Organization and translation of mRNA in sympathetic axons

The long-term survival of most axons relies upon an uninterrupted connection to the neuronal cell body (Cajal, 1991), which provides a continuous supply of proteins and organelles generated there (Grafstein and Forman, 1980; Vallee and Bloom, 1991). However, axons do carry out autonomous synthesis of many macromolecules, notably lipids (Vance et al., 1991; De Chaves et al., 1995) and proteins (Koenig and Giuditta, 1999; Alvarez et al., 2000). Thus, it seems likely that, when physiological changes occur in the axon, often at a great distance from the cell body, the neuron responds not only by altering synthesis in and transport from the cell body (Moskowitz et al., 1993; Moskowitz and Oblinger, 1995; Gillen et al., 1997) but also by regulating synthesis nearby, within the axon itself (Tobias and Koenig, 1975; Eugenin and Alvarez, 1995). However, the relative contributions of axonal and somatic translation and their relative responses to events in the distal axon are almost entirely unknown.

Vertebrate axons contain mRNA (Koenig and Giuditta, 1999) along with organized ribosomes (Koenig and Martin, 1996; Koenig et al., 2000), and both the delivery of mRNA to the axon and the maintenance of its distribution there are microtubule dependent (Bassell et al., 1994b; Olink-Coux and Hollenbeck, 1996). Transport of mRNA into the axon must be selective, because axons do not contain the full neuronal set of transcripts (Olink-Coux and Hollenbeck, 1996). In particular, transcripts for major cytoskeletal proteins have been detected in spinal nerve roots of the rat (Koenig, 1991) and in axons of rat brain neurons (Bassell et al., 1998; Litman et al., 1993), goldfish retinal ganglion cells (Koenig, 1989), goldfish Mauthner cells (Weiner et al., 1996) and chick sympathetic neurons (Olink-Coux and Hollenbeck, 1996). In addition, the recruitment of β-actin mRNA into the axon is regulated by cell signaling events (Bassell et al., 1998; Zhang et al., 1999) and might require the same RNA localization sequence that mediates its asymmetric distribution in polarized fibroblasts (Zhang et al., 2001).

Axonal translation of numerous polypeptides has been demonstrated by metabolic labeling in vertebrate axons (Koenig and Adams, 1982; Koenig, 1989; Koenig, 1991) and the translation of two major cytoskeletal proteins (actin and β-tubulin) has been demonstrated specifically in axons of rat sympathetic neurons (Eng et al., 1999). However, is axonal protein synthesis essential for any axonal function? There are conflicting reports for axonal growth: inhibition of axonal translation has been reported to inhibit axonal regeneration in rodent nerve (Remgård et al., 1992; Edbladh et al., 1994; Gaete et al., 1998) but to have no effect on the growth rate of cultured rat sympathetic neurons (Eng et al., 1999). However, recent reports show that the response to guidance cues of growth cones of Xenopus retinal ganglion cell neurons depends completely on the synthesis and degradation of proteins in the distal axon, on a time scale of minutes (Campbell and Holt, 2001; Ming et al., 2002), and synapse formation in Aplysia neurons also requires axonal protein synthesis (Schacher and Wu, 2002).

To determine exactly how neurons regulate the supply of new macromolecules, it is essential to understand the organization and magnitude of the axon's own synthetic capacity. Which mRNAs are exported to the axon, where do they reside and where is axonally synthesized protein deployed? What is the capacity of axonal translation, what proportion of total neuronal translation synthesis can be carried out there? We have tried to resolve these questions by examining, in a cultured vertebrate neuron that produces bona fide axons, the presence, distribution and translation of total mRNA and of two individual transcripts: β-actin and actin-depolymerizing factor (ADF) in particular. We find that mRNA is delivered very efficiently to axons, that total mRNA and specific species have a similar, non-uniform distribution, and that axonally translated protein is distributed throughout the axon. We immunoprecipitate newly synthesized β-actin, ADF and neurofilament protein from axons free of cell bodies, and show that a significant proportion of the neuron's protein synthesis can occur in axons.

All reagents and supplements were obtained from Sigma (St Louis, MO) unless otherwise specified. All culture media were obtained from Fisher Scientific (Pittsburgh, PA) unless otherwise specified.

