Stochastic and reversible aggregation of mRNA with expanded CUG-triplet repeats

Unlike protein aggregates, which have been extensively studied using biochemical and cytological approaches, little is known about the mechanism of RNA aggregation. More and more RNA aggregates or foci have been identified in different pathologies, such as Myotonic Dystrophy type 1 (DM1) (Davis et al., 1997) and type 2 (DM2) (Liquori et al., 2001), Fragile X-associated tremor/ataxia syndrome (FXTAS) (Tassone et al., 2004), Spinocerebellar ataxia type 8 (SCA8) (Daughters et al., 2009) and Huntington's disease-like 2 (HDL2) (Rudnicki et al., 2007). All these diseases are characterized by microsatellite expansions of CNG or CCTG repeats in specific genes, leading to the accumulation of their transcripts as nuclear RNA foci (Ranum and Cooper, 2006).

The well-studied RNA-mediated disease, Myotonic Dystrophy, is characterized by progressive muscle weakness, muscle wasting and myotonia (Harper, 2001; Ranum and Day, 2004). Myotonic dystrophy type 1 (DM1) is caused by an expansion of CTG trinucleotide repeats in the 3′ untranslated region (UTR) of the protein kinase gene DMPK (Brook et al., 1992; Buxton et al., 1992; Fu et al., 1992; Harley et al., 1992; Mahadevan et al., 1992). This long tract of CUG repeats in the 3′UTR has been shown to cause the retention of this mRNA in the nucleus, where it aggregates as foci (Davis et al., 1997; Taneja et al., 1995). This is thought to be a causative event of DM1. The toxic RNA hypothesis posits that these nuclear foci sequester essential proteins and disrupt their normal function in the cell (Kuyumcu-Martinez and Cooper, 2006; Miller et al., 2000; Nykamp and Swanson, 2004; Ranum and Day, 2004). In support of the toxic RNA hypothesis, a mouse model reproduced key DM1 features by expressing an unrelated mRNA with 250 CUG repeats (Mankodi et al., 2000).

The muscleblind splicing regulatory factor Mbnl1 has been directly implicated in the causation of DM1 because it binds CUG repeats, colocalizes with CUG-repeat RNA foci and regulates the alternative splicing of genes mis-spliced in DM1 patients (Fardaei et al., 2002; Ho et al., 2004; Kino et al., 2009; Mankodi et al., 2001; Osborne et al., 2009). Altered mRNA splicing has been reported in more than 20 genes in DM1 patients (Kalsotra et al., 2008; Lin et al., 2006). In skeletal and heart muscle cells, and neurons of DM1 patients, the recruitment of Mbnl1 into CUG-repeat RNA foci is so extensive that it is depleted from the nucleoplasm (Cardani et al., 2006; Fardaei et al., 2001; Jiang et al., 2004; Mankodi et al., 2005; Mankodi et al., 2003; Miller et al., 2000; Wheeler et al., 2007). In support of the Mbnl1-sequestration hypothesis, disruption of the Mbnl1 gene in mice reproduces both the spliceopathy found in DM1 and the characteristic symptoms of myotonia and myopathy (Kanadia et al., 2003).

The mechanisms that underlie the nuclear retention of mRNA containing CUG-triplet repeats and its aggregation into nuclear foci are not currently well understood. CUG-repeat tracts form hairpins that could affect the processing of the mutant DMPK mRNA and render it incompetent for export. However, it is thought that DMPK mutant mRNA is properly spliced, capped and polyadenylated (Davis et al., 1997; Hamshere et al., 1997). DM1 foci were found to be adjacent to SC35 speckles, suggesting that DMPK mutant mRNA could be blocked from entry into speckles and restricted at this step of nuclear export (Holt et al., 2007; Smith et al., 2007). Mbnl1 has been implicated in the nuclear retention and aggregation of CUG-triplet repeat mRNAs, because depletion of Mbnl1 by RNAi decreases foci accumulation in DM1 myoblast cells (Dansithong et al., 2005). However, it is unclear whether Mbnl1 has a direct role in the aggregation of mRNAs with CUG-triplet repeats.

To explore the mechanism of CUG-repeat RNA foci formation, we directly followed these RNA foci in living myoblast cells using the MS2–GFP system, which tethers fluorescent proteins to transcripts containing MS2 stem-loops (Bertrand et al., 1998; Querido and Chartrand, 2008). With this system, we have successfully tracked the dynamic motion of CUG-repeat transcripts, which show reduced diffusion kinetics compared with transcripts that have five CUG triplets. Our study of Mbnl1 dynamics, colocalization with CUG-repeat transcripts and depletion by RNAi supports the hypothesis that Mbnl1 directly participates in the aggregation of these mRNAs into larger foci. This work demonstrates that CUG-repeat ribonuclear inclusions are labile, constantly forming and disaggregating structures, which display stochastic aggregation and disaggregation behaviour.

