Insights into the roles of local translation from the axonal transcriptome

3.1. Guidance cue responses

Over the past decade, numerous studies have demonstrated that local translation of axonally localized transcripts is important for various functions of developing axons, including axon growth and pathfinding, as well as retrograde signalling (table 1). As mentioned earlier, local protein synthesis was implicated in mediating the responses of growth cones to several guidance cues, such as semaphorin 3A (Sema3A), netrin-1 and brain-derived neurotrophic factor (BDNF) [6]. In each of these cases, turning responses and other morphologic changes to growth cones were blocked by selective application of protein synthesis inhibitors to axons. Local translation was subsequently shown to be required for the adaptation responses of cultured Xenopus spinal neuron growth cones in gradients of extracellular guidance cues [39]. Growth cones undergo cycles of desensitization and resensitization to a specific guidance cue, and inhibition of local protein synthesis prevents resensitization from occurring [39].

Subsequent studies showed that local translation of specific axonally localized transcripts accounted for the protein synthesis requirements of these guidance cues. This was first demonstrated for Sema3A-induced growth cone collapse in axons of rat embryonic sensory neurons [11]. Transcripts encoding RhoA, a monomeric GTPase that regulates actin dynamics, were shown to localize to axons and shown to be translated in response to Sema3A signalling [11]. To definitively demonstrate that axonal RhoA transcripts, and not cell-body-localized transcripts, mediate this effect, a new technique termed ‘axon-specific knockdown’ was developed [40]. This approach, which uses application of siRNA exclusively to axons, showed that knockdown of RhoA mRNA selectively in axons impaired Sema3A-induced growth cone collapse, demonstrating that the axonal RhoA pool mediates the morphological responses of growth cones to Sema3A. Similarly, other studies have implicated local translation of β-actin, cofilin and Par3 in the responses to different guidance cues, such as netrin-1, Slit-2, nerve growth factor (NGF) and BDNF [12,13,15,16]. These studies demonstrated that local translation mediates the responses to numerous axon growth and guidance cues.

3.2. Retrograde signalling

More recent studies have indicated that local translation also has roles in other aspects of axonal signalling. For example, local translation has been implicated as a novel mechanism to convey signals from growth cones to the nucleus, thereby influencing gene transcription. This can be accomplished through local synthesis of transcription factors or adaptor proteins that are retrogradely trafficked to the cell body. Local synthesis of CREB, CEBP-1, STAT3, importins and SMAD transcription factors have all been linked to retrograde signalling mediated by NGF, BMP4 and nerve lesion [27,31–34]. These studies indicate that the consequences of local translation are not limited to localized responses but can extend to other subcellular compartments, such as the nucleus.

3.3. Control of axon-specific protein expression

Local translation may also be important for enabling axon-specific protein expression. For instance, local translation may facilitate the targeting of neuropeptides to axon terminals. Indeed, transcripts encoding vasopressin and oxytocin have been detected in nerve terminals of the posterior pituitary [41–43]. Local translation may also direct the expression of specific receptors in distal axons. For example, the κ-opioid receptor mRNA is localized to axons of mouse sensory neurons [38], and olfactory receptor mRNAs are found in distal axons of olfactory neurons [44,45]. Local translation may also regulate the timing of receptor expression in axons. For example, a reporter construct containing the 3′UTR of the guidance cue receptor EphA2 is upregulated in commissural axons only as the axons reach the midline [46]. Although it is not known whether endogenous EphA2 transcripts are present or regulated in commissural axons, these findings raise the possibility that intermediate targets alter the chemotropic responses of axons by inducing local synthesis of guidance cue receptors. Together, these studies suggest a role for local translation in controlling the selective expression of specific proteins directly within growth cones.

