The long and often branched processes of neurons, axons and dendrites, are specialized compartments used to collect and propagate information, respectively. Information is passed on between neurons through synaptic connections. During development, neurites can grow long distances towards their respective target areas, and several mechanisms ensure that growing neurites find their proper synaptic targets with high precision. However, phases of neurite outgrowth are often also followed by phases of neurite removal during development. Such regressive mechanisms are often collectively referred to as “pruning” (from a gardening expression that means “to weed out branches”) (Schuldiner and Yaron, 2015). For example, after a phase of rapid synapse formation during the first year after birth, the human brain experiences a steady decrease in synapse density during later development until a plateau is reached at the onset of adulthood (Huttenlocher, 1990). But it is not only small synapses that can be removed through pruning, but also long stretches of axon or dendrite. In this review, we discuss four examples of neurite pruning across model organisms in order to explain the general features and cell biological mechanisms of pruning. These examples are: (1) motoneuron axon pruning at the mammalian neuromuscular junction, (2) pruning of retinal ganglion cell synapses in the higher order visual system, (3) pruning of corticospinal tract axons of layer V cortical neurons, and (4) pruning of class IV dendritic arborization (c4da) neurons in the fruit fly Drosophila.
Pruning : circuit specification and more
Why do neurons build synapses or neurites just to degrade them again later on during development? For one, neurons (many billions in the mammalian brain) must make many highly specific synaptic connections (often thousands for a mammalian neuron in the brain). Accordingly, many specificity pathways are known that guide axons to their correct targets and promote synaptic connections only between specific pre- and postsynaptic partners. These systems usually consist of transmembrane receptors and their respective ligands on synaptic partners. The diversity of these receptors is relatively high, and is even increased by alternative splicing (some axon guidance receptors have hundreds of splice isoforms). Yet the question arises if the combinatorial possibilities of a few thousand transmembrane receptors can encode enough specificity for each of these many billions of synapses to be purely genetically encoded. In other words, errors in connectivity during the initial progressive phase of neuronal development might be unavoidable. Subsequent corrective measures such as pruning would be required to achieve the connection specificity needed for proper brain function. Alternatively, mammalian brain circuit development might actually involve a trial-and-error mechanism. In such a scenario, the best-suited out of a number of randomly-made synaptic connections would be selected by functional tests, and others would be removed. In both scenarios, pruning enhances the specificity of brain connectivity.
Another case where pruning is necessary are developmental transitions of the nervous system. A good example for this type of pruning are holometabolous insects, whose life cycles are divided into morphologically very distinct larval and adult phases. The nervous system of a fly larva is adapted to its lifestyle of crawling and eating, while that of the adult fly is adapted to flying and reproducing. Many parts of the larval nervous system are useless in adult life and are therefore removed during metamorphosis, the larval-to-adult transition. Here, pruning serves the removal of developmental intermediates.
A distinction that is often made is between small-scale and large-scale pruning. Small-scale pruning refers to the pruning of single synapses or small stretches of neurite, whereas large-scale pruning can remove long parts of primary axons or dendrites. Small-scale pruning involves retraction as a mechanism, whereas large-scale pruning uses neurite degeneration (i. e., severing and fragmentation) as a mechanism. Large-scale pruning tends to be stereotyped, i. e., the identity of the pruned branch, the timing, and the spatial aspects of the degeneration process are predetermined. Small-scale pruning is usually the result of activity-dependent synapse competition, such that the identity of the pruned synapse or branch is not clear from the beginning. Large-scale pruning events are usually involved in the removal of developmental intermediates such as during insect metamorphosis.
(1) Synapse elimination at the mammalian neuromuscular junction (NMJ)
At birth, skeletal NMJs in mammals are innervated by multiple motoneuron axons. During the first two postnatal weeks, motoneuron axon branches are systematically eliminated until each NMJ is only innervated by one motoneuron axon (Fig. 1A). Pruning axons initially thin out and retract from the NMJ, while the prevailing axon takes over the whole NMJ. Retreating axons often form a characteristic “retraction bulb” at the end, and they leave behind small membrane-bound particles termed axosomes (Bishop et al., 2004). Retraction stops at the next axon branch point.