The MS2–GFP system is designed to study the dynamic movement and localization of RNAs in living cells (Bertrand et al., 1998; Querido and Chartrand, 2008). A tract of 24 MS2 stem-loops was shown to recruit an average of 33 molecules of MS2–GFP out of a maximum possible number of 48 (the MS2 coat protein binds the MS2 RNA stem-loop as a dimer), and this construction was sufficient to detect single transcripts in living cells (Fusco et al., 2003; Shav-Tal et al., 2004). The dynamics and localization of transcripts containing 24 MS2 repeats have been studied and none have found a tendency of these mRNAs to aggregate or become retained in the nucleus. The reporter mRNA used in this study is described in Fig. 1A, and contains the open reading frame of lacZ, 24 MS2 stem-loop sequences, and the 3′UTR of DMPK containing either five or 145 CUG-triplet repeats (lacZ–MS2–5CUG or lacZ–MS2–145CUG). Although five CUG repeats are insufficient to promote RNA aggregation, 145 CUG repeats are known to promote RNA foci formation in DM1 patients (Botta et al., 2008). We chose this CUG-repeat expansion size so that the transcripts would not all be aggregated and we would be able to detect and track single CUG-rich mRNAs (see below).

Stable murine myoblast C2C12 cell lines were established with each construct; lacZ–MS2 mRNA expression was detected after infection of a retroviral vector expressing MS2–GFP. Large MS2–GFP foci were observed in the nucleus of the lacZ–MS2–145CUG cell line, whereas no nuclear aggregates were visible in parental C2C12 Tet-Off cells or the lacZ–MS2–5CUG cell line (Fig. 1B). Fluorescent in situ hybridization (FISH) with probes against the lacZ sequence was used to confirm the colocalization of the MS2–GFP foci with lacZ mRNA in our model system. As seen by FISH on lacZ–MS2–5CUG cells, the majority of lacZ transcript filled the cytoplasm (Fig. 1B, middle panel). In lacZ–MS2–145CUG cells, the lacZ transcript was mostly retained in the nucleus, forming large aggregates that colocalized perfectly with MS2–GFP foci (Fig. 1B, bottom panel). It is known that nuclear RNA foci observed in cells of DM1 patients range in diameter from 0.2 to 2 μm (Jiang et al., 2004). We observed a similar size range in foci formed with the MS2–GFP system in the lacZ–MS2–145CUG cell line, suggesting that this transcript behaves similarly to the endogenous DMPK mutant mRNA in forming ribonuclear inclusions. As expected, β-galactosidase was more abundant in cells expressing the mostly cytoplasmic lacZ mRNA with 5CUG repeats compared with those expressing the nuclear retained lacZ mRNA with 145 CUG repeats (supplementary material Fig. S1A). The three cell lines were also infected with retroviral vector for GFP–Mbnl1 to confirm that the lacZ mRNA foci colocalized with Mbnl1 in the lacZ–MS2–145CUG cells (supplementary material Fig. S2, bottom panel).

To analyze the dynamics of the CUG-rich RNA foci in living cells, the lacZ–MS2–5CUG and lacZ–MS2–145CUG cell lines were infected with retroviral vector expressing MS2–GFP, and live-cell microscopy was performed at a rate of three images per second using a spinning disk confocal microscope, to visualize the dynamics of MS2–GFP tagged RNA particles (see supplementary material Movies 1 and 2). Table 1 summarizes the results obtained from single particle tracking analysis. lacZ–MS2–5CUG transcripts were observed in the nucleus and their particle velocity was analyzed after classifying the particles according to their type of movement. A previous publication revealed that mRNAs follow three patterns of movement in the nucleus: diffusive, corralled and static (Shav-Tal et al., 2004). Diffusive and corralled particles both move by simple diffusion, but corralled particles are constrained in the distance they reach. Stationary or static particles have extremely confined movements (representative graph in supplementary material Fig. S3). As shown in Table 1, the majority of 5CUG transcripts tracked (58%) showed a diffusive movement pattern, with a mean velocity of 0.44 μm/second. A proportion of 38% of 5CUG particles moved in a corralled fashion, with a mean velocity of 0.36 μm/second, and only 4% of particles were static. The diffusion coefficient (D) of the diffusive 5CUG particles was 0.054 μm²/second and 0.043 μm²/second for corralled particles (Table 1), which is consistent with previously published results of 0.045 μm²/second for CFP–MS2–β-globin transcripts in the nucleus (Shav-Tal et al., 2004).

The 5CUG single mRNA particles we tracked had an apparent diameter of less than 200 nm, and their mean speed and diffusion coefficients were consistent with the single mRNA particles described by Shav-Tal and colleagues (Shav-Tal et al., 2004). Thus we assigned these <200 nm particles as representing single mRNPs. Although cells expressing the lacZ–MS2–145CUG mRNA had large nuclear RNA foci, as shown in Fig. 1B, they also contained single particles of <200 nm diameter with the same fluorescence intensity as 5CUG single mRNA particles. Therefore, we could track CUG-repeat mRNA molecules that were not yet incorporated into large aggregates. These 145CUG transcripts had a mean velocity of 0.29 μm/second for diffusive particles and 0.13 μm/second for corralled particles, which was considerably slower than velocities measured for 5CUG mRNPs. Their diffusion coefficients were also strongly reduced, decreasing to 0.025 μm²/second for diffusive particles and 0.006 μm²/second for single 145CUG corralled mRNAs (Table 1).

