Dynamic remodeling of presynaptic endoplasmic reticulum is coordinated through actin and microtubule crosstalk and contributes to defective stimulus-response in Spinal Muscular Atrophy

Background Axonal degeneration and defects in neuromuscular neurotransmission represent a pathological hallmark in spinal muscular atrophy (SMA) and other forms of motoneuron disease. These pathological changes do not only base on altered axonal and presynaptic architecture, but also on alterations in dynamic movements of organelles and subcellular structures that are not necessarily reected by static histopathological changes. In neurons, a highly dynamic endoplasmic reticulum (ER) network exists in the axonal and presynaptic compartment which regulates Ca2+ homeostasis and synapse maintenance. However, the mechanisms of its dynamic regulation and mechanisms of dysfunction that contribute to neurodegeneration remain elusive. Methods Using high resolution microscopy and life imaging of cultured motoneurons from wildtype and a mouse model of spinal muscular atrophy, we investigated the dynamics of the axonal endoplasmic reticulum and ribosome distribution and activation. Results These studies revealed that the dynamic remodeling of ER in axonal lopodia of cultured motoneurons depends mainly on actin cytoskeleton. In Smn -decient motoneurons, movements of ER in lopodia seems to be more affected than in the growth cone core. In addition, ribosome assembly that happens within seconds after exposure to Brain derived neurotrophic factors (BDNF) is reduced in axon terminals of Smn -decient motoneurons, and also the association with ER as a response to extracellular stimuli is highly disturbed. Conclusions These ndings do not only dene a novel function of presynaptic ER in dynamic regulation of local translation. They also implicate impaired dynamic movements of axonal and presynaptic ER as a contributor to the pathophysiology of SMA and possibly also other neurodegenerative diseases.

In neurons, rough cisternal ER (RER), which is the major site for protein synthesis, folding, processing and secretion is mainly restricted to the somatodendritic compartment [11][12][13][14][15]. Axons are traditionally considered to be devoid of RER, as demonstrated by ultrastructure EM and thought to exhibit only smooth ER [16,17]. The function of such dendritic and axonal ER has been suggested to be limited to lipid metabolism, Ca 2+ homeostasis, and to function in contacting membranous organelles to regulate their biogenesis and maintenance [1,6,12,17]. Nevertheless, despite emerging evidence of intra-axonal translation of mRNAs encoding membrane and secreted proteins in axons of isolated neurons, there is no direct evidence for the existence of a rough ER in axons that could process such locally synthesized proteins for integration into the axoplasmic membrane and secretion [18].
Here, we investigated the interaction of ribosomes with ER in the presynaptic terminals of cultured motoneurons and found that ribosomes undergo rapid changes in distribution and structure in response to extracellular cues. In particular, they relocate to the ER where they could accomplish local translation of membrane-associated and secreted proteins. Furthermore, we unraveled the underlying mechanisms of regulation of presynaptic ER dynamic movements and show that fast dynamic elongation of ER into axonal lopodia is regulated mainly by actin, whereas slow ER movement in the growth cone core requires a coordinated actin and microtubule cytoskeleton. Thus, we provide evidence of a novel function of axonal ER in local protein synthesis in addition to its previously described roles.
This observation also seems to be relevant for the pathophysiology of spinal muscular atrophy (SMA), an autosomal recessive motoneuron disease caused by loss of SMN protein [19]. Isolated motoneurons from SMA mouse models show severe defects in mRNA localization and intra-axonal protein synthesis [20,21]. Our data implicate that ER dynamic movements are also impaired in Smn-de cient motoneurons. In particular, the rapid response of ribosomes to extracellular stimuli is not accomplished. Thus, in addition to mRNA mislocalization, aberrant ER dynamics and ribosome response to extracellular signaling could affect axonal growth and presynaptic function and maintenance, thereby contributing to the pathology of SMA.

Materials And Methods
Animals The SMA mouse model is a transgenic mouse expressing two copies of the human SMN2 gene, mated onto a mouse null Smn-/background, resulting in a severe phenotype resembling type I SMA in humans [82]. TrkB knockout mice [83] were obtained from the University of California, Davis (MMRRC: 000188, B6; 129S4-Ntrk2 < tm1Rohr >) and maintained on a C57Bl/6 background. CD1 mice were used for motoneuron cell cultures from WT mice. All mice were housed in the animal facility of the Institute of Clinical Neurobiology, University Hospital of Wuerzburg. All mouse procedures were performed according to the regulations on animal protection of the German federal law and of the Association for Assessment and Accreditation of Laboratory Animal care, approved by the local authorities.

Culture of embryonic mouse motoneurons
Embryonic mouse motoneuron culture from the lumbar spinal cord was performed as previously described [84]. Lumbar spinal cords were dissected from E13.5 mouse embryos, digested with trypsin (Worthington), triturated and then transferred onto a panning plate coated with anti-p75 antibody (MLR2, Abcam) for enrichment of motoneurons. For lentiviral transduction, pSIH-mCherry-KDEL construct expressing lentiviruses were added to the suspension of motoneurons before plating on polyornithine-(PORN) and human merosin-coated (CC085, Merck-Millipore) plates. Motoneurons were cultured in NB medium supplemented with 500 µM Glutamax, 2% heat-inactivated horse serum (Gibco), 2% B27 (Thermo Fisher Scienti c) and 5 ng/ml BDNF for 5 to 6 days in a humidi ed CO2 incubator at 37 o C.
