TLS facilitates transport of mRNA encoding an actin-stabilizing protein to dendritic spines

Dendritic spines are small protrusions on the surface of neuronal dendrites where synaptic sites are formed to receive presynaptic signals and transmit mainly excitatory signals. Morphological heterogeneity of spines may determine their functional diversity (Harris, 1999). Recent publications have shown that spine development directly correlates with synaptogenesis, such that an initial synaptic contact between an axon and dendrites leads to the formation of synapses and the assembly of a postsynaptic complex composed of the proteins PSD-95, Homer, shank and cortactin (Hering and Sheng, 2001). The cytoskeletal signaling postsynaptic complex is involved in changes of spine shape; for example, the spine changes from a filopodia-like structure to a mature mushroom-shaped structure.

The morphological plasticity of dendritic spines is driven by actin filaments concentrated in the spine heads (Matus, 2000). Spines show rapid movements and continuously undergo changes in their shape that are influenced by synaptic activity (Matus, 2005). Actin-depolymerizing reagents have been reported to block spine motility and to interfere with long-term potentiation (LTP) (Kim and Lisman, 1999; Krucker et al., 2000), which is accompanied by an increase in the F-actin content within the dendritic spines (Fukazawa et al., 2003). Therefore, actin filaments have been thought to be a link between activity-induced modulation of synaptic transmission and long-term changes in synaptic morphology associated with memory consolidation. Thus, actin reorganization must be a critical event when spine architecture changes (Okamoto et al., 2004; Star et al., 2002). In recent years, considerable progress has been made in identifying the molecules that control spine growth and maturation (Hering and Sheng, 2001). The cytoskeleton of spines is crucial for their development and stability. An expanding set of actin-binding and/or actin-regulating molecules has been detected in dendritic spines, for example, drebrin, SPAR, epsin, profilin and cortactin (Ackermann and Matus, 2003; Hayashi and Shirao, 1999; Hering and Sheng, 2003; Pak et al., 2001; Sekerkova et al., 2003). Despite the known importance of actin dynamics in synaptic plasticity, how the spines acquire the ability to continuously change their shape in response to synaptic activation has not been fully elucidated.

We recently found that TLS (translocation in liposarcoma), an RNA-binding protein (Crozat et al., 1993) also called FUS (see Mouse Genome Informatics), is translocated to the neuronal dendrites as an RNA-protein complex in response to metabotropic glutamate receptor 5 (mGluR5) activation (Fujii et al., 2005). TLS-null hippocampal neurons displayed abnormal spine morphology as evidenced by their long thin filopodia-like phenotype and a reduced number of spines (Fujii et al., 2005), suggesting that TLS may be involved in the transport of mRNAs to maintain the spine shape when spine remodeling occurs in response to synaptic signals. To identify the RNA targets for TLS and explore the role of TLS in spines, we screened a mouse brain RNA pool for TLS-associated RNAs. Among the candidates isolated, we found mRNAs encoding the actin-stabilizing protein Nd1-L (listed as Ivns1abp, influenza virus NS1a in MGI) (Sasagawa et al., 2002). Its transcript was increased in dendrites upon glutamatergic activation. Here we show that TLS is a component of RNA-transporting complexes in dendrites and associates with mRNA encoding an actin-stabilizer protein. Thereby TLS may regulate the actin cytoskeleton by supplying Nd1-L mRNA to the local translational machinery in dendrites.

Mouse hippocampal neurons were prepared from E16.5 ICR embryonic mice as previously described (Okabe et al., 1999). The neurons were plated on either poly-L-lysine-coated glass-bottom dishes (35 mm; Matsunami, Osaka) at 2-5×10⁵ cells/dish or on 90-mm poly-L-lysine-coated dishes (Sumitomo Bakelite, Tokyo) at 2×10⁶ cells/dish, and on the third day in culture they were treated with 2 μM Ara C (Sigma) to eliminate non-neuronal cells. All animal experiments in this study were regulated by the Animal Research Committee of Osaka Bioscience Institute.

