© Mehau Kulyk/Photo Researchers, Inc.
A typical neuron's axons and dendrites, when loaded with dye and magnified, resemble long, untended tresses on an extremely bad hair day. They extend wildly, usually to one side, and then bend at weird angles as their ends split into branches and sub-branches.
This neuronal coiffure must appear even more chaotic before the nervous system has undergone the developmental equivalent of a crew cut crossed with a topiary trimming. From the late embryonic to early postnatal stage, this pruning process drastically thins out the branches in many axonal and dendritic arbors. Long neuronal offshoots that grew to inappropriate targets simply vanish.
Pruning occurs in probably all vertebrates and in many lower animals; neuroscientists have been aware of it for decades. Nevertheless, this phenomenon has garnered much less attention than that paid to other forces and events shaping the nervous system, such as axon guidance and apoptosis. The reason, researchers say, is that pruning is very hard to monitor and manipulate.
Several recent studies could spur wider interest. Some of these investigations originated in serendipitous findings about knockout animals; others relied on improved experimental tools. The studies uncovered mechanisms and molecules underlying pruning in various animals and in diverse parts of the nervous system.
Experts predict that sorting out how pruning works might eventually help in understanding epilepsy, neurodegenerative diseases, mental retardation, autism, and schizophrenia. So far, however, no definite links have been established between these disorders and pruning.
EVOLUTIONARY VESTIGE The nervous system's need for remodeling suggests that its early development is inefficient and wasteful. But researchers in the field scotch that idea, citing pruning's presumed advantages. They point out that whenever a neuron's electrochemical activity determines which offshoots are pruned, an organism's experiences are, in effect, fine-tuning its nervous system.
A different rationale appears to govern a process dubbed "stereotyped pruning" by Marc Tessier-Lavigne, a former Stanford University biologist who is now at Genentech in South San Francisco. During this developmental process, certain lengthy axonal branches are invariably sheared. "Everybody's best guess is that it's an evolutionary vestige," says Tessier-Lavigne. He characterizes these branches as "scaffolds that are convenient to have at early stages and that then can be jettisoned at later stages."
Dennis D. O'Leary, a molecular neurobiology professor at the Salk Institute for Biological Studies in La Jolla, Calif., elaborates: Suppose a cortical neuron's axon must reach the spinal cord. As the axon extends, it sprouts branches to other targets by splitting into two at various points. But axons often make targeting errors at those points, and few branches reach the spinal cord. O'Leary outlines a better developmental strategy. "You could just elongate a primary axon [to the spine] and let later mechanisms develop the branches along that length at their leisure. So you drastically reduce errors, and you make sure that all targets get sufficient numbers of axons," he says. These branches and even the primary axon are later pruned if their targets are inappropriate.
Pruning research has progressed slowly for various reasons, some technical and some attitudinal. Experiments cannot rely on simple model systems, says Tessier-Lavigne, because pruning events "occur at a time when the nervous system is already quite complex." Moreover, this phenomenon is difficult to detect or confirm. After shearing axonal and dendritic branches, neurons sometimes grow new ones nearby. And a neuron's death or its failure to grow particular branches might cause their absence; pruning is not involved then.
This research field also might lag because it focuses on a negative event. Capturing a vanishing neural connection can be trickier and less exciting than monitoring a budding one. To further explain a preference for the positive, Jeff W. Lichtman, a neurobiology professor at Washington University School of Medicine, argues that the popular notion of memory as a connection between two ideas leads people to assume that the brain has created an actual connection.
Lichtman challenges this assumption. "Michelangelo proved that you can make a beautiful David starting with a block of marble and chipping away all but the thing you want," he analogizes. "There's nothing less logical about a retractive way of generating a specific [neural] circuit than there is about actually forming" a new circuit.
MOLECULAR PLAYERS UNMASKED Work on ocular dominance columns in visual cortex, which garnered the Nobel Prize for David H. Hubel and Torsten N. Wiesel, led to one line of pruning studies that began in the 1970s. Papers on other types of axonal elimination popped up in the 1980s. O'Leary, one of the pioneers, examined the stereotyped pruning of pyramidal-cell axons coursing from Layer 5 of rat cortex to subcortical targets.
