Real-time imaging of translation on single mRNA transcripts in live cells

Visualizing translation on single mRNA molecules

The SunTag system enables the detection of single protein molecules with high signal-to-noise ratio in live cells using multivalent fluorescence amplification (Tanenbaum et al., 2014). Here, we modified this system to fluorescently amplify nascent polypeptide signals for the purpose of detecting translation activity on single mRNA transcripts. To this end, we encoded a tandem array (24 copies) of short V4 peptides (Tanenbaum et al., 2014) in a reporter mRNA transcript, and expressed this construct in HeLa cells stably expressing a GFP-labeled scFv that can specifically bind to the cognate V4 peptide (Figure 1A). During translation, we anticipate that the nascent polypeptides will each bind multiple copies of pre-formed scFv-GFP and become readily detectable. This method thus avoids the time delay caused by the additional time required for GFP maturation, during which the protein product could diffuse away from the translation site. In addition, we fused an ornithine decarboxylase (ODC) to the C terminus of the V4 peptide array (Figure 1A). ODC is one of the most short-lived proteins, which can facilitate fast degradation of the V4 peptide array after translation is completed, potentially reducing the sequestration of scFv-GFP by the translation products and allowing the detection of translation at a relatively high expression level. Consistent with this anticipation, we found that detection of a large number of translation sites became more difficult if ODC was replaced with a stable protein. Finally, for the detection of mRNA molecules independent of translation, we also placed a tandem array (24 copies) of PP7 hairpins after the stop codon such that the mRNA can bind tdTomato-labeled PP7-coat proteins (tdPCP) (Figure 1A) (Wu et al., 2012). The tdTomato-tdPCP used in this work carried a nuclear localization signal such that unbound tdTomato-tdPCP molecules were mostly localized to the cell nucleus to reduce background (Wu et al., 2012).

Successful visualization of the translation sites requires that they are distinguishable from the protein products released after translation. Since an mRNA molecule binds multiple ribosomes simultaneously to form a large polysome complex, which moves much more slowly than the freely diffusing protein products, individual translating polysomes were readily detectable as bright GFP foci in the cytoplasm whereas the rapidly diffusing protein products formed a diffusive fluorescence background at our detection frame rate of 10 Hz (Figure 1B). As a validation, these GFP foci indeed co-localized with tdPCP-tdTomato signals associated with the mRNA (Figure 1B). As controls, when we deleted the PP7 array from the construct, we observed fluorescence foci only in the GFP channel but not in the tdTomato channel (Figure S1A); when we deleted the V4 array from the construct, we observed fluorescence foci only in the tdTomato channel but not in the GFP channel (Figure S1B).

For the full reporter construct containing both V4 and PP7, the distribution of the signals detected from individual foci in the tdTomato channel showed primarily a single peak corresponding to a signal that is equivalent to ~16 purified tdTomato molecules (Figure S1C), which is consistent with the notion that most of the observed foci correspond to single mRNA molecules because the binding efficiency of tdPCP to the PP7 hairpin has been previously shown to be close to 100% (Wu et al., 2012). We note that, in addition to the primary peak, the distribution also showed a longer tail with a small fraction of foci exhibiting signals apparently greater than that of ~24 tdTomato molecules (Figure S1C), which may represent clusters or aggregates of more than one mRNA molecules. In addition, we also determined the distribution of the signals of the detected foci in the GFP channel and compared these signals with the signal expected from a single V4 peptide array to estimate how many ribosomes reside in individual translation foci (Figure S1D, see caption of this Figure for the detailed method). Results of this comparison gave an estimate of ~12 ribosomes per mRNA molecule on average. This number corresponds to an average inter-ribosome distance of ~260 nucleotides on our transcript construct, which is consistent with previously measured values at a transcriptome scale in human cells (Hendrickson et al., 2009), although we note that such value should vary between different transcripts.

