Biophysical neural adaptation mechanisms enable artificial neural networks to capture dynamic retinal computation.

Adaptation is a universal aspect of neural systems that changes circuit computations to match prevailing inputs. These changes facilitate efficient encoding of sensory inputs while avoiding saturation. Conventional artificial neural networks (ANNs) have limited adaptive capabilities, hindering their ability to reliably predict neural output under dynamic input conditions. Can embedding neural adaptive mechanisms in ANNs improve their performance? To answer this question, we develop a new deep learning model of the retina that incorporates the biophysics of photoreceptor adaptation at the front-end of conventional convolutional neural networks (CNNs). These conventional CNNs build on 'Deep Retina,' a previously developed model of retinal ganglion cell (RGC) activity. CNNs that include this new photoreceptor layer outperform conventional CNN models at predicting male and female primate and rat RGC responses to naturalistic stimuli that include dynamic local intensity changes and large changes in the ambient illumination. These improved predictions result directly from adaptation within the phototransduction cascade. This research underscores the potential of embedding models of neural adaptation in ANNs and using them to determine how neural circuits manage the complexities of encoding natural inputs that are dynamic and span a large range of light levels.

Two Sides of the Same Coin: Efficient and Predictive Neural Coding.

Some visual properties are consistent across a wide range of environments, while other properties are more labile. The efficient coding hypothesis states that many of these regularities in the environment can be discarded from neural representations, thus allocating more of the brain's dynamic range to properties that are likely to vary. This paradigm is less clear about how the visual system prioritizes different pieces of information that vary across visual environments. One solution is to prioritize information that can be used to predict future events, particularly those that guide behavior. The relationship between the efficient coding and future prediction paradigms is an area of active investigation. In this review, we argue that these paradigms are complementary and often act on distinct components of the visual input. We also discuss how normative approaches to efficient coding and future prediction can be integrated.

Association of Environmental Factors with Age-Related Macular Degeneration using the Intelligent Research in Sight Registry.

Purpose: Investigate associations of natural environmental exposures with exudative and nonexudative age-related macular degeneration (AMD) across the United States. Design: Database study. Participants: Patients aged ≥ 55 years who were active in the IRIS Registry from 2016 to 2018 were analyzed. Patients were categorized as nonexudative, inactive exudative, and active exudative AMD by International Classification of Diseases 10th Revision and Current Procedural Terminology (CPT) codes. Patients without provider-level ZIP codes matching any ZIP code tabulation area were excluded. Methods: Environmental data were obtained from public sources including the US Geological Survey, National Renewable Energy Laboratory, National Oceanic and Atmospheric Administration, and Environmental Protection Agency. Multiple variable, mixed effects logistic regression models with random intercepts per ZIP code tabulation area quantified the association of each environmental variable with any AMD versus non-AMD patients, any exudative AMD versus nonexudative AMD, and active exudative AMD versus inactive exudative and nonexudative AMD using 3 separate models, while adjusting for age, sex, race, insurance type, smoking history, and phakic status. Main Outcome Measure: Odds ratios for environmental factors. Results: A total of 9 884 527 patients were included. Elevation, latitude, solar irradiance measured in global horizontal irradiance (GHI) and direct normal irradiance (DNI), temperature and precipitation variables, and pollution variables were included in our models. Statistically significant associations with active exudative AMD were GHI (odds ratio [OR], 3.848; 95% confidence interval [CI] with Bonferroni correction, 1.316-11.250), DNI (OR, 0.581; 95% CI, 0.370-0.913), latitude (OR, 1.110; 95% CI, 1.046-1.178), ozone (OR, 1.014; 95% CI, 1.004-1.025), and nitrogen dioxide (OR, 1.005; 95% CI, 1.000-1.010). The only significant environmental associations with any AMD were inches of snow in the winter (OR, 1.005; 95% CI, 1.001-1.009) and ozone (OR, 1.011; 95% CI, 1.003-1.019). Conclusions: The strongest environmental associations differed between AMD subgroups. The solar variables GHI, DNI, and latitude were significantly associated with active exudative AMD. Two pollutant variables, ozone and nitrogen dioxide, also showed positive associations with AMD. Further studies are warranted to investigate the clinical relevance of these associations. Our curated environmental dataset has been made publicly available at https://github.com/uw-biomedical-ml/AMD_environmental_dataset.

Neuroscience: Reliable and refined motion computations in the retina.

