Long-range axon branching: contributions to brain network plasticity and repair

Abstract

Brain function requires exquisitely adapted plasticity at multiple scales, from synapses to whole-brain networks. Evidence for large-scale plasticity in functional brain networks comes from neuroimaging data across a variety of species, particularly during development and following injury. However, how large-scale network remodelling is achieved at the microscopic level is unknown as the growth of entirely new long-distance axons is unlikely to occur. Recent insights from electron microscopic connectome studies and single-cell projectomes of neurons in the brains of multiple model organisms have provided new evidence for the incredible structural complexity of axons and their branches that traverse the brain. This evidence shows highly arborized axonal projections, differentially myelinated branches of the same axon, and axonal regions devoid of synaptic contacts but with the potential to form synaptic connections in new or additional areas. Recent electron microscopic data suggest that these axonal features may be evolutionarily conserved. Here we consider whether these features could enable long-range and large-scale neuroplastic changes at a functional level, particularly following focal brain injury. These insights contribute to our emerging understanding of how the brain undergoes large-scale reorganization to adapt to changing circumstances.

Introduction

Neuroplasticity has been defined as the modifiability of the brain during development and in the adult¹, as well as brain remodelling that can be triggered at any stage in life². For example, injury or lesions of the brain can lead to plastic rearrangements of functional networks via increased synaptic plasticity and the local sprouting of short axonal or dendritic branches^3,4. Less is known about how long-range network plasticity occurs within the brain (for example, how damaged local circuits following focal brain injury repair and reintegrate into global networks). The ability of the brain to adapt to new information and to accommodate environmental changes is fundamental to how the brain functions. Such changes may be subtle, for example, modifying learned information or recalling memories; or more substantial, for example, when areas of the brain alter their processing of new modalities of sensory or motor information following injury or disease. However, we currently lack a unifying understanding of how different types of plasticity can be achieved through structure, although the importance of studying the brain across micro–meso and macro scales has long been recognized⁵.

The ability of the brain’s synapses to change throughout life, and in response to learning, has been shown in both in vitro and in vivo experiments (reviewed in ref. 6). Synaptic plasticity is crucial for the encoding of new information and the updating of stored information. Changes in synaptic spine morphology or the generation of new spines mediate dynamic contacts between neurons. The highly localized nature of synaptic remodelling makes it experimentally tractable and has thus been studied extensively. Likewise, the elaborate dendritic arbours of neurons have been documented in detail in a neuronal cell-type specific manner (for example, in the Allen cell types database⁷ and at NeuroMorpho.org⁸). Neurons have been characterized by the pattern of their dendritic arbours, which span both radially and laterally defined dimensions across tens to hundreds of micrometres (over a few brain sections)^9,10. By contrast, axonal arbours of neurons can span vast distances between areas, regions and hemispheres (over many hundreds of micrometres to millimetres or even centimetres across the brain depending on the species). Although these features make long-range axons much more challenging to study and measure¹¹, they could potentially be key to mediating large-scale functional plastic changes of the brain. Here, we highlight the extent of axonal arborization and long-range connectivity in the brain from recent three-dimensional volume electron microscopy connectome datasets at the nanometre scale, histological data at the micrometre scale and macro-level whole-brain connectivity across a variety of well-known examples of large-scale functional plasticity during development and following brain injury. Combined, this information provides a framework for understanding the structural basis of how local and global connectivity is integrated across scales and how the brain may be able exploit axonal or synaptic plasticity to alter connectivity.

Observing axonal arbours in vivo

Classical methods for studying axonal trajectories relied upon anterograde and retrograde labelling strategies, which inherently require axonal transport or membrane diffusion and usually bulk labelling^12,13. Single-neuron or sparse labelling paradigms have been much less commonly employed due to the difficulty in visualizing entire neurons (including all axonal arbours) labelled in vivo with processes potentially traversing across many sections. Even these studies tended to focus on a specific area of the brain to test an a priori hypothesis¹⁴. Recently, however, electron microscopic connectome reconstructions (a detailed wiring diagram) of the whole brain and nervous systems of different species have become available, in addition to unbiased single-neuron labelling and reconstruction (a diagram of a single neuron and all of its dendrites and axonal branches)^15–18. These methods have demonstrated that the branching and arborization of axons is a key dimension of the neuron’s computational capacity and connectivity across the brain/nervous system.

The findings of in vitro studies, as well as the limited capacity of classical techniques to trace all branches of a single neuron’s axons, have led to the common conception of axons as simple, fixed conduits between the cell body and the target. However, an accumulating body of evidence, reviewed here, suggests the possibility that axons can innervate multiple targets and that information can be differentially propagated along different branches of the axon. Here, we postulate that this could vastly increase the network influence of a single neuron and potentially provide some redundancy within the neuron’s structure that enables plastic changes in connectivity to occur without having to generate entirely new long-range projections. Furthermore, we propose that some silent and/or dormant long-range axonal branches, which appear to be retained even though they lack functional synapses, may provide a key conduit for neuroplastic functional reorganization following injury if active synapses can be formed on these branches.

Structural complexity of long-range axons

In the late nineteenth century, Santiago Ramón y Cajal’s ground-breaking work on neural structures revolutionized our understanding of the nervous system. His meticulous observations led to the ‘neuron doctrine’, which concludes that neurons are discrete functional units, transmitting information unidirectionally from dendrites to axons. Importantly, Cajal also pioneered the idea of a dynamic and changing nervous system, introducing the concept of ‘plasticity’ at a time when prevailing views held that neural circuits were fixed and unchangeable. He proposed that new connections could form through the growth and branching of both dendrites and axons during development and post-trauma functional recovery in adults. This visionary idea laid the foundation for the field of brain plasticity and inspired countless discoveries, including decades of research on neuronal structural plasticity^19,20. Although research on neural plasticity has historically focused on the structure of dendrites and dendritic spines^21–23, unintentionally leading to an oversimplified view of the structure and adaptability of axons, modern single-cell projectomes and electron microscopic-based connectomes are revealing the intricate complexity of axons at both the mesoscale and microscale. This is particularly notable for projection neurons that have long-range connectivity, as the organizational principles that govern local neuronal branching and complexity might also shape macroscale properties of global brain networks. As discussed below, these structural properties are particularly relevant to neuroplasticity and functional recovery following stroke.

Insights from projectome studies in vivo

Studies of axonal morphologies have traditionally made use of different tracing techniques that have accelerated since the 1980s (ref. 24), contributing to leaps in our understanding of brain structure and function²⁵. Traditional single-cell tracing techniques involving tracer or virus injections^24,26,27 enabled the first glimpses of individual axonal projections over long distances in mammalian brains^28–33. These studies revealed surprising complexity in how individual neurons connect across long distances in the brain. For example, researchers using biotinylated dextran amine tracing showed that a single layer V neuron in the rat’s frontal cortex extends its axon all the way to the subthalamic nucleus, emitting collaterals to innervate multiple brain regions along its path³² (Fig. 1a). This highlights the intricate terminal branching patterns and widespread influences a single neuron can exhibit.