Mice heterozygous for Tls/Fus deficiency were generated as previously reported (Hicks et al., 2000). Mice homozygous for the Tls mutation fail to suckle, and die within 16 hours after birth. Therefore, Tls heterozygous littermates were crossbred to obtain homozygous embryos for the preparation of primary cultures of hippocampal neurons.

The human TLS cDNA clone containing the full coding region was a generous gift from D. Ron (NYU, Skirball Institute, New York, USA). The coding region was amplified by PCR and subcloned into KpnI/XhoI sites of the pEGFP-N2 vector (Clontech). The N terminus (amino acids 1-269) and C terminus (amino acids 270-526) of hTLS were also prepared by PCR and subcloned into pEGFP-N2. The coding region of Nd1-L was amplified by PCR from an adult mouse brain cDNA and fused to monomeric red fluorescent protein (RFP) (Campbell et al., 2002). To express RFP or GFP constructs, we injected plasmids [10 ng/μl in TE buffer (10 mM Tris-HCl pH 8.0, 1 mM EDTA)] into the nucleus of cultured neurons under an inverted epifluorescence microscope (Nikon TE300) using glass pipettes connected to a micromanipulator (MMO-202ND; Narishige, Tokyo).

The following antibodies were used: anti-TLS monoclonal mouse antibody (N terminus 1-117 antigen, BD Transduction Labs) for immunoprecipitation, and rabbit TLS polyclonal antibody (TLS-C) was raised against glutathione S-transferase (GST) fused to the C terminus of TLS (amino acids 172-526) (Trans Genic Inc., Kumamoto, Japan).

Primary cultures of hippocampal neurons were fixed in 2.5% PFA in PBS for 20 minutes at room temperature, permeabilized in 0.2% Triton X-100 for 3 minutes at room temperature, and blocked in 5% normal goat serum. Primary antibodies were visualized with Alexa Fluor 350-conjugated goat anti-rabbit IgG. For visualization of F-actin, fixed and permeabilized cells were incubated with Oregon Green-phalloidin (Molecular Probes, 1:20 in 5% normal goat serum) for 1 hour at room temperature.

TLS-GFP-expressing neurons were labeled with 0.5 μM ethidium bromide (EtBr), as previously described, to detect total RNA content (Tang et al., 2001). SYTO14 (Molecular Probes) labeling of neurons was performed as previously described (Kohrmann et al., 1999).

Primary cortical neurons were prepared from either wild-type (WT) or TLS-null (Null) mice. After 12 day in culture (12 DIV: days in vitro), the neurons were incubated in medium containing actinomycin D (5 μg/ml). Total RNA was harvested at four time points (0, 1, 2, and 3 hours) after actinomycin D addition followed by a reverse transcription using random hexamer as described above. RNA was quantitated by real-time PCR using SYBR^®Green PCR Master mix (Applied Biosystems) as described previously (Akashi and Takumi, 2005). Amplification and data collection were performed using the PRISM^® 7900 HT sequence detection system (Applied Biosystems). The obtained cDNA quantities were normalized with respect to the amounts of 18S rRNA.

