Optogenetic control of mRNA localization and translation in live cells

Despite efforts to visualize the spatio–temporal dynamics of single messenger RNAs, the ability to precisely control their function has lagged. This study presents an optogenetic approach for manipulating the localization and translation of specific mRNAs by trapping them in clusters. This clustering greatly amplified reporter signals, enabling endogenous RNA–protein interactions to be clearly visualized in single cells. Functionally, this sequestration reduced the ability of mRNAs to access ribosomes, markedly attenuating protein synthesis. A spatio–temporally resolved analysis indicated that sequestration of endogenous β-actin mRNA attenuated cell motility through the regulation of focal-adhesion dynamics. These results suggest a mechanism highlighting the indispensable role of newly synthesized β-actin protein for efficient cell migration. This platform may be broadly applicable for use in investigating the spatio–temporal activities of specific mRNAs in various biological processes. Kim et al. develop an optogenetic visualization approach that can rapidly and reversibly trap messenger RNA molecules in protein clusters, thereby restricting their access to ribosomes and dampening translation efficiency.

T ranslation is the process of decoding genetic information from messenger RNA to protein. The spatio-temporal regulation of mRNA translation is critical for asymmetric cellular structures and functions involved in many biological processes 1 . The spatio-temporal dynamics of mRNA have been assessed by techniques visualizing mRNA in live cells at the single-molecule level [1][2][3] . Nevertheless, the functional links between mRNA dynamics and cell behaviour remain unclear. Genome-wide analyses have revealed that various mRNAs are concurrently regulated in given cellular contexts 4 , making visualization methods insufficient to determine the mRNAs responsible for specific aspects of a cellular event.
The causal relationships between the translation of specific mRNAs and various biological processes have been directly investigated by manipulating mRNA translation using chemical compounds, oligonucleotide-based approaches and RNA-targeting endonucleases 2,5-7 . However, they cannot determine the roles of the spatial distribution of mRNA in translation, especially at the subcellular level. In addition, technical difficulties using chemical inhibitors, including the need for washout and re-addition, and the long lag times required for manifestation of the effects of oligonucleotides prevent the achievement of translation control with high temporal resolution. Moreover, chemical compounds are not specific enough to allow the study of the contributions of individual mRNAs to cellular functions.
Optogenetic approaches, which can achieve space-and timeresolved control of specific molecules, may circumvent these drawbacks. Optogenetic tools have been found to modulate a wide range of transcriptional and post-translational signalling events 8 . Although a previous module has been shown to activate translation by tethering eukaryotic initiation factor to the 5ʹ region of target mRNAs 9 , this system was applicable to exogenously designed mRNAs and its ability to control cell behaviours has not been validated. Furthermore, modules that inactivate mRNA translation are still in demand to examine the necessity of particular mRNA translation for certain biological functions.
Here, we present an optogenetic method-called mRNAlight-activated reversible inactivation by assembled trap (mRNA-LARIAT)-that can directly perturb the localization and translation of specific mRNAs in live cells. This LARIAT system 10 was combined with RNA-binding protein (RBP)-based mRNA visualization modules to trap specific mRNAs in protein clusters using light. This sequestration of mRNAs restricted their accessibility to ribosomes and markedly reduced translation efficiency. This method was used to manipulate both exogenous and endogenous unmodified mRNAs. Spatio-temporal analysis of endogenous β-actin mRNA sequestration suggested that newly synthesized β-actin protein plays an essential role in cell migration. This technique can be used to study the causal relationships between the translation and localization of particular mRNAs and various cellular functions and diseases.

Results
Design of mRNA-LARIAT. The mRNA-LARIAT system was initially designed using an MS2-based module, the most widely used system for mRNA visualization 1 . The mRNA-LARIAT system consists of a cryptochrome 2-fused anti-green fluorescent protein (GFP) nanobody (V H H(GFP)-CRY2), a cryptochrome-interacting basic-helix-loop-helix 1 (CIB1)-fused multimeric protein (MP; CIB1-MP), a GFP-labelled MS2 coat protein (MCP-GFP) and an MS2-binding site (MBS)-tagged mRNA (target-MBS). The GFP nanobody was employed to adapt this system for use with GFPbased RNA visualization modules. Similar to LARIAT 10 , blue lightinduced CRY2-CIB1 binding triggered interactions among MPs 11 to generate protein 'clusters' . During cluster formation, MCP-GFP and MBS-mRNA complexes are trapped in V H H(GFP)-loaded clusters (Fig. 1a). We removed the nuclear localization signal (NLS) from MCP-GFP for efficient trapping of cytosolic mRNAs.