Chick sympathetic neurons were cultured as described previously (Hollenbeck, 1993a; Olink-Coux and Hollenbeck, 1996). Sympathetic chain ganglia were isolated from 9- to 12-day-old chick embryos and dissociated. For in situ hybridization, 18 mm round No. 1 coverslips were coated with 1% poly-L-lysine (M_r 70,000-120,000) for 12-18 hours, followed by 0.1 μg ml^–1 laminin for 1-18 hours. Coverslips were placed in 12-well plates and dissociated neurons were grown on them in C medium [Leibowitz L-15 medium supplemented with 10% fetal bovine serum (FBS; Gibco BRL, Rockville, MD), 100 U ml^–1 penicillin and 100 μg ml^–1 streptomycin (Gibco BRL, Rockville, MD), 0.6 mM glucose, 2 mM L-glutamine, 50 ng ml^–1 2.5-S mouse nerve growth factor (Alomone Labs, Jerusalem, Israel) and 0.5% methylcellulose (Dow Chemical Company, Midland, MI)] at 37°C. To grow axonal halos for metabolic labeling experiments, 60 mm tissue culture dishes were coated with 1% poly-L-lysine in 0.1 M boric acid buffer (pH 8.4) overnight at 4°C, then approximately 20 to 25 intact sympathetic ganglia were plated onto each dish and grown in C medium in the presence of 0.6 μm Ara-C (cytosine 1-D-arabinofuranoside) for 4-5 days at 37°C to yield axons 4-5 mm in length. For direct autoradiography of ³⁵S metabolically labeled axonal halos, ganglia were cultured on 22 mm square coverslips coated with 1% poly-L-lysine.

Synthetic oligo(dT) 32-mer and oligo(dA) 32-mer were 3′ end-labeled with digoxigenin-11-dUTP using terminal transferase according to the manufacturer's protocol (Roche Molecular Biochemicals, Indianapolis, IN). Probes were purified through Quick Spin columns (Roche Molecular Biochemicals, Indianapolis, IN). Labeling efficiency was confirmed by dot blot analysis using digoxigenin-labeled DNA standards, anti-digoxigenin alkaline-phosphatase conjugate and CSPD, a chemiluminescent substrate for alkaline phosphatase (Roche Molecular Biochemicals, Indianapolis, IN).

The 1690 bp chick β-actin coding sequence plus its 3′ untranslated region (UTR) (nucleotides 36-1726, GenBank L08165) was amplified by PCR from a chick brain cDNA library and subcloned into pGEM-T easy (Promega, Madison, WI) at the SacI and BamHI sites. The 537 bp chick β-actin 3′ UTR (nucleotides 1189-1726, GenBank L08165) was amplified by PCR using a 5′ primer with SalI site and a 3′ primer with a BamHI site, and subcloned into pGEM-T easy. The 723 bp fragment containing the whole coding region and a part of the 3′ UTR of chick ADF (provided by J. R. Bamburg, nucleotides 88-811, GenBank J02912) was subcloned into pGEM-3Z at the BamHI and EcoRI sites. Each construct was linearized with an appropriate restriction enzyme to make a template for in vitro transcription. Linearized constructs with 3′ overhang were filled by treatment with Klenow fragment to avoid generating undesirable turn-around transcripts (Rong et al., 1998). Antisense and sense RNA probes were synthesized and labeled by incorporation of digoxigenin-UTP using a DIG RNA labeling kit (SP6/T7) (Roche Molecular Biochemicals, Indianapolis, IN). Labeled RNA probes were purified and labeling efficiency was confirmed as described above. Denaturing agarose gel electrophoresis was used to determine whether probes had the correct size.

Neurons were fixed with 4% paraformaldehyde in PBS containing 4% sucrose for 30 minutes, washed with PBS three times and then permeabilized in 0.3% Triton X-100 in PBS for 5 minutes at room temperature. Poly(A) mRNAs were detected using a modification of the in situ procedure of Olink-Coux and Hollenbeck (Olink-Coux and Hollenbeck, 1996). Fixed, permeabilized cells were hybridized with 10 ng oligonucleotide probe per coverslip, dissolved in hybridization buffer [2× SSC: 0.2% bovine serum albumin (BSA, molecular biology grade, Roche Molecular Biochemicals, Indianapolis, IN), 10 mM vanadyl adenosine complex, 5 mM EDTA, 10% dextran sulfate, pH 7.0, 15% formamide, 20 μg Escherichia coli tRNA (Roche Molecular Biochemicals, Indianapolis, IN)] in a final volume of 30 μl, overnight at 37°C. Coverslips were incubated in malate buffer (100 mM sodium malate, pH 7.0, 150 mM NaCl) for 5 minutes at room temperature and then in 1% blocking solution in malate buffer (Roche Molecular Biochemicals, Indianapolis, IN). After washing, bound probe was detected with the 2-hydroxy-3-naphtoic acid-2′-phenylanilide phosphate (HNPP) fluorescent detection system (Roche Molecular Biochemicals, Indianapolis, IN), according to the manufacturer's protocol. Alkaline-phosphatase-conjugated anti-digoxigenin antibody was added and the enzymatic reaction was performed by adding HNPP and Fast Red Texas Red. Precipitated fluorescent HNP-Fast Red Texas Red signal was visualized with a Nikon microscope using Texas-Red filter set and a 60× objective and photographed with Kodak TMAX 400 film.