The subcellular fraction containing polyribosomes was prepared as previously described (Ohashi et al., 2002). Briefly, mouse brains were homogenized in TKM buffer (20 mM Tris-HCl, pH 7.5, 50 mM KCl, 2 mM MgCl₂, 250 mM sucrose, 1 mM phenylmethylsulfonyl fluoride and 100 μg/ml cyclohexamide), and centrifuged at 1,000 g for 10 minutes at 4°C to remove nuclei and unbroken tissues. The supernatant was then centrifuged at 10,000 g for 15 minutes at 4°C to yield the post-mitochondrial supernatant (PMS). The PMS was centrifuged at 130,000 g for 1 hour at 4°C to separate the high-speed polyribosomal pellet (P3) including light membranes and polysomes. P3 was resusupended in 150 mM KCl and incubated on ice for 15 minutes, followed by centrifugation at 10,000 g for 5 minutes to remove insoluble materials. P3 was immunoprecipitated with anti-TLS monoclonal mouse antibody and protein G-Sepharose (Amersham Biosciences). A quarter of the immunoprecipitate was eluted by boiling for 5 minutes in SDS-sample buffer and analyzed by SDS-PAGE. The rest was subjected to RT-PCR as described previously (Ben Fredj et al., 2004). Briefly, the rest was incubated in RNP buffer (50 mM Tris-HCl, pH 7.5, 25 mM KCl, 5 mM MgCl₂, 5% glycerol, 0.5 mM dithiothreitol, and 0.2% SDS) at 65°C for 15 minutes to elute RNA. All buffers used for polysomal fractionation and IP were supplemented with RNasin (100 U/ml). The RNA was extracted with TRIzol reagent (Invitrogen), and ethanol-precipitated using Etachine-mate (Wako Nippon Gene, Tokyo). Precipitated RNA was reverse-transcribed by using Superscript II (Invitrogen) according to the manufacturer's instructions, and the first-strand cDNAs obtained were subjected to RT-PCR using random hexamer primers or oligo(dT)₁₇ primers. The resulting cDNAs were amplified by PCR using specific primers: for Nd1-L, 5′-GAT GAA GGT CTG TTT CTG TA-3′ (forward) and 5′-GCC TCT TAA GAC AGA ATT GA-3′ (reverse); and for β-actin, 5′-GAC CTC TAT GCC AAC ACA-3′ (forward) and 5′-TCC ACA TCT GCT GGA AGG T-3′ (reverse).

Polyadenylated RNAs from adult mouse brain were prepared using a mRNA purification kit (Ambion) and reverse transcribed using oligo(dT)-NotI primer from a cDNA synthesis kit (Stratagene). Double-stranded cDNA with a SalI linker at its 5′ end and NotI site at its 3′ end were cloned into pCS2+SN vector [a gift from M. Hibi, RIKEN CDB (Rupp et al., 1994)] at specific SalI-NotI sites. Bacteria were transformed with pCS2+SN plasmid clones and plated to obtain single colonies with single plasmid clones. After restriction mapping and sequencing of over 500 plasmid clones, 103 clones were found to be homologous to known EST sequences. [α-³²P]UTP-labeled RNA probes (ranging from 500-1000 nucleotides) were prepared by using an in vitro transcription system (Promega) with pCS2+SN plasmid clones as a template and examined by conducting the RNA mobility shift assay described below. The binding ability of these RNA was also confirmed by conducting an in vitro binding assay with ³⁵S-labeled TLS. ³⁵S-labeled TLS and non-labeled RNA probes were incubated in binding buffer [20 mM Hepes, pH 7.6, 100 mM KCl, 5% glycerol, 3 mM MgCl₂, 2 mM DTT, RNasin (100U/ml) and 5 mg/ml heparin] for 20 minutes at room temperature, and then incubation with oligo(dT) cellulose. Eluted materials were subjected to SDS-PAGE and autoradiography.

RNA probes were transcribed and labeled with [α-³²P]UTP using an in vitro transcription system (Promega). TLS was prepared by subcloning TLS cDNA to pcDNA3.1/Myc-His vector (Invitrogen) using TNT^® Quick Coupled Reticulocyte Lysate Systems (Promega). RNAs were first incubated in binding buffer [20 mM Hepes, pH 7.6, 100 mM KCl, 5% glycerol, 3 mM MgCl₂, 2 mM DTT, RNasin (2 U/μl), 5 mg/ml heparin, and 0.2 mg/ml yeast tRNA] and then incubated for 20 minutes at room temperature with rabbit reticulocyte lysate expressing TLS. RNA-protein complexes were resolved in non-denaturing 4% polyacrylamide gel in 1× TBE buffer (45 mM Tris-borate pH 8.0, 1 mM EDTA) and visualized by autoradiography. Competition assays were performed by incubating lysates with a 5-, 50- or 500-fold excess of non-radioactive RNA for 10 minutes at room temperature before the radioactive probes were added. Non-labeled transcripts for competition assay were derived from the polylinker region of pcDNA3.1/Myc-His (206 nucleotides) and mouse β-actin 3′-untranslated region (3′-UTR) containing a zip code sequence (780 nucleotides).

