the surround, an equal number of bright and dark bars were shown, and we systematically varied their spatiotemporal properties (bar contrast, ±0.5; Figure 5; Video S3).

Figure 5. The effects of surround motion on spike output vary with ganglion cell type.

(A) Visual stimulus for measuring the effects of surround stimulation on ganglion cell response properties. A spot was windowed over the receptive-field center and spot contrast was randomly drawn from a Gaussian distribution on each frame. A gray annulus separated the center and surround stimuli. The surround contained an equal number of dark and bright bars. Bar spatiotemporal trajectory varied between three conditions: apparent motion (moving), random, and stationary.

(B) Average spike rates for the three surround conditions in (A) for an On smooth monostratified and a broad thorny ganglion cell. These cells showed distinct response patterns to the surround conditions—On smooth monostratified cell spike rates were highest for the moving surround and broad thorny cell responses were largest for the static surround.

(C) Population data for On smooth monostratified (n = 20), On parasol (n = 14), Off parasol (n = 10), Off smooth monostratified (n = 12), and broad thorny cells (n = 4) tested using the surround motion stimulus. Statistical significance is calculated by Wilcoxon signed-rank test (*p < 0.05, **p < 0.01). Broad thorny cell responses were significantly decreased when the data for surround conditions were combined. Error bars indicate mean ± SEM.

To study the effects of spatiotemporal correlations, the position of the surrounding bars changed over time according to apparent motion (moving condition), varied randomly over time (random condition), or were stationary (stationary condition).^29,32 The impact of the surround was determined by comparing changes in a cell’s overall spike rate between the moving and random surround conditions to spike responses when the surround was stationary (Figures 5B and 5C). Both motion and random surround stimulation significantly increased responses of On smooth monostratified cells (p = 8.0 × 10⁻⁴) and significantly decreased responses of broad thorny cells (p = 1.6 × 10⁻²; Figure 5), whereas activation of the surround by preferred-polarity changes in luminance were suppressive for both cell types (Figure 1). The increase in On smooth monostratified responses produced by the motion condition was significantly larger than that produced by the random condition (p = 9 × 10⁻³; Figure 4). Thus, dynamic activation of On smooth monostratified surrounds, particularly by coherent motion, enhances spiking in these cells.

To further quantify the effects of surround stimulation on ganglion cell response properties, we analyzed each cell’s responses to the flickering spot stimulus using a linear-nonlinear (LN) cascade model. In this framework, the relationship between the stimulus and spike output is captured by a linear filter that describes how the neuron integrates the stimulus over space and time, followed by a static nonlinearity that maps the filter output to firing rate.³³ A single linear filter was estimated for each cell using responses to all surround conditions, while separate nonlinearities were computed for each surround condition (Figures 6A–6C; see STAR Methods). Changes in the shape or position of the nonlinearity reflect alterations in a neuron’s sensitivity. For instance, increased sensitivity may appear as a steeper slope (i.e., a gain increase) or a leftward shift of the nonlinearity along the input axis. To determine whether motion in the receptive-field surround modulated the cell’s input-output relationship, we compared nonlinearities across surround conditions. These changes were modeled as alterations in gain or horizontal offset between conditions (see STAR Methods). We found consistent patterns within each cell type, with the relative contributions of gain and offset varying systematically between cell types.

Figure 6. Surround motion modulates ganglion cell response properties.

(A) Linear-nonlinear model framework. The contrast trajectory of the central spot was convolved with a linear temporal filter and the output of that filter (units, filtered contrast) was passed through a nonlinearity that maps the filter output to spike rate. The model was generated only by considering the flickering spot stimulus presented to the receptive-field center. Separate nonlinearities were computed for each of the three surround stimulus conditions.

(B) Example input-output nonlinearities for the three surround conditions from an On smooth monostratified cell. The nonlinearities are horizontally offset relative to one another, with the stationary surround shifted farthest to the right and the moving surround farthest to the left. The left inset displays the temporal filter for the same cell, while the right inset highlights a zoomed region of the nonlinearity. Numbers adjacent to each nonlinearity indicate the average spike rate across the displayed range of filtered contrasts.

(C) Example nonlinearities and corresponding average temporal filters for On and Off parasol, Off smooth monostratified, and broad thorny ganglion cell types. The temporal filters for each cell are shown as insets in their respective panels. For both Off parasol and broad thorny cells, the nonlinearities exhibit a clear reduction in slope under the moving surround condition compared to the stationary surround, indicating a decrease in response gain.

(D) Percentage change in output nonlinearity (spike rate) relative to the stationary surround condition, measured at input values corresponding to 40%–60% filtered contrast. Data are shown for the moving and random surround conditions across five ganglion cell types. Error bars represent mean ± SEM. Asterisks indicate statistically significant differences (p < 0.05; Wilcoxon signed-rank test).

