Topographic Maps: Sperry's Eye and Eph Gradients

Turn an adult newt’s eye upside down in its socket, cut the optic nerve, and let it regrow. When the newt can see again and you hold a lure above its head, it turns its head downward. It keeps making that mistake for as long as anyone watched.

The newt is not blind and not confused. Its vision has come back in perfect order, only rotated, as if the image had been flipped before it reached the brain. That tells you something about how the wiring was rebuilt. The regrowing axons did not connect at random and then get sorted out by use. Each one went back to the place it had come from, even though that place now gave the wrong answer. So the question for this lesson is the one Chris Deppmann asks his class: “How does a neuron know its position and know where to go?”

The idea: an axon finds its place in a map

The retina sends its axons to a target in the midbrain called the optic tectum in fish, amphibians and birds, and the superior colliculus in mammals. The projection is a topographic map: neighbouring ganglion cells in the retina connect to neighbouring places in the target, so the image on the retina is laid out in order on the brain. In the chick, axons from the temporal retina (the side nearer the temple) end in the anterior tectum, and axons from the nasal retina end in the posterior tectum. A second axis runs at right angles: dorsal retina maps to the lateral tectum, ventral retina to the medial tectum.

A growth cone (the moving tip of a growing axon) can sense only what is close to it, yet the map spans the whole target. His lecture sets out three ways a neuron could know where to stop. Its endpoint could be “determined by individual labels”, one matching tag per cell and per target site. It could be set “by gradients”, smooth slopes in the amount of a molecule that give each position a value. Or it could be set “by neighbors”, each axon placing itself relative to the axons around it. A useful picture is seats in a theatre. Every ticket could name a seat, every seat could be numbered by row, or people could simply sit in the order they arrived. The rest of the lesson shows that the retina uses the second and third, together.

Sperry’s classic experiment: eye rotation

Chris puts the experiment as a question: “What happens when you flip structure A around?” Roger Sperry flipped it in 1943, working with adult red-spotted newts, Triturus viridescens. Many textbooks draw this experiment with frogs or goldfish, but Sperry’s 1943 papers used adult newts, and they depended on the fact that an adult newt’s optic nerve regenerates. These were experiments on regrowth, which were then read as a window on how the map is built in the embryo.

In one paper he rotated the eye 180 degrees and left the optic nerve intact. The newts’ responses to moving objects were inverted and reversed, and they stayed that way. That alone proves little, since the old connections were never broken. In the companion paper he also cut the optic nerve and, in one group, cut through the nerve’s sheath as well and twisted the two stumps out of line, so that fibres could not simply follow old channels back. Vision returned after the nerve regrew. Animals with an unrotated eye aimed correctly. Animals with a rotated eye aimed in the wrong direction, by an amount that matched the rotation, and experience did not correct them.

The result argued against the view, as Sperry framed it, that fibres from the retina connect more or less at random and that useful function is then matched to whatever connections form. That was the line of thinking associated with his own doctoral adviser, Paul Weiss. If use did the matching, weeks of missed prey should have taught the newt to aim correctly. Instead the regrown fibres rebuilt the old map. Sperry proposed that retinal cells acquire chemical differences according to their position and that their targets carry matching differences. He developed this into the chemoaffinity hypothesis, set out in full in 1963, and suggested that two crossing gradients, one per axis, could provide the coordinates without a separate label for every cell. He did not have a molecule. He shared the 1981 Nobel Prize in Physiology or Medicine, with David Hubel and Torsten Wiesel, for his separate work on the divided cerebral hemispheres.

Chemospecificity: asking an axon to choose

Chemoaffinity predicts that the tectum itself carries a position-dependent cue that retinal axons can read. Friedrich Bonhoeffer’s group in Germany built an assay to ask about it directly. In 1987 Jochen Walter, Bonhoeffer and colleagues made a striped carpet: membranes from anterior and posterior chick tectum laid down in alternating narrow lanes. A strip of retina was placed across the lanes so that each growing axon met border after border.

Axons from temporal retina grew along the anterior lanes and stayed off the posterior ones. Axons from nasal retina grew on both. On a single kind of membrane, both nasal and temporal axons grew, so posterior membrane was not simply a poor surface. A companion paper found that heating the posterior membranes abolished the temporal axons’ avoidance. That pointed to a heat-sensitive repellent on posterior membranes rather than an attractant on anterior ones.