DIG-labeled cRNA probes specific for β-actin and Nd1-L were synthesized by means of a DIG RNA labeling kit (Roche). Signals were detected using AP-anti-DIG antibody and HNPP (2-hydroxy-3-naphtoic acid-2′-phenylanilide phosphate) fluorescent detection kit (Roche). To detect TLS-containing RNA particles, TLS-GFP expressing hippoccampal neurons (21 DIV) were fixed and permeabilized in 70% ethanol prior to the hybridization. TLS-GFP was indirectly visualized using anti-GFP antibody after cell in situ hybridization.

Immunofluorescent images were obtained using a Fluoview 300 confocal laser-scanning microscope (Olympus). To analyze the RNA content of individual dendrites, we divided the dendritic processes of interest into compartments 20 μm in length and determined the total pixel intensity. Multiple optical sections (12-15 sections and z-spacing of 0.4-0.6 μm) were collected, and recombined using the maximum brightness operation by Metamorph (Roper Scientific).

Maximum-intensity projection images were prepared for each image stack, and these projection images were used for the quantitative analysis. For given optical slices, all projection images were processed identically for fluorescent clusters and for quantitative analysis by fluorescence measurement. Automatic processing of fluorescent images using the Metamorph software enabled us to determine the position and intensity of fluorescent clusters. To quantify the fluorescence intensity of clusters, the binary images were used to specify the cluster domains. The total fluorescence intensity and the average fluorescence intensity in each domain were measured.

To examine whether TLS is colocalized with RNA in dendrites, we labeled mouse hippocampal neurons with the RNA-specific dye SYTO14 (Knowles et al., 1996) and then immunostained them with anti-TLS antibody. The total amount of endogenous TLS in the dendrite was not large, as reported previously (Fujii et al., 2005). Nonetheless, a certain population of endogenous TLS-containing particles was colocalized with SYTO14-labeled RNA particles, suggesting that TLS is a component of RNA-containing large particles in dendrites (Fig. 1A). It has been reported that the C terminus of TLS contains an intrinsic RNA-binding domain (Iko et al., 2004; Zinszner et al., 1997). When GFP was fused to a C-terminal fragment of TLS (C-TLS) or to the full-length TLS (FL) and the fusion protein was overexpressed in neurons, GFP signals could be observed in a distal portion of the dendrite, where they were colocalized with RNA-containing particles that stained with EtBr (Fig. 1B, left and right, and C). C-TLS seemed to form more RNA-TLS clusters in the dendrites than did FL. By contrast, GFP fused to an N-terminal fragment of TLS lacking the RNA-binding domain (N-TLS) was not detected in dendrites and was predominantly localized in the nucleus (Fig. 1B, middle, and C). These results indicate that the RNA-binding domain is necessary for cytoplasmic and/or dendritic localization of TLS. When neurons derived from TLS-null mice were used for examination of RNA content in dendrites, the overall dendritic RNA content was not statistically different between wild-type (WT, +/+) and TLS-null (–/–) neurons (Fig. 2A left and B; control, n=48). However, when the neurons were treated for 60 minutes with DHPG (100 μM), a group-1 mGluR agonist, to stimulate postsynaptic mGluRs, the RNA content was significantly increased in the wild-type neurons; whereas the RNA signal intensity in the TLS-null dendrites remained unchanged (Fig. 2A left and 2B; P<0.001, DHPG, n=54). Exogenously introduced TLS-GFP by itself did not drastically increase the RNA content in either wild-type (Fig. 2A right and C; n=50, 130% of non-transfected wild-type neurons) or TLS-null dendrites (n=60, 125% of non-transfected TLS-null neurons). However, DHPG-treatment significantly increased the dendritic RNA content in both wild-type and the TLS-null neuronal dendrites expressing TLS-GFP (to approx. 160% and 150%, respectively; Fig. 2A right and C, P<0.05, n=50). These results indicate that TLS facilitates RNA translocation upon mGluR activation. Taken together, our data show that an association with RNA is essential for TLS to translocate to cytoplasm and subsequently move to dendrites as an RNA-protein complex in an mGluR-dependent manner.