Medium was changed 24 h after plating and then every other day. Merosin consists of Laminin211 (α2β1γ1), enriched in extrasynaptic basal lamina and Laminin221 (α2β2γ1), which is speci cally expressed at the cleft of NMJs and regulates formation, maturation and maintenance of NMJs [85,86]. Thus, culturing motoneurons on merosin induces the maturation of presynaptic structures in axon terminals. Compartmentalized motoneuron cultures were prepared as described previously [21]. To drive axonal growth into the axonal compartment, 20 ng/ml BDNF and CNTF were added into the axonal compartments, while only 5 ng/ml CNTF was added into the somatodendritic compartments.
Both mCherry-KDEL and GFP-actin were rst ampli ed by PCR and puri ed fragments were inserted into a pSIH-CMV-IRES vector. Lentiviruses were produced in HEK 293T cells using pCMV-VSVG and pCMVΔR8.91 helper plasmids [88]. HEK 293T cells were transfected with calcium-phosphate reagents, and viral supernatants were harvested after 47 h by ultracentrifugation. NSC34 cells were used for virus titer test.

Image acquisition and data analysis
Image acquisition was done with a standard Olympus Fluoview 1000 confocal system with a 60x NA 1.35 oil objective. Structured illumination microscopy (SIM) imaging was performed on a commercial ELYRA S.1 microscope (Zeiss AG). The setup is equipped with a Plan-Apochromat 63x/1.40 immersion-oil based objective and four excitation lasers, a 405 nm diode (50 mW), a 488 nm OPSL (100 mW), a 561 nm OPSL (100 mW) and a 642 nm diode laser (150 mW). For quanti cation of immuno uorescence signals, mean gray values of images were measured from unprocessed raw data after background subtraction using ImageJ-win64. For quanti cation of SIM data represented in Fig. 4, 6 and 9, co-clusters of RPL24/RPS6, Y10B/eEF2 as well as RPL24/RPS6/ER and Y10B/eEF2/ER were counted manually. For this, maximum projections of single SIM channels (RPL24, RPS6 and mCherry-ER or Y10B, eEF2 and mCherry-ER channels) were rst created using ImageJ. An automatic linear adjustment of contrast and brightness was applied to the whole z-projection image of each single channel and these were merged into a RGB image. Co-clusters were de ned as overlapping dots from RPL24, RPS6 and mCherry-ER channels or Y10B, eEF2 and mCherry-ER channels on the RGB image which had a diameter of more than 350 nm. GraphPad Prism 8 software was used for all statistical analyses. Data are shown in scatter dot plots with mean ± SEM, unless otherwise mentioned. For a better visibility, linear contrast enhancement was applied to all representative images using Adobe Photoshop.
Live cell Imaging and data quanti cation Approximately, 40,000 motoneurons were transduced with lentivirus expressing pSIH-mCherry-KDEL or pSIH-GFP-actin-IRES-mCherry-KDEL and cultured on PORN/ merosin-coated 35 mm high µ-dishes (81156, IBIDI) for 6 days. For live cell imaging, an inverted epi uorescence microscopy (TE2000; Nikon) was used that was equipped with a perfect focus system, heated stage chamber (TOKAI HIT CO, LTD) at 37 o C, 5% CO2 and 60x 1.4 NA objective. Time series were captured at a speed of 2 sec per frame over 15 min. 12bit images of 1.024 x 1.024 pixels were acquired with an Orca Flash 4.0 V2 camera (Hamamatsu Photonics), controlled by Nikon Element image software. For quanti cation of ER dynamics, Image Correlation Spectroscopy (ICS) [25] was implemented in python. This approach was used as described previously by Wiseman et,. al. Brie y, molecular movements of ER or molecular transport of other organelles are determined based on ow or diffusion in an image time series , with N, M, T indicating the two dimensions in space and the time dimension respectively. denote the corresponding running indices the basic work ow starts by de ning a space-time window, i.e., a subspace of over ∆t = 10 consecutive frames. This subspace is rasterized over the image with a sampling rate of ∆i = ∆j = 4. For each of those samples a correlation window is computed, describing the overlap of signal for a given shift per time frame. The maximum of this spectrum is the point of maximum correlation and indicates a shift of signal from one image to this next, by . To achieve sub-pixel accuracy a 2D Gaussian function is tted to the correlation spectrum using the Levenberg-Marquardt-Algorithm. The initial parameters were chosen as amplitude , center coordinates in , , center coordinates in , , standard deviation in , , standard deviation in , . An additional parameter rotating the coordinate system by was initialized as 0. The t results for and indicate a directed shift of signal. A larger or indicates an overall lower correlation, i.e., the signal losing its shape. To compute overall dynamics, three distinct subspaces are classi ed: (i) subspaces with a mean intensity less than 10% of the maximum intensity are considered noise. (ii) Subspaces with a mean intensity that is larger than 40% of the maximum intensity are classi ed as core. (iii) Subspaces with an intensity in between is considered as lopodia. The overall dynamics value for a class is the mean dynamics value of all subspaces. Kymograph-analysis of ER dynamics was carried out using Image J as described previously [20]. Brie y, a Z-projection image was made from all the frames of a live cell image, followed by the creation of a multiple kymograph along a line drawn tracking the ER movements in lopodia or core. Distance changed per movement and frequency of movements over 15 min were calculated from multiple kymographs and plotted in a graph. Growth cones without detectable ER movements were classi ed as failure lopodia or core ER movement.

