Quantitative Proteomics Reveals Dynamic Interaction of c-Jun N-terminal Kinase (JNK) with RNA Transport Granule Proteins Splicing Factor Proline- and Glutamine-rich (Sfpq) and Non-POU Domain-containing Octamer-binding Protein (Nono) during Neuronal Differentiation*

The c-Jun N-terminal kinase (JNK) is an important mediator of physiological and pathophysiological processes in the central nervous system. Importantly, JNK not only is involved in neuronal cell death, but also plays a significant role in neuronal differentiation and regeneration. For example, nerve growth factor induces JNK-dependent neuronal differentiation in several model systems. The mechanism by which JNK mediates neuronal differentiation is not well understood. Here, we employed a proteomic strategy to better characterize the function of JNK during neuronal differentiation. We used SILAC-based quantitative proteomics to identify proteins that interact with JNK in PC12 cells in a nerve growth factor–dependent manner. Intriguingly, we found that JNK interacted with neuronal transport granule proteins such as Sfpq and Nono upon NGF treatment. We validated the specificity of these interactions by showing that they were disrupted by a specific peptide inhibitor that blocks the interaction of JNK with its substrates. Immunoprecipitation and Western blotting experiments confirmed the interaction of JNK1 with Sfpq/Nono and demonstrated that it was RNA dependent. Confocal microscopy indicated that JNK1 associated with neuronal granule proteins in the cytosol of PC12 cells, primary cortical neurons, and P19 neuronal cells. Finally, siRNA experiments confirmed that Sfpq was necessary for neurite outgrowth in PC12 cells and that it most likely acted in the same pathway as JNK. In summary, our data indicate that the interaction of JNK1 with transport granule proteins in the cytosol of differentiating neurons plays an important role during neuronal development.

entiation and regeneration. Although stress-induced pro-apoptotic JNK signaling is well characterized, the mechanisms involved in JNK-dependent neuronal differentiation remain enigmatic.
Quantitative mass-spectrometry-based proteomics has emerged as a powerful technology for investigating mammalian signaling pathways (17,18). Specifically, stable isotope labeling by amino acids in cell culture (SILAC) has been shown to be a powerful approach for studying dynamic changes in protein-protein interactions during cell signaling. We reasoned that an unbiased proteomic analysis of JNK interaction partners during neuronal differentiation could provide novel insights into the mechanisms involved. For that purpose, we used PC12 cells as a well-characterized classical model for neuronal differentiation (19). Stimulation of PC12 cells with nerve growth factor (NGF) induces a major shift in phenotype, from proliferating tumor cells to non-dividing neurons showing characteristics of sympathetic neurons such as the growth of long neurites and electrical excitability (19,20). In PC12 cells, NGF binds to neurotrophic tyrosine kinase receptor 1, low-affinity neurotrophin receptor, and fibroblast growth factor receptors (21,22). This leads to a coordinated activation of downstream signaling cascades such as the Ras/Raf/Erk1-2, PI3K, and JNK pathways, resulting in increased expression of genes that are involved in neuronal differentiation (23,24). Furthermore, G-proteins including Ras, Rap, and Cdc42 have been shown to link receptor activity to downstream kinase activation (25)(26)(27).
Here, we used quantitative interaction proteomics to analyze dynamic changes in JNK interaction partners during NGF-induced differentiation of PC12 cells. Our results show that JNK dynamically interacts with G-proteins, cytoskeletal proteins, and RNA binding proteins with distinct kinetic patterns. Intriguingly, several of the identified RNA binding proteins are known components of transport granules involved in mRNA localization and localized translation in neurons. Western blotting and co-localization experiments using confocal microscopy and proximity ligation assays validated the NGFinduced association of JNK1 with two RNA binding proteins, Sfpq and Nono. Furthermore, Sfpq knockdown decreased NGF-induced neurite outgrowth in PC12 cells, supporting the hypothesis that the interaction of JNK and Sfpq may contribute to neuronal differentiation.
Primary cortical neurons were prepared from E18.5 mouse embryos as described elsewhere (28). Briefly, cortices were dissected in Hanks solution under a stereomicroscope and dissociated enzymatically in neurobasal medium (Invitrogen) containing 0.1% trypsin (Sigma-Aldrich) and 0.001% DNaseI (Roche) at 37°C for 15 min. Trypsin was inactivated by 10% FBS in neurobasal media at 37°C for 5 min, and cells were dissociated by trituration with a sterile pipette tip. The resulting suspension was centrifuged at 1000 ϫ g for 3 min, and the supernatant was discarded. Next, neurons were resuspended and plated in neurobasal media containing B27 supplement (Invitrogen) and 100 g/ml penicillin/streptomycin. Neurons were cultured for 4, 6, 8, 12, 24, 36, and 48 h. Cell suspensions of the murine teratocarcinoma cell line P19 (ATCC) were cultured for 4 days in DMEM containing 5% FBS, 1 mM retinoic acid (Sigma), and 100 g/ml penicillin/streptomycin. Subsequently, aggregates of P19 cells were dissociated enzymatically in DMEM (Invitrogen) containing 0.25% trypsin and 0.001% DNaseI at 37°C for 10 min. Trypsin was inactivated by 10% FBS in DMEM at 37°C for 5 min, and cells were dissociated by trituration with a sterile pipette tip. Afterward, cells were resuspended and plated in DMEM containing 5% FBS and 100 g/ml penicillin/streptomycin for 2 days. 48 h after plating, cells were treated for an additional 24 h with 10 M ARAC (Sigma) to eliminate proliferating glial-like cells. Last, P19differentiated neurons were enzymatically dissociated and plated in DMEM containing 5% FBS and 100 g/ml penicillin/streptomycin. Neurons were cultured for 4, 6, 8, 12, 24, 36, and 48 h.