TLS was reported to bind GGUG-containing RNAs in an in vitro study and to be implicated in selective recognition of GGUG splice sites of pre-mRNA (Iko et al., 2004; Lerga et al., 2001). To explore additional functions of TLS in neuronal dendrites, we sought by screening of a mouse brain cDNA pool to identify RNA that could bind to TLS. We prepared radioactive-labeled RNA probes using the cDNA as template and performed an electrophoretic mobility shift assay (EMSA) for RNA. Among the candidate targets (Table 1), we focused on mRNAs encoding β-actin or the actin-stabilizing protein Nd1-L (Sasagawa et al., 2002) as we had observed (Fujii et al., 2005) that TLS-null neurons exhibit abnormal spine morphology such as low spine density or abnormal branching of dendrites that could be due to impaired actin reorganization. Since spine shape can be regulated by actin dynamics, we investigated how mRNAs encoding actin-related proteins could be related to TLS-null phenotypes. We first examined if the contents of these mRNAs were actually decreased in the dendrites of TLS-null neurons. Cell in situ hybridization revealed that both Nd1-L and β-actin mRNAs were increased in dendrites in DHPG-treated neurons by about 4.1-fold and 7.8-fold, respectively compared with those in non-treated WT neurons (Fig. 3A, left, and B). By contrast, the content of Nd1-L mRNAs was decreased in TLS-null neurons, and DHPG treatment did not increase the mRNA content; whereas the content of β-actin mRNA was not significantly decreased, and DHPG-dependent accumulation was still observed (Fig. 3A, right, and B). These data show that dendritic translocation of Nd1-L transcripts depends on TLS expression whereas that of β-actin mRNA does not. This observation is also in good agreement with previous reports that β-actin mRNA directly binds to and is transported by ZBP1, which specifically binds to a short segment of the 3′-untranslated region (3′-UTR) of β-actin mRNA (Gu et al., 2002). The distribution of Nd1-L transcripts was also different from that of β-actin mRNA (Fig. 3C). The latter was evenly distributed in dendrites, whereas the former showed an uneven dendritic distribution (Fig. 3C, arrowheads) and also dense clusters at dendritic branching points (Fig. 3C, arrows).

To exclude the possibility that TLS may affect the stability of the target RNA, we measured the amount of Nd1-L mRNA remaining in neurons by means of quantitative RT-PCR after actinomycin D treatment. The Nd1-L mRNA in TLS-null neurons was observed to be as stable as that seen in wild-type neurons (see Fig. S1 in supplementary material). Taken together, our data suggest that Nd1-L mRNA more selectively associates with TLS than β-actin mRNA does and that TLS does not affect the stability of the mRNAs.

To validate that TLS formed a complex with a specific target RNA, we used anti-TLS antibody to immunoprecipitate a TLS complex from a mouse cortex polysome fraction. Immunoprecipitated RNAs were reverse-transcribed with oligo(dT) used as the primer. The resulting cDNAs were then subjected to PCR using specific primers for β-actin and Nd1-L mRNAs (Fig. 4A). The anti-TLS antibody used in this study was specific (Fig. 4B). Both the amplified PCR products for β-actin and Nd1-L mRNAs (Fig. 4C, anti-TLS) migrated to the same position as the control cDNA products obtained from mouse cortex total RNA (Fig. 4C, cDNA). Co-immunoprecipitation of RNA with TLS from the cortical polysome fraction was expected to be specific for anti-TLS antibody, because no PCR products were detected in immunoprecipitates prepared with control mouse IgG (Fig. 4C, IgG). Association of β-actin mRNA with TLS seemed to be less specific than that of Nd1-L, which is consistent with data from the cell in situ hybridization presented above.

