Branch management: mechanisms of axon branching in the developing vertebrate CNS

Abstract

The remarkable ability of a single axon to extend multiple branches and form terminal arbors allows vertebrate neurons to integrate information from divergent regions of the nervous system. Axons select appropriate pathways during development, but it is the branches that extend interstitially from the axon shaft and arborize at specific targets that are responsible for virtually all of the synaptic connectivity in the vertebrate CNS. How do axons form branches at specific target regions? Recent studies have identified molecular cues that activate intracellular signalling pathways in axons and mediate dynamic reorganization of the cytoskeleton to promote the formation of axon branches.

Introduction

The functions of the vertebrate brain in cognition, learning, memory, sensory perception and motor behaviour require complex neural circuitry. To form this circuitry, individual neurons must connect with multiple synaptic targets, and they do so through extensive branching of their axon and the formation of elaborate terminal arbors^1–6. Branches establish topographic maps in numerous systems, including the retinotectal⁷ and corticospinal systems⁸, in which regions of the retina and sensorimotor cortex are connected to their targets in the optic tectum and spinal cord, respectively. In addition, multiple branches from the same axon can connect widely divergent regions of the nervous system. For example, single descending cortical axons extend branches into the pons and spinal cord⁹, single axons from some regions of the thalamus can ramify widely in the somatosensory, motor and higher-order sensory cortices¹⁰, and single cortical neurons can send axon collateral branches to homotypic and heterotypic regions of the contralateral cortex¹¹. Cajal, after observing the collaterals of callosal axons, commented: “callosal fibers do not simply join structurally and functionally comparable areas in the two hemispheres. They play a broader role, establishing multiple, complex associations that allow activity in one sensory area to influence a number of areas in the contralateral cerebral hemisphere.”¹²

Studies of neural development over the past several decades have focused on mechanisms of axon guidance. Surprisingly, given its importance in establishing neural circuits, axon branching has received less attention. How do axon branches form during development? Branches originate as dynamic protrusions that extend and retract from specific locations on the axon. Some of these protrusions become stabilized into branches that arborize by continued re-branching at target sites, leading to synapse formation. Branching is evoked by local extracellular cues in the target region, which signal through receptors on the axonal membrane to activate intracellular signalling cascades that regulate cytoskeletal dynamics. Axon arbors that form within target regions are highly dynamic but eventually stabilize through competitive mechanisms that can involve neural activity.

In this Review, we examine axon branching in the vertebrate CNS. We present in vivo and in vitro findings that illustrate modes of axon branching and the role of extracellular cues in the development of branches and the shaping of terminal arbors. Moreover, we discuss the role of cytoskeletal dynamics at axon branch points and how intracellular signalling pathways regulate cytoskeletal reorganization. Last, we consider the role of activity in regulating axon branching and shaping the morphology of terminal arbors, and identify areas for future research.

Axon branching and arborization

Growth cones, the expanded motile tips of growing axons, respond to extracellular guidance cues to lead axons along appropriate pathways toward their targets¹³. However, axonal growth cones in the vertebrate CNS do not typically enter their target region. Instead, axons form connections with their target though growth cone-tipped collaterals that branch from the axon shaft and terminal arbors that re-branch from axon collaterals (FIG. 1). In certain circumstances, branches can arise by splitting of the terminal growth cone^4,6, such as at the mouse dorsal root entry zone, where the growth cones of dorsal root ganglion (DRG) axons split to form two daughter branches that ascend or descend and arborize in the spinal cord^14,15.

Figure 1. Stages of axon branching in developing CNS pathways.

In the mammalian corpus callosum and thalamocortical pathway (a) and in the chick retinotectal pathway (b), axons initially extend past their eventual terminal regions (1). After a delay, branches extend interstitially from the axon shaft (2) followed by elimination of the distal axon (3) and formation of terminal arbors in topographically correct target regions (4). Figure c depicts interstitial branching from an individual layer 5 sensorimotor cortical axon into the pons and spinal cord (adapted from ⁹). In the retinotectal system (b), opposing gradients of EPHA–ephrin repellents inhibit retinal axon branching anterior and posterior to the correct target, whereas positive BDNF cues along the tectum promote branching at appropriate topographic positions^{7, 42}. Callosal axons connect homotypic regions of the cortical hemispheres but can also branch to other cortical areas^11,12 (not shown). Thalamocortical axons form branches that arborize topographically in layer 4 of somatosensory cortex ²³ but other thalamocortical axons can branch into multiple cortical areas¹⁹ (not shown).

In the mammalian CNS, axon branches typically extend interstitially, at right angles from the axon shaft behind the terminal growth cone (FIG. 1). This delayed interstitial branching can occur days after axons have bypassed the target¹⁶. Cortical axons in rodents initially bypass the basilar pons⁹, but after a delay, they form filopodia, dynamic finger-like actin-rich membranous protrusions, that can develop into stable branches that arborize in the pons¹⁷. Developing corticospinal axons also bypass spinal targets and later form interstitial branches that arborize once they have entered topographically appropriate target sites¹⁸. Segments of the axons distal to the target are later eliminated^16,19. Callosal axons, which connect the two cerebral hemispheres, also undergo delayed interstitial branching²⁰ beneath their cortical targets, where callosal growth cones collapse and extend repeatedly without advancing forward²¹. Growth cone pausing and interstitial branching have also been observed in dissociated cortical neurons²² where branches develop from dynamic growth cone remnants that remain on the axon shaft in regions where growth cones paused. Developing thalamocortical axons in vivo form layer-specific lower and upper tiers of terminal arbors in the barrel field, which together form a spatial map of the facial vibrissae in the rodent somatosensory cortex ²³. In vivo imaging revealed that branches form de novo along the shaft of the primary thalamocortical axon²⁴ and these give rise to dynamic arbors that extend and retract processes over many days. Thalamocortical axons extend branches at right angles in the target region, showing that interstitial branching is the dominant form of axonal branching in vivo.

