Translation Dynamics of Single mRNAs in Live Cells

Abstract

The translation of mRNA into proteins represents the culmination of gene expression. Recent technological advances have revolutionized our ability to investigate this process with unprecedented precision, enabling the study of translation at the single-molecule level in real time within live cells. In this review, we provide an overview of single-mRNA translation reporters. We focus on the core technology as well as the rapid development of complementary probes, tags, and accessories that enable the visualization and quantification of a wide array of translation dynamics. We then highlight notable studies that have utilized these reporters in model systems to address key biological questions. The high spatiotemporal resolution of these studies is shedding light on previously unseen phenomena, uncovering the full heterogeneity and complexity of translational regulation.

1. INTRODUCTION

Translation is the fundamental biological process by which genetic information stored in mRNA is converted into proteins. This process is a defining feature of life that gives cells the ability to quickly change their phenotype in response to environmental pressures (18, 114). The study of translation is therefore crucial to deciphering the complex gene regulatory dynamics that underlie human health and disease.

Traditionally, translation has been studied with bulk assays, such as western blotting or ribosome profiling. While powerful, these assays only provide average snapshots, making it difficult to detect variations in translation across different cells or mRNAs. Imaging studies have sought to improve resolution by employing fluorescent protein fusion tags like GFP. However, these approaches also present challenges – the slow maturation and dim fluorescence of fusion tags prevents the real-time observation of translation at the single-mRNA level (24).

In 2016 the technological limitations that had prevented the study of single-mRNA translation in live cells were overcome by several groups concurrently (94, 101, 131, 136, 138). The solution turned the paradigm of imaging proteins on its head. Instead of fusing the protein directly to GFP and then waiting for the GFP to be translated, fold, mature, and fluoresce (Figure 1a), preformed GFP could instead be recruited to the protein during its translation via GFP-tagged intrabodies that bind to repeated N-terminal peptide epitopes (Figure 1b).

Figure 1. Imaging single-mRNA translation in living cells.

a) With a traditional GFP tag placed on the N-terminus of a protein of interest, the fluorophore matures too slowly to capture the translation process. b) Using an N-terminal repeat epitope tag, already fluorescent GFP-tagged intrabodies (Fab, scFv, nanobodies) can be recruited to the site of translation quickly. Fluorescence at the translation site is amplified above background because repeat epitopes can recruit more than one GFP-tagged intrabody and also because ribosomes can translate multiple nascent chains from a single mRNA. This creates bright fluorescent puncta that mark the translation sites for tracking in living cells. c) Sample translation sites in living U2OS cells. Addition of puromycin causes translation spots to quickly disappear. Scale bar, 10 µm.

Using this strategy, epitopes in the tag can be fluorescently labeled within seconds of their translation (94, 138). Also, fluorescence is amplified, both from the number of repeated epitopes in the tag and from the number of ribosomes translating the mRNA. This multiplicative amplification produces bright translation sites that can be tracked to accurately quantify translation dynamics from single mRNA (94, 101, 131, 136, 138) (Figure 1b,c). For example, using a widefield microscope with oblique illumination and EMCCD cameras (119), a single translation site can be tracked in multiple colors for thousands of frames (94). Individual translation sites can furthermore be verified by adding the translational inhibitor puromycin (10). This induces the premature release of nascent peptide chains and rapidly depletes translation site fluorescence (94, 101, 131, 136, 138) (Figure 1c).

Since 2016, this core technology has led to a steady stream of papers that have shed light on nearly every aspect of translational regulation. In this review, we summarize the progress, focusing on both new technology and new biology. Our hope is the high resolution this technology offers will lead to a better understanding of the full complexity of mRNA translation in vivo.

2. TECHNOLOGICAL ADVANCES IN LIVE-CELL IMAGING OF TRANSLATION

2.1. An expanding palette of intrabodies for multicolor imaging of mRNA translation

To light up single-mRNA translation, a specific intrabody that can quickly and tightly bind peptide epitopes in live cells is required. In 2016 there were two options: the anti-SunTag single chain variable fragment (scFv) conjugated with the fluorescent protein superfolder GFP (116, 138) (Figure 2, upper-left) or purified anti-FLAG or anti-HA fragmented antibodies (Fabs) conjugated with fluorophores like A488 or Cy3 (94). Although Fabs are relatively straightforward to work with using commercial kits and antibodies, they have not been widely adopted. This is because Fabs need to be digested from full-length antibodies, conjugated to fluorophores, and then physically loaded into cells (29, 87). In contrast, the anti-SunTag scFv is encoded by a plasmid that can be transfected into cells prior to imaging. This makes it easy to share and modify (116).

Figure 2. Intrabodies for single-mRNA translation imaging.