Imaging changes in translation activity caused by translation inhibitors

To further confirm that the observed GFP foci were indeed translation sites, we used translation inhibitors to probe the dynamics of these spots. We first studied the effect of two inhibitors for translation elongation, puromycin (Puro) and cycloheximide (CHX). Mimicking tRNA, Puro binds to the ribosome A site and receives the elongating nascent polypeptide chain from the P site tRNA. Upon Puro treatment, the GFP foci rapidly disappeared (Figure 2A, B, Figure S2B, Movie S1) due to the release of the nascent polypeptides after being transferred to Puro. As expected, the rate with which the GFP foci disappeared (k_obs) depended on the concentration of Puro (Figure 2C, red). CHX, on the other hand, binds to the ribosome E site, preventing the releasing of E site tRNA for subsequent ribosomal translocation (Schneider-Poetsch et al., 2010), which stalls ribosomes on mRNA and should lead to the stabilization of the GFP foci signal. Consistent with this expectation, after addition of CHX, the signals observed from the GFP foci did not change substantially over time: in 100 sec of continuous illumination (1000 frames and 0.1 s per frame), the signal changed by no more than ~10–20% under this condition (Figure 2B). This result also provides a control to show that the changes induced by Puro treatment, or Puro plus CHX treatment (shown later) were not substantially affected by photobleaching. Next, we used Puro to probe CHX inhibition because the activity of Puro depends on tRNA translocation to expose the A site for Puro binding, while CHX binding will trap the ribosome in a pretranslocation state with an occupied A site. Indeed, consistent with previous in vitro studies of the CHX effect (Colombo et al., 1965), we observed a slowdown of the Puro-induced GFP foci disappearance due to CHX in a CHX concentration dependent manner (Figure 2B, C).

We next investigated the effect of a translation inhibitor, homoharringtonine (Homo). Homo inhibits elongation during initial rounds of peptide bond formation, effectively stalling the ribosomes that have just completed the initiation phase and are in the very early elongation phase, whereas the ribosomes that have already entered active elongation phase are not affected by this drug and hence can complete translation and run off the mRNA (Fresno et al., 1977). After the application of Homo, we observed a gradual decay of signals from the GFP foci, reflecting runoff of the ribosomes from the mRNA (Figure S3A, B). The decay kinetics allowed us to estimate the average time needed for a ribosome to translocate through the entire message (T_trans, Figure S3B). As expected, the average T_trans increased linearly with the length of the transcript (Figures S3B, C), from which we obtained an elongation rate of 4 amino acids per second. Our result approximately agrees with the rate of 5.6 amino acids per second for mammalian cells measured previously by ribosome profiling (Ingolia et al., 2011) and with the rate of 3–5 amino acids per second measured by a similar translation imaging method reported in this issue of Cell (Yan et al, 2016).

Taken together, these experiments with translation inhibitors provided further confirmation that the observed GFP foci represent translating polysomes.

Translational response to unfolded protein stress and oxidative stresses

To illustrate the utility of our method to monitor the dynamics of translation regulation in real time, we chose to study the responses of translation activity to two stress conditions: 1) DTT-induced unfolded protein stress in ER and 2) sodium arsenite (NaAsO₂)-induced oxidative stress. DTT- and NaAsO₂-induced stresses lead to the activation of the PERK and HRI kinases, respectively, both of which can phosphorylate the a subunit of eIF2 (eIF2a) (Figure S4A) (Dever, 2002; Gebauer and Hentze, 2004). Translation initiation depends on the GTP-bound form of eIF2 (eIF2-GTP), which is hydrolyzed to eIF2-GDP after translation initiation. Recycling of eIF2-GDP back to eIF2-GTP is catalyzed by eIF2B. During stress, phosphorylated eIF2a (eIF2a(P)) will bind stably to eIF2B, acting as a competitive inhibitor that prevents efficient GDP-to-GTP exchange of eIF2, leading to a decrease of the eIF2-GTP level, and thus a reduction in global translation (Figure S4A) (Dever, 2002; Gebauer and Hentze, 2004). As expected, both DTT and NaAsO₂ treatments induced a decrease in the polysome signals of our reporter construct containing the V4 peptide array in HeLa cells (Figure 3A–C, 3D, E red lines, and Movie S2). As a control, pre-treatment with the translation inhibitor CHX, which prevents ribosome run-off, abrogated the reduction in translation signals under both stress conditions (Figure 3D, E, black lines), which again shows that the measured translation signal decay was not due to photobleaching.