In the vertebrate retinotectal pathway, topographic projections maintain spatial information of the visual field from the retina to the tectum. In the avian²⁵ and rodent²⁶ retinotectal systems, retinal ganglion cell axons initially overshoot their termination zone in the tectum and later emit interstitial branches at correct tectal positions, which is followed by terminal arborization and regression of the distal axon^7,27 (FIG. 1). However, in frogs and fish, growth cones of retinal axons form arbors as soon as they reach the correct region of the tectum²⁸ and axons do not emit side branches on their way to their targets. Live cell imaging of arbor formation at the tips of retinal axons in the frog optic tectum showed that highly dynamic branches extend and retract repeatedly²⁹. Thus, in the vertebrate CNS, axon branches, rather than the primary axonal growth cone, extend into targets at appropriate locations and thereby establish topographic specificity. Terminal arbors that develop from these branches then undergo dynamic remodeling (FIG. 1).

Role of extracellular cues

Axon branching occurs far from the cell body at localized regions of the axon that are close to the axon’s targets, suggesting that target-derived cues regulate axon branching. Families of extrinsic axon guidance cues, growth factors and morphogens can regulate axon branching by determining the correct position of branches or by shaping terminal arbors^2,4.

Guidance cues

Netrins, such as netrin-1, are diffusible guidance cues that attract efferent cortical axons in vivo^30–33 and that evoke axon branching in vitro^34,35. Focal application of netrin-1 can induce localized filopodial protrusions de novo along the axon shaft^34,36 and increase branch length without increasing extension of the primary axon. Netrin-1 also shapes terminal arbors. In the frog optic tectum, injection of netrin-1 enhances development of terminal retinal arbors³⁷ by increasing the dynamic addition and retraction of branches, resulting in an increase in total branch number.

Ephrins are membrane-bound cues that guide axons by repulsion^{38, 39} and they have a role in specifying the locations of axon branches. In vitro, for example, ephrin-A5 and its receptor (EPHA5) can promote or repress branching of various cortical axons growing on membranes from specific cortical layers⁴⁰. Ephrin-As stimulate the branching of thalamocortical axons grown on appropriate layer 4 somatosensory cortical membranes⁴¹ but they repel axons from limbic regions of the thalamus that do not innervate the somatosensory cortex. Ephrin-As also evoke topographically specific branching of retinal axons in tectal membrane stripe assays²⁵. In vivo, the distribution of EPHAs and ephrin-As in both the retina and in the optic tectum determine the position of retinal axon branches and terminal arbors⁷ by forward and reverse signalling⁴². In this system, EPHAs act both as receptors activated by ephrin ligands (forward signalling) and as ligands that activate ephrins (reverse signalling). In a model of retinotectal mapping²⁷, opposing gradients of EPHAs and ephrin-As along the anterior–posterior axis of the tectum inhibit axon branching on the retinal axon segments that are anterior and posterior to the correct target region. A molecule with branch-promoting activity, such as brain-derived neurotrophic factor (BDNF), that was expressed uniformly along the tectum would allow branching in a trough of the repellent gradients⁴³ (FIG. 1). Thus, branching would occur at a unique tectal position for each population of axons based on their temporal or nasal origin in the retina because of their differential sensitivity to the EPHA–ephrin-A repellents. In the dorsal–ventral axis of the tectum, EPHB–ephrinB-1 signalling, perhaps in cooperation with additional cues, regulates topographic mapping of retinal axon branches⁴⁴. Thus, repression of axon branching is an important regulatory mechanism, since axons must often traverse non-target regions on the way to their targets where additional positive cues may evoke branching.

Semaphorins (protein abbreviation SEMA) are repellent axon guidance cues that function in assembly of neuronal circuits⁴⁵. For example, SEMA3A repels cortical axons⁴⁶ and, in vitro, inhibits cortical axon branching^34,47, decreasing cortical branch length without affecting the length of the primary axon³⁴. SEMA3D has no effect on the central branches of Rohon-Beard axons in vivo but does repel and induce branching of the peripheral branches of the same axons⁴⁸. SEMA3A, through its neuropilin–plexin receptors, regulates the pruning of hippocampal axons in vivo and in vitro by selective retraction of transient branches to septal targets⁴⁹. However, SEMA3A can also positively influence axon branching. Recently, SEMA3A was shown to promote branching by cerebellar basket cell axons onto Purkinje cells in the cerebellar cortex⁵⁰. In vivo, mice deficient in SEMA3A show layer-specific reductions in terminal branching by basket cell axons and, in vitro, SEMA3A increases the complexity of basket cell axon arbors. Localized signalling mechanisms in basket cell axons⁵⁰ ensure that SEMA3A promotes specific axon branching without affecting overall axon organization. This might be a general regulatory mechanism whereby molecular cues can influence local branching without affecting axon outgrowth or guidance.