Structures of translation imaging intrabodies bound to their target peptides. Structures were predicted using Alphafold2 and color-coded by low to high confidence (red to blue). Target peptides are shown (prolines in ALFA-Tag not shown), along with binding affinity and live-cell FRAP recovery half-time, if known. Tag accessories (solubilization domain GB1 and tested fluorescent fusion tags) are also indicated, but not shown.

To increase the options for multicolor imaging of mRNA translation, several additional systems complementary to the SunTag have recently been developed, as summarized in Figure 2. First, our lab developed anti-HA (139) and anti-FLAG scFv (83) to bind to the classic 9 aa HA tag (hemagglutinin; YPYDVPDYA) (133) and 8 aa FLAG tag (DYKDDDDK) (65), respectively (Figure 2, lower-left). We called these scFv ‘frankenbodies’ because they were created by grafting HA- (135) and FLAG-specific complementarity determining regions onto a stable scFv framework region from a different scFv, one that had previously been demonstrated to fold and function well in live cells (108). The stability of this scaffold facilitates the folding and solubility of frankenbodies in a variety of settings, including mammalian cell lines (83, 126, 139), yeast cells (120), and models organisms like Drosophila (96, 126) and Zebrafish (139).

The anti-HA frankenbody binds the HA epitope with 35 nM affinity in vitro. In live cells this corresponds to a roughly 3 min binding time as approximated by the fluorescence recovery half time after photobleaching (i.e. FRAP recovery half time) (139). For comparison, the SunTag binds its 19 aa alpha-helical epitope (EELLSKNYHLENEVARLKK) with ~20 pM affinity in vitro and for 5–10 min according to FRAP. We have yet to measure the affinity of the anti-FLAG frankenbody in vitro, but it is certainly weaker than the anti-SunTag or anti-HA scFvs because it binds for just ~30 s in live cells by FRAP (83).

Beyond these two new scFv, two camelid nanobodies have also recently been used to image single-mRNA translation in live cells. These are exciting because they are half the size of scFv (~125 aa versus ~250 aa) and thus take up less space (Figure 2, right). The smaller size should help pack epitopes more tightly and minimize biological interference (97), although some of the size advantage is lost after a GB1 solubilization domain (54 aa) and fluorescent fusion tag are added along with linkers.

The first nanobody used to image translation was developed by the Tanenbaum lab and is appropriately called the ‘MoonTag’ because it complements their SunTag (11). The MoonTag was derived from a nanobody that binds a 15 aa peptide from the gp41 subunit of the HIV envelope protein complex (KNEQELLELDKWASL). This sequence adopts a weak α-helical structure that binds to the MoonTag scFv with ~30 nM affinity in vitro (Figure 2, upper-right)

Most recently, the anti-ALFA-Tag nanobody (52) was also demonstrated to work as a translation imaging probe (7). Like the FLAG tag, the ALFA-Tag (SRLEEELRRRLTE) is synthetic and not found in nature. It is 13 aa in length, takes a strong α-helical structure, is neutral, and lacks primary amines (Figure 2, lower-right), all favorable for minimizing off-target effects. The strong structure also allows for multiple contacts between the three complementarity-determining regions of the nanobody and the full length of the ALFA-Tag epitope, resulting in a very high binding affinity of 26 pM despite the relatively small size of the nanobody (52).

With these intrabodies translation can now in principle be imaged in up to five colors at a time. To facilitate this, a general comparison of the Sun, Moon, HA, and ALFA intrabodies has been published (126). Although the study did not focus on translation, it showed each intrabody can label proteins in diverse environments. It also showed they are orthogonal with the exception of slight cross-reactivity between the Sun and ALFA systems. More recently, the Moon and HA intrabodies were compared in live yeast cells (120). This work establishes a foundation for imaging single-mRNA translation dynamics in yeast.

Looking ahead, an important issue to resolve for multicolor applications is the identification of optimal fluorophores. GFP variants are best, with monomeric GFP reducing dimerization (139) and super-folder GFP assisting with solubilization (116). Although many GFP variants worked with the anti-ALFA-Tag nanobody (7), including mNeonGreen (111), mGreenLantern (20), and mAvic1 (77), the new monomeric superfolder GFP (123) worked best. After GFP, HaloTag is best (53, 54), followed by SNAP tag (139), both of which allow for freedom of color choice. In terms of red fluorescent proteins, the options are limited. mRuby2 works well with scFv (76), but mCherry has been hit (139) or miss (83). Ultimately, the choice may depend on the model system. For example, the yeast-optimized green fluorescent protein Envy (113) was best in yeast.

