The functional organization of axonal mRNA transport and translation

Abstract

Axons extend for tremendously long distances from the neuronal soma and make use of localized mRNA translation to rapidly respond to different extracellular stimuli and physiological states. The locally synthesized proteins support many different functions in both developing and mature axons, raising questions about the mechanisms by which local translation is organized to ensure the appropriate responses to specific stimuli. Publications over the past few years have uncovered new mechanisms for regulating the axonal transport and localized translation of mRNAs, with several of these pathways converging on the regulation of cohorts of functionally related mRNAs — known as RNA regulons — that drive axon growth, axon guidance, injury responses, axon survival and even axonal mitochondrial function. Recent advances point to these different regulatory pathways as organizing platforms that allow the axon’s proteome to be modulated to meet its physiological needs.

Localization and translation of mRNAs in subcellular regions is now recognized as a critical mechanism through which polarized cells establish and maintain the specialized domains that are needed for their polarized functions. This is especially the case for neurons, which have long cytoplasmic projections (dendrites and axons) that are needed for communication between neurons (that is, for neural circuit function) and with target tissues. Formative ultrastructural analyses of the mature rodent hippocampus advanced the idea that translation of dendritic mRNAs promotes synaptic function in this subcellular compartment¹. This local translation provides the dendrite with the ability to rapidly respond to trans-synaptic stimuli that contribute to synaptic plasticity². Recent reviews have nicely outlined the various functions served by and the mechanisms regulating dendritic protein synthesis^3,4.

Dendrites extend only a few millimetres at most from the neuronal soma, whereas an axon can extend anywhere from a few centimetres in rodents to a metre or more in humans. Thus, it was appealing to speculate that axons might also contain mRNAs and be capable of locally synthesizing proteins in order to rapidly change their localized protein composition (that is, the local proteome) in the same manner as dendrites. Although early ultrastructural work failed to detect ribosomes in axons of the mature CNS¹, a handful of laboratories reported the identification of mRNAs and ribosomes in axons of different organisms and neuron types between the 1960s and the late 1990s^5–10 (reviewed in REF.¹¹). Advances in the technology for the detection of mRNAs and newly synthesized proteins through molecular and cell biology approaches began to emerge around 2000, resulting in several publications showing that mRNA translation not only occurs in axons but also provides a variety of functional benefits^12–15 (reviewed in REF.¹⁶). For example, since a single mRNA can be used to generate many copies of a protein, the transport of mRNAs to the axon provides the neuron with the ‘locally sourced’ capacity for proteome modifications, allowing the axon to spatially and temporally regulate the introduction of new proteins. Local translation also affords a means to restrict the activity of proteins to the site at which they are needed by providing an ‘on demand’ mechanism that regulates when and exactly where a protein is generated. Thus, localized protein synthesis allows a subcellular region — such as an axon or even a subaxonal domain — to finely tune gene expression over both space and time¹⁷.

Ribosomes.

RNA–protein complexes composed of 60S and 40S subunits that are responsible for translating mRNAs into protein.

Although dendritic protein synthesis was accepted by the scientific community much earlier than axonal protein synthesis¹⁸, the number of known functions of axonally synthesized proteins may soon exceed those that we know of for dendritically synthesized proteins. The distances that axons extend from the cell body make them more amenable to physical separation from the cell body than dendrites, meaning that knowledge of axonal mRNA populations and the functions that their encoded proteins serve is now advancing at a rapid pace. RNA profiling studies have uncovered hundreds to thousands of mRNAs in distal axons and have demonstrated some differences between axonal mRNA profiles across different neuronal populations, physiological conditions and developmental growth states (reviewed in REF.¹⁹). Undoubtedly, many more functions of axonally synthesized proteins will be uncovered as more specific experimental tools are developed.

RNA profiling.

The measurement of the identities and abundances of RNAs across different cell or tissue populations or subcellular sites and in different physiological conditions or growth states.

Similarly to mRNA translation in dendrites and non-neuronal cells, where and when axonal mRNAs are translated is determined by the interactions of those mRNAs with a combination of RNA-binding proteins (RBPs) and the subsequent modifications of those interactions. mRNAs in the neuronal soma interact with RBPs to assemble into non-membrane-bound ribonucleoproteins (RNPs). Recent publications have provided insights into the nature of these RNPs and have uncovered novel mechanisms for the delivery of mRNAs into axons and the translational regulation of these localized mRNAs. In this Review, we will try to place these mRNA transport and translation mechanisms into the context of the functional outcomes that are linked to the axonally synthesized proteins. RNA regulons are cohorts of mRNAs that are co-regulated by individual RBPs and encode proteins with complementary or synergistic cellular functions²⁰, providing a mechanism through which the neuron can orchestrate the localization and translation of hundreds of different functionally related axonal mRNAs over space and time. Here we argue that RNA regulons extend beyond the initial description of an mRNA family interacting with a single RBP to include groups of mRNAs that interact with combinations of RBPs, subcellular organelles, organelle-like structures and receptors. Intracellular signalling pathways layer onto these RNA regulons to bring further temporal and spatial specificity to axonal mRNA translation.

RNA-binding proteins.

(RBPs). Proteins that can bind RNA via RNA structures (‘motifs’) and that are involved in RNA processing, trafficking, stability and translational regulation.

Ribonucleoproteins.

(RNPs). Complexes of RNA-binding proteins and RNAs that associate, often in the form of a granule or organelle-like structures.

Sorting mRNAs for axon localization

A central theme that has emerged from studies examining subcellular RNA localization, both in neurons and in other systems, is the idea that the cell recognizes which mRNAs to localize on the basis of sequence structures within the mRNAs themselves. These cis elements or motifs, typically (but not exclusively) found in 3′ untranslated regions (UTRs) of the mRNAs²¹, are recognized by RBPs. Together the mRNA and RBP eventually form an RNP that is capable of localization, whether the RNP is linked directly to motor proteins or localizes to a subcellular region by binding to vesicles trafficking on motor proteins (FIG. 1).

Fig. 1 |. ribonucleoprotein (rNP)-coupled transport of mRNAs into axons.

a | Schematic of a neuron showing the anterograde transport of RNPs that are engaged with motor proteins directly (known as transport RNPs) or indirectly (through binding to vesicles) for microtubule-based transport along the axon shaft. Although it is not clear whether transport RNPs also move in a retrograde direction, retrograde vesicle-based RNA transport has been reported recently⁷⁷. b | Some RNA-binding proteins (RBPs) initially interact with newly transcribed RNAs in the nucleus to form an RNP. Drawing from studies in other systems, these nuclear RNPs are thought to exit the nucleus through the nuclear pore, after which they are remodelled through the addition of some new RBPs and the removal of others, with some of the released RBPs returning to the nucleus^27,33. c,d | RNPs can bind kinesin motor proteins as a ‘transport RNP’ (panel c) or can bind to vesicles (endosomes or lysosomes) through molecular tether proteins (panel d). RNPs may also utilize adaptor proteins for binding to kinesin (not shown). Pre-microRNAs (pre-miRNAs) and mature miRNAs have also been shown to associate with vesicles for transport into axons⁸⁴. In addition, the RNA-induced silencing complex (RISC), which is needed for miRNA processing, has been shown to associate with vesicles^76,116, although it is not clear whether it functions directly on or adjacent to the vesicle for processing co-trafficked pre-miRNAs or whether it acts on different pre-miRNAs that already resided in the axon.