3.4. Axonal regeneration

Although axonal localization and translation of mRNAs in developing neurons is well established, the same is not true for adult neurons. Some studies have identified mRNA in mature vertebrate neurons, including Mauthner cells [47], hypothalamic magnocellular neurons [42] and sensory neurons projecting to the olfactory bulb [48], but it is not clear whether axonally localized transcripts are a feature of most types of adult axons. Indeed, mRNA and rRNA levels disappear from axons of hippocampal neurons as they mature [49]. The mechanism by which mRNA and rRNA are lost is not currently understood. Although axons from adult neurons appear to contain reduced levels of mRNA and ribosomes than axons from embryonic neurons, in vitro studies from regenerating axons suggest that adult neurons possess the capacity to restore axonal localization of mRNA. Indeed, axons from adult rat sensory neurons preconditioned with nerve injury contain components of the translational machinery, including ribosomal proteins, rRNA and translation initiation factors both in vivo and in vitro [5]. In culture, these regenerating axons can synthesize new proteins even when severed from their cell bodies [5,26], and intra-axonal protein synthesis is required for proper axon regeneration [50]. Moreover, several studies have demonstrated the functional importance of intra-axonal translation of specific mRNAs in regenerating axons (table 1). Together, these data support the idea that adult axons use local mRNA translation after nerve injury to mediate axon regeneration.

4.1. First-generation axonal mRNA libraries

Several studies (table 2) using both invertebrate and vertebrate neurons have established that a common set of transcripts is found in diverse types of axons. For example, cDNA libraries from squid giant axons [51,52] identified about 150 distinct transcripts by mRNA differential display. Sequence analysis of 27 of these mRNAs revealed a heterogeneous axonal mRNA population that encompassed transcripts for cytoskeletal components, translational machinery (including mRNAs for ribosomal proteins), nuclear-encoded mitochondrial proteins and signalling proteins [52]. Similar results were later obtained by the Twiss group who identified more than 200 different mRNAs in axons from injury-conditioned rat sensory neurons through hybridization to a cDNA microarray containing about 4000 rat cDNAs [53]. These transcripts encoded proteins involved in protein synthesis, intracellular transport, calcium regulation, metabolism and mitochondrial and cytoskeletal functions. cDNA libraries were also prepared from axons of cultured cortical cells which resulted in the identification of more than 300 transcripts [54]. Gene ontology analysis showed that transcripts with a role in translation, mitochondrial function, cytoskeleton regulation and intracellular transport were also enriched in these axons [54].

More recently, serial analysis of gene expression (SAGE) has been used to characterize mRNAs in axons purified from neonatal rat sympathetic ganglia neurons cultured in Campenot devices [22]. SAGE results in the identification of short sequence tags that lie near the poly(A) tail of transcripts and that can uniquely identify individual mRNAs [55]. Almost 50 per cent of the tags from axonal RNA could not be mapped to known transcripts using current SAGE databases, making it unclear which transcripts they correspond to. However, the other 50 per cent identified a total of 317 axonal transcripts. Notably, half of these tags were derived from the 20 most abundant axonal mRNAs, suggesting that just a few mRNAs account for the majority of axonal mRNA. Consistent with earlier studies, these transcripts encoded mitochondrial proteins (30%), signalling proteins (20%), ribosomal proteins (15%), cytoskeletal proteins (5%) and other proteins involved in RNA metabolism and protein synthesis (5%). Interestingly, certain transcripts identified in axons from other types of neurons, such as CREB and cofilin, were not identified in axonal SAGE libraries, although they were also not present in the corresponding cell body SAGE libraries. One valuable feature of this study is that it allows the direct comparison of mRNA distribution in axons and cell bodies, thus making it possible to identify mRNAs that exhibit a high degree of enrichment in axons. These transcripts may be useful for indentifying sequence elements that target mRNAs to axons. In addition, an mRNA that is highly enriched in axons might have unique roles in axons compared with the one that is expressed at lower levels in axons.

Overall, these different libraries showed considerable overlap in the types of transcripts identified in axons, providing evidence that local translation may have common functions in axons from different types of neurons.

4.2. Second-generation axonal mRNA libraries

Two recent large-scale studies have used a microarray analysis to reveal a much more elaborate and complicated landscape of axonal mRNAs, identifying thousands of axonal mRNAs (table 2).