Fig. 1 |. Projectome mapping highlights diverse and variable long-range axonal projections across multiple brain regions.

a, Extensive branching of a single rat corticofugal neuron within the pyramidal tract originating from the cell soma within the lateral agranular cortex. The axon traverses nearly the entire rostro-caudal axis of the rat brain, from the frontal cortex to the brainstem, establishing collateral branches in multiple subcortical regions along its path. b, The majority of hippocampal output neurons project to multiple downstream brain regions. The distribution of projection target counts indicates ~70–90% of CA1, prosubiculum (ProS), subiculum (SUB) and SUB-related areas (SUBr) neurons project to two to ten brain regions. SUBr neurons predominantly innervate two targets, whereas SUB and ProS neurons most frequently contact three targets. Only CA1 neurons commonly have single-target projections. c, Individual hippocampal neurons, even of the same subtype, exhibit remarkably heterogeneous projection patterns. The diagram illustrates the axonal projection strengths of six example neurons (columns) from the hippocampal CA1 and CA2 regions to eight major downstream target areas. Axon projection strength, indicated by the size of the circles, is defined as total axon arbour length per target area. Part a adapted with permission from ref. 32, Society for Neuroscience. Parts b and c adapted with permission from in ref. 41, AAAS.

Recent advances in high-resolution brain-wide imaging platforms, such as MouseLight³⁴ and micro-optical sectioning techniques^35–37, have taken the reconstruction of single neurons to a new level. These technologies combine high-throughput imaging with sparse labelling and sophisticated neuron reconstruction software to generate large-scale ‘single-neuron projectomes’ that identify axonal projections throughout the entire brain. The databases that such studies contribute to, provide detailed projection diagrams of neurons in various brain regions in mice, including the thalamus, striatum, cortex, corpus callosum and hippocampus^15,38–41.

A key finding from these single-neuron projectome studies is the prevalence of long-range, multiregion connections in the nervous system. Far from being rare exceptions, these complex connections are a defining feature of neural networks. For instance, the mouse hippocampal projectome⁴¹ demonstrates that most hippocampal projection neurons from CA1, prosubiculum, subiculum and subiculum-related areas innervate two to ten downstream target regions, whereas fewer than 20% of neurons innervate a single target (Fig. 1b). Strikingly, over a third (34.6%) of hippocampal neurons project their axons to both brain hemispheres. This highlights the widespread influence of individual hippocampal neurons on various brain regions and probably expands their functions. Similar multiregion and bihemispheric projections are also observed in the mouse prefrontal cortex and sensorimotor cortex projectomes^38,40, indicating that single long-range connecting neurons innervating multiple brain areas is a common organizational principle in the brain⁴².

The projectome studies also reveal a surprising degree of connection variability within neuron types defined by morphology or molecular markers¹⁵. Although neurons of the same type generally exhibit similar overall projection patterns, there is a high degree of variability between individual neurons in terms of the subset of downstream targets to which they project (Fig. 1c). This significant individual variation suggests a high degree of flexibility in the circuit wiring that may contribute to functional plasticity beyond the developmental stage.

Recent work that aggregated the findings of previous tracing studies in the rat into a single mammalian inter-regional connectome⁴³ identified different modules within the brain that subserve instinctive and learned behaviours. Modules were defined as integrated structural networks associated with specific functions such as behavioural control and execution⁴³. This analysis also suggested the possibility of modular network compensation allowing modules to compensate for one another in the event of local damage or stress.

Nervous system connectomes at the electron microscopic level provide new insights

At the microscopic level, electron microscopic-reconstructed connectomes have been assembled from whole fruit fly nervous systems to cubic millimetre cortical columns in humans and have unveiled detailed patterns in how axonal branches form synapses onto their target neurons with nanoscale resolution^44–50. These cross-species connectomic data revealed a consistent and surprising degree of multiple synaptic connections on the same dendrite that could be crucial for maintaining functional robustness and resilience of the adult nervous system in the face of injury or disease. A single axon can innervate the same dendrite at multiple spines and many neurons form a disproportionately large number of connections with a small subset of target neurons. This connectivity pattern could play a complex computational role by summing synaptic inputs but may also create the capacity for synaptic rearrangements between the two neurons. For instance, an electron microscopic analysis of the mouse neocortex found that 11 out of 63 axons formed more than one excitatory synapse onto the apical dendrite of a single target neuron⁴⁴. Similar patterns have been observed in the human cerebral cortex, where some axons form more than 50 synapses onto a single dendrite⁴⁸ (Fig. 2a). Although the majority of connections were monosynaptic, with only about 4% of multisynaptic axonal inputs averaged across all target neurons, ~40% of the 2,743 neurons examined within a volume of brain tissue had at least one input that formed seven or more synapses onto the neuron⁴⁸. Thus, multisynaptic inputs to a given dendrite are consistently identified across different species (Fig. 2b).

Fig. 2 |. Examples of axonal complexity, connectivity and plasticity across species revealed by connectomic approaches.

a, An example of a strong, multisynaptic connection between neurons. A single human pyramidal neuron (blue) establishes 53 distinct synaptic contacts onto a neighbouring inhibitory interneuron (yellow). b, The axon branches from one Drosophila neuron (purple) innervating multiple downstream neurons with varying numbers of synaptic contacts, demonstrating the heterogeneity in the number of synaptic contacts made by a single neuron with different postsynaptic partners. c, Electron microscopic-reconstructed two-dimensional axonal diagrams display the complete axonal arbours of three pyramidal neurons from the human superior temporal gyrus (purple) and an interneuron from the rat entorhinal cortex (yellow). This illustration highlights the heterogeneity of axonal arbours, their complex branching and their uneven distribution of synapses even between neurons within the same region and neuronal type. Some branches are densely packed with synapses (red arrows), whereas others are only sparsely populated (blue arrows). Myelinated segments (green) on the rat interneuron appear on specific branches, suggesting a sophisticated mechanism for controlling signal conduction across different branches of the same axon. d, Axonal structural arrangements are subject to modification; axon terminals exhibit structural plasticity in the adult brain, including changes in the size, shape and number of presynaptic elements, as well as branch sprouting and pruning^65,264. e, Differential myelination patterns along axon branches may influence the conduction speed and timing of action potential (AP) propagation throughout different branches of the axon^66,68. Part a adapted with permission from ref. 48, AAAS. Part b adapted from ref. 265, CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Part c adapted from ref. 45, Springer Nature Limited, and adapted with permission from ref. 51, AAAS.

How synapses are distributed along axon branches represents another potential neuroplastic capacity. For example, in the rat medial entorhinal cortex, most axonal branches from both excitatory and inhibitory neurons are densely packed with synapses, connecting to multiple downstream neurons, whereas some axonal regions have very few or even no synapses⁴⁵ (Fig.2c). This uneven distribution of synaptic connections on axonal branches is also evident in connectome data in both Drosophila^47,49 and humans⁵¹ (Fig. 2c), indicating that this may be an evolutionarily conserved feature. This raises an intriguing possibility that these ‘silent’ or ‘dormant’ branches without densely packed synapses could act as reserves, ready to form new connections when learning new information or recovering from injury. It is also possible that if such regions previously formed synaptic connections during development that were then pruned, they may be ‘primed’ to form new connections in the future if the components to assemble a synapse remain available.