Motor proteins.

Proteins that bind to microtubules or microfilaments and transport cargos from one site in a cell to another.

mRNAs are generally thought to exit the nucleus through the nuclear pore complex in the form of an RNP that can include exon junction complex (EJC) components²² (FIG. 1b). EJC proteins bound to 3′ UTR exon junctions are used by the cell to identify mRNAs that will be subjected to nonsense-mediated decay. Given that nonsense-mediated decay has been demonstrated for both axonal and dendritic mRNAs, this suggests that the neuronal mRNAs that are trafficked to subcellular sites can carry along EJC-associated proteins from the nucleus to the axon²³. As discussed below, several RBPs thought to be predominantly nuclear have been observed in rat axons and shown to bind to axonal localization motifs²⁴. It should be noted that some RNPs may be too large to transit through the nuclear pore complex; indeed, an alteRNAtive mechanism of nuclear export known as nuclear envelope budding was recently suggested²⁵.

Nuclear pore complex.

A protein complex that transects the nuclear membrane and controls the exchange of many components between the nucleus and the cytoplasm.

Nonsense-mediated decay.

A mechanism for the selective degradation of mRNAs that contain premature stop codons within, or have exon–exon junctions within, their 3′ untranslated regions.

In Drosophila melanogaster oocytes, the eventual subcellular localization of an mRNA can be determined while the mRNA is still in the nucleus²⁶, and this is likely also to be the case for other organisms and cell types. It is, however, not clear whether an mRNA’s localization fate is determined in the nucleus or cytoplasm in neurons, and there is circumstantial evidence that both scenarios occur. In mice, Actb mRNA requires zipcode-binding protein 1 (ZBP1; also known as IMP1 or IGF2BP1) to bind to a motif in its 3′ UTR in the cytoplasm for localization into axons¹², but also requires that the mRNA is initially bound by ZBP2 (also known as FUBP2 or KHSRP) in the nucleus²⁷. In addition to EJC proteins, a number of heterogeneous nuclear RNPs and other RBPs with well-known roles in RNA splicing and nuclear organization have been shown to bind to axonal mRNA localization motifs, suggesting that these proteins could initiate their interaction with axonally targeted mRNAs in the nucleus^24,28–30 (FIG. 1b). However, stimulation of axons with different trophic and pathfinding stimuli, including neurotropic stimuli and injury, was shown to alter axonal RNA populations of rat sensory neurons in the absence of ongoing transcription³¹, which suggests that co-transcriptional determination of RNA fate through nuclear RBP interactions may not be important for all axonal mRNA localization. In Xenopus oocytes, RNPs are remodelled along the way to their final location with the addition and removal of different RBPs^32,33, implying that the fate of an mRNA may be continually fine-tuned by the complement of proteins bound to the transcript.

Several short RNA motifs or cis elements that are necessary and sufficient for localization of mRNAs into axons have been identified, including the ‘zip code’ motif (present in ACTB mRNA) in chick neurons and the MAIL motif (present in Kpnb1 mRNA) and the AU-rich element (ARE; present in Gap43 and Nrn1 mRNAs) in rat and mouse neurons^12,34–36. However, no consensus sequences have emerged from studies of these and other axonal mRNA localization elements so far, with the exception of the ARE, which is found in many mRNAs. The question of consensus motifs for axonal localization is further complicated by the finding that RBPs can have multiple effects on the mRNAs to which they bind. For example, HU proteins bind to ARE-containing mRNAs and affect both their survival and their translation³⁷, with the neuronal-specific HUD (also called ELAVL4) binding to Gap43 and Nrn1 AREs and stabilizing these mRNAs in rat neurons³⁸. But it has also been shown that the AREs in Gap43 and Nrn1 mRNAs can drive localization of these mRNAs^35,36. The cis elements identified to date range in size from 30–40 nucleotides to a few hundred nucleotides. The longer motifs may represent complex secondary or tertiary structures or binding sites for multiple RBPs that have not yet been defined. It has been shown that even when small motifs of 40–125 nucleotides are used for affinity purification of RBPs from rat peripheral nervous system (PNS) axons, many RBPs can specifically bind to a single RNA motif²⁴. Similar results were observed for the MAIL motif that is responsible for the localization of Kpnb1 mRNA (encoding importin subunit-β1), although it was discovered that nucleolin was the predominant RBP bound to that motif²⁸. The reader is referred to a recent review that outlines known axonal RNA localization motifs and, when known, the RBPs that bind to those motifs³⁹. These RNA localization motifs are by no means unique to neurons, and mRNA localization driving cis elements have been identified in bacteria, metazoans, plants and insects^40–43.

RNA-sequencing analysis of the rat PNS axon mRNAs that co-precipitate with several axonal RBPs has further shown that there are tens to hundreds of different interacting mRNAs for each RBP, suggesting that these different mRNAs may be co-transported by the same RBP²⁴. The interactomes of the axonal heterogeneous nuclear RNPs K, H1 and F have been assessed bioinformatically, showing that these rat RBPs bind to RNA regulons linked to axon growth; however, another axonal RBP, the La (also known as SSB) protein, does not bind to growth-related regulons but binds to a different cohort of axonal mRNAs²⁴ (FIG. 2). While binding of an RBP to axonal mRNAs could represent the formation of a transport RNP, the interaction could represent the formation of an RNP with other functions. Another bioinformatic study took a different approach and searched for established binding sites for the RBP splicing factor proline and glutamine rich (SFPQ) in mRNAs that had been shown to localize into axons; this established an RNA regulon associated with axonal survival that included Bcl2l2 and Lmnb2 mRNAs²⁹. However, as outlined above, RNA profiling studies have shown that other RBPs bind to many different mRNAs, so it is likely that an unbiased test for axonal SFPQ targets would provide a more systematic assessment of an axonal SFPQ RNA regulon. RNA interactomes for ZBP1, FMRP, and HUD — RBPs that have been linked to axonal mRNA localization — have shown hundreds of mRNAs specifically binding to these proteins^44–47. However, these analyses were performed with whole cell or tissue extracts rather than the subcellular compartment(s) relevant for their localizing activities. The RNA interactomes of other RBPs linked to axonal localization, such as nucleolin and hermes, have similarly not been tested at the level of axons. There is clearly a need for more unbiased profiling of RBP–mRNA interactions directly within axons.

Fig. 2 |. Defining axonal mRNA regulons.