The first study not only identified axonal transcripts, but also identified transcripts in specific domains within axons [8]. This group used laser-capture microdissection coupled with microarrays to profile the transcriptome of growth cones and axon shafts from Xenopus retinal ganglion neurons [8]. Unlike previous axonal transcriptome studies, this approach distinguished between mRNAs that are enriched within neuronal growth cones from mRNAs that are enriched in axon shafts. Transcripts that are localized in axon shafts are likely to be translated and affect local processes within axon shafts. Thus, the division of the axonal transcriptome into growth cone and axon shaft-enriched pools may be valuable to predict potential roles of specific axonal mRNAs. The study identified 5105 transcripts that localize to axon shafts of stage 32 neurons and 958 transcripts that localize to the growth cones of the same type of neurons [8]. Of the mRNAs that are present in growth cones, only a minimal portion (6%, 58 transcripts) are enriched in growth cones, while the majority of the transcripts found in growth cones (54%, 523 transcripts) are enriched in axon shafts (figure 1).

What does this mean in terms of axonal biology and local translation? Differential analysis of the growth cone and axonal transcriptomes revealed that growth cones are enriched for transcripts encoding cytoskeletal and motor proteins, proteins involved in translation, transmembrane and cell surface receptors, nuclear proteins, secreted proteins, signalling proteins and metabolic proteins [8]. On the other hand, transcripts encoding proteins with roles in protein degradation, apoptosis, membrane trafficking, cell cycle and transcription factors are enriched in axon shafts. This apparent difference in mRNA localization and distribution within the periphery might underlie fundamental differences between the roles of local translation in axons and growth cones.

The idea that the differences in the axonal and growth cone transcriptomes have functional roles presupposes that translation can occur in both sites. Although translation is often thought to occur in growth cones, translation in axons has also been demonstrated [11]. However, it will be important to confirm that the newly identified axonal mRNAs are indeed translated in axon shafts. If these axonal mRNAs are not translated within axon shafts, then an alternative model could be that axons are storage sites for translationally silent mRNAs and that mRNAs need to be transported to growth cones to be activated. Intriguingly, a similar phenomenon has been observed in dendrites where upon stimulation, dendritic mRNAs translocate to activated synapses within spines, where local translation occurs [56,57].

The study by Zivraj et al. [8] also revealed that the composition of the growth cone transcriptome changes during neuronal maturation. Early embryonic Xenopus retinal ganglion growth cones (stage 24) contained 286 unique transcripts, compared with the 958 contained in growth cones of late embryonic (stage 32) [8]. Functional classification of the mRNAs from the two datasets showed that growth cone transcripts from many different functional categories were upregulated in older axons. The developmentally regulated, axonally localized transcripts may allow local translation to serve different roles at different stages of neuronal development.

This study further showed that an evolutionarily conserved core set of mRNAs is found in axons from neurons of different species. Of the 444 known growth cone transcripts identified in stage 32 Xenopus retinal ganglion neurons, 238 transcripts encoding homologous proteins were also found in growth cones from mouse retinal ganglion neurons of similar developmental stage. This suggests a moderate degree of conservation between the two systems that might point to the existence of conserved functions of local translation.

In the second study, distal axons from embryonic and adult rat sensory neurons were compared [7]. A microarray analysis revealed that 2627 transcripts localize to embryonic axons and 2924 to regenerating adult axons [7]. About 50 per cent of transcripts (1445 mRNAs) were present in both embryonic and adult axons, further supporting the idea that a core set of mRNAs localizes to axons, regardless of the developmental stage. An Ingenuity pathway analysis of the transcript population common to both types of axons revealed that both embryonic and adult sensory axons contained transcripts encoding mitochondrial proteins, proteins linked to neurological diseases and proteins involved in mRNA translation and protein synthesis. Surprisingly, cytoskeletal- and transport-related mRNAs as well as cell cycle mRNAs were a major component of the embryonic axonal transcriptome, whereas inflammatory and immune response mRNAs were present only in adult axons. These findings suggest that additional sets of mRNAs may be present at distinct stages of development to perform stage-specific axonal functions.

Although these studies identified a large population of axonal mRNAs, technical limitations may have prevented a full appreciation of the axonal transcriptome. In particular, the brain is highly enriched in transcripts that lack the poly(A) tail [58,59]. Signal-dependent polyadenylation of transcripts is thought to be a mechanism that regulates local translation in axons [60], which may indicate that transcripts lacking the poly(A) tail are likely to have important roles in mediating translation-dependent responses to signalling molecules in axons. Because the poly(A) tail is required for detection of transcripts in certain library approaches, such as SAGE [22] or cDNA synthesis using oligo(dT) primers as in Zivraj et al. [8], these approaches may miss important subsets of axonal RNAs. This might also explain differences in the apparent content of axonal RNA libraries that have been generated thus far, because different axonal populations may have different pools of polyadenylated mRNAs.