(E) Change in gain versus horizontal offset of the nonlinearity for the moving surround condition relative to the stationary surround, based on simultaneous fits of both parameters in individual cells. Off-dominant cells exhibited minimal horizontal offsets (<0.03% in all cases), while On-type cells showed substantial and statistically significant offsets. Error bars represent mean ± SEM.

(F) Change in horizontal offset (left) and gain (right) for the moving and random surrounds relative to the stationary surround condition. Statistically significant changes are indicated with asterisks. Horizontal offsets are shown for On cell types only, as values for Off-dominant types were minimal (<0.03% on average across all three types). Gain changes are shown only for Off-dominant cells, as the large, statistically significant offsets in On cells limit interpretation of their gain values. Error bars represent mean ± SEM.

In On smooth monostratified cells, the nonlinearity associated with the moving surround was visibly shifted leftward along the x axis compared to that of the stationary surround (Figure 6B). A similar, though smaller, leftward shift was observed for the random surround condition. This shift resulted in increased spike rates across a broad range of input values. For instance, within the filtered contrast range of 0.5–1, the mean spike rate increased by 74.5% during surround motion relative to the stationary condition (Figures 6B and 6D). Functionally, this horizontal shift in the nonlinearity reflects neural sensitization or facilitation and corresponds to a decrease in spike threshold, enabling weaker inputs to elicit spikes.^34–37

For On smooth monostratified cells, variations in the nonlinearities were more accurately accounted for by horizontal offsets than by gain changes, as reflected in significantly lower mean squared error values for the offset-based fits (p = 2.5 × 10⁻⁴; n = 21). Moreover, the magnitude of the horizontal offset was significantly larger in the coherent motion condition compared to the random bar condition, indicating that facilitation was strongest when motion in the surround was spatially coherent (p = 1.1× 10⁻³; Figure 6F). Both the motion and random conditions produced statistically significant leftward shifts in the nonlinearity (p value: motion, 1.1 × 10⁻⁴; random, 1.1 × 10⁻³, n = 21 cells; Figure 6F, left). On parasol cells also exhibited significant leftward offsets under both surround conditions (p value: motion, 2.1 × 10⁻²; random, 2.1 × 10⁻², n = 15 cells; Figure 6F, left).

Surround motion reduced responsiveness in Off smooth monostratified, Off parasol, and broad thorny ganglion cells. Rather than producing a horizontal offset in the nonlinearity, surround motion primarily led to a reduction in its slope—that is, a decrease in gain (Figure 6C). Simultaneously fitting changes in gain and horizontal offset produced minimal changes in horizontal offset (<0.02% filtered contrast) in each of these Off-dominant cell types (horizontal offset: Off smooth monostratified, 0.015% ± 0.005%; Off parasol, 0.016% ± 0.006%; broad thorny, 0.002% ± 0.007%; Figure 6E; see STAR Methods). Thus, we focused our analysis on the magnitude of gain changes in these cell types. Indeed, both Off parasol and Off smooth monostratified cells showed significant decreases in gain relative to the stationary surround condition, reflecting a decrease in sensitivity (Off parasol p value: motion, 2.4 × 10⁻²; random, 2.0 × 10⁻³, n = 11 cells. Off smooth monostratified p value: motion, 1.2 × 10⁻²; random, 3.4 × 10⁻², n = 12 cells; Figure 6F, right).

These experiments reveal that motion in the receptive-field surround has cell-type-specific effects on ganglion cell responsiveness, with On smooth monostratified cells exhibiting particularly strong facilitation. This facilitation was significantly more pronounced under coherent motion compared to random surround conditions. In contrast, Off parasol and Off smooth monostratified cells exhibited decreased responsiveness during surround motion characterized by reductions in gain. Together, these findings demonstrate that coherent motion in the surround can facilitate or suppress ganglion cell output depending on cell type, with On smooth monostratified cells particularly exhibiting enhanced sensitivity via shifts in their input-output nonlinearities.

Facilitation in On smooth monostratified cells is present following motion offset

To assess whether facilitation in the On smooth monostratified cell persisted after the offset of surround motion, we presented a grating in a cell’s receptive-field surround that was either stationary or moving (duration, 1–3 s; Figure 7A). Following the offset of grating motion, a spot was flashed over the receptive-field center to measure the contrast-response function. For On smooth monostratified cells, the contrast-response function following the moving grating was shifted leftward along the x axis relative to the curve measured following the stationary grating (Figure 7C). This persistent leftward shift indicates that motion-induced facilitation continues even after the motion ceases (Figure 7C).

Figure 7. The effects of surround motion also present following motion offset.

(A) Spatial pattern and time course for stimulus designed to test the persistence of effects of surround motion on ganglion cell response properties. A squarewave grating was presented in the receptive-field surround of a cell (bar width, 50–100 μm; contrast, 0.5). The grating either drifted (speed, 1.0 mm s⁻¹; moving surround condition) or was stationary. A spot was flashed (duration, 250 ms) over the receptive-field center 0–50 ms following the offset of grating motion to probe cellular responses.