To detect β-actin or ADF transcripts specifically, fixed permeabilized cells were hybridized with 10 ng antisense probe per coverslip in hybridization buffer containing 20 μg E. coli tRNA, 50% formamide, 2× SSC, 0.2% BSA, 5 mM EDTA, 0.1% Tween 20 and 0.1 mg ml^–1 heparin sodium salt. Hybridization was performed a chamber humidified with 50% formamide/2× SSC at 65°C for 5 hours. Coverslips were washed once in 50% formamide/2× SSC for 20 minutes at room temperature and then twice in 2× SSC for 10 minutes at room temperature. Hybridized probes were detected by an indirect tyramide signal amplification system (NEN, Boston, MA) according to the manufacturer's protocol. Horseradish-peroxidase-conjugated anti-digoxigenin antibody (Roche Molecular Biochemicals, Indianapolis, IN) was added and then first round enzymatic amplification was performed by adding tyramide-biotin. After addition of horseradish-peroxidase-conjugated streptavidin, a second round of enzymatic amplification was performed by adding tyramide-Cy3. Fluorescent Cy3 signals were observed through a Cy3 filter set and photographed with Kodak Technical Pan film or collected as digital images using a Hamamatsu cooled CCD camera (C4742-95) and Metamorph imaging software (Universal Imaging, West Chester, PA). Kodak TMAX 400 or Kodak Technical Pan negatives were scanned using a Sprint Scan 35 (Polaroid, Cambridge, MA) and images were captured with Adobe Photoshop graphic software (Adobe Systems, San Jose, CA). The contrast and brightness of all fluorescent images in each figure were adjusted identically to avoid distortion of relative signal intensities.

Several negative control experiments were carried out to confirm the specificity of in situ hybridization. First, to test for nonspecific binding of RNA probes, sense RNA probes were hybridized in parallel to antisense probes. Second, antisense probes of the neomycin resistance gene (expressed only in prokaryotes) were hybridized. Third, fixed cultures were treated with an RNase mixture (0.2 mg ml^–1 RNase A, 0.2 mg ml^–1 RNase T₁, 100 U ml^–1 RNase T₂) or RNase One (Promega, Madison, WI) for 30 minutes to 2 hours at 37°C prior to hybridization of antisense probes. Fourth, labeled antisense probes that gave positive signals were subjected to competition with unlabeled antisense RNA probes in the same hybridization solution.

Neurons having at least one axon longer than 100 μm were scored in randomly selected fields from at least two different experiments using 60× objective lens and epifluorescent optics. Specific fluorescent puncta were counted in four different regions of each axon: branch points, varicosities, growth cones and the rest of the axonal shaft. The proportion of the puncta found in each subcellular region was calculated. In order to find the number of puncta per unit area or unit volume of each region, actual numbers of puncta at each subcellular region were divided by the average area or volume of each subcellular region calculated from digital images using Metamorph imaging software (Universal Imaging, West Chester, PA). A two-tailed Student's t test was performed to compare the axonal shaft with other subcellular regions and an ANOVA was performed to compare all subcellular regions with each other. For neurons with more than one axon, the number of axons having signal puncta was counted in each cell and a ψ² goodness-of-fit test was performed to see whether signal puncta were distributed randomly among axons.

After explanted ganglia had elaborated extensive radial halos of axons, each cell body mass was removed by microsurgery essentially as previously described (Koenig and Adams, 1982; Olink-Coux and Hollenbeck, 1996). Ganglia were washed three times with methionine/cysteine-free DMEM supplemented with 10% dialysed FBS, 100 U ml^–1 penicillin, 100 μg ml^–1 streptomycin, 0.6 mM glucose, 2 mM L-glutamine and 50 ng ml^–1 2.5-S mouse nerve growth factor. The cell body mass was then excised at a distance of at least two ganglion diameters, sacrificing some axonal mass but ensuring that no cell bodies were included in the axonal halo. After the surgery, axonal halos were incubated in methionine/cysteine-free DMEM containing 0.5-3.33 mCi ml^–1 EasyTag™ EXPRE³⁵S³⁵S protein labeling mix (NEN, Boston, MA) for 2-5 hours and then washed with Hank's balanced salt solution buffer. Axonal halos were detached from dish by gentle pipetting, harvested into microtubes and dissociated in hot SDS sample buffer. Whole ganglia were also labeled and collected in parallel to axonal halos. As controls, metabolic labeling was also performed in the presence of 1 mM chloramphenicol or 1 mM cycloheximide.

Radioactively labeled ganglia or axonal halos were collected in lysis buffer (Roche Molecular Biochemicals, Indianapolis, IN) containing 50 mM Tris-HCl, 150 mM NaCl, 1% Nonidet P40, 0.5% sodium deoxycholate, and passed ten times through a 25-gauge needle. To immunoprecipitate labeled β-actin, AC-15 antibody (Sigma, St Louis, MO), which is specific for β-actin, was incubated with labeled cell extract. In case of ADF immunoprecipitation, rabbit antiserum against chick ADF (provided by J. Bamburg) was used. For neurofilament protein, the lysis buffer was supplemented with 0.1% SDS, 100 μg ml^–1 RNase A and 30 U ml^–1 DNase, and the cell extract was heated for 10 minutes before the addition of monoclonal antibody NR-4 (Sigma), which binds all three neurofilament proteins (NF-L, NF-M and NF-H). The antigen-antibody complex was precipitated using protein-A or protein-G coupled to agarose or Sepharose (Roche Molecular Biochemicals, Indianapolis, IN; Pierce, Rockford, IL) and analysed by SDS-PAGE. Dried gels were exposed to Kodak Biomax MR film.