ER dynamics are regulated by a coordinated actin/microtubule cytoskeleton in growth cones of cultured motoneurons
In growth cones of rat sympathetic neurons, membrane-bound organelles representing ER often colocalize with microtubules and hardly extend beyond microtubules [22]. Nevertheless, growth cone lopodia are enriched in actin laments and mostly lack microtubules [23]. Thus, we wondered whether presynaptic ER could extend into these actin-rich lopodia and associate with actin laments in addition to microtubules. In order to visualize the presynaptic ER, we transduced primary cultured motoneurons with a lentivirus expressing mCherry-KDEL, a well-studied ER marker [24]. Neurons were then xed and immunostained against F-actin and microtubule cytoskeleton using phalloidin and a-tubulin staining, respectively. The association of presynaptic ER with actin and tubulin was then assessed using Structured Illumination Microscopy (SIM) (Fig. 1a). Interestingly, the presynaptic ER was detected not only in the core of growth cones but also in lopodia (Fig. 1b). Notably, ER colocalized with both F-actin and microtubules in the core, while in lopodia, ER overlapped mostly with F-actin, as expected since microtubules are less abundant in lopodia (Fig. 1b, c). To con rm the colocalization of the presynaptic ER with F-actin in lopodia, we monitored the dynamics of ER/actin co-movements in lopodia by transducing motoneurons with a GFP-actin-IRES-mCherry-KDEL lentivirus that allows co-expression of GFP-actin and mCherry-ER. We captured live cell imaging of growth cones from motoneurons that were transduced with GFP-actin-IRES-mCherry-KDEL over 8 min with 2 sec intervals to visualize actin and ER movements simultaneously (Fig. 1d). Two color live cell imaging demonstrated that tubular ER extends into lopodia and retracts along actin laments indicating that ER interaction with actin de nes its rearrangement and remodeling in lopodia (Fig. 1d, Video 1). Representative multiple kymographs indicated coordinated movements of actin and ER in lopodia (Fig. 1e). Based on these observations, we hypothesized that ER extension into lopodia might be actin and not microtubule dependent. To address this, we transduced motoneurons with mCherry-KDEL and treated them with actin and microtubule depolymerizing drugs cytochalasin D (CytoD) or nocodazole, respectively. ER movements were then evaluated in the growth cone by live cell imaging ( Fig. 2a, b, Video 2-7). To measure ER dynamic movements, we implemented an adaption of Image Correlation Spectroscopy (ICS) [25] (Fig. 2c). In addition to ICS-analysis, we used multiple kymographs to measure the frequency and distance moved by presynaptic ER manually. Intriguingly, we found that the dynamic movement of presynaptic ER in lopodia is signi cantly higher than that in the core (Fig. 2d). Besides, the velocity of dynamic movements was higher, as represented in Fig. 2e, and the distance moved by ER in lopodia was greater compared to growth cone core, correspondingly (Fig. 2f). Moreover, we found that upon CytoD treatment, 80% of neurons failed to show ER movements in lopodia, while only 40% failed to show ER movements in the core (Fig. 2g). Treatment with nocodazole resulted in 50% failure in lopodia and 40% failure in core ER movements (Fig. 2g). ICS-analysis revealed that disruption of either actin or microtubule dynamics reduces ER movements signi cantly in both lopodia and core implying that ER dynamic movements depend on a coordinated actin/microtubule cytoskeleton (Fig. 2h, i). Nevertheless, analyzing the frequency of ER movements in lopodia demonstrated that disruption of actin but not microtubules reduced the frequency of ER movements (Fig. 2h). In contrast to lopodia, the frequency of ER movements in core was markedly reduced upon disruption of either actin or microtubule cytoskeleton which is in agreement with the above ICS-analysis (Fig. 2i). This nding was further con rmed by a treatment with both CytoD and nocodazole which severely impaired the ER dynamics in both lopodia and core, as nearly no more ER movements were detectable (Fig. 2g, Supplementary Figure 1A).
Collectively, these data provide evidence of a highly dynamic ER in distal axons that also includes presynaptic compartments of developing motoneurons. The dynamic movements of ER in axonal growth cones could be classi ed into fast movements in lopodia and slower movements in the core. Fast ER remodeling in lopodia is regulated particularly by actin cytoskeleton, whereas slow ER rearrangements in the core require a microtubule and actin crosstalk.

Extracellular stimulation triggers ribosome activation and initiates local translation in growth cones on a time scale of seconds
Deep RNA-seq combined with sensitive uorescence in situ hybridization (FISH) approaches have identi ed numerous mRNAs [26,27] within distal axons which are locally translated [20,28,29]. Locally synthetized proteins are necessary for neural-circuit development, survival and plasticity. In developing axons, local translation mediates an essential response to guidance cues required for path nding [30].