An RNA-binding protein (RBP) can interact with multiple axonal mRNAs that encode proteins with complementary functions, thereby defining an axonal RNA regulon^24,29. The schematic depicts three axonal RNA regulons in which related mRNAs (depicted in different shades of blue, red or purple) are bound by a single shared RBP. Additional RBPs bind to this complex, either in the nucleus or in the cytoplasm, to establish a ribonucleoprotein (RNP) for axonal localization. The regulon-defining RBP is shown here as a nuclear RBP, on the basis of the finding that heterogeneous nuclear RNPs bind to axonal mRNAs that are associated with axon growth²⁴. This does not exclude the possibility of a regulon defined by cytoplasmic RBPs. Axonal RNA regulons for axon growth²⁴ and mitochondrial function (axon survival)²⁹ have been reported, while a regulon for injury signalling in the axon was inferred from work showing that nucleolin is needed for the axonal localization of both Kpnb1 and Mtor mRNAs^28,30.

Interactomes.

Sets of molecular interactions; for example, an RNA-binding protein has both an ‘RNA interactome’ and a ‘protein interactome’ that define the macromolecules that it interacts with.

An RNA consensus motif that retains mRNAs in the neuronal soma, a fate that is conferred by its interaction with Pumilio 2, has recently been uncovered⁴⁸. This finding supports the contention, long held by many in the field, that mRNAs do not lack subcellular localizing activity but are rather actively restricted from subcellular sites through specific RNA motifs. Such cell body retention may differ for individual mRNAs between different neuron types, since Actg mRNA localizes into axons of cultured motor neurons but is retained in the cell body of cortical and sensory neurons⁴⁹. Ironically, the soma-restricting RNA motif is, thus far, better defined as a consensus than the axon-localizing motifs, which suggests that the number of different RBPs that can drive axonal mRNA localization is potentially greater than the number that can restrict mRNAs to the neuronal cell body.

A further consideration when one is attempting to understand the mechanisms that drive the axonal localization of mRNAs is the affinities of mRNAs for individual RBPs. Different motifs can clearly compete for binding to an RBP²³, and it has been shown that heterologous overexpression of axonal RNA localization motifs can block the axonal localization of multiple mRNAs⁵⁰.

mRNA transport into axons

RBPs and mRNAs assemble into granules for transport into axons.

According to a reductionist view, an axonally localizing mRNA is first bound by an RBP and then linked to motor proteins through adaptor proteins for transport into axons. On the basis of several lines of evidence, multiple proteins can bind to an mRNA via localization motifs as well as other regions of the mRNA (FIG. 1c). For example, ZBP1 and heterogeneous nuclear RNP R bind to the 3′ UTR of Actb mRNA^51,52, and a complex containing both HUD and ZBP1 must bind to the ARE of Gap43 for axonal localization, at least in rat PNS neurons³⁵. Proteins that do not directly bind RNA can also be part of a transport granule and have an impact on axonal mRNA transport and translation; for example, SMN helps to assemble RNPs but also localizes into axons and is needed for the proper axonal localization of Actb and Gap43 mRNAs^53,54. As noted above, RNPs are transported into axons and dendrites as membraneless, organelle-like granules that range in size, 100 nm or greater in diameter⁵⁵. Both in axons and dendrites there is evidence that granule populations can be distinguished on the basis of the RBPs that they contain^56–59 and, although not fully characterized, multiplex in situ hybridization studies have shown that there is a limited number of mRNAs within dendritic RNA transport granules⁶⁰. Finally, work in HeLa cells shows that single mRNAs are present in smaller granules containing IMP1 and YBX1 (REF.⁶¹). Although how the different RNP sizes affect function is not clear, the RNA granule size variations may reflect the existence of transport granules with different RNA–RBP compositions and/or the growing and shrinking of granules through the addition or removal of constituents for different needs.

Adaptor proteins.

Proteins containing specific protein-binding sites that facilitate interactions between protein binding partners (for example, proteins linking cargos to motor proteins).

Transport granule.

An RNA-protein complex, or ribonucleoprotein, that is needed for transport of mRNAs to subcellular sites.

RNPs form through a self-assembly process. In some (but not all) RNPs, liquid–liquid phase separation (LLPS) generates phase-separated condensates in which mRNAs and proteins are selectively concentrated⁶². It is therefore appealing to speculate that RNA transport granules may also be generated through LLPS, although it is not clear whether and how motor proteins can tether to phase-separated condensates for the distribution of mRNAs along the axon and it is unlikely that all RNPs self-assemble by LLPS. Several lines of evidence indicate that LLPS occurs as a result of the presence of intrinsically disordered regions or low-complexity regions (LCRs) within select RBPs, which self-oligomerize through non-specific weak interactions (reviewed in REF.⁶³). FUS, TDP43 and TIA1, RBPs the genes of which are subject to mutations causing amyotrophic lateral sclerosis (ALS), each undergo this type of self-assembly through LCRs⁶⁴. Notably, RNP granules are likely to fuse and split, allowing the constant mixing of molecules within these structures as well as their exchange with the cytoplasm⁶⁵ (although the larger pathological aggregates that result from ALS mutations appear less dynamic⁶³). Post-translational modifications can alter the LLPS tendencies of some RBPs. For FMRP, an RBP that localizes to dendrites and axons, phosphorylation increases its LLPS, while methylation decreases this property⁶⁶. In contrast, phosphorylation of G3BP1 decreases its LLPS ability based on work in vitro and in cell lines^67,68. LLPS by axonal G3BP1 is increased immediately following axotomy in rats, with newly synthesized casein kinase 2α phosphorylating G3BP1 and decreasing the numbers of axonal G3BP1 granules⁶⁹.

Liquid–liquid phase separation.

(LLPS). A process in which solutions of macromolecules (proteins and nucleic acids) transition to form a membraneless phase-separated cytoplasmic condensate.

Several lines of evidence point to the RNA transport granule as providing a level of translational control by preventing the translation of neuronal mRNAs while they are in transit to distal neurites⁷⁰ (FIG. 3a,b). For example, ZBP1 binding blocks translation of Actb mRNA in NG108-15 neuroblastoma and HEK293 cells, but phosphorylation of ZBP1 by SRC-family kinases reduces its affinity for mRNA, thereby releasing it for translation⁷¹. The FMRP-containing RNPs that undergo phase separation, as mentioned above, contain eukaryotic translation initiation factor 4E-binding protein 2 and microRNA (miRNA) miR-125b, both of which have the capacity to inhibit generalized cap-dependent and specific mRNA translation⁶⁶. Biochemical purification of RNA granules through either affinity purification for motor protein interaction or differential sedimentation has shown that they contain many different translation factors, RBPs, RNA helicases and ribosome components^58,72,73, indicating that a transport granule is likely to contain many different proteins with the capacity to directly or indirectly regulate translation. It is notable, however, that these approaches result in the analysis of a mixture of different RNA granules, so the published granule protein and mRNA compositions likely reflect combinations of many different granule types rather than an individual granule population. Furthermore, evidence has been provided for the transition of FMRP-containing granules from RNA transport granules to polysome-interacting units⁵⁸. This is likely to reflect multifunctionality of FMRP, a common feature of RBPs, but also emphasizes the dynamic nature of RNPs and underscores the need for better temporal resolution when one is defining RNA granule composition. In addition, the transport of stalled polysomes into dendrites was recently suggested in a study showing that the RNA helicase UPF1 is needed for polysome stalling on Map1b mRNA in the cell soma but also for dendritic Map1b mRNA localization⁷⁴. This raises the possibility that mRNAs that undergo translation that is initiated by multiple ribosomes can be translationally stalled and redirected to subcellular domains as a mechanism for spatial and temporal translational control in neurons.

