Single-molecule imaging of microRNA-mediated gene silencing in cells

MicroRNAs (miRNAs) are small non-coding RNAs, which regulate the expression of thousands of genes; miRNAs silence gene expression from complementary mRNAs through translational repression and mRNA decay. For decades, the function of miRNAs has been studied primarily by ensemble methods, where a bulk collection of molecules is measured outside cells. Thus, the behavior of individual molecules during miRNA-mediated gene silencing, as well as their spatiotemporal regulation inside cells, remains mostly unknown. Here we report single-molecule methods to visualize each step of miRNA-mediated gene silencing in situ inside cells. Simultaneous visualization of single mRNAs, translation, and miRNA-binding revealed that miRNAs preferentially bind to translated mRNAs rather than untranslated mRNAs. Spatiotemporal analysis based on our methods uncovered that miRNAs bind to mRNAs immediately after nuclear export. Subsequently, miRNAs induced translational repression and mRNA decay within 30 and 60 min, respectively, after the binding to mRNAs. This methodology provides a framework for studying miRNA function at the single-molecule level with spatiotemporal information inside cells. For decades, miRNAs have been studied primarily by ensemble methods, where a bulk collection of molecules is measured outside cells. Here, Kobayashi and Singer report methods to image miRNA function at the single-molecule level inside cells.

After the discovery of the first miRNA in 1993 1 , miRNAmediated gene silencing has been studied for decades. However, the function of miRNAs has been monitored primarily by ensemble methods, e.g., luciferase assays, ribosome profiling, and RNA sequencing, where a bulk collection of molecules is measured ex vivo [14][15][16][17][18][19] . Thus, the behavior of individual molecules during miRNA-mediated gene silencing, as well as their spatiotemporal regulation inside cells, remains mostly unknown. Here we report a series of single-molecule methods to visualize each step of miRNAmediated gene silencing: RISC-binding, translational repression, and mRNA decay, in situ inside cells. As our methods visualize the function of miRNAs on a cell-by-cell basis, they enable both singlemolecule and single-cell analysis. These technical advantages, which overcome the limitation of canonical methods, provide novel insights into when, where, and how miRNAs work inside cells.

Results
Visualization of mRNA decay by miRNAs with single-molecule resolution. First, we sought to develop a method to visualize miRNA-mediated mRNA decay with single-molecule resolution. In human U2OS cells, which have been widely used for RNA imaging [20][21][22] , miR-21 is the most abundant miRNA 23 (Supplementary Fig. 1a-c). Thus, we constructed the reporter system that recapitulates mRNA decay by miR-21 (Fig. 1a). In this reporter system, where two different mRNAs were expressed under the control of a bi-directional promoter, firefly luciferase (Fluc) mRNAs represented the internal control. SunTag mRNAs with miR-21 sites were used to monitor miRNA-mediated mRNA decay, while SunTag mRNAs with mutant sites were used as the negative control. The miR-21 sites were designed so that the seed region (miRNA nucleotides 2-8) and the 3′ supplemental region (miRNA nucleotides [13][14][15][16] formed base-pairs with them [11][12][13] , while the mutant sites had mismatches that interrupted basepairing ( Supplementary Fig. 1d). In this method, reporter mRNAs were detected by single-molecule fluorescence in situ hybridization (smFISH) to visualize them with single-molecule sensitivity 24,25 (Fig. 1b and Supplementary Fig. 2).
Validating our method, U2OS cells expressing SunTag mRNAs with miR-21 sites showed a smaller number of mRNAs, compared with the negative control (Fig. 1c). For quantitative analysis, we performed three-dimensional (3D) fluorescence imaging, followed by single-molecule detection in 3D using the FISH-quant algorithm 26 (Supplementary Fig. 2). This analysis confirmed the reduction of mRNA stability when SunTag mRNAs have miR-21 sites (Fig. 1d, e and Supplementary  Fig. 3a-c). Although ensemble methods analyze a bulk collection of mRNAs from numerous cells 14,[16][17][18][19]23 , our method can analyze expression levels of mRNAs on a cell-by-cell basis. In addition, unlike canonical methods, our method can count the absolute number of mRNAs ( Supplementary Fig. 3a, b), thereby allowing the absolute quantification of miRNA function.