Brain-derived neurotrophic factor (BDNF) and its receptor TrkB play a vital role in modulation of local translation [31] and axonal cytoskeleton remodeling [32,33] in motoneurons. We sought to scrutinize the dynamics of ribosome activation in response to BDNF stimulation in the growth cone of motoneurons. Following BDNF binding, TrkB undergoes autophosphorylation and activates MAPK as well as PI3K-AKT signaling pathways [34], leading to activation of ribosomes and induction of local translation [35]. In order to de ne the precise kinetics of TrkB activation, we applied BDNF to motoneurons with a short pulse of 10 sec, 1 min and 10 min, washed it out and immunostained neurons against TrkB and pTrkB. The speci city of TrkB and pTrkB antibodies were con rmed by immunostaining using TrkB knockout mice (Supplementary Figure 2A, B). We could detect phosphorylated TrkB upon 10 sec BDNF pulse in the growth cone, as shown by immuno uorescence assay (Fig. 3a, b). Interestingly, the levels of TrkB also elevated in the growth cone after 1 min BDNF exposure (Fig. 3c). In the whole cell lysate, a corresponding elevation of pTrkB immunoreactivity became detectable rst at 1 min poststimulation, as shown by Western blot (Fig. 3d) and no increase in total levels of TrkB was detectable despite 10 min stimulation, indicating that this short pulse is insu cient to induce the transcription and translation of TrkB in the soma (Fig. 3d, Supplementary Figure 2C). The rapid increase in TrkB immunoreactivity within less than 1 min in the growth cone of immunostained motoneurons could be explained by release of this receptor from intracellular stores or changes in the receptor conformation, which favors antibody binding. The increase in TrkB immunoreactivity in growth cones at a later time point of 10 min poststimulation (Fig.  3c) could be explained by enhanced transport of TrkB from distal axons into growth cones and/or rapid neosynthesis in growth cones since TrkB mRNAs were detected in axons by qRT-PCR assay (Supplementary Figure 2D). Treatment of neurons with either anisomycin, which inhibits the translation, or nocodazole, which blocks the microtubule-based transport, prevents the increase in TrkB signal in growth cones (Fig. 3e, f). These data indicate that BDNF stimulation triggers redistribution of TrkB resulting in an increase in TrkB total immuno uorescence within 1 min, and also induces its local production within 10 min stimulation.
In developing axons, the rapid response to extracellular cues is assured by tight regulation of spatiotemporal changes in mRNA translation [36,37]. Thus, the kinetics of induction of ribosomal changes that trigger translation in growth cones should differ from those in the soma. To study the dynamics of ribosome remodeling, we used Y10B as a general marker for ribosomes and RPL8 as a marker for the 60S ribosomal subunit in order to study the dynamics of movement of these different subunits. We found that induction of BDNF/TrkB signaling leads to a rapid change in the distribution of these ribosomal markers within the growth cone. The immunoreactivities of ribosomal markers Y10B and RPL8 were rapidly altered and appeared increased after 10 sec BDNF pulse (Fig. 3g-i). Interestingly, short pulses of 10 sec and 1 min were not su cient to induce such changes in ribosome distribution in the soma as shown in Supplementary Figure 2E, albeit TrkB phosphorylation was detectable also in this subcellular compartment after 10 sec stimulation (Supplementary Figure 2F). This kinetic distinction suggests that in the growth cone, BDNF/TrkB signaling and downstream mechanisms for modulating ribosome distribution and possibly also activation are different from cell bodies. Increased immunoreactivity for ribosomal subunits in the growth cone within 10 sec BDNF stimulation is striking. Rapid transport of ribosomal subunits from axonal sites to the growth cone appears unlikely since the fastest measured microtubule-dependent axonal transport at a speed of 1 μm/sec cannot provide ribosomes from the axon shaft into growth cones in such a short time [38]. To exclude this, we treated neurons with nocodazole to disrupt microtubule-based axonal transport and found that disruption of axonal transport did not affect BDNF-induced augmentation of ribosomal subunits as illustrated in Fig.  3j. Similarly, inhibition of translation by anisomycin treatment results in a signi cant increase in ribosome immunoreactivity, indicating that this increase does not depend on de novo synthesis of ribosomal proteins within the growth cone (Fig. 3k). Strikingly, CytoD treatment completely prevents the ribosomal response to BDNF stimulation, as no increase in the signal intensities of Y10B and RPL8 were detectable upon either stimulation times (Fig. 3l, m). Based on these observations, we propose that elevated immunoreactivity of ribosomes within 10 sec stimulation might be due to actin dependent conformational and distribution changes in ribosomal subunits which occurs upon assembly of 80S subunits and thus alters accessibility of corresponding epitopes for detection with Y10B and RPL8 antibodies. To investigate this hypothesis, we stained motoneurons with antibodies against RPL24 of the 60S and RPS6 of the 40S subunits (Fig. 4a). Upon assembly of 80S ribosomes, the C-terminus of RPL24 interacts with RPS6, thereby forming a bridge between the two subunits [39]. We employed SIM with a resolution of ~120 nm to examine the interaction of RPL24/RPS6 in growth cones. These experiments were performed both with CD1 WT motoneurons (Supplementary Figure 1B) and Smn+/+; SMN2tgtg motoneurons for a better comparability with Smn-de cient motoneurons used for the experiment shown in Fig. 9. As depicted in Fig. 4b, within 10 sec of stimulation the co-clusters of RPL24/RPS6 increased by 3-fold compared to no stimulation condition. Thus, we reasoned that the rapid increase in Y10B immunoreactivity