Over the centuries, artists, poets, writers and scientists have all attempted to answer a key existential question: what makes us human? Neuroscience has provided us with one exciting possible answer: our brains. To decode the complexities of the brain, many large-scale efforts are aimed at unravelling the cellular, molecular and computational properties of this startlingly complex system. Yet, to date, many important insights towards these problems have come from a surprisingly humble part of the central nervous system – the retina. The retina resides outside the skull within the eye and is responsible for vision. It contains diverse neuron types that detect light and has proven to be a uniquely approachable system for discovering neurobiology principles owing to its inherent organization, wiring and experimental accessibility. In this article, we describe how the retina has been used to make key neuroscience discoveries, and in turn how these principles shed light on how the brain works.

Retina: the right circuit for the job

In 1887, Santiago Ramón y Cajal began to apply Camillo Golgi’s revolutionary Golgi staining techniques to the central nervous system (CNS). Through his work and others, we first learned that the nervous system comprises a dazzlingly diverse array of neuron types (Figure 1a). These observations formed the basis for the ‘neuron doctrine’, the principle that information flows in the brain between individual cells through connections we now know to be synapses. Together, these neural circuits underlie all cognition.

Organization of the retina.

Figure 1
Organization of the retina.

(a) Heterogeneity of retina neuron shapes as illustrated by Ramón y Cajal (Histologie du système nerveux de l’homme & des vertébrés’, 1909). Neurons vary in their shape, branching pattern and laminar targeting. Such drawings helped form the basis for the neuron doctrine. (b) The retina is an extension of the brain that resides in the eye. This neural tissue lines the back of the optic cup (inset) and sends visual information to the brain through retinal ganglion cell (RGC) axons. (c) Retina neuron types, organization and connectivity as viewed in cross section. The retina is laminated and comprises three cellular and two synaptic layers. The outer nuclear layer (ONL) contains rod (blue) and cone (red) photoreceptors that form synapses with horizontal cells (yellow) and bipolar cells (green) in the outer plexiform layer (OPL). The inner nuclear layer (INL) contains horizontal, bipolar and amacrine cells (grey), and the latter two of these neuron types form synapses with RGCs (orange) in the inner plexiform layer (IPL). RGC bodies reside in the ganglion cell layer (GCL). In addition, three glial cell populations are present in the retina (Müller glia, black; microglia, cyan; and astrocytes, blue). (d) Whole mount (en face) view of the retina showing mosaics of RGCs.

Figure 1
Organization of the retina.

(a) Heterogeneity of retina neuron shapes as illustrated by Ramón y Cajal (Histologie du système nerveux de l’homme & des vertébrés’, 1909). Neurons vary in their shape, branching pattern and laminar targeting. Such drawings helped form the basis for the neuron doctrine. (b) The retina is an extension of the brain that resides in the eye. This neural tissue lines the back of the optic cup (inset) and sends visual information to the brain through retinal ganglion cell (RGC) axons. (c) Retina neuron types, organization and connectivity as viewed in cross section. The retina is laminated and comprises three cellular and two synaptic layers. The outer nuclear layer (ONL) contains rod (blue) and cone (red) photoreceptors that form synapses with horizontal cells (yellow) and bipolar cells (green) in the outer plexiform layer (OPL). The inner nuclear layer (INL) contains horizontal, bipolar and amacrine cells (grey), and the latter two of these neuron types form synapses with RGCs (orange) in the inner plexiform layer (IPL). RGC bodies reside in the ganglion cell layer (GCL). In addition, three glial cell populations are present in the retina (Müller glia, black; microglia, cyan; and astrocytes, blue). (d) Whole mount (en face) view of the retina showing mosaics of RGCs.

Since the time of Ramón y Cajal, modern neuron labelling methods have deepened our understanding of the cellular complexity within the CNS. In parallel, studies in model organisms are providing insights into the neural coding of behaviour and the corresponding activity of single neurons and populations. The problem for scientists has been to connect these two rich data sets in order to link neuron identity to circuit outcomes. This has been challenging for several reasons.