Motoneuron synapses are eliminated by a competitive mechanism. For each NMJ, the presynaptic terminal that releases more neurotransmitter – and hence activates the muscle better – is stabilized, while the weaker presynaptic terminal is destabilized. This can be mimicked experimentally by activating or inhibiting single axons, which enhances or decreases their chances of prevailing at the NMJ. Stabilization and destabilization involve retrograde feedback from the muscle to the individual presynaptic axon terminals. It has been postulated that mature and immature isoforms of Brain-derived Neurotrophic Factor (BDNF) can act as such stabilizing and destabilizing feedback signals, respectively (Je et al., 2012). Weak axon branches are systematically destabilized. They contain fewer stabilizing cytoskeletal elements, and retraction bulbs contain lysosomes (membrane degrading organelles). Axosomes are often engulfed by myelinating Schwann cells. NMJ synapse elimination serves to specify muscle inputs and is an example of small-scale pruning.
(2) Eye-specific segregation of retinal ganglion cell axons
Retinal ganglion cells (RGCs) transfer visual information from the eye into the brain for information processing in the central nervous system (CNS). One of the first relay centers in this pathway is the dorsal Lateral Geniculate Nucleus (dLGN), a thalamic center in the diencephalon. Each dLGN hemisphere receives major input from the optic nerves and from both the ipsi- and the contralateral eye (Fig. 1B). It could be demonstrated through dye injection experiments that ipsi- and contralateral RGC projections in the dLGN are strictly segregated into patch-like structures. This segregation is important for the calculation of stereo vision in the brain. Segregation of inputs arises developmentally through pruning of wrongly targeted inputs. RGC segregation is probably driven by competition between RGC axons. If neuronal transmission is experimentally decreased in one eye, RGC inputs in the dLGN from this eye are preferentially pruned, and segregation defects occur (Schafer et al., 2012). Weak RGC synapses are marked by components of the complement cascade. Complement proteins are part of the innate immune system in the blood that bind to bacteria and mark them for phagocytosis. During development, they are also found in the CNS. Just as complement-marked bacteria in the blood are engulfed and digested by phagocytes, complement-marked RGC synapses are phagocytosed by microglia, the CNS immune cells. Apart from microglia, synapses can also be engulfed by astrocytes. This type of glial cell does not use complement to detect weak synapses, but the receptor MEGF10 that normally recognizes apoptotic (i. e., dying) cells (Chung et al., 2013). Lack of the above receptors leads to segregation and connectivity defects in the dLGN. RGC axon segregation is an example of small-scale pruning and serves to specify visual circuits.
(3) Selective elimination of layer V corticospinal tracts
The mammalian neocortex is organized into five layers of neurons. Pyramidal neurons in layer V (L5) signal cortical output to other regions in the CNS. Layer V neurons from different cortical areas project axons to different targets: L5 neurons in the visual cortex project to the Superior Colliculus (SC), a visual computation center, whereas L5 neurons in the motor cortex project axons to the spinal cord. In mice, L5 neuron axons start to grow around birth. All L5 neuron axons extend out of the cortex through a common pathway, and all L5 neuron axons initially extend to the spinal cord. Next, collateral axon branches start to grow out of the primary axon to other targets (Fig. 1C). Specificity is eventually achieved through pruning of non-specific projections, i. e., spinal cord projections of visual cortex L5 neurons and SC projections of motor cortex L5 neurons are removed. This specification occurs within the next week after birth when the different areas in the cortex acquire specific functions. The wrongly projected axonal branches (some centimetres in length) become fragmented and eventually disappear. Thus, L5 axons are pruned through a degenerative mechanism. Wrongly targeted L5 axons are probably repulsed by their targets: layer V neurons from visual cortex, but not from motor cortex, express the repulsive axon guidance receptors neuropilin-2 and PlexinA3/A4, whose repulsive ligand Semaphorin 3F is expressed in the spinal cord (Low et al., 2008). Activation of this receptor leads to repulsion from the spinal cord. Mutants for PlexinA3/A4 cannot prune their visual corticospinal projections. In this way, visual L5 axons in the spinal cord are pruned but motor L5 axons are not. L5 neuron axon pruning is large-scale and stereotyped, as the timing and the identity of the pruned branches can be precisely predicted.