Visualization of translational repression by miRNAs with single-molecule resolution. Second, we attempted to develop a method to visualize miRNA-mediated translational repression with single-molecule resolution. To this end, we took advantage of the technique called single-molecule imaging of nascent peptides (SINAPS) 22 , where translation of reporter mRNAs can be visualized with single-molecule resolution. Based on the principle of SINAPS, we constructed the reporter mRNA that recapitulated translational repression by miR-21 (Fig. 2a). This reporter has the SunTag sequence, consisting of 24 tandem repeats of the GCN4 epitope 27 , in the ORF. Through immunofluorescence (IF) with anti-GCN4 antibodies, SunTag allows us to visualize nascent peptides being translated from mRNAs with single-molecule sensitivity. To inhibit the accumulation of SunTag throughout the cytoplasm, which dramatically increases background fluorescence, the degron sequence was added to the C terminus of the ORF 28 . To monitor miRNAmediated translational repression, miR-21 sites were inserted in the 3′ UTR. Notably, miRNAs also trigger mRNA decay (Fig. 1c), which causes a non-negligible reduction in the number of mRNAs for analysis. To overcome this problem, we added the anti-decay sequence, A 114 -N 40 , which protects mRNAs against deadenylation, the first step of miRNA-mediated mRNA decay [29][30][31][32] , to the end of the 3′ UTR. In this method, reporter mRNAs and their translation were visualized by smFISH and IF, respectively, with single-molecule resolution ( Fig. 2b and Supplementary Fig. 4).
U2OS cells expressing reporter mRNAs without miR-21 sites showed bright SunTag signals on mRNAs (Fig. 2c, left panels, white arrowheads), indicating that translation was successfully visualized. In agreement with this, SunTag signals on mRNAs almost completely disappeared upon treatment with puromycin, an inhibitor of translation ( Supplementary Fig. 5a-c). We also confirmed that adding the A 114 -N 40 sequence did not have a significant impact on translation ( Supplementary Fig. 5d-f). Importantly, U2OS cells expressing reporter mRNAs with miR-21 sites did not show bright SunTag signals on mRNAs (Fig. 2c, right panels, black arrowheads). The reduction of translational efficiency by miR-21 was confirmed by quantitative analysis (Fig. 2d, e and Supplementary Fig. 6a, b). Together, these results indicated that our method made it possible to visualize miRNAmediated translational repression with single-molecule resolution. When we used this method for single-cell analysis, we found that miRNAs can completely halt translation of target mRNAs within some cells, (Fig. 2e, see cells at the bottom), while some other cells looked insensitive to the silencing (Fig. 2e, see cells at the top). This variability was independent of expression levels of target mRNAs ( Supplementary Fig. 6c) and of nuclear sizes (Supplementary Fig. 6d), which correlated with the cell cycle 33,34 .
Canonical methods, where a bulk collection of mRNAs is analyzed, are sufficient to monitor translational repression by miRNAs [15][16][17][18][19][29][30][31][32] . However, even if these methods detect a 50% reduction in translational efficiency, they cannot address how miRNAs accomplished the 50% silencing inside cells; the number of translated mRNAs may be reduced to 50%, or the number of ribosomes on translated mRNAs may be reduced to 50%. Taking advantage of single-molecule methods, we addressed this issue. To identify the number of translated mRNAs, we performed 3Dcolocalization analysis between mRNAs and SunTag. In this analysis, the 3D positions of mRNAs and SunTag were localized at sub-pixel resolution by 3D Gaussian fitting 26 . Subsequently, based on the colocalization with SunTag, which is determined by the 3D distance, all mRNAs were classified into "untranslated" or "translated" mRNAs ( Supplementary Fig. 4). This analysis revealed that miRNAs reduced the number of translated mRNAs within cells (Fig. 2f, g). In SINAPS experiments, dim SunTag signals that do not colocalize with mRNAs (free SunTag) represent single SunTag peptides released from ribosomes 22 (Fig. 2c, arrows). On the other hand, bright SunTag signals consist of multiple SunTag peptides being translated by multiple ribosomes (Fig. 2c, white arrowheads). Thus, using the fluorescence intensity of free SunTag and that of SunTag on mRNAs (Fig. 2h), the number of ribosomes on translated mRNAs could be roughly estimated 22 (Supplementary Fig. 4). This analysis revealed that miRNAs also reduced the number of ribosomes on translated mRNAs ( Supplementary Fig. 6e).