First, the complexity of brain circuitry has made detailed analysis very difficult. In addition, the marked heterogeneity of neuron shapes can obscure even large alterations in specific subsets. These challenges are compounded by the fact that the contribution of individual neurons to circuit function is largely unknown in brain, neuron type diversity is only beginning to be mapped, and markers to label and manipulate many brain neuron subtypes are limited. Many of these hurdles have been overcome in the retina, the neural tissue responsible for vision that resides in the back of the eye (Figure 1b). The retina has proven to be an excellent system for linking neuron structure to function for several reasons (Box 1). First, retina neurons connect in precise synaptic layers termed lamina, in which synaptic partner choice can be readily assayed by looking for alterations in layer stratification (Figure 1c). Second, much is known about retinal circuit formation, and these cells and circuits exist in 3D patterned arrays called mosaics (Figure 1d). Third, retinal neurons themselves are outside the skull, and thus more readily accessible to study in vivo. Fourth, neuroscientists have developed a rich retina toolbox that consists of a large number of genetic model systems, antibodies, cell type–specific viruses, directed electrophysiology and gene addition or deletion methods for experimental manipulation (Box 2). Finally, the retina shares general neuron and synapse features with the brain, and the function of molecular pathways uncovered in retina often extends to brain neurons and circuits.

Box 1. The retina is an excellent system to study the nervous system
  • Accessible part of the central nervous system

  • Highly organized into distinct layers called lamina

  • Most cell types and synaptic partners are known and can be manipulated

  • Scientists have developed a large set of experimental tools

  • Shares key neuron and synapse features with the brain

Box 2. A rich toolbox for study of the retina
  • Numerous mouse lines for neuron-specific manipulation

  • Many neuron subtypes and synapse antibodies for visualizing cells and connections

  • Straightforward methods for gene deletion or overexpression

  • Numerous approaches to visualize and reconstruct single neurons of a given type

  • Methods to manipulate and record from identified single neurons or populations

Neural circuitry in the retina

There are five general neuron types in the retina that are arranged into three nuclear layers and two synaptic layers (Figures 1c and 2a). Photoreceptors reside in the outermost nuclear layer and consist of rods and cones that detect photons and convert them into an electrical signal. Rods respond to dim light and are responsible for night vision, while cones respond to brighter light and are responsible for daylight vision and colour detection. Photoreceptors relay information to interneurons (horizontal, bipolar and amacrine cells) that process light signals and connect with retinal ganglion cells (RGCs). RGCs integrate all information and send it to the brain through their long axons, which comprise the optic nerve. Retinal neurons can be further subdivided into ~100 distinct subtypes, with most diversity found in the amacrine and RGC populations. While this number of subtypes is comparable to that of other brain regions, markers for a majority of these cells have been well defined only in retina. Moreover, specific types of retinal neurons pattern and connect in defined nuclear and synaptic layers (Figure 2b). This precise organization occurs across species, and insights using retina have been garnered from a host of model organisms including mice, rabbits, cats, zebrafish, chicks, salamander, fruit flies and frogs. While some specific differences do occur, the largely conserved neural patterning of the retina makes this system particularly amenable to molecular and mechanistic studies. Below, we highlight some of the many key discoveries made in retina regarding the molecular determinants of neuron type, shape, organization and connectivity (Box 3).

Box 3. Key discoveries in neuroscience aided by the retina
  • Comprehensive mapping of neuron diversity in adulthood and development

  • Molecular determinates of neuron shape and organization

  • Relationships between neuron form and function

  • Principles that underlie neural connectivity

  • Molecular determinates of synaptic specificity

The retina is composed of a large number of neuron types.

Figure 2
The retina is composed of a large number of neuron types.

(a) Different neuron types can be readily visualized in the retina using cell-specific antibodies which reveal that specific neuron subtypes target distinct retina synapse lamina. (b) Retinal neurons have diverse morphologies and form synapses in distinct layers to which their neurites project. (c) Retinal ganglion cells are highly diverse, comprising many structurally and functionally distinct subtypes. (d) The strong relationship between neuronal structure and function is demonstrated here by the J-RGC subtype which is direction selective and has a distinctly asymmetric arbor. J-RGC electrophysiological responses are strongest to light moving in the direction in which their arbors point (orange arrow, inset).