These studies demonstrated that the axonal transcriptome is complex and highly regulated. With the number of axonal transcripts jumping from hundreds to thousands of mRNAs, these studies allow broader conclusions about the major roles for local translation in axons. Some of these are discussed subsequently.

10.1. What transcripts are found in the axons of living animals?

The studies that have sought to determine the roles of local translation have largely focused on cultured neurons. Therefore, it is possible that the phenomenon of mRNA trafficking to axons could reflect an adaptation to the culturing environment. Furthermore, the specific mRNAs that are detected by library approaches may be influenced by the culturing conditions. Recent studies have provided evidence for the physiological occurrence of axonal targeting of mRNA. Recently, the Twiss group described a genetically engineered mouse that expresses an RNA that interferes with the function of ZBP-1, a protein that is important for trafficking β-actin mRNA to axons [85,86]. These mice exhibit defective axonal regeneration, supporting the idea that local translation has roles in this process. In another study, our group showed that BDNF induces the intra-axonal synthesis of SMAD1/5/8 [33]. In embryos deficient in BDNF, SMAD1/5/8 is depleted from axons, but not from cell bodies, supporting the idea that BDNF physiologically regulates intra-axonal protein synthesis of SMAD1/5/8. However, despite these few examples, future studies are needed to ascertain whether the various roles ascribed to axonal mRNAs based on studies in cultured neurons can be validated in vivo.

To understand the role of local translation in vivo, it will be important to determine which mRNAs are localized to actively growing axons during embryogenesis, in adult animals, and after axonal injury. New genetically engineered animals may help to address this question. We recently described the use of the EGFP-L10a mouse [87] to explore ribosome localization in adult and regenerating axons in animals [88]. These mice express a fluorescent form of the ribosomal protein L10a under the control of the Glt25d2 promoter that drives transgene expression in layer 5b cortical neurons [89]. Using these mice, we found EGFP-L10a labelling at the nodes of Ranvier in uninjured mice, and prominent labelling in the growth cones of regenerating spinal axons [88]. These transgenic mice are especially useful to selectively image ribosomes in long projection axons, without the problem of detecting ribosomes from nearby non-axonal tissue. This general approach can also be used to selectively harvest ribosomes and their bound mRNAs from axons. To do this, the corticospinal tract can be dissected and subjected to anti-EGFP immunoprecipitation. This will result in the selective recovery of axonal ribosomes since EGFP-L10a is expressed only in cortical neurons and their descending axons in the spinal cord. Ribosomes from the surrounding myelin or other supporting cells will not be recovered because they will not express EGFP-L10a. This approach can be used to identify transcripts bound to ribosomes in normal adult corticospinal tract axons, regenerating axons after injury or even in growing axons during development. The use of different promoters to drive the expression of the transgene could potentially allow the isolation of axonal ribosomes from other neuronal populations, as long as the axons are physically distant from the cell bodies that express EGFP-L10a. These experiments could reveal which mRNAs are localized to axons during pathfinding and in other physiological contexts.

The use of in vivo systems will also be important to better understand the role of the supporting glial cells in influencing intra-axonal translation. A recent study using a ribosomal protein reporter found that Schwann cells can transfer polyribosomes and, potentially, mRNAs to axons severed from cell bodies [90]. This finding suggests that neighbouring glial cells can directly modulate axonal gene expression by regulating the content of axons. It will be important to understand whether the exchange of mRNA and ribosomes between glia and neurons occurs only in response to axonal injury or whether it is a constitutive physiological pathway used to influence axonal biology.

