Semaphorins and Repulsion

A protein from chick brain makes the spread tips of growing sensory axons fold up and stop at a concentration of about ten picomolar. Why would a developing brain make something that stops growth?

Because a map needs walls as well as roads. An axon on its way to a target passes many places where it must not go: the wrong side of the spinal cord, the wrong layer, the wrong nerve. Chris Deppmann titles this part of his course “Repulsive Cues: Don’t Go There.” This lesson follows the semaphorins, the largest and best-studied family of such cues, from a collapsing growth cone in a dish to a receptor with a missing wire, and then to a result that complicates the word “repellent”.

Repulsive cues: don’t go there

You met the growth cone as both sense organ and motor: a fan of actin-filled filopodia and lamellipodia at the axon’s tip. It turns toward the side where actin builds and away from the side where actin falls apart. The guidance cues lesson sorted cues into four kinds: contact attraction, contact repulsion, long-range attraction and long-range repulsion. Semaphorins supply both kinds of repulsion. Some are secreted and act at a distance; others sit in membranes and act on contact.

Why is a repellent as useful as an attractant? Think of a cattle drive. Riders can lead the herd from the front, or they can ride the flanks and turn back any animal that strays. Flanking is often the cheaper way to keep a long column on course. An axon that is pushed away from every wrong place ends up in the right one, even if nothing pulls it there.

A repellent needs an assay. In 1990 Jonathan Raper and Josef Kapfhammer, at the University of Pennsylvania, grew chick dorsal root ganglion neurons on laminin and added a suspension of embryonic brain membranes. Within minutes the growth cones collapsed, folding from a spread fan into a thin tip. The effect reversed when the membranes were gone. Membranes from liver or fibroblasts did nothing. The activity was destroyed by heat and by trypsin, so it was a protein. Counting the fraction of collapsed growth cones gave a fast, graded readout that could follow the activity through purification.

Finding the repellent

In 1993 Yuling Luo, David Raible and Raper reported the protein: a glycoprotein of about 100 kilodaltons from chick brain that collapsed sensory growth cones at roughly 10 picomolar. Retinal growth cones did not respond, an early sign that a repellent acts only on axons equipped to hear it. They named it collapsin. Its first half resembled fasciclin IV, a protein that Alex Kolodkin, Corey Goodman and colleagues had cloned the year before in the grasshopper.

Fasciclin IV had its own story. In the grasshopper limb bud it marks bands of epithelial cells, and one band lies where the growth cones of the Ti1 pioneer neurons make a sharp turn. Embryos cultured in an antibody against fasciclin IV built that pathway abnormally. So one guidance protein was a membrane protein in an insect and its relative was a secreted repellent in a bird.

Later in 1993 Kolodkin, David Matthes and Goodman described related genes in flies, in humans and in two viruses, all sharing one large region now called the sema domain. They named the family the semaphorins, after the semaphore, a way of signalling with flags. Fasciclin IV became Semaphorin I, and collapsin is now Sema3A. In vertebrates the class 3 semaphorins are secreted, classes 4 to 6 span the membrane, and class 7 is held by a lipid anchor. Semaphorins also work far outside the nervous system, in immune cells, heart, blood vessels and bone.

The receptor with the missing wire

The next question was who receives Sema3A. In the August 1997 issue of Cell two papers answered it back to back. Zhigang He and Marc Tessier-Lavigne used expression cloning: they put pools of cDNA into cells and looked for cells that bound tagged Semaphorin III. Alex Kolodkin, David Ginty and colleagues reached the same protein by a different route, from binding studies. He and Tessier-Lavigne also showed that antibodies to neuropilin block Semaphorin III repulsion of sensory axons. The protein was neuropilin, first described in the frog nervous system by the laboratory of Hajime Fujisawa. Credit for the receptor belongs to both groups.

Neuropilin presented a puzzle. It bound Sema3A tightly, but its tail inside the cell was only about forty amino acids, with nothing that looked like an enzyme. A receptor that catches a signal must also pass it on. Where was the wire?