(B) Example spike rates for an On smooth monostratified cell for flashes of 0% (left) and +25% contrast (right; delay, 50 ms following motion offset). Shaded gray region indicates the timing of the test flash. Flashes elicited larger spike responses following the offset of the moving surround grating (red) than for the stationary control (black).

(C) Spike rate as a function of contrast for the On smooth monostratified cell in (B) for the moving and stationary surround conditions. The contrast-response curve for the moving surround (red) was shifted to the left relative to the stationary curve (black).

(D) Horizontal offset of the stationary surround curve relative to the moving surround curve across a population of On smooth monostratified (n = 12) and On parasol ganglion cells (n = 6). The offset values were statistically significant for On smooth monostratified cells (p = 9.8 × 10⁻⁴). Gray circles are individual cells. Error bars represent mean ± SEM.

(E) Whole-cell, voltage-clamp recordings of excitatory synaptic inputs from the On smooth monostratified cell in (B). Recordings are shown for 0% (left) and +25% contrast (right) flashes (delay, 50 ms following motion offset).

(F) Inhibitory synaptic currents to the same stimuli and in the same cell as in (E).

(G) Excitatory synaptic currents recorded at five different delays relative to motion offset in a population of On smooth monostratified cells (n = 5; contrast, 0.25–0.3). Delay values are given for the onset of the flash (duration, 250 ms) relative to motion offset. Currents were elevated for the moving surround relative to the stationary surround at delays <200 ms.

(H) Inhibitory synaptic currents were recorded at two different delay periods (n = 5 cells; contrast, 0.25–0.3). On average, inhibitory currents during surround motion were elevated relative to excitatory currents, but these differences were not statistically significant (p > 0.05).

(I) Synaptic current recordings from an On smooth monostratified ganglion cell. A moving grating was presented to the receptive-field surround while a constant gray stimulus was presented over the center region. Shaded region indicates period of surround motion. Inward excitatory synaptic currents were observed at motion onset and offset, consistent with surround modulation of presynaptic glutamate release from bipolar cells.

(J) Serial-inhibition circuit model for generating surround facilitation that is consistent with synaptic recordings in (I). An amacrine cell located in the receptive-field surround inhibits another amacrine cell, which, in turn, provides inhibition to bipolar cells that provide excitatory synaptic input to an On smooth monostratified cell. Activation of the surround amacrine cell inhibits the second amacrine cell, which then produces disinhibition at the level of the bipolar cell, resulting in an increased glutamate release onto the ganglion cell dendrites.

We quantified the horizontal offset between the contrast-response functions for the moving and stationary surround conditions using the same nonparametric fitting approach used in Figure 6. For On smooth monostratified cells, we observed a significant offset of +10.5% ± 2.5% contrast, indicating that approximately 10% more flash contrast was required under the stationary surround condition to elicit the same spike output as under the moving surround (p = 9.8 × 10⁻⁴; n = 12 cells; Figure 7D). In contrast, On parasol cells did not show a significant horizontal offset (mean offset: −6.3% ± 3.3% contrast; p = 0.22; n = 6 cells; Figure 7D).

Interactions between excitatory and inhibitory circuits play a critical role in shaping neural tuning and feature selectivity.^38–42 To assess the relative contributions of excitatory and inhibitory synaptic inputs to surround facilitation, we conducted whole-cell voltage-clamp recordings in On smooth monostratified cells during presentation of either moving or stationary gratings in the receptive-field surround. Consistent with the spiking responses, surround motion enhanced excitatory synaptic currents evoked by a test flash—producing larger inward currents for a given flash contrast (Figure 7E). Inhibitory synaptic currents were also elevated during surround motion (Figure 7F); however, such increases would typically suppress spiking. Moreover, due to the reduced driving force for inhibition, which is approximately 2–3 times less than excitation near spike threshold, inhibitory currents would need to be substantially larger than excitatory currents to appreciably impact spike output. Therefore, the enhanced spike output observed in response to surround motion is primarily driven by increased excitatory synaptic input.

We also observed a robust increase in spike rate following surround motion, even when the center stimulus had zero contrast. This response pattern reflects a well-established form of neural facilitation or sensitization, consistent with disinhibition occurring either presynaptically at bipolar cell terminals or postsynaptically at ganglion cell dendrites.^{34–36,43,44} Supporting this interpretation, excitatory synaptic recordings showed a corresponding inward current following the offset of surround motion under zero-contrast conditions, while inhibitory input was weak and delayed (Figures 7E and 7F). These findings indicate that the observed facilitation arises primarily from surround-driven modulation of bipolar cell output onto the dendrites of On smooth monostratified ganglion cells. The likely circuit mechanism underlying this effect is described further in the discussion.