10.2. Are transcripts that require endoplasmic reticulum and Golgi processing synthesized and targeted in axons?

A major challenge in understanding whether axons can produce secreted and membrane-bound proteins is to demonstrate that these proteins can be locally processed in the absence of typical ER and Golgi structures. Despite the apparent absence of ER and Golgi, several studies have suggested that axon growth cones may have ER/Golgi-like functionality. The strongest data supporting this idea were the demonstration of axonal protein glycosylation by Flanagan and co-workers [46]. These investigators examined the fate of placental alkaline phosphatase (PLAP) transcripts transduced into chick retinal axons using Sindbis virus, an RNA virus that is translated by the host ribosomes [91]. PLAP is a glycosyl phosphatidylinositol (GPI)-anchored protein that is enzymatically active only when exported to an extracellular location [46]. Addition of the GPI anchor, which is typically thought to occur in the ER, allows the docking of the protein to the outer surface of the cell membrane. Therefore, detection of surface immunoreactivity for the PLAP reporter could serve as an indicator of GPI anchor addition. Indeed, heterologous PLAP expression resulted in detectable surface phosphatase activity, suggesting that PLAP can be locally glycosylated and inserted in the membrane [46]. Interestingly, a review of the second-generation mRNA libraries shows that mRNA for at least one well-characterized GPI-anchored protein, NCAM, can be found in axons.

However, it is difficult to extrapolate from the PLAP experiment whether the more complex patterns of membrane insertion or glycosylation seen in most other receptor proteins can occur in axons. Experiments monitoring the trafficking of locally translated membrane-targeted proteins during different stages of protein processing in axons are needed. It will be important to directly measure axonal glycosylation and determine whether it results in similar patterns of glycosylation complexity as those seen in cell bodies. To further establish the extent to which axonally synthesized proteins are locally glycosylated, recently described azide- and alkyne-tagged glycosylation precursors labelled with ‘click chemistry’ [92] might allow the monitoring of glycosylation in isolated axons or in compartmentalized but intact neurons. These experiments will provide evidence for the presence of functional ER- and Golgi-like structures in axons.

10.3. Do axonal transcripts have different sequence features than transcripts in cell bodies?

Although many of the transcriptomic studies have identified novel mRNAs that are present in axons, the specific transcript variants that are localized are not yet known. All transcript are susceptible to alternative splicing and it is estimated that more than 50 per cent mammalian transcripts use alternative poly(A) sites [93], which influences the 3′UTR length. In addition, in some cases, the use of alternative promoter can lead to different 5′UTRs. Techniques such as next-generation sequencing (NGS) are particularly useful for determining transcript structures [94], and therefore may identify specific variants that are selectively trafficked to axons. Indeed, recent studies have provided specific examples of distinct variants of the same transcript that are differentially transported to axons [22,23,28]. Application of NGS to the axonal transcriptome will be useful to efficiently identify transcript variants that are preferentially enriched in axons. Ultimately, this may help in the identification of sequences or elements that confer axonal targeting.

These variants may also be differentially regulated, as variant-specific sequences may contain microRNA-binding sites or sites for the association of RNA-binding proteins. The potential role for microRNAs in regulating local translation was suggested by the finding that proteins involved in microRNA-mediated translational silencing are present in embryonic axons, and that introduction of small interfering RNA into axons leads to degradation of target mRNAs [40]. Several microRNAs have since been shown to be enriched in distal axons [95], and miR-9 has been shown to locally regulate BDNF-dependent axon extension and branching [96].

10.4. How do the diverse axonal transcripts get targeted to axons?

Only a few cis-acting elements have been identified that target mRNAs to axons. Of these, the β-actin zipcode sequence is one of the best characterized [97,98]. This element, a conserved bipartite sequence present at the beginning of the β-actin 3′UTR [97,99], has been shown to interact with the RNA-binding protein ZBP-1 to mediate axonal localization of β-actin [4,100]. However, the canonical zipcode sequence is found only in a limited number of transcripts [99], raising the question of what mechanisms are used to target other mRNAs to axons. One possibility is that the binding of an mRNA to ZBP-1 is not exclusively determined by the presence of the canonical zipcode sequence in the mRNA. Rather, specific recognition of the mRNA might depend on the secondary and tertiary structure that the mRNA assumes when folded. It is therefore possible that ZBP-1 has many more targets that can be predicted using current algorithms. Indeed, a recent report found that competing away ZBP-1 by overexpressing a β-actin 3′UTR transgene reduces not only axonal levels of β-actin, but also GAP43 mRNA, which lacks a canonical zipcode [85]. Further studies are needed to understand which cis-acting elements and proteins determine the axonal localization of mRNAs that do not use ZBP-1.