The movements of large aggregates of lacZ–MS2–145CUG mRNA, which ranged in size from 200 nm to 1250 nm in diameter, were also tracked. The average velocity of diffusive aggregates was 0.22 μm/second, whereas corralled aggregates had a mean velocity of 0.14 μm/second. It is interesting that the mean velocity of 145CUG particles was very similar, both when single particles or RNA aggregates were measured, indicating that single transcripts already had a reduced diffusion velocity, even when not aggregated with other CUG-repeat mRNAs. A third of the nuclear 145CUG foci were static, because they displayed very little measurable displacement. These data demonstrate that the speed and the diffusive movements of single CUG-rich mRNPs are greatly restricted in the nucleus, compared with that of single 5CUG transcripts or mRNA particles characterized previously (Shav-Tal et al., 2004).

One hypothesis for the role of RNAs with expanded CUG triplets in the etiology of DM1 is that these RNAs sequester splicing factors in nuclear RNA foci, affecting the alternative splicing of several transcripts in DM1 patients (Ranum and Day, 2004). However, it is not yet clear how these RNA foci are formed and whether they constitute irreversible aggregates. To determine whether these RNA foci are stable structures, we performed fluorescence recovery after photobleaching (FRAP) experiments on RNA foci tagged with MS2–GFP. The MS2 coat protein and RNA stem-loop used in this study are variants that interact with extremely high affinity and have a very low exchange rate, so this complex has a half-life of several hours (Boireau et al., 2007). Previous studies have shown that one can measure the dynamics of the MS2-tagged RNA itself, and not the exchange rate of the MS2–GFP molecule on the MS2 stem-loop, by using FRAP (Boireau et al., 2007; Mollet et al., 2008; Shav-Tal et al., 2004). With this system, we were thus able to perform FRAP analysis on the CUG-repeat RNA foci to get a representation of the aggregation of this RNA in living cells.

Three-dimensional time-lapse microscopy was performed on a spinning disk confocal microscope equipped with a photobleaching device at an acquisition rate of two image stacks per minute. After the prebleach series, a 2-μm-diameter region was positioned over the foci and bleached. Recovery of MS2–GFP RNA in the foci was then quantified for 20 minutes. A FRAP experiment on a lacZ–MS2–145CUG foci tagged with MS2–GFP is shown in Fig. 2A (see supplementary material Movie 3). For each photobleached foci, the fluorescence recovery curve was traced. Fig. 2B shows the FRAP curves of five individual MS2–GFP foci of various prebleach diameters. The recovery curves of these different foci were not averaged as usually done with this type of analysis, because they were not consistent. Interestingly, each individual focus behaved differently in terms of the percentage and half-time of fluorescence recovery (Fig. 2B). Indeed, although some foci showed almost no recovery, an individual focus was observed to recover to 350% of its initial value (Fig. 2D). Although there was no consistent result for half-time of recovery from FRAP analysis of MS2–GFP-labelled RNA foci, the median half-time of recovery was 2.7 minutes. Moreover, there was no relationship between the half-time of recovery and the initial prebleach diameter of the MS2–GFP foci (Fig. 2C), and this was also true for the percentage of recovery achieved (Fig. 2D). Altogether, this wide range of values suggests an aggregation mechanism that is stochastic in nature. We also performed fluorescence loss in photobleaching (FLIP) experiments, in which we repeatedly photobleached the free nucleoplasmic lacZ–MS2–145CUG RNA tagged with MS2–GFP, and measured the effect on the fluorescence of MS2–GFP-labelled RNA foci (supplementary material Fig. S4). We observed a decrease in the fluorescence of MS2–GFP-labelled RNA foci, suggesting an exchange between the free nucleoplasmic and aggregated lacZ–MS2–145CUG RNA. Moreover, variability in the rate of fluorescence loss during FLIP confirmed the results obtained by FRAP, corroborating a stochastic aggregation-disaggregation model for CUG-repeat mRNA foci.

Heterogeneity in the FRAP recovery might occur because some of the photobleached MS2–GFP foci correspond to lacZ–MS2– 145CUG RNA transcription sites, which would have different kinetics of recovery compared with that of RNA aggregates. To determine the number of lacZ–MS2–145CUG transcription sites in this C2C12 cell line, DNA FISH was performed using a mixed LNA-DNA probe complementary to the MS2 antisense sequence. With this assay, only one lacZ–MS2–145CUG transcription site was observed in this cell line (supplementary material Fig. S5). Because more than 10 MS2–GFP foci with a diameter larger than 400 nm were observed per nucleus on average, this means that less than 10% of the foci studied by FRAP correspond to transcription sites.

When analyzing time-lapse 3D live cell microscopy films, we were able to observe the spontaneous formation or disappearance of lacZ–MS2–145CUG foci labelled with MS2–GFP. One can observe the formation of such a foci in isosurface view to represent the 3D image (Fig. 3A and supplementary material Movie 4). Fig. 3B displays a graph of corresponding total fluorescence intensity of six foci. Because the acquisitions were performed in 3D, we were able to eliminate the possibility that these foci might simply come from out-of-focus preformed foci. The half-time of spontaneous aggregation events was measured (Fig. 3C) and also showed a large variability, with a median half-time of 2.1 minutes, consistent with the half-times measured for FRAP recovery of MS2–GFP foci. Spontaneous disaggregation of foci was also observed within the time frame of imaging. These disappearances were not due to acquisition photobleaching, because neighbouring foci in the same nucleus remained stable throughout the acquisition time. Time-lapse imaging allowed us to observe that RNA aggregates can split into smaller foci (Fig. 3D), and also coalesce and fuse with other foci (Fig. 3E). Aggregates of CUG-repeat RNA displayed a globular joining and disjoining dynamic similarly to that of a viscous liquid, rather than a solid aggregate. Altogether, these results confirmed that CUG-rich RNA foci are dynamic and follow a stochastic behaviour of aggregation-disaggregation.

