Search

Almost every neuron in the early visual system has some form of an antagonistic receptive field surround. But the function of the surround is poorly understood, especially under naturalistic stimulus conditions. Anatomical and functional characterization of the surround of retinal ganglion cells suggests that surround signals combine with the excitatory feedforward pathway upstream of an important synaptic nonlinearity between bipolar cells and postsynaptic ganglion cells. This circuit architecture suggests at least two unexplored consequences for receptive field structure. First, center and surround inputs should interact nonlinearly, even prior to the summation across visual space that forms the full receptive field center. Second, inputs to the surround may alter the sensitivity of the center to spatial contrast in a scene. We test these hypotheses using both synthetic and naturalistic visual stimuli, and find that the statistics of natural scenes promote these nonlinear interactions. This work shows that, unexpectedly, the surround modulates spatial contrast encoding based on visual context.

Adaptation is a salient property of sensory processing. All adaptational or gain control mechanisms face the challenge of obtaining a reliable estimate of the property of the input to be adapted to and obtaining this estimate sufficiently rapidly to be useful. Here, we explore how the primate retina balances the need to change gain rapidly and reliably when photons arrive rarely at individual rod photoreceptors. We find that the weakest backgrounds that decrease the gain of the retinal output signals are similar to those that increase human behavioral threshold, and identify a novel site of gain control in the retinal circuitry. Thus, surprisingly, the gain of retinal signals begins to decrease essentially as soon as background lights are detectable; under these conditions, gain control does not rely on a highly averaged estimate of the photon count, but instead signals from individual photon absorptions trigger changes in gain. DOI: http://dx.doi.org/10.7554/eLife.00467.001, eLife digest To process the sights and sounds around us, our senses must be attuned to a huge range of signals: from barely audible whispers to deafening rock concerts, and from dim glimmers of light to bright spotlights. Sensory neurons face the challenge of encoding this huge range of inputs within their much more restricted response range. Thus, neurons in our eyes and ears must continually adjust their gain or sensitivity to match changes in the light and sound inputs. These gain control processes must operate rapidly to keep up with the ever-changing input signals, but must also operate accurately so as not to distort the inputs. The trade-off between rapid and accurate gain control can be illustrated by considering how the retina processes information at low light levels. There are two main types of light-sensitive cells in the retina: rods and cones. Vision at night relies on the ability of the rods to detect single photons—the smallest unit of light. In starlight, an individual rod will register photons only rarely, and most of the time, the majority of the rods will not register any photons. Neurons in the retinal circuits that read out the rod signals receive input from hundreds or thousands of rods, and those rod inputs are highly amplified to allow detection of the responses produced when a tiny fraction of the rods absorbs a photon. But this amplification is dangerous, as it could easily saturate retinal signals when light levels increase. Gain control mechanisms are needed to avoid such saturation. Schwartz and Rieke now add to our understanding of this process by examining how the retinas of non-human primates behave in low light. They reveal that levels of background light that can only just be detected behaviorally trigger retinal gain controls; these gain controls operate when less than 1% of rods absorb a photon. Under these conditions, the physics of light itself will cause considerable variability in the stream of photons arriving at the retina, leading to high variability in the gain of retinal responses. Nonetheless, changes in gain occurred rapidly following changes in background, indicating that the underlying mechanisms spend little time averaging incident photons. Taken together, these findings will require revisiting our ideas about how adaptational mechanisms balance the competing demands of speed and reliability to help us see the world around us. DOI: http://dx.doi.org/10.7554/eLife.00467.002