Research Article

Real-time quantification of single RNA translation dynamics in living cells

See allHide authors and affiliations

Science  17 Jun 2016:
Vol. 352, Issue 6292, pp. 1425-1429
DOI: 10.1126/science.aaf0899

The when, where, and how of translation

High-resolution single-molecule imaging shows the spatial and temporal dynamics of molecular events (see the Perspective by Iwasaki and Ingolia). Wu et al. and Morisaki et al. developed an approach to study the translation of single messenger RNAs (mRNAs) in live cells. Nascent polypeptides containing multimerized epitopes were imaged with fluorescent antibody fragments, while simultaneously detecting the single mRNAs using a different fluorescent tag. The approach enabled a direct readout of initiation and elongation, as well as revealing the spatial distribution of translation and allowing the correlation of polysome motility with translation dynamics. Membrane-targeted mRNAs could be distinguished from cytoplasmic mRNAs, as could single polysomes from higher-order polysomal complexes. Furthermore, the work reveals the stochasticity of translation, which can occur constitutively or in bursts, much like transcription, and the spatial regulation of translation in neuronal dendrites.

Science, this issue p. 1430, p. 1425; see also p. 1391


Although messenger RNA (mRNA) translation is a fundamental biological process, it has never been imaged in real time in vivo with single-molecule precision. To achieve this, we developed nascent chain tracking (NCT), a technique that uses multi-epitope tags and antibody-based fluorescent probes to quantify protein synthesis dynamics at the single-mRNA level. NCT reveals an elongation rate of ~10 amino acids per second, with initiation occurring stochastically every ~30 seconds. Polysomes contain ~1 ribosome every 200 to 900 nucleotides and are globular rather than elongated in shape. By developing multicolor probes, we showed that most polysomes act independently; however, a small fraction (~5%) form complexes in which two distinct mRNAs can be translated simultaneously. The sensitivity and versatility of NCT make it a powerful new tool for quantifying mRNA translation kinetics.

At the core of all gene regulatory networks are the processes of DNA transcription and RNA translation. Although transcription is now regularly quantified in real time in vivo with single-gene resolution (1, 2), the same cannot be said for translation. In principle, fluorescent tags such as green fluorescent protein (GFP) could be used to do this, but in practice, these tags take too long to fluoresce or their signals are too weak to visualize translation of a single RNA in real time (3). Recently, some of these problems were overcome by using a RNA biosensor that is dislodged by translating ribosomes (4). However, fluorescence is lost on the first round of translation with this assay, so ongoing translation dynamics cannot be visualized.

Developing a method to visualize the translation of single mRNAs in living cells

To visualize translation of a single RNA in real time, we developed a system based on bright photostable small-molecule dyes, antibody enhancement, and multi-epitope protein tags. We constructed a plasmid encoding the large nuclear protein KDM5B, N-terminally tagged with a 10× FLAG tag (which we refer to as the spaghetti monster, SM) (5) and containing a 24× MS2 tag (6) in the 3′ untranslated region (UTR) (Fig. 1A). The FLAG SM tag created a highly avid site for binding of fluorescently labeled fragments of antibody to FLAG (anti-FLAG Fab), whereas the MS2 stem-loop repeat allowed visualization with labeled MS2 coat protein (MCP) (2, 6) (Fig. 1B). We transiently transfected this plasmid into U2OS cells that were subsequently bead-loaded with Cy3-labeled anti-FLAG Fab and Halo-tagged MCP [labeled with the far-red JF646 fluorophore (7)]. Twenty-four hours after transfection, MCP marked mRNA in the cytoplasm and Fab marked KDM5B in the nucleus, suggesting that neither the FLAG SM tag nor the presence of Fab interfered with mRNA and KDM5B production and localization (Fig. 1C). To see how soon Fab could mark protein, we imaged ~6 hours after transfection. At this early time, Fab only lightly marked the nucleus, suggesting that very little KDM5B had been synthesized (Fig. 1D). Fab also colocalized and co-moved with many MCP-labeled mRNAs in the cytoplasm (Fig. 1, D and E, and movie S1). These bright co-moving spots displayed RNA-like diffusivity, were very stable—lasting for 2 hours or more (movie S2)—and could only be seen in the cytoplasm of cells that were both transfected with FLAG SM–tagged KDM5B and bead-loaded with Fab.

