A crucial role for hnRNP K in axon development in Xenopus laevis

Axonal outgrowth depends on coordinating the expression of functionally interrelated genes. For example, synthesis of cytoskeletal proteins and the associated factors that regulate and organize their assembly must be matched with bouts of axonal extension to ensure that the axon is neither under- nor oversupplied. The rapidly shifting dynamics of this outgrowth makes it difficult to envision how this balance is achieved through transcriptional control alone. Placing the synthesis of cytoskeletal proteins under direct translational control would both complement transcriptional regulatory mechanisms and offer advantages to the growing axon for matching expression with demand. The dramatic shift from untranslated to translated pools undergone by the middle neurofilament (NF-M) mRNA in retinal ganglion cells during optic nerve regeneration indicates that proteins involved in building the axon are indeed under strong translational control(Ananthakrishnan et al., 2008). However, the identities of specific elements involved in the translational control of such messages and the degree to which such elements are necessary for axon outgrowth remain unresolved issues.

The life histories of mRNAs as they pass through the cell are largely governed by continually evolving sets of ribonucleoproteins (RNPs)(Moore, 2005). One idea is that RNPs may act as components of regulatory modules targeting subsets of specific functionally related messages, providing an additional level of control beyond transcription (Keene and Tenenbaum, 2002). In the nervous system, RNPs are already known to play important roles in neural plasticity and development(Perrone-Bizzozero and Bolognani,2002; Si et al.,2003; Yao et al.,2006). Heterogeneous nuclear RNP K (hnRNP K) is the founding member of the K-homology domain family of RNPs. Although abundantly expressed in nervous system, its role there is not understood. It binds directly to the NF-M 3′-untranslated region (3′-UTR), along with additional RNPs, including HuB and hnRNPs E1 and E2(Antic et al., 1999; Thyagarajan and Szaro, 2004). hnRNP K is thought to function as a central scaffolding component of evolving mRNP complexes whose compositions are modulated by multiple kinases, thereby providing a mechanism whereby mRNA fate can be coupled with cell-signaling events (Bomsztyk et al.,2004).

Two observations highlight the potential importance of the interactions of hnRNP K with neuronal mRNAs during development. In neuroblastoma cells, hnRNP K directly antagonizes Hu binding to p21 mRNA to promote proliferation over differentiation (Yano et al., 2005). In mammalian brain, it associates with the mRNAs of three NF subunits [light (NF-L), medium (NF-M) and heavy (NF-H)] more strongly in postnatal than in adult neocortex(Thyagarajan and Szaro, 2008). To explore its role in the intact developing nervous system, we first examined the cellular distribution of hnRNP K in developing Xenopus and then suppressed its expression with antisense morpholino oligonucleotides (MOs)injected into blastomeres. Suppressing hnRNP K expression both blocked axonal outgrowth and inhibited the translation of NF-M. These observations are consistent with the hypothesis that hnRNP K plays an essential role in the translation of a subset of proteins involved in building the axon.

Two-cell stage periodic albino Xenopus laevis embryos(Hoperskaya, 1975) were microinjected into one or both blastomeres (10 nl each; 1 ng/nl) with MO(Gervasi and Szaro, 2004). The following MOs (Gene Tools) were used: antisense hnRNP K MO1, 5′GCT GCT ACC TTT CTC CTA CGC CGA C3′ starting from nucleotide -54; a second,non-overlapping antisense hnRNP K MO (MO2), 5′TC TTC CTG CTC TGT CTC CAT CTT CT3′ (the initiation codon is underlined); antisense hnRNP E MO, 5′GGC TAT TGT CAG AAG CTG TAC AAA G3′ starting from nucleotide -50; and a standard control MO, 5′CCT CTT ACC TCA GTT ACA ATT TAT A3′. Fluorescent dextrans (FDx) were co-injected to label cells descended from the injected blastomere (0.75% w/v lysinated fluorescein-isothiocyanate Dextran; 0.1-0.25% Cascade Blue Dextran, Molecular Probes).

For rescue experiments, cDNA spanning the complete coding sequence of X. laevis hnRNP K (GenBank BC044711) was obtained by RT-PCR(Superscript II RT; Platinum Pfx DNA polymerase, Invitrogen) from total RNA of stage 29/30 embryos using the following primers, which had NotI sites added to their 5′-end to facilitate cloning: 5′ATA GCG GCC GCT AAA AGA AGA TGG AGA CAG AGC AGG3′ (forward); and 5′ATA GCG GCC GCG AAG TCA AAA CCC ATA AGA ATA ATC3′ (reverse). PCR products were ligated into the NotI site of a modified pGem3Z expression vector (pSP6-XhnRNPK),downstream of the SP6 promoter and upstream of the 3′ UTR of rabbitβ-globin (Lin and Szaro,1996). The insert sequence was confirmed to match the GenBank sequence. The plasmid was subsequently linearized with XhoI for use as template for in vitro transcription of 5′-capped RNA (mMESSAGE mMACHINE SP6, Ambion). Rescue was performed by co-injecting two-cell embryos with hnRNP K-MO1 (10 ng) plus hnRNP K RNA (50-1000 pg).