Polyacrylamide gels containing radioactive proteins were fixed in 30% acetic acid, 10% methanol for 1 hour and completely dried in a vacuum gel dryer. Fluorography was performed using Kodak Biomax MR film and an intensifying screen at –80°C. For direct autoradiography of axonal halos, fixed axonal halos were dehydrated by successive incubation in 50%, 75%, 90%, 95% and 100% ethanol. Mounted coverslips were dipped in 42°C Kodak NTB2 emulsion solution and air dried for 3 hours under a safe light. Emulsion solution coated coverslips were stored in a light-tight box with desiccant for 2 weeks at 4°C and then developed with Kodak D-19. Coverslips were observed under 10× objective using dark field optics and photographed with Kodak TMAX 100 film.

We have previously demonstrated the presence of poly(A) mRNA in axons of sympathetic neurons using fluorescence in situ hybridization (Olink-Coux and Hollenbeck, 1996). We first re-examined these cells using a detection method that is significantly more sensitive and has a higher signal-to-noise ratio (Takizawa et al., 1997) (see Materials and Methods) to allow us to determine the distribution of total mRNA. The poly(A) mRNA signal was bright in all cell bodies and also in most axons, where its pattern could be clearly seen to be punctate (Fig. 1A,B). The mRNA was distributed along the length of the axons in an irregular pattern and, although it tended to be denser in the more proximal regions of the axon, it was also found in the most distal regions, including the growth cones. The specificity of the antisense (oligo dT) probe for mRNA was shown by its elimination by RNase treatment and its absence when a sense probe was used (Fig. 1C-F). In addition, signal was still detected in cell bodies and proximal axons when RNase T₂, which preferentially cleaves at the 3′ end of A residues (Egami and Nakamura, 1969), was omitted from RNase mix (not shown). This indicates that the fluorescent signal resulted from specific binding of antisense oligo (dT) probes to poly(A) tracts.

A range of transcripts have been identified in the axoplasm of several kinds of neurons (reviewed by Van Minnen, 1994; Mohr, 1999; Mohr and Richter, 2000; Alvarez et al., 2000) but, to understand their physiological role in the axon, it is essential to determine the presence, abundance and distribution of specific transcript species. Because the structure that is most remote from the cell body in a growing axon is the dynamic, actin-rich growth cone, actin and its accessory proteins are good candidates for local axonal translation during outgrowth (Bamburg and Bray, 1987; Bassell et al., 1998). We have previously detected β-actin mRNA in axons of chick sympathetic neurons indirectly using RT-PCR (Olink-Coux and Hollenbeck, 1996), and Bassell et al. (Bassell et al., 1998) have detected its presence directly in axons of rat cortical neurons. Here, we set out to examine directly the axonal localization and subcellular distribution of mRNAs for β-actin and ADF, an actin binding protein that is essential for normal actin organization and is involved in growth cone function (Bamburg, 1999; Meberg and Bamburg, 2000).

Using antisense probes and two-stage enzymatic amplification, we performed in situ hybridization against β-actin and ADF mRNAs. Two RNA probes were used to detect β-actin transcripts: a 537 bp RNA probe complementary to the β-actin 3′ UTR and a 1690 bp RNA probe complementary to the entire β-actin sequence. The former was used because its sequence is unique and thus lacks similarity to that of other actin isoforms; the latter, because its greater length provides increased signal. The hybridization conditions were more stringent than those used for the oligo d(T) or oligo d(A) probes in order to denature the long RNA probes completely and thereby maximize the specificity of binding. Both the 1690 bp and the 537 bp probes successfully detected β-actin transcripts in axons of chick sympathetic neurons (Fig. 2A-D). Cell bodies were very brightly stained and discrete, punctate signals were detected in axons. These appeared most commonly at axonal branch points, varicosities and growth cones, but also along the length of the axon shaft (Fig. 2B,D). As with oligo dT probes, the signal was strongest in the 100 μm or so of axon nearest the cell body but was found along the length of longer axons. The overall pattern of staining was indistinguishable between the 1690 bp and 537 bp probe, implying that the signals resulted from hybridization to the same pool of mRNA (see also Table 2). The higher signal intensity provided by the 1690 bp was probably due to the increased signal provided by the longer probe, rather than cross-hybridization with γ-actin transcripts, which have been reported to be absent from the axon (Bassell et al., 1998).

The ADF antisense RNA probe of 723 bp was complementary to 500 bp of coding sequence and 223 bp of 3′ UTR. This region was selected because of its unique sequence. As for β-actin, ADF probes revealed a discrete, punctate mRNA distribution in axons, with signal detected at branch points, varicosities, growth cones and along the axonal shaft (Fig. 2E,F). However, the number and intensity of the signal puncta were less than those seen with β-actin probes (Table 1). Staining of the cell body was less intense than β-actin staining as well, probably because of the lower level of expression of ADF in chick sympathetic neurons. The specificity of our single-transcript detection was confirmed by RNase treatment (Fig. 2G,H) and sense probe controls (Fig. 2I,J), as well as by a complete lack of signal when an antisense RNA probe for the bacterial neo gene was used, or excess amount of unlabeled antisense RNA probe was added to compete out labeled antisense probe (not shown).