Like semaphorins, SLITs, which act on ROBO receptors, are repulsive cues and they function, for example, in the guidance of commissural axon pathways in the mammalian forebrain⁵¹. However, SLITs can also promote the collateral axon branching of mammalian sensory DRG axons in vitro⁵² and the branching of zebrafish trigeminal axons in vivo⁵³. By contrast, Slit1a inhibits arborization of retinal ganglion cell axons in the zebrafish optic tectum, thereby preventing the premature maturation of terminal arbors⁵⁴. In this study, in vivo imaging showed that genetic removal of Slit–Robo signalling increases the size and complexity of axon arbors by decreasing branch dynamics and increasing stable branch tips. SLITs, like semaphorins, exhibit context-dependent promotion or repression of axon branching, depending on the particular population of neurons. Moreover, the balance between factors that promote and inhibit arborization may determine final arbor size.

Growth factors

Growth factors, such as fibroblast growth factor (FGF), and neurotrophins, such as nerve growth factor (NGF) and BDNF, can also evoke axon branching and arborization. For example, in vitro, NGF promotes DRG axon branching^55,56, BDNF evokes filopodial and lamellipodial protrusions along frog spinal axons⁵⁷, and FGF-2 and BDNF stimulate branching in cortical axons^58,59. In hippocampal slice cultures, BDNF transfection specifically increases axon branching in dentate granule cells⁶⁰. Moreover, BDNF has been shown to have a prominent role in axon arbor formation in retinal axons in the frog optic tectum^61,62. For example, in vivo imaging⁶³ showed that acute BDNF application induces rapid extension of new branches on retinal axon arbors. Thus, target-derived cues shape terminal arbors by regulating the dynamic extension and retraction of their branches. Interestingly, comparisons between the effects of netrin-1 and BDNF on retinal arbors³⁷ suggest that although both factors increase arbor complexity, BDNF promotes the addition and stability of axon branches whereas netrin-1 induces new branch growth but not branch stabilization. Different cues can, therefore, produce a similar outcome on final arbor morphology by different dynamic strategies.

Morphogens

WNTs comprise a diverse family of secreted morphogens that shape embryonic development. Several WNTs also function as axon guidance cues^64,65 or regulate axon branching. For example, WNT7A⁶⁶, secreted by cerebellar granule cells, induces remodeling of pontine mossy fibre axons by inhibiting their elongation, enlarging their growth cones, inducing spreading of regions along the axon shaft and consequently increasing axon branching. WNT3A, secreted by motor neurons, regulates terminal differentiation in a subset of presynaptic arbors of spinal sensory neurons by inducing axon branching^67,68, whereas WNT5A, a repulsive axon guidance cue for cortical axons in vivo^69–71, increases the elongation of cortical axons and branches⁷² but not the numbers of axon branches⁷³. However, WNT5A increases axon branching of developing sympathetic neurons acutely and over the long term also enhances axon extension⁷⁴.

These findings suggest that extracellular cues can promote or repress axon branching without affecting the growth and guidance of the parent axon. Differences between local signalling and cytoskeletal mechanisms at the growth cone and along the axon shaft could account for different effects of the same molecule on axon guidance, elongation and branching.

Reorganization of the cytoskeleton

The initiation, growth and guidance of axon branches in response to extracellular cues require regulation of actin and microtubule dynamics (FIG. 2), which share similarities^5,75,76 with cytoskeletal dynamics during growth cone navigation^75,77–79. Filopodial and lamellipodial protrusions along the axon serve as precursors to branches. Filopodia contain long unbranched bundled actin filaments, whereas lamellipodial veils contain a meshwork of branched actin filaments. An initial step in branch formation is the focal accumulation of actin filaments at axon membrane protrusions^34,80. Filopodia can emerge from transient accumulations of actin filaments (termed actin patches) on sensory axons^81–83, although only a fraction of actin patches gives rise to axonal filopodia³.

Figure 2. Cytoskeletal reorganization at different stages of axon branching.

a | Membrane protrusion requires the accumulation of actin filaments, which can form patches^81–83. In (a) membrane protrusion requires actin nucleation (ARP2/3^82,83 and Cordon-bleu⁸⁸), actin branching (ARP2/3) and actin elongation (Ena/VASP)³⁵. Cofilin is important for the turnover of actin filaments⁹² and Septin 6 is localized to the actin patch¹¹⁰. MTs are stabilized by MAPs (Tau//Doublecortin)^78,99,103 in the axon but are also capable of extending at their plus ends (APC/EB proteins)^100,101. b | Filopodia, which contain bundled actin filaments, emerge from the axon shaft accompanied by localized splaying and fragmentation of MTs^{55,80,93,96,97}; these invade the filopodium and extend along actin bundles. MTs are severed by Spastin and Katanin⁹⁸. Septin 7 promotes the entry of MTs into the filopodium¹¹⁰. c, d | Stable MTs enter the nascent branch (c), which continues to mature and extend; this growth is led by a motile growth cone (c, d). MAPs protect some of the MTs in the axon shaft from severing (c, d). Within the axon branch MTs become bundled and stabilized by MAPs (d).

Actin dynamics

Actin filaments are polymers with barbed ends, facing the plasma membrane, where actin monomers are added and pointed ends, facing away from the membrane, where actin filaments disassemble. Membrane protrusion and motility are driven by dynamic cycles of actin polymerization and depolymerization^77,78. In growth cones and filopodia, actin filaments undergo continual cycles of retrograde flow and polymerization⁸⁴. The balance between these two processes determines whether membrane protrusions extend or withdraw.