Quantification of Neurite Outgrowth-Neurites of PC12 cells were measured essentially as described elsewhere (29,30). PC12 cells were grown in six-well plates, and five pictures per well were randomly taken 24 h after treatment using a camera (Canon, Krefeld, Germany) linked to an inverted microscope with phase contrast illumination (Olympus, Hamburg, Germany). Each visual field contained on average 40 living cells. The number of neurites per visual field was measured using Adobe Photoshop (Adobe Systems, San Jose, CA). This number was divided by the total number of cells per visual field and expressed either as the percentage of untreated cells or as the percent inhibition of NGF-treated cells within the corresponding experiment.
Immunoprecipitation-For immunoprecipitation followed by LC-MS/MS analysis, PC12 cells were lysed for 10 min on ice in 50 mM Tris HCl (pH 7.4), 150 mM NaCl, 1% Triton X-100, 0.25% sodium deoxycholate, 1 mM EDTA, and all the protease and phosphatase inhibitors as described above. The protein concentration was determined using the Coomassie Plus Protein Assay Kit (Pierce). JNK1 (Biovision, Milpitas, CA) and JNK2 (Novus Biologicals) were incubated for 1 h at 4°C with protein A Sepharose beads (Biovision) and then crosslinked for 1 h at room temperature in 200 mM sodium borate (pH 9.0), 25 mM dimethyl pimelimidate (Sigma-Aldrich). After three washes in 200 mM sodium borate, the antibody-bead complexes were blocked in 200 mM ethanolamine (pH 8.0) for 2 h at 4°C. Equal amounts of protein extracts were incubated with bead-coupled antibody overnight at 4°C. After two washes with lysis buffer and one with 5 mM Tris HCl (pH 7.4), proteins were eluted with 100 mM glycine (pH 3.0). For immunoprecipitation followed by immunoblot analysis, cells were lysed for 30 min on ice in 20 mM HEPES (pH 7.9), 400 mM NaCl, 0.025% Nonidet P-40, 1 mM EDTA, 1 mM EGTA, and all the protease and phosphatase inhibitors as described above. The protein concentration was determined using the Coomassie Plus Protein Assay Kit (Pierce). JNK1/3 or Nono antibody (both from Santa Cruz Biotechnology) was bound to protein A Sepharose beads (Biovision) in PBS for 3 h at 4°C. Equal amounts of protein extracts were incubated with bead-coupled antibody in lysis buffer without Nonidet P-40 overnight at 4°C. After three washes with lysis buffer without Nonidet P-40, samples were boiled in LDS sample buffer (Invitrogen) and further processed via immunoblot analysis as described above.
LC-MS/MS-LC-MS/MS analysis was performed as described previously (33). Peptide mixtures were separated via reversed phase chromatography using the Eksigent NanoLC-1D Plus system (Eksigent, Dublin, CA) on in-house-manufactured 10-cm fritless silica microcolumns with an inner diameter of 75 m. Columns were packed with ReproSil-Pur C 18 -AQ 3-m resin (Dr. Maisch GmbH, Ammer-buch, Germany) (34). Separation was performed using a 10% to 60% acetonitrile gradient (155 min) with 0.5% acetic acid at a flow rate of 200 nl/min. Eluting peptides were directly ionized via electrospray ionization and transferred into the orifice of an LTQ-Orbitrap hybrid mass spectrometer (Thermo Fisher, Waltham, MA). Mass spectrometry was performed in the data-dependent mode with one full scan in the Orbitrap (m/z ϭ 300 -1700; r ϭ 60,000; target value ϭ 1 ϫ 10 6 ). The five most intense ions with a charge state greater than 1 were selected (target value ϭ 5000; monoisotopic precursor selection enabled) and fragmented in the LTQ using collision-induced dissociation (35% normalized collision energy, wideband activation enabled). The dynamic exclusion for selected precursor ions was 60 s.
MS Data Processing-The MaxQuant software package (version 1.3.0.5) was used to identify and quantify proteins (35). SILAC triplets or duplets were quantified using the following settings: heavy-label Lys-8, Arg-10 and Lys-4, Arg-6; a maximum of three labeled amino acids per peptide; and top 10 MS/MS peaks per 100 Da. Carbamidomethylation of cysteine was selected as a fixed modification, and oxidation of methionine and acetylation of the protein N terminus were used as variable modifications. Trypsin was selected as the protease (full specificity) with a maximum of two missed cleavages. MS/MS spectra were searched using the Andromeda search engine (36) against a UniProt rat database (release 2012-06, 37,327 entries) with an additional 248 common contaminants. All protein sequences were also reversed to generate a target-decoy database. A first search with a precursor mass tolerance of 20 ppm was used, and the main search was performed with a mass tolerance of 6 ppm (37). A mass tolerance of 20 ppm was selected for fragment ions. A minimum of six amino acids per identified peptide and at least one peptide per protein group were required. The false discovery rate was set at 1% at both the peptide and the protein level. Protein ratios were calculated from the median of all normalized peptide ratios using only unique peptides or peptides assigned to the protein group with the greatest number of peptides (razor peptides). Only protein groups with at least three SILAC counts were considered for further analysis.