The stripe assay also gave biochemists a test to follow during purification. In 1995 two papers appeared back to back in Cell. Uwe Drescher, Bonhoeffer and colleagues purified RAGS, a 25 kilodalton protein anchored to the membrane by a lipid (a GPI anchor), found in the posterior tectum, that collapsed and repelled retinal growth cones in culture. Hwai-Jong Cheng, John Flanagan and colleagues reported that ELF-1, a ligand of the Eph receptor family, rose in a gradient from anterior to posterior across the chick tectum, while its receptor Mek4 ran the opposite way in the retina, high in temporal ganglion cells and low in nasal ones. Under the names agreed in 1997, RAGS is ephrin-A5, ELF-1 is ephrin-A2, and Mek4 is EphA3.

The molecular answer: opposing Eph and ephrin gradients

Eph receptors are the largest family of receptor tyrosine kinases (that lesson covers how such receptors are switched on). Their ligands, the ephrins, are themselves membrane proteins. Ephrin-As are held to the outer surface by a GPI anchor and bind mainly EphA receptors; ephrin-Bs cross the membrane and bind mainly EphB receptors. Because both partners sit on cell surfaces, signalling needs contact, and it can run both ways: the receptor signals forward into its own cell, and the ephrin can signal in reverse into the cell that carries it (Egea and Klein reviewed this bidirectional signalling in 2007; that lesson explains how).

Along the anterior-posterior axis the logic is a pair of opposite slopes. Temporal axons carry the most EphA, so a little ephrin-A repels them, and they stop in the anterior target. Nasal axons carry little EphA, tolerate more ephrin, and travel into the posterior target. The gradients fit the map; proving them necessary took a mouse.

In 2000 David Feldheim, Jonas Frisén, John Flanagan and colleagues removed the genes for ephrin-A2 and ephrin-A5. With either gene gone, the map in the superior colliculus was moderately abnormal; even mice with one working copy of each gene had a defect. With both gone, axons from one small patch of retina ended at several places scattered along the axis. Chris gives his class this map with a prompt: “Here’s a retinotectal map from an ephrin-A knockout mouse. What went wrong in the projection pattern? What does this tell you about how gradients encode position?” The careful answer is that the map loses much of its anterior-posterior order, but not all of it: termination zones are ectopic and multiple, the projection still reaches the colliculus, and some order remains. The defect grows as ligand is lowered, which says that the amount of ephrin, not merely its presence, carries position. The other axis uses the other class. EphB receptors, high in ventral retina, and ephrin-B ligands, high in the medial target, help sort axons along it. (Ephrin-B also acts at the optic chiasm, in that lesson.)

Big Question 2: How do local rules create global order?

A repulsive gradient read against a fixed threshold has a problem. The same receptor level could be high in one animal and low in another, and every axon is repelled, so something must spread them out. Arthur Brown, Greg Lemke and colleagues tested how the gradient is read in 2000. They put the coding sequence of EphA3, which mouse ganglion cells do not normally make, under the control of the gene Isl2, active in about half of the ganglion cells and scattered across the retina. At every retinal position, some cells now carried extra EphA and their neighbours did not.

Design problem

In this knock-in mouse, a random half of the retinal ganglion cells carry extra EphA3 on top of their normal gradient. If each axon reads an absolute level of EphA against the ephrin gradient, what map do you predict? What if each axon compares itself with its neighbours?

In 2004 Michael Reber, Patrick Burrola and Lemke combined this allele with a knockout of EphA4 to vary receptor levels further and fitted the results with a relative signalling model: each axon’s ending depends on its EphA signal relative to that of its neighbours, so axons compete for space and the map is the outcome. This is the “by neighbors” mode working inside a gradient. Local rules, each growth cone sensing only a short distance, produce global order because every axon is also sensing the others.

Gradients stop being enough at the fine scale. Early on, a retinal axon’s branches in the colliculus spread over a broad zone around the right place. Correlated spontaneous activity in the retina then sharpens them. In mice lacking the β2 subunit of the nicotinic acetylcholine receptor, whose retinal waves are abnormal, axons reach roughly the right region but their termination zones stay large and diffuse, as McLaughlin and colleagues found in 2003 and Cang and Feldheim reviewed in 2013. Chemistry sets the map, and activity refines it inside that frame, which is where the next unit begins.