The mRNA-LARIAT system was tested by co-expressing V H H(GFP)-LARIAT (mCherry-labelled CRY2-V H H(GFP) and CIB1-MP), MCP-GFP and MBS-tagged infrared fluorescent protein (iRFP) 12 in HeLa cells. The fluorescence signal seemed evenly dispersed before blue light stimulation (Fig. 1b). Light stimulation induced clustering of the V H H(GFP)-LARIAT signal co-localized with the MCP-GFP signal, and the process was reversed by the withdrawal of light (Supplementary Video 1). The fluorescence intensity and localization of iRFP did not change, indicating that expression of the LARIAT components did not reduce protein synthesis and clusters did not trap the translated protein (Fig. 1b,c and Supplementary Video 2). Pulsatile light illumination repeatedly induced the formation of clusters with assembly and disassembly kinetics (T 1/2 ) of 41.4 ± 2.4 s and 534 ± 37.8 s, respectively (Fig. 1d). Cluster formation was quantitatively controlled by variations in light density and pulse number (Fig. 1e,f).

Trapping of specific mRNAs into the cluster by mRNA-LARIAT.
The presence of MBS-labelled mRNAs in the clusters was verified by fluorescent in situ hybridization (FISH) of iRFP-MBS. A FISH probe against MBS showed significant co-localization of target mRNAs with clusters only when MCP-GFP was present (Fig. 2a,b). In contrast, FISH probes against non-target mRNA-such as GAPDH, tubulin 1α, Arp2 and Arp3-did not elicit any co-localization (Fig. 2c). Quantitative analysis indicated that the amount of the target transcript was significantly higher in clusters with MCP than those without MCP (Fig. 2d). The fraction of sequestered mRNA in individual clusters depended on cluster size (r = 0.389) and varied from 0.01 to 18% (Fig. 2e). The conventional MS2 module containing NLS in MCP-GFP was also compatible with mRNA-LARIAT (Extended Data Fig. 1a-d). Moreover, light illumination of a subcellular region triggered local clustering without affecting the unilluminated areas (Extended Data Fig. 1e).
Because fluorescence amplification of mRNA-LARIAT through clustering enabled clear detection of mRNA-RBP (MBS-MCP) complexes, the ability of mRNA-LARIAT to visualize the interactions of mRNA with endogenous proteins was tested (Extended Data Fig. 2). Staining of the mRNA-loaded clusters using antibodies against small (rpS6) and large (RPL10A) ribosomal proteins, and with FISH probes targeting 18S or 28S ribosomal RNA revealed that these proteins and RNAs localized in clusters containing the target transcripts (Fig. 2f-h and Extended Data Fig. 3a,b). To test whether the trapped mRNAs could dynamically interact with ribosomes, cells were either serum-starved or treated with puromycin to dissociate the mRNA from the ribosomes (Extended Data Fig. 2). The cells were subsequently illuminated with light, along with serum stimulation 13 or washout of puromycin, which triggered the reassociation of the mRNAs with ribosomes (Extended Data Fig.  3a,b). The application of light before serum addition or puromycin removal did not induce recruitment of ribosomes to the clusters, although the target transcripts resided in the clusters. The treatment of cells with puromycin while trapping translational (ribosomebound) mRNA showed that ribosomal components were present in the clusters, indicating that the release of ribosomes associated with the trapped mRNAs outside the clusters was restricted (Extended Data Fig. 3f,g). In addition, the molecular exchange between the inside and outside of the clusters was limited (Fig. 2i), indicating that there is minimal replacement of the sequestered components by molecules outside the clusters. These findings indicate that light-mediated clusters markedly restrict the diffusion of molecules  associated with the target mRNAs, probably because of steric hindrance by the large clusters of MPs 10 , attenuating the dynamic interactions between ribosomes and mRNAs.

Inhibition of translation by light-induced mRNA sequestration.
These findings suggested that mRNA-LARIAT could influence mRNA translation. Before testing that, we examined whether tagging MBS to mCherry mRNA and co-expression of MCP-GFP had any background effect on protein synthesis. Compared with cells expressing mCherry without the MBS tag, mCherry expression was lower in cells expressing mCherry with an MBS tag, with or without the co-expression of MCP-GFP. The correlation between mCherry intensity and the number of mCherry-encoding mRNAs was also reduced, indicating that the attachment of MBS to mCherry mRNA and co-expression of MCP inhibited translation (Extended Data Fig. 4). The effect of mRNA-LARIAT was therefore assessed using only MBS-tagged mRNA and co-expressed MCP-GFP. The effect of mRNA sequestration on translation was tested by introducing mRNA-LARIAT components containing a tetracycline-responsive promoter element (TRE)-driven MBS-labelled mCherry plasmid into the NIH3T3 Tet-On stable cell line (Fig. 3a). Treatment with doxycycline and light markedly reduced the level of mCherry protein (approximately 90%; Fig. 3b,c) but not mRNA (Extended Data Fig. 5a). The reduction in mCherry expression was   dependent on the doxycycline concentration, with higher doxycycline concentrations increasing the expression of the target transcripts (Extended Data Fig. 5b) and decreasing the ratio of trapped mRNA (Fig. 3d), indicating that the efficiency of translation inhibition was dependent on the ratio of target mRNAs to clustering components. In addition, the abundance of target mRNAs affected the size but not the number of clusters (Extended Data Fig. 5c,d). Treatment with both light and doxycycline dramatically attenuated the translation of the target (TRE-iRFP682-MBS) but not the nontarget (TRE-mCherry) transcript (Fig. 3e), reflecting the specificity of mRNA-LARIAT. Light alone had no effect on the global translation processes, as shown by measurements of the phosphorylation of the translation-initiating factor eIF2α 14 (Extended Data Fig. 6a,b).