Fig. 3 |. The fate of mRNAs on reaching their axonal destinations.

a | Schematic of a neuron showing the transport of mRNA and pre-microRNAs (pre-miRNAs) to their axonal destinations. Ribosomes and translation factors accumulate in growth cones and at spots along the axon shaft, including at branch points in cultured neurons^6,15,153; these regions have been referred to as ‘hotspots’ for translation in axons. Axonal mRNAs are presumably released from a transport ribonucleoprotein (RNP) or vesicle (endosome or lysosome) at or near those hotspots. However, axonal translation is likely not limited to these regions. The released mRNAs can be translated or sequestered into axonal mRNA storage depots. b | mRNAs in the transport granule are thought to be translationally suppressed during their transport^72,154 and single RNA-binding proteins (RBPs) have been shown to prevent mRNA translation when bound⁷¹. On arrival in axons and release from RBPs, mRNAs can be immediately translated or they can be stored in stress granule-like compartments as seen for G3BP1 granules in rodents and TIAR-2 granules in Caenorhabditis elegans^90,91. c | Vesicle-coupled mRNAs are thought to be similarly translationally repressed by RBPs during their transport. On release of the RNP from the vesicle and subsequent release of RBPs, the mRNA can be immediately translated or stored in axonal stress granule-like compartments. RNPs bound to lysosomes include G3BP1 (REF.⁷⁷), which has also been shown to store mRNAs in axons⁹¹, suggesting that the transport RBP may be remodelled on release to become a storage depot for axonal mRNAs. d | RNA-induced silencing complex (RISC) components have been shown to associate with vesicles^76,116, suggesting that vesicle co-transported pre-miRNAs may provide a second means to suppress the translation of local mRNAs until needed. That is, an ‘on demand’ maturation of the pre-miRNA that happens before or after its release from the vesicle could place it and RISC components in close proximity to the released mRNAs and allow translational silencing.

Polysome.

A complex of multiple ribosomes bound to an mRNA that is typically regarded as a site of active translation.

Some axonal RNAs associate with vesicles for transport and translation.

Several recent publications have demonstrated a previously unrecognized association of some axonal mRNAs with membrane bound organelles — including endosomes, multivesicular bodies, lysosomes, and proteins involved in endoplasmic reticulum (ER)–Golgi complex trafficking — that have been linked to RNA regulation across different cellular systems (BOX 1). In neurons, recent studies have shown that endosomes, lysosomes and mitochondria contribute to axonal RNA transport and/or translational regulation^75–77 (FIG. 1d). In the case of RNA transport, RNPs that bind to vesicles have been described as ‘hitchhiking’, which implies a somewhat random process in which a passing vehicle picks up a passenger in search of a ride⁷⁸. This mechanism is energetically advantageous, having the potential to transport many axonal cargoes at once, and so is undoubtedly driven by specific interactions that have undergone centuries of evolution. A similar hitchhiking mechanism has been documented in numerous other organisms that show polarized subcellular organization (BOX 1), such as the filamentous fungus Ustilago maydis⁷⁹.

Box 1 |. organelle-based mRNA transport and translation mechanisms.

mRNAs have been shown to ‘hitchhike’ onto components of the endocytic membrane transport pathway in a number of polarized cells types. endosomes, lysosomes, and multivesicular bodies are components of the endocytic membrane transport pathway¹⁵⁵. These membrane-bound organelles have a characteristic luminal pH and are defined by their specific rib GtPase protein content. endosomes can originate by budding from the Golgi complex or by inteRNAlization from the plasma membrane. endocytosis of plasma membrane proteins can generate ‘signalling endosomes’ for intracellular signalling or that can be recycled back to the plasma membrane to regulate membrane components such as receptors and ion channels. Other endosomes are transported between different regions of the cell (in a mechanism known as transcytosis), such as axonal and dendritic compartments. early endosomes (expressing RAB5A) can undergo maturation to become late endosomes and multivesicular bodies (expressing RAB7), which have key roles in the sorting of endocytosed cargos and can fuse with the lysosomes that mediate protein degradation. Multivesicular bodies can also fuse with the plasma membrane to release exosomes. together, these components bring flexibility to the control of RNA localization and translation in specific subcellular regions.

The yeast Saccharomyces cerevisiae provided early evidence for differential RNA transport in specific cellular subcompartments. ASH1 and several other mRNAs were shown to localize and co-migrate with the cortical endoplasmic reticulum (er) during daughter cell budding, using the RNA-binding protein (rBP) she2 as a tether^156–159. COPI proteins drive localization of mRNAs for translation at or near mitochondria in yeast, and these mRNAs mislocalize to the ER when the mRNA–COPI protein complex is disrupted¹⁶⁰. the transition from a yeast-like cell to a polarized hyphal growth in the fungus Ustilago maydis requires the transport of CDC3 mRNA, which hitchhikes on RAB5A-positive endosomes in a complex with the RBP RRM4 and the adaptor protein uPa1 (REFS^79,113). ribosomes associated with these vesicles provide an efficient means to asymmetrically deliver and establish polar growth^79,161. endosomes were also recently implicated in transport of GluB mRNA to the cortical ER in endosperm cells of rice¹⁶².

Recent work in mammalian cells extends the roles of the ER in the synthesis of cytoplasmic proteins. the translation of mRNAs encoding secreted and membrane-bound proteins is well established to occur at the rough ER, but it was recently shown that translation of a broad population of mRNAs encoding cytoplasmic proteins also occurs at the ER¹⁶³. another recent study found distinct populations of ribosome-bound mRNAs interacting with the ER proteins LRRC59 and SEC61B in HEK293 cells, with particular enrichment of mRNAs encoding organelle-associated proteins¹⁶³. this suggests that the ER provides a compartmentalized subcellular environment for the translation of specific mRNAs.

