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.

DART.2: bidirectional synaptic pharmacology with thousandfold cellular specificity.

Precision pharmacology aims to manipulate specific cellular interactions within complex tissues. In this pursuit, we introduce DART.2 (drug acutely restricted by tethering), a second-generation cell-specific pharmacology technology. The core advance is optimized cellular specificity-up to 3,000-fold in 15 min-enabling the targeted delivery of even epileptogenic drugs without off-target effects. Additionally, we introduce brain-wide dosing methods as an alternative to local cannulation and tracer reagents for brain-wide dose quantification. We describe four pharmaceuticals-two that antagonize excitatory and inhibitory postsynaptic receptors, and two that allosterically potentiate these receptors. Their versatility is showcased across multiple mouse-brain regions, including cerebellum, striatum, visual cortex and retina. Finally, in the ventral tegmental area, we find that blocking inhibitory inputs to dopamine neurons accelerates locomotion, contrasting with previous optogenetic and pharmacological findings. Beyond enabling the bidirectional perturbation of chemical synapses, these reagents offer intersectional precision-between genetically defined postsynaptic cells and neurotransmitter-defined presynaptic partners.

Local synaptic inhibition mediates cerebellar granule cell pattern separation and enables learned sensorimotor associations.

The cerebellar cortex has a key role in generating predictive sensorimotor associations. To do so, the granule cell layer is thought to establish unique sensorimotor representations for learning. However, how this is achieved and how granule cell population responses contribute to behavior have remained unclear. To address these questions, we have used in vivo calcium imaging and granule cell-specific pharmacological manipulation of synaptic inhibition in awake, behaving mice. These experiments indicate that inhibition sparsens and thresholds sensory responses, limiting overlap between sensory ensembles and preventing spiking in many granule cells that receive excitatory input. Moreover, inhibition can be recruited in a stimulus-specific manner to powerfully decorrelate multisensory ensembles. Consistent with these results, granule cell inhibition is required for accurate cerebellum-dependent sensorimotor behavior. These data thus reveal key mechanisms for granule cell layer pattern separation beyond those envisioned by classical models.

Population encoding of stimulus features along the visual hierarchy.

The retina and primary visual cortex (V1) both exhibit diverse neural populations sensitive to diverse visual features. Yet it remains unclear how neural populations in each area partition stimulus space to span these features. One possibility is that neural populations are organized into discrete groups of neurons, with each group signaling a particular constellation of features. Alternatively, neurons could be continuously distributed across feature-encoding space. To distinguish these possibilities, we presented a battery of visual stimuli to the mouse retina and V1 while measuring neural responses with multi-electrode arrays. Using machine learning approaches, we developed a manifold embedding technique that captures how neural populations partition feature space and how visual responses correlate with physiological and anatomical properties of individual neurons. We show that retinal populations discretely encode features, while V1 populations provide a more continuous representation. Applying the same analysis approach to convolutional neural networks that model visual processing, we demonstrate that they partition features much more similarly to the retina, indicating they are more like big retinas than little brains.

GABAergic Inhibition Controls Receptive Field Size, Sensitivity, and Contrast Preference of Direction Selective Retinal Ganglion Cells Near the Threshold of Vision.

Information about motion is encoded by direction-selective retinal ganglion cells (DSGCs). These cells reliably transmit this information across a broad range of light levels, spanning moonlight to sunlight. Previous work indicates that adaptation to low light levels causes heterogeneous changes to the direction tuning of ON-OFF (oo)DSGCs and suggests that superior-preferring ON-OFF DSGCs (s-DSGCs) are biased toward detecting stimuli rather than precisely signaling direction. Using a large-scale multielectrode array, we measured the absolute sensitivity of ooDSGCs and found that s-DSGCs are 10-fold more sensitive to dim flashes of light than other ooDSGCs. We measured their receptive field (RF) sizes and found that s-DSGCs also have larger receptive fields than other ooDSGCs; however, the size difference does not fully explain the sensitivity difference. Using a conditional knock-out of gap junctions and pharmacological manipulations, we demonstrate that GABA-mediated inhibition contributes to the difference in absolute sensitivity and receptive field size at low light levels, while the connexin36-mediated gap junction coupling plays a minor role. We further show that under scotopic conditions, ooDSGCs exhibit only an ON response, but pharmacologically removing GABA-mediated inhibition unmasks an OFF response. These results reveal that GABAergic inhibition controls and differentially modulates the responses of ooDSGCs under scotopic conditions.