Figure 2
The retina is composed of a large number of neuron types.

(a) Different neuron types can be readily visualized in the retina using cell-specific antibodies which reveal that specific neuron subtypes target distinct retina synapse lamina. (b) Retinal neurons have diverse morphologies and form synapses in distinct layers to which their neurites project. (c) Retinal ganglion cells are highly diverse, comprising many structurally and functionally distinct subtypes. (d) The strong relationship between neuronal structure and function is demonstrated here by the J-RGC subtype which is direction selective and has a distinctly asymmetric arbor. J-RGC electrophysiological responses are strongest to light moving in the direction in which their arbors point (orange arrow, inset).

Mapping neuron types

A neuron type is generally defined as a set of cells that have a unified morphology, function, organizational arrangement and molecular signature. The advent of single-cell sequencing has helped propel efforts to map neuron diversity. Based on these and related studies, the retina is the only part of the mammalian CNS in which we are approaching a full account of neuron types. In mouse, this ‘simple’ circuit consists of ~46 RGC types, ~63 kinds of amacrine cells, 15 bipolar types, three photoreceptor subsets, one horizontal cell type and three kinds of glia (Figures 1c and 2c). In retina, scientists have the unique opportunity to validate that cell types identified by sequencing represent true cell types in tissue due to the ability to readily assay each cell type for well-defined criteria (e.g. morphology, organization, molecular signatures and function). The substantial retina neuron diversity may be a lesson for similar studies in brain where neural subtypes are beginning to be mapped in diverse brain regions. Such cell type information is useful for understanding how the nervous system works because there is a clear relationship between a neuron’s type, its structure and its function. One particularly notable example of this relationship occurs in the ganglion subtype called the ‘J-RGC’. This cell has a distinctly asymmetric arbor (Figure 2d). Recording from J-RGCs revealed that they respond to light moving only in one direction, the direction in which its neural dendrites point (Figure 2d inset). Thus, from the retina, we have learned that neuron type, and in turn neuron shape, can directly influence neuron function.

Molecules that dictate neuron locations and shapes

How might J-RGCs and other neurons know how to send their neurites to the right location and array them in the correct shape? Given the large variety of neuron shapes, surprisingly few answers are known to this question. In mammals, most examples of these pathways have been worked out in retina. Here, perhaps the best understood cell type is the starburst amacrine. Starburst amacrines help encode direction-selective light responses, and as their name implies, the starburst cell looks like a star with equally spaced processes that radiate out from the nucleus (Figure 3a). Like other retina neurons, these neurites display three key organizational properties. The first is laminar specificity, where starburst neurites reside in only a subset of synapse layers in the retina. The second is mosaic distribution, where the cell bodies of these neurons are organized in a lattice-like array when viewed from the top down. The third is self-recognition, where a given neurite appears to repel neurites arising from the same cell, but overlap with ‘non-self’ neurites arising from otherwise identical starburst amacrines. For this neuron type (but as yet, few others), we know some of the molecules involved in each of these features. Starburst laminar specificity has been mapped in part to the transmembrane protein semaphorin 6A and its receptor plexin A2. Deletion of these molecules results in mislaminated processes and a corresponding defect in direction selectivity (Figure 3b). The transcriptional molecule Fezf1 also participates in this process, ensuring that starburst amacrines arrive at the right location during development. In contrast, self-recognition is controlled by protocadherins, a unique protein superfamily that creates great diversity through alternative splicing. This gives each starburst amacrine its own protocadherin ‘coat of arms’ with which to tell itself from its neighbours (Figure 3c). Finally, starburst cell mosaic patterning has been attributed to the transmembrane proteins MEGF10 and MEGF11. Deletion of Megf10 and Megf11 turns the starburst nuclear distribution pattern from a lattice into a random array of cells that no longer tile the retina appropriately (Figure 3d). Notably, with a few exceptions, we have yet to discover what these pathways may be in the other ~99 neuron types in retina and know almost nothing about similar properties in brain. Thus, even in retina, our knowledge of which molecules dictate neuron locations and shapes is far from complete.

Molecular regulators of starburst amacrine cell connectivity, neurite organization and mosaic formation.