(4) Dendrite pruning of Drosophila c4da neurons
Class IV dendritic arborization (c4da) neurons are nociceptive sensory neurons in the skin of larvae of Drosophila fruit flies. These neurons have long and branched sensory dendrites which cover the whole larval body wall. These dendrites are in close association with epidermal cells. During metamorphosis, larval epidermal cells are replaced by new adult skin cells, and the c4da neuron dendrites lose their substrate. They are therefore pruned at the onset of the pupal phase (Fig. 2A), while the cell body and axon of the neuron stay intact. Because of the ease of genetic manipulation and the accessibility of c4da neurons for microscopic analysis, c4da neuron dendrite pruning is one of the best-understood pruning models. C4da neuron dendrite pruning is induced by the pupariation hormone ecdysone. Ecdysone activates the ecdysone receptor EcR-B1, a transcription factor which activates a relatively small number of pruning genes (Fig. 2B). These include Sox14, a transcription factor which in turn activates the transcription of the MICAL gene (Kirilly et al., 2009). MICAL encodes an enzyme that can sever actin filaments, stabilizing cytoskeletal components. At the cell biological level, signs of dendrite pruning can be detected at three to five hours after the onset of the pupal phase (hours after puparium formation, h APF). Firstly, the cytoskeleton (in particular, microtubules) are disassembled in the proximal dendrites (Lee et al., 2009) (Fig. 2B). Microtubule disassembly requires the activity of a microtubule severing enzyme known as Katanin p60-like1. Secondly, endocytic vesicles are observed in proximal dendrites (Kanamori et al., 2015). These vesicles retrieve membrane material from the plasma membrane. The combination of microtubule disassembly and membrane retrieval leads to thinning and destabilization of proximal dendrites. Proximal thinnings act as diffusion barriers between the distal dendrites and the soma. These distal parts display spatially confined pulses (“transients”) of Ca2+ influx. Their appearance is a good indicator that the dendrite will soon break off (Kanamori et al., 2015). At about 6h APF, dendrites start to break at the proximal thin sites. At this time, active caspases can be detected in the dendrites (Fig. 2A, B). These proteases are usually activated in cells undergoing regulated cell death (Williams et al., 2006). Caspase activation leads to dendrite fragmentation between six and ten hours APF. The dendritic remnants are then phagocytosed by the epidermal cells surrounding the dendrite (Han et al., 2014) (Fig. 2A, B). To this end, epidermal cells use the receptor Draper, a homolog of the glial synapse receptor MEGF10. By 16 hours APF, c4da neuron dendrite pruning is completed. During the next two days, the epidermis and the muscle cells surrounding c4da neurons are renewed, after which the c4da neurons extend new dendrites that are adapted to the adult skin. C4da neuron dendrites are developmental intermediates, and their pruning is an example of stereotypic large-scale pruning.
How is pruning initiated?
There are two principle ways to trigger pruning. In the case of synapse elimination at the NMJ and RGC axon pruning in the dLGN, pruning is induced if the synaptic connection does not hold up to a certain standard, or is weaker than its neighbors. At the NMJ, sufficient activity leads to positive feedback in the form of growth factors (also known as trophic support), while insufficient activity leads to withdrawal of support, or even to a “punishment” signal. How these support or punishment signals are applied locally is not known, nor how they induce retraction. In the developing dLGN, the complement component C1q is found on presynaptic terminals that are not closely appositioned onto postsynapses. In both cases, a functional test determines whether a synapse should be pruned or not.