Some initial hints that vesicle-based RNA transport is present in neurons could be inferred from work showing that COPB1, a subunit of the coatomer complex that coats vesicles that are transported between the ER and the Golgi complex, contributes to the microtubule-based transport of murine Kor (also known as Oprk1) mRNA into sensory axons⁸⁰. Similarly, a protein component of the endosome sorting complex required for transport (ESCRT), ESCRTII, was shown to be an RBP for Drosophila bicoid mRNA⁸¹. ESCRT plays a role in endosome sorting and multivesicular body formation, but has many additional functions, including receptor signalling⁸². More recent work shows that ESCRTII colocalizes with actb mRNA, the axon guidance protein receptor Dcc and early endosomes in the growth cones of Xenopus retinal ganglion cells⁸³. These and other studies provided links between vesicle trafficking-associated proteins and axonal mRNAs that could impact axonal mRNA transport and/or translation. Consistent with these hints, recent studies from several groups have provided direct links between axonal RNPs and the endosomal pathway, spanning from early endosomes to lysosomes, pointing to these vesicles as transport and translational control vehicles for RNAs^75–77,84 (FIG. 1d).

Endosomes and lysosomes are well suited as hitchhiking platforms for axons since they can undergo long-range bidirectional transport along microtubules⁸⁵. Adaptor proteins, such as pleckstrin homology domain-containing family M member 2 (PLEKHM2; also known as SKIP) — in combination with ADP-ribosylation factor-like protein 8 (ARL8) — and HOOK proteins, link the vesicles to kinesin and dynein motor proteins for anterograde and retrograde axonal transport, respectively^86,87. The movement of endosomes and lysosomes is characterized by directional changes and pauses, and many of these organelles undergo retrograde and anterograde transport in relatively equal proportions⁸⁸. Although we generally think of axonal mRNAs as moving anterogradely, there are instances of retrograde movements of RBPs and RNAs. For example, the axonal RBP La is anterogradely transported in its native form and retrogradely transported after sumoylation⁸⁹. Other examples of the retrograde movement of RNPs are mentioned below. Even though the function of mRNAs moving retrogradely is currently not clear, having an ‘on demand’ mechanism to relocalize mRNA cohorts from one region to another within the axon could be advantageous in responding to different stimuli, provided that there is some specificity to the response with regard to the delivered mRNAs. Thus, bidirectional transport of vesicles within axons could provide the means to reposition mRNAs and translational machinery along the axon.

It was recently shown that the RBPs G3BP1, TDP43, and caprin 1 hitchhike on lysosomes in the axons of cultured rodent cortical neurons⁷⁵. G3BP1 and caprin 1 are components of stress granules, organelle-like structures used to store translationally silent mRNAs during periods of cellular stress⁶⁸. Notably, intra-axonal stress granule-like structure formation by G3BP1 (in rats and mice) or the stress granule protein TIAR-2 (in Caenorhabditis elegans) has been shown to decrease axonal regeneration after axotomy by decreasing axonal mRNA translation^90,91. G3BP1-containing RNA granules interact with lysosomes through the adaptor protein annexin A11 (ANXA11) for transport along axons in rat, zebrafish and stem cell-derived human neurons⁷⁵. Annexins are a protein family that show conserved membrane binding and contain Ca²⁺-interacting domains⁹². ANXA11’s interaction with lysosome and endosome compartments occurs via its carboxy terminus and is Ca²⁺ dependent. Disruption of this interaction through the introduction of protein encoded by a mutant ANXA11 gene linked to ALS blocks transport of G3BP1-containing RNPs and Actb mRNA⁷⁵. G3BP1 contains three LCRs and is known to self-assemble via LLPS. Recent work indicates this self-assembly requires interactions with RNAs^67,68. Furthermore, the amino-terminal region of ANXA11 also contains an LCR sequence that promotes LLPS that may enable its interaction with RNPs⁷⁵.

The relationship between the ANXA11-carrying lysosome linked G3BP1 granules that are transported along axons and G3BP1’s role in stress granules is not clear. For example, it is unknown whether the G3BP1 RNA transport granule is distinct from the axonal G3BP1-containing stress granule-like structure or whether the transport granule matures into an axonal mRNA storage depot once it has docked in the axons. Further work will clearly be needed to resolve this question, but the RNP modifications that are documented to occur in transit in Xenopus oocytes³³ provide an appealing mechanism for changes in granule function (FIG. 3c). Overall, these recent studies provide a provocative mechanism for the movement of large RNA granules rapidly across axons, and the proteins linking RNPs to vesicles have clear links to human diseases. Notably, neurons cultured to model ALS, spinal muscular atrophy or peripheral neuropathies have shown alterations in axonal mRNA localization^93–97.

Notably, recent studies have reported that a significant number of organelles labelled by the lysosome marker LAMP1 lack major lysosomal hydrolases⁹⁸, so it remains unknown whether those lysosomes linked to RNA transport are actually active lysosomes. One can speculate that there are different vesicle types linked to axonal RNPs, in a manner similar to the mixture of different RNA granules present in the biochemical preparations outlined above. Further work will be needed to define the nature and potential variations of the vesicles associated with RNA localization. Furthermore, Rab7a-carrying late endosomes have recently been implicated as sites of mRNA translation in Xenopus laevis retinal ganglion cell axons⁷⁷. These Rab7a vesicles frequently paused at axonal mitochondria and transported mRNAs that encode mitochondrial proteins, thus providing a delivery (and potentially synthesis) platform to support axonal mitochondrial function. The introduction of rab7a mutations identical to those that cause Charcot–Marie–Tooth 2B neuropathy in humans in these neurons decreased both intra-axonal protein synthesis and neuronal mitochondrial function⁷⁷. It is important to mention that most of the axonal mRNAs examined in this study were transported into axons in a vesicle-independent fashion. In combination with the work in G3BP1 granules outlined above, this emphasizes the fact that vesicle-based RNA localization and translation mechanisms account for the localization of only a subset of axonal mRNAs.

Vesicle-associated transport extends to axonal miRNAs.

Work in several different neuronal systems has shown that non-coding RNAs (ncRNAs) localize into both axons and dendrites. These include small miRNAs and PIWI-interacting RNAs^19,99. miRNAs modulate translation and/or survival of mRNAs in axons^100–106. Although not as well studied as miRNAs, PIWI-interacting RNAs are also presumed to regulate axonal mRNA survival on the basis of what is known of their functions in other systems¹⁰⁷. There is evidence that pre-miRNAs, the precursors that must be processed for miRNA functionality, can be transported into both axons and dendrites and then processed locally in response to different stimuli^108,109. This provides a mechanism to locally modulate the axonal translatome, but it is not clear how these ncRNAs actually get into axons. There are reports of the transfer of ncRNAs as well as ribosomes to axons from glial cells¹¹⁰ (BOX 2), but many of the experimental systems in which axonal ncRNAs were initially observed used compartmentalized cultures in which glia are not included in the axonal compartment. This suggests that mechanisms have evolved to move ncRNAs into axons, and that these mechanisms exhibit specificity that means that not all ncRNAs make their way into axons and dendrites.

Box 2 |. Axonal ribosome dynamics.