β-Actin and ADF mRNAs were detected in all of the cell bodies and most, but not all, of the axons in sympathetic cultures (70-81%; Table 1). Because recent studies have indicated that the presence of β-actin mRNA in axons of cultured neurons is not constitutive but is regulated by neurotrophin signaling (Zhang et al., 1999), we examined the distribution of mRNA in sympathetic neurons containing more than one axon. In cells selected at random, each axon was scored for whether or not it contained β-actin or ADF mRNA (Table 2). If the presence or absence of mRNA in an axon is a property of the neuron as a whole then there should be a bias towards neurons having either all or none of their axons containing mRNA. However, if the entry of mRNA into each axon is a property of the axon alone then the pattern of presence or absence of mRNA in axons of multiple-axon neurons should follow a binomial distribution. We found that the data were biased significantly towards neurons having all or none of their axons containing β-actin and ADF mRNAs (Table 2).

The distribution of total mRNA and recent results on the arrangement of ribosomes in the axon (Koenig and Martin, 1996; Koenig et al., 2000) both suggested that the distribution of individual mRNA species in the axon would be non-uniform. We examined our data to test the hypothesis that mRNA for β-actin and ADF are not only distributed irregularly but are also concentrated specifically in actin-rich regions of the axon. The signal was clearly concentrated in two such regions, axonal branch points and growth cones, as well as in varicosities (Fig. 2). We quantified this distribution in 100 randomly selected neurons for each hybridization protocol and found that 78-85% of the β-actin and ADF mRNA was concentrated in these three regions, with the remainder distributed along the entire axon shaft (Table 1). To quantify the β-actin and ADF mRNA concentrations in these different regions of the axon, we measured the projected areas occupied by each region in sympathetic cultures and calculated the concentrations of mRNA signal per μm² of projected area. The concentrations in branch points, growth cones and varicosities were all significantly higher than that in the axon shaft (Table 3). In addition, when the relative thickness of the growth cone was factored in to calculate the relative concentrations of mRNA signal per μm³, the branch points, growth cones and varicosities all had β-actin and ADF mRNA concentrations approximately 100 times higher than those of the axon shaft as a whole (Table 3).

When whole sympathetic ganglia are cultured for 4-5 days, they extend radial halos of axons that extend 4-5 mm (Fig. 3). By maintaining the ganglia on a poly-L-lysine-treated substratum and exposing them intermittently to the DNA synthesis inhibitor Ara-C, the presence of non-neuronal cells in the halos can be suppressed (Fig. 3A-D). In order to determine whether sympathetic axons are capable of translating their mRNA independently of the cell body, we performed [³⁵S]-methionine/cysteine metabolic labeling on ganglion explants with and without prior surgical removal of the cell body mass (see Materials and Methods). When labeled cultures were subjected to emulsion autoradiography, dense silver grains were detected throughout not only the intact ganglia (Fig. 3E) but also the axonal halos that had been exposed to [³⁵S]-methionine/cysteine in the absence of cell bodies (Fig. 3F). There was no obvious nonuniform distribution of axonally synthesized proteins along the length of axons in the halos. Autonomous axonal protein synthesis was abolished when axonal halos were treated with the eukaryotic translation inhibitor cycloheximide, but remained at control levels when they were treated with the prokaryotic and mitochondrial translation inhibitor chloramphenicol (Fig. 3G,H). Thus, the axonal protein synthesis results from cytoplasmic, not mitochondrial, translation.

To compare mRNA translation in the axons versus cell bodies we subjected the explant cultures, either with or without removal of their cell body mass, to [³⁵S]-methionine/cysteine metabolic labeling for 2-3 hours and then harvested the explants. When these were analysed for their content of newly synthesized proteins by SDS-PAGE and autoradiography, dozens of polypeptides were apparent in both preparations (Fig. 4A, lanes 1,2). By this measure, new protein synthesis in both whole ganglia and axonal halos was eliminated by cycloheximide treatment (Fig. 4A, lanes 3,4) during [³⁵S]-methionine/cysteine exposure but not by chloramphenicol treatment (Fig. 4A, lanes 5,6). Thus, many individual mRNA species must be both present and actively translated in these axons. However, the proteins synthesized in the two preparations were not identical: autoradiograms showed that synthesis of several proteins was confined to or highly enriched in the cell bodies; there were also several proteins whose synthesis was nearly or entirely confined to the axonal compartment (Fig. 4B). This indicates that certain transcripts are selectively and almost entirely transported into the axon, that they are selectively translated there or both. The identities of these proteins are not known at present but they include two prominent polypeptides with estimated sizes of 24 kDa and 62 kDa. As described below, the stoichiometry of synthesis of the neurofilament proteins also differed substantially between cell bodies and axons (Fig. 5).