To determine whether the clusters are associated with cellular structures, such as stress granules, which could translationally regulate non-target transcripts 15 , the dynamics of stress granules were monitored using FusionRed-tagged G3BP1. No clusters or stress granules were observed before light stimulation. Light illumination induced cluster formation but had no effect on the stress-granule marker (Extended Data Fig. 6c). Treatment with NaAsO 2 generated stress granules-which were disassembled by the addition of cycloheximide (CHX) 15 -without any perturbation of light-induced clusters. Interestingly, a partial overlap of stress granules and clusters was observed about 20 min after NaAsO 2 treatment. However, a time-course analysis revealed that the stress granules formed initially at locations without clusters (Extended Data Fig. 6d). These findings indicate that light-induced clusters differ molecularly from stress granules and that translational inhibition by mRNA-LARIAT is highly specific to target transcripts.
Next, mRNA-LARIAT was applied to primary rat hippocampal neurons. The 3ʹ untranslated region (3ʹ UTR) of Ca 2+ /calmodulindependent kinase IIα (CaMKIIα) mRNA contains sufficient information for dendritic localization and brain-derived neurotrophic factor (BDNF)-dependent translation 16 . A diRFP-3ʹCaMKIIα-MBS expression vector containing the CaMKIIα 3ʹ UTR inserted between an iRFP reporter and MBS was synthesized (Fig. 3f). An unstable iRFP fusion protein (diRFP) was generated by conjugating iRFP with the destabilizing domain of ornithine decarboxylase to reduce background fluorescence. Following light stimulation, clusters formed throughout the cytoplasm of hippocampal neurons co-expressing mRNA-LARIAT components plus diRFP-3ʹCaMKIIα-MBS ( Fig. 3g and Supplementary Video 3). Treatment with BDNF increased the iRFP fluorescence only when the target mRNAs had not been recruited into clusters. However, the trapping of target mRNAs into clusters significantly reduced the iRFP signal, even in the presence of BDNF. This reduction was comparable to that induced by treatment with the translation inhibitor anisomycin (Fig. 3h,i). These results demonstrate that the light-induced sequestration of target mRNAs into clusters effectively inhibits translation.

Sequestration of non-engineered endogenous mRNAs using the RCas9 module.
To expand the versatility of our platform, we used the RNA-targeting Cas9 (RCas9) system to manipulate endogenous mRNAs 3 . This system contains a CRISPR-associated Cas9 protein lacking nuclease activity (dCas9), a single guide RNA (sgRNA) and a programmed PAM-presenting oligonucleotide (PAMmer; Fig. 4a). Light illumination efficiently generated clusters without affecting cell proliferation and viability ( Fig. 4b-d and Supplementary Videos 4,5). The trapping of β-actin and GAPDH mRNA in these clusters was confirmed by FISH (Fig. 4e,f and Extended Data Fig.  7a,b). Omission of any single component significantly reduced the efficiency of target-mRNA sequestration in clusters. However, a modest, but significant, recruitment of target mRNAs into clusters was observed, even in the absence of PAMmer, indicating that the sgRNA is critical for target binding and that the PAMmer increases the targeting efficiency 3 . Staining clusters of β-actin mRNA with FISH probes against non-target mRNAs showed no detectable signal (Extended Data Fig. 7c), and quantitative analyses indicated that endogenous target mRNAs were trapped in clusters (Fig. 4g). Moreover, RCas9-based mRNA-LARIAT exhibited a positive correlation between cluster size and the relative quantity of trapped mRNA (Fig. 4h). Other than β-actin and GAPDH mRNA, relatively low-abundant target transcripts such as Arp2 and Arp3 mRNA 17 were also efficiently trapped in clusters (Extended Data Fig. 7d). The conventional RCas9 module with an NLS on dCas9-GFP was able to trap target transcripts in clusters (Extended Data Fig. 8).
The RCas9-based mRNA-LARIAT was functionally validated using a TRE-driven mCherry reporter gene with a λ bacteriophage sequence tag at its 3ʹ UTR 3 . Similar to the results for MBS tag and MCP co-expression, the λ-tagged gene expressed lower levels of mCherry protein than the untagged gene (Extended Data Fig. 4). The effect of light illumination on mRNA-LARIAT was therefore assessed using only mCherry-λ. Light illumination of doxycyclinetreated cells containing sequestered reporter mRNA yielded lower levels of mCherry protein than cells left in the dark (Fig. 4i,j).