Early electron microscopy studies showing that mature CNS axons do not contain polysomes led to the unfortunate conclusion that axons were not capable of protein synthesis¹⁶⁴. More recent work has shown that translation in CNS axons largely occurs through single ribosomes bound to mRNAs (monosomes)¹⁶⁵, which may help to explain why it was missed by the classical electron microscopy work⁶ and has implications for our understanding of how the neuron regulates axonal protein synthesis. Consistent with this, there are now several reports of mRNAs and translational machinery in CNS axons, even in the mature CNS^166–169. indeed, recent work examining axonal translation in the forebrain proposed that translation in CNS axons is likely to be “more a rule than an exception”¹⁶⁶. in addition, there are instances in which polysomes have been detected in vertebrate peripheral nervous system (PNS) axons, suggesting that translation could occur in axons in a manner similar to that in dendrites^6,170.

A study in 2008 raised the intriguing possibility that ribosomes could be transferred to distal PNS axons from surrounding schwann cells¹¹⁰. although transfer was initially observed in distal severed axons (that is, in the segment that will go on to degenerate), it was more recently shown that transfer to injured PNS axons still connected to the soma is also possible^171,172. it is not clear whether this mechanism extends to the CNS, but it is intriguing to consider it in light of the distances that axons extend. very recent work suggested that the transfer may be microtubule dependent and include glial-derived mRNAs¹⁷³. At least in vitro, Schwann cell-derived exosomes can transfer miRNAs to neurons and modify axon outgrowth¹⁷⁴. One therefore has to keep in mind that at least some of the ribosomes and RNAs identified in axons could derive from other cells. This is particularly the case in vivo, where axons extend manyfold longer distances than they do in culture conditions (where much of the work on axonal mRNA transport/translation has been performed). Other cellular systems were recently shown to transfer mRNA from one cell to another through membrane nanotubes¹⁷⁵; the close relationship of glia to axons could facilitate such a mechanism.

Axons contain an extensive population of mRNAs, including many that encode ribosomal proteins (RPs)¹⁹. this is curious, given that biogenesis of ribosome subunits was thought to occur strictly in the nucleolus. However, there is evidence that ribosomes are not a homogeneous population, and several lines of evidence point to the existence of functionally distinct ribosome cohorts^176,177. This raises the possibility that local translation of axonal RP mRNAs could alter ribosome function if the proteins could be added to the ribosome subunits locally in axons. indeed, a recent article showed RNA motif-dependent translation of axonal RP mRNA codons and the addition of the nascently translated rP to the axonal ribosome¹⁷⁸. Other recent work showed that different RPs associate with the guidance cue receptors DCC, neuropilin 1 (NRP1) and ROBO2 through an RNA-dependent mechanism¹³⁴. Proteosome-dependent degradation of some axonal ribosome constituents as axons initiate synapse formation has been shown, which emphasizes the dynamic nature of axonal ribosomes¹⁷⁹.

Similarly to mRNAs, pre-miRNAs have been shown to be transported through interactions between RBPs and vesicles^76,84 (FIG. 1d). Neuronal pre-miR-134 localizes into dendrites via the DEAH-box RNA helicase DHX36, which directly binds to pre-miR-134’s terminal loop¹¹¹. So far, RBPs that are required for axonal localization of pre-miRNAs or miRNAs have not been reported. However, it is known that pre-miRNA-338’s localization into rat sympathetic axons is microtubule dependent, and mass spectrometry of proteins co-precipitating with pre-miRNA-338 identified RBPs, translational machinery constituents, components of the 40S and 60S ribosomal subunits and important motor proteins¹¹². Pre-miR-338 associates with mitochondria and modulates translation of cellular mRNAs encoding mitochondrial proteins. This association with mitochondria is an intriguing parallel with coding RNAs, and raises the possibility of vesicle-mediated trafficking of axonal pre-miRNAs and/or miRNAs. Indeed, recent work in Xenopus provided evidence that axonal pre-miRNAs can localize into axons through association with vesicles⁸⁴. This study showed that Xenopus pre-miR-181a-1 co-traffics with late endosome and/or lysosome compartments, with the pre-miRNA showing both anterograde and retrograde movements along axons to ultimately reside in the central domain of the retinal ganglion cell growth cone. Importantly, super-high-resolution microscopy showed that the pre-miRNA molecules are tethered to the outer membrane of the late endosomes and lysosomes rather than inside the vesicle⁸⁴. G3BP1 RNPs were similarly seen adjacent to the outer membrane of lysosomes by electron microscopy of human U2OS cells⁷⁵. Although details on how pre-miR-181a-1 docks onto the vesicular membrane are still missing, it is intriguing to speculate this role may be served by the molecular adaptor ANXA11, as is the case for the axonal G3BP1-containing RNA granules⁷⁵, or the protein UPA1, as is the case for the link between the RBP RRM1 and endosomes in filamentous fungi^79,113. Notably, it has been shown that both the enzyme Dicer, a ribonuclease needed for miRNA maturation that had been shown to localize into axons previously^114,115, and miR-124 co-traffic along with acidic vesicles in murine motor axons⁷⁶. This raises the possibility that pre-miRNAs may co-traffic with the machinery needed to cleave them into mature miRNAs, which is analogous to sending the ‘factory’ as a single unit into the axons rather than simply bringing the parts to the factory (FIG. 3c).

A critical unanswered question is whether the pre-miRNA processing occurs entirely separately from the transported granule or whether vesicle-associated pre-miRNA is processed during transport to regulate mRNAs within the granule. These possibilities are not mutually exclusive, so both may occur depending on the pre-miRNA and RNA cargos, which would bring substantial flexibility for regulating dynamics of axonal mRNAs and translation. Also, the translational activity of different mRNAs could change depending on the granule’s subcellular environment, meaning that there may be an advantage to selectively attenuating the translation of some but not all mRNAs that are adjacent to or being transported by a vesicle compartment. Notably, the RNA-induced silencing complex (RISC) that is needed for miRNA function depends on the endosomal pathway¹¹⁶, supporting the idea miRNA functionality is linked to transported vesicles.

Organizing translation ‘on demand’

Axonally synthesized proteins support developmental and regenerative axon growth, axon maintenance and axon survival, and play a role in neurodegeneration through their local functions in the axon or by generating retrograde signals that are sent to the soma^117,118 (FIG. 4). Many of these retrograde signals modify transcriptional programmes in the nucleus, indicating that axonally synthesized proteins can broadly impact the neuronal transcriptome (reviewed in REFS^119,120). There have been hints at the specificity of the axon’s translational responses to different stimuli. For example, attractive cues increase translation of actb mRNA in growth cones, whereas repulsive cues stimulate translation of cofilin mRNA in Xenopus laevis^121,122. The links between axonal RNAs and vesicles as well as the association of axonal RNAs with other organelles and organelle-like structures discussed earlier raise the question of whether these structures help to organize the translational specificity of the axon’s response to different stimuli and physiological states. Indeed, there is evidence for a higher level of organization for translational regulation that is likely to help to match the local translatome to the temporal and spatial physiological needs of the axon’s domains and subdomains. RNA granules isolated from neurons and the brain of different species have, in some cases, been shown to contain translation factors and ribosome subunits in addition to RBPs and mRNAs^{58,59,72,73,123}. This, coupled with the translational suppression or stalling of transported mRNAs outlined earlier herein⁷⁴, suggests that the RNA transport granule — on its own or in association with vesicles — could provide a platform to mobilize mRNA translation once the RNP reaches the correct location in the axon (FIG. 3). It is clear that different stimuli can newly activate or increase translation of different mRNAs to provide an ‘on demand’ mechanism to change the axonal proteome. In the following sections, we suggest that the regulatory platforms described above provide a level of organization for translational control in axons that parallels the concept of RNA regulons.