Imaging of RISC-binding with single-molecule resolution. Third, we sought to establish a method to image RISC-binding with single-molecule resolution. To this end, the reporter mRNA for translational repression, harboring eight miR-21 sites (Fig. 2a), was repurposed to image RISC on mRNAs. In this method, RISC was imaged by IF with anti-AGO antibodies, while reporter mRNAs were imaged by smFISH with single-molecule resolution (Fig. 3a, and Supplementary Fig. 7). Although there are four AGO proteins (AGO1-4) in humans 12 , AGO2 is predominantly expressed in U2OS cells 35 (Supplementary Fig. 8a), hence we focused on AGO2 in this study. To ensure the reliability of IF with anti-AGO2 antibodies, we PonA bi-promoter, PonA-inducible bi-directional promoter. b Schematic of the smFISH experiment to visualize miRNA-mediated mRNA decay at single-mRNA resolution. Green and magenta spots represent SunTag and Fluc mRNAs, respectively. c The images of U2OS cells expressing the 8× miR-21 mutant reporter (left) and the 8× miR-21 reporter (right). Fluc mRNAs (magenta, top) and SunTag mRNAs (green, bottom) were labeled by smFISH. Nuclei (blue) were stained by DAPI. Scale bar, 5 μm. d, e mRNA decay mediated by miR-21. Images were analyzed using CellProfiler and FISH-quant. Then, mRNA stability (the relative abundance of SunTag mRNAs to Fluc mRNAs) was calculated as described in Supplementary Fig. 2 (see also Methods), followed by normalization to the value of the negative control. The results of bulk analysis (d) and single-cell analysis (e) are shown. In e, each circle represents a single cell (n = 50 for each condition), while red lines represent the medians. The p-value of one-tailed Mann-Whitney test is shown. Source data are provided as a Source Data file.
tested two distinct monoclonal antibodies (the 4G8 mouse antibody and the 11A9 rat antibody), which have already been wellcharacterized [36][37][38] , and confirmed that both antibodies showed similar IF patterns ( Supplementary Fig. 8b), which is consistent with previous studies looking at endogenous AGO2 39,40 . For our method, it was crucial to eliminate the unwanted RISC-binding to reporter mRNAs independent of miR-21. Thus, we carefully removed the potential miRNA sites (8mer, 7mer, and 6mer) 11 of the top 30 most abundant miRNAs from the reporter mRNAs ( Supplementary  Fig. 1b, c).
Notably, in some cell lines, AGO2 was found in the nucleus and in the cytoplasm at the comparable levels 41,42 . In U2OS cells, however, the majority of AGO signals was observed in the cytoplasm ( Supplementary Fig. 8c), hence we focused on cytoplasmic mRNAs in our analyses. As with translational efficiency (Fig. 2e), our method highlighted the cell-to-cell variability in RISC-binding efficiency (Fig. 3d); some cells showed high RISC-binding efficiency, while RISC-binding in some other cells was comparable to the negative control. This variability was also independent of expression levels of target mRNAs (Supplementary Fig. 8d), expression levels of AGO ( Supplementary  Fig. 8e), and nuclear sizes ( Supplementary Fig. 8f). Although RISC-binding was successfully imaged at the single-mRNA level, our reporter mRNAs have eight miR-21 sites, so it was still unclear whether our method can detect single-RISC molecules. To investigate the sensitivity of this method to RISC, we constructed reporter mRNAs harboring a single miR-21 site or its mutant site (the negative control). When we compared AGO signals on them, there was a small difference, which was statistically significant ( Supplementary Fig. 8g, h), suggesting that our method can detect at least a part of single-RISC molecules. However, given that the difference was quite small, our method would not be robust enough to detect all single-RISC molecules unequivocally. Therefore, in addition to~30% of mRNAs that are RISC-positive (Fig. 3f, right), more mRNAs would likely be bound by RISC.
Simultaneous visualization of single mRNAs, translation, and RISC-binding. Since we developed a series of methods to visualize each step of miRNA-mediated gene silencing with single-molecule resolution, we next explored the relationship between these steps at the single-mRNA level. To this end, we visualized single mRNAs, translation, and RISC-binding simultaneously ( Fig. 4a and Supplementary Fig. 9), using the reporter mRNA harboring miR-21 sites (Fig. 2a). Based on 3Dcolocalization analyses, all mRNAs were classified into four classes: (1) RISC-negative untranslated mRNAs, (2) RISCnegative translated mRNAs, (3) RISC-positive untranslated mRNAs, and (4) RISC-positive translated mRNAs (Fig. 4b, c). Validating our method, reporter mRNAs with miR-21 sites were classified as "RISC-positive" and "untranslated" more often than the negative control (Fig. 4c). A small portion (~10%) of reporter mRNAs without miR-21 sites were classified as "RISC-positive" (Fig. 4c, left). However, since the potential miRNA sites of major miRNAs were already removed from our reporter mRNAs ( Supplementary Fig. 1b, c), this could be attributed to the nonspecific colocalization between mRNAs and AGO signals by chance. In line with this, reporter mRNAs without miR-21 sites were not translationally repressed even when we focused on RISC-positive mRNAs ( Supplementary Fig. 10a). It is noteworthy that a substantial amount of reporter mRNAs with miR-21 sites were classified as RISC-negative untranslated mRNAs (Fig. 4c, right). We speculate that this class would include mRNAs that are already silenced and released from RISC. Supporting this hypothesis, reporter mRNAs with miR-21 sites were translationally repressed even when we focused on RISC-negative mRNAs ( Supplementary Fig. 10b). When we used the classification data for single-cell analysis, RISC-binding efficiency and translational efficiency showed a trend of negative correlation (Fig. 4d), which is consistent with the function of RISC.