A recent study has revealed that this uneven distribution of connection strengths (how many synapses form between two neurons), known as heavy-tailed connectivity, is widespread across species⁵². Among the connected neuron pairs in the Drosophila central brain, 50% are weakly linked by a single synapse, whereas others are connected by more than a thousand synapses⁴⁷. Similarly, heavy-tailed connectivity is also present in the connectomes of the roundworm Caenorhabditis elegans^53,54 and the mouse retina⁵⁵. These observations challenge the idea that brain wiring is purely stochastic and based on the distance between axons and dendritic spines (that is, Peters’ rule⁵⁶). Instead, heavy-tailed and clustered connectivity might arise from a preferential growth of synapses and may be driven by activity-dependent mechanisms^52,57. Thus, the brain’s built-in capacity for neuroplasticity may not simply be a result of random growth during development but rather a product of dynamic processes that might continue to shape the brain throughout life.

An interesting rodent model for examining axon diversity and branch reorganization is the mesoaccumbens system. Exquisite single-axon labelling of neurons in the ventral tegmental area projecting to the nucleus accumbens has demonstrated important circuit differences among dopaminergic neuron types^58,59. Using anterograde single-cell viral labelling and reconstruction, the authors observed broadly different categories of dopaminergic projections to the cortex, basal forebrain, accumbens, caudate putamen and brainstem⁵⁸. A recent electron microscopic connectome study in the mesoaccumbens system reported large-scale axonal rearrangements of branches following cocaine exposure in mice⁶⁰. Large axonal swellings (or ‘bulbs’) were identified along axonal branches and at branch points, as well as increased axon branching in cocaine treated animals. Moreover, because this study was conducted at the electron microscopic level, the authors reconstructed varicosities and shaft synapses along axonal branches and quantified the number and size of vesicles within such varicosities. Even in controls, many varicosities contained no vesicles, indicating the potential to form new contacts with their targets.

Overall, electron microscopic connectome data have provided ‘ground truth’ evidence that extensive branching of long-range projecting axons is a key feature of the nervous system that is evolutionarily conserved. Moreover, the number and location of synaptic contacts between neurons differs between neurons and appears to be nonrandom. Finally, these datasets have identified that regions of the axon are devoid of synaptic connections, thus providing the potential to form new synaptic contacts during normal functions and to be deployed for brain repair.

Long-range axonal plasticity through branch activation and reorganization

The intricate structure of long-range axonal branches revealed by projectome and connectome projects suggests a fascinating dimension of neuroplasticity: even without large-scale rewiring, the brain may be able to functionally adapt and reorganize by combining stable long-range branches with dynamic local modifications. Indeed, advances in in vivo time-lapse two-photon imaging have revealed the remarkable structural plasticity of axon terminals in adult mouse brains^61,62. Presynaptic axonal boutons exhibit dynamic changes in size, shape and number^63,64. Axons can also sprout new branches or prune existing ones, forging new connections and eliminating old ones to reshape neural circuits^60,65. These plasticity mechanisms could significantly influence synaptic strength and the flow of information between long-range projection neurons and their downstream networks (Fig. 2d). Beyond structural adaptations, axons display functional plasticity in how they generate and conduct electrical signals⁶⁶. Changes in the expression and function of voltage-gated ion channels along the axon fine-tune excitability and conduction velocity⁶⁷. In addition, it has been shown that oligodendrocytes can dynamically regulate the differential myelination of axons and could thus directly affect the conduction speed and timing of action potentials across axonal branches⁶⁸ (Fig. 2e). Interestingly, individual axonal branches can even be differentially myelinated^45,69 (Fig. 2c), suggesting that the strength and speed of signals transmitted by different branches could differ, potentially resulting in differences in their functional connections and influence across local or distant networks. These forms of structural and functional plasticity could dynamically influence spike-timing-dependent plasticity (strengthening synaptic connections between neurons based on the timing of their action potentials⁷⁰) and network activity and thus expand the repertoire of outputs an axon may have. This interplay between long-range projections and local plasticity provides a potentially powerful mechanism for neural adaptation, as it could enable both learning and recovery from injury by dynamically adjusting network arrangements without requiring extensive long-distance axon regeneration (see ‘Neuroplasticity of long-range connections after stroke‘ below).

Axon branch formation and stabilization

During early development, the initial outgrowth of the main axon shaft is modulated and guided by molecular cues, a detailed discussion of which is beyond the scope of this review, but the interested reader is directed to other studies on these topics^71–73. Instead, we focus here on the mechanisms that mediate axonal branching and stabilization and, in particular, the role of spontaneous neuronal activity in these processes.

Distinct spatial and temporal patterns of spontaneous activity emerge at early stages of development and are thought to initiate circuit formation⁷⁴. During circuit development, patterned spontaneous activity occurs in several regions^74,75 (for example, neocortex^74,76, thalamus⁷⁷, hippocampus⁷⁸, auditory and visual systems^79–81 and spinal cord^82,83). During nascent circuit formation, axon branches randomly extend or retract to explore the environment⁷³. In amphibians and fish, newly formed synapses appear to drive both the initiation and stabilization of axonal branches⁸⁴. In the Xenopus visual system, synapses are strengthened when ipsilaterally projecting retinal ganglion cells (RGCs) receive synchronous, rather than asynchronous, stimulation from both eyes⁸⁵. However asynchronous stimulation caused enlarged axonal arbours and dynamic branching suggesting that axons were exploring the environment to identify synchronously activated synaptic partners⁸⁵. By contrast, axons extending into functionally inappropriate territories exhibit firing patterns distinct from those of nearby inputs and fail to maintain stable functional and structural contacts. Such inappropriate connections have been thought to be pruned back^86–88. However, recent evidence from the electron microscopic-level connectomes described above, suggests that some axonal branches without connections remain into adulthood (referred to as silent/dormant connections here; see the glossary for more information).

In addition to stabilizing axonal branches, early circuit activity can passively enable more molecular developmental processes to guide neural connections or direct and shape these connections more instructively⁸⁹. One prominent example of this occurs within the mouse visual system. Here, spontaneous retinal waves exhibit a distinct propagation direction bias during development that corresponds to optic flow generated during forward locomotion^74,89. Thus, before the onset of visual experience, spontaneous activity in the developing mouse retina is already structured in a manner that enables the development of circuits that mediate visual responses to environmental stimuli yet to be experienced.

The positions and strength of the presynaptic sites that are stabilized by patterned spontaneous activity further determine where additional axon branches are added or eliminated as axons continue to develop. For example, recent direct in vivo evidence elegantly demonstrated how endogenous patterns of spontaneous activity (in this case, retinal waves) instructed, through a Hebbian mechanism, the anatomical position of mouse retinocollicular synapses, which in turn determined where single axons were added or eliminated⁷⁹. Eliminated axon branches occured in arbour regions having low levels of local synchronization between single-axon firing and waves of spontaneous activity in neighbouring neurons, whereas the addition of axon branches occured in regions exhibiting high levels of local synchronization⁷⁹.