Immunofluorescence Analysis-PC12 cells were incubated on poly-l-lysine-coated glass coverslips, fixed with 4% paraformaldehyde, and blocked for 1 h in PBS containing 10% goat serum and 0.3% Triton-X 100. The cells were incubated at a dilution of 1:500 with antibodies against JNK1 (Novus Biologicals), Nono, and Sfpq (Santa Cruz Biotechnology) overnight at 4°C. Alexa Fluor 488 goat antirabbit IgG and Alexa Fluor 568 goat anti-mouse IgG (Invitrogen) were used as secondary antibodies at a dilution of 1:1000. PC12 cells were counterstained with DAPI and mounted with ProLong Gold Antifade (Invitrogen). Nonspecific staining was assessed through incubation of cells in the absence of primary antibodies. Images were taken with a Leica TCS SP5 confocal microscope (Leica Microsystems, Wetzlar, Germany) using a ϫ63 objective and recorded by Leica LAS AF software. Images were processed using Adobe Photoshop (Adobe Systems). P19 neuronal cells and primary cortical neurons were fixed with 4% paraformaldehyde in 0.1 M sodium phosphate buffer (pH 7.4). Primary antibodies against JNK1, Sfpq (as described above), and doublecortin (1:2000; Santa Cruz) were used. Secondary antibodies conjugated to Cy2, Cy3, or Cy5 were used at a dilution of 1:500 (Jackson ImmunoResearch Laboratories, Suffolk, UK). Fluorescence was imaged with a ϫ63 objective on a Zeiss LSM 700 confocal microscope, and images were processed using Adobe Photoshop software. Primary cortical and P19 neuronal cultures were visualized with a ϫ40 objective, and co-localization signals of JNK1 and Sfpq were quantified on at least 10 randomly taken visual fields containing ϳ15 to 20 cells each.
Proximity Ligation Assay-PC12 cells were incubated on poly-llysine-coated 18-well -Slides (ibidi, Martinsried, Germany) and then subjected to primary antibody incubation as described above. The interactions between Sfpq-Nono and JNK1-Sfpq were assessed using the Duolink proximity ligation assay (Olink, Uppsala, Sweden) according to the manufacturer's instructions. Images were taken with a Leica TCS SP5 confocal microscope (Leica Microsystems) using a ϫ63 objective. Fluorescence spots were counted using ImageJ software (National Institutes of Health). Negative controls using JNK1 antibody alone with both secondary antibodies or secondary antibodies only did not show any signal.
siRNA Transfection-Transfection was carried out using the Amaxa Cell Line Nucleofector system according to the manufacturer's instructions with PC12-cell-specific settings (Lonza, Cologne, Germany). In brief, 2 ϫ 10 6 cells were transfected with 200 nmol Sfpq siRNA (Thermo Fisher) and seeded in a six-well plate. After 4 days, cells were starved for 24 h and then differentiated as described above. Sfpq knockdown efficiency was confirmed by immunoblot before NGF treatment.
Statistical Analysis-Band intensities of Western blots were evaluated densitometrically using Scion Image 4.0 (Scion Corporation) and are expressed as a percentage relative to unstimulated cells ("relative pixel density"). Comparisons between multiple groups were performed via one-way analysis of variance followed by a Bonferroni post hoc test, and comparisons between two groups were assessed via the standard Student's t test (Prism 4.0, GraphPad Software, San Diego, CA). All data are expressed as mean Ϯ S.E. Differences were considered statistically significant at p Ͻ 0.05. Cluster analysis was performed using Perseus (MPI, Munich, Germany) and PANTHER molecular function terms in the DAVID bioinformatics database (DA-VID Bioinformatics Resources, National Cancer Institute, Frederick, MD) (38).

RESULTS
Nerve Growth Factor Induces JNK-dependent Neurite Outgrowth in PC12 Cells-In order to study JNK interaction partners during neuronal differentiation via quantitative mass spectrometry, we decided to use the very well-characterized rat pheochromocytoma cell line PC12 (19). The hallmark of PC12 cells is that treatment with NGF induces a major shift in phenotype, from proliferating tumor cells to non-dividing neurons. For our proteomics experiments, PC12 cells were grown in SILAC media containing dialyzed horse and calf sera and lower concentrations of lysine and arginine than in standard RPMI media. 24 h before differentiation, cells were starved in SILAC media containing 1% dialyzed horse serum. To assess the role of JNK in NGF-induced neuronal differentiation, we stimulated PC12 cells with a single dose of NGF (100 ng/ml) in the presence or absence of the ATP-competitive JNK inhibitor SP600125 (Fig. 1A). Without the inhibitor, NGF-induced neurite outgrowth was clearly visible at both 6 and 24 h after stimulation. When NGF was not replenished, neurites were largely degenerated after 48 h. Pretreatment of cells with SP600125 decreased NGF-induced neurite outgrowth by ϳ60% at 6 h and 24 h. This result is in line with previous studies showing that JNK is an important mediator for neuronal differentiation in PC12 cells (23,26,39,40). To analyze JNK activity, we quantified phosphorylation levels via Western blotting (Fig. 1B). We found increased JNK activation beginning at 1 h after treatment, with a maximum at about 2 h. Total JNK levels did not change significantly. Western blot results also showed that both JNK isoforms (p54 and p46) were phosphorylated in response to NGF. In summary, these results confirm that JNK activity is required for neurite outgrowth and establish PC12 cells as a model for the study of NGF-induced JNK-dependent signaling via SILAC-based quantitative mass spectrometry. NGF treatment had no effect on total JNK levels. Arrows indicate the two different p54 and p46 JNK isoforms. **p Ͻ 0.01 versus unstimulated (n ϭ 4 to 7).