Fig. 4 |. Distinct axonal translational control mechanisms have different functional outcomes.

The concept of RNA regulons (FIG. 2) can be extended to encompass groups of axonal mRNAs that exhibit translational regulation on the basis of the functional effects of the encoded proteins. The schematic illustrates examples of these regulatory mechanisms. a | Late endosome-coupled mRNA transport has been linked to several mRNAs involved in the support of mitochondrial function, which is needed for collateral branching of axons^77,141,142. b | Some axonal mRNAs that encode injury response-associated and axon regeneration-associated proteins are translationally silenced by storage in G3BP1 granules (stress granule-like structures) in mammalian peripheral nervous system axons; phosphorylation of G3BP1 (G3BP1^PS149) triggers disassembly of these granules and the release of the mRNAs for translation^67,68,91. TIAR-2 granules have similarly been linked to the storage of growth-associated axonal mRNAs in the axons of Caenorhabditis elegans, with phosphorylation of TIAR-2 favouring granule disassembly⁹⁰. c | Cell surface receptors can sequester mRNAs and ribosome subunits in the axoplasm, releasing these for synthesis of proteins presumably needed for axon guidance on the receptor binding to those guidance cues^133,134. d | Increased levels of axoplasmic Ca²⁺ following injury can trigger endoplasmic reticulum (ER) stress-associated phosphorylation of eukaryotic translation initiation factor 2A (EIF2A), which can suppress generalized protein synthesis but paradoxically upregulate the synthesis of proteins needed for response to injury^{119,144,145,147,148}. Motifs in both the 5′ untranslated region (UTR) and the 3′ UTR of axonal mRNAs have been shown to modulate both their transport into axons and their translation within axons; these post-transcriptional mechanisms are mediated by RNA-binding proteins and by translation initiation factors interacting with those motifs. Such a mechanism could therefore underlie the injury response. Notably, signals activated by trophic and tropic stimuli can converge on translation initiation factors to regulate axonal mRNA translation in addition to the receptor-mediated storage mechanism depicted in panel d. e | Although not as clearly linked to RNA regulons, the repression of translation in axons as well as potentially ‘on demand’ degradation of some axonal mRNAs can be regulated through microRNAs (miRNAs). Thus far, this regulatory mechanism has been linked to mitochondrial function^100,112. f | Several of the axonally translated proteins that are regulated by these mechanisms can contribute to axon-to-soma signalling to further regulate gene expression¹¹⁷.

Platforms to store mRNAs within axons until needed.

The axon must make a choice of whether to translate or store an mRNA once its transport ceases. It was recently shown that axonal G3BP1 granules are used to store mRNAs in mature rat PNS axons⁹¹ (FIG. 4b). Kpnb1 and Nrn1 mRNAs are stored in these granules in mammalian axons, but Gap43 mRNA is not stored and its translation is not affected by G3BP1 overexpression or aggregate disassembly. The translation of mRNAs stored in G3BP1 granules is needed for optimal axon regrowth after axotomy, since G3BP1 granule disassembly as a result of the application of a cell-permeable peptide from G3BP1’s acidic domain in rodents or the knockdown of TIAR-2 in C. elegans increases axon growth^90,91. Thus, the sequestering of axonal mRNAs in stress granule-like structures can be thought of as a general mechanism for modulating an RNA regulon for axon growth after injury, at least as tested thus far. The finding that Gap43 mRNA is not regulated by G3BP1 points to the existence of distinct cohorts of mRNAs in axons, regulated by different cellular platforms. From comparison of the published profiles of axonal mRNAs from rat and mouse neurons^124–126 with those of mRNAs co-precipitating with G3BP1 from whole cell or tissue lysates of various species^127–130, there are potentially many G3BP1-interacting axonal mRNAs beyond Kpnb1, Nrn1 and Actb mRNAs that have been published to date^75,91. However, it is not known whether mRNA storage by G3BP1 and TIAR-2/TIA1 is something that the neuron utilizes for normal functions and thus whether the response to axotomy simply takes advantage of a normal physiological mechanism. If such non-stressed roles exist for G3BP1 and TIAR-2/TIA1 granules in neurons, then we should stop referring to these axonal structures as ‘stress granules’.

Another mechanism for storing axonal mRNAs has been uncovered that provides potential for direct modulation by extracellular ligands. Adenomatous polyposis complex (APC), a microtubule plus-end-binding protein, was shown to bind mRNAs in pseudopodia of motile fibroblasts¹³¹. APC prominently localizes to rat neuronal growth cones, and it has been shown that growth cone mRNAs and axonal microtubules are linked through their interactions with APC¹³². It was unclear whether APC plays a role in transport of the mRNAs, but the association of APC with microtubules and its concentration in the growth cone raise the more likely possibility that it serves as a storage depot for some localized mRNAs.

Further evidence for storage of mRNAs in growth cones came from a study showing that ribosome components and translation factors bind the cytoplasmic tail of the guidance cue receptor DCC in murine growth cones, with ligand binding to DCC activating protein synthesis adjacent to the receptor¹³³. This provided a compelling mechanism for directly coupling translation to extracellular stimuli – essentially, DCC is ‘geared up and ready’ to rapidly generate new proteins for the growth cone once activated (FIG. 4c). This concept was recently extended to include other guidance cue receptors, including neuropilin 1 (NRP1) and ROBO2, but not EPHB2 (REF.¹³⁴). This ribosome–receptor coupling is mRNA dependent, and distinct RBPs and mRNA populations co-precipitate with each receptor¹³⁴. The ligand-dependent assembly of polysomes along the receptor-associated mRNAs requires endocytosis, thereby providing a link to the vesicular transport mechanisms outlined above. It will be of interest to determine whether other ligand–receptor systems are similarly geared up for translation. Nonetheless, the ribosome–receptor coupling and stress granule-like storage outlined above point to stimulus-dependent intra-axonal mechanisms that are specifically tuned to release cohorts of mRNAs for translation.

Axonal mitochondria and adjacent sites bring a mechanism for translational regulation.