Fig. 1 A system for imaging single RNA translation kinetics.

(A) Our plasmid encodes KDM5B with an N-terminal 10× FLAG SM tag followed by a 24× MS2 tag in the 3′ UTR (aa, amino acids). (B) Schematic of the system. RNA (red) is marked by MCP (labeled with JF646) that binds to repeated MS2 stem loops in the 3′ UTR; protein (green) is labeled by Fab (conjugated to Cy3) that binds to peptide epitopes in the N terminus. (C) A deconvolved image showing that the FLAG SM–tagged KDM5B protein (“SM-KDM5B”) localizes to the nucleus, whereas its mRNA localizes to the cytoplasm 24 hours after transfection and bead-loading with Fab and MCP. (D) Six hours after transfection, FLAG SM–tagged KDM5B colocalizes with mRNA in punctae (arrows). (E) Example co-movement of FLAG SM–tagged KDM5B and mRNA punctae circled in yellow in (D). (F) Addition of puromycin (50 μg/ml) leads to a loss of the FLAG SM–tagged KDM5B signal in punctae; this does not occur in control cells loaded with vehicle. (G) Quantification of the loss of the FLAG SM–tagged KDM5B signal from punctae (lower green curve) in a single cell as a function of time after addition of puromycin (which takes effect at 0 s), compared with the signal from a control cell (upper gray curve). Curves are normalized to pre-puromycin levels. Scale bars, 10 μm.

We wondered whether the co-moving protein-mRNA spots were bona fide translation sites. To test this, we treated cells with 50 μg/ml of puromycin, an inhibitor of translation that releases nascent chains from ribosomes (8). Within minutes of drug addition, the number of co-moving spots dropped exponentially (Fig. 1, F and G, and movie S3). By tracking co-moving spots, we could see the disappearance of Fab-labeled protein, despite the persistence of the mRNA (Fig. 1F). The time of complete protein disappearance varied among different mRNAs, but there was a well-defined exponential decay in the number of protein spots (Fig. 1G). To further confirm that protein-mRNA spots were translation sites, we treated cells with 4 μg/ml cycloheximide to slow elongation and load more ribosomes per transcript, and the spots became brighter (fig. S1). Together, these data suggest that the co-moving spots were indeed translation sites.

KDM5B is a large protein (1544 amino acids), so its translation should be relatively easy to detect. To see whether we could also detect translation of smaller proteins, we constructed two plasmids (Fig. 2A) encoding either β-actin (374 amino acids) or the core histone H2B (125 amino acids). As with KDM5B, neither the FLAG SM tag nor Fab disturbed the localization of these proteins (Fig. 2, B and C), and Western blots confirmed that tagged proteins were of the expected length (fig. S2). Furthermore, we could again observe translation sites ~6 hours after transient transfection (Fig. 2D and movies S4 and S5), indicating that our system is useful for imaging the translation of protein-coding genes of varying sizes.

Fig. 2 Quantifying the mobility, ribosomal content, and structure of translation sites.

(A) Plasmids encoding β-actin (SM-β-actin) and H2B (SM-H2B), analogous to the FLAG SM–tagged KDM5B construct in Fig. 1A. (B and C) Deconvolved images showing that SM-β-actin [(B), green] and SM-H2B [(C), green] localize to the cytoplasm and nucleus, respectively, whereas their mRNAs (red) both localize to the cytoplasm 24 hours after transfection and bead-loading with Fab and MCP. (D) Six hours after transfection, SM-β-actin translation sites can be seen (arrows) and tracked (yellow circle and inset). (E and F) Quantification of the frequency (E) and intensity (F) of translation sites. Normalization to the number of ribosomes is shown on the right axis in (F) (a.u., arbitrary units). (G) Measured mean squared displacements (MSD) of tracked polysomes as a function of time. (H) Histograms of the Gaussian fit distances between mRNA and protein in tracked polysomes. Error bars show SEM. Scale bars, 10 μm.