Actin arrays and dynamics are highly regulated by actin-associated proteins, which have important roles in axon guidance and branching⁷⁷ (FIG. 2). For example, Ena/VASP proteins (enabled/vasodilator-stimulated phosphoprotein) bind to the barbed ends of actin filaments and enhance actin filament elongation by antagonizing the capping proteins that inhibit F-actin elongation⁸⁵. When Ena/VASP proteins are sequestered to the mitochondria in hippocampal neurons, the axons of these neurons have fewer filopodia containing fewer actin filament bundles³⁵. Interestingly, Ena/VASP depletion in frog retinal ganglion neurons reduces filopodial formation, retinotectal axon branching and arborization in the tectum in vivo⁸⁶. The formins, which nucleate (i.e. promote the formation of actin filaments de novo, rather than adding to existing filaments) actin filaments at barbed ends, are essential in neurite initiation and the formation of filopodia. In cortical neurons lacking Ena/VASP, filopodia can be restored by ectopic expression of the formin mDIA2 ⁸⁷. A recently identified actin nucleator, cordon-bleu⁸⁸, promotes actin polymerization of unbranched filaments and, when overexpressed in hippocampal neurons in vitro, increases axon branching without affecting axon length. Actin-related protein 2/3 (ARP2/3) nucleate new branched filaments from the sides of existing actin filaments⁷⁷ and are required for NGF-induced actin patches, filopodia formation and branching of sensory axons^82,83. One study has shown that ARP2/3 depletion in hippocampal neurons⁸⁹ reduces filopodial and lamellipodial dynamics, although another study⁹⁰, using a dominant negative approach to inactivate ARP2/3, saw no decrease in axonal filopodia.

Dynamic axonal filopodia not only protrude but also withdraw. Gelsolin severs actin filaments and caps free barbed ends and its depletion in hippocampal neurons increases filopodia through reduced retraction⁹¹. The actin-severing protein family ADF/cofilin increases actin filament depolymerization but it also promotes filament assembly by increasing the available pool of actin monomers and free barbed ends. Thus, ADF/cofilin activity contributes to BDNF-induced filopodia on retinal growth cones⁹². Myosin II, a motor protein involved in actomyosin contraction of antiparallel filaments, drives retrograde actin flow and aids in disruption of F-actin networks⁸⁴. Although the exact role of many actin-associated proteins remains to be determined in vertebrate axon development, dynamic remodeling of the actin cytoskeleton during cycles of polymerization and depolymerization is an essential event in the initiation of branches by filopodia and lamellipodia as they probe the extracellular environment.

Microtubule dynamics

Actin filament polymerization is essential for the pushing force that drives membrane protrusion but further development and stabilization of axon branches require the entry of dynamic microtubules (MTs) into filopodia^5,93. MTs are hollow tubular structures that comprise linear arrays of α and β tubulin^78,94. In growth cones and developing branches, the MT ‘plus’ ends, where tubulin dimers are added and subtracted, face outward toward the membrane, whereas MTs are generally stabilized at their minus ends, which are oriented away from the membrane. MTs undergo dynamic instability⁹⁵ that results in continuous cycles of growth and shrinkage at their plus ends. These cycles of polymerization and depolymerization enable MTs to reorganize and explore the growth cone periphery and extend into developing axon branches. MTs in axons are organized in stable parallel bundles, so important early steps in axon branch formation are the splaying of bundled MTs, their local fragmentation at axon branch points followed by entry of short MT fragments into filopodia. This dramatic MT reorganization, observed in chick sensory axons as well as rodent hippocampal and cortical neurons^{55,80,93,96,97} leads to stabilization of elongating axon branches containing MTs. As shown by time lapse imaging, only those filopodia that contain MTs develop into branches⁹³, although even long branches containing MTs can eventually retract.

How are MT dynamics and organization regulated during axon branch formation? Microtubule-associated proteins (MAPs) — such as stabilizing, plus-end-tracking, severing, destabilizing and motor proteins — have roles in growth cone steering^77,78,94 and some of these MAPs have also been investigated for a role in axon branching. For example, overexpression of MT severing proteins (either spastin or katanin), in cultured hippocampal neurons, increase axon branching⁹⁸, while depletion of either protein reduces axon branching. Interestingly, tau, a MAP that stabilizes MTs against depolymerization, protects MTs against the severing effects of katanin, so that when tau is depleted axon branching is increased⁹⁹. This suggests that detachment of tau from MTs might be a mechanism for regulating branch formation. One possibility is that as short MT fragments enter nascent branches, structural MAPs such as MAP1B, which also interacts with actin filaments⁷⁸, could stabilize them.

Further growth of MTs at their dynamic plus ends is regulated by a family of plus-end tracking proteins (+TIPs), such as the end-binding proteins EB1 and EB3 and adenomatous polyposis coli protein (APC), which contribute to MT elongation and axon outgrowth¹⁰⁰. WNT3A-induced remodeling of DRG axons⁶⁸ — comprising a decrease in the growth rate of the axon and increases in branching and growth cone size — results from changes in the directionality of MTs that lead to MT looping; such looping has also been observed in large paused cortical growth cones^80,93. Since WNT3A decreases the levels of APC, APC is thought to have a role in determining the direction of MT growth, although exactly how this contributes to regulation of axon branching is unclear. APC-deficient mice in vivo show severe misrouting of cortical axons and, in vitro, cortical axons lacking APC show excessive axon branching, which is caused by MT instability and disruption of MT organization at branch points¹⁰¹. The growth cones of APC-deficient cortical neurons frequently split to produce multiple branches¹⁰², and this results from MT splaying and disorganization. Deletion of the gene encoding doublecortin (DCX), another MAP, not only causes migratory defects in adult mouse forebrain neurons but also evokes excessive branching of the primary neurite¹⁰³, which was attributed to a failure in the maintenance of MT cross linking. The MT-destabilizing kinesin, KIF2A, also has a role in axon branching¹⁰⁴. Hippocampal and cortical neurons in vivo and in vitro from mice lacking KIF2A have abnormally long axon collateral branches that form more secondary and tertiary branches. Concomitantly, in neurons lacking KIF2A, MT-depolymerizing activity is diminished, suggesting that KIF2A normally suppresses collateral growth. These results emphasize the importance of regulated cytoskeletal dynamics in repressing axon branching during the development of neural circuitry.

