Synapse Elimination: One Muscle Fiber, One Axon

A rat is born with most of its muscle fibers wired to two or more motor neurons. Two weeks later, each fiber listens to one.

Nothing is wrong with the extra inputs. They release acetylcholine, each evokes an endplate potential in the muscle fiber, and they work. Then they are removed. Chris Deppmann puts the general version of this on one line: “Functional Synapses are Eliminated.” This lesson asks how anyone counted the inputs, where the lost ones go, and what decides which axon keeps the fiber. The answer to the last question turns out to be activity, read locally, and the molecules behind it are still being worked out.

The idea: refinement removes working connections

Deppmann frames the wiring of a mature circuit as “Neural Circuit refinement,” a change in three quantities: the number of neurons, the number of synapses, and the strength of synapses. Earlier lessons covered the first. Neurons that fail to get enough target-derived factor die, as you saw in competition for targets. This lesson is about the second and third, on neurons that stay alive.

The best place to study it is the neuromuscular junction, the synapse between a motor axon and a skeletal muscle fiber. Its assembly, with the receptor patch that agrin and its partners build, is covered in the neuromuscular junction. Here what matters is that a typical adult mammalian muscle fiber has one junction and one axon. The postsynaptic site is called the endplate.

A useful measure is convergence, the number of axons that innervate one postsynaptic cell. Under “Refinement of Convergence,” convergence falls with age in the systems where it has been measured. At the neonatal rodent junction it falls from two or more to one. In their review of synapse elimination at the mammalian NMJ, Lichtman and Colman point out why muscle allows it: the fibers of one muscle are nearly interchangeable targets, so a motor neuron can first diverge onto many equally suitable fibers and then sort out which it keeps. The diffuse, redundant early wiring becomes a specific one.

Counting inputs and counting neurons

You cannot see an axon’s identity in a stain of a junction with several terminals tangled together. The first count was electrical. In 1970 P. A. Redfern recorded from fibers of newborn rat diaphragm with an electrode inside the cell. He increased the strength of the stimulus to the phrenic nerve in small steps. The endplate potential grew in discrete jumps, each with its own threshold and latency, because each jump was another axon reaching threshold. This method is called stepwise recruitment. Most responses at birth were complex, built from two to four steps. By 16 to 18 days they were single steps.

That count left an obvious question. Were the extra axons lost because their motor neurons died? Michael Brown, Jan Jansen and David Van Essen answered it in rat soleus muscle in 1976. Fibers lost their extra inputs by the end of the second week, and the force produced by each motor unit (one motor neuron and all the fibers it supplies) fell sharply as a share of the muscle’s twitch. Yet the number of motor neurons supplying the soleus did not change significantly. The only way both are true is that each surviving neuron withdrew some of its terminals. Synapse elimination is branch loss, the subject of axon pruning, not neuron death.

Fluorescent proteins let later work see the branches. In 2001 Cynthia Keller-Peck, Jeff Lichtman and colleagues used transgenic mice in which only a small, random subset of motor axons glowed, and traced every branch of single axons. Each axon was losing branches, and its branches withdrew at different times, at different junctions, without an obvious spatial pattern. If the neuron were running one program for its whole arbor, the branches would go together. They did not, which points to a separate contest at every junction.

Watching the contest

In 2003 Mark Walsh and Lichtman watched one. They used mice in which motor axons carried two colours of fluorescent protein, labelled the receptors with fluorescent alpha-bungarotoxin, and imaged the same junction in the sternomastoid muscle of an anaesthetised pup, let it recover, and found the junction again days later.

Typically, one axon withdrew from the receptor sites while the other extended into the sites it left, so the movie looks like a takeover. In rare cases the winner did not move into a vacated site, and the receptors there quickly disappeared. Two findings were surprising. The relative areas shifted in favour of one axon and then the other, so competition was not always a steady retreat. And the outcome could not be predicted from how much territory each axon held at the start.