As neurons start to form synaptic connections and functional circuits begin to emerge during neurodevelopment, spontaneous activity becomes correlated across large groups of neighbouring cells⁷⁵. Through correlated neuronal activity across a developing network, short-range and long-range axonal branches are stabilized and facilitate the establishment and maintenance of functional connectivity within and across hemispheres (reviewed in ref. 90). As described in the following sections, connectivity remains malleable throughout the lifespan. This attribute permits functional reorganization in the context of normal brain function, such as during learning increasingly difficult tasks⁸⁴, as well as in response to injury, thereby contributing to behavioural recovery.

Evidence of axonal plasticity from developmental lesions

Experiments in a variety of species including rodents⁹¹, ferrets⁹², hamsters⁹³ and frogs⁹⁴ have demonstrated a high degree of axonal plasticity during development. In a series of elegant studies, Sur and colleagues manipulated the projections of RGC axons to innervate the medial geniculate (auditory) nucleus of the thalamus of the newborn ferret. This was principally achieved by lesioning the visual targets of these axons (the superior colliculus and/or the lateral geniculate nucleus) as well as parts of the visual cortex, allowing the RGC axons to expand within the medial geniculate. The medial geniculate cells that innervate the auditory cortex thus carried visual information from the rewired RGC axons⁹². These ‘rewired’ ferrets demonstrated visual responses in their auditory cortex, including crude orientation and direction selectivity maps⁹⁵ as well as visually evoked behavioural responses⁹⁶. Further studies investigated the RGC cell types innervating the medial geniculate and identified that these were principally W cells, which in their ‘rewired’ state carried visual information such as orientation tuning to the auditory cortex^96–99. Thus, if manipulated early enough in development, W-type RGC axons could generate new axonal connections in the medial geniculate nucleus. However, neurons within the deafferented inferior colliculus were no longer able to reinnervate the medial geniculate¹⁰⁰. Although these studies were aimed at identifying whether sensory specific regions of the thalamus and cortex could be reprogrammed to process information of another sensory modality, they also revealed information about long-range axonal plasticity during development. An overarching principle from these experiments is that some, but not all, neurons in the central nervous system exhibit developmental plasticity, and this can occur by initially forming extensive axonal branches that may or may not be pruned and could be repurposed to prioritize inputs from another sensory modality or, alternatively, through the ability to generate new axonal branches.

Evidence from developmental brain disorders

A remarkable example of diverse long-range axonal connections in the developing human brain occurs when the corpus callosum fails to develop typically, a condition known as corpus callosum dysgenesis (CCD)¹⁰¹. The corpus callosum is the largest fibre tract in the brain of placental mammals. It connects neurons between the cerebral hemispheres and is responsible for integrating information entering the brain in a lateralized manner and is engaged when the brain processes complex information¹⁰². During fetal development, the telencephalon undergoes cellular remodelling of the septal midline, which results in a fused substrate that allows axons to cross the midline¹⁰³. When interhemispheric remodelling fails to occur, callosal axons that approach the midline cannot cross the interhemispheric fissure and instead grow in large axonal ‘Probst bundles’ on either side of the midline¹⁰⁴. Probst bundles occur frequently in CCD and are one of the most robust examples of population level changes of ectopic axonal projections in the brain of mammals¹⁰³.

People who have undergone a callosotomy as adults to control intractable epilepsy¹⁰⁴ can experience a split-brain syndrome where they are unable to integrate information entering from each side of the body. Surprisingly, this phenotype is absent or much less pronounced in people with CCD or those who undergo callosotomy as children, indicating a remarkable degree of long-range axonal plasticity during a critical developmental period that appears to compensate for the lack of hemispheric connectivity¹⁰⁵. Another example of extreme functional plasticity in the developing brain occurs in children who have undergone a hemispherectomy to control severe seizures. Such patients recover from having a significant region (sometimes most) of one cerebral hemisphere removed and can develop remarkably normally, albeit with some deficits^106,107. The exact mechanisms by which functions can develop following this dramatic loss of tissue is currently not known but presumably involves a new organization of connections to provide such functional networks in the remaining and/or differently wired brain. Finally, individuals who become blind early in life display a remarkable reorganization of functional activation in cortical areas providing further evidence that repurposing of networks is possible during development¹⁰⁸.

Long-distance axonal projections in brain network communication

Across mammalian species, mapping stimulus-evoked or spontaneous activity reveals large-scale functional network architecture that corresponds to known structural connectivity patterns^109–115 and suggests a basic functional connectional topology that is conserved across rodents, primates and humans^116,117. For example, global patterns of brain activity measured using functional magnetic resonance imaging (fMRI)¹¹⁸ or optical techniques¹¹⁹ are temporally correlated within resting-state networks (RSNs) (Fig. 3). The functional significance of RSN topographies derives from the observation that they correspond to known sensory, motor and other higher-order cognitive functional systems^120,121. RSN topographies are usually determined by evaluating the zero-lag correlation of spontaneous activity between brain regions, which acts as a measure of their functional connectivity¹²² (Fig. 3b,c). In adults, functional connectivity structure is remarkably stable within individual subjects over long time scales^123,124, persists during wake and sleep¹²⁵ and is consistent across subjects at the population level¹²⁶. Metrics of functional connectivity are themselves powerful biomarkers of brain integrity. Functional connectivity strength in regions subserving a particular function (for example, sensorimotor) predicts future task performance^127–129, changes with learning^127,130 and is associated with behavioural changes or disease vulnerability^131–133.

Fig. 3 |. Long-range axons provide a substrate for functional brain network connectivity.

a, Axonal projections of three neurons in the mouse motor cortex exhibit highly branched, long-range and short-range connections. Some neurons with cell bodies in left motor cortex (red) or right motor cortex (blue) exhibit ipsilateral, cortico-cortical projections and subcortical projections, whereas other neurons of which one example is shown here (dark green) project to both hemispheres. Note the interhemispheric callosal projections of the single dark green axon and their highly complex branching structure within the contralateral hemisphere. The neuronal projection data are from the Mouse Light database (https://ml-neuronbrowser.janelia.org/). b, Axonal projection connectivity facilitates functional brain network communication. The same neurons in part a are overlaid on the mouse cortex and regions of interest displayed in the left motor (red circle) right motor (blue circle) and right visual (yellow circle) cortex. The long-range and short-range structural connectivity provides the substrate for functional connectivity within and across brain regions. For example, spontaneous activity in the left motor cortex exhibits a high degree of temporal synchrony with activity in the right motor cortex (strong correlation, SC), whereas neither region exhibits similar activity to that in right visual cortex (weak correlation, WC). These regional differences in synchrony are used to define global resting-state networks (see text). Part b adapted with permission from ref. 266, Elsevier.