2.2. Designer repeat epitope tags

A diversity of tagging arrangements are possible when working with repeat epitope tags. The most straight-forward is to place epitopes one after the other in a linear sequence, as in the SunTag (116, 138) (Figure 3a, left). Short and flexible GS-rich linkers are used to spread the epitopes out so neighboring epitopes can be bound simultaneously. Flanking prolines can also help distinguish epitopes from nearby structure, as in the ALFA-Tag (7, 52). With flexible linkers in between, repeated epitopes should form a linear 2D string, although interactions between neighboring epitopes is possible. For example, AlphaFold2 (69, 90) predicts many arrangements of the 24 alpha helical SunTag epitopes, sometimes pairing them (Figure 3a, left), other times placing them in random bundles (not shown). It remains unclear what subset of arrangements the epitopes take in live cells.

Figure 3. Tag arrangements for imaging translation.

a) Left, epitopes are commonly arranged one after the other in a linear sequence for translation imaging, as in the 24×SunTag (left box). Right, linear epitopes can alternatively be embedded within structures to help spread them out and create binding hotspots, as in the 10×FLAG spaghetti monster tag (right box). b) Two types of epitopes can be arranged into reporters to compare the translation kinetics of (i) two transcripts, (ii) canonical and non-canonical (IRES) translation, (iii) translation from 0 and ±1 frames, or (iv) barcoded proteins with unique fluorescence fluctuations for multiplexed imaging.

Alternatively, epitopes can be embedded within a scaffold to help spread them out, create binding hotspots, and improve their solubility. This was the premise for the spaghetti monster tags, each of which includes a core structure containing ten repeated epitopes (either HA, FLAG, Myc, V5, OLLAS, or strep II) (94, 128) (Figure 3a, right). The core scaffolds come from diverse lineages of fluorescent proteins, either sfGFP (100), mRuby (76), or mWasabi (3). Cores are typically mutated to a non-fluorescent yet stable ‘dark’ form (4), although this is not required.

Individual epitopes in tags can be combined in interesting ways for multiplexed imaging (Figure 3b). To compare the translation kinetics of different transcripts, each can be tagged with complementary epitopes for co-imaging (7, 11, 30, 94, 139) (Figure 3b, i). Epitopes can also be placed within a single transcript to compare the translation kinetics of, say, two open reading frames in a bicistronic reporter (75) (Figure 3b, ii).

A pair of complementary epitopes can be frameshifted with respect to one another to enable accurate quantification of translation from the 0 and ±1 frames at the same time. The multiframe tag (86) and Moon and Sun Hybrid (MASH tag) tag (11) both utilized this approach, shuffling 0-frame epitopes between ±1-frame epitopes (Figure 3b, iii).

More generally, epitopes can be placed in any sort of pattern to create epitope ‘barcodes’ (Figure 3b, iv). Theoretically these barcodes create unique fluorescent signals through time during the translation process. If the epitopes are placed in distinct positions so their temporal signatures during translation are unique, their fluctuations can theoretically be distinguished, even if both signals use the same fluorophore (104).

2.3. Accessories for single-mRNA translation imaging

Many accessories can be added to single-mRNA translation reporters to assist imaging. The most common accessory is an mRNA tag (Figure 4, right). By tagging mRNA, translation sites can be monitored even when their translation signals are absent or too low to detect. The mRNA tag also makes it possible to calculate the efficiency of translation (the fraction of mRNA that are translating), to check if translation signals arise from single mRNA or from clusters of mRNA, and to determine if translation sites have unique mobilities or subcellular localizations.

Figure 4. Modular accessories for improved single-mRNA translation imaging.

A variety of accessory modules can be added to the core translation imaging tag to, from left to right, induce translation, degrade protein faster, immobilize nascent chains in the ER membrane, or light up transcripts with an mRNA tag. The mRNA tag can also be used to assay mRNA degradation (using the TREAT assay), recruit translation regulatory factors, or immobilize mRNA to the plasma membrane.

There have been many reviews recently of mRNA tags (14, 102, 107, 109, 117, 121), so here we just briefly discuss tags for single-mRNA imaging. The most common are the MS2 (8) and PP7 tags (23) developed in the Singer lab. These tags contain repeated mRNA stem loops bound by coexpressed fluorescent MS2 or PP7 bacteriophage coat proteins. Stem loops are placed in the 3’ untranslated region (UTR) for translation imaging as placement in the 5’UTR can interfere with translation initiation (63, 99). MS2 has improved over the years. Notably, stem loop sequences were randomized in version 5 to make cloning easier (137). In version 6 the tag was redesigned to bind coat proteins with reduced affinity to prevent degradation artifacts (122), especially for overexpressed reporters in yeast (45, 46, 56, 59). Most recently, fusing a translation termination factor to the MS2 coat protein was used to prevent inappropriate nonsense-mediated decay (80).

Beyond stem-loop-based tags, the mRNA Mango aptamer (40) and the cobalamin riboswitch-based riboglow tag (13, 16) have been placed in tandem repeat arrays to amplify fluorescence for single-mRNA tracking in live cells (15, 21, 22). These tags are compact since they only recruit small fluorophores rather than coat proteins. Furthermore, the Mango aptamer is bound by a cell permeable dye that is fluorogenic to reduce background.