Dissociated embryonic spinal cord-myotomal cultures were prepared from spinal cord, including extreme caudal hindbrain together with surrounding myotomes, of stage 22 embryos. Each culture was prepared from a single embryo and grown at 22.5°C on 35 mm polystyrene dishes (Δ plastic, Nunclon Nalge International), as described (Tabti and Poo, 1991; Undamatla and Szaro, 2001).

The following antibodies were used (#1-7, mouse monoclonal; #8, rabbit antiserum): (1) Xenopus NF-M, 2 μg/ml [RMO270(Szaro et al., 1989; Wetzel et al., 1989)]; (2) a neuronal β-tubulin isotype, 1:100 [JDR.38B, Sigma(Moody et al., 1996)]; (3)hnRNP K, 1:100 (Santa Cruz Biotechnology); (4) HNK-1, undiluted hybridoma supernatant (Nordlander, 1989; Szaro et al., 1991); (5) a phosphorylated epitope of the C-terminal tail domain of Xenopus NF-M,1:200 [S6 (Szaro and Gainer,1988)]; (6) Xenopus MAP-1, 1:200 [8D-12(Lin and Szaro, 1996)]; (7)glyceraldehyde phosphate dehydrogenase (GAPDH), 5 μg/ml (clone 6C5,Ambion); and (8) Xenopus peripherin, 1:2000(Gervasi et al., 2000). Alexa Fluor phalloidin 555 and DAPI (Molecular Probes) were used at 25 μl/ml and 300 nM to stain F-actin filaments and nuclei, respectively.

The numbers of embryos injected with each MO and analyzed in whole-mount are presented in Table 1. For whole-mount immunohistochemistry, embryos were fixed in phosphate-buffered(0.1 M sodium phosphate, pH 7.4) 10% formalin and processed as described,using biotinylated secondary antibodies and fluorescent avidins(Dent et al., 1989; Gervasi et al., 2000). Fluorescence was quantified from confocal microscope images using Metamorph software (version 4.5.6, Molecular Devices). For frozen sections (12 μm transverse), juvenile frogs were anesthetized, perfusion fixed with paraformadehyde, and processed for immunoperoxidase staining using diaminobenzidine with nickel chloride as the chromogen(Szaro and Gainer, 1988). For dissociated cell cultures, cells were fixed 24 hours after plating for 2 hours in phosphate-buffered 1% formaldehyde/2% sucrose at room temperature and processed for immunofluorescence procedures using biotinylated secondary antibodies and rhodamine-avidin (Undamatla and Szaro, 2001). For phalloidin staining, cultures were fixed in phosphate-buffered 4% paraformaldehyde/2% sucrose and processed as described(Smith et al., 2006).

For western blots, 60-80 stage 29/30 embryos per sample were homogenized in SDS-urea sample buffer, of which 45 μg protein per lane was separated by 7.5% SDS-PAGE, transferred overnight (60V, 12°C) to nitrocellulose, and processed for immunoperoxidase staining(Zhao and Szaro, 1994). Blots were processed separately with primary antibodies to either hnRNP K or NF-M then stripped and reprocessed for GAPDH. Staining was visualized by chemiluminescence (SuperSignal West Pico; Pierce) and imaged on a ChemiDoc XRS System (Biorad, Hercules, CA). Band intensities were quantified using Quantity One software (Biorad).

Digoxigenin-labeled cRNA probes were transcribed using a Genius 4 Kit(Roche) from plasmids containing Xenopus peripherin (2.0 kb) and NF-M cDNAs (clone 6B-1a of NF-M1, 1.7 kb)(Gervasi et al., 2003). Peripherin sense cRNA was used to control for non-specific hybridization. Whole-mount in situ hybridization was performed on stage 37/38 tadpoles (Shain and Zuber,1996): hybridization, 18 hours at 60°C; probe concentration,1-2 mg/ml; high stringency washes (50% formamide/2×SSC at 60°C for 30 minutes). Anti-digoxigenin Fab fragments coupled to alkaline phosphatase were used to visualize staining (Genius 3 Kit; Roche). Fluorescence in situ hybridization (FISH) was performed using the above probes, hybridization and wash conditions, followed by immunohistochemistry with peroxidase-conjugated anti-digoxigenin and tyramide-FITC(Davidson and Keller, 1999). For combined FISH-immunohistochemistry, preparations were processed for FISH first, then for rhodamine-conjugated immunofluorescence and finally for DAPI staining.