Testing of the ability of RCas9-based mRNA-LARIAT to visualize endogenous mRNA (β-actin mRNA)-protein complexes demonstrated that ribosomal components were trapped in β-actin mRNA clusters, whereas serum starvation before light illumination resulted in clusters lacking ribosomes (Extended Data Fig. 9). In addition to the ribosomal components, β-actin mRNA was trapped together with insulin-like growth factor 2 mRNA-binding protein 1 (IMP-1)-a transporter of β-actin mRNA 18 (Fig. 4k,l)suggesting that this technique may be able to disrupt β-actin mRNA localization as well as translation. In contrast, fragile X mental retardation protein (FMRP), which does not bind β-actin mRNA 19 , did not localize in the clusters. These findings indicate that RCas9-based mRNA-LARIAT efficiently targets endogenous mRNAs and permits the visualization of endogenous mRNAmolecular complexes. Light-induced sequestration of endogenous β-actin mRNA during cell migration. Fibroblast migration has been extensively used to understand the roles of asymmetric distribution and local translation of β-actin mRNA 1,20,21 . The mRNA-LARIAT system was therefore utilized to sequester endogenous β-actin mRNA in fibroblasts.
The β-actin mRNA appeared near the leading edge both in the dark and in the absence of V H H(GFP) with light illumination. Treatment with the translation inhibitor CHX did not alter the localization of the β-actin mRNA, which is consistent with previous findings 22 .
In contrast, trapping of the β-actin mRNAs markedly shifted their localization to the perinuclear region, as shown by the polarization and dispersion indexes 23 ( Fig. 5a and Extended Data Fig. 10). Cell migration during the sequestration of the β-actin mRNA was monitored by light illumination for 6 h. Persistent trapping of the β-actin mRNA significantly reduced the total path length compared with the negative controls but did not alter directionality (Fig. 5b-d and Supplementary Video 6). This defect in cell migration manifested within a relatively short time period (approximately 20 min; Fig. 5e). Additional treatment with CHX during the sequestration of the β-actin mRNA did not further reduce cell migration (Fig. 5f). Considering the non-specific inhibition of mRNA translation by CHX, these results suggest that β-actin translation plays a prominent role in cell migration. When the cells that had been illuminated were returned to the dark, the migration rate returned to unstimulated levels accompanied by the disassembly of clusters (Fig. 5g). To assess whether the defect in migration can be rescued by exogenous expression of β-actin, mCherry-labelled β-actin with a zipcode at the 3ʹ UTR was expressed so that mCherry-β-actin mRNA would have similar dynamics to endogenous β-actin mRNA but would not be targeted by mRNA-LARIAT. The mCherry-β-actin mRNAs were incorporated into F-actin fibres (Fig. 5h) and restored cell migration even when endogenous β-actin mRNA was targeted (Fig. 5i). Clusters targeting β-actin mRNA did not co-localize with F-actin structures, implying that the clusters are not physically associated with the actin cytoskeleton. These results demonstrate the specificity of our system and suggest the importance of newly synthesized β-actin for efficient cell migration.
The leading edges of cells were illuminated to examine the effect of local perturbation of the translation of β-actin mRNA on cell migration. In the absence of the sgRNA and PAMmer components of RCas9, light illumination did not affect persistent membrane protrusion (Fig. 6a,b). The iRFP-Lifeact signal did not aggregate following cluster induction, which means that the clusters per se did not affect the pre-existing F-actin structures. In contrast, locally induced clusters targeting β-actin mRNA disrupted constant membrane protrusion (Fig. 6c,d and Supplementary Videos 7,8). The unilluminated regions of the membrane showed new protrusions, whereas selective illumination of those sites inhibited further protrusion. Similarly, illumination of whole cells resulted in a rapid turnover of membrane protrusions (Supplementary Video 9).
Given that the light-induced inhibition of β-actin synthesis disrupted cell migration but not the protrusion activity, it was unclear whether the newly synthesized and previously synthesized β-actin molecules play distinct roles in actin nucleation. Because actin nucleation during cell migration involves Arp2/3 complexes and formin proteins 24 , cells were treated with CK-666 or SMIFH2 to block Arp2/3-or formin-driven actin polymerization, respectively, while the β-actin mRNA was sequestered. Measurements of edge velocities and changes in cell area revealed that CK-666 effectively blocked membrane protrusion, whereas SMIFH2 did not (Fig. 6e,f and Supplementary Video 9), suggesting that pre-existing β-actin proteins utilized by the Arp2/3-mediated pathway are sufficient for actin polymerization required for membrane protrusion, a process negligibly affected by mRNA-LARIAT. The importance of forminmediated actin polymerization for cell migration 24 indicated that newly synthesized β-actin proteins may be preferentially utilized by formin-mediated pathways to generate cell movement.