Although communication within and across RSNs occurs through neuronal connections, the exact role of monosynaptic long-range projecting axons, versus short-range polysynaptic connectivity in orchestrating functional connectivity is unclear. Studies integrating connectivity information across scales have demonstrated that regional neuronal architecture at the microscale can be tuned to that region’s role in the global brain network^134–137. Further, the local complexity of neuronal branching in mammals is associated with the local number of macroscale white matter projections^135,136,138. Whether examined microscopically or macroscopically, connectivity is unevenly distributed in the brain^47,52,54. Short distance connections predominate^139–143 but brain regions/networks also exhibit a small proportion of long-range axonal connections^144,145. Decades of observations demonstrate a strong correspondence between the covariance structure of temporally evolving neural activity (that is, functional connectivity) and the properties of underlying structural connectivity^146–149. However, functional connectivity does not require direct (monosynaptic) connectivity. Homotopic functional connections between corresponding structures across hemispheres are typically the strongest subset of functional connections in the brain¹⁵⁰, but not all homotopic functional connections are supported by direct callosal projections¹⁵¹, and strong homotopic functional connections exist in individuals completely lacking callosal connections^{149,152–155}.

It is appealing to view long-range axonal projections from one region to another as a bridge linking brain areas¹¹⁷. This type of architecture could reduce topological distance between neuronal elements^117,156,157 or increase the capacity of interareal communication^117,158. Although local computation is performed by individual neural elements, their organization into circuits and networks supports a richer, more diverse functional repertoire^117,158. Thus, the computational capacity in which a neural element participates depends on the set of outgoing and incoming connections to and from other brain regions^159,160. Indeed, recent work suggests that the computational output of local circuits is signalled to other brain regions through long-range axonal projections⁴². Long-distance structural connections would allow distant brain areas to interact and, in the process, support more complex brain dynamics through novel configurations of inputs and outputs^117,159,161. Communication pathways formed by long-distance connections exhibit redundancies that may promote intranetwork and internetwork robustness, preserving functionality in cases where connectivity might be severed¹¹⁷. This same spatiostructural architecture, including the presence of long-distance connectivity, exists across Drosophila¹⁶², mice¹⁶³, macaques¹⁶⁴ and humans¹¹⁷, suggesting an evolutionarily conserved mechanism of interareal communication (and its preservation) across species. An alternative network strategy is for communication across polysynaptic chains of neurons. However, the unique roles of each strategy in brain network communication are not entirely clear. Studies examining axonal morphologies across species suggest that evolutionary pressures to balance the high metabolic cost of long-range connections with the computational efficiencies afforded by these structures might drive whether regions are mono-synaptically or poly-synaptically connected (Box 1).

Box 1 |. Long-range axons versus polysynaptic neuronal chains, balancing cost and efficiency.

Whether at the neuronal, regional or network level, connectivity tends to be concentrated on a small number of highly connected brain elements (that is, hubs)¹³⁵. Brain hubs exhibit dense interconnectivity amongst themselves, forming a ‘rich club’ that integrates information across anatomically distributed (spatially distant) neural systems while conferring a level of network resilience^116,142. Rich-club brain architecture appears to be a fundamental organizational property of the brain and is observed in Caenorhabditis elegans, flies, rodents, macaques and humans^{50,116,267,268}. Further, alterations to this ‘rich club backbone’¹¹⁶ have significant impacts on the brain’s functional integrity. For example, damage to these regions (for example, following focal ischaemia²⁶⁹) produces severe and more widespread neurological dysfunction compared with lesions to other brain areas²⁷⁰. Thus, mechanisms underlying brain organization must balance communication efficiency with high wiring cost and potential effects for vulnerability and resilience.

Local and distant communication between brain hubs can be achieved either through long-range axons or polysynaptic chains of neurons. What are the considerations that determine how networks are wired? Several studies across species and scales, comparing neuronal and network features, suggest that evolutionary pressures balance a trade-off between wiring costs^139,140 and computational efficiency¹⁴². Long axons are developmentally and metabolically expensive, requiring complex developmental programmes to wire appropriately and high energy to maintain^271–273. However, long axons are more efficient at communicating over long distances compared with polysynaptic neurons: action potentials travel along axons without synaptic delays²⁷⁴ (~1 ms per synapse), fewer synapses mean more signal fidelity due to reduced synaptic noise²⁷⁵ and long axons support small-world network topology (short path length and high clustering)¹⁴². In addition to the lower wiring costs of polysynaptic neuronal chains, there are advantages of greater computational flexibility — each intermediate neuron permits the integration of inputs from other local neurons that help modulate the transmission of signals and facilitate learning and adaptation²⁷⁶. The dichotomy of long axons versus polysynaptic chains are two extremes in a spectrum of possibilities, and the assumptions in modelling their costs/efficiencies are imperfect. For example, as described here, the morphology of long axons can be very complex: they have complex branched morphologies^45,69 capable of innervating multiple targets, they can be differentially myelinated, and some receive synapses whereas others do not^45,47,49,51. Thus, the computational power of long axons might rival that of polysynaptic neuronal chains while maintaining advantages of speed and signal fidelity. Indeed, higher species (especially mammals and primates) tend to have more long-range axons, with more complicated branching and myelination compared with simpler organisms^51,277. An important corollary to the higher metabolic cost to maintain long axons is their greater vulnerability to injury. Long axons have immense energy demands, making them metabolically fragile²⁷³. Disruption of bioenergetics, due to mitochondrial dysfunction or interference with glial support, can lead to axonopathy²⁷⁸. Thus, long axons may be more vulnerable to ageing²⁷⁹, neurodegenerative processes²⁸⁰ and chronic ischaemia²⁸¹, requiring processes to reduce such impacts (for example, by eliminating some branches while preserving the rest of the axonal arbour).

Functional network plasticity

Axonal plasticity might be a key feature of brain network reorganization. The degree to which brain regions are functionally connected can be modified by experience and appear to reflect the effects of Hebbian neural plasticity. For example, during visual development in mice, monocular deprivation results in pruning of thalamocortical presynaptic terminals and postsynaptic dendritic spines in deprived visual cortex, whereas thalamocortical projections increase their arborization and synaptic territory in non-deprived visual cortex^165–167. These deprivation-induced changes in synaptic pruning and circuit refinement are associated with reduced homotopic functional connectivity within the visual network¹⁶⁸. By contrast, visual deprivation with binocular lid suturing not only increases homotopic visual functional connectivity but also increases anticorrelation between the visual network and several networks outside of the visual cortex¹⁶⁸. These observations suggest that sensory deprivation within a particular system can induce long-distance, cross-modal plasticity across RSNs. The deployment of additional axonal branches including those that contain silent and/or dormant capacity for additional or altered synaptic connections may contribute to these forms of mesoscale plasticity (this concept is illustrated in figure 5 of ref. 61) and may parallel the developmental plasticity observed following lesions described above.

Experience-driven changes in functional connectivity structure can occur in the adult brain as well. For example, functional brain networks appear to reconfigure dynamically during motor learning, starting with distributed network activity that evolves to more efficient modular architecture after learning¹⁶⁹. Similar findings were demonstrated in working memory training¹⁷⁰. More dramatic changes were observed in studies with dominant arm immobilization (via casting) for 2 weeks. Cortical and cerebellar regions controlling the disused extremity functionally disconnect (that is, exhibit reduced functional connectivity strength) within 48 h from the rest of the somatomotor system, whereas internal connectivity is maintained within the disused subcircuit¹⁷¹. These functional connectivity changes are also reversible, as cast removal causes disconnected RSNs to reconnect. An unexpected observation during the casting period was the presence of spontaneous large amplitude blood oxygenation level-dependent signals (termed ‘plasticity pulses’) within patterns of resting-state activity that propagated through the disused somatomotor sub-circuit¹⁷¹. The role, if any, of these pulses is unclear. However, it has been proposed¹⁷² that disuse-driven spontaneous activity pulses might help preserve functionally disconnected subcircuits (at least for a limited time) in a manner similar to emergent spontaneous activity during nascent circuit development and axonal exploration/branching.