Another common accessory is the addition of a localization element for easier tracking. One way to do this is to encode a signal peptide like the 29aa CytERM (33) at the N-terminus. When CytERM is translated, the translating nascent chain is embedded into the ER membrane, where the transcript becomes immobilized (136) (Figure 4, middle). Another approach is to modify the MS2 or PP7 coat proteins so they contain a CAAX sequence (Figure 4, far-right). CAAX is prenylated and targeted to the plasma membrane (44, 57), so newly transcribed mRNA labeled by CAAX-fused coat proteins will also be targeted to the plasma membrane (138). Here, background fluorescence from unbound coat proteins in the plasma membrane can be avoided by combining MS2 and PP7 tags, using one for transcript labeling and the other for membrane localization. This clever combination was recently used to do long-term tracking of single mRNAs, so multiple cycles of bursting translation could be imaged (84).

Complementary stem loops can also be combined in reporters (Figure 4, right). In the ‘TREAT’ assay, a viral pseudo-knot in the 3’UTR is flanked by PP7 and MS2 stem loops (66). The pseudo-knot blocks mRNA degradation by XRN1, so intact mRNAs will have both PP7 coat proteins (PCP) and MS2 coat proteins (MCP) signals while degraded mRNAs will only have MCP. Likewise, we used boxB stem loops (35, 78) to recruit the regulatory factor Argonaut to mRNA and better quantify its impact on translation (30).

Finally, accessory elements have been added to control gene expression. An iron response element can be placed in the 5’UTR so that translation can be turned on or off (Figure 4, far-left) (37). Also, degrons can be added to maintain low levels of protein so translation sites can be easily seen (Figure 4, left). The ornithine decarboxylase (ODC; 467 aa) (131), an auxin-inducible degron (AID; 68 aa) (136), and a destabilized FK506 binding protein (107 aa) (129) have all been used for this purpose. Beyond degrons, the Wu lab recently devised a system to temporally control degradation by modifying the MS2 coat protein so it dimerizes with an mRNA decay factor in a chemically inducible way (9).

3. QUANTIFICATION OF SINGLE-mRNA TRANSLATION IMAGING STUDIES

3.1. Subcellular localization and mobility of translation sites

The direct imaging of single-mRNA translation makes it possible to identify the specific locations where translation preferentially occurs in live cells (Fig 5a-i). This has revealed differences in translation efficiency between the cytoplasm, the endoplasmic reticulum (ER) (129), and other subcellular membraneless condensates (89, 91, 132).

Figure 5. Measuring single mRNA translation dynamics:

a Some measurables of single-mRNA translation reporters: i. translation subcellular location (random vs. nuclear periphery), ii. mobility (motored vs. diffusive), iii. polysome density (from low to high), and iv. translation efficiency (fraction of mRNA with translation signals). b. Fluorescence signals from single-mRNA translation reporters naturally fluctuate up and down through time t as ribosomes initiate and terminate translation. c. The elongation time can be quantified in a variety of ways: i. fluorescence autocorrelation G(τ) of natural fluctuations, ii. treatment with the translation initiation inhibitor harringtonine (leads to steady loss in average fluorescence signal I(t) as ribosomes run off transcripts one by one without further initiation), iii. treatment with the elongation inhibitor cycloheximide (freezes translation, reducing fluctuations), and iv. fluorescence recovery after photobleaching (FRAP).

The mobility of translation sites can also be determined using single-mRNA translation reporters. By fitting diffraction-limited translation sites with a 2D (or 3D) Gaussian function, the coordinates of the translation site at each time point can be measured. Analyzing the displacements of these coordinates over time makes it possible to categorize the mobility of translation sites as diffusive, motored, or bound (Figure 5a-ii) (94, 101, 136). In terms of diffusion, translating mRNAs have generally been observed to be less mobile than non-translating, although the range is broad, from 0.01 – 6 μm²/s (71, 94, 136, 138). For comparison, ribonucleoproteins have cytoplasmic diffusion coefficients ranging from 0.04 μm²/s to 0.4 μm²/s versus freely diffusing GFP ~ 25 um²/s (43, 71, 93, 98, 115). Differences may reflect cell type (98, 139), protein secretion, subcellular location, or subcellular interactions (106, 131, 136), all of which make it difficult to predict translation status from mobility alone (94).

3.2. Ribosome density and translation efficiency

The fitting of the translation site signal to a Gaussian not only provides its spatial coordinates but also its magnitude. As the magnitude is proportional to the number of nascent chains, calibration of the signal makes it possible to estimate the number of ribosomes per mRNA (Figure 5a-iii). Different research groups have consistently measured ribosome densities ranging from one ribosome every 200 to 1000 nucleotides (94, 101, 131, 136, 138).