The answer came in 1999, again in two papers in one issue of Cell. Takahashi, Strittmatter and colleagues found that neuropilin-1 and a plexin form a complex that binds Sema3A more tightly than neuropilin alone. In engineered cells, neither protein alone let Sema3A shrink the cell; together they did. A neuropilin without its tail still supported collapse in other work, while a plexin stripped of its inner part blocked Sema3A responses in sensory neurons. Tamagnone, Comoglio and colleagues described the plexins as a large family: some bind membrane semaphorins directly, while the class 3 semaphorins reach plexins through neuropilin. Neuropilin catches; plexin signals.

What does plexin do inside? Its inner part contains a domain that acts as a GAP, a protein that speeds GTP hydrolysis and so switches a small GTPase off (see GTPase switches and Ras to MAPK). In the best-studied case, Plexin-B1, the target is the Ras-family GTPase R-Ras, which helps hold integrin adhesions to the substrate. Plexin signalling also engages Rho-family GTPases and proteins that act on actin and microtubules. The net result is local loss of actin and adhesion, the cytoskeletal response described in integrins and Rho GTPases. On one side of a growth cone that means a turn; everywhere at once it means collapse.

Keeping axons out of the ventral horn

Deppmann’s lectures make the point in one line: “Sema3A excludes some axons from the ventral horn.” Sensory neurons in a dorsal root ganglion are of several kinds. Small neurons that carry pain and temperature depend on NGF, carry its receptor TrkA, and end in the dorsal horn. Large neurons that carry muscle stretch depend on NT-3, carry TrkC, and run on into the ventral horn to reach motor neurons. Both enter the cord at the same place. How do the pain fibres know to stop?

The ventral cord was known to release something that inhibits sensory axon growth. In 1995 Elizabeth Messersmith, Kolodkin and colleagues, with explants of embryonic rat tissue, showed that this activity repels axons grown in NGF but has little effect on axons grown in NT-3. Cells secreting Semaphorin III did the same, and the protein is made in the ventral cord. In chick, Raper’s group found that antibodies against collapsin neutralized the ventral cord’s repellent, and their data suggest that muscle afferents lose their sensitivity as they grow into the ventral cord.

Design problem

You culture a dorsal root ganglion in a collagen gel next to a piece of ventral spinal cord, once in medium with NGF and once in medium with NT-3. Predict the outgrowth in each case, and say what it tells you about how the ventral horn is kept free of pain fibres.

That caveat matters. Mice lacking Sema3A have clearly disorganized peripheral nerves, and some sensory axons enter spinal cord regions they normally avoid. Other cues probably share the work in the living cord.

Attraction and repulsion can flip

Deppmann’s lectures warn: “Attraction and repulsion can flip.” In 1998 Hongjun Song, Mu-ming Poo and colleagues put single Xenopus spinal neurons in culture and set up a small Sema3A gradient with a micropipette. The growth cones turned away. When the researchers raised activity of the cyclic GMP pathway, the same growth cones turned toward Sema3A. The senses and cGMP lesson introduced cGMP as a second messenger made by guanylate cyclases. Here its level sets the sign of the answer. A similar switch with cAMP governs netrin, covered in netrin and DCC.

Does a living neuron use this? In the developing cortex, Sema3A is high near the pial surface. Franck Polleux and Anirvan Ghosh, with colleagues, had shown that it repels cortical axons downward. In 2000 Polleux, T. Morrow and Ghosh showed it also attracts the apical dendrite of the same pyramidal neurons upward. Soluble guanylate cyclase sits mainly in the developing apical dendrite and is needed for that attraction. One cue, one neuron, two answers, set by where the enzyme is.

Semaphorins do not retire when wiring ends. “Repulsive cues are essential in sculpting the CNS (same repulsive cues that participate in the initial assembly),” as Deppmann puts it, and he asks, “Are they important in maintaining the circuit in the adult?” The same family later prunes axon branches, covered in axon pruning, and limits dendritic spines, covered in CNS synapses.