Co-immunoprecipitation (co-IP) of hnRNP K and RNAs was carried out on samples from juvenile frog brain using protein A-sepharose beads (CL4B, Sigma)coated with purified monoclonal antibodies to hnRNP K (clone 3C2; Santa Cruz Biotechnology) or β-galactosidase (Z3783; Promega)(Tenenbaum et al., 2002; Thyagarajan and Szaro, 2008). Prior to IP, 10% of the sample was used (total input control, TIC) to ascertain mRNA levels in the original sample. For reverse transcriptase polymerase chain reaction (RT-PCR), the primers for the first round of PCR (30 cycles of 95°C for 1 minute; 55°C for 30 seconds; 72°C for 1 minute) were as follows (sense and antisense, respectively): NF-M,5′ATT CAG AGG AGC AAA AGG AT3′ and 5′TCC TGC TGC TTG CTG CAA TT3′; peripherin, 5′TTC ATG AGG AGG AAC TCA AT3′and 5′TCC TCC GAC TCT TGT ACG TT3′; EF-1α,5′ACC CTG CTG GAA GCT CTT GA3′ and 5′ GCA GAC GGA GAG GCT TAT CAG T3′. For the second round of amplification for peripherin (15 additional cycles: 95°C for 1 minute; 55°C for 30 seconds; 72°C for 1 minute), 5 μl of the first reaction and a second, nested set of peripherin primers was used: 5′GGT CCG CTA CCT GGA GCA GC3′ and 5′ACA TTG AGC AGG TCT TGG TA3′. PCR products were visualized on a 1% agarose-TBE gel stained with ethidium bromide.

Embryos were collected at stages 25 and 29/30 (5 embryos per sample) and processed for polysomal profiling by ultracentrifugation on a 5-56% linear sucrose gradient (Meyuhas et al.,1996; Ananthakrishnan et al.,2008). Cellular nuclei were first separated from the cytosol by centrifugation (285 g, 4°C, 2 minutes). To quantify nuclear and cytosolic peripherin and NF-M mRNA, total RNA was isolated from the nuclear pellet and from 10% of the cytosolic fraction(RNeasy Micro Kit, Qiagen). The remainder of the cytosol was used for polysome profiling. For each fraction, a spectrophotometric reading for total RNA (2μl, A₂₆₀; NanoDrop Technologies) was taken, and peripherin and NF-M mRNA levels were quantified from the remainder by quantitative real-time RT-PCR (qRT-PCR; Applied Biosystems, ABI SDS 7900HT; ABI SDS software, version 2). The following primers were used(forward and reverse): NF-M, 5′ACC GAA GAA GTC TTC AGG AAT AGG TA3′ and 5′CAC TGT CCG GCT CAT GCA3′; peripherin,5′GCT GTA AGC GAC TTT GGT GCT A3′ and 5′CAA GGC ATG ATG GGA CAG AGT3′. Taqman probes were as follows: NF-M, 5′-6FAM CTG ACC GAG GCT GCA GA MGBNFQ-3′; peripherin: 5′-6FAM CAT TCC CTG CCT CTG TGT ATT GGC TGG TAMRA-3′.

Previous biochemical studies had shown that hnRNP K in Xenopus is initially expressed as a maternal protein in oocytes and is present throughout early development, with expression increasing as cells differentiate(Marcu et al., 2001; Iwasaki et al., 2008). Results from whole-mount immunocytochemistry were consistent with these earlier biochemical studies and further found that hnRNP K was present in all the germ layers from blastulae stages onwards (illustrated at stages 10, 15, 22, 37 and 42 in Fig. 1). Throughout development, hnRNP K was prominently expressed in nuclei. From gastrula stages onwards, hnRNP K became progressively more highly expressed in ectodermal and mesodermal derivatives, such as neural plate and somites, than in endodermal ones. This trend continued through early tadpole stages, where hnRNP K staining in differentiating tissues, especially in nervous system, became increasingly more cytoplasmic (Fig. 1F).

To improve resolution of intracellular structures, we next examined hnRNP K expression in dissociated embryonic spinal cord-myotomal co-cultures. At 24 hours after plating, hnRNP K immunostaining colocalized with DAPI-stained nuclei in both neurons and muscle cells(Fig. 2A-C); it was detectable neither in neurites nor in growth cones(Fig. 2C). In histological sections of post-metamorphic frog nervous system(Fig. 2D,E), the relative intensity of nuclear versus cytoplasmic staining varied extensively among cell populations. In retinal ganglion cells, for example, immunostaining was generally more intense in cytoplasm than in nuclei(Fig. 2D), and in spinal cord motoneurons, intensities of cytoplasmic when compared with nuclear staining were more comparable (Fig. 2E). Nuclear versus cytoplasmic staining varied considerably among other brain and spinal cord cell types, with numerous examples of cells having greater nuclear than cytoplasmic staining, and vice versa. These data are consistent with the idea that hnRNP K shuttles between the nucleus and cytoplasm, escorting its RNA targets to assist their recruitment into mRNP granules and ribosomes(Piñol-Roma and Dreyfuss,1992; Moore,2005). hnRNP K immunostaining was not found in axons (e.g. optic nerve and ventral spinal cord white matter; see Fig. S1 in the supplementary material), indicating that hnRNP K is unlikely to play a major role in axonal transport of mRNAs.