To characterize the time course of motion-induced facilitation in excitatory synaptic inputs, we repeated this experiment while varying the interval between the offset of surround motion and the onset of the test flash. A flash was presented at one time point before grating offset (delay: −400 ms) and at four time points after offset (delays, +50, +100, +200, and +400 ms; flash contrast, 0.25–0.3; Figures 7G and 7H). Across cells, excitatory responses were significantly elevated during coherent surround motion (delay, −400 ms; p = 1.2 × 10⁻²). These responses peaked shortly after motion ceased (delay, +50 ms; p = 6.2 × 10⁻³) and gradually declined at longer delays (delay, +100 ms; p = 1.4 × 10⁻²), becoming statistically indistinguishable from the stationary surround by +200 ms (p > 0.05). These data indicate that facilitation persists following the offset of surround motion for at least 100 ms, a substantial fraction of the time window of visual fixation following ballistic eye movements.

DISCUSSION

A central pursuit of systems neuroscience is to understand how circuit structure shapes the computational properties of neural circuits. In this study, we demonstrate that the parallel output pathways in the primate retina exhibit notable differences in the strength and functional impact of their receptive-field surrounds. While our recordings were limited to a subset of ganglion cell types, the observed diversity in surround effects suggests that distinct retinal circuits may support specialized computations and stimulus selectivities. For example, broad thorny ganglion cells show a strongly suppressive surround that is highly sensitive to the spatial scale of motion correlations, reminiscent of object-motion-sensitive cells in other vertebrate retinas (Figures 2, 3, 4, and 5).^15–19 By comparison, the surround of the On smooth monostratified cell is suppressive for stationary, spatially uniform stimuli but facilitatory for motion—a functional profile that, to our knowledge, has not been described outside the primate retina (Figures 1, 2, 6, and 7).

Diversity of receptive-field surround strength in the primate retina

We observed substantial heterogeneity in both the strength of receptive-field surround antagonism and in its functional consequences across ganglion cell types (Figures 1, 2, 3, 4, 5, 6, and 7). Broad thorny and On smooth monostratified cells appeared to anchor opposite ends of this spectrum: broad thorny cells exhibited strong, suppressive surrounds reminiscent of object-motion-sensitive cells in other vertebrates, while On smooth monostratified cells displayed context-dependent modulation, with surround activation suppressing responses to stationary stimuli but enhancing responses to motion. Despite these categorical differences, the influence of the spatial scale of motion correlations on spike output varied continuously across cell types, suggesting a graded distribution of surround-mediated computations rather than discrete functional classes (Figure 4).

This continuum of surround effects likely arises from the intricate and varied ways in which amacrine cells—an exceptionally diverse interneuron population—interact with bipolar and ganglion cells, as well as with each other.^45–47 These interactions can modulate both excitatory and inhibitory signaling in a cell-type-specific manner, thereby shaping the temporal dynamics, spatial selectivity, and feature tuning of ganglion cell responses.^{15,16,18,39,48–52} However, the underlying circuit mechanisms remain incompletely understood. The functional diversity of amacrine cell types, their synaptic connectivity patterns, and their role in mediating context-dependent modulation in the primate retina are active areas of investigation, and understanding these diverse functions will be essential for a comprehensive understanding of how retinal circuits support parallel processing of complex visual scenes.

Source of surround facilitation in On smooth monostratified cells

The component of the receptive-field surround that originates from horizontal cells in the outer retina is common to all ganglion cell types and contributes to classical center-surround antagonism through lateral inhibition at the photoreceptor-to-bipolar cell synapse.^53,54 However, the diverse effects of surround activation and motion observed across ganglion cell types are unlikely to arise solely from this shared outer retinal mechanism. Instead, they likely reflect differences in the inner retinal circuitry, particularly in the influence of inhibitory amacrine cells within the inner plexiform layer.^55,56

One plausible circuit mechanism underlying the facilitation observed in On smooth monostratified cells during surround motion involves serial inhibition, a well-established motif in which one amacrine cell inhibits another.^57–59 In this scenario, a first amacrine cell provides direct inhibitory input to the bipolar cells that excite the ganglion cell and/or to the ganglion cell itself (Figure 7). This first amacrine cell is, in turn, inhibited by a second amacrine cell, which is activated by motion in the receptive-field surround. Activation of the second amacrine cell by surround motion would suppress the inhibitory influence of the first, thereby producing disinhibition—a net increase in excitatory drive at the level of the bipolar cell and ganglion cell. This disinhibitory pathway could account for the increased spiking observed in On smooth monostratified cells in response to dynamic surround stimulation (Figures 5, 6, and 7).