The axonal mRNA profiling studies can also offer insights into some of the potential mechanisms regulating axonal mRNA transport. For example, these studies found that mRNAs for ribosomal proteins are enriched in axons. A common feature of all ribosomal transcripts is the cis-element at the beginning of their 5′UTR known as 5′-terminal oligopyrimidine tract (5′-TOP), which consists of a C followed by a tract of 6–10 pyrimidines [101]. It is possible that this shared feature accounts for the selective enrichment of the ribosomal transcripts in axons. Consistent with this idea, other transcripts that contain 5′-TOP sequences, such as all the five elongation factors and the initiation factors eIF3e, eIF3f and eIF3h [102] are also enriched in axons. How could 5′-TOP sequences be targeted to axons? The La protein, which has previously been shown to bind 5′-TOP sequences [103], is known to be trafficked into axons [104], making La a candidate for axonal transport of these mRNAs.

In addition to ribosomal transcripts, it is also unknown how transcripts encoding mitochondrial proteins are targeted to axons. Transcripts encoding mitochondrial proteins have not been shown to share a common sequence motif, which might be implicated in axonal targeting. However, these transcripts associate with the outer surface of the mitochondria [105,106]. How could this explain the axonal localization of mitochondrial mRNAs? One model is that these mRNAs are transported to axons together with the mitochondria. mRNA association with the mitochondrial surface is co-translational, and mediated by a mitochondrial targeting signal on the nascent protein that is recognized by receptors on the mitochondrial outer surface [107]. As with proteins targeted to the ER, the complex of the nascent mitochondrial protein, its corresponding mRNA and ribosomes are bound in close proximity to mitochondrial pores that mediate the translocation of the newly translated protein inside the mitochondria [107].

10.5. Why are only some axonal proteins synthesized locally?

One of the major insights revealed from the axonal transcriptome libraries is that many proteins that are known to be present in axons are not synthesized locally. For example, the components of the PAR polarity complex have all been localized to distal axons and growth cones by immunofluorescence [16]. However, axonal mRNA for PAR-6 and aPKCζ was not detected by FISH [16] and did not appear in any of the axonal mRNA libraries. Thus, these proteins were most likely synthesized in the cell body and transported to the growth cone. In contrast, PAR3, another component of the PAR complex, is locally synthesized from axonally localized PAR3 transcripts [16]. The idea that only a subset of axonal proteins are synthesized locally is supported by metabolic labelling experiments using axonally applied [³⁵S]-methionine/cysteine [108]. These studies showed that only approximately 5–10% of neuronal proteins are locally synthesized in axons, while the majority of axonal proteins are synthesized in cell bodies and anterogradely transported into axons. Thus, only certain proteins in axons are locally translated while other proteins are derived from the cell body. These findings raise the question of the underlying logic behind why some proteins are locally synthesized and why others are made in the cell body and subsequently transported to axons. Conceivably, certain common regulatory or functional aspects of locally translated proteins may explain why these proteins are preferentially translated within axons.

An additional consideration is that some axonal proteins may derive from both local synthesis and transport from the cell body. For example, selective application of protein synthesis inhibitors to axons showed that less than 1 per cent of the axonal actin and tubulin derive from local synthesis [109]. Why would both cell-body-derived and locally synthesized proteins both be needed in axons? One possibility is that the locally translated protein may have a different set of post-translational modifications than the cell-body-derived proteins, as has been suggested for the local synthesis of β-actin [9].

10.6. What is the significance of the low-abundance and ‘non-enriched’ axonal transcripts?

Most of the information on the roles and importance of local translation comes from studies that look at transcripts that are highly abundant in distal axons. However, the second-generation axonal libraries have also uncovered the presence of a subset of transcripts that localize to axons but have low abundance compared with other axonal mRNAs. It is possible that low-abundance transcripts represent mRNAs that are not actively targeted to axons but localize to axons through diffusion or other passive mechanisms, thus representing transcriptome ‘noise’. However, it is also possible that low-abundance or non-enriched transcripts are locally translated and are required for specific axonal functions. Indeed, β-actin transcripts are not enriched in axons [22], although local translation of β-actin has important roles in axonal turning [12,13]. Axon-specific knockdown will be a valuable tool to assess the roles of these non-enriched axonal transcripts.