The splicing factor Mbnl1 has been shown to have a role in accumulation of CUG-rich mRNA foci in DM1 myoblasts (Dansithong et al., 2005). However, it is not clear whether Mbnl1 directly participates in foci formation or if it has a role in stabilizing RNA aggregates once they have formed. Is Mbnl1 associated with single CUG-rich mRNAs or does it bind only to RNA aggregates? To explore this question, we used our capacity to image single lacZ–MS2–145CUG mRNA particles and investigated whether these single mRNAs colocalized with Mbnl1 protein in living cells. Cells expressing lacZ–MS2–145CUG mRNAs were co-infected with retroviral vectors expressing MS2–GFP and mCherry–Mbnl1, allowing the detection of both Mbnl1 protein and CUG-rich mRNA in vivo (Fig. 4A). To characterize the colocalization of these signals, all the MS2–GFP-labelled lacZ–MS2–145CUG particles were identified and sorted according to particle diameter in each nucleus, and the mCherry–Mbnl1 images were then overlaid to detect the presence of a corresponding Mbnl1 particle in the same location. As shown in Fig. 4B, 100% of MS2–GFP foci that had a diameter larger than 400 nm were found to contain mCherry–Mbnl1. Smaller 145CUG foci that were between 200 and 400 nm in diameter had detectable mCherry–Mbnl1 in 94% of cases, whereas <200-nm-diameter single MS2–GFP particles had Mbnl1 in 73% of cases. To eliminate the possibility that mCherry–Mbnl1 could not be detected in some of the MS2–GFP particles because mCherry is not as bright as EGFP (Shaner et al., 2005), the fluorescent proteins were inverted by tagging MS2 with mCherry, and using a GFP–Mbnl1 retroviral vector (Fig. 4C). Colocalization analysis showed that GFP–Mbnl1 could be detected in 100% of MS2–mCherry foci larger than 200 nm in diameter, and that only 7% of single particles smaller than 200 nm in diameter lacked GFP–Mbnl1 (Fig. 4D). In contrast to the colocalization observed in foci containing lacZ–MS2–145CUG and MS2–GFP, no colocalization with mCherry–Mbnl1 was observed in foci containing lacZ–MS2–5CUG and MS2–GFP (supplementary material Fig. S6). This live-cell analysis reveals the association of Mbnl1 with single nucleoplasmic CUG-repeat mRNA and not only with RNA aggregates.

To study the role of Mbnl1 in aggregation of CUG-repeat RNA, we also looked at Mbnl1 dynamics using FRAP. The C2C12 cell line expressing lacZ–MS2–145CUG RNA was infected with retroviral vector encoding GFP–Mbnl1. To ensure that we did not overexpress GFP–Mbnl1, we compared the rate of photobleaching of GFP–Mbnl1 and MS2–GFP in the nucleoplasm in regions distinct from CUG-repeat foci. Unlike the free, fast diffusive MS2–GFP protein, which was quickly photobleached, GFP–Mbnl1 photobleached less rapidly, suggesting that most GFP–Mbnl1 in the nucleus was associated with proteins or transcripts (either endogenous or 145CUG RNAs) (supplementary material Fig. S7). Photobleaching was performed on lacZ–MS2–145CUG RNA foci tagged with GFP–Mbnl1 with the same 2-μm-diameter bleach region used on the MS2–GFP foci (Fig. 5A). For GFP–Mbnl1 FRAP, 20 image stacks per minute were acquired and fluorescence recovery was measured for 5 minutes (see supplementary material Movie 5). Fig. 5B shows relative fluorescence recovery curves for five GFP–Mbnl1 foci with various prebleach diameters. Interestingly, similarly to the MS2–GFP curves, they could not be averaged, as they diverged too widely between individual foci. The median half-time for GFP–Mbnl1 fluorescence recovery was faster than the recovery of MS2–GFP foci (1.2 minutes vs 2.7 minutes) and the half-times of recovery varied less widely than they did with MS2–GFP FRAP (Fig. 5C). The percentage of recovery of GFP–Mbnl1 foci was charted in Fig. 5D, showing that 60% of foci had 0–50% fluorescence recovery after photobleaching and 40% had more than 50% recovery. A very similar distribution was observed with MS2–GFP FRAP. As in the case of MS2–GFP-tagged RNA foci, there was no consistent correlation between half-time of recovery and prebleach diameter of the GFP–Mbnl1 foci (Fig. 5E). Altogether, these results show that association of Mbnl1 with the CUG-rich RNA foci follows a stochastic process, as observed with lacZ–MS2–145CUG RNA.

