A chromatic feature detector in the retina signals visual context changes

Abstract

The retina transforms patterns of light into visual feature representations supporting behaviour. These representations are distributed across various types of retinal ganglion cells (RGCs), whose spatial and temporal tuning properties have been extensively studied in many model organisms, including the mouse. However, it has been difficult to link the potentially nonlinear retinal transformations of natural visual inputs to specific ethological purposes. Here, we discover a novel selectivity to chromatic contrast in an RGC type that allows the detection of transitions of the horizon across a retinal region. We trained a convolutional neural network (CNN) model on large-scale functional recordings of RGC responses to natural mouse movies, and then used this model to search in silico for stimuli that maximally excite distinct types of RGCs. This procedure predicted centre colour-opponency in transient Suppressed-by-Contrast RGCs (tSbC), a cell type whose function is being debated. We confirmed experimentally that these cells indeed responded very selectively to Green-OFF, UV-ON contrasts, which we found to be characteristic of transitions from ground to sky in the visual scene, as might be elicited by head- or eye-movements across the horizon. Because tSbCs reliably detected these transitions, we suggest a role for this RGC type in providing contextual information (i.e. sky or ground) necessary for the selection of appropriate behavioural responses to other stimuli, such as looming objects. Our work showcases how a combination of experiments with natural stimuli and computational modelling allows discovering novel types of stimulus selectivity and identifying their potential ethological relevance.

Asymmetric distribution of color-opponent response types across mouse visual cortex supports superior color vision in the sky

“…Our results extend these previous studies by demonstrating that the asymmetry across visual cortex can be explained by the asymmetric distribution of response types with distinct color tuning in their RF center, and by linking them to a neuronal computation relevant for the upper visual field, namely the detection of aerial predators. At the level of the mouse retina, color-opponency is largely mediated by center-surround interactions (Joesch and Meister, 2016;Szatko et al, 2020;Khani and Gollisch, 2021) and only very few neurons exhibit color-opponency in their center (Höfling et al, 2022). In line with this, we found that the pronounced representation of color by the center component of V1 RFs was not solely inherited by color-opponency present in the RF center of reti-nal output neurons.…”

Modeling responses of macaque and human retinal ganglion cells to natural images using a convolutional neural network

“…• Modeling responses of additional cell types, such as small bistratified cells [16,21] and smooth monostratified cells [15,22]; • Testing whether a CNN-based model can be used to identify cell types [11] that are not easily classified by reverse correlation with white noise; • Testing if a CNN-based model can reproduce specific ethologically relevant nonlinear computations in the retina, as was observed in the salamander retina [6]. For example, future work could focus on whether such a model can predict motion sensitivity [23][24][25][26] and direction selectivity [27,28] of RGCs in the primate retina; • Modeling responses to continuous natural movies, potentially with varying contrast to engage gain control mechanisms [10]; • Testing to what extent model performance generalizes and how RGC receptive field properties vary across stimulus classes (i.e.…”

“…Furthermore, analysis of the CNN model parameters revealed similarities between the learned convolutional filters and the receptive fields of bipolar cells [6]. More recently, several studies used CNN-based models to capture contrast adaptation in white noise in the macaque and rat retina [10], identify the chromatic properties of mouse RGCs [11] and test how effectively a model trained on a single stimulus type can capture both white noise and natural image responses in the marmoset retina [12]. However, it remains unclear how effectively CNN-based approaches can capture the responses of RGCs in the macaque and human retinas to natural scenes, and the visual information conveyed by those responses.…”

Scalable gaussian process inference of neural responses to natural images

“…To demonstrate the advantages of GP processes to model neural data, we started by considering neural responses to a static image. In the future it would be interesting to extend our model to neural responses to dynamic stimuli, such as natural movies [25]. This would of course increase the dimensionality of the input, and thus the difficulty of inference.…”