To study its function during development, we suppressed hnRNP K expression with antisense MOs. The bilateral symmetry of the embryo enabled the application of anatomical criteria to identify affected cellular subtypes in unilaterally injected embryos; FDx was co-injected to identify the injected side. To control for inadvertantly targeting genes other than hnRNP K, we injected two separate hnRNP K antisense MOs targeting non-overlapping regions of the 5′-UTR (Heasman,2002). As an additional control, we injected a third MO targeting the 5′-UTR of Xenopus hnRNP E(Gravina et al., 2002), the closest relative of hnRNP K(Ostareck-Lederer et al.,1998). A standard control MO was added to control for non-specific MO effects.

Up to gastrula stages, neither hnRNP K MO had any discernible effect on hnRNP K expression (Fig. 3A1,A2). This was most likely because of a large store of maternal protein (Iwasaki et al.,2008). By neural plate stages, however, expression was strongly suppressed throughout the injected half of the embryo(Fig. 3B1,B2). The efficacies of the two hnRNP K MOs at suppressing hnRNP K expression were similar and the resulting phenotypes (described later) were indistinguishable. Neither hnRNP K MO had any effect on hnRNP E expression (not shown),nor did the hnRNP E and standard control MOs have any effect on hnRNP K (see Fig. S2 in the supplementary material). Suppression of hnRNP K expression persisted through at least stage 40(Fig. 3C-E). The suppression of hnRNP K was also assessed by western blots of stage 29/30 embryos (see Fig. S3 in the supplementary material). Neither unilateral nor bilateral injections of hnRNP K MO1 had any effect on GAPDH expression. Embryos receiving control MO expressed 97±5% (s.d.) (three blots) as much hnRNP K as uninjected ones, and embryos receiving unilateral or bilateral injections of hnRNP K MO1 expressed 57±6% and 8±3%, respectively, as much as uninjected ones.

By external criteria, hnRNP K MO-injected embryos developed normally through stage 20. At stage 22, the injected sides of hnRNP K MO-injected animals began to constrict, leading to a `bent' phenotype. Despite this bending and additional mild defects in anterior cranial and optic vesicle formation, rates of survival through stage 40 (assayed in 19 spawnings) were indistinguishable between hnRNP K MO-injected embryos and controls:hnRNP K MO1, 91±4% of 2042 embryos; hnRNP K MO2, 92±3% of 1039;control MO, 94±5% of 1739; FDx alone, 93±6% of 1106; and uninjected embryos, 94±4% of 2598. None of these survival rates differed significantly (P>0.1, one-way ANOVA).

We next examined effects of hnRNP K knockdown in neurons by whole-mount immunostaining of tadpoles (stage 37/38) for several neuronal antigens typically found in both perikarya and axons. In all cases, phenotypes obtained with either of the two hnRNP K MOs were indistinguishable. We first looked at a phosphorylation-independent epitope of NF-M (RMO270). In contrast to uninjected and control MO-injected animals(Fig. 4A), NF-M immunostaining of optic (not shown), motor and sensory nerves was severely compromised on the hnRNP K MO-injected side (MO1, Fig. 4B: MO2, Fig. 4C). Rostrocaudal spinal tracts on the injected side were also stained less intensely than on the uninjected side (separated by the broken line, Fig. 4B,C). Because these tracts contain decussated axons, we believe that such staining on the injected side may have originated from the uninjected side. The absence of any NF-M immunostained perikarya on the injected side supported this possibility and further indicated that NF-M protein expression was also likely to have been affected. Western blots of stage 29/30 embryos were used to confirm this suppression (see Fig. S3 in the supplementary material). Whereas control MO-injected animals expressed 88±7% (±s.d.) (three blots) as much NF-M as uninjected ones, animals unilaterally and bilaterally injected with hnRNP K MO1 expressed only 56±4% and 9±4% as much,respectively.