To further explore the role of Mbnl1 in CUG-rich RNA foci formation, we designed a short hairpin RNA interference (shRNA) retroviral vector targeting Mbnl1. The shMbnl1 vector achieved efficient depletion of Mbnl1 protein in cells, whereas the control shMbnl1mut vector containing two point mutations in the shRNA did not change Mbnl1 levels (Fig. 6A). To determine the effect of Mbnl1 depletion on CUG-rich RNA foci formation, a FISH experiment with a Cy3-labelled (CAG)₁₀ probe was performed on lacZ–MS2–145CUG cells stably expressing shMbnl1 or control vector. Using quantitative fluorescence microscopy, the nuclear surface area covered by RNA foci was measured, which revealed a decrease of 2.5 times the median value of the surface area occupied by RNA foci between Mbnl1-depleted cells versus control cells (Fig. 6B). Measurement of fluorescence intensity of the RNA foci instead of their surface area also resulted in a two times decrease in RNA foci accumulation in Mbnl1-depleted cells versus control cells (data not shown). Compared with control cells, shMbnl1 cells displayed higher β-galactosidase activity without a change in lacZ–MS2–145CUG mRNA levels, suggesting an increased cytoplasmic accumulation and translation of this transcript (supplementary material Fig. S8). To determine whether depletion of Mbnl1 affects aggregation of foci containing CUG-rich RNA, FRAP analysis was performed on MS2–GFP-labelled RNA foci in Mbnl1-depleted cells. As a positive control for the inhibition of CUG-rich RNA foci formation, cells were treated with the RNA polymerase II inhibitor α-amanitin. Depletion of another splicing factor, hnRNP A1, by shRNA was also performed to control for the specificity of Mbnl1 knockdown in the RNA aggregation process (Fig. 6A). The recovery of 145CUG RNA foci after FRAP was reduced in Mbnl1-depleted cells, because the median half-time of recovery was two times lower, indicating that aggregation time was shorter than in control cells (Fig. 6C). Moreover, the percentage of foci recovery after photobleaching was altered towards less recovery in Mbnl1-depleted cells compared with control cells (Fig. 6D). Cells treated with α-amanitin had less lacZ–MS2–145CUG mRNA because transcription was inhibited before FRAP analysis, and they displayed even lower recovery of foci after FRAP than Mbnl1-depleted cells. In both Mbnl1-knockdown and α-amanitin-treated cells, RNA aggregation after FRAP did not last as long as it did in WT cells, leading to shorter half-times of recovery. A mutated shRNA against Mbnl1 or depletion of hnRNP A1 had no effect on half-time of recovery and percentage of foci recovery (Fig. 6C,D). These results strongly suggest that the nuclear retention and foci formation of CUG-rich mRNA is not simply a self-aggregation of the RNA, but is based, in part, on a protein machinery that involves the Mbnl1 protein.

Unlike protein aggregates, RNA aggregates have been poorly studied, despite the fact that they are now implicated in several pathologies (DM1, DM2, HDL2, SCA8, FXTAS) (Ranum and Cooper, 2006). Thus, the question of the role of RNA aggregates in the etiology of these diseases is very relevant. The current model posits that these aggregates can sequester RNA-binding proteins and disrupt their normal biological functions (Ranum and Day, 2004). Because most studies on these RNA foci are based on fluorescent in situ hybridization in fixed cells, it is still unclear whether the foci behave as insoluble, static aggregates or whether they are dynamic structures.

To answer this question, we used the MS2–GFP system to follow the dynamics and aggregation of mRNAs with expanded CUG triplets in living cells. Using single-particle tracking, we observed that even single non-aggregated mRNPs with 145CUG triplets are considerably slowed in their diffusive movement compared with mRNPs with 5CUG. Indeed, single mRNPs containing 145 CUG-triplet repeats had an average diffusion coefficient of 0.025 μm²/second for diffusive single transcripts, and 0.014 μm²/second for diffusive aggregated mRNPs, compared with 0.054 μm²/seconds for diffusive single transcripts containing only 5CUG. This is probably not due to the increased size of the mRNA caused by the addition of 140 CUG-triplets (420 nucleotides), because this represents an increase in size of only 6% for this long lacZ transcript. According to the Stokes–Einstein law of diffusion, the decreased diffusion coefficient might reflect a change in the radius or shape of the mRNP (i.e. the CUG triplets adopt long stem-loop structures) and/or the mass of the mRNP has increased (we show that Mbnl1 binds single 145CUG transcripts and not the 5CUG). This reduced diffusion of the CUG-triplet repeat mRNPs might provide an explanation for the failure of the majority of these mutant transcripts to reach the nuclear pores and be exported to the cytoplasm.

Using FRAP and FLIP, we showed that the MS2–GFP-labelled CUG-rich mRNAs form dynamic aggregates, which does not support the hypothesis that these foci are insoluble and static. The FRAP and FLIP experiments also revealed that fluorescence recovery and loss of the various RNA foci vary and do not follow a predictable pattern, suggesting that the aggregation process of these RNA foci is stochastic. Surprisingly, we observed spontaneous aggregation-disaggregation events during 4D imaging, which shows that these RNA foci are unstable. The observation of spontaneous aggregation, suggesting the presence of a lag phase before aggregation, and the stochasticity in the aggregation process, support a nucleation model in which the formation of a stable aggregate nucleus would be the rate-limiting step. Interestingly, similar kinetics has been reported for the aggregation of huntingtin and β-amyloid peptide (Colby et al., 2006; Hortchansky et al., 2005).

