Michael B. Manookin

Inhibitory networks shape information processing in the brain by integrating signals across space, regulating neuronal responses, and controlling the flow of signals through neural circuits. Understanding these computations requires linking the structure of individual inhibitory neuron types to their functional properties and network connectivity. The retina provides a tractable system for meeting this challenge, yet the organization of many retinal inhibitory circuits remains poorly understood. Here, we identify a population of polyaxonal amacrine cells recorded using multielectrode arrays as A1 amacrine cells, a morphologically defined interneuron type in primate retina. This identification is supported by a correspondence between properties of morphologically-identified A1 cells from single cell recordings and putative A1 amacrines in the array recordings. We then identified recursive bistratified ganglion cells in the array recordings based on their electrical coupling to A1 cells, linking this inhibitory network to retinal output pathways. A1 and recursive bistratified cells also share an asymmetry in receptive-field organization, with larger On than Off centers. Finally, A1 spike amplitudes decline during bursts, and analyses of their electrical images suggest that this change may cause some spikes to fail to propagate through their extensive axonal arbors. These findings establish the cellular identity and network organization of a long-range inhibitory circuit in the primate retina and reveal how neuronal architecture, connectivity, and spike dynamics may jointly shape the transmission of visual information.

The retina encodes visual information through parallel pathways formed by distinct retinal ganglion (RGC) types. Current understanding of how each cell type contributes to downstream visual processing and perception is limited by a lack of tools for selectively manipulating the activities of individual RGC types. Here, we introduce a computational framework called STORM (SpatioTemporal Orthogonal Response Modulation) that creates visual stimuli that aim to selectively modulate the activities of specific RGC types while minimally affecting the activities of nearby RGCs of non-targeted types. To illustrate the approach, demonstrate its flexibility, and identify conditions under which each simulated RGC type could be selectively manipulated, we first applied this framework to simplified models of RGC function, such as Linear-Nonlinear-Poisson (LNP) and LN-LN cascade models. We also deployed this approach on more complex (and more accurate) models of RGC responses, based on convolutional neural networks. We demonstrate that differences in spatial receptive fields, temporal dynamics, response polarities, and nonlinearities can provide sufficient degrees of freedom for designing stimuli that selectively manipulate individual RGC types. We also generalize this approach to conditions where the receptive field locations are unknown for in vivo and clinical applications.

Strategies to stimulate the regeneration of neurons in the adult central nervous system can offer universal solutions for neurodegenerative diseases. Taking lessons from naturally regenerating species, such as the zebrafish, we have previously shown that vector-mediated expression of proneural transcription factors can stimulate neurogenesis from the resident Müller glia (MG) population in the adult mouse retina, both in vitro and in vivo . To bring this closer to translation, we now show that vector-mediated expression of the proneural transcription factor ASCL1 can reprogram adult macaque MG into functional neurons. To this end, we established purified MG cultures and show they retain a mature transcriptomic profile that correlates with foveal and peripheral MG. Importantly, MG-derived neurons express retinal ganglion cell markers, can fire action potentials and have a transcriptome that overlaps with developing human and adult macaque retinal ganglion cells. To refine this approach for clinical application, we incorporated microRNA-124 target sites in the reprogramming cassette and show that this restricts expression to MG in mixed primary cultures and intact explant cultures of adult macaque retina. Regulating ASCL1 expression with microRNA-124 target sites maintained the reprogramming efficiency from adult MG cultures and improved the yield of RGC-like neurons from infant MG cultures. Most importantly, with this vector cassette we successfully reprogrammed macaque MG from both adult and infant retina into HuC/D+ neurons. Our findings demonstrate that ASCL1 can induce neurogenesis from macaque MG across ages and provide a targeted, effective strategy for potential clinical translation in retinal repair.

The visual image transmitted by the retina to the brain has long been understood in terms of spatial filtering by the center-surround receptive fields of retinal ganglion cells (RGCs). Recently, this textbook view has been challenged by the stunning functional diversity and specificity observed in ∼40 distinct RGC types in the mouse retina. However, it is unclear whether the ∼20 morphologically and molecularly identified RGC types in primates exhibit similar functional diversity, or instead exhibit center-surround organization at different spatial scales. Here, we reveal striking and surprising functional diversity in macaque and human RGC types using large-scale multi-electrode recordings from isolated macaque and human retinas. In addition to the five well-known primate RGC types, 18-27 types were distinguished by their functional properties, likely revealing several previously unknown types. Surprisingly, many of these cell types exhibited striking non-classical receptive field structure, including irregular spatial and chromatic properties not previously reported in any species. Qualitatively similar results were observed in recordings from the human retina. The receptive fields of less-understood RGC types formed uniform mosaics covering visual space, confirming their classification, and the morphological counterparts of two types were established using single-cell recording. The striking receptive field diversity was paralleled by distinctive responses to natural movies and complexity of visual computation. These findings suggest that diverse RGC types, rather than merely filtering the scene at different spatial scales, instead play specialized roles in human vision.

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.

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.

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 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.