A substantial fraction of nascent RNAs are elongating

Recovery of Exo1 transcription sites showed that the synthesis of new RNAs occurred in <3 min and that virtually no RNAs were retained at the transcription site for a long time (immobile fraction of 3%; Fig. 2 D). A second clone expressing the same MS2-tagged RNA yielded similar kinetics (Exo2; Fig. 2 E), indicating that this was principally a property of the reporter.

RNA biogenesis occurs in a series of steps: transcription, 3′-end formation (cleavage and polyadenylation), and release from the transcription site (Fig. 2 F). To test whether elongation was rate limiting, we used a slow version of RNAPII, the hC4 mutant (de la Mata et al., 2003). This α-amanitin–resistant mutant was transfected in Exo1 cells, endogenous enzymes were inactivated with α-amanitin, and remaining transcription sites were analyzed by FRAP. As a control, we used another α-amanitin–resistant version of RNAPII that has no elongation defect (wild type [WT^res]; de la Mata et al., 2003). FRAP curves obtained with WT^res or the endogenous polymerase were nearly identical (Fig. 3). In contrast, recoveries were markedly slower with the hC4 mutant: 129 s for half-recovery versus 75 s for the WT^res enzyme. To confirm that transcriptional elongation took a substantial part of the time that nascent RNAs spent at their transcription sites, we used camptothecin. This drug targets topoisomerase I, and it transiently cross-links it to DNA. A previous study has shown that this creates steric road blocks for polymerases, which result in increased pausing and a net slowdown of transcriptional elongation (Capranico et al., 2007). Indeed, when cells were treated with this drug, the recovery of nascent RNAs was slower: 170 s for half-recovery in Exo1 cells compared with 65 s in the case of untreated cells (Fig. 4 A).

Open in a separate window

Figure 3.

Effect of mutants of RNAPII on the synthesis of nascent RNAs. Exo1 cells were transfected with vectors expressing mutant forms of RNAPII, Tat, and MS2-GFP and were treated with α-amanitin before FRAP. (top left) Recovery curves of the endogenous and transfected WT^RES RNAPII (orange and blue curves, respectively). (top right) Comparison between the WT^RES (blue), hC4 (red), and ΔCTD (green) RNAPII. (bottom) Fit of the two-step model (straight line followed an exponential) with WT^RES (left) and hC4 curves (right). Error bars represent SD.

Open in a separate window

Figure 4.

Elongation is a limiting step in the recovery of nascent RNAs. (A) Recoveries of MS2-GFP in Exo1 cells treated with camptothecin. Cells were transfected with vectors expressing Tat and MS2-GFP, treated or untreated with camptothecin, and analyzed by FRAP. (left) Comparison of treated (green) with untreated (blue) cells. (right) Treated cells were fitted with the two-step model. Pink, straight line; green, exponential component. (B) Comparison of the MS2-GFP recovery curves for the short and long HIV-1 reporters. ExoLong cells were transfected with vectors expressing Tat and MS2-GFP and were analyzed by FRAP. (top left) Comparison of the recovery curves of Exo1 cells (blue), Exo2 (red), and ExoLong (green). (top right) Fit of the long reporter with the two-step model. (bottom left) Comparison of Exo1 and ExoLong recovery curves after normalization by the initial slope (initiation rate). (bottom right) Comparison of experimental (blue) and computer-simulated (green) FRAP curves for the long reporter. (C) Recoveries of MS2-GFP in HIV-1 reporters lacking U3. pTRIP_1_13 cells expressing the ΔU3 reporter (pTRIP) were transfected as described in A and analyzed by FRAP. Error bars represent SD.

To further establish that elongation was rate limiting, we developed a cell line (ExoLong) that integrated a second reporter, pExo-MS2 × 24-Long, which differed by 2.4 kb in the length of the sequence separating the MS2 repeat and the end of the gene. If elongation is rate limiting, the recovery should be slower for the long reporter, and the difference should represent the time taken to transcribe the additional sequences. As expected, transcription sites of ExoLong recovered more slowly: 126 s for half-recovery versus 65 s for the short reporter (Fig. 4 B). Altogether, these observations demonstrated that elongation took a substantial fraction of the time dedicated to mRNA production.

A two-step model comprising elongation and 3′-end processing can describe recoveries of nascent RNAs

In kinetic terms, transcription elongation is the repetition of an elementary step: the addition of one nucleotide. The stochastic basis of the process implies that the time taken to add a single nucleotide is variable and follows an exponential described by the rate constant k_el. In contrast, when polymerases transcribe a large number of nucleotides, the repetition of the elementary step creates a statistical averaging such that the time taken to synthesize n nucleotides will be virtually constant for all polymerases and equal to n/k_el. This feature of transcription elongation is distinctive, and it predicts a linear increase in signal during FRAP recovery, whereas single-step processes should result in exponentials (Fig. S3 B, available at http://www.jcb.org/cgi/content/full/jcb.200706018/DC1).