Actin–microtubule interactions

Neither actin filaments nor MTs act alone during the formation of axon branches. Dynamic MTs and actin filaments are thought to interact bidirectionally in growth cone motility^78,105, as dynamic MTs extend along and copolymerize with actin filaments in the growth cone periphery^80,106,107. Actin–MT interactions are also essential for branch formation, as MTs extend into filopodia along actin filament bundles (FIG. 2). Thus, drugs that attenuate the dynamics of either actin filaments or MTs inhibit branch formation but not extension of cortical axons in vitro⁸⁰. Mechanisms that link actin filaments and dynamic MTs may have implications for axon branch formation. For example, in cultured cortical neurons, the interaction between the F-actin-associated protein drebrin and EB3 occurs at the base of filopodia, which is important for the exploration of filopodia by MTs during neuritogenesis¹⁰⁸. Genetic ablation of ADF/Cofilin in cortical and hippocampal neurons, disorganizes the actin network and impairs MT bundling and protrusion into filopodia, leading to inhibition of neurite initiation¹⁰⁹. Since ADF/Cofilin regulates actin retrograde flow by actin disassembly and turnover, the ability of MTs to coalesce and form neurites was thought to depend on actin retrograde flow driven by actin-severing proteins. In this study, the authors concluded that MTs follow the lead of dynamic actin filaments. Coordination of actin and MT dynamics is also required for axon branching in chick DRG neurons and mammalian hippocampal neurons. For example, branch formation in sensory axons involves several septin proteins that interact with both cytoskeletal components¹¹⁰. SEPT6 localizes to actin patches and recruits cortactin, an ARP2/3 regulator, to trigger formation of filopodia, whereas SEPT7 promotes the entry of axonal MTs into filopodia, enabling branch formation through as yet unknown mechanisms. It seems likely that future studies will reveal additional molecules that coordinate reorganization of the actin and MT cytoskeletal to promote development of axon branches.

Signalling pathways

During axon pathfinding, the growth cone must interpret and integrate numerous extracellular cues simultaneously through activation of multiple signal transduction pathways^111,112. Some of these pathways also regulate axon branching. RhoGTPases^78,113 and the protein kinase glycogen synthase kinase 3β (GSK3β)^114–117 have emerged as signalling nodes that regulate actin and MT dynamics through multiple extracellular cues and cytoskeletal effectors. Although no single pathway from receptor to the cytoskeleton has been completely defined, we will present several examples of signalling pathways that regulate axon branching (FIG. 3).

Figure 3. Signalling pathways that promote axon branching.

a | Different pathways leading to MT destabilization (top) at early branching stages or by MT stabilization at later branching stage (bottom) can each promote axon branching by opposing effects on MT stability. Differential effects on MT stability are shown in two examples of branching induced by BDNF and its TRKB receptor (top) in cortical neurons⁵⁸ or the morphogen WNT7A (bottom) in pontine mossy fibers^66,120 b | BDNF and netrins increase terminal arborization of frog retinotectal axons^{37, 63}. In this signalling pathway¹²³, NEDD4 ubiquitinates PTEN and PTEN degradation leads to increased terminal axon branching by promotion of PI3K, which is known to regulate cytoskeletal dynamics. By contrast, inhibition of NEDD4 leads to an increase in PTEN levels and hence inhibition of axon branching. c | NGF-induced TRKA signalling promotes axon branching in chick sensory (DRG) axons⁸². It does this by activating PI3K and, in turn, RAC, which activates actin associated proteins to increase actin polymerization and the formation of actin patches. Cortactin, recruited by Septin6, promotes the emergence of filopodia from actin patches. Abbreviations: brain-derived neurotrophic factor (BDNF), mitogen activated protein kinase phosphatase (MKP-1), MAPK c-jun N-terminal kinase (JNK), stathmin (STMN), phosphoinositide 3-kinase (PI3K), ubiquitin ligase E3 (NEDD4), the phosphatase phosphatidylinositol 3,4,5 trisphosphate (PTEN), nerve growth factor (NGF), actin related protein (ARP 2/3).

NGF promotes the formation of filopodia and branches in chick sensory axons⁵⁵ by increasing the number of actin patches through an increase in the rate of patch formation⁸¹. This effect is driven by phosphoinositide 3-kinase (PI3K)^82,83 and requires the actin nucleating complex ARP2/3. In a model of the NGF signalling pathway⁸², the NGF receptor TRKA activates PI3K, which in turn activates RAC1 to drive activity of WAVE1 that activates ARP2/3. The actin-associated protein cortactin, which stabilizes branched actin filaments nucleated by ARP2/3¹¹⁸, does not regulate actin patch formation but promotes the emergence of filopodia from actin patches. Cortactin is also involved in axon branching¹¹⁹ by promoting actin polymerization and membrane protrusion. In hippocampal neurons, calpain, through proteolysis of cortactin, represses actin polymerization and maintains the neurite shaft in a consolidated state, which can be de-repressed by factors such as netrin-1 and BDNF to permit branching. These examples illustrate how actin-nucleating proteins and associated stabilizers regulate specific steps in branch formation as well as suppression of axon branching.