Unexpectedly, when we analyzed each class of mRNAs at the single-molecule level, we found that translated mRNAs tend to be bound by RISC, compared with untranslated mRNAs ( Fig. 4e and Supplementary Fig. 11a). This tendency was confirmed by quantitative analysis, where we analyzed the fluorescence intensity of AGO on untranslated mRNAs and translated mRNAs (Fig. 4f). In line with this, RISC-positive mRNAs tended to be translated, compared with RISC-negative mRNAs ( Fig. 4g and Supplementary Fig. 11b). The quantitative analysis confirmed that the mRNAs bound by RISC are efficiently translated (Fig. 4h). Given that RISC does not activate translation under these conditions 43 , these data indicate that RISC preferentially binds to translated mRNAs rather than untranslated mRNAs. To validate this finding, we quantified AGO signals on mRNAs in the presence or absence of a translation inhibitor, cycloheximide. This analysis confirmed that RISC-binding to reporter mRNAs was attenuated by the treatment with cycloheximide (Supplementary Fig. 11c, d). As cycloheximide immobilizes ribosomes on mRNAs, cycloheximide treatment did not change SunTag signals on mRNAs ( Supplementary Fig. 11e). The results of single-mRNA analysis ( Fig. 4e-h) and single-cell analysis (Fig. 4d) may appear to be contradictory at first glance. Although RISC showed a preference for translated mRNAs at the single-mRNA level, RISC is the molecule that represses translation. We speculate, therefore, cells with high or low RISC-binding efficiency corresponded to cells with low or high translational efficiency at the single-cell level, respectively. Single-molecule imaging of miRNA-mediated translational repression. a Schematic of the reporter construct to recapitulate miRNA-mediated translational repression. PonA promoter, PonA-inducible promoter. b Schematic of the SINAPS experiment to visualize miRNA-mediated translational repression at single-mRNA resolution. Green and magenta spots represent SunTag peptides and reporter mRNAs, respectively. c The images of U2OS cells expressing the 8× miR-21 mutant reporter (left) and the 8× miR-21 reporter (right). Reporter mRNAs (magenta, top) and SunTag peptides (green, middle) were labeled by SINAPS. Merged images are shown at the bottom. White and black arrowheads indicate translated and untranslated mRNAs, respectively, while white arrows indicate "free" SunTag peptides. Scale bar, 1 μm. d, e Translational repression mediated by miR-21. Images were analyzed using CellProfiler and FISH-quant. Then, translational efficiency was calculated by division of the total SunTag intensity on mRNAs by the number of mRNAs as described in Supplementary Fig. 4 (see also Methods). The results of bulk analysis (d) and single-cell analysis (e) are shown. In e, each circle represents a single cell (n = 50 for each condition), while red lines represent the medians. The p-value of one-tailed Mann-Whitney test is shown. f Reduction of the fraction of translated mRNAs by miR-21. The fraction of translated mRNAs was calculated as described in Supplementary Fig. 4 (see also Methods). Each circle represents a single cell (n = 50 for each condition), while red lines represent the medians. The p-value of one-tailed Mann-Whitney test is shown. g The ratio of untranslated and translated mRNAs. All mRNAs were classified into untranslated or translated mRNAs based on 3D-colocalization analysis. Spatiotemporal analysis of RISC-binding, translational repression, and mRNA decay by single-mRNA imaging. Even though RISC preferred translated mRNAs over untranslated mRNAs, if RISC repressed translation immediately, most of RISCpositive mRNAs should be untranslated mRNAs. As our data showed the opposite (Fig. 4e-h), we speculated that RISC needs a relatively long time to repress translation. Taking advantage of our methods, which can visualize RISC-binding, translational repression, and mRNA decay, at the single-mRNA and single-cell levels, we next explored the time-course of miRNA-mediated gene silencing.