By labeling mRNAs in conjunction with translation sites, the fraction of translating mRNAs can easily be measured. When combined with ribosome density estimates, this makes it possible to understand what contributes to translation efficiency (75, 86). To illustrate, a 30% translation efficiency can mean 30% of mRNAs are actively translating, all mRNAs translate with 30% productivity (ribosome density), or a combination of both (Figure 5a-iv).

3.3. Translation elongation and initiation rates

Measurements of translation elongation and initiation rates all rely on fluctuations in the fluorescence of translation sites that occur as ribosomes initiate and terminate translation. When initiation is infrequent, such as when a reporter contains the 5’UTR of Emi1, an mRNA will usually only accommodate one ribosome at a time. In such scenarios, the elongation rate can be estimated by observing the speed of intensity ramping or the time period during which the translation signal persists (138). However, an mRNA typically accommodates multiple ribosomes simultaneously (in polysomes), resulting in the convolution of fluorescence signals as ribosomes come and go (Figure 5b). In this case, Fluorescence Correlation Spectroscopy (FCS) can be used to estimate translation elongation rates (41, 95), with the correlation time providing a rough estimate for the elongation time (94) (Figure 5c-i).

Although FCS is powerful, it requires long-term tracking and interpreting the correlation curve is complicated (2, 34). Translation inhibitor experiments have therefore become popular as they enable an easier, more intuitive measurement of ribosome dynamics. The translation initiation inhibitor harringtonine can be used to measure elongation. After addition of harringtonine, only pre-loaded ribosomes continue translation. This leads to a gradual loss in fluorescence as the ribosomes terminate one by one (Figure 5c-ii) (101, 131, 138). As a control, the elongation inhibitor cycloheximide can be used to halt ribosomes and basically lock nascent peptide chains in place. At sufficiently high concentrations, this results in the loss of fluorescence fluctuations (Figure 5c-iii) (84, 94, 131, 138).

As an alternative to inhibitors, ribosome dynamics can also be measured by Fluorescence Recovery after Photobleaching (FRAP). Here a translation site is photobleached and the recovery of fluorescence is monitored over time. Provided the fluorescent intrabodies used to label the site do not themselves unbind during the experiment, the recovery is complete once all bleached nascent chains are replaced by newly-translated chains (Figure 5c-iv). If ribosomes are uniformly distributed, then this occurs at roughly the elongation time (94, 136).

Using these methods, single-mRNA elongation rates have now been measured by many groups independently. Overall the measured rates have been in decent agreement, ranging from 2 to 10 aa/sec (94, 101, 131, 136, 138), consistent with ribosome profiling estimates (~6 aa/sec, (67)). There are some notable exceptions. The Bertrand lab estimated up to 18 aa/sec in mammalian cells using FRAP (101) while the Lagha lab estimated 35 aa/sec in living fly embryos using FCS (41).

With the elongation rate and ribosome density in hand, it is possible to also estimate the initiation rate. To illustrate, if a transcript on average contains 10 ribosomes and elongation takes 100 sec, then one ribosome will terminate every 10 seconds on average. In this case, one ribosome must also initiate every 10 seconds to maintain steady state. Using this simple approach, initiation rates have been estimated and have so far ranged tightly from 2 to 5 ribosomes per minute, even when the studies that measured exceptionally high elongation rates are included (41, 94, 101, 131, 136, 138). It remains unclear what regulatory mechanisms control initiation and elongation rates. A recent study from the Rissland lab suggests codon usage may be linked to initiation (5). Also, differential elongation rates have been measured for β- vs γ-actin, two genes that are nearly identical in amino acid sequence and length, but differ by ~13% in nucleotide sequence (124).

3.4. Computational modeling

Integrating computational modeling with experimental data is arguably the most powerful approach to fitting translation dynamics and obtaining a deeper understanding of the underlying regulatory mechanisms. Computational tools facilitate the extraction of meaningful information from large datasets and enable the modeling of more realistic translation processes such as translational bursting (75, 84, 86) and heterogeneity (11, 64). Of note, RiboFitter from the Tanenbaum lab can fit complicated fluctuations and predict when ribosomes initiate (11). Also, rSNAPsim (RNA Sequence to NAscent Protein simulator) continues to be refined by the Munsky lab so a wide range of single-mRNA translation experiments can be simulated and fitted for arbitrary genes and tag placements (2, 104).

4. BIOLOGICAL APPLICATIONS

The ability to track the translation dynamics of individual mRNA in live cells has led to a wave of biological studies and prompted a re-examination of long-standing models (Figure 6). An emerging theme is the intrinsic heterogeneity of translational regulation which previously went undetected. In the remaining sections, we offer a summary of the new studies and discuss their key findings.

Figure 6. Biological applications of single-mRNA translation reporters in live cells.