Next, we studied the effect of a potent inhibitor of the PERK kinase, GSK2606414. Addition of this kinase inhibitor prevented translation shutdown induced by DTT, and also inhibited the NaAsO₂-induced translation shutdown but to a weaker extent (Figure 3D, E, blue lines), consistent the known weaker activity of this inhibitor for the HRI kinase (Axten et al., 2012). We further studied the effect of ISRIB on the translation responses to stresses. ISRIB is an integrated stress response inhibitor that enhances memory performance (Sidrauski et al., 2013) and has been recently shown to compensate the eIF2a(P)-induced inhibition effect of eIF2B (Sekine et al., 2015; Sidrauski et al., 2015). Addition of ISRIB nearly completely eliminated DDT-induced translation reduction (Figure 3D, magenta line), and also inhibited the NaAsO₂-induced translation reduction but again to a weaker extent (Figure 3E, magenta line). This could be due to the fact that NaAsO₂ induced a higher level of eIF2a phosphorylation (Figure S4B, C), which could reduce the counteracting effect of ISRIB (Sidrauski et al., 2015). These experiments also indicate that the decay of translation signals in the presence of DTT and NaAsO₂ could be largely accounted for by the eIF2a phosphorylation pathway and was unlikely a reflection of the generic decay of cell health.

Notably, the kinetics of translation signal changes upon DTT and NaAsO₂ treatments were similar, as observed both in single-cell time profiles (Figure 3B, C) and in ensemble-average curves of many cells (Figure 3D, E, red line). Several possible factors could potentially contribute to the observed kinetics: 1) the time required for the stress agents to enter the cell, 2) the time required to activate the kinases and phosphorylate eIF2a to inhibit eIF2B, 3) the time required to consume existing eIF2-GTP in the cell, and 4) the run-off time of ribosomes that were on and actively translating the mRNAs. The first factor was unlikely to be rate limiting because when we added Puro to inhibit translation, the translation signal decrease started within several seconds, much shorter than the 200–300 lag times observed in the DTT and NaAsO₂ cases. Our observation that the kinetics of translation signal decays were largely similar for DTT and NaAsO₂ treatments with the DTT-induced decay being only slightly slower (Figure 3D, E, red lines), in spite of the substantially higher level of eIF2a phosphorylation induced by NaAsO₂ (Figure S4B, C), suggests that the second factor – the time required to reach a sufficient level of eIF2a phosphorylation to cause substantial eIF2B inhibition – was also unlikely a major rate limiting factor in our measured kinetics. This is probably because eIF2B is expressed at a much lower level compared to eIF2a and hence a slight eIF2a phosphorylation can lead to efficient inhibition of eIF2B (Dever, 2002). The third factor could potentially be a rate-limiting factor: after the activity of eIF2B is inhibited, the time required to consume existing eIF2-GTP in the cell to sufficiently low levels to cause reduction in the translation initiation rate may contribute significantly to the observed lag times and decay kinetics in the translation signals. Finally, the fourth factor could also contribute significantly to the observed kinetics: even if translation initiation was shut down immediately, it would still take a substantial amount of time (~300s) for those ribosomes already on the mRNA to finish translation (Figure S3B); this ribosome run-off time would convolve with the translation initiate rate decay to give the overall decay kinetics in our observed translation signals. The third and fourth factors should not be dependent on the eIF2a phosphorylation level, which could explain the apparent similar kinetics in the observation translation signal decays in the DTT and NaAsO₂ treatments.

Although global translation is repressed under unfolded protein and oxidative stresses, some specific genes are known to be up-regulated, exemplified by ATF4, whose translation is controlled by two upstream open reading frames (uORFs) (Harding et al., 2000). After translation of the first short ORF (uORF1), under the normal condition when eIF2-GTP is abundant, the second ORF (uORF2) will be translated through re-initiation, suppressing the expression of the ORF of the ATF4 gene (ORF3) (Figure 4A). During stress when the eIF2-GTP level is reduced, uORF2 will not be as efficiently re-initiated, which will allow ORF3 to be translated and cause an up-regulation of ATF4 (Lu et al., 2004; Vattem and Wek, 2004). To study this effect, we made a reporter construct containing the ATF4 uORFs and the V4 peptide array in ORF3 (Figure 4A). Notably, we observed that both DTT and NaAsO₂ treatments induced only a transient increase in translation signals, which was then followed by a substantial signal decrease both in individual cells (Figure 4B–E and Movie S3) and in the ensemble average of many cells (Figure. S5). Such a pulse-like translation up-regulation was observed for most cells (23 out 23 cells in the NaAsO₂ case and 21 out 25 cells in the DTT case). The kinetics of the pulse-like up-regulation were also largely similar for the two treatments, with the DTT-induced change being only moderately slower, consistent with the rate-limiting mechanisms described above. The decrease in translation following the initial increase is likely because, under our stress conditions, the eIF2-GTP level eventually decreased to a level too low to support efficient translation initiation of ORF3, suggesting that the ATF4 requires a certain range of eIF2-GTP levels for optimum expression. It is thus possible that the up-regulation would be sustained under weaker stress conditions. Consistent with this interpretation, a previous ensemble, steady-state study showed that effective up-regulation in the translation of endogenous ATF4 or a reporter construct controlled by ATF4 uORFs only occurs at an intermediate range of stress agent concentration and eIF2a phosphorylation level, and when the concentrations of the stress agent and the phosphorylation level of eIF2a become too high, the translation of ATF4 (or the reporter construct) is no longer up-regulated (Lu et al., 2004). uORFs have been found to regulate many other genes during stress responses (Andreev et al., 2015) and synaptic plasticity (Di Prisco et al., 2014). Similar to ATF4 uORFs, these other uORFs might also exert a temporal regulation of their down-stream genes under certain environmental stimuli. Overall, these results demonstrate the ability of our imaging method to capture, in real time, the transient changes in translation activity in response to environmental stimuli.