Nascent chain tracking to quantify polysome mobility and size

We tracked translation of single molecules by using a technique that we call nascent chain tracking (NCT). With NCT, we followed individual H2B, β-actin, and KDM5B translation sites for hundreds to thousands of seconds in full cell volumes (300 time points × 2 colors × 13 z planes = 7800 images per movie). We adjusted laser powers to focus exclusively on translation sites rather than on fully translated single protein products (which could interfere with tracking). As shown in the inset of Fig. 2D, this allowed us to accurately compare (i) the appearance frequency and brightness, (ii) the mobility, and (iii) the size of translation sites. Of these parameters, frequency and brightness varied the most, tending to increase with construct length. We detected the translation of 86 ± 2% of KDM5B mRNA but just 19 ± 4% of β-actin mRNA and 4 ± 1% of H2B mRNA (error, SEM) (Fig. 2E). Furthermore, KDM5B translation sites, as marked by Fab, were over 1.5 times as bright as β-actin sites, which themselves were nearly 1.5 times as bright as H2B sites (Fig. 2F).

One explanation for the difference in brightness in translation sites could be a difference in the number of nascent peptide chains per mRNA, as would be the case in polysomes (9). To determine precisely how many nascent chains exist per site, we calibrated fluorescence by imaging a new β-actin plasmid containing a single 1× FLAG tag rather than the 10× FLAG SM tag (fig. S3A). With this plasmid, only one Fab can be bound per peptide chain, allowing a direct comparison between translation site fluorescence and single Fab fluorescence. By imaging cells transfected with this plasmid at high laser powers, both single Fabs and translation sites could be detected and tracked (fig. S3, B and C), revealing translation sites to be on average 3.1 ± 0.5 times as bright as single Fabs (fig. S3D). We therefore estimate that there are 3.1 ± 0.5 nascent peptide chains per β-actin translation site, 2.1 ± 0.4 per H2B site, and 5.1 ± 0.9 per KDM5B site (Fig. 2F, right axis). Combining these data and assuming one ribosome per nascent chain, we conclude that detected translation sites are polysomes that can contain as few as one ribosome every 900 mRNA bases or as many as one ribosome every 200 mRNA bases.

In addition to differences in their brightness, NCT also exposed differences in the mobility of polysomes. We quantified this by measuring the mean squared displacement of tracked polysomes as a function of time. For the nuclear proteins H2B and KDM5B, mobility was modeled well by diffusion. Not only did displacement increase linearly with time for at least 20 s, the less massive H2B polysomes also moved significantly faster than KDM5B polysomes (Fig. 2G). In contrast, β-actin polysomes displayed constrained diffusion, with jump sizes that were initially between those of KDM5B and H2B (consistent with diffusion) but that ultimately lagged behind both at longer times. This constrained movement of β-actin could be due to interactions with cytoplasmic binding partners. Despite these trends, there was substantial variability in mobility between mRNA, so that we sometimes saw rapidly moving KDM5B polysomes (up to 6 μm2/s) and nearly immobile H2B polysomes (~0.01 μm2/s; fig. S4). This made it nearly impossible to identify a translated mRNA based on mobility alone, implying that the translation machinery only weakly alters mRNA movement in our system.

Unlike their brightness and mobility, the size of polysomes was less variable between constructs. To quantify sizes, we measured the distance between the 3′ UTR of polysomal mRNA (labeled with MCP) and the nascent peptide chains (labeled with Fab). The fluorescence from polysomes was within diffraction-limited spots, so we determined their mean positions with superresolution by using Gaussian fitting (fig. S5). According to hairpin models of polysome organization (10), this distance should grow as the length of the mRNA grows. Instead, we found that the distance was shortest in KDM5B polysomes, typically around 65 nm, compared with 85 nm for H2B and 105 nm for β-actin (Fig. 2H). This suggests that the polysomes that we imaged are organized in a globular shape rather than an elongated shape, consistent with recent atomic force microscopy images (11).