BDNF induces axon branching of mouse cortical neurons through its receptor TRKB. Recently, negative regulation of mitogen-activated protein kinase (MAPK) by the MAPK phosphatase MKP-1 was shown to have a role in BDNF-induced cortical axon branching⁵⁸. MAPK is also involved in cortical axon branching downstream of netrin-1-induced calcium signalling³⁶. In vivo overexpression or down regulation of MKP-1 in layer II–III cortical neurons increases or decreases their terminal axon branching, respectively, in the contralateral cortex. MKP-1 is induced transiently at sites of BDNF activity and inactivates the MAPK c-jun N-terminal kinase (JNK) by dephosphorylation, which in turn reduces the phosphorylation of the JNK substrate stathmin-1 (STMN1), thereby activating STMN1. This study also showed that MKP-1 overexpression increased tyrosinated tubulin (associated with dynamic MTs) and decreased detyrosinated and acetylated tubulin (associated with stable MTs). Since STMNs destabilize MTs⁹⁴, in this model⁵⁸, the authors propose that activating STMNs through BDNF signalling destabilizes MTs and, consequently, increases cortical axon branching (FIG. 3).

In other models of axon branching, extracellular cues have been shown to regulate axon branching by influencing MT dynamics through GSK3β signalling. For example, WNT7A⁶⁶ induces filopodia and spread regions along the axon shaft of pontine mossy fibres where stable MTs become unbundled¹²⁰, leading to increased axon branching. WNT7A inhibits GSK3β activity, which normally phosphorylates MAPs such as tau, APC and MAP1B⁶⁵. Activation of the WNT pathway therefore decreases MAP1B phosphorylation. Since phosphorylated MAP1B maintains MTs in a dynamic state¹²¹, decreased MAP1B phosphorylation would increase MT stability, leading to terminal remodeling of mossy fibres and axon branching. This seems to contradict results showing that axon branching is dependent on pathways that destabilize MTs⁵⁸. However, axon branching initially involves destabilizing MTs at branch points (FIG. 2) but further growth of branches requires entry of stable MTs into the nascent branch. GSK3β has also been shown to function in a novel signalling pathway that inhibits axon branching of granule cell neurons in the rodent cerebellar cortex¹²². In this study, knockdown of the JNK-interacting protein kinase 3 (JIP3) stimulates granule cell interstitial axon branching by decreasing levels of GSK3β. Suppression of axon branching by JIP3–GSK3β signaling involves phosphorylation of a novel GSK3β substrate, DCX. However, the downstream cytoskeletal mechanisms by which the JIP3–GSK3β–DCX signalling pathway actively suppresses axon branching are unknown. Manipulation of GSK3s α and β levels and its association with different MAP substrates in DRG and hippocampal neuronal cultures¹¹⁷ have led to various outcomes on axon outgrowth and axon branching, highlighting the complexity of GSK3 function in the regulation of MT organization and dynamics.

Netrin-1 and BDNF regulate terminal branching of retinal axons in the frog optic tectum^37,63. A recent in vivo study¹²³ identified a signalling pathway involving the ubiquitin-protein ligase E3 NEDD4 and the phosphatase phosphatidylinositol 3,4,5- trisphosphate (PTEN) that negatively regulates the PI3K signalling pathway (FIG. 3). Inhibition of NEDD4 increases the levels of PTEN, resulting in simple sparsely branched retinal axon arbors in the tectum but no effect on axon navigation. Thus, PTEN overexpression inhibits axon branching, whereas PTEN knockdown restores axon branching. In this study, the authors proposed that, in response to external signals, NEDD4 ubiquitinates PTEN, thereby targeting it for degradation, which promotes the PI3K pathway and the downstream cytoskeletal reorganization that leads to axon branching. Importantly, this study shows how signalling pathways can specifically regulate terminal branching without affecting long-range axon pathfinding.

Activity-dependent mechanisms

Neuronal activity can also regulate axon branching (FIG. 4). Spontaneous electrical activity during early development¹²⁴ generates transient fluctuations in intracellular calcium, an essential second messenger. Calcium transients, activated by cues such as netrin-1 and BDNF in neuronal growth cones, regulate axon outgrowth and guidance¹²⁵ by signalling to downstream effectors such as calcium/calmodulin-dependent protein kinase II (CaMKII) and the calcium/calmodulin-dependent phosphatase calcineurin that ultimately regulate the growth cone cytoskeleton. Calcium signalling also has an important role in axon branching. For example, local application of netrin-1 to cortical axons in vitro³⁶ increases the frequency of calcium transients in small regions of the axon and simultaneously evokes rapid localized filopodial protrusion and axon branching through CaMKII and MAPK activation. Thus, local Ca²⁺ transients can promote branching within discrete regions of the axon.

Figure 4. Competition shapes the morphology of terminal arbors.

a | The terminal arbor of a retinal axon in the zebrafish optic tectum with normal electrical activity has a complex highly branched morphology (left), and it can outcompete a neighbouring retinotectal arbor in which electrical activity has been silenced (right), resulting in a smaller less elaborate morphology in the silenced axon¹³⁴. b | A retinal axon arbor in a normal control zebrafish tectum (left) has a smaller less complex morphology when all other retinal gangion cells (RCGs) are present. In the mutant zebrafish with only a single RCG axon (right), arbors terminate in appropriate tectal regions but are larger and more complex in the absence of competitive interactions with neighboring axons (figure adapted from ¹³⁶). Anterior posterior tectum (A, P), nasal temporal retina (N, T).