A critical feature of the observed facilitation is its dependence on motion—not merely on surround activation—suggesting that at least one element of the underlying circuit must be motion sensitive. In the proposed serial-inhibition motif, this implies that the first amacrine cell in the disinhibitory chain must exhibit motion sensitivity. Previous studies in mouse, salamander, rabbit, and primate retinas have shown that motion sensitivity in certain amacrine cell and ganglion cell types is largely inherited from their excitatory synaptic inputs.^{16–18,29,32,60,61} In these cells, motion sensitivity often arises through a combination of mechanisms, including spatiotemporal filtering of excitatory inputs, rectified transmission from bipolar or excitatory amacrine cells, and electrical coupling.^29,32,62 Amacrine cell-mediated inhibition can further sharpen or shape motion tuning, as demonstrated in prior work.^16,18,29,63 Together, these observations support the plausibility of a motion-sensitive interneuron as a critical element in the disinhibitory circuit underlying motion-dependent facilitation in On smooth monostratified ganglion cells.

While serial inhibition provides a parsimonious and mechanistically plausible explanation for motion-dependent facilitation, it remains one of several possible circuit configurations. Additional mechanisms—such as changes in amacrine cell transmitter release dynamics, modulation of bipolar cell gain, or plasticity in inhibitory feedback loops—may also contribute. Indeed, facilitation was evident both during surround motion (Figures 5, 6, and 7) and after motion offset (Figures 7), and in both cases was present in the excitatory synaptic inputs to the ganglion cell. This indicates that the underlying mechanism(s) act by modulating the excitatory outputs of diffuse bipolar cells, either through the interaction of multiple neural pathways or through the dynamics of a single circuit motif. Future studies combining connectomics and targeted recordings will be essential to resolve the precise circuit motifs and synaptic interactions that support motion-dependent computation in On smooth monostratified cells and other retinal ganglion cell types. These investigations will be critical for advancing our understanding of how diverse inhibitory pathways shape sensory processing in the primate retina.

Implications of surround motion sensitivity during natural viewing

Correlated motion can occur across a range of spatial scales during natural vision. For example, self-motion caused by eye movements or an observer’s movements through the environment generates long-range motion correlations across the visual field. In contrast, local motion correlations arise when an object moves relative to the observer, with the spatial extent of those correlations on the retina depending on the object’s size and distance. Sensitivity to these correlations offers clear ethological advantages, aiding in navigation, predator avoidance, prey capture, and other visually guided behaviors.

Our experiments revealed a broad spectrum of sensitivities to spatial-motion correlations across retinal ganglion cell types (Figure 4). Broad thorny ganglion cells exhibited strong suppression in response to medium- and long-range motion correlations, and several other ganglion cell types showed similar but weaker suppressive effects. In contrast, cells such as the On smooth monostratified ganglion cell were largely insensitive to the spatial extent of motion correlations and maintained robust responses even when correlations extended over large retinal distances. Natural scenes lack a characteristic spatial scale because objects of a given size can appear at varying distances from an observer. The same principle applies to motion, suggesting that the retina’s diverse sensitivities to motion correlation length may support efficient encoding of the wide range of motion signals encountered in natural vision.

The broad thorny as a candidate object-motion-sensitive cell in primates

Detecting objects moving independently of an animal’s own motion is a critical and behaviorally relevant visual computation. Ganglion cells with this type of motion selectivity have been identified in the retinas of salamanders, rabbits, and mice.^15–19 Our findings reveal that broad thorny ganglion cells in the primate retina exhibit similar stimulus selectivity: they display strong surround suppression, sluggish response kinetics, and respond most robustly to differential motion between the center and surround regions of the receptive field (Figures 1, 2, 3, and 4). Moreover, voltage-clamp recordings showed that this local motion sensitivity arises primarily from modulation of excitatory synaptic input—consistent with the mechanisms reported for object-motion-sensitive cells in non-primate species (Figure 3).^16–19

In addition to their functional similarities, broad thorny cells share key morphological features with object-motion-sensitive ganglion cells in other species, including comparable dendritic architecture and stratification within the inner plexiform layer.^19,20,64 Broad thorny cells may also utilize a similar circuit motif. In mice, object-motion-sensitive ganglion cells receive excitatory synaptic input from vGlut3-positive amacrine cells—a critical element of their motion sensitivity.^17,18 Recent electron microscopy studies of the macaque retina have shown that broad thorny ganglion cells receive synaptic contacts from an amacrine cell that resembles vGluT3-positive amacrine cell in the baboon retina.^65,66 However, further investigation is needed to clarify the extent of circuit similarities and differences between broad thorny cells and the object-motion-sensitive ganglion cells across species.

Feature selectivity in the dorsal visual pathway of primates

The classical center-surround receptive field has been a foundational concept in sensory neuroscience for more than a century.^1–4 Indeed, with few exceptions, models of the early vision in primates have been based on the idea that retinal responses can be adequately described as a simple weighted sum of the stimuli falling within the receptive-field center and surround regions.^67–71 However, over time, findings in several sensory systems, including the vertebrate retina, have challenged this traditional model of receptive-field organization.^72–80

Notably, in the medial temporal (MT) area of the primate cortex, cells have been found with center-surround receptive fields akin to those reported here for the broad thorny and On smooth monostratified ganglion cells—motion in the receptive-field surround strongly suppressed the responses of one set of cortical neurons and facilitated the responses of another set of neurons.⁷⁴ Other aspects of motion processing, such as motion direction, similarly appear to occur both at the retinal and cortical levels in primates^81,82 and non-primates.⁸³

The parallel computation of motion in the retina and cortex likely indicates its ethological importance, such as its roles in reflexive eye movements that center the fovea on specific image structure and in higher-level computations such as parsing a scene into objects. Here, we show that interactions of the receptive-field center and surround in primate retinal ganglion cells are not restricted to spatial filtering but instead also contribute to the extraction of visual motion.