What proportion of the sympathetic neurons' total capacity for protein synthesis resides in the axons of these explant cultures? In several experiments, we determined the number of [³⁵S]-methionine/cysteine-labeled axonal halos necessary to equal the amount of labeled protein in a single [³⁵S]-methionine/cysteine-labeled whole ganglion (Fig. 4, lanes 1,2). This number ranged between 20 and 68, meaning that axonal protein synthesis in these explants accounted for 1.5-5.0% of total neuronal protein synthesis. This estimate is conservative for the same reason that it is variable: in removing the cell body masses from the ganglion explants, we worked under a microscope and were careful to remove all of the cell bodies, even at a sacrifice of a variable portion of the axonal halo. The radius of the axonal halo averaged 5 mm and we removed on average the most proximal 0.5 mm of the halo in the course of excising the cell body mass. Thus, 1.5-5.0% places a lower limit on the contribution of axonal protein synthesis to these neurons. We calculate the specific translation activity per unit volume of axonal cytoplasm as follows: in our cultures, micrographs show that the average length of our axonal halos after removal of the cell body mass was 4.5 mm, the average diameter of the cell body is 12 μm and the average nuclear diameter is 8 μm; electron micrographs show that the average axon diameter is 0.4 μm (Hollenbeck, 1993b). Therefore, the non-nuclear volume of cytoplasm in the cell body is 637 μm³, or 53% of total cytoplasmic volume in the neuron, whereas that of the axon is 565 μm³, or 47% of the total volume. Thus, in these sympathetic cultures, the cytoplasmic volume of the axon is approximately equal to that of the cell body. Because these axons contain 1.5-5.0% of the total protein synthesis in approximately 50% of the non-nuclear volume of the cell, the amount of protein synthesis per unit volume in the axoplasm is 3-10% that of the cell body cytoplasm.

Although there is significant amount of olfactory marker protein mRNA in axons of rat olfactory neurons, there is no evidence that they are actually translated to produce protein (Wensley et al., 1995). In order to see whether β-actin and ADF transcripts detected in axons are actively translated to produce their corresponding proteins, immunoprecipitation was performed using ³⁵S-labeled axoplasmic extract. After surgery to remove their cell body masses, axonal halos were incubated in [³⁵S]-methionine/cysteine-containing medium for 5 hours, then lysed and subjected to immunoprecipitation and autoradiography to detect specific newly synthesized axonal proteins. We were able to detect newly synthesized 43 kDa β-actin in both ganglia (Fig. 5, lane 1) and axons (Fig. 5, lane 2) using AC-15 antibody, which is specific for β-actin (North et al., 1994). This result contrasts with previous studies that reported no β-actin synthesis in rat sympathetic axons (Eng et al., 1999). An unknown protein with a molecular weight of 55 kDa precipitated with β-actin. Rabbit antiserum against chick ADF also successfully precipitated labeled 18.5 kDa ADF from both ganglia (Fig. 5, lane 3) and axons (Fig. 5, lane 4), along with a minor band of 43 kDa, presumably actin (Fig. 5, round dot).

To demonstrate the axonal translation of an exclusively neuronal protein, we also immunoprecipitated neurofilament protein from metabolically labeled ganglia and axonal halos. The monoclonal antibody NR-4 binds to all three neurofilament subunit proteins: NF-L (apparent molecular weight 66 kDa); NF-M (95-100 kDa); NF-H (110-115 kDa) (Lee and Cleveland, 1996). This was also confirmed by immunoprecipitation experiments with cold chick brain extract (data not shown). All three subunit proteins were synthesized in whole ganglia, with NF-M being present in highest stoichiometry and NF-L in the lowest, representing a very small proportion of the neurofilament protein synthesized in the metabolic labeling period (Fig. 5, lane 5). However, in axons, the major neurofilament protein synthesized during the labeling period was the NF-L, with NF-M and NF-H virtually undetectable (Fig. 5, lane 6). This result not only confirms that synthesis in axonal halos is not due to non-neuronal cells but also reinforces the finding (Fig. 4) that the pattern of synthesis in the axons differs from that in cell bodies.

Even the prodigious translation capacity of the neuronal cell body and the large volume of anterograde axonal transport of protein (Grafstein and Forman, 1980; Vallee and Bloom, 1991) cannot entirely support the needs of the axon. Important events in the distal axon require local protein synthesis that is locally regulated (Campbell and Holt, 2001; Ming et al., 2002; Schacher and Wu, 2002). In this respect, the axon might have more than previously imagined in common with dendrites, where local protein synthesis plays a pivotal role in synaptic plasticity (Steward, 1997; Schuman, 1999; Martin et al., 2000). But how is mRNA organized in the axon to support local synthesis?