Late gene therapy limits the restoration of retinal function in a mouse model of retinitis pigmentosa.

Retinitis pigmentosa is an inherited photoreceptor degeneration that begins with rod loss followed by cone loss. This cell loss greatly diminishes vision, with most patients becoming legally blind. Gene therapies are being developed, but it is unknown how retinal function depends on the time of intervention. To uncover this dependence, we utilize a mouse model of retinitis pigmentosa capable of artificial genetic rescue. This model enables a benchmark of best-case gene therapy by removing variables that complicate answering this question. Complete genetic rescue was performed at 25%, 50%, and 70% rod loss (early, mid and late, respectively). Early and mid treatment restore retinal output to near wild-type levels. Late treatment retinas exhibit continued, albeit slowed, loss of sensitivity and signal fidelity among retinal ganglion cells, as well as persistent gliosis. We conclude that gene replacement therapies delivered after 50% rod loss are unlikely to restore visual function to normal. This is critical information for administering gene therapies to rescue vision.

Decomposition of retinal ganglion cell electrical images for cell type and functional inference.

Identifying neuronal cell types and their biophysical properties based on their extracellular electrical features is a major challenge for experimental neuroscience and the development of high-resolution brain-machine interfaces. One example is identification of retinal ganglion cell (RGC) types and their visual response properties, which is fundamental for developing future electronic implants that can restore vision. The electrical image (EI) of a RGC, or the mean spatio-temporal voltage footprint of its recorded spikes on a high-density electrode array, contains substantial information about its anatomical, morphological, and functional properties. However, the analysis of these properties is complex because of the high-dimensional nature of the EI. We present a novel optimization-based algorithm to decompose electrical image into a low-dimensional, biophysically-based representation: the temporally-shifted superposition of three learned basis waveforms corresponding to spike waveforms produced in the somatic, dendritic and axonal cellular compartments. Large-scale multi-electrode recordings from the macaque retina were used to test the effectiveness of the decomposition. The decomposition accurately localized the somatic and dendritic compartments of the cell. The imputed dendritic fields of RGCs correctly predicted the location and shape of their visual receptive fields. The inferred waveform amplitudes and shapes accurately identified the four major primate RGC types (ON and OFF midget and parasol cells), a substantial advance. Together, these findings may contribute to more accurate inference of RGC types and their original light responses in the degenerated retina, with possible implications for other electrical imaging applications.

Population encoding of stimulus features along the visual hierarchy.

The retina and primary visual cortex (V1) both exhibit diverse neural populations sensitive to diverse visual features. Yet it remains unclear how neural populations in each area partition stimulus space to span these features. One possibility is that neural populations are organized into discrete groups of neurons, with each group signaling a particular constellation of features. Alternatively, neurons could be continuously distributed across feature-encoding space. To distinguish these possibilities, we presented a battery of visual stimuli to mouse retina and V1 while measuring neural responses with multi-electrode arrays. Using machine learning approaches, we developed a manifold embedding technique that captures how neural populations partition feature space and how visual responses correlate with physiological and anatomical properties of individual neurons. We show that retinal populations discretely encode features, while V1 populations provide a more continuous representation. Applying the same analysis approach to convolutional neural networks that model visual processing, we demonstrate that they partition features much more similarly to the retina, indicating they are more like big retinas than little brains.

Large-scale interrogation of retinal cell functions by 1-photon light-sheet microscopy.

Visual processing in the retina depends on the collective activity of large ensembles of neurons organized in different layers. Current techniques for measuring activity of layer-specific neural ensembles rely on expensive pulsed infrared lasers to drive 2-photon activation of calcium-dependent fluorescent reporters. We present a 1-photon light-sheet imaging system that can measure the activity in hundreds of neurons in the ex vivo retina over a large field of view while presenting visual stimuli. This allows for a reliable functional classification of different retinal cell types. We also demonstrate that the system has sufficient resolution to image calcium entry at individual synaptic release sites across the axon terminals of dozens of simultaneously imaged bipolar cells. The simple design, large field of view, and fast image acquisition make this a powerful system for high-throughput and high-resolution measurements of retinal processing at a fraction of the cost of alternative approaches.