Firstly, using the methods we developed (Figs. 2a and 4a), we performed spatiotemporal analysis of RISC-binding and  [44][45][46][47][48] (Fig. 2a). After the pulse of PonA treatment, reporter mRNAs were observed by single-mRNA imaging at three different time points (Fig. 5a). During these experiments, the outlines of the nuclei and cells, visualized by DAPI and the non-specific background signals of smFISH probes, respectively, were automatically detected by the CellProfiler algorithm 49 (Supplementary Fig. 9). Validating our experiments, the ratio of the number of cytoplasmic mRNAs to that of nuclear mRNAs increased over time after nuclear export (Fig. 5b). Notably, single-cell analysis revealed that RISC binds to cytoplasmic mRNAs immediately after the completion of PonA treatment ( Fig. 5c and Supplementary Fig. 12a). These cytoplasmic mRNAs should be the mRNAs immediately after nuclear export, because most mRNAs are still in the nucleus at this time point (Fig. 5b). Single-mRNA analysis that analyzed the intensity of AGO on cytoplasmic mRNAs, confirmed that RISC-binding took place instantly ( Fig. 5d and Supplementary Fig. 12c). When focusing on translational repression, however, single-cell analysis showed that RISC did not repress translation until 30 min after the PonA pulse ( Fig. 5e and Supplementary Fig. 12b). This was confirmed by single-mRNA analysis, where the number of ribosomes on translated mRNAs in the cytoplasm was analyzed ( Fig. 5f and Supplementary Fig. 12c). Together, these results indicated that RISC bound to mRNAs immediately after their transport from the nucleus to the cytoplasm, followed by translational repression within 30 min. The data of spatiotemporal analysis also confirmed that RISC preferred translated mRNAs over untranslated mRNAs (Fig. 5g, h).
Finally, we performed spatiotemporal analysis of mRNA decay using our reporter system (Fig. 1a). Under the control of the PonAinducible bi-directional promoter, Fluc mRNAs, the internal control, and SunTag mRNAs, the reporter to monitor mRNA decay, were expressed for time-course experiments (Fig. 6a). As with the analysis for RISC-binding and translational repression (Fig. 5b), the ratio of the number of cytoplasmic mRNAs to that of nuclear mRNAs was increased over time (Fig. 6b), indicating that our spatiotemporal analysis worked well. Unlike translational repression, mRNA decay was not observed at 30 min after the PonA pulse (Fig. 6c). Instead, single-cell analysis showed a reduction of mRNA stability at 60 min after the PonA pulse, indicating that RISC induced mRNA decay within 60 min after the binding to mRNA. Additional time points will reveal a more precise time-course of miRNA-mediated gene silencing, e.g., the time lag between translational repression and mRNA decay, in the future.

Discussion
Historically, miRNAs have been studied primarily by ensemble methods, where a bulk collection of molecules was measured outside cells [14][15][16][17][18][19]23,[29][30][31][32][41][42][43] . Although recent studies provided several valuable methods to analyze the function of miRNAs more precisely [50][51][52][53][54][55] , the behavior of individual molecules inside cells, as well as their spatiotemporal regulation, remains mostly unknown. In this study, we developed a series of single-molecule methods, which enabled us to image each step of miRNAmediated gene silencing: RISC-binding, translational repression, and mRNA decay, inside cells. Our methods, which overcame the limitation of canonical methods, provided novel insights into when, where, and how miRNAs work inside cells (Fig. 7): (1) RISC bound to mRNAs immediately after their transport from the nucleus to the cytoplasm; (2) RISC preferred translated mRNAs over untranslated mRNAs; (3) RISC repressed translation of mRNAs within 30 min after the binding; (4) RISC reduced both the number of translated mRNAs and the number of ribosomes on translated mRNAs; (5) RISC completely halted translation of mRNAs in some cells; (6) RISC induced mRNA decay within 60 min after binding to mRNAs.
We found that RISC bound to mRNAs immediately after nuclear export (Fig. 5 and Supplementary Fig. 12). Notably, the basal level of translation was not fully active at this time point (Fig. 5e and Supplementary Fig. 12b, see the data of the negative control). Together, these findings indicate that RISC bound to mRNAs when translation was not fully active, and kept them from being activated. This way of silencing would be more efficient than after translation was underway. We also found that RISC preferentially bound to translated mRNAs rather than untranslated mRNAs ( Fig. 4 and Supplementary Fig. 11). Were RISC to recognize translated mRNAs and untranslated mRNAs with the same efficiency, this would be a waste of RISC, as RISC does not need to bind to untranslated mRNAs. Thus, we speculate that this preference would contribute to the economical use of RISC, the number of which is limited inside cells 35,56 . The mechanism that enables RISC to preferentially bind to translated mRNAs should be investigated in the future. It is noteworthy that small interfering RNAs (siRNAs) also prefer translated mRNAs 57 ; a recent study revealed that translating ribosomes unfold mRNA structures and unmask siRNA sites, thereby promoting interactions between target mRNAs and siRNAs. Although most miRNA sites are located within the 3′ UTR, where ribosomes do not access, as base-pairing within RNAs can occur over large distances 58 , miRNAs might prefer translated mRNAs through a similar mechanism.