Although not known as a storage depot, mitochondria are emerging as another platform for regulating translation of localized mRNA cohorts (FIG. 4a). The local translation of nuclear-encoded mitochondrial protein mRNAs was initially seen in the squid giant axon¹³⁵. This was later replicated in mammalian neurons, where mRNAs encoding mitochondrial proteins are translated adjacent to or even on the mitochondrial membrane^136–138. For example, the axonally synthesized products of CoxIV (also known as Cox4i1) and Lmnb2 mRNAs maintain mitochondrial function and are needed for axon survival^138,139, but it is not presently known whether these mRNAs are translated at or near axonal mitochondria. These and other examples suggest that there is a mitochondrial-linked RNA regulon that includes mRNAs that encode proteins needed for mitochondrial function in axons. It has been shown that the axonally synthesized protein BCL-2L2 (also called BCL-W) supports axon survival by inhibiting inositol 3-phosphate receptor 1 (IP3R1) on the axonal ER membrane⁹⁵. Since ER Ca²⁺ released by IP3R1 activation can precipitate the release of mitochondrial Ca²⁺ that ultimately causes axon degeneration¹⁴⁰, this brings Bcl2l2 mRNA into the mitochondrial-linked RNA regulon and implicates SFPQ, the protein that cotransports axonal Bcl2l2 and Lmnb2 mRNAs²⁹, in the regulation of this regulon.

Mitochondrial respiration has also been shown to support localized protein synthesis. For example, protein synthesis-dependent collateral branching of axons requires that transported mitochondria stall along axons to generate a localized source of ATP for translation¹⁴¹. Similarly, physiological stimuli that activate synaptic plasticity require a focal source of ATP (mitochondria) to support mRNA translation at the synapse, creating a compartmentalized region of localized protein synthesis adjacent to the activated synapse¹⁴². It is appealing to speculate that the filopodia that transition into an axon branch when induced by nerve growth factor (NGF)^141,143 similarly represent compartmented regions of the axon that support mRNA translation. This is also a further indication that extracellular stimuli can converge on this perimitochondrial translation compartment.

Translational regulation in axons by calcium and stress signalling.

After axonal injury in the rat PNS, an influx of extracellular Ca⁺² into the injured axon and subsequent Ca²⁺ release from the axonal ER have been shown to mobilize the translation of stored axonal mRNAs³⁹ (FIG. 4d). This increased level of axoplasmic Ca²⁺ activates the translation of Kpnb1, Ranbp1, Stat3 and Vim (encoding vimentin) mRNAs within a few hours after crush injury of the sciatic nerve. The release of Ca²⁺ from axonal ER stores also increases translation of Calr mRNA in axons through phosphorylation of the translation initiation factor EIF2A¹⁴⁴. It was recently shown that the axotomy-induced increase in Calr mRNA translation requires activation of PERK, thereby pointing to a contribution of localized ER stress¹⁴⁵. Since the ER chaperone proteins needed for folding nascently synthesized proteins are Ca²⁺ dependent¹⁴⁶, depletion of ER Ca²⁺ triggers ER stress and results in an unfolded protein response. Activation of the unfolded protein response in axons promotes translation of Calr, Grp78 (also known as Hspa5 and Bip) and Luman (also known as Creb3) mRNAs in axons^144,145,147 and promotes axon regeneration after PNS nerve injury¹⁴⁸. Together, these observations point to ER-associated signalling as a platform to regulate local synthesis of axonal injury-associated proteins. It will be of interest to determine whether this translational regulatory mechanism is used to respond to other types of axonal stress and whether this involves a shared ‘stress RNA regulon’. Axonal translation in response to the SEMA3A guidance cue was recently shown to similarly require phosphorylation of EIF2A¹⁴⁹, suggesting that this ER-modulated RNA regulon likely has functions beyond injury. These signalling pathways likely layer onto the translational platforms outlined above to bring further temporal and spatial specificity. Additionally, ribosomes themselves may provide some specificity to axonal translation, as a result of both the mechanisms through which they get into axons and their composition (BOX 2).

Unfolded protein response.

A molecular response that occurs when levels of unfolded proteins increase in the endoplasmic reticulum; this reduces overall protein synthesis to decrease continued load of unfolded proteins and allows the cell to respond to different types of stress.

Summary and future perspectives

As we hope the reader will glean from this Review, neurons tightly regulate mRNA transport into axons and translation within axons to enable many different functions. Thankfully, the field has been able to rapidly transition from proving that translation occurs in axons to understanding the biology of the system. Mutations in genes encoding proteins needed for axonal mRNA transport, storage and translation have been linked to human neurological disorders associated with axon degeneration, including peripheral neuropathies and motor neuron degeneration, emphasizing the importance of the functions that axonally synthesized proteins serve. Indeed, expression of ALS-associated mutations in the gene encoding TDP43 in mice precipitate broad but specific changes in the axonal mRNA and miRNA populations⁷³.

Publications emerging over the last few years point to a higher level of organization for regulating the axonal proteome through localized translation. RNPs and organelle-like RNA granules (some assembled through LLPS and some true membrane-bound organelles) have been shown to provide molecular platforms that play a role not only in the transport of mRNAs but also in their translation. By co-packaging ribosome subunits, translation factors and pre-miRNAs with some of these transport vessels, the neuron can effectively send out much of the factory for translation to sites in the axon or just deliver the mRNA templates. Storage of mRNAs once localized, in granules and associated with receptors, brings a compartmentalizing feature for translation ‘on demand’ independent of the neuronal cell body. Cohorts of mRNAs sharing these different platforms — that is, RNA regulons — bring the opportunity to functionally organize mRNA transport and translation in order to drive the axon’s response to different stimuli and physiological conditions to optimally and autonomously fill its needs.

While we imply that these axonal RNA regulons are specific for different axonal responses and functions, there are undoubtedly some overlapping mRNAs between these regulons. It is intriguing to speculate that different mechanisms are used for transport or translation of these overlapping mRNAs depending on the functions they serve. There is perhaps some evidence for this already with Calr mRNA having distinct 3′ UTR motifs for ligand-dependent versus constitutive transport into axons¹⁵⁰. It is also conceivable that the axon makes use of combinations of just a few RNA regulons to mount specific responses to different stimuli and physiological states. To fully understand the functional organization of axonal RNA regulons implied here, we need to better systematically define the different mRNAs that support different functions of the axon at both the transport level and the translational level. As we learn more about these mechanisms, there will undoubtedly be connections between these and additional regulatory layers for axonal functions and responses. There are some hints of connections in the literature now. For example, transport of G3BP1 granules through ANX11A was suggested to require Ca²⁺ release from lysosomes⁷⁵, and G3BP1 is deacetylated by HDAC6, with acetylated G3BP1 blocking its interaction with RNAs¹⁵¹. Axonal HDAC6 is activated by Ca²⁺ and deacetylates MIRO1 to decrease mitochondrial transport in axons¹⁵². It will be of interest to determine how this may affect the compartmentalized synthesis of mitochondrial proteins outlined above and how the axon balances this and other post-translational modifications to fine-tune axonal protein synthesis.