We used FRAP to study the dynamics of Mbnl1 recruitment to foci containing CUG-rich RNA. The recovery that can be observed in the GFP–Mbnl1 foci might be due to three possible mechanisms: (1) exchange of free nucleoplasmic GFP–Mbnl1 with the GFP–Mbnl1 in CUG-rich RNA foci; (2) aggregation of single transcripts of 145CUG RNA bound by GFP–Mbnl1; and (3) aggregation with neighbouring foci containing GFP–Mbnl1. The last two mechanisms are stochastic in nature, as we discussed above. Because (1) the FLIP experiment shows a direct exchange between CUG-rich RNA foci and the free nucleoplasmic 145CUG RNA, (2) these nucleoplasmic 145CUG RNA are associated with Mbnl1, and (3) similarly to the MS2–GFP-tagged RNA foci, GFP–Mbnl1 FRAP recoveries were stochastic, these results are consistent with the hypothesis that a large portion of the GFP–Mbnl1 signal recovers as the protein is carried along and aggregates with mRNAs into foci in a stochastic process.

A previous study in which FRAP was performed on GFP–Mbnl1 foci in cells expressing a CUG-triplet repeat transgene and in DM1 cells reported very different results from ours (Ho et al., 2005). The authors measured a half-time of recovery of 3.5 seconds for GFP–Mbnl1 foci, in contrast to the median half-time of recovery of 1.2 minutes that we observed. In addition, the recovery curves and measured percentages of recovery reported by Ho and colleagues were very consistent between foci, whereas we observed a heterogeneous recovery of GFP–Mbnl1 foci. The differing results might be explained by the level of expression of GFP–Mbnl1 in our two systems. Using FRAP on free GFP–Mbnl1 in the nucleoplasm, Ho and co-workers measured a fast recovery (~3 seconds), suggesting the presence of a considerable amount of free diffusive GFP–Mbnl1. In our study, we expressed low levels of GFP–Mbnl1 compared with endogenous Mbnl1 in the C2C12 myoblasts (supplementary material Fig. S7). We also used FLIP to measure the rate of GFP–Mbnl1 photobleaching in the nucleoplasm and observed little free diffusing GFP–Mbnl1 (supplementary material Fig. S7), suggesting that most of it was associated with low-diffusing mRNPs or proteins. Therefore, the data from Ho and colleagues might reflect the fast exchange between unbound GFP–Mbnl1 in the nucleoplasm and GFP–Mbnl1 bound to RNA aggregates. Our FRAP data are not consistent with that model, but better fit with the hypothesis that recovery of GFP–Mbnl1 signal occurs as the protein is carried with newly aggregating CUG-rich mRNAs. However, the faster half-times of recovery measured with GFP–Mbnl1 compared with MS2–GFP (median of 1.2 minutes versus 2.7 minutes, respectively) suggest that some exchange of free GFP–Mbnl1 also participates in the recovery.

The depletion of Mbnl1 had previously been shown to affect the accumulation of CUG-rich RNA foci in the nucleus (Dansithong et al., 2005). However, this study looked at RNA foci accumulation by FISH and did not assess the role of Mbnl1 in the RNA aggregation process. We observed that the depletion of Mbnl1 decreased by 2.5 times the accumulation of CUG-repeat RNA foci, as measured by the quantification of nuclear lacZ–MS2–145CUG RNA foci by FISH. The observations that Mbnl1 was already associated with single 145CUG transcripts in the nucleus and that the depletion of Mbnl1 reduced the recovery of CUG-rich RNA foci after FRAP suggest a direct role of Mbnl1 in RNA aggregation. Mbnl1 oligomerization on CUG stems might directly cause RNA aggregation (Yuan et al., 2007). However, this role is not essential because the depletion of Mbnl1 causes only a partial defect in CUG-rich RNA foci formation. Therefore, other proteins might also be involved in this process. For instance, the Mbnl1 paralogues Mbnl2 and Mbnl3 are expressed in C2C12 myoblasts (Holt et al., 2009), and they might, in part, be complementing the function of Mbnl1 in RNA aggregation (Fardaei et al., 2002; Miller et al., 2000). Other RNA-binding proteins, such as hnRNP H, have been shown to be recruited to DM1 foci, and might also contribute to the aggregation of CUG-rich RNA (Kim et al., 2005; Paul et al., 2006).

Our data support a model in which Mbnl1 is directly involved in the aggregation of CUG-rich mRNA transcripts into large nuclear foci. In DM2, HDL2, SCA8 and FXTAS RNA gain-of-function pathologies, Mbnl1 has been found to colocalize with the nuclear RNA foci (Daughters et al., 2009; Iwahashi et al., 2006; Mankodi et al., 2001; Tassone et al., 2004). In SCA3, Mbnl1 has been shown to increase CAG-repeat RNA toxicity in a Drosophila model for the disease (Li et al., 2008). It will be interesting to investigate whether Mbnl1 also has a direct role in RNA aggregation in these diseases, similarly to its role in DM1.

This work demonstrates that nuclear aggregates of CUG-repeat RNA are labile, constantly forming and disaggregating structures. This strongly suggests that it will soon be possible to inhibit or reverse (CUG)n RNA foci formation in patients. Indeed, recent studies used (CAG)₂₅ oligos in mouse models of myotonic dystrophy to cause the disaggregation of RNA foci and achieved reversal of symptoms (Mulders et al., 2009; Wheeler et al., 2009). A thorough understanding of the aggregation mechanism of toxic RNA should help achieve effective therapies for DM1 and other RNA gain-of-function diseases.