Next, we tested whether or not TLS directly bound Nd1-L mRNA. We identified 1300-nucleotide-long transcripts (2010-3325) corresponding to the 3′-UTR of Nd1-L mRNA associated with TLS in vitro (see Fig. S2 in supplementary material). To further specify the sequence responsible for TLS binding, we prepared various nucleotide (nt) fragments, each approximately 200 nt in length and used them as EMSA probes. A prominent mobility shift was observed with fragments covering a relatively long region spanning 729 nt (2597-3325) of the 3′-UTR (see Fig. S2 in supplementary material). This result suggests that TLS does not appear to recognize a specific short sequence, such as a zip code sequence for β-actin 3′-UTR (Gu et al., 2002), within the RNA. Rather, TLS seems to interact with RNA at multiple RNA-binding motifs or recognize a partial conformation of RNA. Therefore, we used the 493-nt region spanning 2448-2940 in the 3′-UTR as a Nd1-L RNA probe, and conducted EMSA (Fig. 5). The results of a competition assay for TLS binding using non-radioactive transcripts showed that the presence of a 50-fold or greater molar excess of cold probe completely prevented the binding of TLS to radioactive-labeled Nd1-L RNA probe (Fig. 5A), whereas non-specific RNA competitors, pcDNA3.1/Myc-His polylinker (205 nt) and mouse β-actin 3′-UTR containing a zip code sequence (780 nt) did not inhibit the binding (Fig. 5B). This observation suggests that TLS may selectively associate with Nd1-L mRNA to form a protein-RNA complex although the association does not seem to be sequence specific.

Next, we tested whether the transport of Nd1-L mRNA depended on TLS. To exogenously express TLS, we injected TLS-null neurons with TLS-GFP-bearing adenovirus and then examined the localization of Nd1-L mRNA by in situ hybridization (Fig. 6). Whereas TLS-null neurons hardly expressed Nd1-L transcripts (Fig. 6A), TLS-expressing neurons showed a restored Nd1-L mRNA localization in their dendrites (4.76±0.67-fold; Fig. 6B) although morphological abnormalities were not rescued. TLS-GFP overexpressed in mouse hippocampal neurons was present in a punctate distribution in their dendrites, and was also colocalized with Nd1-L mRNA in the necks and heads of spines (Fig. 6C, top). When TLS-GFP was expressed in the TLS-null neurons, TLS-GFP in spine-like structures was also colocalized with Nd1-L mRNA (Fig. 6C, arrows, bottom), indicating that TLS associates with the mRNA and that transport of Nd1-L mRNA to the dendritic spines might be facilitated by TLS.

When actin filaments in spines were visualized by Oregon Green-conjugated phalloidin staining of F-actin, the mature neuronal dendrites showed typical mushroom-shaped spines (Fig. 7A1-2, arrowheads). However, dendritic spines of TLS-null neurons displayed long, thin filopodia-like protrusions or abnormal branching (Fig. 7B1-3, arrows). Since the actin organization of the spines was impaired in TLS-null neurons, we tested to see if overexpression of Nd1-L would affect the spine morphology. When Nd1-L fused to RFP (Nd1-L-RFP) was exogenously expressed in WT neurons, the protein appeared in dendrites and was also localized in spines (Fig. 8). These transfected cells were then exposed to an actin-destabilizing agent, cytochalasin D (2 μM, for 48 hours). Neurons exposed to cytochalasin D were damaged and the spines and actin filaments in dendritic shafts were destroyed (Fig. 8, vector/CytD). However, neurons overexpressing Nd1-L-RFP had abundant F-actin-rich spines, were resistant to cytochalasin D and exhibited mature mushroom-shaped spines (Fig. 8, Nd1-L/CytD). We observed a similar effect of Nd1-L on mouse hippocampal neurons treated with cytochalasin B (1 μM, for 48 hours; data not shown). Our data suggest that Nd1-L functions in actin stabilization (`less motile but active') and that without Nd1-L, spine maturation may be disturbed and the spines may remain as filopodia-like processes (`motile but inactive and immature'), as observed in TLS-null neurons.

