Awake responses suggest inefficient dense coding in the mouse retina

Abstract

The structure and function of the vertebrate retina have been extensively studied across species with an isolated, ex vivo preparation. Retinal function in vivo, however, remains elusive, especially in awake animals. Here we performed single-unit extracellular recordings in the optic tract of head-fixed mice to compare the output of awake, anesthetized, and ex vivo retinas. While the visual response properties were overall similar, we found that awake retinal output had 1) faster kinetics with less variability in the response latencies across different cell types; and 2) higher firing activity, by ~20 Hz on average, for both baseline and visually evoked responses. Notably, unlike the other conditions, many awake ON cells did not increase firing in response to light increments due to high baseline activity near saturation. Instead, they encoded light intensity fluctuations primarily by decreasing firing upon light decrements. In either condition, the visual message remains the same: the more spikes, the higher light intensity. The awake response patterns, however, violate efficient coding principles, predicting that sensory systems should favor firing patterns minimizing energy consumption. Our findings suggest that the retina employs dense coding in vivo, rather than sparse efficient coding as suggested from previous ex vivo studies.

Topographic axonal projection at single-cell precision supports local retinotopy in the mouse superior colliculus

“…Most identified axonal patches had RFs within the stimulation screen (±22° in elevation and ±36.5° in azimuth from the mouse eye; Boissonnet et al, 2022). In accordance with ex vivo retinal physiology (Baden et al, 2016), the average RF size of the RGC axons was 6.6° ± 3.9°…”

“…To map the RF of the RGC axons, we computed the response-weighted average of the presented random checkerboard stimuli (frame rate, 4 Hz; rectangular fields 3.7° in width and 2.9° in height; “reverse-correlation”; Chichilnisky, 2001) and fitted a 2D Gaussian at the peak latency to characterize the spatial RF profile (Figure 2C,D). Most identified axonal patches had RFs within the stimulation screen (±22° in elevation and ±36.5° in azimuth from the mouse eye; Boissonnet et al, 2022). In accordance with ex vivo retinal physiology (Baden et al, 2016), the average RF size of the RGC axons was 6.6° ± 3.9° ( N = 443), estimated as the mean of the long- and short-axis diameters of the 2D Gaussian profile at 1 SD.…”

“…Visual stimuli were presented to the subject animals as described previously (Boissonnet et al, 2022). In short, a custom gamma-corrected digital light processing device was used to project images (1280-by-720 pixels; frame rate, 60 Hz) to a spherical screen (radius, 20 cm) placed ~20 cm to the contralateral side of an animal's eye, stimulating the visual field ±22° in elevation and ±36.5° in azimuth (Figure 2A and Supplemental Figures 2A and 3A).…”

Atypical cortical feedback underlies failure to process contextual information in the superior colliculus of Scn2a+/-autism model mice

“…We presented visual stimuli to the mice as described previously (Boissonnet et al, 2023). We used a custom gamma-corrected digital light processing device to project images onto a spherical screen placed approximately 20 cm from the left eye of the animal.…”

“…Pupil detection and eye-tracking were performed with a region-based convolutional neural network (R-CNN; He et al 2017). The network was trained by a set of 260 mouse eye video frames with manually segmented pupil outlines (Boissonnet et al, 2023). The detected pupil was parameterized by a 2-dimensional Gaussian curve fit to the pixel map.…”