RISC generally targets the 3′ UTR of mRNAs. It has been proposed that this is because RISC targeting regions other than the 3′ UTR may be removed from mRNAs by translating ribosomes before it represses translation 13,59,60 . Notably, this model is based on the assumption that RISC may need a longer time to repress translation than the speed of translation. However, Fig. 4 Simultaneous visualization of single mRNAs, translation, and RISC-binding. a Schematic of the SINAPS-IF-FISH experiment to visualize translation and RISC-binding simultaneously at single-mRNA resolution. Magenta, green, and cyan spots represent reporter mRNAs, SunTag peptides, and RISC, respectively. b The images of a RISC-negative untranslated mRNA (first row), a RISC-negative translated mRNA (second row), a RISC-positive untranslated mRNA (third row), and a RISC-positive translated mRNA (fourth row) in U2OS cells are shown. Reporter mRNAs (magenta, first column), SunTag peptides (green, second column), and AGO (cyan, third column) were labeled by SINAPS and IF-FISH. Merged images are shown on the right side. Scale bar, 1 μm. c The ratio of RISC-negative untranslated (magenta), RISC-negative translated (green), RISC-positive untranslated (cyan), and RISC-positive translated (orange) mRNAs. All mRNAs were classified into these four classes based on 3D-colocalization analysis. d A trend of negative correlation between translational efficiency and RISC-binding efficiency at the single-cell level. Images were analyzed using CellProfiler and FISH-quant. Then, translational efficiency and RISC-binding efficiency were calculated as described in Supplementary Fig. 9 (see also Methods). Each circle represents a single cell (n = 50 for each condition). e, f Translated mRNAs tend to be RISC-positive mRNAs. The fraction of RISC-positive mRNAs (e) and the intensity of AGO on mRNAs (f) are shown. In f, the means with SEM and the p-value of one-tailed Mann-Whitney test are shown. Magenta and green bars represent the values of untranslated and translated mRNAs, respectively. g, h RISC-positive mRNAs tend to be translated mRNAs. The fraction of translated mRNAs (g) and the number of ribosomes on mRNAs (h) are shown. In h, the means with SEM and the p-value of one-tailed Mann-Whitney test are shown. Magenta and cyan bars represent the values of RISC-negative and RISC-positive mRNAs, respectively. Source data are provided as a Source Data file.
the time span from RISC-binding to translational repression has been unknown. In this study, our spatiotemporal analysis revealed that RISC needs~30 min to repress translation after the binding to mRNAs (Fig. 5). As this is much slower than translation initiation rates, typically faster than~1 per min 61 , our finding provides a missing piece to explain why RISC generally uses the 3′ UTR of mRNAs: translational repression by RISC is too slow to compete with ribosomes. Nonetheless, translation initiation rates differ depending on each mRNA species 61 and are dynamically regulated in response to cellular contexts [62][63][64] . *** *** *** *** *** Fig. 5 Spatiotemporal analysis of RISC-binding and translational repression by single-mRNA imaging. a Schematic of spatiotemporal analysis of RISCbinding and translational repression by single-mRNA imaging. In pulse-chase experiments, IF-FISH and SINAPS were performed to visualize RISC-binding and translation at single-mRNA resolution. Magenta, green, and cyan spots represent reporter mRNAs, SunTag peptides, and RISC, respectively. b Transport of reporter mRNAs from the nucleus to the cytoplasm. In pulse-chase experiments, reporter mRNAs were labeled by smFISH. The ratios of the number of cytoplasmic mRNAs to that of nuclear mRNAs are shown. Each circle represents a single cell (n = 50 for each condition), while red lines represent the medians. cyto, cytoplasmic. nuc, nuclear. c-f Time-course analysis of RISC-binding (c and d) and translational repression (e and f) by single-mRNA imaging. Images were analyzed using CellProfiler and FISH-quant. Then, RISC-binding efficiency (c), the intensity of AGO on mRNAs (d), translational efficiency (e), and the number of ribosomes on translated mRNAs (f) were calculated as described in Supplementary Fig. 9 (see also Methods). In c and e, each circle represents a single cell (n = 50 for each condition), while red lines represent the medians. The p-values of Dunn's multiple comparisons test are shown. *** and n.s. represent p < 0.001 and not significant (p > 0.05), respectively. p-values in c were 3.0 × 10 −10 (0 min), 1.6 × 10 −13 (30 min), and 6.1 × 10 −13 (60 min), while those in e were > 0.99 (0 min), <1.0 × 10 −15 (30 min), and < 1.0 × 10 −15 (60 min). g Translated mRNAs tend to be RISC-positive mRNAs at all time points. The fraction of RISC-positive mRNAs is shown. Magenta