TLS has been characterized as an RNA-binding protein predominantly localized in the nucleus. We recently demonstrated that TLS is specifically localized in the neuronal dendrites of mature neurons and recruited into the spines upon mGluR5 activation (Fujii et al., 2005). Several lines of evidence have revealed that spine shape is mainly regulated by actin filaments (Matus, 2005; Okamoto et al., 2004; Star et al., 2002). TLS-null neurons display abnormal spine morphology, implying a link between TLS and actin remodeling upon synaptic activation. In this study, we identified several mRNAs in TLS-associated complexes in neuronal dendrites, including those encoding molecules that could be responsible for the abnormal spine phenotypes of TLS-null neurons. Among them, we focused on an actin-stabilizing protein, Nd1-L, as a candidate that is involved in dendritic spine morphogenesis and remodeling.

TLS was earlier reported to specifically bind to a GGUG-type exon-intron junction and to form splicing ribonuleocomplexes in the nucleus of fibroblasts (Yang et al., 1998). However, our EMSA revealed that TLS preferentially associated with a 493-nucleotide-long sequence of the 3′-UTR of Nd1-L mRNA in which a GGUG-type motif was absent (Fig. 5). It was previously reported that TLS forms an RNP complex with a microtubule-dependent kinesin motor protein, KIF5, specifically with KIF5B (Kanai et al., 2004) and is transported in dendrites. Recently we demonstrated that TLS in neurons is translocated to the distal part of dendrites and recruited to the spines upon mGluR5 activation (Fujii et al., 2005). This dendritic distribution of TLS in neurons has also been reported by others (Belly et al., 2005). Thus, TLS in neurons cannot be just a nuclear-cytoplasmic shuttling protein. TLS was colocalized with RNA-containing particles that labeled with SYTO14 (Fig. 1). In addition, a specific anti-TLS monoclonal antibody co-immunoprecipitated Nd1-L mRNA (Fig. 4). However, Nd1-L mRNA does not have the GGUG exon-intron junctions in its prespliced RNA, suggesting that TLS in neuronal dendrites associates with certain types of mRNA other than by a specific sequence recognition. Iko et al. reported that the GGUG-containing RNAs (oligoribonucleotides) have higher affinity than other RNAs for the zinc finger motif of TLS and that the RNA recognition motif (RRM) or the RGG-rich domains in the C terminus of TLS are not necessary for binding to the GGUG junction (Iko et al., 2004). Moreover, the C-terminal fragment of TLS fused to GFP was still able to translocate to a distal part of the dendrite. We observed that neurons transfected with cDNA encoding the C-terminal fragment of TLS formed large clusters of RNA-containing particles rather than inhibited RNA transport. Binding of mRNAs to TLS may be mediated by multiple RNA-binding domains to recognize a conformational structure of a certain region within mRNA, or by interaction with other proteins including KIF5. TLS contains RanBP2-type ZnF (Zinc finger motif), which was identified as a KIF5B/C-interacting domain in the presence of RanGDP (Cai et al., 2001). It would be interesting to further examine if any mutation or deletion within the RNA-binding domain or ZnF of TLS actually inhibits the association with the target mRNAs or KIF5. A genetic approach, such as deletion of one of these RRMs, should provide more direct answers.