We have previously demonstrated that the axons of sympathetic neurons contain poly(A) mRNA and that its presence there requires intact microtubules (Olink-Coux and Hollenbeck, 1996). Here, we have shown that neurons transport mRNA into their axons very efficiently, despite their extreme asymmetry and the absolute distances involved. We detected both poly(A) mRNA and individual transcripts in 70-100% of axons, often at a distance of hundreds or thousands of micrometers from the cell body (Figs 1,2; Olink-Coux and Hollenbeck, 1996). Thus, neurons display a more spatially polarized distribution of mRNA than is seen in cell types of more modest dimensions, such as migrating fibroblasts, in which β-actin mRNA is distributed preferentially towards the leading edge of the lamellipodia in 30-60% of the cells (Sundell and Singer, 1990; Kislauskis et al., 1994). Neurons probably achieve this by using microtubule-based axonal transport of mRNA over long distances (Olink-Coux and Hollenbeck, 1996), as they do for the transport of organelles (Lane and Allan, 1998; Goldstein and Yang, 2000). This is similar to the mechanism of myelin basic protein mRNA transport in oligodendrocytes (Carson et al., 1997) and several maternal mRNAs in Drosophila early embryo (Yisraeli et al., 1989; Theurkauf and Hazelrigg, 1998; Lall et al., 1999; Brendza et al., 2000; Wilkie and Davis, 2001), but contrasts with the proposed F-actin-based mechanism of β-actin mRNA transport in fibroblasts (Sundell and Singer, 1991; Taneja et al., 1992; Bassell et al., 1994a) and the brown alga fucus (Bouget et al., 1996), and with Ash1 mRNA transport in yeast (Takizawa et al., 1997; Takizawa and Vale, 2000).

Which of the thousands of transcripts produced in neurons are present in axons? Judging from the SDS-PAGE analysis of metabolically labeled neurons, many of them: axons showed dozens of newly synthesized axonal polypeptides. Indeed, relatively few major polypeptides were synthesized mainly or exclusively in the cell body (Fig. 4). More surprising is that a similar small number were limited to the axon (Fig. 4). Other studies have detected many axonally synthesized polypeptides in rat motor (Eng et al., 1999) and sensory (Koenig, 1991) neurons, although they reported no significant difference between the compositions of newly synthesized proteins in the axons and cell bodies. However, Zheng et al. reported several proteins whose synthesis is enriched in either axons or cell bodies (Zheng et al., 2001) with molecular weights of 40-65 kDa, as seen in this study (Fig. 4), despite the different species used. Our immunoprecipitation results with newly synthesized neurofilament protein (Fig. 5) show that, even when the same group of proteins is translated in both axons and cell bodies, the pattern of synthesis may differ substantially. However, despite the number and diversity of axonal translation products, there are many proteins that are unlikely to be synthesized there, including those involved in nuclear structure, in chromatin formation and in endoplasmic-reticulum-dependent translation and protein folding.

In the case of cytoskeletal proteins, we have previously shown by reverse-transcription PCR that mRNA encoding one major protein (β-actin) was present in axons, whereas that encoding α-tubulin was not (Olink-Coux and Hollenbeck, 1996). Here, we have shown directly that mRNAs for β-actin and ADF are present and widespread in most axons, and that they, along with the NF-L transcript, are actively translated to produce their respective proteins. Because β-actin (Herman, 1993) is abundant in neuronal growth cones (Bassell et al., 1998), as it is at the leading edge of other moving cells (Hoock et al., 1991; Bassell et al., 1998), the presence of its mRNA and its active translation there probably reflect a requirement for rapid regulation of β-actin translation where its concentration and dynamics are greatest (Lawrence and Singer, 1986; Kislauskis et al., 1997). The presence and translation of both β-actin and ADF mRNA in the axon of sympathetic neurons supports this idea, because ADF, like β-actin, is concentrated in the growth cone (Bamburg and Bray, 1987), where it is believed to modulate the behavior of the actin cytoskeleton (Bamburg, 1999; Meberg and Bamburg, 2000).

However, are mRNAs limited to the growth cone region of the distal axon, or to the regions where the proteins that they encode are most concentrated? Our data show that the densities of both β-actin and ADF transcripts are about two orders of magnitude higher in the growth cone region than they are in the axon as a whole. However, they are also highly concentrated in axonal branch points and small varicosities, sometimes at a considerable distance from the growth cone (Fig. 2, Table 3). This mRNA could have been in transit toward the distal axon at the time of fixation. Axonal mRNA transport is microtubule dependent, whereas these regions are actin rich and microtubule poor. Thus, mRNAs could dwell longer there, as do many organelles undergoing axonal transport. It is also possible that mRNAs are specifically docked in actin-rich regions, as has been proposed in fibroblasts (Bassell et al., 1994b), Xenopus oocytes (Yisraeli et al., 1989) and cortical zone of goldfish Mauthner cells (Muslimov et al., 2002), although axonal mRNA distribution has been shown to be largely dependent on microtubules in the chick sympathetic neurons (Olink-Coux and Hollenbeck, 1996).