Many aspects of translation have now been studied using single-mRNA translation reporters, including: a) the structural organization of single-mRNA during translation, b) local translation in special sub-cellular locations such as the ER, c) the clustering of translation sites in translation factories, d) endogenous translation using genome editing techniques, e) viral translation using increasingly realistic viral translation reporters such as replicons, f) translation quality control which monitors for ribosomal traffic jams, g) translational silencing during stress and the relocalization of mRNA to stress granules (where translation may not always be off), h) nonsense mediated decay and siRNA-mediated translation silencing, and i) miRNA-mediated translational silencing, which may involve recruitment to P-bodies.

4.1. Structural organization of translating mRNA

How mRNA are structurally organized during translation has been difficult to experimentally assess. According to electron microscopy, ribosomes can be arranged in a variety of configurations during translation, from elongated hairpins to more globular spirals (27). To determine if we could see the same heterogeneity in live cells, we measured the distance between the average position of translating ribosomes and the 3’UTRs of mRNAs. This work supported a globular structure (94), suggesting elongated hairpins may be exceptional. Whatever the overall geometry, it was generally assumed, based on the classic closed loop model, that the 3’ and 5’ ends of mRNA were in close proximity during translation (125). Surprisingly, recent experiments utilizing single-molecule fluorescence in situ hybridization (smFISH) demonstrated compaction of mRNAs upon translation inhibition (1, 72), in opposition to the closed-loop model. Further, smFISH with immunofluorescence against the SunTag showed the distance between the 5’ and 3’ ends of mRNA actually spreads out as translation ramps up (1) (Figure 6a). We later confirmed this spreading in live cells using a dual-color bicistronic single-mRNA translation reporter (75). Taken together, these data suggest the closed loop configuration may be a more transient structure than previously thought (73, 125).

4.2. Local translation and mRNA transport

Local translation refers to an mRNA being translated where its protein is needed. Evidence for local translation is well documented, especially in neurons where protein synthesis has been detected in distal dendrites and axons (19, 38, 49, 61, 112). Detection traditionally involves the incorporation of noncanonical amino acid analogs or puromycin into nascent chains, often coupled with a proximity ligation assay (puro-PLA) (39, 55, 85). While informative, these procedures are mostly static, with limited access to dynamics.

Single-mRNA translation reporters have recently been used to more directly image local translation. This has uncovered translation in diverse settings, including the ER (51, 129) (Figure 6b), at early and late endosomes (31), and near mitochondria (103). Somewhat surprisingly, these reporters have also shown puro-PLA does not always pinpoint the site of translation, highlighting the need for complementary methods (42, 62). In this spirit, a dual protein-mRNA localization screen identified mRNA accumulations at various subcellular sites, including centrosomes, the golgi, the nuclear periphery, cell protrusions, and concentrated within foci (26). Foci of dynein mRNAs were shown to be active sites of translation, termed translation factories (101) (Figure 6c), while those of β-catenin turned out to be sites of protein degradation. More recently, stimulus-dependent translation ‘hot spots’ of the Arc protein have been detected in dendrites. Intriguingly, these sites persist across multiple distinct transcription cycles despite the short lifetime of Arc, suggesting a role for local translation in memory consolidation (36). Finally, in new work, single-mRNA translation reporters helped show that co-translational partner binding specificity, and consequential protein destination, may be dictated not only by the subcellular location of mRNA translation, but also by the speed at which translation occurs (48).

Translation dependence for the targeting of mRNAs to specific subcellular compartments is an emerging theme (47). In a recent screen, localized targeting of at least 10 transcripts was translation dependent (26). Other recent demonstrations include transcripts for dlg-1 (118) and erm1 (134) in C-elegans, and the centrosomal protein pericentrin in human and zebrafish (110), whose localization to apical cell adhesion junctions, plasma membrane and centrosomes, respectively, relies on an intact nascent peptide-ribosome complex. Notably, XBP1 transcripts were shown to be targeted to the ER in a translation-dependent manner, where non-canonical splicing occurred (51). One study implies a prominent role for microtubules in the translation-dependent targeting to mitotic centrosomes of both ASPM and Numa (106). ASPM polysomes associated with microtubules also showed a mobility that corresponded to movement of the entire filament. Such ‘hitchhiking’ modes of directed transport may also occur through mRNA-ribosome association with endosomes (31, 60), mitochondria (32, 58) and lysosomes (81). Additional studies should help clarify the origins of these distinct modes of transport.