Whether at the cellular or population level, spontaneous neuronal activity is regulated by the constraints imposed by anatomy (for example, axons) and constraints imposed by the context and history of neural events that sculpt functional networks (for example, experience¹⁷¹). Spontaneous activity at both the micro and macroscopic scales might therefore reflect the results of long-term (and long range) Hebbian-based functional reorganization but may also contain information regarding processes associated with neuroplasticity itself (plasticity pulses). From these observations, it is reasonable to posit that large-scale measurements of ongoing, intrinsic brain activity (for example, via blood oxygenation level-dependent signals, fMRI or fluctuations in voltage or calcium via wide-field optical imaging (WFOI)) can provide information related to homeostatic and consolidative processes associated with functional network plasticity¹⁷². To the extent that this activity per se serves a function, it is perhaps to maintain the brain’s functional organization over time (network stability) while also permitting experience-dependent remodelling/reconfiguration (neuroplasticity). Spontaneous activity might also represent the brain’s ongoing generative and predictive models of the environment and self, continuously anticipating sensory, motor and cognitive outcomes (on these topics, the interested reader is directed to two excellent opinion pieces^172,173).

Patterns of activity observed in specific brain regions are regulated by the cell types within those regions and their respective connections^174–181. These relationships are also species specific as different types of neuronal dynamics are linked to specific structural network topologies across organisms^182–185. A question remains as to whether, after network reorganization, a region’s existing structural connections are sufficient to enable patterns of activity to support a function that would not otherwise be subserved or prioritized by that region. Such a capacity would suggest that functions are not merely localized to specific brain regions that are ‘predefined’ during development but, instead, that any region can support a function provided the proper patterns of activity can propagate in that region. This hypothesis is supported by the functional rerouting experiments of Sur and colleagues¹⁰⁰ described above, as well as decades of work examining the effects of sensory deprivation and cross-modal plasticity in multiple species^186–189. Another extreme but illustrative example involves a child who sustained large, bilateral perinatal strokes 3 weeks postnatally but nevertheless experienced typical neurodevelopment¹⁹⁰. These endogenous ‘fail-safe’ mechanisms for supporting function in the event of catastrophe appear to be laid out during development. For example, L4 neurons extend transient callosal axons that are subsequently eliminated during typical sensory experience but can remain callosal, maturing and forming functional connections in response to atypical changes in thalamic input¹⁹¹. Thus, in some circumstances, long-range and short-range axons provide a means of structural and functional plasticity that endows the cortex with the ability to generate alternative circuits in non-canonical circumstances¹⁹¹.

Neuroplasticity of long-range connections after stroke

An extreme test of neuroplasticity is how well the brain recovers after brain injury, especially in the elderly adult where neuroplasticity may be limited. The fact that most patients who suffer acute brain injury experience some degree of recovery suggests the existence of endogenous brain repair mechanisms. Indeed, the neuroplastic mechanisms described above are probably drivers of recovery. However, for complete recovery to occur, it stands to reason that not only should local circuits (for example, sensorimotor circuits) be repaired, but the reintegration of these ‘repaired’ local circuits into global brain networks is needed to permit recovery across higher levels of integrated processing (that is, leveraging the computational capacity afforded by richer network connectivity). It is the reintegration of repaired circuits into global networks that probably involves long-range axonal plasticity. Thus, full recovery does not only involve recovery of strength to a paralysed limb, for example, but the reintegration of that limb to the full spectrum of brain processes involved in motor planning, smooth volitional movement and motor programmes reactive to environmental stimuli.

The most common cause of acute brain injury in humans is ischaemic stroke — brain injury caused by the lack of blood flow to a region of the brain. Focal ischaemia results in permanent loss of brain tissue. Lesions cause direct structural damage that disconnects local brain circuitry and indirectly disrupts global RSNs^192–195 (Fig. 4a). Brain dysfunction in brain regions remote from the lesion (termed ‘diaschisis’¹⁹⁶ and, more specifically, as ‘connectomal diaschisis’ within the context of widespread changes in functional connectivity^194,195) can result in neurological deficits spanning multiple domains. Recovery from stroke is associated with neuronal and synaptic plasticity — the formation of new structural and functional contacts that take over the role of those lost due to infarction^197,198. Evidence suggests that these changes probably occur locally (within peri-infarct tissue) and also at distant sites where connectivity changes might occur (for example, with the contralesional hemisphere). Although a direct causal link between axonal plasticity, local/global circuit reconnection and neurological recovery has not been firmly established, several lines of evidence (such as those described above and below), point towards the potential recruitment (and redeployment) of local and distal axonal branches being critical mediators of these processes.

Fig. 4 |. Poststroke plasticity facilitated by short axonal branch extension.

a, Top left: a schematic of healthy organization of somatosensory forepaw circuitry in the rodent involving thalamocortical (purple) and cortico-cortical (blue) connectivity. Middle: stimulation (lightning bolt) of the right forepaw results in activation of the receptive field in somatosensory cortex and perception. Bottom: intact functional connectivity between left and right somatosensory forepaw cortex. b, Acute ischaemia (red shaded area, top) causes direct ablation of the forepaw somatosensory cortex, and structural damage that disconnects both the local thalamocortical circuit (purple shaded projection) and interhemispheric connections within the somatosensory network (red dashed projection, top, and dashed arrow, bottom). Middle: forepaw stimulation does not result in somatosensory activation, and therefore, there is no perception. c, Top: recovery is associated with repair of the thalamocortical circuit due to either branching axons or the strengthening of pre-existing connections (purple axons, top), as well as the repair of interhemispheric connections within the somatosensory network (green axon branch). Mature infarct is shown in black. Top and middle: stimulation of the forepaw results in activation of the cortex adjacent to the infarcted area (green area). Bottom: new patterns of interhemispheric connectivity return, albeit weaker, between left somatosensory forepaw cortex and remapped regions of perilesional cortex (green area).

Local circuit repair

In the weeks to months following stroke, functional neuroimaging (fMRI and WFOI) studies demonstrate that local representations of function lost to infarction ‘remap’ to peri-infarct cortex¹⁹⁹ at distances of up to several millimetres away. For example, human and animal studies have shown that peri-infarct regions become more responsive to stimulation of somatomotor regions with which they are not generally associated (for example, during tactile stimulation of the hand in humans²⁰⁰ or via electrical and/or mechanical stimulation of the forelimb in rodents¹⁹⁷) (Fig. 4b). In rodents, remodelling of local circuitry in peri-infarct cortex (that is, remapping) correlates temporally with behavioural recovery^{197,201–203}, and behavioural deficits can be reinstated following ablation of these remapped regions²⁰⁴. Thus, remapping is the spatial representation of a repaired circuit — effectively repurposing viable, usually adjacent, brain circuits to take over function lost to the stroke. In human and animal studies, perilesional remapping is more prominent and tightly focused in subjects exhibiting more complete recovery; those with poorer recovery exhibit weaker or diffuse activation patterns that can involve both hemispheres^205–209, indicating that how this tissue remodels is important. Although these collective observations indicate that functionally remodelled peri-infarct cortex takes over the function of brain regions lost to stroke^{197,200,201,210,211}, how structural changes in the brain facilitate this process is less clear.