Late gene therapy limits the restoration of retinal function in a mouse model of retinitis pigmentosa.

Retinitis pigmentosa is an inherited photoreceptor degeneration that begins with rod loss followed by cone loss and eventual blindness. Gene therapies are being developed, but it is unknown how retinal function depends on the time of intervention. To uncover this dependence, we utilized a mouse model of retinitis pigmentosa capable of artificial genetic rescue. This model enables a benchmark of best-case gene therapy by removing the variables that complicate the ability to answer this vital question. Complete genetic rescue was performed at 25%, 50%, and 70% rod loss (early, mid and late, respectively). Early and mid treatment restored retinal function to near wild-type levels, specifically the sensitivity and signal fidelity of retinal ganglion cells (RGCs), the 'output' neurons of the retina. However, some anatomical defects persisted. Late treatment retinas exhibited continued, albeit slowed, loss of sensitivity and signal fidelity among RGCs, as well as persistent gliosis. We conclude that gene replacement therapies delivered after 50% rod loss are unlikely to restore visual function to normal. This is critical information for administering gene therapies to rescue vision.

Scalable Variational Inference for Low-Rank Spatiotemporal Receptive Fields.

An important problem in systems neuroscience is to characterize how a neuron integrates sensory inputs across space and time. The linear receptive field provides a mathematical characterization of this weighting function and is commonly used to quantify neural response properties and classify cell types. However, estimating receptive fields is difficult in settings with limited data and correlated or high-dimensional stimuli. To overcome these difficulties, we propose a hierarchical model designed to flexibly parameterize low-rank receptive fields. The model includes gaussian process priors over spatial and temporal components of the receptive field, encouraging smoothness in space and time. We also propose a new temporal prior, temporal relevance determination, which imposes a variable degree of smoothness as a function of time lag. We derive a scalable algorithm for variational Bayesian inference for both spatial and temporal receptive field components and hyperparameters. The resulting estimator scales to high-dimensional settings in which full-rank maximum likelihood or a posteriori estimates are intractable. We evaluate our approach on neural data from rat retina and primate cortex and show that it substantially outperforms a variety of existing estimators. Our modeling approach will have useful extensions to a variety of other high-dimensional inference problems with smooth or low-rank structure.

Cones and cone pathways remain functional in advanced retinal degeneration.

Most defects causing retinal degeneration in retinitis pigmentosa (RP) are rod-specific mutations, but the subsequent degeneration of cones, which produces loss of daylight vision and high-acuity perception, is the most debilitating feature of the disease. To understand better why cones degenerate and how cone vision might be restored, we have made the first single-cell recordings of light responses from degenerating cones and retinal interneurons after most rods have died and cones have lost their outer-segment disk membranes and synaptic pedicles. We show that degenerating cones have functional cyclic-nucleotide-gated channels and can continue to give light responses, apparently produced by opsin localized either to small areas of organized membrane near the ciliary axoneme or distributed throughout the inner segment. Light responses of second-order horizontal and bipolar cells are less sensitive but otherwise resemble those of normal retina. Furthermore, retinal output as reflected in responses of ganglion cells is less sensitive but maintains spatiotemporal receptive fields at cone-mediated light levels. Together, these findings show that cones and their retinal pathways can remain functional even as degeneration is progressing, an encouraging result for future research aimed at enhancing the light sensitivity of residual cones to restore vision in patients with genetically inherited retinal degeneration.

Neuroscience: Visual restoration with optogenetics.

Treating photoreceptor degenerative diseases is an exciting application of optogenetic technologies. However, there are significant challenges, such as producing normal visual signaling as the retina rewires in response to photoreceptor death. However, a new study shows remarkable functional stability in retinal circuits that can be engaged by optogenetics following photoreceptor loss.

An optical approach for mapping functional connectivity at single-cell resolution in brain circuits.