Figure 3
Molecular regulators of starburst amacrine cell connectivity, neurite organization and mosaic formation.

(a) Starburst amacrine cells tile the retina and form a star-like shape for which they are named, where equally spaced processes radiate from the nucleus. (b) Starburst amacrines send their dendrites to two distinct sublamina in the IPL. Starburst amacrine synapse organization and function is disrupted with the loss of Semaphorin 6a and its receptor Plexin A2. (c) Starburst amacrines display self/non-self recognition, where neurites from a single starburst cell (self) are repelled by isoneuronal interactions, while neurites from different starburst cells (non-self) show extensive interaction. Loss of protocadherin (Pcdh) diversity results in altered self-recognition and neurite crossing. (d) Starburst amacrines are distributed in a mosaic pattern resulting in organized equal neuron spacing. MEGF10 and MEGF11 help drive starburst amacrine mosaic formation, and their loss results in random cell spacing.

Figure 3
Molecular regulators of starburst amacrine cell connectivity, neurite organization and mosaic formation.

(a) Starburst amacrine cells tile the retina and form a star-like shape for which they are named, where equally spaced processes radiate from the nucleus. (b) Starburst amacrines send their dendrites to two distinct sublamina in the IPL. Starburst amacrine synapse organization and function is disrupted with the loss of Semaphorin 6a and its receptor Plexin A2. (c) Starburst amacrines display self/non-self recognition, where neurites from a single starburst cell (self) are repelled by isoneuronal interactions, while neurites from different starburst cells (non-self) show extensive interaction. Loss of protocadherin (Pcdh) diversity results in altered self-recognition and neurite crossing. (d) Starburst amacrines are distributed in a mosaic pattern resulting in organized equal neuron spacing. MEGF10 and MEGF11 help drive starburst amacrine mosaic formation, and their loss results in random cell spacing.

Synapse connectivity

While neuron diversity is complex, the intricacies of the synapses that wire these cells together is even more breathtaking. The brain has over 100 trillion synapses, and these connections arise during development when neurons grow, extend neurites and form contacts with their partners. The processes by which neurons choose synapse partners and then maintain or eliminate these connections has garnered much attention. We have made some progress in understanding how neural activity influences whether a synapse is maintained or lost, and these data show that active synapses are preferentially preserved. What is much less clear is how different neuron types choose which neural partners to connect with in the first place. Synapse connectivity decisions are generally thought to be governed by cell surface molecules that selectively recognize and bind receptors on the postsynaptic cell. The fly and mouse retina have been useful for mapping these molecules, since synapses here are laminated so that the position of dendritic and axonal arbors determines the partners from which a neuron has to choose (Figure 4a, b). The connectivity outcomes from these interactions can be influenced by downstream repulsive and adhesive cues and by activity-dependent processes (Figure 4c, d). These mechanisms can also act regionally even within a given nerve terminal, leading to subcellular synapse specificity (Figure 4e). Synapse recognition molecules themselves can be grouped into protein subfamilies that include cadherin family proteins, leucine-rich repeat containing proteins and immunoglobulin family proteins. Each of these families contains several synapse specificity molecules. For instance, cadherin 8 and cadherin 9 control the laminar position and synapse function of two different subtypes of bipolar cells. Deleting either protein or mis-expressing them disrupts proper bipolar terminal positioning, and thus connectivity. In another example, the immunoglobulin superfamily recognition molecule sidekick 2 was found to regulate connectivity between a particular subset of amacrines and a small RGC type responsible for object motion sensing. These and other examples are beginning to shed light on the complex cellular and subcellular molecular code responsible for precisely connecting neurons together.

Principles governing synapse specificity.

Figure 4
Principles governing synapse specificity.

(a) Axons and dendrites of retinal neurons are arranged in precise layers, and neurites within the same lamina have the potential to form synapses. (b) Neuron terminals within a lamina utilize a molecular code to select synaptic partners from a subset of the available neurites. (c, d) Molecular cues can guide synapse formation and refinement through distinct downstream signalling pathways that include selective adhesion with one postsynaptic partner (c) or repulsion from one partner that in turn favours connections with another (d). (e) Once a synaptic partner is selected, neurons can also exhibit subcellular specificity, where synapses are formed on a distinct region of the target neuron’s arbor.