At the molecular level, local circuit disconnection within the infarct manifests within hours of an ischaemic insult in rodent stroke models, as a marked decrease in dendritic spine density^212–215. Over the following days to weeks, a molecular growth programme is initiated that supports the formation of new local circuitry including several genes associated with plasticity^3,216. In this environment of heightened plasticity cues, spine turnover within perilesional tissue increases substantially, with spine generation outpacing spine removal, especially in regions of remapping¹⁶⁸. This process results in spine accumulation and gradual recovery of perilesional spine density²¹⁵, observations that correlate with behavioural recovery²¹⁷. Although synaptic changes reflected by spine density appear to be important for recovery of function, enhanced perilesional spine loss and/or generation also suggest that old connections are replaced with new ones. This process appears crucial for changes in receptivity in perilesional neurons. For example, in surviving peri-infarct areas during early remapping periods, normally highly limb-selective neurons exhibit remarkable flexibility and begin to process sensory stimuli from multiple limbs²¹⁸. Over several weeks, neurons within remapped regions develop a stronger response preference to a particular limb. These cellular-level changes manifest at the meso and/or macroscopic level in sensory maps of the affected limb²¹⁸. For example, using WFOI, remapping of forepaw cortex in rodent stroke models following photothrombosis requires activity-dependent synaptogenesis involving the immediate early gene aArc (activity-regulated cytoskeletal associated protein). Knocking out the Arc gene prevented remapping and arrested behavioural recovery¹⁶⁸, demonstrating the importance of synaptic plasticity in recovery after brain injury.

Concurrent with synaptic plasticity, local circuit reconfiguration is probably facilitated by axonal plasticity — either the sprouting of new axonal branches or the reconnection of those already present (that is, silent/dormant or en passant axons). Evidence that axonal plasticity plays a role in local circuit brain repair comes from the peri-infarct transcriptome in rodent models demonstrating expression of several axonal-growth-promoting genes including those involved in axon guidance and regeneration (BASP1, CAP23, GDF10, ephrin A5, CREB and so on)²¹⁶. The fact that many of these axon-growth-promoting genes were expressed in localized regions of the contralesional hemisphere suggests that axonal plasticity occurs even at sites distant from the focal injury. Similarly, growth-associated protein 43 (GAP-43) is robustly upregulated following ischaemic stroke in rodents and non-human primates^219–221. GAP-43 is a cytosolic protein selectively trafficked to the growth cones in sprouting axons that facilitates growth cone motility, navigation and neuronal branching^222–226. Before the appearance of mature synapses, robust GAP-43 expression is observed throughout cortical and subcortical brain regions in the peri-infarct region²²⁷. Surprisingly, increased expression of GAP-43 (albeit in a smaller area) is observed in a mirror region of the opposite (uninjured) hemisphere, suggesting that axonal branching from long axons occurs at distant sites within the same somatomotor network. As mature synaptic markers appear, GAP-43 levels decline²²⁰, suggesting GAP-43 mediated axonal sprouting may be a prerequisite for synapse formation and recovery. The vector trajectory of axonal projections/branches after stroke is directed towards perilesional tissue^3,216,228, suggesting that damaged neurons may seek new postsynaptic targets in remapped cortex. Interventional strategies designed to promote axonal sprouting and stabilize new axon formation improve motor function after stroke in rodents^229–232. In the contralateral hemisphere, GAP-43 expression might initiate new growth cones in callosally projecting axons terminating in perilesional tissue²³³. As outlined in the next section, this evidence of local and distant axonal plasticity after stroke might not only permit local circuit repair but also facilitate reconnection of global brain networks to support recovery of function.

Based on the above mapping data and transcriptional profiles, there is substantial evidence supporting the concept that axonal sprouting plays an important role in poststroke circuit reorganization and remapping. However, this view has been challenged²³⁴. Evidence of remapping largely comes from mesoscopic or macroscopic functional neuroimaging methods, which collect ensemble activity integrated over large populations of cells. At the microscopic level, evaluating individual neuronal responses might also erroneously exaggerate map changes. For example, neuronal responses might be preferentially tuned to a specific feature whereas the entire response profile could be broader and reflect multiple inputs, a subset of which encode salience. This ‘winner-takes-all’ assignment of function fails to recognize that neurons already have the capacity to process non-preferred stimulus features, which can be shifted and strengthened during ‘remapping’. It has therefore been suggested that remapping may not be due to axonal sprouting but rather result from the strengthening of pre-existing connections²³⁴. Considering evidence described above, brain remodelling after stroke probably requires utilizing pre-existing connections, silent and/or dormant connections, as well as growing new, short-range axonal branches depending on the size and location of injury.

Long-distance network repair

Examining distributed patterns of synchronized, resting-state activity throughout the brain reveals that global patterns of functional connectivity^235,236 are altered after stroke (Fig. 4c). Shortly after focal ischaemia in humans and mice, disruption of interhemispheric, homotopic functional connectivity predicts poor motor and attentional recovery^237–239. Over the course of the following months in humans — or weeks to months in animal models — restoration of homotopic functional connectivity correlates with improved behavioural performance^{195,240–243}. These processes occur in parallel with remapping; brain networks returning towards more normal patterns of intrinsic organization after stroke (that is, restored functional connectivity) (Fig. 4c, bottom) also exhibit more normalized patterns of activation (that is, stimulus-evoked activity during task²⁰⁵). Further, and consistently reported across mammalian species, is the observation that better functional recovery after stroke is associated with a return to balanced intrinsic (for example, resting-state) activity. Patients with impaired hand movement experience abnormally high interhemispheric inhibition. This arises from neurons in the contralesional motor cortex projecting to perilesional motor cortex, with the degree of inhibition increasing with deficit severity^244,245. Similarly, abnormal intrinsic brain rhythms that give rise to abnormally strong patterns of task activation result in reduced variability of motor patterns after stroke²⁴⁶. Such abnormal brain activity appears to reflect maladaptive plasticity — an attempt to link disconnected regions either by increasing neural activity upstream of the lesion or by rerouting activity through accessory regions^246,247.

