Skip to main content

Deciphering the computations underlying visual processing: Repurposing of retinal cells and how they are decoded by the visual thalamus

Periodic Reporting for period 2 - RetinalRepurposing (Deciphering the computations underlying visual processing:Repurposing of retinal cells and how they are decoded by the visual thalamus)

Reporting period: 2019-04-01 to 2020-09-30

A key feature of the retina is its parallel processing: The visual information is split into >30 different channels via various types of retinal ganglion cells (RGCs), the output neurons of the retina. Each RGC encodes a specific modality in the visual field, such as edges, color, or motion. The modality assigned to each RGC is classically thought to be a fixed, hardwired property, one that does not change over the wide range of visual inputs the animal detects throughout its lifespan. For example, even though the mean light intensity can differ by a factor of one trillion over the course of the day – retinal neurons can adjust their gain to maintain sensitivity and modality preference to accommodate these massive changes in light levels. Unexpectedly, recent evidence reveals that retinal neurons can, in fact, change their core computations under several circumstances. For example, direction selective (DS) RGCs, which respond to motion in one (preferred) direction, can reverse their directional preference following a short repetitive stimulation. RGCs were also found to change their polarity preferences (e.g. from displaying an ‘On’ response to light increment to displaying an ‘Off’ response to light decrement) or change their spatial integration mode (i.e. linear vs. nonlinear summation) due to changes in ambient light level. Notably, all these examples describe changes in a property that is traditionally thought to characterize the retinal cell and define its identity. We refer to these newly discovered dramatic changes in RGC encoding properties as repurposing. We aim to establish the triggers and dimensions over which RGCs dynamically change their encoding properties, and decipher the mechanisms allowing for changes in hardwired neuronal circuits. By doing so, we will also shed light on the mechanisms underlying the ‘original’ encoding properties of the RGCs, before they dynamically changed and repurposed. Finally, we aim to resolve how the dynamic changes in RGCs’ core computations are being integrated and interpreted by retinal targets.
Differential encoding of the visual field in the mouse retina:
The finding that RGCs can dynamically change their light responses with changes in the visual input led us to hypothesize that RGCs may also change their light responses with their location within the retina. Conventionally, RGCs belonging to the same subtype are expected to share the same light responses all over the retinal area in order to homogeneously encode a specific visual property in the entire visual field. However, the visual input is often non-homogenic. For example, a small animal like the mouse views the environment from close to the ground, so the upper visual field, falling on the ventral retina, primarily represents the sky while the lower visual field, falling on the dorsal retina, primarily represents the ground. We studied one RGC subtype, the transient-Off-alpha RGC, and found that neurons belonging to this subtype change their light response duration with retinal location, as they gradually became more sustained along the ventral-dorsal axis, revealing >5-fold-longer duration responses in the dorsal retina. This change resulted from differential input from retinal interneurons called AII cells to the transient-Off-alpha RGCs. Our findings show that RGCs of the same subtype may differently process the visual information in different retinal locations, suggesting that retinal networks adjust to the prevailing visual image to enable optimized sampling of the visual scene.

Revealing the mechanisms allowing for dynamic changes in encoding properties of DS-RGCs:
DS-RGCs encode the direction of motion in the visual field: they respond strongly to an object moving in one (preferred) direction but not in the opposite direction. Extensive studies have demonstrated that the asymmetric response is hardwired and relies on asymmetric inhibitory inputs from starburst amacrine cells (SACs). Yet, we previously found that DS-RGCs can reverse their directional preference following a short repetitive visual stimulation. What are the mechanisms that allow DS-RGCs to overcome their circuit anatomy and reverse their directional preference?
On-SACs, like many other neurons in the visual system display antagonistic center-surround receptive field organization: they depolarize in response to light increment in the center, but depolarize in response to light decrement in the surround receptive field. We found that repetitive visual stimulation eliminates SAC center response and enhances SAC surround response. This change in receptive field organization causes a shift in SAC response time which underlies reversal in DS-RGCs. Thus, we identified antagonistic center-surround mechanism for retinal direction selectivity, in which center-mediated stimulation evokes preferred-direction responses whereas surround-mediated stimulation evokes null-direction responses.
Our findings demonstrate unexpected processing in a primary sensory organ, the retina. First, neurons of the same subtype may display distinct visual responses, potentially adjusting their encoding properties to the prevailing stimulus characteristics. This finding impacts the ongoing efforts to define neuronal cell types in the retina and in any other brain structure. Moreover, they should be considered when studying visual processing in the retina and along the entire visual pathways.
Second, our results imply that dynamic changes in encoding properties of RGCs may result from changes in the balance between center and surround receptive field. It was previously shown that surround strengths in the light adapted retina. This was suggested to sharpen the cell’s responses and enhance visual acuity. We suggest that on top of that, dynamic center-surround balance may underlie changes in the encoding properties of RGCs, allowing DS-RGCs to overcome anatomical constrains and reverse their directional preference.
Finally, these findings challenge the core of the established view of parallel processing in the retina and the entire visual system, highlighting the need to understand how the dynamic visual information obtained in the retina is transferred to and decoded by retinal targets.