JNK Interacts with Proteins of Different Molecular Functions during Neuronal
Differentiation-Phosphoproteomics is a powerful approach for the study of cellular signal transduction (41). However, in the case of JNK, this strategy is complicated by several factors. First, the consensus JNK phosphorylation site is a typical proline-directed phosphorylation motif shared by several other kinases such as ERK1/2 and cyclin-dependent kinases. Second, it has been shown that JNK sites can also be phosphorylated by other kinases (6). These two factors make it difficult to discern JNK substrates from substrates of other kinases in a classical phosphoproteomic experiment. Finally, focusing exclusively on JNK substrates would not provide any information about upstream signaling events that lead to JNK activation. Because of these limitations, we pursued an alternative proteomic approach. We took advantage of the fact that JNK physically associates with effectors via a docking site, the so-called JNK binding domain (JBD). JBDs mediate the interaction of JNK with substrates, as well as with adaptor proteins and upstream activators (6). Therefore, studying dynamic changes in JNK interaction partners can provide insights into both upstream and downstream signaling events.
Our experimental strategy was to precipitate JNK from stimulated PC12 cells and characterize dynamic changes in the JNK interactome via SILAC-based quantitative shotgun proteomics ( Fig. 2A). Metabolic labeling of PC12 cells with either "medium" (Lys-4, Arg-6) or "heavy" (Lys-8, Arg-10) SILAC media for six passages resulted in essentially complete labeling (data not shown). Labeled cells were then treated with NGF for different lengths of time. To cover the dynamic changes in JNK-associated proteins, we performed three parallel triple SILAC experiments with one time point as an internal reference. This procedure allows analysis of temporal changes of the interactome (42). Cells were lysed, and JNK and associated proteins were immunoprecipitated with beadcoupled JNK1-and JNK2-specific antibodies. Immunoprecipitates from the three corresponding SILAC states were combined, washed, and analyzed via high-resolution quantitative shotgun proteomics. Protein data from the three timeseries measurements were combined into a single time course using the common 2-h time point as a reference (i.e. by calculating the ratio of ratios). To further increase the robustness of the data, we combined data from two independent biological replicates, and only proteins that were quantified with at least three SILAC ratio counts at each time point were considered. This analysis resulted in a list of 728 proteins. The list of identified proteins was expected to contain three different groups of proteins: (i) NGF-regulated JNK interaction partners, (ii) constitutive JNK interactors, and (iii) nonspecific contaminants. As our aim was to assess the role of JNK in NGF-induced neurite outgrowth, we focused our analysis on only the first group. To extract potential NGFregulated interaction partners, we selected all proteins with a log 2 fold change Ն 0.7 and/or Յ Ϫ0.7 at any time point (n ϭ 201) (supplemental Table S1). We then used hierarchical clustering to classify these proteins according to their dynamic behavior. The abundance profiles of proteins fell into three major clusters (Fig. 2B).
The first cluster contained proteins that showed increased abundance at both 0.5 to 1 h and 24 h of NGF stimulation with transiently decreased levels between 2 and 6 h. This cluster mainly contained actin-binding cytoskeletal proteins such as tropomodulin-2, tropomyosin ␣-4 chain, and several members of the myosin family involved in cytoskeleton reorganization and ATP-dependent movement along microfilaments. Furthermore, the motor protein kinesin light chain was found in this cluster, which is in line with previous studies demonstrating that JNK is involved in kinesin-dependent transport processes and microtubule dynamics in neurons (11,13,43).
The third cluster contained proteins that belonged to the families of large G-proteins and small GTPases including guanine nucleotide-binding protein G(o) subunit ␣, guanine nucleotide-binding protein G(i) subunit ␣-2, Rap1b, Rac1, RhoA, and Cdc42. These proteins showed increased levels 0.5 h to 2 h after NGF treatment followed by a steady decrease over time below control levels at 24 h. This finding suggests that NGF-dependent JNK activation might be induced via multiple G-proteins. Indeed, it has been shown that several G-proteins are involved in JNK activation during PC12 cell differentiation (25,26) and that Rap1 is involved in NGF-induced PC12 cell differentiation (44,45).
The second cluster contained proteins that showed delayed association with JNK starting 2 to 24 h after NGF stimulation. We were surprised to find that this cluster was strongly enriched in RNA binding proteins. For example, this subset contained proteins involved in mRNA splicing such as Fus and SF1 (46) and RNA helicases such as Ddx-1, -5, and -17. Intriguingly, two mRNA binding proteins in this cluster showed transient JNK binding at 2 to 3 h of stimulation. This kinetic behavior is similar to the activation profile of JNK (Fig. 1B). These two proteins, namely, non-POU domain-containing octamer-binding protein (Nono) and splicing factor proline-and glutamine-rich (Sfpq), are components of mRNA transport granules in dendrites (47). Additionally, we identified paraspeckle component 1 (Pspc1) in the cluster. Pspc1 is a paralogue of Nono and a known component of neuronal transport granule protein (47). Pspc1, Nono, and Sfpq are the three mammalian members of the Drosophila behavior and human splicing (DBHS) family involved in different aspects of RNA biology (48,49). As components of paraspeckles and neuronal granules, they are found in the nucleus and in the cytosol, respectively. DBHS proteins have been linked to neuronal function. Sfpq mediates neuronal development in zebrafish, and Nono is required for normal vision and courtship behavior in fruit flies (50,51). FIG. 2. Quantitative mass spectrometry timecourse analysis of potential JNK interacting proteins after NGF treatment. A, PC12 cells were grown in light, medium, or heavy SILAC media and were stimulated with NGF for different lengths of time. After JNK immunoprecipitation, samples were combined, digested with LysC and trypsin, and analyzed via LC-MS/MS. NGF-induced changes in JNK protein binding were calculated by using medium (2-h NGF-treated) cells as an internal reference. The illustration shows an example for a protein that transiently binds to JNK with the highest affinity at 6 h after NGF treatment. B, data from two independent biological experiments were combined and diated by JBDs. To this end, we used the peptide inhibitor D-JNKI-1, which blocks interaction of JNK with its substrates (52). D-JNKI-1 consists of the protease-resistant D-retroinverso form of a 10-amino-acid HIV-Tat sequence, which allows for cellular uptake, and the 20-amino-acid sequence of the JBD of JNK interacting protein 1 (53)(54)(55). Thus, D-JNKI-1 specifically blocks interactions of JNK mediated by JBDs. D-JNKI-1 has been shown to be highly specific and has been used to block JNK activity in several neuronal disease model systems (56 -58).