and green bars represent the values of untranslated and translated mRNAs, respectively. h RISC-positive mRNAs tend to be translated mRNAs at all time points. The fraction of translated mRNAs is shown. Magenta and cyan bars represent the values of RISC-negative and RISC-positive mRNAs, respectively. Source data are provided as a Source Data file. In pulse-chase experiments, smFISH was performed to visualize mRNA decay at single-mRNA resolution. Green and magenta spots represent SunTag and Fluc mRNAs, respectively. b Transport of reporter mRNAs from the nucleus to the cytoplasm. In pulse-chase experiments, reporter mRNAs were labeled by smFISH. The ratios of the number of cytoplasmic mRNAs to that of nuclear mRNAs are shown. Each circle represents a single cell (n = 50 for each condition), while red lines represent the medians. cyto, cytoplasmic. nuc, nuclear. c Time-course analysis of mRNA decay by single-mRNA imaging. Images were analyzed using CellProfiler and FISH-quant. Then, mRNA stability (the relative abundance of SunTag mRNAs to Fluc mRNAs) was calculated as described in Supplementary Fig. 2 (see also Methods), followed by normalization to the value of the negative control at each time point. Each circle represents a single cell (n = 50 for each condition), while red lines represent the medians. The results of Dunn's multiple comparisons test are shown. *** and n.s. represent p < 0.001 and not significant (p > 0.05), respectively. p-values were 0.24 (0 min), 0.11 (30 min), and 6.4 × 10 −10 (60 min). Note that time-course data of RISC-binding, translational repression, and mRNA decay were analyzed using exactly the same statistical test (Dunn's multiple comparisons test). Source data are provided as a Source Data file. Therefore, our model does not exclude noncanonical miRNA sites within the 5′ UTR and the ORF. The reason why RISC needs 30 min to repress translation should be addressed in future work. mRNA regulation is a fundamental step in gene regulation, thereby influencing a wide variety of biological processes and diseases. To regulate mRNAs, various RNA-binding proteins (RBPs) associate with mRNAs, which trigger the stabilization or degradation of mRNAs, the activation or repression of translation, or the translocation of mRNAs to specific areas. The methods to analyze such mRNA regulation in situ, however, have been limited. Since our methods, which make it possible to visualize single mRNAs, translation, and RISC-binding simultaneously with single-molecule resolution inside cells, can potentially be customized for other RBPs, this methodology could be a valuable framework for studying mRNA regulation in the future.

Plasmid construction
The reporter plasmid for miRNA-mediated mRNA decay. The plasmid pPonA-BI-Gl-NORM-LacZA TER-LacZB (Addgene plasmid # 86212), which expresses two different mRNAs (Gl-NORM-LacZA and Gl-TER-LacZB) under the control of the PonA-inducible bi-directional promoter 46 , was used as the backbone sequence. The Gl-TER-LacZB sequence was replaced by the Fluc sequenc 65,66 , while Gl-NORM-LacZA sequence was replaced by the SunTag sequence followed by the AID degron 28 . The degron sequence was inserted in order to visualize translational repression, but not to visualize of mRNA decay. Eight miR-21 sites (or eight miR-21 mutant sites) were inserted into the 3′ UTR of the SunTag mRNA to recapitulate mRNA decay by miRNAs (Fig. 1a). The miR-21 sites were designed so that they form base-pairs with both the seed region and the 3′ supplemental region of miR-21 [11][12][13] (Supplementary Fig. 1d, left). Although the seed region is primarily responsible for miRNA targeting, a recent study revealed that RISC can bind to mRNAs only by base-pairing of the 3′ region of miRNAs without involvement of the seed region 67 . As such, the miR-21 mutant sites were designed so that neither the seed region nor the 3′ supplemental region forms base-pairs with them ( Supplementary Fig. 1d, right). Although the backbone sequence originally had the SV40 poly(A) signals, they were replaced by the bGH poly(A) signals. The reporter plasmid for miRNA-mediated translational repression. The plasmid for miRNA-mediated mRNA decay was used as the backbone sequence. The bGH poly(A) signal for the SunTag mRNA was replaced by the A 114 -N 40 -HhR sequence, which protects mRNAs from decay; the A 114 -N 40 sequence consists of a 114-nt poly(A) sequence (A114) followed by a 40-nt unrelated sequence (N40), which protects mRNAs against deadenylation, the first step of miRNA-mediated mRNA decay [29][30][31][32] . To eliminate the unwanted RISC-binding to reporter mRNAs independent of miR-21, all potential miRNA sites (8mer, 7mer, and 6mer) of top 30 most abundant miRNAs ( Supplementary Fig. 1b, c) were removed from the 3′ UTR of the reporter mRNAs.