in response to extracellular stimulation involves extremely fast assembly of ribosomes into 80S subunits. The increased interaction of RPL24 and RPS6 ribosomal proteins and consequently the assembly of 80S ribosomes suggest that this conformational change could correlate with translation initiation. To visualize actively translating ribosomes, we co-stained neurons against Y10B as a structural component of ribosomes and the elongation factor eEF2, which associates only with actively translating ribosomes (Fig. 4c). Again, we found a signi cant increase in the number of Y10B/eEF2 co-clusters within 10 sec of stimulation, indicating that the response of ribosomes in growth cones to activation of TrkB receptors is very fast (Fig. 4d). For a direct comparison of Smn+/+; SMN2tgtg motoneurons with motoneurons from CD1 mice that were used as WT controls in other experiments, we counted the coclusters of RPL24/RPS6 as well as Y10B/eEF2 in motoneurons obtained from mice with CD1 genetic background. We observed a signi cant increase in RPL24/RPS6 co-clusters at 10 sec poststimulation and a tendency at 1 min poststimulation (Supplementary Figure 1B). Likewise, co-clusters of Y10B/eEF2 were increased signi cantly within 10 sec as well as 1 min BDNF stimulation (Supplementary Figure 1C).
To con rm that the rapid activation of ribosomes depends on BDNF/TrkB signaling, we stained motoneurons from TrkB knockout mice against Y10B and found that in TrkB knockout motoneurons ribosomes fail to respond to BDNF stimulation (Supplementary Figure 2G, H). Next, we questioned whether this rapid activation of ribosomes in response to extracellular stimuli associates with a temporal change in protein synthesis in axonal growth cones. To examine the dynamics of local translation, we applied a time series after BDNF pulse to motoneurons, xed them and stained against b-actin (Fig. 5a).
Remarkably, we observed a signi cant increase in b-actin signal intensity within 1 min of stimulation suggesting that either the velocity of protein synthesis is extremely high in growth cones or actin undergoes conformational changes in such a way that this leads to enhanced immunoreactivity (Fig. 5b).
To distinguish these two possibilities and to con rm that elevated levels of b-actin rely on protein synthesis and not transport from the axon, we pretreated neurons with anisomycin and nocodazole and applied BDNF pulses for 10 sec to 10 min. Inhibition of protein synthesis prevented the increase of b-actin protein levels, while disruption of microtubule-dependent axonal transport had no in uence (Fig. 5c, d). In addition, we used puromycin to measure the rate of newly synthetized proteins (Fig. 5e, f). Similar to bactin, within 1 min of stimulation, synthesis of de novo proteins was accomplished as detected by marked increase in puromycin immunoreactivity, supporting the observation of an extremely fast rate of local translation in growth cones (Fig. 5g). In contrast, 1 min likewise 10 min stimulations did not trigger translation in the soma under the same conditions, pointing again towards different translation kinetics in these distinct compartments (Supplementary Figure 2I). In line with that, inhibiting translation but not axonal transport prohibited elevated protein synthesis in response to extracellular stimuli as demonstrated by the puromycin assay (Fig. 5h, i). Taken together, these data indicate that in presynaptic compartments, BDNF/TrkB signaling activates the translational machinery within seconds and thus regulates local protein synthesis at an extraordinary high speed.
Rough ER is present in the presynaptic compartment and contributes to Brain-derived neurotrophic factor induced presynaptic protein translation Transcripts encoding membrane and secreted proteins as well as ER resident proteins have been reported to be present in axons, and the corresponding proteins are synthetized and delivered into the axoplasmic membrane [36,40,41]. Although initial ultrastructural approaches conducted more than 40 years ago have only identi ed smooth ER in neuronal processes [17], uorescence microscopy approaches have identi ed ER-associated proteins implicated in protein translocation, folding, and post-translational modi cations as well as proteins of the Golgi apparatus in distal axons suggesting that axons might contain RER [18]. Accordingly, we considered whether the presynaptic compartment harbors RER, thereby participating in processing of locally produced membrane and secretory proteins. To address this hypothesis, we transduced Smn+/+; SMN2tgtg motoneurons with lentiviruses encoding the ER marker mCherry-KDEL and co-stained against RPL24 and RPS6 as markers of entirely assembled 80S ribosomes (Fig. 6a). In contrast to unstimulated neurons, a signi cant number of RPL24/RPS6 co-clusters colocalized with ER in presynaptic compartments of stimulated neurons, and this colocalization increased after 10 sec as well as 1 min BDNF pulses (Fig. 6b). Next, we examined the attachment of actively translating ribosomes on presynaptic ER by staining mCherry-KDEL expressing neurons against Y10B and eEF2 as depicted in Fig. 6c. Similarly, 10 sec and 1 min stimulations caused an increase in colocalization of Y10B/eEF2 co-clusters with presynaptic ER indicating that ribosomes in the elongation stage of translation attach to the presynaptic ER (Fig. 6d). To con rm these results in vivo, we stimulated NMJs of P5 WT mice with BDNF and performed IHC using an ER marker, calreticulin, and Y10B as ribosomal marker (Fig. 6e). Consistent with in vitro data, we could detect ribosomes which associated with ER within the presynaptic area labeled by synaptophysin. These ndings imply that RER exists both in the growth cone of developing neurons and in the nerve terminals in vivo and thus support a role for presynaptic ER in processing and surface delivery of axonally synthetized membrane and secretory proteins.