To see whether D-JNKI-1 blocks NGF-induced neurite outgrowth, we pretreated PC12 cells with D-JNKI-1 for 1 h. D-JNKI-1 dose-dependently decreased NGF-induced neurite outgrowth as assessed 24 h after stimulation (Fig. 3A). The inhibitory effect of D-JNKI-1 at a concentration of 10 M was ϳ60%, which is comparable to the inhibitory effect of the ATP competitive inhibitor SP600125 at the same concentration. Pretreatment with the control peptide D-TAT alone had no effect on NGF-induced neurite outgrowth, indicating that the inhibitory effect of D-JNKI-1 on neurite outgrowth is JNK specific. Additionally, we tested the inhibitory effect of D-JNKI-1 via phospho-specific Western blot analysis of the major JNK downstream target c-Jun. As expected, D-JNKI-1 (10 M) significantly inhibited NGF-induced c-Jun phosphorylation at 3 h after NGF treatment (data not shown). These results corroborate our previous findings and suggest that JNK interaction with its substrates is necessary for NGFinduced neurite outgrowth.
To investigate the effect of D-JNKI-1 on the JNK interactome, we focused on 3-h NGF stimulation, as this time point coincides with high JNK activation. Light cells were left untreated, medium-heavy cells were pretreated for 1 h with D-JNKI-1 (10 M), and heavy cells were pretreated with the control peptide D-TAT. Both medium-heavy and heavy cells were then stimulated with NGF for 3 h (Fig. 3B). JNK was precipitated from all three cell populations, and JNK-associated proteins were identified via mass spectrometry as described above. As a control we analyzed supernatants from the immunoprecipitations, and we did not observe global effects of NGF or D-JNKI-1 on the PC12 proteome (supplemental Fig. S1A). For JNK-associated proteins, we plotted ratios of NGFϩD-TAT treated samples versus unstimulated samples (x-axis) and of NGFϩD-TAT treated samples versus NGFϩD-JNKI-1 treated samples (y-axis) against each other (Fig. 3C). The resulting plot shows which proteins interacted with JNK in an NGF-and/or D-JNKI-1-dependent manner.
We found that NGF-induced and D-JNKI-1-inhibited interactions were clearly correlated. Specifically, most of the NGF-induced interactions were inhibited by D-JNKI-1. Thus, NGF-induced interactions seem largely JBD dependent. Interestingly, interactors fell into two groups. The first group was strongly NGF dependent and moderately affected by the inhibitor (highlighted in green in Fig. 3C). This set of proteins was enriched in actin-binding cytoskeletal proteins such as drebrin, a protein that is involved in axonal growth (59). This subset also included filamin-B, an actin regulatory protein that has been described as a JNK scaffold protein that tethers Rac1 and a JNK-specific module (60). The second group of proteins was strongly inhibited by D-JNKI-1 and showed a moderate response to NGF (highlighted in red in Fig. 3C). This group contained the above-mentioned DBHS family proteins Sfpq, Nono, and Pspc1 and several other known components of RNA transport granules (Pur ␣ and Pur ␤, several FMRPs, DDX1, DDX5, DDX6, and DDX17). To test whether other proteins in our dataset were also involved in RNA transport, we compared our data to a proteomic analysis of RNA transport granules (47). 26 out of 42 RNA transport granule proteins from a study by Kanai et al. were found in our dataset (supplemental Fig. S1B), and 18 out of these 26 were classified as JNK-specific interaction partners (log 2 fold change Ն 0.7). This is a highly significant enrichment (p ϭ 2.81E-07, hypergeometric test) for mRNA transport granule proteins in the NGF-and D-JNKI-1-dependent JNK interactome. In summary, our experiments with the JNK-specific blocking peptide D-JNKI-1 confirm that JNK interacts with cytoskeletal proteins and RNA transport granule proteins in an NGF-dependent manner. The interactions can be specifically blocked with D-JNKI-1 and are thus apparently mediated by JBDs. JNK interacting proteins from this experiment are listed in supplemental Table S2.
JNK1 Interacts with the Nono-Sfpq Heterodimer in an RNAdependent Way-To the best of our knowledge, the interaction of JNK with RNA transport granule proteins had not been described at the time of this study. We therefore performed follow-up experiments to validate and further characterize this observation. First, we sought to validate the mass spectrometry data via immunoprecipitation and Western blotting. We observed that JNK1 co-immunoprecipitated with Sfpq, Nono, and Pspc1 (Fig. 4A). Whereas NGF treatment increased the interaction, pretreatment of cells with D-JNKI-1 decreased the NGF-induced interaction to baseline levels. These results confirm that (i) JNK interacts with DBHS family proteins in PC12 cells, (ii) the interaction is increased by NGF stimulation, and (iii) the interaction is JBD dependent.