4.3. Endogenous translation and model organisms

To better understand how translation contributes to physiology, it will be important to image translation from endogenous genes using in vivo models. Ideally, both mRNA and encoded nascent proteins should be labeled for the unambiguous evaluation of translation initiation. The feasibility of tagging endogenous mRNAs has been demonstrated in different organisms including yeast (79), Dictyostelium (28), flies (68), and for an essential gene, ACTB, in mice (82). More recently, RNA tagging has been extended by simultaneously tagging protein with repeat epitopes for translation imaging (Figure 6d). This enabled live imaging of translation in human cells (101), and in Drosophila embryos from a hunchback (hb) transgene reporter (127) and from the endogenous twist (twi) locus (41). Both fly studies reported overall good temporal coordination between transcript and protein expression during development. Local translation was seen in nuclear cycle (nc) 14 embryos, when a reduction in translation for hb mRNAs was observed in anterior versus posterior regions. Interestingly, spatial heterogeneity of translation was also observed for twist. However, the heterogeneity was along the apico-basal axis of nc 14 embryos. Increased basal translation of Twi was associated with basal accumulation of twi mRNA reminiscent of translation factories (101). Altogether, a clear theme that emerges from these pioneering studies is that translation of individual transcripts is highly regulated in a spatial manner. Defining mechanisms that control translation at this level should be an active area of future research.

4.4. Viral translation measurements

An exciting application of single-mRNA translation imaging technology is to the study of viruses. Viruses are generally too small to encode bulky translation machinery, so they have evolved a variety of strategies to hijack host translation during infection. By real-time single-mRNA reporter imaging, researchers are beginning to uncover a surprising degree of heterogeneity during viral infection.

Early applications of single-mRNA translation reporters to virology were focused on specific viral RNA elements. Our lab, for example, showed frameshifting at the HIV-1 frameshift sequence can occur in stochastic bursts of activity that involve relatively long queues of ribosomes (86). We similarly showed translation of the internal ribosomal entry site from the encephalomyocarditis virus occurs in bursts, the frequency of which are upregulated during stress when cap-dependent translation is down regulated (75).

More recently, viral translation reporters have better mimicked the true viral infection process. For example, to better explore the relationship between viral translation and packaging, a virion containing a minimal HIV-1 single-mRNA translation reporter was used to create stable cell lines that were later co-transfected with compensatory plasmids encoding tag-free versions of viral packaging elements (25). This revealed HIV-1 mRNA packaging occurs largely independently from translation, in contrast to models in which the HIV-1 Gag protein autoregulates its own translation. Similarly, virions have now been developed for positive-strand RNA viruses that never pass through the nucleus, such as coxsackievirus B3 (12) (Figure 6e). To prevent interference with the infection process, the virion encoded a minimal 5x SunTag epitope that is cleaved post-translationally. Furthermore, an mRNA tag was avoided altogether. Using this system, the earliest stages of viral replication were imaged in real time, allowing the authors to pinpoint this step as a bottleneck to infection. A subsequent study expanded on this approach by co-imaging anti-viral signaling (17). Given the power and broad applicability of these methods, we anticipate they will become increasingly common in the field of virology.

4.5. Co-translational quality control

During translation, if ribosomes collide or stall, the ribosome-associated quality control pathway is triggered. Although the pathway’s mechanism and key players have been extensively investigated through biochemical assays (70), the dynamics of this process remained unclear. Through use of single-mRNA translation reporters containing a problematic sequence (poly[A]), the Green lab found that the removal of stalled ribosomes from mRNAs occurs at a very slow rate (50) (Figure 6f). This slow rate enabled the differentiation between problematic stalls and transient stalls.

The Nonsense-Mediated Decay (NMD) pathway is a quality control mechanism responsible for cleaving and degrading mRNAs that contain premature termination codons (PTCs). PTCs can result from errors in mRNA splicing or nonsense mutations in the coding region. To study NMD, the model gene Triose phosphate isomerase (TPI) was introduced into a single-mRNA translation reporter. NMD was then monitored via the physical separation of translation and RNA signals (64) (Figure 6h). The study measured the processivity of the exonuclease XRN1 and also showed the efficiency of NMD is influenced by the location of the PTC and its surroundings. Notably, cleavage events were observed only after the first ribosome reached the PTC, and each terminating ribosome had an equal chance of inducing NMD. Recently, similar kinetics were independently measured using an iron-inducible translation plus TREAT reporter, validating the results (37).

4.6. Stress response

Several groups have used single-mRNA translation reporters to visualize translation shutdown and recovery in response to various forms of cellular stress (51, 89, 91, 92, 131, 132). Early work explored the impact of drug treatments on translation shutdown, revealing pretreatment with GSK2606414 (a PERK inhibitor) and ISRIB (an integrated stress response inhibitor) prevents shutdown (131).