Distinct mobilities of individual polysomes in different subcellular compartments

Our ability to observe translation of mRNA transcripts in real time also allowed us to track the movement of individual translating polysomes. Previous studies have revealed diverse mobility behaviors of mRNAs (Park et al., 2014), although it is unknown whether the mRNAs are undergoing translation in these experiments. We analyzed the movements of individual polysomes in HeLa cells. The motion of polysomes could be mainly categorized into three categories: stationary, sub-diffusive, and diffusive (Figure 5A), consistent with previous mobility analysis using fluorescently labeled mRNAs (Fusco et al., 2003). Occasionally, rapid and directed motion could be observed for the polysomes, but at a very low frequency (less than 0.1% of the polysomes). The polysomes displayed a broad range of mobility (measured in terms of the mean-squared displacement (MSD) at 0.5 s time delay, Figure 5B), which had little correlation with fluorescence signal intensity (correlation coefficient -0.006), suggesting that the number of translating ribosomes on the mRNA was not a main determinant of the polysome mobility. We measured the mobilities of polysomes translating cytosolic proteins using both our original construct containing the V4 peptide array and ODC, as well as a construct containing a blue fluorescent protein (BFP) encoding sequence upstream of the V4 peptide array. Both constructs showed very similar polysome mobility behaviors. The average diffusion coefficients of the polysomes in individual cells varied from 0.01 to 0.06 μm²/s, and have averaged values at 0.021± 0.002 μm²/s and 0.021± 0.001 μm²/s (mean ± S.E.M. n = 42 and 45 cells) for the two constructs, respectively. These values are similar to previously reported mRNA mobility in human cancer cell lines (Halstead et al., 2015; Mor et al., 2010).

In comparison, we studied the mobility of polysomes translating secreted and transmembrane proteins by placing a secreted form of BFP (sBFP) or Smoothened receptor transmembrane domain (SMOTM) upstream of the V4 peptide array (Figure 5C and Figure S6A, B). We observed much smaller mobilities for these polysomes (Figure 5C), with a substantially reduced fraction of diffusive polysomes and much smaller average diffusion coefficients (0.0013 ± 0.0001μm²/s for sBFP and 0.0017 ± 0.0001μm²/s for SMOTM, n = 42 and 44 cell, respectively) (Figure S6C, D). The polysomes translating secretory or transmembrane proteins are known to be anchored to the ER through interactions with translocons (Johnson and van Waes, 1999). Such anchorage likely accounts for the observed low mobility.

Notably, the polysomes translating cytosolic proteins exhibit a heterogeneous mobility distribution, displaying a peak with low mobility values and a long tail of higher mobility values (Figure 5B). Furthermore, the polysomes in the perinuclear region exhibited a much larger percentage of such low-mobility polysomes, whereas this low-mobility population was substantially reduced, but not fully eliminated, in the non-perinuclear polysomes (Figure 6A, B and Movie S4). As a result, the average diffusion coefficient of the perinuclear polysomes, 0.0053±0.0003 μm²/s, was much smaller than those of the polysomes not in the perinuclear region, 0.028± 0.002 μm²/s (Figure 6C). We speculate that this subpopulation of slowly moving polysomes may be interacting with cellular structures, giving rise to their low mobility. Given the enrichment of this slow-moving population in the perinculear region, such cellular structures may also be enriched in the perinuclear region, with possible candidates being ER (Reid and Nicchitta, 2015), the nuclear envelop (Mahamid et al., 2016) or other structures.