Although co-immunoprecipitation of endogenous proteins is considered the gold-standard assay for protein-protein analyzed with MaxQuant. Proteins with a log 2 fold change Ն 0.7 and/or log 2 fold change Յ Ϫ0.7 at any time point given were extracted and z-scored. Proteins with similar JNK binding kinetics were clustered using Perseus software (Euclidean distance, average linkage, n ϭ 201). Analysis of the molecular function of potential JNK interacting proteins using Protein Analysis through Evolutionary Relationships (PANTHER) revealed an enrichment for cytoskeletal protein (cluster 1, n ϭ 25), nucleic acid binding (cluster 2, n ϭ 73) and G-proteins (cluster 3, n ϭ 104). The three most significant PANTHER terms per cluster are shown. interactions, it is still affected by possible antibody crossreactivities. To verify the specificity of the assay, we performed the inverse experiment and used an antibody against Nono for immunoprecipitation. In addition, we used RNase A treatment to test whether the interactions were RNA dependent. In these experiments we found that NGF treatment led to a ϳ2.5-fold induction of JNK1 binding to Nono (Fig. 4B). RNase A treatment decreased the interaction, indicating that it was RNA dependent. The interaction seemed to be specific for JNK1, as JNK2 was not detected in the Nono pull-downs. Sfpq was also observed to co-precipitate with Nono, and this interaction was not affected by NGF or RNase A treatment. This result is in line with a previous study demonstrating that Nono and Sfpq form a stable, RNase-resistant, mRNAbinding heterodimer in neurons (47). Collectively, these results further confirm the interaction of JNK and mRNA transport granule proteins and indicate that JNK1, rather than JNK2, is the main interactor. Because we observed that the interaction was RNA dependent, we suggest that JNK1 associates with RNA-containing multi-protein complexes rather than with individual proteins.
NGF Treatment Induces JNK1 Interaction with Sfpq in the Cytosol-DBHS family proteins and JNK can be found in the nucleus and the cytosol. We therefore analyzed the subcellular distribution of Nono and Sfpq in PC12 cells via confocal microscopy. Representative images of three independent experiments showed that Nono and Sfpq were most abundant in the nucleus but were also found in the cytosol of PC12 cells (Fig. 5A, upper panel). This result was also confirmed by Western blot analyses of cytosolic and nuclear fractions from two independent experiments (Fig. 5B). NGF treatment for 3 h did not significantly alter the distribution of the total cellular Nono and Sfpq pools (Fig. 5B). To analyze the interaction of JNK1 and Sfpq, we stained stimulated and unstimulated PC12 cells with JNK1-and Sfpq-specific antibodies. Interestingly, increased co-localization of JNK1 and Sfpq was observed in outgrowing neurites of NGF-treated cells relative to unstimulated cells (Fig. 5A, lower panel). Evaluation of JNK1 stained cells alone revealed that JNK1 was mainly expressed in the cytosol of untreated and NGF-treated cells (Fig. 5A,  lower panel, arrowheads). Consistent results were obtained in Western blots of cytosolic and nuclear extracts (Fig. 5B). These results are also supported by a previous study showing that JNK1 is mainly located in the cytosol in PC12 cells (61). Similar to JNK1, JNK2 was mainly detected in the cytosol, but it also showed considerable nuclear localization. NGF-induced phosphorylation of both JNK isoforms occurred mainly in the cytosol (Fig. 5B).
Unbiased assessment of changes in protein co-localization via microscopy is challenging. Therefore, to confirm the increased co-localization of JNK1 and Sfpq, we employed the proximity ligation assay (PLA) (62). PLA uses the enzymatic ligation of connector oligonucleotides that are attached to different antibodies and rolling circle amplification to increase the signal derived from individual interactions. With this method one can detect the co-localization of endogenous proteins in an unbiased and quantitative manner. As a positive control, we tested the interaction of Nono and Sfpq. Evaluation of PLA signals in cells co-stained with DAPI revealed that the Nono-Sfpq heterodimer was located mainly in the nucleus and to a minor extent in the cytosol (Fig. 6A). Next, we used PLA to quantify the co-localization of JNK1 and Sfpq (Fig. 6B). Quantitative analysis of PLA signals revealed that NGF treatment led to a significant induction (ϳ5-fold) of JNK1-Sfpq interaction. In line with our previous findings, PLA signals of the JNK1-Sfpq interaction in NGF-treated cells were mainly observed in the cytosol, whereas JNK1-Sfpq interaction in unstimulated cells was found in the nucleus. Together, these results indicate that the Nono-Sfpq heterodimer is mainly located in the nucleus, but JNK1 interacts with Nono-Sfpq in an NGF-dependent manner mainly in the cytosol.
JNK1 and Sfpq Transiently Co-localize in Primary Cortical Neurons and P19 Neuronal Cells-To further study the association of JNK1 and Sfpq in the cytosol of differentiating neurons, we made use of two additional neuronal models, (i) mouse primary cortical neurons and (ii) neurons derived from the murine embryonic carcinoma cell line P19. Primary cortical neurons spontaneously form neurites after plating, continuing in vitro their differentiation process (28). P19 cells differentiate into neurons upon treatment with retinoic acid and produce numerous neurites similar to primary cortical neurons (63).