The reporter plasmid for RISC-binding. To investigate the sensitivity of our method to RISC, the 8× miR-21 sequence within the plasmid for miRNA-mediated translational repression was replaced by a single miR-21 site or its mutant site. Reporter plasmids were purified using NucleoBond Xtra Midi kit (MACHEREY-NAGEL, 740410) for the purpose of nucleofection.
Nucleofection. U2OS cells were briefly rinsed with DPBS (Corning, 21-031-CV), followed by the treatment with Trypsin-EDTA (Thermo Fisher, 25300-054) to detach them. After adding culture media to neutralize trypsinization, cells were centrifuged at 300 × g for 1 min.  Supplementary Fig. 8b, anti-AGO2 Rat antibody (MilliporeSigma, MABE253, clone 11A9) was also tested at a 1:50 dilution. Cells were washed with the blocking buffer, followed by incubation with the blocking buffer supplemented with 2 μg/ml Goat anti-Mouse IgG conjugated with Alexa Fluor 647 (Thermo Fisher, A-21236) for 30 min. After washing with 1× PBS, immunostained cells were pre-hybridized and hybridized to perform smFISH as described in the smFISH section. For the experiments to visualize single mRNAs, translation, and RISC-binding simultaneously (Figs. 4 and 5), immunostained cells were pre-hybridized and hybridized to perform SINAPS as described in the SINAPS section. Image analysis. Images were analyzed using FISH-quant 26 , an algorithm implemented in MATLAB. Briefly, after background subtraction, FISH-quant automatically detects fluorescent spots and localizes them in 3D at sub-pixel resolution by fitting 3D Gaussians. This provides the number of spots inside cells, the intensity of each spot, and the 3D position of each spot 26 . To distinguish nuclear and cytoplasmic areas, the nuclei and cells were visualized by DAPI and the non-specific background signals of smFISH probes, respectively. Their outlines were automatically detected by the CellProfiler algorithm 49 , followed by conversion to the outline files compatible with FISH-quant. Based on these outlines, all spots were classified into "nuclear" and "cytoplasmic". To prepare images for figures (Figs. 1c, 2c, 3b, 4b and Supplementary Fig. 8b), raw images were processed using ImageJ software (Version: 2.1.0/1.53c).
Colocalization analysis. Colocalization in 3D was analyzed using FISH-quant 26 . First, the average drift between different colors was calculated and corrected. Then, using the 3D positions of spots, their 3D distances were calculated. When two spots in different colors were localized within the maximum allowed distance (mRNA-SunTag, 500 nm; mRNA-AGO, 250 nm), these spots were considered colocalized. Compared with SunTag IF spots ( Supplementary Fig. 6b, median,~250 spots per cell), AGO has many more IF spots ( Supplementary Fig. 8c, median,~2000 spots per cell), which cause the non-specific coincidence of the 3D positions of mRNAs and AGO by chance. To minimize such false colocalizations, we adopted the shorter maximum allowed distance for mRNA-AGO colocalization. Based on the 3D-colocalization with SunTag and AGO, mRNAs were classified into "translated", "untranslated", "RISC-positive", and "RISC-negative". Likewise, SunTag and AGO spots were also classified into "on mRNAs" and "free", depending on the colocalization with mRNAs.

Data analysis
Profiling miRNAs expressed in U2OS cells. The small RNA-seq data of U2OS cells (GEO accession: GSM416754) 23 was reanalyzed as described previously 65 .
Data analysis for miRNA-mediated mRNA decay. The mRNA stability (the relative abundance of SunTag mRNAs to Fluc mRNAs) of each cell, which is used for single-cell analysis (Figs. 1e and 6c), was calculated by Eq. 1, where S(k) is the mRNA stability of the k th cell. M Fluc,cyto (k) and M Sun,cyto (k) are the number of Fluc mRNAs and SunTag mRNAs, respectively, in the cytoplasm of the k th cell.