Courtesy of Hollis T. Cline
Researchers have implicated several neurotrophic factors and GTPases in visual-system remodeling. But the molecules underlying stereotyped pruning remained a mystery. Then, several years ago, Stanford biological sciences professor Susan K. McConnell stumbled across a tantalizing finding: Mice that do not make a functioning version of the transcription factor Otx1 also fail to prune their Layer 5 axons.1 In addition, such mice exhibit abnormal spinning behavior and epileptic seizures.
"It's hard to imagine that all of those excess connections that should have been pruned away wouldn't result in some kind of abnormal behavior," says McConnell. But she acknowledges no proof exists that the excess connections form functioning synapses or contribute directly to abnormal phenotypes.
Tessier-Lavigne has identified two other proteins that contribute to stereotyped pruning, plexin-A3 and neuropilin-2. These combine to form receptors for semaphorins, molecules that guide axons by repelling them. A few years ago, his lab discovered that a hippocampal pathway, the infrapyramidal bundle (IPB), is unusually long in plexin-A3 mutant and neuropilin-2 knockout mice. In a recent paper, Tessier-Lavigne, former graduate student Anil Bagri, former postdoc Hwai-Jong Cheng, and colleagues reported that the overlong IPB results from a failure to prune. The plexin-A3 mutant mice also do not shear an axonal tract extending from the hippocampus to the brain's septum.2 These mice have seizures.
Courtesy of Jeff W. Lichtman
"It's well documented that semaphorins activating these neuropilin/plexin receptor complexes can lead to growth-cone repulsion," notes Tessier-Lavigne. "It can also lead to collapse of growth cones. It's not too much of a stretch to imagine that what can cause collapse of a growth cone can also cause the retraction of an axon, with just some minor tinkering of the signal- transduction apparatus downstream of the receptor."
Neither observations in vivo nor experiments on explant cultures and dissociated neurons could determine decisively whether IPB and hippocampo-septal pruning occurs by axonal retraction or degradation. To answer that question, Cheng, now an assistant neuroscience professor at the University of California, Davis, is setting up hippocampal slice cultures labeled with green fluorescent protein. He hopes that their neurons will prune so predictably that he can catch them in the act.
DEGENERATION AND METAMORPHOSIS Liqun Luo, an associate biological sciences professor at Stanford, knows how pruning occurs, albeit not in mice hippocampi but in the g-neurons of the Drosophila pupa's mushroom body, a brain center for olfaction, learning, and memory. Working on the same floor as Tessier-Lavigne and McConnell (who playfully calls it "Pruning Central for the World"), Luo and graduate student Ryan J. Watts used time-course analyses to establish that g-neuron axons do not retract; they degenerate. Mutation experiments indicated that this kind of pruning requires ubiquitin-mediated proteolysis.3
Luo and Watts observed a morphological resemblance between g-neuron pruning and Wallerian degeneration (WD), the fragmentation of an axon after it is severed from a neuron's cell body. But he stresses, "We don't know enough about either process to know exactly how similar they are mechanistically."
Nevertheless, WD might provide insights into pruning in higher animals, argues Martin C. Raff, a retired University College London cell biologist who coauthored a recent review on axonal self-destruction.4 Wlds mutant mice, developed in the 1980s, exhibit much slower WD; if pruning in such mice were also delayed, then it likely would involve degeneration. Such studies, if they exist, have not been published. Raff admits that he "can't understand why this mouse that's been around for years hasn't been more widely used."
James W. Truman's University of Washington lab has long investigated pruning during Drosophila metamorphosis. Elimination of axonal and dendritic branches generally requires that the steroid hormone 20-hydroxyecdysone bind to a heterodimer consisting of the ecdysone receptor and Ultraspiracle, an ortholog of the vertebrate retinoid X receptor (RXR). In mammals, notes Truman, RXR dimerizes with so many other receptors that teasing out its role in axonal remodeling would be extremely difficult. But he insists that fly metamorphosis remains a valuable pruning model.