Confocal microscopy of primary cortical neurons and P19 neuronal cells revealed that JNK1 and Sfpq, like in PC12 cells, were present in the nucleus and in the cytosol, particularly during the first 12 h of differentiation, which coincides with the polarization and outgrowth of primary neurites in both neuronal models (Figs. 7A-7C and 7E-7G; arrowheads in insets 7AЈ-7CЈ and 7EЈ-7GЈ). Confocal images showed that JNK1 was mainly localized in the cytosol (Figs. 7AЉ-7HЉ), whereas Sfpq localized in the cytosol only during the polarization and outgrowth of primary neurites (Figs. 7Aٞ-7Hٞ). At later stages, Sfpq was found exclusively in the nucleus (Figs. 7D, 7H, 7Dٞ, and 7Hٞ). Quantification of neurons stained for JNK1 and Sfpq revealed a maximal co-localization of both molecules in the cytosol of immature neurons during the first 12 h of analyzed via LC-MS/MS. C, 701 proteins were quantified with at least three ratio counts. PANTHER analysis showed an enrichment for cytoskeletal proteins (n ϭ 65) and nucleic acid binding proteins (n ϭ 94), which specifically co-immunoprecipitated with JNK after NGF treatment. A log 2 fold change of 0.7 was applied as a cutoff. The three most significant PANTHER terms per cluster are shown. Highlighted proteins are discussed in the "Results" section. neuronal differentiation, whereas no co-localization was observed after 24 h (Fig. 7I).
Together, these results are consistent with the data obtained from PC12 cells. The fact that transient co-localization of JNK1 and Sfpq occurred in the same subcellular location and with similar kinetics in three distinct neuronal models strongly suggests that interaction of JNK with RNA transport granules is a generic feature during neuronal differentiation (Fig. 7J).
Sfpq Contributes to NGF-induced Neurite Outgrowth in PC12 Cells-To further investigate whether the JNK interaction partner Sfpq is involved in neuronal differentiation, we treated PC12 cells with Sfpq siRNA for 4 days. Western blot results for PC12 whole cell lysates showed that Sfpq was dose-dependently knocked down (Fig. 8A). Semi-quantitative analysis of band intensities revealed that knockdown efficiency was almost complete (97%) at a dose of 200 nmol siRNA. We then treated Sfpq knockdown cells with NGF (100 ng/ml) and measured neurite outgrowth 24 h after stimulation (Fig. 8B). Sfpq knockdown led to a significant decrease (ϳ50%) in NGFinduced neurite outgrowth, demonstrating that Sfpq is involved in the neuronal differentiation of PC12 cells. In parallel, we pretreated mock-transfected or Sfpq siRNA transfected cells with JNK inhibitor SP600125 (10 M) for 1 h before NGF stimulation. SP600125 significantly decreased NGF-induced neurite outgrowth by ϳ60% in mock-transfected cells. Thus, Sfpq knockdown and JNK inhibition have a similar effect on neurite outgrowth. Interestingly, treating cells with both SP600125 and Sfpq siRNA did not further increase the inhibition of neurite outgrowth. The lack of a cumulative effect is consistent with the hypothesis that JNK and Sfpq mediate NGF-induced neurite outgrowth via the same pathway and that interaction between JNK and RNA transport granule proteins is involved in neuronal differentiation.

DISCUSSION
Signal transduction cascades are typically represented as graphs that begin with the activation of a cell surface receptor and end with the activation of a transcription factor. However, it is now clear that gene expression is also regulated at the posttranscriptional level. For example, microRNAs and RNA binding proteins interact with mRNAs and control their splicing, export, stability, localization, and translation (64,65). In fact, translation efficiency appears to be the single best predictor of cellular protein abundance (66). Posttranscriptional regulation is particularly relevant in highly compartmentalized cells such as neurons, where specific functions occur in restricted subcellular domains (67). Although the link between signaling cascades and transcriptional control is quite well studied, the connections between cell signaling and posttranscriptional regulatory events are still poorly characterized.
It is known that JNKs play an important role in both neuronal cell death and neuronal differentiation (5)(6)(7)(8). Whereas apoptosis induction appears to depend on nuclear JNK, cy- tosolic JNK activity is required for neurite outgrowth (5,11). Here, we have shown that JNK dynamically associates with RNA binding proteins in an early phase during neuronal differentiation. First, we showed that JNK activity is required for NGF-induced neurite outgrowth, consistent with previous reports (23,24). We then used quantitative interaction proteomics to identify proteins associating with JNK in an NGF-dependent manner. Surprisingly, we observed that several RNA-binding transport granule proteins including Nono, Sfpq, and Pspc1 temporally associated with JNK. We validated the specificity of these interactions by showing that they could be blocked by D-JNKI-1, a highly specific peptide inhibitor of JNK. Moreover, we used Western blotting to show that Nono, Sfpq, and Pspc1 co-precipitate with JNK1 but not JNK2 in an RNA-dependent manner. Using fractionation experiments, confocal microscopy, and proximity ligation assay, we found that this interaction mainly occurred in the cytosol of differentiating PC12 cells. Importantly, we were able to replicate our findings in P19-derived and primary neurons, strongly suggesting that the interaction is a generic feature of neuronal , and nuclei (blue) co-stained PC12 cells show that Nono and Sfpq were expressed in the nucleus as well as in the cytosol. Co-localization mainly occurred in the nucleus, where the greater fraction of Nono and Sfpq is located (upper panel). The cellular distribution of JNK1 (green) showed that JNK1 was mainly localized in the cytosol (lower panel). NGF treatment for 3 h led to an increased co-localization signal of JNK1 (green) and Sfpq (red) in the cytosol. Scale bar: 6 m. B, Western blot analysis of cytosolic and nuclear fractions revealed that NGF treatment for 3 h led to JNK phosphorylation mainly in the cytosol. Reprobing with JNK1-and JNK2-specific antibodies showed that JNKs were mainly localized in the cytosol, whereas the JNK2 p54 isoform was also found in the nucleus. Reprobing with Nono-and Sfpq-specific antibodies corroborated the cellular distribution of Nono and Sfpq noted in confocal microscopy analysis. Representative immunoblots are shown (n ϭ 2).
differentiation. The fact that the temporal interaction was observed during the early stages of neurite extension suggests that this interaction might play a role in the initiation of posttranscriptional gene regulation events.