In the current issue of Cell Reports Methods, Spampinato et al. demonstrate a multiplexed system combining holographic photo-stimulation and functional imaging that may offer a generalizable approach for revealing how signals interact in complex neural circuits.

Robust cone-mediated signaling persists late into rod photoreceptor degeneration.

Rod photoreceptor degeneration causes deterioration in the morphology and physiology of cone photoreceptors along with changes in retinal circuits. These changes could diminish visual signaling at cone-mediated light levels, thereby limiting the efficacy of treatments such as gene therapy for rescuing normal, cone-mediated vision. However, the impact of progressive rod death on cone-mediated signaling remains unclear. To investigate the fidelity of retinal ganglion cell (RGC) signaling throughout disease progression, we used a mouse model of rod degeneration (Cngb1neo/neo). Despite clear deterioration of cone morphology with rod death, cone-mediated signaling among RGCs remained surprisingly robust: spatiotemporal receptive fields changed little and the mutual information between stimuli and spiking responses was relatively constant. This relative stability held until nearly all rods had died and cones had completely lost well-formed outer segments. Interestingly, RGC information rates were higher and more stable for natural movies than checkerboard noise as degeneration progressed. The main change in RGC responses with photoreceptor degeneration was a decrease in response gain. These results suggest that gene therapies for rod degenerative diseases are likely to prolong cone-mediated vision even if there are changes to cone morphology and density.

Our sense of sight depends on the retina, a thin layer of cells at the back of each eye. Its job is to detect light using cells called photoreceptors, then send that information to the rest of the brain. The retina has two kinds of photoreceptors: rods (active in dim light) and cones (which detect colour and work in bright light). We rely heavily on cone cells for vision in our daily lives. Retinitis pigmentosa is a progressive eye disease affecting photoreceptors. In the early stages of this disease, rods gradually die off. Next, cone cells start to die, inevitably resulting in blindness. There is currently no cure, although some experimental treatments are being developed that aim to prevent rod death or replace missing rod cells. However, it is unclear if these therapies will be effective, because we do not fully understand how rod death affects cone cells ­ for example, whether or not it damages the cones irreversibly. Scalabrino et al. therefore set out to track how the signals that cones send to the brain changed over time during progression of the disease using genetically altered mice that reproduced the symptoms of retinitis pigmentosa. In these mice, rod cells die off over several months, followed by complete loss of cones a few months later. Initial microscopy experiments looking at the shape and appearance of the cone cells revealed that the cones started looking abnormal long before all the rods died. Next, to determine if these unhealthy cones had stopped working, Scalabrino et al. measured the activity of the mice's retinal ganglion cells (RGCs) in bright light ­ in other words, when cones are normally active. RGCs transmit signals from photoreceptors to the brain, like a 'telephone line' between our brains and eyes. Applying a technique called information theory ­ which was originally used to determine how efficiently signals travel down telephone lines ­ to these experiments revealed that the RGCs still sent high-quality visual information from the cones to the brain. This is was surprising because the cones appeared to be dying and were surrounded by dead rods. This study sheds new light on the biological processes underpinning a devastating eye disease. The results suggest that treatments to restore vision could work even if given after a patient's cones start looking unhealthy, giving hope for the development of new therapies.

Efficient coding, channel capacity, and the emergence of retinal mosaics.

Among the most striking features of retinal organization is the grouping of its output neurons, the retinal ganglion cells (RGCs), into a diversity of functional types. Each of these types exhibits a mosaic-like organization of receptive fields (RFs) that tiles the retina and visual space. Previous work has shown that many features of RGC organization, including the existence of ON and OFF cell types, the structure of spatial RFs, and their relative arrangement, can be predicted on the basis of efficient coding theory. This theory posits that the nervous system is organized to maximize information in its encoding of stimuli while minimizing metabolic costs. Here, we use efficient coding theory to present a comprehensive account of mosaic organization in the case of natural videos as the retinal channel capacity-the number of simulated RGCs available for encoding-is varied. We show that mosaic density increases with channel capacity up to a series of critical points at which, surprisingly, new cell types emerge. Each successive cell type focuses on increasingly high temporal frequencies and integrates signals over larger spatial areas. In addition, we show theoretically and in simulation that a transition from mosaic alignment to anti-alignment across pairs of cell types is observed with increasing output noise and decreasing input noise. Together, these results offer a unified perspective on the relationship between retinal mosaics, efficient coding, and channel capacity that can help to explain the stunning functional diversity of retinal cell types.