Limitations of the study

The specific neural circuitry that provides synaptic input to broad thorny and smooth monostratified ganglion cells remains incompletely understood. In particular, the identity and connectivity of the presynaptic bipolar and amacrine cell types that shape their response properties have not yet been established. Due to the current lack of reliable genetic tools for targeting defined cell types in the primate retina, it is not yet possible to selectively manipulate or record from individual circuit elements to determine their precise contributions. As a result, while our findings reveal key physiological mechanisms underlying the feature selectivity of these ganglion cell types, the detailed circuit motifs that generate these computations cannot be definitively resolved. Future studies combining connectomics, targeted recordings, and circuit-level modeling will be essential to map these pathways and fully elucidate how specific synaptic interactions give rise to the observed response properties.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources should be directed to the lead contact, Michael Manookin (manookin@uw.edu).

Materials availability

This study did not generate any new materials.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Retinal tissue was obtained from macaque monkeys (M. fascicularis, M. mulatta, and M. nemestrina) of either sex through the tissue distribution program at the University of Washington National Primate Research Center in accordance with procedures approved by the University of Washington Institutional Animal Care and Use Committee.

METHOD DETAILS

Tissue preparation

Macaque monkey eyes were enucleated and hemisected, and vitreous humor was removed by hand. Retinas were dissected out of the eye under infrared light and kept in a ~32° C–34° C Ames bicarbonate solution (Sigma) bubbled with 5% CO₂ and 95% O₂. Before experimental use, retinas were dark adapted for at least 1 h. Pieces of pigmented epithelium-attached retina (2–8 mm, ~10°–30° foveal eccentricity at temporal or nasal positions relative to the eye) were mounted ganglion cell-side up onto a poly-D-lysine-coated coverslip (BD Biosciences) and placed in a dish perfused at ~6–8 mL min⁻¹ warmed (32° C) Ames solution. The tissue was illuminated by a 940 nm LED for visualization.

Electrophysiology and data analysis

To obtain spiking responses from ganglion cells, cell-attached, extracellular recordings were performed with borosilicate glass pipettes filled with Ames solution. Data was acquired at 10 kHz using a Multiclamp 700B amplifier (Molecular Devices) and Symphony data acquisition software. All analysis of spiking data was performed using customized scripts in MATLAB (R2019 and R2022b, Mathworks). Wavelet filtering was used to amplify spikes relative to the noise. Spike detection was performed using k-means clustering on spike waveforms or by choosing a manual threshold of standard deviation relative to baseline.

Cellular recordings were made using borosilicate glass pipettes filled with either Ames’ medium for extracellular spike recordings or a cesium-based internal solution for whole-cell recordings. The internal solution contained (in mM): 105 CsCH₃SO₃, 10 TEA-Cl, 20 HEPES, 10 EGTA, 2 QX-314, 5 Mg-ATP, and 0.5 Tris-GTP; pH was adjusted to ~7.3 with CsOH, and osmolarity to ~280 mOsm. Series resistance (typically 3–9 MΩ) was compensated online by 50%. Membrane potentials were corrected offline to account for the ~11 mV liquid junction potential between the intracellular and extracellular solutions. To isolate excitatory and inhibitory synaptic inputs, whole-cell, voltage-clamp recordings were performed by holding the cell at the estimated reversal potentials for chloride (approximately −70 mV) and cation-mediated excitation (0 mV), respectively.

We can unambiguously identify On and Off parasol cells, On/Off smooth monostratified cells, and broad thorny cells, and several additional cell types based on well-established physiological characteristics. These included characteristic light response properties (e.g., response polarity, transience, and contrast sensitivity), receptive-field organization, and response kinetics. Our identification protocols are grounded in prior published work from our group and others.^20,21,30, and these criteria allow for reliable cell-type classification in the macaque retina.

In addition, for a subset of the current experiments, we included 0.5–1.0% EZ-Link biocytin or biocytin hydrazide (ThermoFisher) in the internal recording solution to allow post hoc morphological recovery following fixation (4% paraformaldehyde) and immunostaining (streptavidin, Alexa Fluor 488; ThermoFisher). This approach provided further confirmation of dendritic stratification patterns and morphology in select cases. Cells for which the physiological identity could not be clearly established were excluded from the dataset.