"I think that within the developing nervous system of more complex organisms, you also have the same types of themes going on that we see in the fly," Truman says. "In the fly, [these themes are] really exaggerated because they're in the context of this whole-organism metamorphosis, rather than a single-cell-type metamorphosis." One example of a metamorphosis-like event in higher animals, he asserts, occurs during visual-system development, when certain neurons break one set of connections and form a new set.
NEURONAL BOXING BOUTS Because of its accessibility, the neuromuscular junction (NMJ) is a popular model of axonal-arbor pruning. Early in development, each axon innervates many muscle fibers, and each fiber is innervated by many axons. Later, one axon can still contact numerous fibers, but each fiber is contacted by only one axon.
Courtesy of Jeff W. Lichtman
Pruning is accordingly extensive, as shown by a murine study presented at this year's Society for Neuroscience meeting. Washington University's Lichtman and his graduate student, J.D. Wylie, found that the number of neck-muscle fibers contacted by a single axon's branches plunged 10-fold during the first two postnatal weeks, from roughly 200 to 20.5
Given the complex biochemistry likely to underlie pruning, Lichtman opines, "It will be very unlikely you're going to find a molecule that explains it." Instead, his lab applies novel axon-labeling and time-lapse-imaging techniques to investigate NMJ physiology. Acquiring images every few seconds over six or seven hours, the lab is making a movie of NMJ pruning. The film already reveals an unsuspected combination of retraction and degeneration. As the axon's tip moves backwards, "little pieces of cytoplasm are shed," says Lichtman.
He recently published two papers elucidating the competition between NMJ axons that results in all but one being eliminated from a muscle fiber. The first study6 found that "if you have two neurons duking it out with each other on one muscle fiber, and the same two neurons are duking it out on another muscle fiber, the outcome will be identical and take about the same amount of time," recalls Lichtman. One axon always wins because of "the pattern or amount of action potentials coming out of that axon."
The second study,7 done in collaboration with Washington University colleague Joshua R. Sanes, explored this action-potential hypothesis further. If it were true, Lichtman continues, "then if you could essentially manipulate the activity pattern of one of the competitors so that it wasn't punching very hard anymore, presumably that should change the outcome. And indeed it did."
But a few hard-slugging axons do not, in fact, win all the muscle-fiber trophies. Each axon eventually seems to innervate a reasonably large number of fibers. The reason, Lichtman says, lies in his lab's finding that "when a neuron that had a lot of branches competed against a neuron that had a few branches, the neuron with a few branches always was competitively stronger." The less-branched neuron might prevail in such match-ups because it can devote more resources, such as neurotransmitter molecules, to each branch.
Another new NMJ study implicates a single molecule in pruning: reelin, a secreted protein that plays a well established role in embryonic neuronal migration. A year ago, Gabriella D'Arcangelo, an assistant pediatrics professor at Baylor College of Medicine in Houston and postdoc Carlo C. Quattrocchi, were surprised to observe that NMJ neurons also express reelin.8
The researchers next examined the adult reeler mutant mouse whose reelin gene is disrupted. At 43% of sampled synapses, they saw that two axons, not one, innervated a single spot on a muscle fiber. The prevalence of these abnormal synapses dropped by about 50% two days after reelin was injected into muscles. The abnormal synapses likely did not result from initial axonal overgrowth, D'Arcangelo explains, because such excessive sprouting typically leads to many axons per muscle fiber and to axons that grow past the fibers.
Working earlier with Flavio Keller in Rome, Quattrocchi detected reelin's serine protease activity in vitro. He and D'Arcangelo now are exploring a model based on this protease activity. D'Arcangelo speculates: "Let's say there are two axons going into the same synapse. One axon is more active than the other. The one that's more active will endocytose or inhibit or somehow remove reelin from underneath its surface. So that axon will be protected, and it will stay. The other axon that's less active is not able to neutralize this protease activity of reelin. It gets cut off because reelin will act to detach it."
THE CRUCIAL SUBPLATE Neuroscientists have proposed two general models to explain how pruning helps create the ocular dominance columns (ODCs), areas in Layer 4 of visual cortex that respond to one eye or the other. One model is dominant, says Carla J. Shatz, chair of Harvard Medical School's neurobiology department. It posits that before vision has begun (in humans at birth; in cats when their eyes open), spontaneous waves of brain activity create ODCs through pruning and other processes. During a later critical period, lack of input from an eye can lead to renewed pruning and changes in the columns.