The pRevTRE-lacZ–MS2–145CUG retroviral vector was constructed by subcloning a fragment of the plasmid pR26EGFP+200 (Amack and Mahadevan, 2001) containing EGFP and the DMPK 3′UTR into the pRev TRE plasmid (Clontech Laboratories, Mountain View, CA). The EGFP sequence was replaced with the coding sequence of lacZ followed by 24 MS2 stem-loops from the RSV-Z–MS2–24 plasmid (Fusco et al., 2003). The MS2–GFP fusion protein without an NLS was sub-cloned into the pBabe puro retroviral vector (Morgenstern and Land, 1990). The MS2–mCherry fusion protein was constructed by cloning the mCherry sequence into the pcDNA3-HA–MS2–linker plasmid. The MS2 coat protein used is the V29I-dIFG double mutant, with increased affinity for the MS2 stem-loop (Lim and Peabody, 1994). The GFP–Mbnl1 fusion protein was subcloned from the pDEST53–GFP–Mbnl1-42kDa plasmid (Dhaenens et al., 2008) into the pBabe puro vector. To make the mCherry–Mbnl1 fusion protein, a linker sequence (GGGGSGGGGS) was inserted between mCherry and the Mbnl1 sequence from pDEST53–GFP–Mbnl1-42kDa.

Cell lines expressing the lacZ–MS2–5CUG or 145CUG transcripts were generated by infecting C2C12 Tet-Off cells with the desired pRevTRE plasmid and selecting for stable clones. The cell line 145–8, which expresses high levels of lacZ–MS2–145CUG mRNA (supplementary material Fig. S1), was used throughout this study. All the experiments were performed 48 hours after removal of doxycycline to fully induce transcription from the Tet-response element (TRE) promoter. For live-cell microscopy, cells were plated on 35 mm glass-bottom plates (Fluorodish, World Precision Instruments) and transiently infected with retroviral vector(s) expressing the desired fluorescent fusion protein(s). The spinning disk confocal microscopes were equipped with environment chambers for the control of CO₂ concentration, temperature (chamber and objective heater) and humidity, adjusted to maintain a reading of 37°C in the cell medium. The α-amanitin treatment was done by adding 100 μg/ml of drug to the medium. Live-cell experiments were performed from 1.5 to 3 hours after adding α-amanitin (A2263 Sigma).

Time-lapse microscopy was performed at McGill University on the Cell Imaging and Analysis Network (CIAN) SD2 spinning disk confocal microscope assembled by Quorum Technologies (Guelph, Ontario, Canada) on a Leica DMI6000B platform. Images were acquired using a 100× oil objective NA 1.40 and the Hamamatsu C9100-12 EMCCD camera (Hamamatsu, Higashi Japan). Exposure time was set at 333 ms at maximum speed and a single confocal image was acquired per time point. Tracking was done with the MetaMorph object-tracking module, using the Threshold Result algorithm, which detects the center of the intensity peak for each object and determines which object is closest in the preceding image. Tracks were visually inspected to ensure that the particle could be not be confused with neighbouring particles. The mean square displacement (MSD) for each tracked particle was calculated with the formula: Distance to origin (DTO) is the straight-line distance between the particle's current position and the first point in the track, calculated by the MetaMorph object tracking module. The diffusion coefficient (D) for each particle was then derived by measuring the average slope of the MSD (μm²) over time. To determine particle diameters, four measures of the diameter were taken at different angles and averaged.

The lacZ–MS2–5CUG or 145CUG cells were co-infected with retrovirus for MS2–GFP and mCherry–Mbnl1, or with MS2–mCherry and GFP–Mbnl1 to reverse the fluorophores. Live cells were visualized by microscopy on a spinning disk confocal microscope (either the CIAN SD2 or Perkin-Elmer UltraView Vox). GFP and mCherry images were acquired sequentially, and the images from the two channels were then overlaid for analysis. The absence of bleed-through between GFP and mCherry was verified by visualizing cells infected with only one of the fluorescent fusion proteins. The RNA particles were first identified in the image of the MS2 fluorophore, their diameter was measured and the MS2 RNA particles were then scored for the detection of an overlapping peak in the Mbnl1 signal by applying a linescan through both red and green channels with the MetaMorph software. Additional particles could be observed in the Mbnl1 channel, which did not colocalize with any MS2 signal. These were not included in the analysis because they could represent endogenous transcripts bound by Mbnl1.