A direct causal link between axonal plasticity, local and/or global circuit reconnection, and changes in behavioural function has not been established, but a mounting body of evidence suggests that this occurs. The ‘structural growth programme’ initiated shortly after focal ischaemia^3,216 can be modified by modulating genes and associated pathways supporting synaptic plasticity, axon guidance and growth signalling more generally^{229–231,248,249}. These manipulations subsequently result in differential sprouting responses across brain regions in a manner that depends on the intervention but correlates with somatomotor function. For example, in the healthy mouse brain CCR5 transcripts are undetectable in cortical neurons but differentially upregulated after stroke²³⁰. Neuronal knockdown of CCR5 in mouse premotor cortex after stroke leads to early recovery of motor control²³⁰. These behavioural improvements are associated with the preservation of perilesional dendritic spines, contralesional axonal sprouting, and an increase in cortical axonal projections from the ipsilesional hemisphere to the contralateral hemisphere²³⁰. Contralateral stimulation also markedly alters several genes (for example, RTN4R, BASP1, GDF10) coding for proteins involved in axonal sprouting and growth²⁰⁵. Whether locally remapped circuits reintegrate into global networks to facilitate recovery is an ongoing question. Circumstantial evidence suggests this to be true²⁰⁵. Structural changes probably act in concert with, and are influenced by, other mechanisms supporting plasticity^250,251. Axonal sprouting, branching and the potential deployment of previously silent/dormant axons and their subsequent connectivity probably require coordination amongst several molecular cues to ensure proper synaptic targeting to facilitate reintegration between the peri-infarct and remote brain areas after stroke^192,194,195.

Conclusion and perspectives

New evidence from sparse labelling and electron microscopic connectome data have revealed the extraordinary reach and complex branched morphology of long-range axons in species from flies to humans. This work also reveals that axonal branches can be differentially myelinated, and, together with work demonstrating the capacity for axons to differentially propagate axon potentials along specific branches, suggests additional ways that neurons could influence multiple downstream targets.

Long-range axonal connectivity and branching of single axons into multiple areas suggests a structural link between submicrometre-level synaptic plasticity within local circuits, single neuron plasticity and regional plasticity at a network level. This also suggests ways that additional areas could be recruited during normal brain function as task complexities increase. For example, it is tempting to speculate that if information processing could be mediated by a subset of axonal branches under certain tasks then additional branches of the axon could help to recruit additional areas for information processing during more complex tasks.

Such axonal branch complexities also suggest that there may be differential responses to lesions such as those during development or following stroke or brain injury in adult brains. The recovery of function is probably the result of a hierarchy of repair mechanisms at both local circuit (involving synaptic plasticity) and global network levels (involving long-range axonal plasticity). The reason that some people suffer devastating loss of function without recovery whereas others can have remarkable recovery could be reflected in how well axons can adapt to integrate and recruit other areas to accommodate new functions.

Understanding the brain’s remarkable ability to adapt and change as circumstances or the environment are altered is one of the most fascinating questions in neuroscience. Here, we postulate that understanding the scope and breadth of possibilities of axonal functions and how axonal branch complexities arise and are maintained or sculpted throughout life will have significant impacts on our understanding of brain development, function and our ability to promote plastic changes in response to injury or disease. The findings from electron microscopic connectome data showing that there are regions of the axon and axonal branches that remain, even though they have few synaptic contacts, is intriguing. Previous work has shown that axons that fail to make synaptic connections are eliminated²⁵²; therefore, how these silent/dormant axonal branches remain is yet to be understood. One possibility is that axonal branch pruning occurs only during development when axons fail to make any synaptic contacts. Those that do make synaptic contacts might subsequently lose synaptic contacts if they are not used, thus giving rise to silent/dormant axon branches as described here. Although this hypothesis has yet to be tested, the latest advancement of electron microscopic-based^253–255 and light-microscopy-based²⁵⁶ connectomic techniques now provide the capability to reconstruct and directly compare synapse distribution on the axonal branches of same neuron types in animals under different experimental conditions. Careful mapping of axonal trajectories in vivo is also required to advance this field, particularly as it is unlikely that many of the processes that regulate axonal branching into different regions of the brain may not be amenable to studies in vitro. Moreover, differential action potential propagation has only been observed in vivo⁶⁶. Therefore, new methods to observe and study axonal growth, branching and myelination, as well as action potential propagation in vivo, are required.

Other ways of furthering our knowledge of neuroplasticity processes required for maintaining brain function throughout life could capitalize on multimodal, cross-scale examinations of connectivity. Light sheet microscopy²⁵⁷ and wide-field two-photon imaging^258–260 can be leveraged to map the structural and functional connectivity of specific neuronal populations across multiple brain regions⁷⁹. Such technological developments are vital to understand the role of multi-area axonal branches in mediating activity and the recruitment of cortical areas during cognition and behaviour. Construction of brain-wide gene expression atlases allows spatial variations in gene expression to be correlated with distributed properties of connectome structure and function²⁶¹. For example, transcriptomes of well-connected hubs are characterized by genes that support oxidative metabolism and mitochondrial function, whereas non-hub brain regions express genes regulating neuronal development and synaptic signalling²⁶¹. That similar observations extend across mice²⁶², macaques²⁶¹ and humans^261,263 implies a cross-species preservation of broad spatial transcriptional gradients that track regional variations in neuronal microcircuitry and inter-regional connectivity.

To understand how long-range axonal complexity contributes to network repair after focal brain injury, these methods will need to be explored in animal models. For example, combining high-resolution connectomic techniques with high-resolution spatial transcriptomic techniques following experimental stroke might reveal specific pathways that shed light on molecular mechanisms underlying long-range axonal plasticity. Moreover, genetic manipulation of these pathways following experimental stroke could be used to investigate their role in network repair. In human stroke, blood biomarkers of long-range axon repair will need to be developed, which can then be measured during stroke recovery. These blood biomarkers can then be correlated with poststroke changes in functional connectivity to examine their role in brain network repair. Such studies may identify potential drug targets to enhance repair and recovery after stroke.

Glossary

Axonal plasticity

Any change to an axon; including an increase or decrease in activity, activation or loss/pruning of a specific branch, growth of a new branch (probably over a few micrometres) or synapse, a significant change in axonal trajectory from neurotypical (usually during development).

Connectome

A comprehensive, ultradetailed wiring diagram of each neuron and its individual synapses within a circuit or an entire nervous system, typically reconstructed from electron microscopy data. Although the concept of ‘connectome’ extends to macroscale structural and functional connectivity (for example, the human connectome), it specifically refers to the nanoscale diagrams in this review.

Global networks

Global networks comprising multiple interconnected local circuits, often across distant regions of the brain working together to perform more complex cognitive functions and behaviours.

Local circuits

Local circuits are populations of interconnected colocalized neurons that carry out specific, localized functions.

Long-range axon

Axons that project out of a given cortical functional domain (for example, a primary sensory area) or nuclei to another area in the same or opposite hemisphere or in a descending/ascending manner.

Peri-infarct region

The area of tissue surrounding a completed stroke (after the phase of active ischaemia) where the region of damage has a stabilized and is no longer expanding.

Projectome

A detailed map with single-neuron resolution of the axonal projections linking distinct brain regions, which reveals the target patterns of different neuron types and the large-scale architecture of the brain’s network.

Remapping

The brain’s ability to reorganize or reprioritize its neural connections and function in response to injury. Remapping describes a phenomenon but not necessarily a mechanism.

Silent or dormant long-range axonal branches

Axonal branches or regions of the axon shaft lacking, or with low density of, synaptic connections at a given time. These regions of the axon may or may not have previously had synaptic boutons. Previously active regions of the axon may be better primed for neuroplasticity.

Related links

MouseLight: https://ml-neuronbrowser.janelia.org/

NeuroMorpho.org: https://neuromorpho.org/