Our data make it unlikely that the distribution of mRNA indicates the sites within the axon where protein synthesis occurs. The total mRNA, β-actin mRNA and ADF mRNA that we detected are highly concentrated in a few regions (Table 3), whereas the axonally synthesized protein is uniformly dispersed along the axon (Fig. 3). However, mRNA exists in neurons in several forms and the distribution of actively translated mRNA could be different from the distribution of total mRNA found here. Morphological studies have shown that complexes of ribosomes reside in the cortical region of vertebrate axons and perhaps contain translatable mRNA (Koenig et al., 2000). In addition, mRNA granules detected by microscopy in rat cortical neurons (Knowles et al., 1996) and mouse oligodendrocytes (Barbarese et al., 1995) contain both mRNA and components of the translation machinery such as ribosomes and elongation factor. mRNA-containing structures such as these could be either translationally active or quiescent; some evidence suggests that they might be competent for translation upon an appropriate physiological cue (Krichevsky and Kosik, 2001; Zhang et al., 2001). Finally, evidence from live neurons indicates that mRNA and other translation components can undergo rapid transport in a particulate form (Knowles et al., 1996). The mRNA that we have detected in axons could be in one or more of these forms.

Neurons with several axons were more likely than expected by chance assortment to have β-actin or ADF transcripts in all or none of their axons. This suggests that the transport of individual transcripts into the axon is a property of the entire neuron rather than an autonomous feature of each axon. Furthermore, it suggests that the proposed mechanism for recruiting β-actin transcripts to the axon, neurotrophin signaling (Zhang et al., 1999; Zhang et al., 2001), is likely to act at the level of the cell body rather than on individual axons. This would be consistent with evidence that some intracellular signaling events stimulated by neurotrophin treatment of sympathetic axons are mediated through changes occurring in the cell body (MacInnis and Campenot, 2002).

Whatever the nature of the signal that stimulates mRNA transport into the axon, what qualifies a particular transcript to respond? Cis-acting elements in mRNAs play a key role in their transport and localization in a range of oocytes, embryos and somatic cell types (Palacios and Johnston, 2001; Jansen, 2001). In particular, β-actin localization in leading edge of fibroblast requires specific sequences in the 3′ UTR, the `zipcode' (Kislauskis et al., 1994), and these sequences might be responsible for the transport of β-actin mRNA into axons (Olink-Coux and Hollenbeck, 1996; Zhang et al., 2001) (M. Olink-Coux and P.J.H., unpublished). However, the ADF transcript, which has neither the β-actin zipcode nor any other homology with the UTR or coding region of β-actin mRNA, is nonetheless efficiently transported into neurons. This, along with the sheer number of different transcripts present in axons (Fig. 4) (Koenig, 1991; Eng et al., 1999; Zheng et al., 2001) indicates that there are likely to be several different sequence motifs that can mediate mRNA transport into the axon. Furthermore, the rather small number of transcripts that seem to be confined to the cell body raises the possibility that specific sequence motifs could also prohibit entry into the axon.

Although some vertebrate axons contain abundant mRNA but little or no protein synthesis (Wensley et al., 1995), sympathetic neurons clearly contain mRNA that is actively translated (Figs 3, 4). However, how large is the contribution of axonal translation to the neuron as a whole? We found that synthesis in the axons of our cultures ranged from 1.5% to 5.0% of the total for the neurons, a similar value to the 1.4-4.1% previously reported for goldfish Mauthner axons (Alvarez and Benech, 1983) and a higher value than the 0.5% for rat sympathetic neuronal culture (Eng et al., 1999), probably because of differences in the lengths of the axons and the proportions of the cell volume contained there. We further calculated the specific translation activity per unit volume of axonal cytoplasm to be 3-10% that of cell body cytoplasm. Axons that contain just one-tenth of the protein synthetic capacity per unit volume of the cell body must nonetheless explain the fact that local protein synthesis is required to support axonal growth and development, and synaptic plasticity (Martin et al., 1997; Casadio et al., 1999; Campbell and Holt, 2001; Zheng et al., 2001; Beaumont et al., 2001; Ming et al., 2002; Schacher and Wu, 2002; Brittis et al., 2002; Zhang and Poo, 2002). Furthermore, if total axonal translational capacity scales up proportionally with axonal volume then sympathetic axons with length much greater than 4.5 mm would carry out a larger proportion of the neuron's total protein synthesis. For example, for these neurons, the ratio of axonal to non-nuclear cell body volume reaches 10:1 at an axon length of ∼50 mm. If linear scaling occurs, this is the axon length at which the ratio of axonal to cell body translation would reach 1:1. It is possible that, as axon length increases or as neurons form synapses and mature, the specific translation capacity of axonal cytoplasm declines. However, the discovery of ribosomes and translation factors in adult vertebrate axons (Koenig et al., 2000; Zheng et al., 2001) makes it unlikely that axonal protein synthesis is a property only of developing or regenerating neurons.

We thank J. R. Bamburg for his generous gift of antiserum against ADF and S.-J. Kang for performing emulsion autoradiography experiments. This work was supported by grant NS27073 from the NIH.