Courtesy of Marc Tessier-Lavigne
The alternative model, Shatz adds, proposes that ODCs are not activity-driven but develop because of "a kind of lock-and-key mechanism" that wires axons from the thalamus' lateral geniculate nucleus (LGN) up to Layer 4. Little if any pruning is involved. A recent study from Shatz's lab, however, could require a revision of this theory. She and colleagues removed cats' subplate neurons. These short-lived cells relay thalamic activity into the cortical plate before thalamic axons arrive in Layer 4. After subplate removal, ODCs did not form, even though thalamic axons reached Layer 4.9
This finding presents a problem for the alternative model, Shatz contends, "because the argument [there] has been that it's, say, a right-eye set of molecules in part of Layer 4 that matches up to right-eye LGN neurons, and left eye matches up to left eye. But we didn't touch the LGN neurons, and we didn't touch Layer 4 neurons. We just got rid of the subplate. So what this means is that either the subplate is needed to provide activity in the circuit, or the subplate is where all the molecular information lies."
PRUNING AND PEOPLE Researchers cannot study neural pruning directly in humans. Two approaches, however, provide indirect evidence.
Peter R. Huttenlocher, a pediatrics and neurology professor at the University of Chicago, has long examined synaptic density in the human cerebral cortex. He found that density peaks from ages 1 to 2, declines until age 16, and then levels off. At maximum, a neuron in prefrontal cortex has about 80,000 synapses; the adult total is approximately half that figure.
"Considerable evidence" exists, says Huttenlocher, that the density drop results from a thinning-out of synapses formed by dendrites, not--in young people, at least--from neuronal death or cortical expansion. Synaptic loss, he adds, is linked far more to the disappearance of dendritic spines (short lollipop-shaped protuberances) than to the pruning of whole dendrites.
A brain imaging project at the US National Institute of Mental Health identifies a different timeline for the thinning of the brain's neuronal cell bodies and dendrites. Three thousand magnetic resonance imaging scans of 1,200 people, most under the age of 20, show that gray matter peaks at age 11 in girls and 12 in boys, and decreases during adolescence.
"We're not positive it's pruning, but that's certainly the obvious guess," says project director Jay N. Giedd. Autopsies of children are too rare to confirm this hypothesis, but he hopes that better imaging technologies will eventually provide a more detailed picture of the gray matter. If pruning is confirmed, Giedd wonders whether "it's the use-it-or-lose-it principle that's guiding it. If so, then that has a lot of implications for parents, teachers, or society. What can we do to move it in the right directions?"
Douglas Steinberg (firstname.lastname@example.org) is a freelance writer in New York City.
1. J.M. Weimann et al., "Cortical neurons require Otx1 for the refinement of exuberant axonal projections to subcortical targets," Neuron, 24:819-31, 1999.
2. A. Bagri et al., "Stereotyped pruning of long hippocampal axon branches triggered by retraction inducers of the semaphorin family," Cell, 113:285-99, May 2, 2003.
3. R.J. Watts et al., "Axon pruning during Drosophila metamorphosis: evidence for local degeneration and requirement of the ubiquitin-proteasome system," Neuron, 38:871-85, June 19, 2003.
4. M.C. Raff et al., "Axonal self-destruction and neurodegeneration," Science, 296:868-71, 2002.
5. J.D. Wylie, J.W. Lichtman, "Massive branch loss in early postnatal life as motor units are pruned during synapse elimination," poster #457.15, Society for Neuroscience meeting, 2003.
6. N. Kasthuri, J.W. Lichtman, "The role of neuronal identity in synaptic competition," Nature, 424:426-30, July 24, 2003.
7. M. Buffelli et al., "Genetic evidence that relative synaptic efficacy biases the outcome of synaptic competition," Nature, 424:430-4, July 24, 2003.
8. C.C. Quattrocchi et al., "Reelin promotes peripheral synapse elimination and maturation," Science, 301:649-53, August 1, 2003.
9. P.O. Kanold et al., "Role of subplate neurons in functional maturation of visual cortical columns," Science, 301:521-5, July 25, 2003.