Visualizing local translation in the dendrites of neurons

Last, we demonstrated the ability of our method to probe the dynamics of local translation in complex, polarized cells, such as neurons. Spatial regulation of translation is critical for the functions of neurons (Buxbaum et al., 2015; Jung et al., 2014; Sutton and Schuman, 2006). We expressed the scFv-GFP and the reporter construct containing the V4 peptide array in cultured primary hippocampal neurons using lentivirus. Fluorescent translation foci were readily detected in neurons and the disappearance of these foci upon Puro treatment confirmed their identity as translation sites (Figure 7A). The polysomes in neurons displayed a substantially lower mobility compared to those in HeLa cells: the average diffusion coefficient of polysomes was 0.006 ± 0.001 μm²/s in neurons versus 0.021± 0.002 μm²/s in HeLa cells (p-value = 0.001 using unpaired t test, n = 35 cells for neurons, n = 42 cells for HeLa cells), suggesting different interactions of polysomes with local cellular environment in different cell types. This result is in line with previously observed low mobility of mRNAs in neurons (Park et al., 2014).

It has been shown previously that the 3’UTR of some mRNAs, such as that of the Arc gene, harbor ‘zipcode’ sequences that are responsible for transporting these mRNAs to dendrites for local translation (Bramham et al., 2008; Buxbaum et al., 2015). We thus placed the 3’UTR of the Arc gene, after the stop codon of our V4 peptide array construct (Figure 7B). As anticipated, more translating polysomes were observed in the dendrites after addition of the Arc 3’UTR (Figure 7B, C). The average distance to the cell body was 42 ± 2 μm for the polysomes with the Arc 3’UTR, substantially larger compared to the average distance (19 ± 2 μm) observed for the polysomes without the Arc 3’UTR (p-value = 0.0002, unpaired t test, n = 26 and 21 cells for construct with and without Arc 3’UTR, respectively).

Surprisingly, we also observed a substantially larger fraction of translating polysomes exhibiting rapid and directed motion in the dendrites of neurons than in HeLa cells (7% in neurons versus <0.1% in HeLa cells) (Figure 7D and Movie S5). These polysomes could move in both anterograde (56%) and retrograde (44%) directions. The speed of the directed motion was distributed around 1 – 3 μm/s (Figure 7E), with the average value being 2.0 μm/s. Since we detect the nascent polypeptide signal here, these results show that active transport of mRNA can occur after translation has already started and the mRNA has already entered the polysome state. We note that the reporter constructs used in these experiments did not contain the PP7 array, hence untranslated mRNAs were not detected and it is thus unknown what fraction of the mRNAs undergoing active transport were translated. It has been previously suggested that locally translated mRNAs are kept in a translationally silent state before being transported to the proper destination by the binding of repressing factors that block translation initiation, translation is repressed during mRNA transport, and such inhibition is lifted after these mRNAs are localized to their destinations by local signaling or locally enriched trans-acting factors (Besse and Ephrussi, 2008). Our observation that active transport of mRNA can occur after the mRNA has already started translation and entered the polysome state suggests that the engagements of mRNA with transport machinery and with translation machinery are in fact not mutually exclusive, at least for some mRNAs, highlighting the complexity of spatial regulation of translation. We cannot, however, rule out the possibility that the already initiated translation of the polysomes could become momentarily repressed during transport, for example by ribosome stalling.

Strengths and limitations of the method

In this paper, we developed an imaging approach that allows translation of individual mRNA molecules to be directly visualized in live cells and demonstrated the capabilities of this method to reveal spatiotemporal dynamics of translation regulation under different environmental stimuli and in different cell types. A similar in vivo single-molecule translation imaging method based on SunTag is reported in the same issue of Cell (Yan et al, 2016).