Newly synthesized β-actin is indispensable to cell migration.
Despite the presence of vast quantities of pre-existing β-actin molecules, cell migration may require newly synthesized β-actin to stabilize focal-adhesion complexes 21 . The effects of β-actin mRNA manipulation on adhesion dynamics were evaluated by expressing mCherry-labelled paxillin (paxillin-mCherry) with mRNA-LARIAT components. The number of stable focal adhesions were considerably reduced after sequestration of the β-actin mRNA or CHX treatment, as evidenced by decreased numbers of large and thick paxillin-mCherry foci compared with those generated in the dark (Fig. 7a,b). Nevertheless, nascent focal adhesions (indicated by small paxillin-mCherry foci) and membrane protrusions were generated under all conditions. Because the migration defect caused by β-actin mRNA sequestration became evident within a relatively short time (<20 min; Fig. 5e), we assessed whether a change in adhesion dynamics was involved in this process. The trapping of β-actin mRNA elicited a rapid turnover of protrusions and the generated clusters showed a significant spatial overlap with adhesion signals, whereas trapping GAPDH mRNA led to stable membrane protrusions and a continuous increase in adhesion signals without any spatial overlap with clusters (Fig. 7c,d). Analysis of the temporal cross-correlation between clusters and focal-adhesion movement showed that clusters targeting β-actin mRNA were more strongly correlated with focal-adhesion movement than clusters targeting GAPDH mRNA. The movement of β-actin mRNA-targeting clusters was slightly slower than that produced by paxillin (Fig. 7e).  These results indicate that inhibition of the translation of β-actin mRNA impedes the maturation but not the generation of focal adhesions, thereby reducing persistent membrane protrusions and migration.

Discussion
We describe the development of the mRNA-LARIAT technique, which allows optogenetic control over the localization and translation of specific exogenous and endogenous mRNAs in live cells. The sequestration of mRNA inhibits translation by limiting the dynamic interactions of the mRNA with the ribosomes. We verified the functionality of mRNA-LARIAT by showing that light reduced protein synthesis in fibroblasts and hippocampal neurons to levels comparable to those induced by a translation inhibitor. The mRNA-LARIAT system could sensitively identify endogenous RNA-binding complexes at the single-cell level, enabling our method to better address RNA-protein interactions and the heterogeneity of mRNA granules as well as unidentified cis-acting elements in particular mRNAs.
Green-fluorescent-protein nanobody-based mRNA-LARIAT can be adapted for use with previously developed RNA-visualization modules. This platform would be applicable to transgenic models that stably express tagged mRNAs and fluorescent protein-labelled RBPs 25 . Direct conjugation of CRY2 to RBP could reduce the number of components and robustly control mRNA translation.
Even in a transient expression system that results in heterogeneous expression of individual components, mRNA-LARIAT suppressed target translation by up to 90%. However, the efficiency was dependent on the ratio of target transcripts to clustering components. For example, an increase in the quantity of target transcript results in a significant proportion of mRNAs that are not trapped. The use of cell lines stably expressing mRNA-LARIAT components may minimize the heterogeneity of outcomes and increase robustness. Previous knowledge of the half-life of the target proteins may enable optimization of light illumination to effectively control the total amount of protein.
Tagging with MBS and co-expression of RBPs have been reported to inhibit mRNA degradation, especially for short-lived mRNAs 26 . Other studies reported that the MS2 system did not affect the levels of mRNA and protein 25,27 , indicating that the outcomes were dependent on the experimental conditions and target mRNAs. We found that MBS tagging of mCherry and co-expression of MCP-GFP somewhat inhibited protein synthesis (Extended Data Fig. 4), suggesting that care should be taken to assess the background effect of the mRNA-tagging method before the application of mRNA-LARIAT.
Optogenetic aggregation of intrinsically disordered protein regions was recently reported as sufficient to drive liquid-liquid phase separation 8,28 . Although mRNA-LARIAT also induced the accumulation of RNA-protein complexes, the clusters were independent of membrane-less organelles such as stress granules, indicating that the clusters generated by mRNA-LARIAT differed molecularly from stress granules. Phase separation may be driven by our method if a particular target transcript is associated with RBPs rich in intrinsically disordered regions.