We can distinguish between the direction and speed of a moving object effortlessly, but this is actually a very challenging computational task. A new study demonstrates that this process begins at the first stages of visual processing in the retina.

Predictive encoding of motion begins in the primate retina.

Predictive motion encoding is an important aspect of visually guided behavior that allows animals to estimate the trajectory of moving objects. Motion prediction is understood primarily in the context of translational motion, but the environment contains other types of behaviorally salient motion correlation such as those produced by approaching or receding objects. However, the neural mechanisms that detect and predictively encode these correlations remain unclear. We report here that four of the parallel output pathways in the primate retina encode predictive motion information, and this encoding occurs for several classes of spatiotemporal correlation that are found in natural vision. Such predictive coding can be explained by known nonlinear circuit mechanisms that produce a nearly optimal encoding, with transmitted information approaching the theoretical limit imposed by the stimulus itself. Thus, these neural circuit mechanisms efficiently separate predictive information from nonpredictive information during the encoding process.

Selectivity to approaching motion in retinal inputs to the dorsal visual pathway.

To efficiently navigate through the environment and avoid potential threats, an animal must quickly detect the motion of approaching objects. Current models of primate vision place the origins of this complex computation in the visual cortex. Here, we report that detection of approaching motion begins in the retina. Several ganglion cell types, the retinal output neurons, show selectivity to approaching motion. Synaptic current recordings from these cells further reveal that this preference for approaching motion arises in the interplay between presynaptic excitatory and inhibitory circuit elements. These findings demonstrate how excitatory and inhibitory circuits interact to mediate an ethologically relevant neural function. Moreover, the elementary computations that detect approaching motion begin early in the visual stream of primates.

Wide-field amacrine cell inputs to ON parasol ganglion cells in macaque retina.

Parasol cells are one of the major types of primate retinal ganglion cells. The goal of this study was to describe the synaptic inputs that shape the light responses of the ON type of parasol cells, which are excited by increments in light intensity. A connectome from central macaque retina was generated by serial blockface scanning electron microscopy. Six neighboring ON parasol cells were reconstructed, and their synaptic inputs were analyzed. On average, they received 21% of their input from bipolar cells, excitatory local circuit neurons receiving input from cones. The majority of their input was from amacrine cells, local circuit neurons of the inner retina that are typically inhibitory. Their contributions to the neural circuit providing input to parasol cells are not well-understood, and the focus of this study was on the presynaptic wide-field amacrine cells, which provided 17% of the input to ON parasol cells. These are GABAergic amacrine cells with long, relatively straight dendrites, and sometimes also axons, that run in a single, narrow stratum of the inner plexiform layer. The presynaptic wide-field amacrine cells were reconstructed, and two types were identified based on their characteristic morphology. One presynaptic amacrine cell was identified as semilunar type 2, a polyaxonal cell that is electrically coupled to ON parasol cells. A second amacrine was identified as wiry type 2, a type known to be sensitive to motion. These inputs likely make ON parasol cells more sensitive to stimuli that are rapidly changing outside their classical receptive fields.

Photopharmacologic Vision Restoration Reduces Pathological Rhythmic Field Potentials in Blind Mouse Retina.

Photopharmacology has yielded compounds that have potential to restore impaired visual responses resulting from outer retinal degeneration diseases such as retinitis pigmentosa. Here we evaluate two photoswitchable azobenzene ion channel blockers, DAQ and DAA for vision restoration. DAQ exerts its effect primarily on RGCs, whereas DAA induces light-dependent spiking primarily through amacrine cell activation. Degeneration-induced local field potentials remain a major challenge common to all vision restoration approaches. These 5-10 Hz rhythmic potentials increase the background firing rate of retinal ganglion cells (RGCs) and overlay the stimulated response, thereby reducing signal-to-noise ratio. Along with the bipolar cell-selective photoswitch DAD and second-generation RGC-targeting photoswitch PhENAQ, we investigated the effects of DAA and DAQ on rhythmic local field potentials (LFPs) occurring in the degenerating retina. We found that photoswitches targeting neurons upstream of RGCs, DAA (amacrine cells) and DAD (bipolar cells) suppress the frequency of LFPs, while DAQ and PhENAQ (RGCs) had negligible effects on frequency or spectral power of LFPs. Taken together, these results demonstrate remarkable diversity of cell-type specificity of photoswitchable channel blockers in the retina and suggest that specific compounds may counter rhythmic LFPs to produce superior signal-to-noise characteristics in vision restoration.

Neural sensitization improves encoding fidelity in the primate retina.