As the absence of NF-M stained axons might therefore have been due to the suppression of NF-M expression, we next stained animals for three additional neuronally expressed antigens normally present in neuronal perikarya and axons(Fig. 4D1-H). Two of these were cytoskeletal proteins, neuronal β-tubulin (visualized with antibody JDR.38B) and peripherin (visualized with a rabbit antiserum). These proteins are abundant in all developing Xenopus axons from the time of neurite initiation onwards (Moody et al.,1996; Gervasi et al.,2000; Undamatla and Szaro,2001). The HNK-1 antibody targets an extracellular carbohydrate moiety, and in Xenopus developing spinal cord, it stains terminally differentiated neurons and their axons(Nordlander, 1989). On the uninjected side, all three exhibited patterns of staining that are typical for these antibodies: i.e. clusters of cell bodies in spinal cord and peripheral ganglia, as well as robustly stained CNS axon tracts, motor axons and peripheral sensory axons. On the injected side, clusters of neuronal perikarya were well stained, indicating that the hnRNP K MO had little effect on their overall expression. This conclusion was confirmed by comparing the relative intensities of peripherin immunofluorescence quantitatively from confocal microscopic images. On the injected side, the mean pixel intensity per cell was 95% that of the uninjected side [142±33 (s.d.) in 120 cells versus 150±26 in 114 cells, respectively]. Thus, the absence of axons cannot reasonably be attributed to faint immunostaining. On the hnRNP KMO-injected side, spinal cord CNS tracts were only very lightly stained, and only a very few short wispy peripheral axons were seen emanating from spinal cord and cranial ganglia. In all control MO-injected cases, staining with these antibodies was bilaterally symmetric and indistinguishable from that of uninjected animals. The incidence of these effects is summarized in Table 2. Because staining with three independent markers failed to reveal axons but was unaffected in perikarya, we concluded that hnRNP K knockdown compromised axonal outgrowth but not neuronal determination.

To confirm that effects on axon outgrowth and NF-M expression were specific to hnRNP K, we co-injected in vitro transcribed Xenopus hnRNP K RNA lacking the MO targeting site (50-1000 pg) plus hnRNP K MO1 (10 ng). Injection of 1000 pg of RNA proved lethal (33 embryos), but at 100 and 250 pg per embryo, hnRNP K expression was largely restored (30 embryos each). Co-injected animals continued to bend towards the injected side to varying degrees among animals, but less so than in animals receiving hnRNP K MO alone,providing external confirmation of partial rescue. hnRNP K immunostaining intensity on the injected side also varied among animals but was generally distributed as in normals (Fig. 5A). For example, hnRNP K immunostaining on the injected side appeared in somitic myocyte nuclei, although myotomes were generally smaller and nuclei were less well aligned than normal.

Next, embryos co-injected with hnRNP K MO and 100-250 pg of RNA were immunostained in whole mount for neuronal β-tubulin and NF-M. Both peripheral axon outgrowth and NF-M expression were effectively restored(Table 2; Fig. 5B,C), with the degree of restoration inversely correlating with the severity of the `bent' phenotype. Thus, MO effects on axon outgrowth and NF-M protein expression can be attributed to the suppression of hnRNP K expression.

hnRNP K has been implicated in multiple aspects of post-transcriptional regulation (Bomsztyk et al.,2004). To gain insights into its specific role in post-transcriptional gene expression, we examined effects of hnRNP K knockdown on NF-M expression, a known target of hnRNP K. We compared these effects with those on peripherin, a functionally similar protein whose RNA is not a target. NF-M is a Type IV and peripherin is a Type III neuronal intermediate filament protein. Although both are found together in developing axons, peripherin is generally expressed earlier than NF-M, and thus their expressions are separately controlled (Gervasi et al.,2000).

hnRNP K has been shown in vitro in both Xenopus and rat to bind directly to the NF-M 3′-UTR, but, as of yet, this interaction has only been demonstrated to occur endogenously in rat (Thyagarajan and Szaro, 2004; 2008). To confirm an endogenous association in Xenopus, hnRNP K-containing RNP-RNA complexes extracted from juvenile frog brain were co-immunoprecipitated with anti-hnRNP K, then assayed by RT-PCR and agarose gel electrophoresis for NF-M (Fig. 6A). Co-IP NF-M PCR product was readily visible and enriched (lane 3) over pre-IP product (TIC, lane 2). Product was missing from IPs performed with purified anti-β-galactosidase monoclonal antibody (lane 4), confirming the specificity of the hnRNP K antibody. Xenopus EF-1α, which is not an hnRNP K target (lane 5), was also missing from the anti-hnRNP K co-IP, confirming the specificity of the interaction with NF-M mRNA. No peripherin RT-PCR product was detected in the hnRNP K-IP, even after two rounds of amplification with nested sets of primers(Fig. 6B; lanes 4 and 5). Thus,in Xenopus brain, NF-M mRNA associates endogenously with hnRNP K but peripherin mRNA does not.