The application of mRNA-LARIAT to fibroblasts showed that newly synthesized β-actin is critical for cell-migration efficiency. The light-induced inhibition of β-actin translation effectively and reversibly attenuated cell motility. Restoration of the migration defect by expression of non-target β-actin supports the specificity of mRNA-LARIAT. Despite large quantities of pre-existing long-lived (half-life of 48-72 h) β-actin protein, the sequestration of β-actin mRNA significantly perturbed cell migration within 20 min, suggesting that a minute fraction of newly synthesized β-actin can have a profound effect on this process. The results from orthogonal control using light and chemical compounds indicated that newly synthesized β-actin proteins are preferentially utilized by the formin-dependent pathway involved in the maturation of focal adhesion rather than by Arp2/3-driven membrane protrusion. The observed defect in migration was probably not due to the disruption of pre-existing F-actin structures by clusters or co-trapping of focaladhesion molecules. These results suggest a model that qualitatively distinguishes between newly synthesized and pre-existing β-actin proteins, with the former required for active cell migration (Fig. 7f). However, we cannot rule out possible secondary effects induced by co-trapping of proteins that bind to β-actin mRNA.
The mRNA-LARIAT system is generally adaptable and can spatio-temporally manipulate the translation of target transcripts to determine their physiological roles. The mRNA-LARIAT system should be applied to in vivo models and targeting of various endogenous mRNAs. The use of MCP-MBS transgenic mice 25 , design of various PAMmer sequences 7 and engineering of programmable RBPs 29 may further expand the versatility of this technique.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41556-020-0468-1.
Cell culture and transfection. HeLa (ATCC), NIH3T3 (ATCC) and NIH3T3 Tet-On 3G (Clontech) cells were cultured in DMEM medium (Gibco) supplemented with 10% fetal bovine serum (FBS; Invitrogen) at 37 °C in a humidified 10% CO 2 environment. All cells were confirmed to be free of mycoplasma using an e-Myco mycoplasma PCR detection kit (iNtRON). Cells were plated on 96-well plates (μ-Plate 96-well ibiTreat; ibidi GmbH) for live-cell imaging. For the NIH3T3 cells, the wells were coated with fibronectin (1:100; Invitrogen). The cells were transfected using either a microporator (Neon transfection system, Invitrogen) or Lipofectamine LTX (Invitrogen) according to the manufacturer's instructions. Lipofectamine RNAiMax (Invitrogen) was used for the transfection of PAMmer. Specifically, for RCas9-based mRNA-LARIAT, the plasmid was transfected into the cells; after the cells had settled on the bottom of the plate, they were transfected with PAMmer. The transfection efficiencies of HeLa and NIH3T3 cells were enhanced by two electroporation pulses of 980 V for 35 ms and three pulses of 1,450 V for 10 ms, respectively. For the immunoblot analyses, the NIH3T3 Tet-on 3G cells were electroporated before being plated onto six-well plates.
Hippocampal neuron preparation and transfection. Hippocampal cultures were prepared from embryonic day 18 (E18) Sprague Dawley rats. The embryos were placed in Hank's balanced salt solution (HBSS)-HEPES (Gibco), and their hippocampi were dissected and incubated in 0.25% trypsin for 15 min at 37 °C with agitation (tapping) at 5-min intervals. The plates were washed three times with HBSS-HEPES supplemented with FBS and triturated with fire-polished Pasteur pipettes. Dissociated tissues in neurobasal plating (Gibco) containing 10% horse serum (Invitrogen), 1% penicillin-streptomycin (Gibco) and 2% GlutaMAX (Gibco) were plated onto plates coated with 1 mg ml −1 poly-l-lysine (Sigma) and incubated at 37 °C in a humidified 5% CO 2 incubator for 1 h. The plating medium was replaced with neurobasal medium containing 2% B-27 (Gibco), 1% penicillinstreptomycin and 2% GlutaMAX. Neurons were transfected using Lipofectamine LTX according to the manufacturer's instructions. All experimental procedures involving animals were approved by the Animal Ethics Committee of the Korea Advanced Institute of Science and Technology. This study is compliant with all of the relevant ethical regulations regarding animal research.
Live-cell imaging and electronics. Live cells were imaged using a Nikon A1R confocal microscope (Nikon Instruments) mounted onto a Nikon Eclipse Ti body equipped with a Nikon CFI plan apochromat VC objective (numerical aperture (NA) ×60/1.4 or ×40/0.75; Nikon Instruments) and digital-zooming Nikon imaging software (NIS-element AR 64-bit version 3.21; Laboratory Imaging). The environmental conditions were maintained at 37 °C and 10 or 5% CO 2 (Live Cell Instruments) using a Chamlide TC system placed on a microscope stage. Unless otherwise indicated, the cells were photostimulated with single 1-s loop using a 488-nm laser at a light-power density of 490 μW mm −2 using a photostimulation module in the Nikon imaging software (NIS-elements). The blue-light intensity was measured using an optical power meter (ADCMT), with a custom 96-well LED array (Photron) to ensure uniform blue-light stimulation across all wells. For the FISH and immunofluorescence experiments, 10-s pulses of light were delivered for 1 h at intervals of 5 min.