Altered regulation of dynamics of local translation and aberrant presynaptic ER dynamics are involved in axon pathology in spinal muscular atrophy As demonstrated by previous studies [21,42,43], aberrant axonal mRNA localization and impaired local translation are important contributors to disease manifestation and clinical pathology in SMA and ALS.
Therefore, we sought to investigate the importance of our observations on dynamic regulation of local translation and contribution of the presynaptic ER in context of SMA. We used a mouse model that resembles severe type I SMA in humans. First, we measured the time course of BDNF-induced ribosome activation using Y10B immunostaining in Smn-de cient motoneurons (Fig. 7a). Interestingly, even 10 min stimulation did not elicit any alteration in ribosome distribution in Smn-de cient growth cones as the uorescence intensity for Y10B did not increase, in contrast to WT control motoneurons (Fig. 7b). In line with that, we did not detect increased immunoreactivity of TrkB as a response to BDNF stimulation, indicating that the mechanisms that underly these changes in wildtype cells are defective in Smnde cient motoneurons (Fig. 7c, d). Total levels of TrkB were similar in control and Smn-de cient neurons under basal (no BDNF stimulation) condition in both growth cones and soma (Fig. 7e). BDNF-stimulated Smn-de cient motoneurons also failed to show alterations in puromycin immuno uorescence indicating that lack of Smn protein could be responsible for disturbed kinetics of local translation and impaired dynamic regulation of growth cone proteome (Fig. 7c, f). Next, we measured dynamic movements of presynaptic ER in Smn-de cient motoneurons and found that while dynamic movements are not disturbed in the growth cone core, the movements are markedly reduced in lopodia (Fig. 8a, b, Video 8 -9). Consistent with these defects, SIM image analyses of ER colocalization with F-actin and a-Tubulin in growth cones of Smn-de cient motoneurons revealed that the altered distribution of actin laments in lopodia correlates with disturbed ER entry into lopodia, which could explain the mechanism underlying defective ER dynamic remodeling in growth cone lopodia (Fig. 8c).
Finally, we investigated the dynamics of ribosome activation and their association to presynaptic ER upon BDNF induction by SIM. We found that in Smn-de cient motoneurons, BDNF stimulation had no or only minor effects below detection limit on the assembly of 80S ribosomes (Fig. 9a). As shown in Fig. 9b, we did not detect elevated levels of RPL24/RPS6 co-clusters in BDNF-stimulated Smn-de cient motoneurons compared to unstimulated Smn-de cient neurons. Smn+/+; SMN2tgtg motoneurons showed elevated levels of RPL24/RPS6 co-clusters upon BDNF induction compared to unstimulated Smn+/+; SMN2tgtg neurons ( Fig. 9b and Fig. 4b). Likewise, the number of RPL24/RPS6 co-clusters associated with ER was not altered in growth cones of BDNF-stimulated Smn-de cient motoneurons compared to unstimulated Smn-de cient neurons (Fig. 9a, c). To con rm these results, we performed immunostaining with Y10B and eEF2 in mCherry-ER expressing motoneurons to evaluate ER and ribosomes and their association in Smn-de cient motoneurons (Fig. 9d). Compared to unstimulated Smnde cient neurons, BDNF-stimulated Smn-de cient neurons did not show increased levels of Y10B/eEF2 co-clusters (Fig. 9e). In contrast, BDNF-stimulated Smn+/+; SMN2tgtg motoneurons displayed signi cant increase in RPL24/RPS6 co-clusters that associated with ER compared to unstimulated Smn+/+; SMN2tgtg neurons (Fig. 9b), as previously illustrated in Fig. 4b. Furthermore, no enhanced levels of Y10B/eEF2 co-clusters associated with ER that are prominent in Smn+/+; SMN2tgtg neurons (Fig. 6d, Fig.  9f) were detected in Smn-de cient neurons after a 10 sec or 1 min BDNF pulse (Fig. 9f). Thus, Smn de ciency leads to impaired BDNF/TrkB responses both in ER dynamics and in ribosome assembly, which in turn could cause disturbed translation initiation and ER attachment of ribosomes in the presynaptic compartment.

Discussion
In neurons, ER forms a highly dynamic network in dendrites and axons. The regulation of axonal ER dynamic movements as well as its function in development and maintenance of synapses have gained emerging interest [44][45][46]. In this study, we developed a technique to visualize and quantify the dynamic movements of ER in presynaptic terminals of cultured motoneurons and show that motoneurons harbor a rough ER in the axonal growth cone which extends into lopodia. Its integrity and dynamic remodeling are regulated mainly by the actin cytoskeleton, and this dynamic remodeling is markedly disturbed in motoneurons from a mouse model of spinal muscular atrophy.