An animal's motion through the environment can induce large and frequent fluctuations in light intensity on the retina. These fluctuations pose a major challenge to neural circuits tasked with encoding visual information, as they can cause cells to adapt and lose sensitivity. Here, we report that sensitization, a short-term plasticity mechanism, solves this difficult computational problem by maintaining neuronal sensitivity in the face of these fluctuations. The numerically dominant output pathway in the macaque monkey retina, the midget (parvocellular-projecting) pathway, undergoes sensitization under specific conditions, including simulated eye movements. Sensitization is present in the excitatory synaptic inputs from midget bipolar cells and is mediated by presynaptic disinhibition from a wide-field mechanism extending >0.5 mm along the retinal surface. Direct physiological recordings and a computational model indicate that sensitization in the midget pathway supports accurate sensory encoding and prevents a loss of responsiveness during dynamic visual processing.

Unusual Physiological Properties of Smooth Monostratified Ganglion Cell Types in Primate Retina.

The functions of the diverse retinal ganglion cell types in primates and the parallel visual pathways they initiate remain poorly understood. Here, unusual physiological and computational properties of the ON and OFF smooth monostratified ganglion cells are explored. Large-scale multi-electrode recordings from 48 macaque retinas revealed that these cells exhibit irregular receptive field structure composed of spatially segregated hotspots, quite different from the classic center-surround model of retinal receptive fields. Surprisingly, visual stimulation of different hotspots in the same cell produced spikes with subtly different spatiotemporal voltage signatures, consistent with a dendritic contribution to hotspot structure. Targeted visual stimulation and computational inference demonstrated strong nonlinear subunit properties associated with each hotspot, supporting a model in which the hotspots apply nonlinearities at a larger spatial scale than bipolar cells. These findings reveal a previously unreported nonlinear mechanism in the output of the primate retina that contributes to signaling spatial information.

Neural Mechanisms Mediating Motion Sensitivity in Parasol Ganglion Cells of the Primate Retina.

Considerable theoretical and experimental effort has been dedicated to understanding how neural circuits detect visual motion. In primates, much is known about the cortical circuits that contribute to motion processing, but the role of the retina in this fundamental neural computation is poorly understood. Here, we used a combination of extracellular and whole-cell recording to test for motion sensitivity in the two main classes of output neurons in the primate retina-midget (parvocellular-projecting) and parasol (magnocellular-projecting) ganglion cells. We report that parasol, but not midget, ganglion cells are motion sensitive. This motion sensitivity is present in synaptic excitation and disinhibition from presynaptic bipolar cells and amacrine cells, respectively. Moreover, electrical coupling between neighboring bipolar cells and the nonlinear nature of synaptic release contribute to the observed motion sensitivity. Our findings indicate that motion computations arise far earlier in the primate visual stream than previously thought.

Photopharmacological control of bipolar cells restores visual function in blind mice.

Photopharmacological control of neuronal activity using synthetic photochromic ligands, or photoswitches, is a promising approach for restoring visual function in patients suffering from degenerative retinal diseases. Azobenzene photoswitches, such as AAQ and DENAQ, have been shown to restore the responses of retinal ganglion cells to light in mouse models of retinal degeneration but do not recapitulate native retinal signal processing. Here, we describe diethylamino-azo-diethylamino (DAD), a third-generation photoswitch that is capable of restoring retinal ganglion cell light responses to blue or white light. In acute brain slices of murine layer 2/3 cortical neurons, we determined that the photoswitch quickly relaxes to its inactive form in the dark. DAD is not permanently charged, and the uncharged form enables the photoswitch to rapidly and effectively cross biological barriers and thereby access and photosensitize retinal neurons. Intravitreal injection of DAD restored retinal light responses and light-driven behavior to blind mice. Unlike DENAQ, DAD acts upstream of retinal ganglion cells, primarily conferring light sensitivity to bipolar cells. Moreover, DAD was capable of generating ON and OFF visual responses in the blind retina by utilizing intrinsic retinal circuitry, which may be advantageous for restoring visual function.

Neuroscience: Peeking Under the Sombrero.

A recent study has introduced a new analytical approach to understanding neural circuits which has revealed previously hidden neural interactions in a large population of cells in the primate retina. The neural circuit described likely contributes to encoding visual motion.

Distinctive receptive field and physiological properties of a wide-field amacrine cell in the macaque monkey retina.