Visual stimulation

Stimuli were displayed on either an OLED monitor (eMagin) or an Lightcrafter 4500 digital light projector (Texas Instruments) before being focused via a microscope objective onto the photoreceptors of each retinal specimen. Light levels were in the mid-to high-photopic range (4 × 10³ to 4 × 10⁴ cone photoisomerizerizations s⁻¹). Symphony (http://symphony-das.github.io) and Stage (http://stage-vss.github.io) software packages were used to generate visual stimuli. Stimuli were presented at 60 Hz refresh rates. Peristimulus time histograms of spike responses were calculated by convolving the binned spike rate with a Gaussian kernel (mean, 0.0; standard deviation, 6 ms).

Stimuli were centered on a cell’s receptive field using a sinusoidally modulated bar in either horizontal or vertical orientation (bar size, 50 × 1000 μm). The peak response along each axis was used as the x and y coordinates for centering subsequent stimuli. To estimate the size of the receptive-field center, we presented annuli with progressively increasing inner diameters (see Figure S1). The boundary between center and surround was defined as the smallest inner diameter at which the cell’s response either reversed in polarity (e.g., from On to Off) or was abolished entirely—indicating that the surround dominated the response.

Because the surround is engaged via an additional synapse relative to the center, it is typically delayed by at least 10–20 ms. If the receptive-field center were still significantly activated by an annulus, one would expect to observe a transient excitatory response to a preferred-polarity stimulus preceding the onset of surround suppression. The complete absence of such a response to a high-contrast annulus provides strong evidence that the center was not appreciably engaged. To further minimize potential overlap between center and surround activation, a mean-luminance buffer zone (25–50 μm wide) was added between the center and surround regions in subsequent experiments.

The average inner diameters used, pooled across cells and eccentricities, were as follows (mean ± standard deviation): On parasol, 356 ± 137 μm (n = 24); Off parasol, 368 ± 129 μm (n = 23); broad thorny, 437 ± 124 μm (n = 14); On smooth monostratified, 646 ± 199 μm (n = 28); Off smooth monostratified, 545 ± 49 μm (n = 12). These values exceed the published dendritic tree diameters for the same cell types at comparable eccentricities (mean ± standard deviation): On/Off parasol, 209 ± 70 μm; broad thorny, 369 ± 63 μm; On smooth monostratified, 423 ± 105 μm; Off smooth monostratified, 405 ± 127 μm.^20,68

To measure the effects of spatial correlations on ganglion cell responses to motion, we presented dynamic stimuli composed of black and white squares (equal in number) over a uniform gray background (Figure 4). Each square was initialized at a random location, drawn from a uniform distribution spanning the x- and y-dimensions of the display area. Square sizes ranged from 40 to 50 μm, and the spatial density was held constant at 200 squares mm⁻².

On each frame presentation, every square moved from its previous position to a new position in two-dimensional space. The initial displacement (D₀) on each axis (x and y) was randomly drawn from a zero-mean Gaussian distribution, with a standard deviation corresponding to a specified motion speed (500–1000 μm s⁻¹).

$$D_0=\left[\begin{array}{c}\Delta x_1\\ \Delta y_1\end{array}\begin{array}{c}\Delta x_2\\ \Delta y_2\end{array}\begin{array}{c}\mathrm{\dots }\\ \mathrm{\dots }\end{array}\begin{array}{c}\Delta x_i\\ \Delta y_i\end{array}\right]$$

where x_i and y_i indicate the x and y locations of the ith square, respectively. To introduce spatial correlations in motion, we imposed pairwise correlations in both the direction and magnitude of motion steps across squares. These correlations decayed exponentially with the Euclidean distance between square centers, such that nearby squares tended to move in more similar directions than distant ones. These correlations were represented with a covariance matrix (Σ) in which the off-diagonal elements represented the covariances between different pairs of squares. The entries of the covariance matrix were:

$$cov\left(i,j\right)=exp\left(\frac{-d_{i,j}}{\lambda }\right)$$

where d_i,j is the pairwise Euclidean distance between the i-th and j-th squares and λ is the correlation length constant that determines how rapidly the correlations decay over space. The resulting displacements (D) were calculated by taking the matrix product of the initial random displacements (D₀) and the covariance matrix:

$$D=D_0\Sigma$$

To prevent the accumulation of local clustering over time—a natural consequence of spatial correlations—we periodically reset the positions of all squares. Without this reset, the correlated motion causes squares to converge into tight clusters, degrading the uniformity of spatial coverage. The correlation time refers to the temporal window over which these spatial correlations persist before being reset; in our experiments, we tuned this value to maintain stable stimulus statistics while preserving the temporal structure of correlated motion (correlation time, 0.2 s).