As apposite to microtubules [24], the interaction of ER with actin cytoskeleton especially in the axon and presynaptic area has not been investigated yet in much detail. In somatodendritic compartments, kinesin and dynein-based ER sliding along microtubules as well as interaction between STIM1 on the ER and EB1 on the microtubule plus end have shown to mediate ER translocation [3,47]. However, studies with cultured cells have revealed that rst, nocodazole treatment does not retract ER from periphery immediately [48], second, ER tubules can form in the absence of microtubules [49][50][51] and third, disruption of actin laments in hippocampal neurons affects Ca 2+ release from ER in soma [52]. These observations indicate that ER dynamic movements might not depend only on microtubules but also on actin cytoskeleton. Indeed, the rst indirect evidence for the role of actin in ER translocation into dendrites comes from a study by Wagner et al., showing that myosin-Va, which is an actin-based motor, mediates ER movement in dendritic spines of Purkinje neurons [7]. Axonal growth cones consist of highly dynamic membrane protrusions; " lopodia and lamellipodia", and a more stable central region; "growth cone core" [53]. All these domains are transient and undergo constant growth and retraction in shape and structure [54]. In contrast to microtubule-rich core, actin is predominant in lopodia and lamellipodia. Therefore, dynamics of lopodia exceed those in the core [55]. In line with that, our time-lapse recordings show that the dynamics of ER differs in growth cone center and periphery. In the core of growth cones, the velocity of ER dynamic movements is much lower than that in actin-rich lopodia. Importantly, we found that ER movements in the core are regulated by both actin and microtubule cytoskeleton. Previous studies have demonstrated that F-actin and microtubules interplay during neuronal polarization and development [56][57][58]. This crosstalk is mediated by macromolecules such as Drebrin E, which physically links F-actin and microtubule plus end binding protein or other regulatory proteins such as Rho GTPases, MAP2 and tau [56,58]. Thus, such mechanisms could be implicated in the coordinated actin/microtubule regulation of ER dynamics in the growth cone core. Strikingly, actin depolymerization destroys ER dynamic movements in lopodia of 80% of imaged cells. In addition, two-color live cell imaging data revealed that ER moves along actin in lopodia and F-actin withdrawal results in nearly 99% retraction of ER from lopodia.
These data provide the rst direct evidence of an actin-dependent ER remodeling and integrity in axon terminals of neurons. ER entry into lopodia suggests a membrane contact site between the ER and the plasma membrane [1]. This physical tethering could be essential for the surface delivery of newly synthesized receptors, membrane proteins or lipids. Further studies are required to investigate whether myosin-Va might be also involved in such actin-dependent ER dynamic regulation in axons. Regulation of dynamic remodeling of axonal ER recently became attentive through studies identifying mutations in ER regulatory proteins that associate with human neurodegenerative diseases including hereditary spastic paraplegia, Alzheimer's disease (AD) and ALS. Mutations in ER-shaping proteins such as Spastin, ATL1, RTN2 or REEP1 that couple tubular ER network with microtubule dynamics cause hereditary spastic paraplegia [59,60]. AD patients carry mutations in Rab10, another protein involved in microtubule and ER coupling [61,62]. Similarly, mutations in vesicle-tra cking protein A and B (VAPA and VAPB) cause lateonset SMA and ALS [8,63]. VAPA and VAPB proteins reside in ER and facilitate anchoring of membranous organelles to the ER [8] and regulate tubular ER morphogenesis and dynamics [46]. Such mutations that impair ER dynamic remodeling do not only disturb ER functions but also impact mitochondria and endosomes through ER MCSs [10]. Our data unravel that loss of SMN protein associates with impaired dynamic remodeling of ER in actin-rich growth cone lopodia in cultured motoneurons. Interestingly, we did not observe defective ER dynamics in the growth cone core of Smnde cient motoneurons, indicating that impaired actin and not microtubule cytoskeleton underlies the defective ER dynamics in the growth cone of Smn-de cient motoneurons. In line with that, SIM data show altered colocalization of ER with F-actin in growth cone lopodia of Smn-de cient motoneurons, again con rming disturbed actin-dependent ER extension into lopodia. Actin cytoskeleton appears to play a central role in neurodegeneration through involvement in key axonal functions and synapse maintenance [64]. Indeed, there is emerging evidence for a defective actin cytoskeleton in axons and presynaptic terminals of SMA patients as well as animal models. Loss of SMN causes impaired mRNA localization and local translation of actin isoforms Actα and Actβ resulting in disturbed dynamic of F-actin polymerization in the growth cone [20]. Pro lin-2 is a neuronal speci c regulator of F-actin polymerization which has been shown to directly interact with SMN protein [65,66]. SMN interaction with Pro lin-2 reduces its sequestering activity on F-actin, thereby promoting actin polymerization, whereas SMN release allows Pro lin-2 binding to actin monomers, leading to reduced actin polymerization [67]. Plastin-3, an actin bundling protein that also regulates actin turnover, is a protective modi er of SMA and its overexpression has been shown to ameliorate defects in axon growth [68], NMJ maturation and motor functions in SMA animal models [69]. Notably, SMA mouse models exhibit elevated levels of microtubules and neuro laments in the presynaptic terminals [70]. Therefore, a compensatory mechanism involving microtubules could be considered for the observed intact dynamic remodeling of ER in the growth cone core of Smn-de cient motoneurons.