At early stages of visual processing, receptive fields are typically described as subtending local regions of space and thus performing computations on a narrow spatial scale. Nevertheless, stimulation well outside of the classical receptive field can exert clear and significant effects on visual processing. Given the distances over which they occur, the retinal mechanisms responsible for these long-range effects would certainly require signal propagation via active membrane properties. Here the physiology of a wide-field amacrine cell-the wiry cell-in macaque monkey retina is explored, revealing receptive fields that represent a striking departure from the classic structure. A single wiry cell integrates signals over wide regions of retina, 5-10 times larger than the classic receptive fields of most retinal ganglion cells. Wiry cells integrate signals over space much more effectively than predicted from passive signal propagation, and spatial integration is strongly attenuated during blockade of NMDA spikes but integration is insensitive to blockade of NaV channels with TTX. Thus these cells appear well suited for contributing to the long-range interactions of visual signals that characterize many aspects of visual perception.

Broad thorny ganglion cells: a candidate for visual pursuit error signaling in the primate retina.

Functional analyses exist only for a few of the morphologically described primate ganglion cell types, and their correlates in other mammalian species remain elusive. Here, we recorded light responses of broad thorny cells in the whole-mounted macaque retina. They showed ON-OFF-center light responses that were strongly suppressed by stimulation of the receptive field surround. Spike responses were delayed compared with parasol ganglion cells and other ON-OFF cells, including recursive bistratified ganglion cells and A1 amacrine cells. The receptive field structure was shaped by direct excitatory synaptic input and strong presynaptic and postsynaptic inhibition in both ON and OFF pathways. The cells responded strongly to dark or bright stimuli moving either in or out of the receptive field, independent of the direction of motion. However, they did not show a maintained spike response either to a uniform background or to a drifting plaid pattern. These properties could be ideally suited for guiding movements involved in visual pursuit. The functional characteristics reported here permit the first direct cross-species comparison of putative homologous ganglion cell types. Based on morphological similarities, broad thorny ganglion cells have been proposed to be homologs of rabbit local edge detector ganglion cells, but we now show that the two cells have quite distinct physiological properties. Thus, our data argue against broad thorny cells as the homologs of local edge detector cells.

NMDA and AMPA receptors contribute similarly to temporal processing in mammalian retinal ganglion cells.

Postsynaptic AMPA- and NMDA-type glutamate receptors (AMPARs, NMDARs) are commonly expressed at the same synapses. AMPARs are thought to mediate the majority of fast excitatory neurotransmission whereas NMDARs, with their relatively slower kinetics and higher Ca(2+) permeability, are thought to mediate synaptic plasticity, especially in neural circuits devoted to learning and memory. In sensory neurons, however, the roles of AMPARs and NMDARs are less well understood. Here, we tested in the in vitro guinea pig retina whether AMPARs and NMDARs differentially support temporal contrast encoding by two ganglion cell types. In both OFF Alpha and Delta ganglion cells, contrast stimulation evoked an NMDAR-mediated response with a characteristic J-shaped I-V relationship. In OFF Delta cells, AMPAR- and NMDAR-mediated responses could be modulated at low frequencies but were suppressed during 10 Hz stimulation, when responses were instead shaped by synaptic inhibition. With inhibition blocked, both AMPAR- and NMDAR-mediated responses could be modulated at 10 Hz, indicating that NMDAR kinetics do not limit temporal encoding. In OFF Alpha cells, NMDAR-mediated responses followed stimuli at frequencies up to ∼18 Hz. In both cell types, NMDAR-mediated responses to contrast modulation at 9-18 Hz showed delays of <10 ms relative to AMPAR-mediated responses. Thus, NMDARs combine with AMPARs to encode rapidly modulated glutamate release, and NMDAR kinetics do not limit temporal coding by OFF Alpha and Delta ganglion cells substantially. Furthermore, glutamatergic transmission is differentially regulated across bipolar cell pathways: in some, release is suppressed at high temporal frequencies by presynaptic inhibition.

Specialized synaptic pathway for chromatic signals beneath S-cone photoreceptors is common to human, Old and New World primates.

The distribution of the soluble NSF-attachment protein receptor protein syntaxin-4 and the Na-K-Cl cotransporter (NKCC) were investigated in the outer plexiform layer of human retina using immunohistochemistry. Both proteins, which are proposed to be components of a gamma-aminobutyric acid mediated feed-forward circuit from horizontal cells directly to bipolar cells, were enriched beneath S-cones. The expression pattern of syntaxin-4 was further analyzed in baboon and marmoset to determine if the synaptic specialization is common to primates. Syntaxin-4 was enriched beneath S-cones in both species, which together with the human results indicates that this specialization may have evolved for the purpose of mediating unique color vision capacities that are exclusive to primates.