Downstream of calcium activity and activation of CaMKII, which acts as an intracellular calcium sensor, the Rho family of small GTPases regulates actin filament assembly and organization in growth cones^78,125. The RhoGTPase RhoA has also been shown to have an important role in cortical axon branching¹²⁶. In this study branching of horizontal axons arising from upper layer cortical neurons, was observed in cortical slices where introduction of constitutively active RhoA increased axon branching through increased dynamic addition and loss of branches while RhoA inhibition decreased branching. Blockade of neural activity decreased active RhoA, suggesting that RhoA is a positive regulator of activity dependent axon branching in cortical neurons. This is only one possible regulator of activity-dependent axon branching, and further study will be required for understanding molecular mechanisms that mediate activity-dependent processes.

Branches can extend while their axons stall or retract¹⁹. In vivo, axons and branches are likely to encounter different types or concentrations of factors that evoke different levels of Ca²⁺ activity, which could independently regulate the outgrowth of different processes from the same axon. For example, in NGF-mediated axon outgrowth of sympathetic neurons, axon branches that are depolarized to induce calcium entry are favoured for growth at the expense of unstimulated branches of the same axon¹²⁷. In cortical neurons, imposition of Ca²⁺ activity in an axon versus its branches, causes a process with higher frequency Ca²⁺ transients to grow faster at the expense of another process with lower levels of Ca²⁺ activity that stalls or retracts¹²⁸. How growth of one axonal process is coordinated with retraction of another is unknown.

Activity-dependent mechanisms are also important in shaping terminal axon arbors^129,130. For example, patterned neural activity in the frog retinotectal pathway shapes the morphology of retinal axon arbors by governing axon branch dynamics¹³¹. Callosal axons in mice expressing an inward rectifying potassium channel to suppress neural activity extend through the contralateral cortex with little branching and end in immature growth cone-like structures¹³². Similarly, suppression of spontaneous firing of thalamic or cortical neurons in organotyptic co-cultures reduces thalamocortical axon branching¹³³. Conversely, neural activity enhances the branch dynamics of thalamocortical axons that are biased toward branch addition and elongation in cortical target layers¹³⁴.

The effects of neural activity can also involve competition among neighbouring axon arbors. For example, in the zebrafish retinotectal system¹³⁵, silencing electrical activity in one axon relative to neighboring axons reduces the size of the terminal arbor of the ‘silenced’ axon (FIG. 4). Arbor size and complexity of the neuron with reduced activity is restored when activity of neighboring axons is silenced, demonstrating a competitive activity-dependent mechanism. However, another study in the zebrafish retinotectal system¹³⁶ showed that singly silenced axons have larger than normal arbors, although silencing neighbouring axons also restores normal arbor size to the activity-suppressed axon. In this study suppression of presynaptic activity in retinal ganglion cells and live cell observations of axon arbor formation over days showed that presynaptic activity is not required for development of arbors with an appropriate number of stable branches. However, presynaptic activity is required for arbor maturation, as shown by increased production of dynamic filopodia. These results suggest that activity normally arrests formation of filopodia during arbor maturation. Moreover singly silenced axons fail to arrest growth of branches, leading to expansion of the arbor territory on the tectum. Restoration of normal arbor size by silencing of neighboring retinal axons demonstrates that, although the cellular mechanisms underlying the competition are unknown, the ability of axons to arrest their growth is an activity-dependent competitive process. To further demonstrate effects on axon arbors in the absence of competition, zebrafish with a single retinal ganglion cell were created¹³⁷. Arbors of this single axon terminate in appropriate tectal regions but are larger and more complex than normal. Although an understanding of the exact mechanisms requires further study, these examples show the importance of competitive interactions in shaping axon terminal arbors.

Conclusions and future directions

Over the past decade, we have gained significant insight into mechanisms that regulate axon branching. The discovery of new branching mechanisms, such as the kinase signalling pathway that controls immobilization of mitochondria at nascent presynaptic sites to regulate cortical axon branching¹³⁸, will continue to expand and modify our understanding of axon branching. However, many issues remain unresolved. First, we still lack a complete understanding of why individual axons branch only at specific target locations. The presence of gradients of molecular cues and the suppression of inappropriate branching have been postulated as mechanisms that underlie the specificity of branching in the retinotectal system⁷ but whether these apply more generally to vertebrate systems will require further study. Second, although the cytoskeleton is the final target for the convergence of signalling pathways, we still lack a clear understanding of how the consolidated axonal cytoskeleton is transformed into dynamic MTs and actin filaments that drive membrane protrusion at axon branch points. The improvements in genetically encoded labeling techniques and high resolution time lapse microscopy should now make it possible to observe longer term cytoskeletal reorganization and dynamics that occur at specific stages of axon branching (BOX 1). Third, to understand mechanisms of axon branching in vivo, it will be important to investigate signalling events and cytoskeletal changes during branching in preparations of the CNS such as slices or explants that recapitulate the complex in vivo environment. These approaches can be combined with genetic manipulations of signalling components and cytoskeletal effectors. It is clear from this review that our understanding of how signalling pathways transduce effects of extracellular cues on the axonal cytoskeleton to promote axon branching is at an early stage. At present, it is unclear how the same guidance cue can have different effects on branching in different neuronal populations or differentially affect growth of an axon and branches of the same neuron. In future studies, it will be important to connect intracellular signalling to its effects on cytoskeletal dynamics at branch points in living neurons of different types. Moreover, mechanisms for suppression of axon branching to create topographic specificity are poorly understood. Live cell imaging has provided an exciting window into dynamic remodeling of axon terminal arbors in vivo and it will be important in revealing how electrical activity through calcium signalling regulates arbor formation and competitive interactions among neighbouring axons in a target region. Addressing such questions with new techniques over the next decade will inform our understanding of the remarkable ability of a single axon to branch into specific regions of the developing nervous system.