Subcellular control of axon growth, synapse differentiation and plasticity depend on regulatory mechanisms that ensure processing of local information and subsequent responses on a time scale of seconds. mRNA localization and local protein synthesis are conserved mechanisms that modify axonal proteome at a fast temporal and spatial scale, thereby maintaining plasticity capacity of axonal synapses [27]. In addition to cytoplasmic proteins, mRNAs encoding secreted and membrane proteins have been identi ed in axonal transcriptomes [21,36,71]. On-site synthesis and glycosylation of such membrane proteins would require additional RER and Golgi machinery in the distal axon. However, previous studies have suggested that in axons ER is considered to be devoid of ribosomes and thus no direct evidence of its potential functions related to local protein synthesis has been reported yet [11,16,17]. Interestingly, Merianda et al., 2009 reported glycosylation patterns of transmembrane proteins being locally translated in axons of sensory neurons [18]. We used a BDNF pulse approach combined with super-resolution uorescence imaging by SIM and demonstrated that stimulation of motoneurons triggers the assembly of 80S ribosomes and initiates the translation in axon terminals within 10 sec.
Ribosomes in the elongation phase of translation attach to the presynaptic ER and locally synthetized proteins become detectable at 1 min poststimulation. Contrary to axons, the response of soma to BDNF/TrkB activation was considerably slower and happened in longer time scales of several minutes.
These distinctive response rates could be due to different intracellular levels of second messengers such as cAMP [72]. It remains unknown whether similar or distinct signaling pathways downstream to BDNF/TrkB, such as mTOR, MAPK and PI3K pathways, are involved in activation of ribosomes and translation initiation in the growth cone and its counterparts. Thus, our data uncover the rate of dynamic local translation at presynaptic terminals of cultured motoneurons and identify a rough ER in axon terminals that participates in post-translational processing of locally synthetized proteins in response to extracellular stimuli. This fast rate of translation indicates that ribosomes only transiently associate with presynaptic ER in response to stimuli and explains why previous studies failed to detect RER in axons using unstimulated tissues.
Translational defects and disturbed mRNA recruitment onto polysomes previously reported in SMA mouse models implicate a master function of SMN in translational regulation [20,73,74]. Indeed, SMN is not only involved in the assembly of snRNP particles but also associates directly with actively translating polysomes, as shown in cultured cells, spinal cord and brain tissues [75,76]. In our study, Smn-de cient motoneurons failed to respond to BDNF stimuli by eliciting the fast assembly of 80S ribosomes, resulted in delayed translation initiation and perturbed association of ribosomes with the presynaptic ER after BDNF stimulation. In motoneurons, BDNF modulates diverse neuronal functions such as survival, growth and differentiation [77][78][79]. Strikingly, we did not detect any reduction in TrkB levels in both growth cone and soma of Smn-de cient motoneurons. In WT motoneurons, TrkB levels elevate in the growth cone within 1 min after BDNF stimulation, which is indicative of a rapid increase of cell surface translocation of TrkB receptors. This so called "self-enhancement" of BDNF-TrkB signaling seems to be critical for synaptic plasticity and maintenance [80]. Notably, upon BDNF pulse the levels of detectable TrkB immunoreactivity did not increase in Smn-de cient motoneurons even at 10 min poststimulation. Interestingly, CytoD treatment has been shown to inhibit TrkB translocation to the cell surface [81]. Our data also show that in growth cones of WT neurons, CytoD-treatment abolishes the ribosomal response to BDNF pulse, indicating that an actin-based motor moves ribosomes and possibly tethers them to the ER. Thus, an impaired actin cytoskeleton affecting cell surface TrkB translocation, ER dynamic remodeling and ribosome/ER anchoring appears to underlie the mechanism of disturbed BDNF responsivity and synaptic defects in SMA.

Conclusions
In summary, this work identi es a novel function for axonal ER in regulation of stimulus-induced local translation and discloses the mechanism of dynamic regulation of presynaptic ER by an actinmicrotubule crosstalk. Our study on Smn-de cient motoneurons uncovers molecular determinants involved in impaired sensing of changes in the extracellular environment leading to a delayed response and explains why SMN de ciency makes motoneurons vulnerable to axon and synapse differentiation aberrations.

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication
All authors read and approved the nal manuscript.

Availability of data and materials
The primary datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare no competing nancial interests.      Ribosomes attach to the presynaptic ER and promote local translation in response to BDNF stimulation.
Graphs are shown in scatter dot plot with mean ± SEM. Statistical analyses: One-way ANOVA with Dunn's post-test.  Fig. 6d and included here again to allow direct comparison. All data are normalized to no BDNF group of the corresponding genotype. Graphs are shown in scatter dot plot with mean ± SEM. Statistical analyses: One-way ANOVA with Dunn's post-test.