Scene statistics and noise determine the relative arrangement of receptive field mosaics.

Many sensory systems utilize parallel ON and OFF pathways that signal stimulus increments and decrements, respectively. These pathways consist of ensembles or grids of ON and OFF detectors spanning sensory space. Yet, encoding by opponent pathways raises a question: How should grids of ON and OFF detectors be arranged to optimally encode natural stimuli? We investigated this question using a model of the retina guided by efficient coding theory. Specifically, we optimized spatial receptive fields and contrast response functions to encode natural images given noise and constrained firing rates. We find that the optimal arrangement of ON and OFF receptive fields exhibits a transition between aligned and antialigned grids. The preferred phase depends on detector noise and the statistical structure of the natural stimuli. These results reveal that noise and stimulus statistics produce qualitative shifts in neural coding strategies and provide theoretical predictions for the configuration of opponent pathways in the nervous system.

High-resolution light-field microscopy with patterned illumination.

Light-field fluorescence microscopy can record large-scale population activity of neurons expressing genetically-encoded fluorescent indicators within volumes of tissue. Conventional light-field microscopy (LFM) suffers from poor lateral resolution when using wide-field illumination. Here, we demonstrate a structured-illumination light-field microscopy (SI-LFM) modality that enhances spatial resolution over the imaging volume. This modality increases resolution by illuminating sample volume with grating patterns that are invariant over the axial direction. The size of the SI-LFM point-spread-function (PSF) was approximately half the size of the conventional LFM PSF when imaging fluorescent beads. SI-LFM also resolved fine spatial features in lens tissue samples and fixed mouse retina samples. Finally, SI-LFM reported neural activity with approximately three times the signal-to-noise ratio of conventional LFM when imaging live zebrafish expressing a genetically encoded calcium sensor.

Properties of multivesicular release from mouse rod photoreceptors support transmission of single-photon responses.

Vision under starlight requires rod photoreceptors to transduce and transmit single-photon responses to the visual system. Small single-photon voltage changes must therefore cause detectable reductions in glutamate release. We found that rods achieve this by employing mechanisms that enhance release regularity and its sensitivity to small voltage changes. At the resting membrane potential in darkness, mouse rods exhibit coordinated and regularly timed multivesicular release events, each consisting of ~17 vesicles and occurring two to three times more regularly than predicted by Poisson statistics. Hyperpolarizing rods to mimic the voltage change produced by a single photon abruptly reduced the probability of multivesicular release nearly to zero with a rebound increase at stimulus offset. Simulations of these release dynamics indicate that this regularly timed, multivesicular release promotes transmission of single-photon responses to post-synaptic rod-bipolar cells. Furthermore, the mechanism is efficient, requiring lower overall release rates than uniquantal release governed by Poisson statistics.

Inter-mosaic coordination of retinal receptive fields.

The output of the retina is organized into many detector grids, called 'mosaics', that signal different features of visual scenes to the brain1-4. Each mosaic comprises a single type of retinal ganglion cell (RGC), whose receptive fields tile visual space. Many mosaics arise as pairs, signalling increments (ON) and decrements (OFF), respectively, of a particular visual feature5. Here we use a model of efficient coding6 to determine how such mosaic pairs should be arranged to optimize the encoding of natural scenes. We find that information is maximized when these mosaic pairs are anti-aligned, meaning that the distances between the receptive field centres across mosaics are greater than expected by chance. We tested this prediction across multiple receptive field mosaics acquired using large-scale measurements of the light responses of rat and primate RGCs. ON and OFF RGC pairs with similar feature selectivity had anti-aligned receptive field mosaics, consistent with this prediction. ON and OFF RGC types that encode distinct features have independent mosaics. These results extend efficient coding theory beyond individual cells to predict how populations of diverse types of RGC are spatially arranged.