Box 1. New technologies for studying axon branching.

Much of the organization of axon branch points is obscured because the small diameter and cylindrical structure of the axon. Many of the live-cell imaging studies of cytoskeletal dynamics at branch points and the growth cone described in this review have employed transfection of neurons with plasmids encoding fluorescent fusion proteins, such as enhanced green fluorescent protein (EGFP) and mCherry fused to cytoskeletal components such as actin and tubulin. To visualize fluorescently labeled actin filaments and microtubules and their associated proteins many investigators are currently using high resolution microscopy such as total internal reflection fluorescence (TIRF) microscopy and spinning disk confocal microscopy. However, new advances in imaging — specifically structural illumination microscopy (SIM)¹³⁹, stimulated emission depletion (STED) microscopy¹⁴⁰, photoactivated localization microscopy (PALM)¹⁴¹ and stochastic optical reconstruction microscopy (STORM)¹⁴² — will allow future studies of cytoskeletal dynamics and signalling at branch points at resolutions on the order of tens of nanometers. Importantly, all of these imaging techniques can be used on living samples. The study of axon branching would also greatly benefit from the availability a greater diversity of ‘caged’ proteins and compounds, so that local uncaging could be performed at specific sites along an otherwise consolidated axon, in a similar way to the uncaging of neurotransmitters at individual dendritic spines. Optogenetic control over genetic or functionally defined populations of or individual neurons will undoubtedly allow much more sophisticated experiments on the role of activity in axon branching at high temporal and spatial precision¹⁴¹. Such techniques would also allow researchers to test the role of activity in individual branches or arbors of the same neuron. To understand axon branching, it is also important to reconstruct, with high precision, the full extent of an axon arbor. In the past, this has required laborious thin section and/or electron micrograph reconstruction. The recent advent of tissue clearing techniques, such as ScaleA2 ¹⁴³, CLARITY¹⁴⁴, SeeDB ¹⁴⁵ and Clear^{T/T2 146}, which allow fluorescently labeled axons and branches to be reconstructed in their full extent, will help in understanding how axon branching is affected in normal, transgenic and disease models.

Nature Reviews glossary

1. terminal arbor

The highly branched tree-like structure at the end of an axon that innervates target regions

2. collateral branch

A branch that extends from the sides of an axon, often interstitially, that can innervate a target by re-branching to form a terminal arbor

3. cytoskeletal dynamics

Cycles of polymerization and depolymerization resulting in growth and shrinkage of microtubules and actin filaments allowing their reorganization

4. forward and reverse signaling

EPHAs act both as receptors activated by ephrin ligands (forward signaling) and as ligands that activate ephrins (reverse signaling)

5. neurotrophic factor

Molecules such as BDNF and NGF that regulate neuronal growth and survival

6. lamellipodium

A thin sheet like veil of cytoplasm at the growth cone periphery comprised of actin filament networks

7. filopodium

A finger like membrane protrusion containing bundled actin filaments. Filopodia extend transiently from the growth cone, the axon shaft and from axon branches

8. +TIPS

Plus end tracking proteins such as EB1 and EB3 that associate with the growing plus ends of dynamic microtubules

9. RhoGTPases

A family of molecules that regulate cytoskeletal dynamics downstream of guidance cue receptors

10. calcium transient

A transient elevation in the level of intracellular calcium that can occur repetitively at different frequencies

Biographies

Katherine Kalil received her Ph.D at the Massachusetts Institute of Technology, Cambridge, USA, and has spent her research career at the University of Wisconsin-Madison, USA, where she is a professor in the Department of Neuroscience. Her research focuses on cytoskeletal and signaling mechanisms in axon growth and guidance.

Erik Dent received his Ph.D at the University of Wisconsin-Madison, USA, and conducted postdoctoral training in the Department of Biology at Massachusetts Institute of Technology, Cambridge, USA. He returned to Madison and is now an Associate Professor in the Department of Neuroscience at the University of Wisconsin-Madison, USA. His research focuses on the actin and microtubule cytoskeleton in neuronal development and plasticity.

Nature Reviews Online Summary

Axon branching connects single neurons with multiple targets, which, along with the formation of highly branched terminal arbors, underlies the complex circuitry of the vertebrate CNS.

Axon collateral branches extend interstitially from the axon shaft as dynamic filopodia that develop into branches at appropriate targets regions to form functional maps. Extrinsic guidance cues, growth factors and morphogens regulate axon branching and shape terminal arbors that develop from axon branches.

Growth and guidance of axon branches in response to extracellular cues require dynamic reorganization of the actin and microtubule cytoskeleton. Cycles of cytoskeletal polymerization and depolymerization are highly regulated by actin and microtubule associated proteins during branch formation.

Complex signaling pathways activated by extracellular cues through their receptors regulate axon branching. The ultimate target of signal transduction pathways is the cytoskeleton, which can reorganize by changes in dynamics to promote or suppress axon branching.

Neuronal activity, often stimulated by extracellular cues, can regulate axon branching by transient fluctuations in intracellular calcium that acts as a second messenger to activate downstream cytoskeletal effectors. Effects of neural activity can involve competition among neighboring axon arbors such as in the retinotectal system where competitive activity-dependent mechanisms regulate arbor size and complexity.

Future directions in the study of axon branch formation will involve the use of preparations of the vertebrate CNS that recapitulate the complexity of the in vivo environment. Improvements in labeling techniques and high resolution time-lapse microscopy should facilitate such studies.