As a direct visualization method, this imaging method has several advantages in the investigation of translation regulation. First, it provides a real-time readout of translation activity, which allows the dynamics of transient translational regulation events to be readily captured. For example, we detected an intriguing pulse-like translation up-regulation for the transcripts containing the ATF4 uORFs. Yan et al. detected translation initiation, elongation, ribosome stalling, and rapid and reversible transitions between active and inactive translation states of individual mRNA molecules in real time (Yan et al, 2016). Second, this imaging method preserves accurate spatial information of translation sites, which enables the study of the movement dynamics of individual polysomes and the local translation activities in specific subcellular compartments. Taking advantages of this capability, we observed that polysomes translating different types of proteins (cytosolic or membrane/secretary) had distinct mobilities and that polysomes translating cytosolic proteins even exhibited distinct mobilities in different subcellular locations. Moreover, we directly captured local translation in neuron dendrites and observed that mRNAs that already started translation in the polysome state can undergo active transport. High detection sensitivity is a third merit of this approach. The ability to detect translation activity of single mRNA molecules makes this method a valuable tool to study translation regulation of lowly expressed genes.

It is however worth noting that an imaging method like this one also has its own limitations. For example, the method has a relatively low throughput, typically allowing the study of only one or two mRNA species at a time, and can be challenging for studying those special cell types that are difficult to introduce recombinant genes. In comparison, ribosome profiling has much higher throughput since it provides information about translation activity of endogenous genes at the whole transcriptome level. Moreover, ribosome profiling also provides translation rate with sub-ORF and even sub-codon resolution, whereas our imaging method does not have such resolution since the translation activity is reported by adding a reporter tag (V4 peptide array) to the mRNA of interest. In addition, the introduction of the reporter tag sequence into the target mRNA could potentially leads to perturbation. In this current study, our experimental observations, such as the measured translation elongation rate, the observed translation responses to inhibitors and environmental stresses, and the observed 3’UTR-dependent local translation in dendrites are all consistent with previous results, suggesting that the perturbation induced by the reporter tag was not substantial. However, we cannot formally rule out the possibility that, when attached to certain specific transcripts, the reporter tag could potentially exert appreciable perturbation and that additional measures may need to be taken to minimize such perturbation.

Future development of the method

DNA constructs

DNA constructs used in this study (pFUGW-scFv-sfGFP-GB1, pFUGW-NLS-tdPCP-tdTomato, pEF-24xV4-ODC-24xPP7, pEF-24xV4-MBP-CAAX-24xPP7, pEF-24xV4-TagBFP-ODC-24xPP7, pEF-24xV4-MBP-ODC-24xPP7, pEF-ATF4 uORFs-24xV4-ODC-24xPP7, pEF-sBFP-P2A-24xV4-ODC-24xPP7, pEF-SMOTM-P2A-24xV4-ODC-24xPP7, pFUGW-24xV4-ODC, pFUGW-24xV4-ODC-Arc 3’UTR) were generated using Gibson assembly from different insert fragments and linear vector backbones. A detailed description is provided in Supplemental Experimental Procedures.

HeLa cell line stably expressing scFv-GFP and tdPCP-tdTomato

Hela cells were infected with lentiviruses encoding scFv-GFP and NLS-tdPCP-tdTomato. A stable cell line that expresses both components at optimal levels for the detection of translation foci was selected as described in Supplemental Experimental Procedures.

Primary hippocampal neuron culture

Primary hippocampal neurons were infected at 1 day in vitro (DIV 1) by lentiviruses encoding scFv-GFP and translation reporter constructs. The translation signals were visualized at DIV 14–16. The neuronal culture conditions were provided in Supplemental Experimental Procedures.

Live cell imaging

The objective (Olympus UPlanSApo 100×, 1.4 NA) was maintained at 37 °C throughout the experiment by objective heater (Bioptechs). A 488-nm krypton gas laser (Innova 70C, Coherent) and a 561-nm solid-state laser (Sapphire, Coherent) were used to excite GFP and tdTomato, respectively. The emissions of GFP and tdTomato were separated by a quad-view system (photometrics) onto different quadrants of an electron-multiplying CCD camera (iXon897, Andor). In the quadview system, the emissions of GFP and RFP were separated by a 560-nm long pass dichoic mirror, and further passed emission filters ET525/50m and ET595/50m, respectively. The GFP and tdTomato channels were aligned using 100 nm fluorescent microspheres (TetraSpeck, Invitrogen) immobilized on a glass coverslip. The images were recorded at a camera frame rate of 10 Hz and, in some experiments, the laser illumination was only turned on in a subset of frames as detailed in Supplemental Experimental Procedures.

Image analysis

Individual translation foci were identified using a previously described peak-finding algorithm (Bates et al., 2007), which provides the intensity as well as the centroid position of each fluorescence foci. The analyses of translation activity during drug or stress treatments, as well as detailed polysome tracking procedures, are provided in Supplemental Experimental Procedures.