The stimulus used for the linear-nonlinear model analysis (Figures 5 and 6) consisted of a flickering spot centered over a cell’s receptive field. Spot intensity was drawn from either a Gaussian or binary distribution. For Gaussian-distributed stimuli, the luminance values were drawn from a normal distribution with a mean equal to the background intensity and a standard deviation equal to 0.3–0.5 times the mean intensity (i.e., 30–50% root-mean-square contrast). For binary-distributed stimuli, the contrast was defined as the difference between the high and low luminance values divided by their sum, and was also set to 0.3–0.5 (i.e., 30–50% Michelson contrast). Binary flicker was used exclusively for broad thorny ganglion cells, which exhibit a strong preference for motion and respond weakly to stationary stimuli.^20,30

Analysis

We used a linear-nonlinear (LN) cascade model to assess how surround stimulation affects ganglion cell response properties. Importantly, this analysis was performed solely on the contrast trajectory of the flickering spot over the receptive-field center and information about the surround stimulus was not used in generating the model. The linear filter (F)—representing the cell’s impulse response to a brief light flash—was computed by cross-correlating the cell’s spike train (R) with the contrast trajectory of the flickering stimulus (S).³³ In the frequency domain, this was accomplished by multiplying the Fourier transform of the spike response with the complex conjugate of the Fourier transform of the stimulus. To account for stimulus correlations, the result was normalized by the stimulus autocorrelation.^29,84

$$\widehat{F}\left(\omega \right)=\frac{\widehat{S*}\left(\omega \right)\widehat{R}\left(\omega \right)}{\widehat{S}*\left(\omega \right)\widehat{S}\left(\omega \right)}$$

where $\widehat{S}\left(\omega \right)$ is the Fourier transform of S(t), $\widehat{R}\left(\omega \right)$ is the Fourier transform of R(t) and * denotes the complex conjugate. The inverse Fourier transform was then applied to return the linear filter to the time domain. The temporal filter was normalized to unit length by dividing by its vector magnitude (L2 norm). Because the different surround conditions had minimal impact on the filter’s shape, a single, common filter was computed across all conditions. This approach allowed us to isolate the effects of surround stimulation by examining changes in the nonlinearity alone.

For each of the three surround conditions, we computed the cell’s static nonlinearity by first convolving the temporal filter with the contrast trajectory of the flickering stimulus, producing a linear prediction expressed in units of filtered contrast. These filtered contrast values were then sorted and divided into 100 bins containing equal numbers of samples, and the average spike rate within each bin was calculated.

To quantify the effects of surround stimulation, we modeled changes in the nonlinearity as either a difference in slope (gain change) or a horizontal offset. Gain changes were estimated by scaling the x-axis (filtered contrast) of the stationary-surround nonlinearity to align it with that of the comparison condition (moving or random) using a nonparametric method published previously.⁸⁵ For each candidate scale factor, filtered contrast values were re-binned, and the mean squared error (MSE) between the resulting binned spike rates and those of the comparison nonlinearity was calculated. To avoid local minima, we performed an initial coarse search over a wide range of scale factors and then used the best value as the starting point for a nonlinear least-squares fit (using the Nelder-Mead algorithm via the fminsearch function in MATLAB).

Horizontal offsets were estimated using an analogous approach, except the fitted parameter was an additive shift applied to the x-axis rather than a multiplicative scale. For consistency, the offset or gain factor was applied to the nonlinearity from the moving or random surround condition so that positive values corresponded to an increase in responsiveness. Conceptually, the offset value indicates the amount of filtered contrast that must be added to the stationary-surround condition to produce the same spike rate as observed in the comparison condition. Finally, to assess whether gain changes or horizontal offsets best accounted for the observed differences in nonlinearities, we fit models that simultaneously included both parameters using the same optimization procedure.

Supplementary Material

Supplemental information can be found online at https//doi.org/10.1016/j.celrep.2025.116832.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Streptavidin, Alexa Fluor 488	ThermoFisher Cat# S11223	RRID:AB_2336881
Biological samples
Macaque retina	Regional Primate Research Center	N/A
Chemicals, peptides, and recombinant proteins
Ames’ medium	Sigma cat # 1420	N/A
Software and algorithms
Stage	Rieke Lab, U. of Washington	N/A
Symphony	Rieke Lab, U. of Washington	N/A
MATLAB	Mathworks	N/A
Adobe Illustrator	Adobe	N/A
Fiji/ImageJ	https://fiji.sc/	N/A
Igor Pro	Wavemetrics	N/A

Highlights.

• Receptive-field surround motion differentially modulates primate ganglion cell responses

• Broad thorny cells exhibit classical object-motion sensitivity enhanced by uncorrelated motion

• On smooth monostratified cells show motion-dependent facilitation from the surround

• Surround effects deviate from classical models and resemble cortical-like processing

Data and code availability

• Data from this study are available upon reasonable request from the lead contact.

• All analysis code was written in MATLAB version >R2018b. Code used in the manuscript is available from GitHub: https://github.com/mikemanookin/SurroundMotionPaper.git.

• Additional information needed to reanalyze the data is also available from the lead contact.