In Drosophila c4da neurons, the trigger for dendrite pruning is the pupariation hormone ecdysone whose levels rise sharply at the onset of metamorphosis. Ecdysone induces the expression of pruning genes. This set of genes induces an intrinsic dendrite pruning program, i. e., no information from other cell types is needed. One trigger for L5 neuron axon pruning are repulsive cues in the environment. Interestingly, the specificity here arises through the tissue-specific expression of repulsion receptors. The repulsive ligand Sema 3F is expressed in the spinal cord. The repulsive receptors are expressed in neurons from visual cortex, but not in neurons from the motor cortex. Therefore, only axons of L5 neurons from visual cortex can be repulsed (and pruned), the axons of motor cortex neurons are insensitive to the signal. Thus, the cause for pruning is a tissue-specific difference in gene expression. Whether activity also plays a role in L5 axon pruning has not been determined. Taken together, stereotypic large-scale pruning is often induced through changes in gene expression.
Local effects on cytoskeleton and plasma membrane
Both pruning by retraction and pruning through degeneration must involve alterations of the cytoskeleton and plasma membrane of the affected neurite. During motoneuron synapse elimination, retracting axons have fewer microtubules and disorganized neurofilaments compared to stable branches (Bishop et al., 2004). It is unknown how these effects are induced by negative feedback from the muscle.
In c4da neurons, microtubules are selectively destroyed in the proximal regions where the dendrite will eventually break. The plasma membrane of c4da neurons is also strongly altered at the onset of metamorphosis. In particular, a local increase in endocytosis in proximal dendrites thins dendrites locally and generates presumptive break sites. How microtubule removal and endocytosis are induced by ecdysone is not yet known.
Membrane and cytoskeletal destabilization are sure to play a role during L5 axon pruning. It will be interesting to see whether single RGC synapses need to be destabilized before microglial phagocytosis.
Fragmentation by caspases and phagocytosis
During stereotypic large scale pruning, neurites are usually fragmented before their phagocytosis. Fragmentation of neurites also occurs during neuronal cell death, indicating mechanistic similarities. Indeed, fragmentation depends on caspases in c4da neurons, and caspases have been implicated in other pruning models as well (Schuldiner and Yaron, 2015). It is interesting to speculate that also motoneuron axosome shedding and fragmentation of L5 axons might involve caspases.
Phagocytosis is a very important mechanism during pruning. It is required for removal of dendritic debris during c4da neuron dendrite pruning (Han et al., 2014), and also for RGC synapse pruning in the dLGN – apparently, these synapses cannot be pruned through retraction. Inhibition of phagocytosis in the dLGN results in hyperinnervation (Schafer et al., 2012). Thus, phagocytosis actively promotes synapse disassembly. Potential roles for phagocytosis during motoneuron axon pruning or L5 axon pruning have not been assessed. Based on the observation that L5 axons prune by fragmentation, at least in this model a phagocytic mechanism is likely.
Several pruning mechanisms have now been described and much can be inferred from comparisons as seen above. One important open question is that of spatial regulation – how are severing sites determined, and what determines where pruning stops? The mechanisms underlying synaptic competition are also not yet well understood. The elucidation of the signaling pathways downstream of pruning triggers (i. e., how do hormones or punishment signals signal to the cytoskeleton?) will probably give new insights into these questions.