In this study, we demonstrated that Nd1-L mRNA was localized in dendrites. Localization of the Nd1-L transcript was similar to the distribution of TLS in the dendrite (Figs 3 and 6; e.g. punctate distribution along dendrites and occasional localization in spines). The expression pattern of both Nd1-L and TLS during mouse development was also very similar from neonatal day through to the adult (our unpublished data). We demonstrated that TLS does not move in the dendritic shaft of cytochalasin-treated neurons (Fujii et al., 2005). The evidence led us to the idea that TLS can `sense' actin cytoskeletal changes in spines, in which TLS may interact with actin filaments via an actin based-motor protein such as myosin V. Subsequently, myosin V was found in the TLS-containing protein complex (our unpublished data). In our current model, when actin cytoskeletal organization collapses, TLS may be released from actin filaments and become free to associate with mRNA required for actin reorganization, which is then transported to the local translational machinery in the spines. Our present data suggest that there must be a defined combination of multiple proteins assembled into the RNA-transporting complex in dendrites including other transporters (e.g. KIF family, Pur-α, staufen, FMRP, etc.) (Antar and Bassell, 2003; Kiebler et al., 1999; Miyashiro et al., 2003; Ohashi et al., 2002; Tang et al., 2001). When and where (e.g. signal-dependency, in nucleus/cytoplasm/dendrite) such RNA transporting assembly is completed are interesting issues to address.

Among at least actin-binding and/or actin-regulatory proteins, there might be multiple regulatory proteins participating in activity-dependent spine remodeling. For example, cortactin, a well-described F-actin-binding protein at postsynaptic sites, has been reported to regulate spine morphogenesis. Cortactin-deficiency causes diminished spine density, whereas overexpression of cortactin induces longitudinal spine growth (Hering and Sheng, 2003). SPAR (spine-associated RapGAP) reorganizes the actin cytoskeleton and recruits PSD-95 to F-actin. Dominant-negative SPAR causes narrowing and elongation of spines (Pak et al., 2001). Similarly, the spines are significantly longer in drebrin (another actin-binding protein)-expressing neurons than in controls (Hayashi and Shirao, 1999). Moreover, NMDA receptor activation initiates changes in the actin cytoskeleton of dendritic spines that stabilize synaptic structure. These changes are mediated by profilin, a regulator of actin polymerization (Ackermann and Matus, 2003). In this study, we have shown a role for Nd1-L in actin stabilization, as overexpression of Nd1-L increased the number of normal spines (Fig. 8) whereas TLS-null neurons, which are unable to transport Nd1-L mRNA to their spines, displayed low spine density (Fig. 7). Combining these data with those reported previously (Fujii et al., 2005), we propose that mGluR5 activation induces translocation of TLS into spines and regulates spine morphology to stabilize the synaptic structure, probably through an actin network containing protein products of an RNA cargos delivered by TLS. Our investigation indicates, for the first time, that an RNA-binding protein facilitates translocation of mRNA encoding actin-regulatory protein and provides new insights into the signaling pathway for spine remodeling through mGluR.

A recent publication reported that a novel intracellular structure termed the “cell spreading center” is composed of a ribonucleoprotein core containing rRNA and TLS, which is associated with actin filaments near the leading edge of cells (de Hoog et al., 2004). Similarly, in the polysomal fraction of mouse brain, we found TLS to be associated with 28S rRNA (see Fig. S3 in supplementary material). This association could be mediated by the Pur-α/KIF5B complex (Kanai et al., 2004; Ohashi et al., 2002). Also, in mature hippocampal neurons, TLS was mainly localized in dendritic spines as a relatively large cluster, suggesting that TLS might associate with rRNA. These findings argue that TLS might be involved in translation. However, Nd1-L contains the BACK (BTB and C-terminal Kelch) domain (Stogios and Prive, 2004). Many kelch-repeat proteins are involved in organization of the cytoskeleton via interaction with actin; whereas BTB domains have multiple cellular roles, including recruitment to E3 ubiquitin ligase complexes. Additional information on both local protein synthesis and protein degradation of TLS and its cargo Nd1-L in neurons awaits further biochemical and cell biological studies.

We thank D. Ron for providing the human TLS cDNA clones, G. Hicks for heterozygous TLS-deficient mice, M. Hatano for Nd1-L antibody, S. Okabe for valuable discussions, P. Delgado for comments on writing the manuscript, and M. Hibi and T. Yoshioka for their generous support. The excellent technical assistance of Y. Shima, M. Tanaka, A. Yamamoto and S. Yoshiba is also greatly acknowledged. This work was supported in part by grants from MEXT, Mitsubishi Pharma Research Foundation and Sony Corporation.