A lingering question in the field has been the role of stress granules and P-bodies in stress-induced translation repression. Using reporters harboring β-actin UTRs, we showed translation is shutdown outside of SGs and P-bodies upon oxidative stress (91). This suggests a more passive than active role, with SGs and P-bodies preferentially interacting with already silent mRNA via diverse multivalent interactions (Figure 6g, mRNA 1). A follow-up study further revealed VCP (Valosin-containing protein, a quality control factor) is required for the release of transcripts from stalled ribosomes, enabling their partitioning into SGs (92). Further evidence for a more passive role comes from the Chao lab, who showed the translation and degradation of mRNAs post-stress does not necessarily depend on their localizations in SG or P-bodies prior to stress (132). Further, the Chao lab recently showed some mRNAs can surprisingly be translated in SGs, including transcripts that are upregulated during stress like ATF4 (131), as well as those that are downregulated (particularly those with a 5’TOP motif) (89) (Figure 6g, mRNA 2). While this contradicts the preferential interactions we observed between silent mRNAs and SGs or P-bodies, collectively these new studies call into question the traditional model that SGs and P-bodies play an active, causal role in translation shutdown during stress (6, 88).

4.7. siRNA-guided gene silencing

Small interfering RNAs (siRNAs) are small RNAs that target specific mRNAs for degradation or translation repression. The TREAT assay was used to directly observe siRNA-directed cleavage using an siRNA site between PP7 and PKs (66). This approach allowed for the visualization and characterization of siRNA-guided cleavage events.

The process of siRNA-guided cleavage by Argonaute2 (AGO2) in the context of translation activity has also been evaluated (105). The impact of target site accessibility on mRNA cleavage was investigated by insertion of target sites between epitope tags and the 3’UTR of reporter mRNAs. Cleavage events resulted in the separation of translation and RNA signals, with the majority of cleavages occurring within 10 minutes of translation initiation (Figure 6h). Interestingly, cleavage was delayed by insertion of spacer sequences, suggesting the arrival of translocating ribosomes at the siRNA site stimulated cleavage. The reporter in which target sites were placed just downstream of the stop codon was successfully cleaved in a similar time frame as the one placed just upstream of the stop codon, implying that the unfolding of mRNA higher-order structures is the key process rather than the displacement of RNA binding proteins by ribosomes to open up the AGO2 binding site. Structural rearrangements of AGO2, catalysis, and fragment release were observed to take place in less than 1 minute.

4.8. miRNA-mediated gene silencing

Two studies recently use single-mRNA translation reporters to investigate the role of AGO2 in miRNA-mediated gene silencing (30, 74). Kobayashi and Singer used an array of miR-21 miRNA response elements (MREs) to recruit AGO2 to reporters, while we used BoxB stem loops or miR-26-5p MREs. Both studies found evidence of progressive translational silencing over 30–40 minutes, with miR-21 preferentially targeting mRNAs undergoing strong translation. Post-silencing, mRNAs targeted by miR-21 were decayed, while mRNAs tethered to AGO2 or targeted by miR-25-5p went to P-bodies, where mRNA decay was difficult to detect (Figure 6i). Despite the general agreement of the two studies, it remains unclear why the reporters had different long-term fates. Adding to the complexity, a concurrent study from the Chao lab using an iron-inducible translation plus TREAT reporter showed mRNAs targeted by miR-21 are decayed in a translation-independent manner (37). At the same time, the study showed that translation in general tends to increase mRNA decay, a result that is consistent with their earlier work showing translation inhibition stabilizes mRNA (66). Given the complexity of these kinetics and the discrepancies amongst studies from different groups, future work is needed to better understand the precise relationship between miRNA-mediated gene silencing and mRNA translation and decay. For example, new work from the Wu lab shows mRNA decay in P-bodies is affected by photoxic stress from imaging (9), highlighting the need for additional controls and complementary methods.

5. OUTLOOK AND CONCLUSIONS

The rapid development and diverse applications of single-mRNA translation reporters prove their immense potential. With a plethora of probes and tags to now choose from, we believe the technology is poised for application in a wider range of model systems. For example, low-hanging fruit now within reach is the imaging of single-mRNA translation in yeast, where the necessary tools have already been optimized (120). As well, it will be interesting to see if single-mRNA translation reporters can be adapted for bacteria, where the imaging of tagged ribosomal subunits has been successful (130).

As we look to the future, there is still a need for less invasive tags and probes. Particularly when tagging endogenous genes in model organisms, the potential for interference is high. For example, the study of the endogenous twi locus utilized 32 repeats of the SunTag epitope and 128 MS2 stem loops. It is likely that such large insertions will alter the functionality of some genes, especially when the tags are saturated. A case in point is the tagging of endogenous RNA pol II with 56 SunTag epitopes. Although this enabled translation imaging, the tagged pol II subunit failed to localize in the nucleus (101). The development of smaller, cleavable tags will therefore be beneficial. To help achieve this, the imaging of viral translation presents a lot of opportunities due to the abundance of cleavage sites within viral polypeptides. By placing a cleavage site after a repeat epitope tag, the tag can undergo natural cleavage during viral infection (12). It will be interesting to see how this concept is leveraged in future virology applications.

Whatever the future may hold, single-mRNA translation reporters have certainly come of age and we certainly look forward to seeing where they take us next.