Cell viability test. HeLa cells transfected with a microporator were plated onto 96-well plates. After 24 h, the growth medium was replaced with fresh medium (DMEM with 10% FBS) and the plate was placed on a 96-well LED array in a culture incubator. The cells were illuminated with blue light for 24 h at intervals of 1 min. The plates that were not illuminated were wrapped in aluminium foil to avoid exposure to light and placed in the same culture incubator. Dead cells were stained by the addition of 1 μM SYTOX blue dead cell stain for 5 min to each well. After staining, fluorescence imaging was performed to identify the dead transfected cells.
Imaging processing and analysis. Images were analysed using Nikon imaging software (NIS-elements AR 32-bit version 3.21; Laboratory Imaging), MetaMorph software (version 7.8.1.0; MDS Analytical Technologies), ImageJ software (version 1.50b; U.S. National Institutes of Health; http://imagej.nih.gov/ij/) or Imaris (Imaris x64 version 8.4.1, Bitplane). The average fluorescence intensity was measured by integrating densities using the Nikon imaging software. The FISH images were deconvoluted using the Nikon imaging software. Cluster formation was quantified by first defining clusters as discrete puncta with a fluorescence intensity of 1,500-4,090 a.u. (excluding saturated pixels), size of >0.2 μm 2 and circularity of 0.5-1.0 a.u. The number of clusters per cell were determined using the 'Objective Count' tool in the Nikon imaging software. After defining each cluster, co-localization between fluorescence channels was quantified and the Pearson's correlation coefficient was applied using the 'Co-localization' tool in the Nikon imaging software, followed by Fisher's transformation. The dendrite staining intensity was measured using the 'Intensity Profile' tool in the Nikon imaging software. Changes in dendrite intensities were calculated as the differences in iRFP682 intensities before and after treatment, with these differences normalized to the FusionRed signal. Migration was analysed using available macro sequences, as described 37 . Dendrite length was analysed using the 'FilamentTracer' module in Imaris. For focal-adhesion analysis, paxillin-mCherry puncta with a fluorescence intensity of 1,500-4,090 a.u. and 0.5-10 μm in size were determined using the 'Threshold' tool in ImageJ. The number of stable focal adhesions were counted and the cell areas were measured using the ' Analyze Particle' tool. To analyse the celledge dynamics, the cell edges were parameterized and 60 windows were defined and followed to quantify edge velocity and draw activity maps 38 .
Ratio of target mRNA to ribosome subunits trapped in the clusters. The fluorescence intensity of the MBS or β-actin FISH signals inside the clusters as well as the total fluorescence after subtraction of background-defined as the median intensity within the boundaries of each cell 23 -were measured to estimate the ratios of mRNA located in clusters per cell. Because mRNA-LARIAT generates clusters of various sizes, with substantial differences between the intensities of the large and small clusters, all clusters could not be captured and analysed in a single image. Accordingly, the imaging conditions were set so that the intensity of large-cluster signals was not saturated, which inevitably resulted in the loss of signals from some small clusters. Thus, the calculated ratios of trapped mRNA were probably underestimates.
Correlation between cell-edge dynamics and focal adhesion. Cells were identified from the background signals and their edges were enhanced by applying an unsharp mask. The cell masks were defined by automatic thresholding. The edge velocities of protrusion and retraction were determined by tracking the boundary points from frame to frame. The cell edges and focal adhesion signal vectors were computed as the product of the vector with varied time lags as described previously 38 . Cross-correlation values were computed as a function of offset in time using Pearson's correlation coefficients. those shown in Figs. 5a, 5b-e and 7b, which were performed four, six and four times, respectively, with similar results. Significant differences between two variables were analysed by either a two-tailed Student's t-test using the Microsoft Excel 2013 and 2016 software or a one-way ANOVA with Tukey's multiple comparisons test using the GraphPad Prism 6 software. Unless otherwise indicated, a P < 0.05 was considered statistically significant. Specific P values are indicated in the figures.
Step-by-step protocols. The step-by-step protocol developed in this study can be found at Protocol Exchange 41 .
Polarization and distribution index analysis. Cells were identified after background subtraction by determining the median intensity within the boundaries of each cell. The polarization and distribution of each FISH signal were quantified as previously described 23 .
Immunoblot analysis. Whole-cell lysates were prepared in the dark using PRO-PREP solution (iNtRON Biotechnology). The proteins in the lysates (15 μg total protein per sample) were resolved by SDS-PAGE on a NuPAGE Novex 4-12% Bis-Tris gel (Invitrogen). The proteins were transferred to a nitrocellulose membrane using an iBlot transfer stack and an iBlot gel transfer device (Invitrogen) according to the manufacturer's instructions. The membranes were probed with the following primary antibodies: mouse anti-mCherry (1:2,000; Abcam), rabbit anti-β-actin (1:2,000; Cell Signaling Technology), rabbit anti-p-eIF2α (1:1,000; Cell Signaling Technology), mouse anti-eIF2α (1:1,000; Cell Signaling Technology) and/or rabbit anti-GAPDH (1:200; Santa Cruz). After washing, the membranes were incubated with goat anti-rabbit IRDye 680RD (1:15,000; LI-COR) or goat anti-mouse IRDye 800CW (1:15,000; LI-COR Biosciences), as appropriate. The blots were scanned with an Odyssey CLx infrared imaging system (LI-COR Biosciences) and the band intensities were measured using the Nikon imaging software with the 'ROI Statistics' tool.