Unilaterally injected hnRNP K MO tadpoles (stage 39/40) expressed NF-M mRNA in neuronal perikarya on both sides(Fig. 6C), indicating that the loss of NF-M protein expression was not due primarily to transcriptional effects. Although in some instances NF-M mRNA staining was less intense on the injected side, this was not always the case(Fig. 6D, seen with FISH). Because, in dissociated cell culture, NF-M expression increases when spinal neurites contact muscle (Undamatla and Szaro, 2001), this reduction was possibly due to the loss of such contacts. Peripherin mRNA was robustly and equivalently expressed on both sides of hnRNP K-knockdown animals(Fig. 6E), indicating that hnRNP K knockdown did not lead to generalized reductions in transcriptional activity.

Dissociated embryonic spinal cord-myotome cell cultures were used to determine whether the effects on axon outgrowth seen in intact embryos were cell-autonomous and to provide independent confirmation of the antibody staining in whole mount that axon outgrowth is compromised; in culture, it is readily seen under phase-contrast illumination without immunostaining. Suppression of hnRNP K expression was as effective in culture as in whole mount, persisting through 2 days of culture (corresponding to 3 days of development, not illustrated). Only a few cells descended from hnRNP K MO-injected blastomeres (identified by FDx) retained any residual,albeit barely detectable, hnRNP K 24 hours after plating: 3±2% (mean of four dishes) of FDx-labeled cells for hnRNP K MO-injected cultures when compared with 87±3% (mean of three dishes) of FDx-labeled cells for control MO-injected cultures (significant at P<0.01, t-test). Conversely, hnRNP K immunostaining in control MO cultures was comparable with uninjected cultures.

To assay whether hnRNP K knockdown compromised neuronal determination and survival rates in culture, two markers for neurons were used: immunostaining for neuronal β-tubulin and FISH for NF-M. Embryos were bilaterally injected to ensure that all cells were descended from injected blastomeres. As identified by neuronal β-tubulin immunostaining, the numbers of `neurons' per dish were (±s.e.m., four dishes at 24 hours): hnRNP K MO-injected, 245±17; control MO-injected,269±17; uninjected, 287±16. Comparable data were obtained when NF-M FISH was used to label neurons (four dishes): hnRNP KMO-bilaterally injected, 234±20; hnRNP K MO-unilaterally injected, 256±16; control MO-injected, 290±12; uninjected cultures, 261±13. In no case did counts differ significantly(P>0.1, one-way ANOVA).

Using NF-M FISH to identify neurons, we quantified neurite outgrowth (Fig. 7) under phase-contrast illumination (Fig. 7A,B). The percentages of `neurons' with one or more neurites were indistinguishable between uninjected and control MO-cultures(Fig. 7C). As might be expected, unilaterally injected hnRNP K MO cultures had approximately half as many neurite-bearing `neurons' as uninjected and control MO-cultures,whereas bilaterally injected hnRNP K MO cultures had virtually none(∼2%). These differences were statistically significant(P<0.01, one-way ANOVA), thus confirming that the effects on axon outgrowth are cell autonomous. In addition, as reported in chick spinal cord(Bennett and DiLullo, 1985),control cultures had the occasional cell expressing NF-M protein(Fig. 7D-E′) but lacking neurites, indicating that having a neurite is not a necessary precondition for NF-M protein expression. In addition, because cells lacked even early signs of neurite initiation, we concluded that inhibition of outgrowth began as early as axon initiation.

Because hnRNP K has been implicated in shuttling RNAs from the nucleus(Bomsztyk et al., 1997; Bomsztyk et al., 2004; Mikula et al., 2006), we used FISH to localize NF-M RNA intracellularly during hnRNP K knockdown. In whole mount, on the uninjected side of hnRNP K MO tadpoles, spinal cord neuronal nuclei were better defined by circumferential cytoplasmic FISH-staining than were those on the injected side(Fig. 8A,A′). For better resolution of this phenomenon, we performed NF-M double FISH-immunohistochemistry in culture. In control cultures, NF-M protein extended into neurites and the RNA surrounded the nucleus (labeled with DAPI),as is normal (Fig. 8B1-3). In unilaterally injected hnRNP K MO-cultures, such `normal' neurons could be seen adjacent to cells that were unstained for NF-M protein and had RNA-staining that appeared to overlap nuclei(Fig. 8C1,2). When counterstained with DAPI, this staining was clearly seen to overlap with nuclei (Fig. 8D1-4) in both unilaterally and bilaterally injected cultures. Moreover, in bilaterally injected cultures, NF-M protein was absent from virtually all (>98%) NF-M RNA positive cells, and these cells had no neurites.