Extracting translation kinetics from NCT data

Having measured the basic physical properties of polysomes, we next focused on translation kinetics. In particular, we wondered how the number of ribosomes in polysomes is controlled. This number reflects a balance between incoming and outgoing ribosomes and therefore depends on the ribosome elongation rate. One way to noninvasively estimate this rate is to examine the correlation of fluctuations in NCT data by means of fluorescence correlation spectroscopy (FCS), similar to how transcription elongation rates have been measured by using MS2 fluorescence fluctuations (12). Before doing so, however, we needed to ensure that Fabs would bind polysomes quickly and irreversibly on the time scale of translation. If not, their dynamics would contribute to the fluctuations, and this would distort our FCS measurements of elongation times.

To measure how quickly Fabs bind polysomes, we microinjected them into cells transfected 6 hours earlier with our KDM5B construct and preloaded with MCP. Just 3 s after microinjection (as soon as we could image), many polysomes were labeled by Fab, implying that the binding time is less than 3 s (Fig. 3, A and B). To measure the lifetime of Fab binding, we performed fluorescence recovery after photobleaching (FRAP) experiments in cells transfected with the H2B construct and bead-loaded with Fab and MCP 24 hours earlier. We chose H2B because it is known to remain bound for hours (13, 14), so any fluorescence recovery on the time scale of minutes would be exclusively due to Fab turnover. As Fig. 3C shows, there was little FRAP recovery in 4 min, implying that most Fabs are bound much longer. These binding kinetics (Fig. 3D) make Fabs ideal tools for measuring translation elongation times on time scales ranging from ~10 s to ~5 min.

Fig. 3 Quantifying the translational kinetics of tracked polysomes.

(A) Sample cell transfected with FLAG SM–tagged KDM5B and loaded with MCP before Fab microinjection. Many mRNAs (red) can be seen. Their fluorescence does not bleed into the green channel (inset). (B) Three seconds after microinjection, Fabs (green) co-localize with mRNAs (arrows and inset). The site of microinjection can be seen on the right (bright green smear). (C) Sample FRAP experiment on a cell transfected with SM-H2B and bead-loaded with MCP and Fab 24 hours earlier. There is little recovery in 200 s (lower curve; int., intensity). Error bars show SEM. (D) A cartoon of results from (A) to (C) showing fast on rates and slow off rates for Fab (green Y shapes) binding to SM epitopes (triangles) as they emerge from a ribosome (circle) translating mRNA (thick line). (E) The intensity of a tracked FLAG SM–tagged KDM5B polysome (yellow circle and inset) can be measured as a function of time. The cartoon below shows how movement of ribosomes along mRNA and the emergence of elongating peptide chains can produce intensity fluctuations at the indicated times t1, t2, and t3 (AAA, poly-A tail). (F) The average correlation curves calculated from intensity fluctuation data for each construct (error bars show SEM; G, autocorrelation function). The time at which the correlation hits zero can be obtained from fits (dashed lines) to estimate the elongation dwell time. (G) Calculated elongation rates (amino acids per second) from fits in (F) (error bars show 95% confidence intervals). Scale bars, 10 μm.

Knowing the limits of Fab, we analyzed the fluorescence fluctuations of polysomes by extracting their intensity time series from our tracking data. We began with KDM5B polysomes because these were the brightest and most numerous. The intensity of polysomes fluctuated with time (Fig. 3E), reflecting changes in the number of elongating ribosomes. From each intensity time series, we computed the correlation curve, and we averaged these together. There was high variability among mRNAs (fig. S6), but the average correlation curve revealed a clear linear drop in correlation that went to zero at 149 ± 20 s (Fig. 3F, upper panel). In direct analogy to transcription correlation analyses (12), the time at which the correlation drops to zero marks the total elongation dwell time. To confirm this, we treated cells with 100 μg/ml of cycloheximide to stall translation. As expected, the correlation disappeared (fig. S7 and movies S6 and S7).

