Strong and specific connections between retinal axon mosaics and midbrain neurons revealed by large scale paired recordings

Abstract

The superior colliculus (SC) is a midbrain structure that plays important roles in visually guided behaviors. Neurons in the SC receive afferent inputs from retinal ganglion cells (RGC), the output cells of the retina, but how SC neurons integrate RGC activity in vivo is unknown. SC neurons might be driven by strong but sparse retinal inputs, thereby reliably transmitting specific retinal functional channels. Alternatively, SC neurons could sum numerous but weak inputs, thereby extracting new features by combining a diversity of retinal signals. Here, we discovered that high-density electrodes simultaneously capture the activity and the location of large populations of retinal axons and their postsynaptic SC target neurons, permitting us to investigate the retinocollicular circuit on a structural and functional level in vivo. We show that RGC axons in the mouse are organized in mosaics that provide a single cell precise representation of the retina as input to SC. This isomorphic mapping between retina and SC builds the scaffold for highly specific wiring in the retinocollicular circuit which we show is characterized by strong connections and limited functional convergence, established in log-normally distributed connection strength. Because our novel method of large-scale paired recordings is broadly applicable for investigating functional connectivity across brain regions, we were also able to identify retinal inputs to the avian optic tectum of the zebra finch. We found common wiring rules in mammals and birds that provide a precise and reliable representation of the visual world encoded in RGCs to neurons in retinorecipient areas.

Awake responses suggest inefficient dense coding in the mouse retina

“…Our approach cannot obtain some RGC types that do not send their axons to OT, such as the one projecting to the suprachiasmatic nucleus via the retinohypothalamic tract (Li and Schmidt, 2018). Nevertheless, the fact that we obtained diverse types of visual responses, better than previous in vivo studies (Hong et al, 2018; Liang et al, 2018; Schröder et al, 2020; Sibille et al, 2021) if not all the ∼30 RGC types observed ex vivo (Baden et al, 2016; Jouty et al, 2018), validates our recording methods to monitor retinal outputs in vivo .…”

“…We next asked if RGC responses differ between in vivo and ex vivo conditions. This is a critical comparison because retinal physiology has been mostly studied and best characterized in an isolated preparation (Gollisch and Meister, 2010;Sanes and Masland, 2015), but little is known about it in awake animals (Weyand, 2007;Liang et al, 2018Liang et al, , 2020Schröder et al, 2020;Sibille et al, 2021). Here we exploited stimulus ensemble statistical techniques ("reverse correlation"; Meister et al, 1994;Chichilnisky, 2001) to systematically characterize the visual response properties and make a direct comparison across different recording conditions in the linear-nonlinear cascade model framework (see Methods for details).…”

“…Specifically, the electrodes were placed in the optic tract (OT), a bundle of nerve fibers composed of RGC axons as they project from the eye to their main targets: dLGN and SC. These recordings are superior to direct in vivo retinal recordings (e.g., with epiretinally implanted mesh electrodes; Hong et al, 2018) because the retinal circuits and the eye optics remain intact, and also to those extracellular recordings in SC (Sibille et al, 2021) or dLGN (Weyand, 2007) because RGC axonal signals do not need to be disambiguated from those of local axons or somata. This advantage also exists for calcium imaging recordings of the retina directly in the eye of immobilized zebrafish larvae (Esposti et al, 2013) or those of RGC axon terminals in dLGN (Liang et al, 2018, 2020) or SC (Schröder et al, 2020).…”

“…Optical methods were also developed to image the activity of retinal neurons directly through the pupil of a live animal (Geng et al, 2012; Yin et al, 2013, 2014). In contrast, thus far only a handful of studies have reported awake recordings from the retina (Esposti et al, 2013; Hong et al, 2018) or its outputs (Weyand, 2007; Liang et al, 2018, 2020; Schröder et al, 2020; Sibille et al, 2021). Thus, despite a long history of research on the retina, it still remains unclear what exactly the eye tells the brain in awake animals.…”

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

“…A fundamental question is then if RGC axons can nevertheless find precise targets even after they get “lost” in the long-range projection (Figure 1A), or if target neurons need to reconstruct the topography by elaborating local circuitry (Figure 1B). A recent study using Neuropixel probes has demonstrated a more precise one-dimensional mapping of RGC projections to the mouse SC surface (Sibille et al, 2021); however, local two-dimensional (2D) organizations of RGC axons remain elusive in any retinorecipient area because such long-range projections of dense axonal fibers preclude a precise anatomical characterization of the projection patterns at a single-cell resolution (Hong et al, 2011).…”