Nono, Sfpq, and Pspc1 are the mammalian members of the DBHS family involved in many aspects of RNA biology (48,49). Most research on these proteins so far has focused on their role in the nucleus, where they are components of subnuclear bodies termed paraspeckles. Although the function of paraspeckles is not well understood, they have been reported to retain mRNAs in the nucleus to allow their rapid release into the cytosol upon stress (68). In neuronal cells, all three proteins have been described as components of RNA transport granules (47). Sfpq has been shown to be involved in neuronal development in zebrafish (50), and we show here that knocking down Sfpq impaired neurite outgrowth. Nono is required for normal vision and courtship behavior in fruit flies (51). Thus, DBHS proteins have neuronal functions in vivo. We observed that JNK1 interacted with Sfpq in the cytosol of PC12 cells. Therefore, our data suggest that JNK1 mediates neuronal differentiation at least partially via interaction with DBHS family proteins in the cytosol. Interestingly, we found that the NGF-induced interaction of JNK1 with Nono/Sfpq was RNA-dependent. This suggests that JNK1 associates with entire transport granules in an NGF-dependent manner. In fact, the NGF-induced JNK interactome was strongly enriched in known components of RNA transport granules. It should therefore be stressed that we have no evidence that JNK1 is a direct binder of Sfpq and/or Nono. In fact, given that the interaction was RNA dependent, it appears more likely that JNK1 associates with higher order complexes containing these proteins and RNAs, possibly via its low-complexity region (see below). Both JNK and neuronal granule proteins are known to be involved in neuronal differentiation. However, to the best of our knowledge, a link between JNK signaling and neuronal granules has not yet been described.
RNA granules are particularly important in neurons for the targeting of mRNAs to distinct cellular compartments for localized translation (67,69). How proteins and RNAs are targeted to and released from RNA granules is not yet entirely understood. Recently, it has been reported that proteins can assemble into RNA granules via low-complexity regions (70,71). How this assembly and disassembly are regulated is largely unknown. One possible mechanism involves phosphorylation. For example, the kinase DYRK3 has been shown to interact with stress granules, presumably via a low-complexity region (72). DYRK3 inhibition stabilizes the granules, suggesting that this kinase can disassemble granules via the phosphorylation of specific target proteins. Similar to DYRK3, JNK1 has a predicted low-complexity region that is disordered in the crystal structure (73). It is tempting to speculate that JNK1 might be involved in regulating transport granules. For example, granule-associated JNK1 might phosphorylate substrate proteins in the granules to induce their release. As for DYRK3, these putative JNK1 substrates remain to be identified. In addition to RNA binding proteins, we found that JNK1 also interacted with proteins involved in dendritic and/or axonal transport via Co-localization and cellular distribution of Nono, Sfpq, and JNK1 assessed via proximity ligation assay (PLA). PC12 cells were incubated with Nono-and Sfpq-specific antibodies followed by PLA secondary antibodies (red) and were analyzed via confocal microscopy. Nuclear counterstaining with DAPI (blue) revealed that the Nono-Sfpq heterodimer was mainly located in the nucleus (upper panel). Incubation with JNK1-and Sfpq-specific antibodies followed by PLA staining showed a ϳ5.4-fold increase in JNK1-Sfpq interaction after NGF treatment for 3 h (lower panel). **p Ͻ 0.01 versus unstimulated (n ϭ 8 to 9). Nuclear counterstaining with DAPI showed that JNK1 and Sfpq mainly interacted in the cytosol of NGF-treated cells. Results were obtained from three independent experiments. Scale bar: 6 m. microfilaments and microtubules (kinesin light chain, filamin-B, and drebrin). It is therefore possible that JNK1 controls the transport of RNA granules. Consistently, JNK1 is known to control the phosphorylation of microtubuleassociated proteins and to be required for the maintenance of neuronal microtubules (13). In summary, our unbiased proteomic analysis of the dynamic JNK interactome reveals that JNK dynamically associates with RNA transport granule proteins during neuronal differentiation. Although the exact mechanisms involved in the interaction of JNK with neuronal granules remain to be investigated, blocking the interaction with the highly selective peptide D-JNKI-1 inhibits neurite outgrowth. Thus, interactions mediated via JBDs appear to be functionally important for neuronal differentiation. Furthermore, we show that Sfpq, a member of RNA transport granules, is involved in neuronal differentiation in PC12 cells. However, several important questions remain. First, it is not known which RNAs are located in the JNK-associated neuronal granules. Furthermore, it might be informative to see which proteins are phosphorylated by JNK upon NGF treatment, although the identification of direct kinase substrates in vivo is challenging. We also note that several well-described JNK interaction partners such as microtubule-associated protein, JNK interacting protein, and c-Jun were not identified in our dataset. It is not clear whether this is due to the specialized function of JNK in PC12 cells or to limitations of our experimental approach. Finally, JNK is not the only mediator of neuronal differentiation. How neurons integrate different signaling events in a spatio-temporal manner remains to be explored.