Figure 4
Principles governing synapse specificity.

(a) Axons and dendrites of retinal neurons are arranged in precise layers, and neurites within the same lamina have the potential to form synapses. (b) Neuron terminals within a lamina utilize a molecular code to select synaptic partners from a subset of the available neurites. (c, d) Molecular cues can guide synapse formation and refinement through distinct downstream signalling pathways that include selective adhesion with one postsynaptic partner (c) or repulsion from one partner that in turn favours connections with another (d). (e) Once a synaptic partner is selected, neurons can also exhibit subcellular specificity, where synapses are formed on a distinct region of the target neuron’s arbor.

Concluding comments

The retina provides a fertile ground for the discovery of processes, molecules and pathways that dictate neuron form and function. Yet, much remains to be learned even about this obliging and ‘simple’ system. In the coming years, continued use of inventive genetic, biochemical, imaging and enhanced single-cell omics approaches will enable continued discovery of not only how the CNS works, but also what enables it to generate such a vast array of neuron-type specific cellular and connectivity patterns. Particularly exciting is emerging work that aims to shed light on brain diseases using the retina. Perhaps one day, such efforts will enable a complete understanding of the neurobiology that occurs in this small, but crucial window into the brain.

Further reading

  • Masland, R.H. (2012) The neuronal organization of the retina. Neuron.76, 266–280. DOI: 10.1016/j.neuron.2012.10.002

  • Lefebvre, J.L., Sanes, J.R., and Kay, J.N. (2015) Development of dendritic form and function. Annu. Rev. Cell Dev. Biol. 31, 741–777. DOI: 10.1146/annurev-cellbio-100913-013020

  • Sanes, J.R., and Zipursky, S.L. (2020) Synaptic specificity, recognition molecules, and assembly of neural circuits. Cell.181, 536–556. DOI: 10.1017/j/cell.2020.04.008

  • Albrecht, N.E., Alevy, J., Jiang, D., et al., (2018) Rapid and integrative discovery of retina regulatory molecules. Cell Rep. 24, 2506–2519. DOI: 10.1016/j.celrep.2018.07.090

  • Sun, L.O., Jiang, Z., Rivlin-Etzion, M., et al. (2013). On and off retinal circuit assembly by divergent molecular mechanisms. Science342, 1241974. DOI: 10.1126/science.1241974

  • Lefebvre, J.L., Kostadinov, D., Chen, W.V. et al. (2012) Protocadherins mediate dendritic self-avoidance in the mammalian nervous system. Nature488, 517–521. DOI: 10.1038/nature11305

  • Kay, J.N., Chu, M.W., and Sanes, J.R. (2012) MEGF10 and MEGF11 mediate homotypic interactions required for mosaic spacing of retinal neurons. Nature483, 465–469. DOI: 10.1038/nature10877

  • Kim, I.J., Zhang, Y., Yamagata, M., Meister, M., Sanes, J.R. (2008) Molecular identification of a retinal cell type that responds to upward motion. Nature452, 478–82. DOI: 10.1038/nature06739

Authors information

graphic

Nicholas Albrecht is a PhD candidate in the laboratory of Dr. Melanie Samuel at Baylor College of Medicine. Nicholas' research is focused on developing new super resolution imaging techniques to study disease and age-related changes in neurons and synapses. Email: nalbrech@bcm.edu

graphic

Courtney Burger is a PhD candidate in the laboratory of Dr. Melanie Samuel at Baylor College of Medicine. Courtney’s research is aimed at understanding the molecular mechanisms that regulate the development and maturation of cells found in the outer retina. Email: caburger@bcm.edu

graphic

Melanie Samuel, PhD is a faculty member in the Department of Neuroscience and Huffington Center on Aging at Baylor College of Medicine. Dr Samuel’s interdisciplinary research group utilizes genetic and molecular approaches to identify novel pathways that generate and maintain neurons and synapses. Email: msamuel@bcm.edu

Published by Portland Press Limited under the Creative Commons Attribution License 4.0 (CC BY-NC-ND)