To corroborate these measurements, we performed FRAP on KDM5B polysomes. We photobleached a large section of the cytoplasm where many polysomes were present (fig. S8). By tuning the powers of the photobleaching laser, we could selectively photobleach just the Fab, leaving the mRNA bright. This allowed us to monitor the fluorescence recovery of the relatively slower-moving polysomes. On average, polysomes recovered 80 to 90% of their fluorescence in 125 to 180 s, although there was again high variability among mRNAs, just as with FCS. Nevertheless, the average recovery time was on the same time scale as our FCS measurements.

Given this consistency, we next performed FCS on the shorter β-actin and H2B constructs. Again, the correlation curves were linear and went to zero at distinct elongation dwell times, whereas random background spots showed no correlations from frame to frame (fig. S9). As expected, the dwell times decreased with mRNA length, being 32 ± 9 s for β-actin and just 16 ± 7 s for H2B (Fig. 3F, lower panels). Importantly, for all constructs, the correlation vanished at times greater than the dwell time. This implies that initiation is random, so there is no memory between initiation events, similar to transcriptional initiation (12) and in contrast to bursting (15, 16).

To calculate the elongation rate, we divided the length of the encoded protein by the elongation dwell time of each construct. The calculated rates were all within error (Fig. 3G), yielding a single consistent elongation rate of 10 ± 2.3 amino acids/s, which is fairly close to what has been measured using genome-wide ribosomal profiling (5.6 amino acids/s) (17, 18). The difference is probably due to single-molecule variability (as shown in figs. S4 and S6) or differences in mRNA sequence and codon usage (19).

With a consistent elongation rate, we can unify our observations. First, assuming KDM5B elongation occurs at 10 ± 2.3 amino acids/s, a new ribosome would have to initiate on average every 30 ± 9 s to maintain the measured 5.1 ± 0.9 ribosomes per polysome. From this we can predict that 96 ± 3% of KDM5B mRNA will be translated by polysomes, 3.1 ± 2.5% of KDM5B mRNA will be translated by a single ribosome, and 0.6 ± 0.6% will be untranslated. Moreover, using the same initiation and elongation rates for the other constructs (because they have the same 5′ and 3′ UTRs), we can predict that 35 ± 9% of β-actin mRNA will be translated by polysomes containing 2.5 ± 0.1 ribosomes on average (36 ± 2% translated by single ribosomes and 29 ± 7% untranslated), whereas just 6.5 ± 2.5% of H2B mRNA will be translated by polysomes containing 2.2 ± 0.1 ribosomes on average (27 ± 3% translated by single ribosomes and 66 ± 5% untranslated).

These predictions (detailed in the supplementary materials and summarized in figs. S10 and S11) are consistent with our earlier measurements of polysomes (19 ± 4% of β-actin mRNA containing 3.1 ± 0.5 ribosomes and 4 ± 1% of H2B mRNA containing 2.1 ± 0.4 ribosomes). They are also in agreement with independent measurements of the number of ribosomes per polysome obtained by polysome profiling (figs. S10 and S11). Although we found a lower density of ribosomes within polysomes than others have (20, 21), the consistency of our live-cell and biochemical data suggests that the difference is due to mRNA variability rather than to experimental stress that might artificially lower the density. In particular, other mRNAs with different 5′ and 3′ architectures (22, 23) will probably have different polysome occupancies and dynamics, depending not only on the balance of elongation and initiation but also on the metabolic status of the cell and the local environment.