Immunofluorescence. HeLa cells expressing mRNA-LARIAT components were illuminated with an LED array at a power of 0.7 mW mm −2 (470 nm) for 1 h at 37 °C in a 10% CO 2 incubator. The cells were immediately fixed with 4% paraformaldehyde at room temperature for 15 min and washed with PBST (PBS containing 0.5% Tween-20). Following cell permeabilization with PBS containing 0.1% Triton X-100 (Sigma) for 10 min and washing with PBST, the cells were blocked by incubation for 1 h in PBST containing 1% BSA, washed again in PBST and incubated overnight at 4 °C with the primary antibodies rabbit monoclonal anti-RPL10A (1:100; Abcam), rabbit anti-rpS6 (1:1,000; Cell Signaling Technology), rabbit anti-FMRP (1:100; Abcam) and mouse monoclonal anti-IMP-1 (1:50; Santa Cruz Biotechnology) diluted in PBST containing 1% BSA. The cells were washed in PBST and incubated at room temperature for 1 h with Alexa Fluor 633-conjugated goat anti-rabbit or Alexa Fluor 647-conjugated donkey anti-mouse secondary antibody (1:2,000; Invitrogen), as appropriate. After washing with PBST, the cells were imaged by confocal microscopy. The mitochondria in live cells were stained using MitoTracker (100 nM; Invitrogen) according to the manufacturer's instructions.

FISH.
Stellaris FISH probes against MBS sites-designed as described previously 27 -were labelled with CAL Fluor Red 635 or CAL Fluor Orange 560 (LGC Biosearch Technologies). Stellaris FISH probes targeting mCherry labelled with Quasar 670 dye were purchased (LGC Biosearch Technologies). The FISH probes recognizing human GAPDH and β-actin were purchased and labelled with Quasar 670 (LGC Biosearch Technologies). Mouse tubulin 1α, Arp2 and Arp3 probes were labelled with Quasar 670 (LGC Biosearch Technologies), as was the previously described mouse β-actin probe 27 . The mouse 18S and 28S rRNA FISH probes-designed as previously described 39 -were labelled with Quasar 670 and 570 (LBC Biosearch Technologies), respectively. Twenty-four hours after transfection, cells in 96-well plates (ibidi) were fixed in 4% paraformaldehyde solution (Electron Microscopy Sciences) for 10 min at room temperature, washed twice with 1×PBS (Roche) for 5 min each and permeabilized by immersion in 70% ethanol (EMD Millipore) for at least 24 h at 4 °C. These cells were washed with 20% Stellaris RNA FISH wash buffer A (LGC Biosearch Technologies), 70% nuclease-free water (Invitrogen) and 10% deionized formamide (Sigma) at room temperature for 5 min, distributed into wells, and incubated for 16 h in the dark at 37 °C with a 250-nM working probes solution prepared in 90% Stellaris RNA FISH hybridization buffer (LGC Biosearch Technologies) and 10% deionized formamide. The cells were subsequently washed again with 20% Stellaris RNA FISH wash buffer A (LGC Biosearch Technologies), 70% nuclease-free water (Invitrogen) and 10% deionized formamide (Sigma) at 37 °C for 30 min in the dark, followed by a wash with Stellaris RNA FISH wash buffer B (LGC Biosearch Technologies) at room temperature for 5 min. Finally, a drop of fluorescence mounting medium (Dako) was added to each well.
Quantitative real-time PCR. The collected cells were homogenized and their total RNA isolated using a PureLink RNA mini kit (Ambion). The RNA was reverse transcribed using a Superscript III first-strand synthesis system (Invitrogen) to generate cDNA. For real-time quantitative PCR, each reaction mixture contained EvaGreen smart mix (Solgent). The relative mRNA levels were calculated using the CFX96 real-time PCR Detection System (BIO-RAD). The primers used in this study have been described 40 .  Fig. 2 | experimental scheme for analysing trapping of ribosomal components with target mRNAs in clusters. Sequestration of ribosomal components with target mRNAs was tested in various conditions depending on serum or puromycin treatment. Cells were either pre-treated or not treated with puromycin for 4 h. Then all cells were stimulated by blue light with an LED at 3-min intervals for 3 h. Cells with puromycin were either washed before or while light stimulation for 1 h. For serum starvation, cells were starved for 24 h. Then serum was re-added before or while light stimulation for 2 h. Endogenous ribosomes were visualized by staining with antibodies (Ab) against small and large ribosomal subunits or rRNA FISh probes.