Horizontal cell feedback without cone type-selective inhibition mediates "red-green" color opponency in midget ganglion cells of the primate retina.

The distinctive red-green dimension of human and nonhuman primate color perception arose relatively recently in the primate lineage with the appearance of separate long (L) and middle (M) wavelength-sensitive cone photoreceptor types. "Midget" ganglion cells of the retina use center-surround receptive field structure to combine L and M cone signals antagonistically and thereby establish a "red-green, color-opponent" visual pathway. However, the synaptic origin of red-green opponency is unknown, and conflicting evidence for either random or L versus M cone-selective inhibitory circuits has divergent implications for the developmental and evolutionary origins of trichromatic color vision. Here we directly measure the synaptic conductances evoked by selective L or M cone stimulation in the midget ganglion cell dendritic tree and show that L versus M cone opponency arises presynaptic to the midget cell and is transmitted entirely by modulation of an excitatory conductance. L and M cone synaptic inhibition is feedforward and thus occurs in phase with excitation for both cone types. Block of GABAergic and glycinergic receptors does not attenuate or modify L versus M cone antagonism, discounting both presynaptic and postsynaptic inhibition as sources of cone opponency. In sharp contrast, enrichment of retinal pH-buffering capacity, to attenuate negative feedback from horizontal cells that sum L and M cone inputs linearly and without selectivity, completely abolished both the midget cell surround and all chromatic opponency. Thus, red-green opponency appears to arise via outer retinal horizontal cell feedback that is not cone type selective without recourse to any inner retinal L versus M cone inhibitory pathways.

NMDA receptor contributions to visual contrast coding.

In the retina, it is not well understood how visual processing depends on AMPA- and NMDA-type glutamate receptors. Here we investigated how these receptors contribute to contrast coding in identified guinea pig ganglion cell types in vitro. NMDA-mediated responses were negligible in ON alpha cells but substantial in OFF alpha and delta cells. OFF delta cell NMDA receptors were composed of GluN2B subunits. Using a novel deconvolution method, we determined the individual contributions of AMPA, NMDA, and inhibitory currents to light responses of each cell type. OFF alpha and delta cells used NMDA receptors for encoding either the full contrast range (alpha), including near-threshold responses, or only a high range (delta). However, contrast sensitivity depended substantially on NMDA receptors only in OFF alpha cells. NMDA receptors contribute to visual contrast coding in a cell-type-specific manner. Certain cell types generate excitatory responses using primarily AMPA receptors or disinhibition.

Distinct expressions of contrast gain control in parallel synaptic pathways converging on a retinal ganglion cell.

Visual neurons adapt to increases in stimulus contrast by reducing their response sensitivity and decreasing their integration time, a collective process known as 'contrast gain control.' In retinal ganglion cells, gain control arises at two stages: an intrinsic mechanism related to spike generation, and a synaptic mechanism in retinal pathways. Here, we tested whether gain control is expressed similarly by three synaptic pathways that converge on an OFF alpha/Y-type ganglion cell: excitatory inputs driven by OFF cone bipolar cells; inhibitory inputs driven by ON cone bipolar cells; and inhibitory inputs driven by rod bipolar cells. We made whole-cell recordings of membrane current in guinea pig ganglion cells in vitro. At high contrast, OFF bipolar cell-mediated excitatory input reduced gain and shortened integration time. Inhibitory input was measured by clamping voltage near 0 mV or by recording in the presence of ionotropic glutamate receptor (iGluR) antagonists to isolate the following circuit: cone --> ON cone bipolar cell --> AII amacrine cell --> OFF ganglion cell. At high contrast, this input reduced gain with no effect on integration time. Mean luminance was reduced 1000-fold to recruit the rod bipolar pathway: rod --> rod bipolar cell --> AII cell --> OFF ganglion cell. The spiking response, measured with loose-patch recording, adapted despite essentially no gain control in synaptic currents. Thus, cone bipolar-driven pathways adapt differently, with kinetic effects confined to the excitatory OFF pathway. The ON bipolar-mediated inhibition reduced gain at high contrast by a mechanism that did not require an iGluR. Under rod bipolar-driven conditions, ganglion cell firing showed gain control that was explained primarily by an intrinsic property.