Simultaneous multicolor imaging of distinct mRNAs being translated in a single cell

One advantage of using Fab to mark translation sites is the large number of high-affinity antibodies for multicolor applications. To demonstrate this, we generated new Fabs from hemagglutinin (HA) antibodies and labeled these with Alexa488 dye. In parallel, we engineered a new KDM5B construct with a 10× HA SM tag (HA-KDM5B) (5) to complement FLAG SM–tagged KDM5B (hereafter, FLAG-KDM5B), as shown in Fig. 4A. As a first application of this technology, we wanted to test whether polysomes interact with each other to form higher-order structures that can translate two distinct mRNAs at the same time. For this, we cotransfected cells with HA- and FLAG-KDM5B and bead-loaded them with MCP and anti-HA and anti-FLAG Fab. As anticipated, cotransfected cells contained two types of polysomes in equal fractions (Fig. 4, B to E), one type labeled by anti-HA Fab (HA-KDM5B) and the other labeled by anti-FLAG Fab (FLAG-KDM5B) (Fig. 4B and movie S8). For the most part, there was little interaction between the two, providing direct evidence that the vast majority of KDM5B polysomes act independently of one another. However, a small fraction (~5%) of KDM5B polysomes formed complexes that co-moved for hundreds of seconds (Fig. 4, C and D, and fig. S12) and that produced both HA- and FLAG-tagged nascent peptide chains. By measuring the distance between the nascent HA and FLAG chains, we found the complexes to be roughly twice the size of a single polysome (Fig. 4F), suggesting that the component polysomes remain compartmentalized. These complexes could reflect a more general strategy to either assemble higher-order complexes cotranslationally (24) or co-regulate the expression of two genes.

Fig. 4 Three-color imaging of the translation of two distinct proteins and mRNAs.

(A) Complementary plasmids for imaging the translation of FLAG-KDM5B (green) and HA-KDM5B (blue). (B) Sample tracks from a cell expressing FLAG- and HA-KDM5B that was bead-loaded 3 hours earlier with MCP and anti-FLAG and anti-HA Fab. A translation site (circled) harboring both FLAG- and HA-KDM5B polysomes is tracked in (C). (D) The fluorescence from the spot in (C) (cropped and centered on mRNA) shows strong spatial overlap of FLAG- and HA-KDM5B. (E) The percentage of FLAG- and HA-KDM5B mRNA in polysomes and the percentage of mRNA in multi-polysome complexes. Error bars show SEM. (F) The Gaussian fit distance between the colocalized FLAG- and HA-KDM5B in the circled spot in (B) as a function of time. The distribution of these distances is shown on the right. It peaks at ~130 nm, twice the distance reported in Fig. 2H between KDM5B nascent chains and mRNA in a single polysome (~65 nm, dashed line).

This work is similar to a companion study by Wu et al. (25) that combined the SunTag (26) with MS2 to image single mRNA translation kinetics in live cells. Their measurements and our measurements of translation elongation and initiation rates are within a factor of 2, indicating that the general approach is reproducible between laboratories. Although the techniques are similar, the combination of anti-FLAG Fab and anti-HA Fab enables multicolor experiments that are not possible with the SunTag system. Also, the FLAG and HA epitopes (8 and 9 amino acids, respectively) are just over one-third the size of the SunTag epitope (22 amino acids), so spatiotemporal resolution can be up to three times as good and imaging is potentially less invasive. These advantages will make NCT a powerful new tool for detecting, tracking, and quantifying translation dynamics and for dissecting gene regulatory networks in vivo.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S12

References (2733)

Movies S1 to S8

References and Notes

  1. Acknowledgments: This work was supported by start-up funds from Colorado State University. We thank R. Cohen, F. Scavone, H. Kimura, S. DiPietro, A. Ambrosio, L. Stargell, A. Thurston, and J. Chao for help with experiments. We thank Wu et al. (25) for sharing their preliminary data and the Transcription Imaging Consortium at the Janelia Research Campus of the Howard Hughes Medical Institute for creating the environment that made our interactions and discussions possible. T.M. and T.J.S. planned, analyzed, and carried out all experiments except for microinjections. T.J.S., T.M., and K.L. wrote the manuscript. K.L. helped with plasmid construction and transfections, cell culture, analysis, Western blotting, and figures. K.F.D. and J.G.D. performed microinjections. B.P.E. provided guidance for the multicolor single-molecule microscopy. Z.Z. purified Halo-MCP. L.L.L. and S.V. provided HA and FLAG SM constructs. L.D.L. and J.B.G. provided fluorophores. T.L. provided help with experimental design.
View Abstract

Navigate This Article