Activity decides the winner

If size does not decide, what does? Experiments in neonatal rat muscle gave the first clue. Reducing activity, for example by blocking nerve impulses with tetrodotoxin, slowed elimination. Increasing muscle activity by stimulating it sped elimination up. So activity drives the process. But all of these experiments changed activity everywhere at once, and they could not say whether inactivity marks the loser or whether the loser is the input that is less active than its neighbour.

Brown, Jansen and Van Essen had already found a hint that the contest is local: in their experiments with a foreign nerve implanted into a muscle, endplates on one fiber that sat more than about 1 mm apart could both persist. Elimination seemed to act only between synapses close together.

Mu-ming Poo’s group then reduced the contest to three cells. In 1991 Yi-Jun Lo and Poo made Xenopus cultures in which one embryonic muscle cell was innervated by two spinal neurons. A brief burst of stimulation to one neuron immediately suppressed the synapse made by the other. When both neurons were stimulated together, the suppression was largely absent. This is heterosynaptic depression: activity at one synapse weakens an inactive neighbour on the same cell.

Design problem

You can apply alpha-bungarotoxin, which binds acetylcholine receptors and blocks them, to part of one mouse endplate or to the whole endplate, and then image the junction over weeks. Predict what happens to the receptors and to the overlying nerve terminal in each case, and say what each result would tell you about the rule.

”We know very little about the molecular mechanisms”

Deppmann’s own verdict on synapse elimination is blunt: “WE KNOW VERY LITTLE ABOUT THE MOLECULAR MECHANISMS THAT MEDIATE THIS.” It is fair. Several candidate signals have good evidence, mostly from culture.

Kinases in the muscle. Phillip Nelson’s laboratory at the NIH built compartmented cultures in which two groups of neurons innervated the same myotubes. Stimulating one group weakened the other’s input, and that heterosynaptic loss depended on protein kinase C (PKC), the kinase switched on by diacylglycerol and calcium that you met in the PLC lesson. An activity-dependent action of protein kinase A (PKA), the cAMP-driven kinase of cAMP and PKA, stabilized the stimulated input. Both worked through the number of acetylcholine receptors at the synapse. In rat muscle in vivo, other work has linked PKC activity to the normal pace of elimination.

A neurotrophin with two faces. BDNF is made as a pro-form, proBDNF, which can be cut to mature BDNF. As Trk and p75 explains, the pro-form signals through p75 and the mature form through TrkB. Bai Lu’s group, working at the NIH, found in Xenopus nerve-muscle cultures that muscle stimulation releases proBDNF, which depresses synapses and then causes the presynaptic terminal to retract, through p75 on the nerve. Cutting proBDNF to mature BDNF at the synapse turned the same molecule into a signal that potentiates. With two neurons on one muscle cell, proBDNF and p75 promoted retraction of the less active terminal, while mature BDNF and TrkB stabilized the active one. In developing mouse junctions, adding proBDNF promoted elimination, infusing mature BDNF delayed it, and blocking the cutting step sped it up through p75. The group called the two forms “punishment” and “reward” signals.

Deppmann’s summary ties these together: when the winning axon’s transmitter activates the muscle, the muscle generates two intracellular signals, one that punishes nearby inactive sites and one that protects the active site. In culture, the heterosynaptic depression is reported to need a rise of calcium in the postsynaptic cell. What remains unknown is how these signals are confined to one patch, and how they relate to the cytoskeletal steps inside a retreating branch.

The same rule in the brain

The neuromuscular junction is not unusual. In the rodent cerebellum, each Purkinje cell is innervated by several climbing fibers, the axons from the inferior olive, in the first postnatal week, and by about the end of the third week most have one. Recordings of stepwise climbing-fiber responses, the same counting method Redfern used, track the change, and that literature, from groups such as Masanobu Kano’s, is separate from the muscle work. The detailed mechanisms differ, and glia and microglia take part in the brain, as microglia and complement describes. The logic of starting with several working inputs and keeping the most effective one recurs across the nervous system.