To validate our conclusion that hnRNP K plays an important role in shuttling NF-M message effectively from the nucleus, we quantified NF-M and peripherin RNA from nuclear and cytoplasmic subcellular fractions of bilaterally injected hnRNP K MO- and control(stage 29/30) tadpoles. The cytoplasmic/nuclear ratio (determined asΔC_T from qRT-PCR) was significantly less for NF-MmRNA in hnRNP K MO embryos than for either RNA in the other groups(P<0.01, one-way ANOVA). Approximately 2.4% (∼2 ^ΔC_T) of the NF-M mRNA that was expressed in hnRNP K MO embryos was retained in the nucleus, when compared with∼0.03% for other groups, an ∼80-fold reduction in the efficiency of nucleocytoplasmic export.

To test whether NF-M RNA translation was directly affected, we performed polysomal profiling by sucrose gradient ultracentrifugation. In this assay, more actively translated mRNAs fractionate with polysomes, which lie to the left of the monosomal peak seen in A₂₆₀ plots(Fig. 9). Messages in monosomal or lighter fractions (to the right) are considered translationally silent. In control embryos, NF-M mRNA underwent a marked, leftward shift into polysomal fractions between stages 25 (early axonal outgrowth) and 29/30 (the peak of spinal axonal outgrowth), indicating increased translational efficiency. A comparable shift occurs during optic nerve regeneration(Ananthakrishnan et al., 2008),suggesting that this shift in embryos is associated with increased axonal outgrowth. At stage 25, peripherin mRNA was already present in polysomal fractions and exhibited a less dramatic shift at stage 29/30,consistent with its being expressed earlier than NF-M(Gervasi et al., 2000). In hnRNP K MO animals, the leftward shift was virtually eliminated for NF-M mRNA, leaving the mRNA in monosomal and lighter fractions. By stark contrast, the peripherin mRNA profile was unaffected, confirming that hnRNP K is not essential for its translation. Such a selective effect indicates that hnRNP K is essential for the translation of only a subset of mRNAs.

Earlier work in Xenopus showing that loss of NF-M attenuates axon elongation without completely abolishing it(Szaro et al., 1991; Lin and Szaro, 1995; Lin and Szaro, 1996; Walker et al., 2001), along with results from knockout mice (Elder et al., 1998), argue that more than a loss of NF-M protein must be at work to account for the loss of axons in hnRNP K knockdown animals. To gain some further insights into this issue, we thus looked at effects in `neurons'(as identified by peripherin and neuronal tubulin staining) on the three axonal cytoskeletal elements whose synthesis does not require hnRNP K. In bilaterally injected hnRNP K MO cultures, peripherin formed twisted filaments and aggregates, and neuronal β-tubulin formed densely packed,ring-like structures, ∼2-6 μm in diameter in perikarya(Fig. 10D,F). By contrast,their distributions appeared normal in control MO cultures(Fig. 10C,E). F-actin staining with fluorescent phalloidin was also abnormal. Normally in Xenopuscultures, even newly initiated neurites and filopodia stand out with fluorescent phalloidin-staining (Smith et al., 2006), but in bilaterally injected hnRNP KMO-cultures, neither filopodia nor neuritic processes were detectable in any cells (Fig. 10B). Instead,F-actin staining was generally circumscribed around the perikaryal periphery,usually with more on one side of the cell than the other, suggesting cells may be polarized despite their failure to initiate neurites. By contrast, neuritic and growth cone lamellipodia and filopodia in control cultures were well stained, with perikaryal hot-spots forming, which is normal(Fig. 10A). Thus, hnRNP K is likely to be important in the post-transcriptional regulation of a set of proteins needed for organizing the cytoarchitecture to form axons.

hnRNP K-knockdown by antisense MOs inhibited axon outgrowth, dramatically reducing otherwise robust nerves to the occasional wispy fiber in intact animals and virtually eliminating neuritic outgrowth altogether in culture. This deficiency arose not from inhibiting neuronal differentiation per se, as the expressions of several neuronal markers were unaffected. Instead, it most probably arose from the loss of specific key axonal cytoskeletal proteins whose mRNAs are influenced by hnRNP K, leading to the disrupted cytoarchitectures of essential axonal polymers. Comparing the differential effects on peripherin and NF-M further indicated at the molecular level that this RNA binding protein in developing Xenopusis needed for efficient nucleocytoplasmic shuttling and loading of selective target mRNAs onto polysomes for translation.