3D time-lapse confocal microscopy was performed on an UltraView Vox spinning disk confocal microscope equipped with a PhotoKinesis device for FRAP (Perkin Elmer, Massachusetts USA). Images were acquired using a 100× NA 1.40 oil objective and the Hamamatsu C9100-50 EMCCD camera. Exposure time was set at 50 ms and a Z-stack sufficient to cover all foci in the nucleus was acquired for each cell, with a slice thickness of 0.25 μm. Two image stacks were acquired per minute. To minimize phototoxicity and photobleaching, the camera was set to a bin of two and the 488 nm laser power was reduced to 8% of maximum. After a prebleach series, a 2-μm-diameter bleach area was positioned over the chosen foci and bleached for 20 iterations at 100% laser power. Data acquisition was continued for 20 minutes to measure recovery. For fluorescence loss in photobleaching (FLIP), fields were chosen where two adjacent cells containing MS2–GFP RNA foci were visible. Exposure time was kept at 50 ms, but time intervals were set to 3 seconds. The 2-μm-diameter bleaching area was positioned in a region of the nucleus that did not contain foci. A prebleach series was obtained and then, every 3 seconds, a new image stack was acquired, followed by 20 iterations of bleaching at 100% laser power in the bleach area. Total duration of data acquisition for FLIP was 5 minutes. For the observation of the CUG-repeat mRNA foci with the GFP–Mbnl1 protein, exposure time was set at 100 ms and 488 nm laser power was boosted to 12%. Time intervals were set to 3 seconds, and the 2 μm circular bleach area was positioned on the chosen GFP–Mbnl1 foci immediately before bleaching. GFP signal in neighbouring cells was monitored to control for incidental photobleaching during image acquisition; no decrease in fluorescence intensity was observed with these settings.

The resulting films were analyzed using the MetaMorph program. The 3D images were thresholded to remove signal from MS2–GFP or GFP–Mbnl1 molecules not sequestered in CUG-repeat RNA aggregates. An isosurface rendering of the RNA foci was performed to define them as singular objects, and the total fluorescence intensity was obtained for each object at every time point. FRAP recovery curves were analyzed with the GraphPad Prism program (GraphPad Software, La Jolla, CA). The majority of the FRAP recovery curves fit the one-phase exponential association equation, indicating a single population of RNA. The recovery curves that did not fit this equation occurred when there was no recovery, and their half-time and % recovery were set as zero. The percentage of fluorescence recovery and the half-times of recovery were calculated by the GraphPad Prism software. For MS2–GFP foci analyzed by FLIP, 18 out of 22 foci fit a one-phase exponential decay equation. The half-lives of the foci were calculated using GraphPad Prism software.

Fluorescent in situ hybridization (FISH) on RNA was performed according to the protocol described previously (Querido and Chartrand, 2008). Briefly, cells were fixed in PBS containing 4% paraformaldehyde, dehydrated for 2 hours in 70% ethanol, and incubated overnight with 10 ng of probe in 2×SSC, 40% formamide at 37°C. To visualise the localization of lacZ mRNA, five Cy3-labelled probes designed to hybridize in different regions of the lacZ coding sequence were used, as described previously (Long et al., 1995). To visualize foci of CUG-triplet-repeat mRNA, a single probe with the sequence (CAG)₁₀ was labelled with Cy3 or Cy5 and used for FISH, as originally described (Taneja et al., 1995). A control Cy3-labelled probe with the sequence (CTG)₁₀ gave no hybridization signal in cells expressing lacZ–MS2–145CUG mRNA (data not shown). For simultaneous DNA and RNA FISH, cells were fixed and dehydrated as above. Chromatin was denatured by treatment with 70% formamide, 2×SSC at 70°C for 10 minutes, followed by two washes in ice-cold 2×SSC. Hybridization with 50 ng of LNA-DNA probe complementary to the MS2 antisense sequence and 10 ng (CAG)₁₀ probe was performed overnight at 37°C in 50% formamide, 2×SSC, 10% dextran sulfate, 1 mg/ml BSA, 20 mM VRC. Three washes were performed with 50% formamide, 2×SSC at 42°C for 5 minutes each, followed by three washes with 2×SSC at 42°C for 5 minutes each. Images for FISH were acquired on a transmitted light Nikon TE2000U microscope with 100× oil objective of 1.40 NA and the CoolSNAPfx CCD camera (Photometrics, Tucson, AZ) using the MetaMorph software (Molecular Devices). To quantify the CUG repeat RNA foci visualized by FISH, a MetaMorph script was designed to first outline the nuclei using the DAPI-stained images, then threshold the CUG-repeat RNA foci in the nuclei to define them as objects, and finally measure and sum the total surface area covered by RNA foci in each nucleus or the total fluorescence intensity of the nuclear RNA foci.

We thank Mani S. Mahadevan (University of Virginia, VA) for providing the DMPK 3′UTR plasmids, Nicolas Charlet-Berguerand (Université de Strasbourg, France) for GFP–Mbnl1 plasmids, Stephen W. Michnick (Université de Montréal, Canada) for the mCherry plasmid, Charles Thornton (University of Rochester, NY) and Glenn E. Morris (Wolfson Centre for Inherited Neuromuscular Disease, UK) for Mbnl1 antibodies, and Benoit Chabot (Université de Sherbrooke, Canada) for hnRNP A1 antibody. We are grateful to P. Lapointe from Perkin-Elmer Canada and G. Hickson for access to the UltraView Vox microscope. We thank J. Lacoste from the Cell imaging and analysis network (CIAN) at McGill University for helpful discussions. We are also thankful to M. Vasseur of the Université de Montréal for expert technical assistance. E.Q. was supported by a development grant from the Muscular Dystrophy Association USA (MDA). F.G. was supported by a fellowship from the Terry Fox foundation of the National Cancer Institute of Canada. This research was funded by grants from the Canadian Institute of Health Research (NDS62501) to P.C., and the MDA (E.Q.). P.C. is a Senior Scholar from the Fonds de la Recherche en Santé du Québec.