Chung, W.S., Clarke, L.E., Wang, G.X., Stafford, B.K., Sher, A., Chakraborty, C., Joung, J., Foo, L.C., Thompson, A., Chen, C., Smith, S.J., Barres, B.A. (2013). Astrocytes mediate synapse elimination through MEGF10 and MERTK pathways. Nature 504, 394–400. CrossrefPubMedWeb of ScienceGoogle Scholar
Han, C., Song, Y., Xiao, H., Wang, D., Franc, N.C., Jan, L.Y., Jan, Y.N. (2014). Epidermal cells are the primary phagocytes in the fragmentation and clearance of degenerating dendrites in Drosophila. Neuron 81, 544–560. CrossrefWeb of SciencePubMedGoogle Scholar
Je, H.S., Yang, F., Ji, Y., Nagappan, G., Hempstead, B.L., Lu, B. (2012). Role of pro-brain-derived neurotrophic factor (proBDNF) to mature BDNF conversion in activity-dependent competition at developing neuromuscular synapses.Proc. Natl. Acad. Sci. USA 109, 15924–15929. CrossrefGoogle Scholar
Kanamori, T., Yoshino, J., Yasunaga, K., Dairyo, Y., Emoto, K. (2015). Local endocytosis triggers dendritic thinning and pruning in Drosophila sensory neurons. Nat. Comm. 12, 6515. Google Scholar
Kirilly, D., Gu, Y., Huang, Y., Wu, Z., Bashirullah, A., Low, B.C., Kolodkin, A.L., Wang, H., Yu, F. (2009). A genetic pathway composed of Sox14 and Mical governs severing of dendrites during pruning. Nat. Neurosci. 12, 1497–1505. PubMedCrossrefWeb of ScienceGoogle Scholar
Lee, H.H., Jan, L.Y., Jan, Y.N. (2009). Drosophila IKK-related kinase Ik2 and Katanin p60-like 1 regulate dendrite pruning of sensory neuron during metamorphosis. Proc. Natl. Acad. Sci. USA. 106, 6363–6368. CrossrefWeb of ScienceGoogle Scholar
Low, L.K., Liu, X.B., Faulkner, R.L., Coble, J., Cheng, H.J. (2008). Plexin signaling selectively regulates the stereotyped pruning of corticospinal axons from visual cortex. Proc. Natl. Acad. Sci. USA 105, 8136–8141. Web of ScienceCrossrefGoogle Scholar
Schafer, D.P., Lehrman, E.K., Kautzman, A.G., Koyama, R., Mardinly, A.R., Yamasaki, R., Ransohoff, R.M., Greenberg, M.E., Barres, B.A., Stevens, B. (2012). Microglia sculpt postnatal neural circuits in an activity- and complement-dependent manner. Neuron 24, 691–705. Web of ScienceGoogle Scholar
Williams, D.W., Kondo, S., Krzyzaowska, A., Hiromi, Y., Truman, J.W. (2006). Local caspase activity directs engulfment of dendrites during pruning. Nat. Neurosci. 9, 1234–1236. PubMedCrossrefGoogle Scholar
About the article
Dr. Sebastian Rumpf (*1975) studied Biology in Heidelberg, PhD at the Max Planck Institute of Biochemistry (2006). Postdoctoral work at the University of California San Francisco in the laboratory of Prof. Yuh Nung Jan, research group leader at the University of Münster, Germany since 2013. Dr. Rumpf focuses on pruning mechanisms in Drosophila.
Sandra Rode (*1987) studied Biology in Münster. Master thesis in the lab of Prof. C. Klämbt on functional aspects of glial cells. Since 2013 PhD student at the University of Münster where she studies the mechanism of dendrite pruning.
SvendeHerzmann (*1988) studied Biology in Münster. Master thesis in the lab of Prof. C. Klämbt on development of glial cells. Since 2013 PhD student at the University of Münster where she studies the mechanism of dendrite pruning.
Rafael Krumkamp (*1987) studied Biology in Münster. Master thesis in the lab of S. Rumpf where he is also a PhD student since 2015 and studies the mechanism of dendrite pruning.
Published Online: 2017-02-10
Published in Print: 2017-02-01