The ubiquitous and early expression of hnRNP K originally led us to anticipate that its knockdown would probably produce an early embryonic lethal. In retrospect, the survival of embryos through early stages of development may have been a fortuitous consequence of the persistence of maternal hnRNP K through gastrula stages, possibly allowing embryos to bypass earlier defects on such processes as cellular proliferation. hnRNP Kand its Drosophila homolog bancal, for example, have both been implicated in promoting cell proliferation in neuroblastoma cells and developing appendages (Charroux et al.,1999; Yano et al.,2005). Even so, we saw no evidence of any overall inhibition of cellular proliferation in Xenopus, even after hnRNP K expression was thoroughly suppressed: (1) tail formation, which requires cell proliferation,was overtly normal; (2) proliferating-cell-nuclear-antigen (PCNA) was expressed in hnRNP K MO-injected descendants (data not shown); and(3) neuronal cell counts (i.e. cells expressing neuronal β-tubulin and NF-M mRNA) in dissociated cell culture did not differ significantly between hnRNP K MO- and control cultures. Experiments in culture also ruled out any significant cell-autonomous effects on neuronal cell survival,although we can rule out neither secondary trophic effects in the intact animal owing to the loss of axons nor the possibility that hnRNP K plays additional roles at later times.

At the molecular level, hnRNP K knockdown significantly reduced the efficiency of NF-M mRNA export from the nucleus and inhibited its loading onto polysomes for translation, leading to suppression of NF-M protein expression. These data, along with its cellular localization, support a role for hnRNP K as a nucleocytoplasmic RNA shuttling protein in Xenopus,despite the apparent absence of the KNS nuclear localization signal of human hnRNP K (Siomi et al., 1993; Michael et al., 1997). Shuttling could nonetheless be mediated by alternative sequences in Xenopus hnRNP K. For example, it shares an ERK1/2 phosphorylation site with human hnRNP K, which has been implicated in cytosolic localization in HeLa cells (Habelhah et al.,2001). Drosophila bancal also exhibits shuttling behavior, despite lacking the KNS motif. Instead, it has an M9 motif, which is known to mediate shuttling of human hnRNPs A1 and A2/B1(Charroux et al., 1999).

Because hnRNP K has been implicated in multiple, sometimes seemingly contradictory facets of RNA regulation, drawing an explicit connection with any one aspect of mRNA regulation has historically proved difficult(Bomsztyk et al., 1997; Bomsztyk et al., 2004; Ostareck-Lederer et al.,1998). The contemporary view is that hnRNP K serves as the central element of a docking platform that accompanies RNAs from nucleus to cytoplasm,selectively recruiting various elements to the RNA to dictate its fate at any moment (Bomsztyk et al., 2004; Mikula et al., 2006). Our results support such an idea. This raises future questions about how hnRNP K function is regulated. In addition to changes in hnRNP K expression, its actions can be modulated by competition with other RNPs, such as hnRNP E1(Ostareck et al., 1997; Ostareck-Lederer and Ostareck,2004) and Hu proteins (Yano et al., 2005), both of which also target NF-M mRNA(Antic et al., 1999; Thyagarajan and Szaro, 2004). Such proteins could therefore participate jointly in NF-Mpost-transcriptional regulatory modules. In other instances, hnRNP K actions are regulated by phosphorylation, potentially linking its actions to cell signaling events (Ostareck-Lederer et al.,2002). Our results indicate that the developmental changes in hnRNP K-NF-M RNA interactions seen during postnatal cortical development(Thyagarajan and Szaro, 2008)may underlie regulated changes in NF-M translation.

The most surprising result was the loss of axon outgrowth. The near ubiquitous penetrance of this effect in both intact animals and culture makes it exceedingly unlikely that only a few neuronal subtypes were targeted. Equally unlikely is that axonal loss was caused solely by the loss of NF-M. Instead, the failure of neurites to be initiated, together with the disrupted cytoarchitectures of essential axonal polymers, argue that the axonal defects result from the failure of one or more proteins needed for organizing these polymers to be expressed. These findings raise the intriguing possibility that the translational control of multiple proteins involved in building the axon may be linked through shared elements such as hnRNP K. Our earlier findings that hnRNP K binding to NF-M RNA is developmentally regulated and that NF-M itself is under strong translational control during optic axon regeneration are consistent with hnRNP K being part of such a regulatory module rather than a constitutive facet of NF-M RNA metabolism(Thyagarajan and Szaro, 2004; Thyagarajan and Szaro, 2008; Ananthakrishnan et al., 2008). Future studies to identify the co-factors of hnRNP K, its additional targets during axonogenesis and its mode of regulation should provide the necessary information to distinguish between whether hnRNP K acts as part of a regulatory module influencing several targets via the same mechanism or as one shared element of multiple modules differentially affecting each target separately. Regardless of the outcome of such studies, the current work affirms the importance of post-transcriptional control of selective mRNAs for axon development and identifies hnRNP K as one of the crucial elements.

We thank Drs Amar Thyagarajan and Lakshminarayanan Ananthakrishnan for technical help with the immunoprecipitations and polysome profiling,respectively, and Dr John Schmidt and Kurt Gibbs for editorial comments. The work was supported by NSF IOS 643147 and NYSPHS (New York State Spinal Cord Injury Research Trust) C020940.