Elongation is the first step of mRNA biogenesis that can be observed in our FRAP experiments. Indeed, once the polymerase reaches the polyA site, the pre-mRNA is cleaved, polyadenylated, and released. These processes occur as a series of single steps and, thus, may be modeled as exponentials. We attempted to fit the FRAP curve with two components: a straight line for elongation followed by a single exponential for 3′-end formation and release. This two-step model fitted experimental data substantially better than a single exponential (Figs. 2, G and H; and 3). The short reporter gave 65 to 73 s for elongation and 44 to 31 s for the half-time of the exponential in Exo1 and Exo2 cells, respectively (Table I). The long reporter yielded 148 s for elongation and 65 s for the exponential. This translated into similar elongation rates of 1.79–2.03 kb/min. Cells treated with camptothecin or transfected with the slow mutant of RNAPII could also be fitted to this two-step model (Figs. 3 and 4 A). For the hC4 mutant, it yielded an elongation rate more than twofold slower than with WT RNAPII (0.8 kb/min), which is in agreement with in vitro data (Coulter and Greenleaf, 1985).

The proportion of time dedicated to elongation versus 3′-end processing can also be estimated by direct comparison of the short and long reporters. Indeed, if initiation rates are equal, the intensity of the MS2 signal at various transcription sites should be proportional to the time that the RNA spends there. When the curves of the short and long reporters were normalized to their initiation rate, that is, to their initial slopes, the long reporter accumulated 1.5 times more RNAs at the transcription site (Fig. 4 B). If one assumes that the time required for elongation is proportional to the RNA length, whereas other steps are identical for the two reporters, this value corresponds to half of the time spent on elongation for the short reporter and 71% for the long one (see Materials and methods). This was in agreement with the values obtained by fitting the FRAP curves with the two-step model (Table I).

Estimation of the relative rates of elongation and 3′-end formation by quantitative in situ hybridization

Next, we performed quantitative in situ hybridization with oligonucleotide probes that hybridized along the gene. Incompletely transcribed RNAs should yield more signals with probes hybridizing at their 5′ end, whereas full-length RNAs should yield equal signals with 5′ and 3′ probes (Femino et al., 1998). Because incompletely transcribed RNAs correspond to elongating molecules, whereas full-length RNAs are at the stage of 3′-end processing or transcript release, the ratio of 5′ to 3′ probes can be used to estimate the relative time taken by elongation versus 3′-end processing and release. Four sets of Cy3-labeled probes were used (Fig. 5): the first hybridized in exon 1 (E1), the second in the intron (I), the third at the splice acceptor site (I-E2), and the last in exon 2 (E2), immediately before the active polyA site. When signals were normalized with a Cy5-labeled probe against the MS2 repeat, 5′ probes gave 2.2-fold more signal than 3′ probes (Fig. 5). To confirm this, we hybridized E1-Cy5 and E2-Cy3 probes simultaneously, and signals at the transcription sites were normalized to signals in the cytoplasm. Again, we found twofold more E1 probe at transcription sites (Fig. S4, available at http://www.jcb.org/cgi/content/full/jcb.200706018/DC1). From these values, we could estimate that half of the polymerases that have reached the MS2 repeat are elongating toward the polyA site, whereas the other half are at the stage of 3′-end formation (see Materials and methods), which is in agreement with FRAP experiments.

Open in a separate window

Figure 5.

Measurements of nascent RNA species by quantitative in situ hybridization. (A) Schematic of the short reporter with the position of the probes indicated. The relative amount of probe corresponds to the Cy3/Cy5 (MS2) ratio measured at the transcription site of Exo1 cells. The predicted ratios were estimated from an elongation rate of 2.03 kb/min and 63.5 s for 3′-end formation (see Materials and methods). (B) Polyadenylated HIV-1 mRNAs do not accumulate at the transcription site of Exo1 cells. (top) Cells were hybridized with a mixture of pA+-Cy3 and MS2-Cy5 probes. (bottom) Cells expressing PABN1-GFP were hybridized with an MS2-Cy3 probe. Insets are magnifications of the transcription sites (2.5 × 2.5 μm). (C) Images of single cells with the various HIV-1 probes. Exo1 cells were transfected with Tat and hybridized with a mixture of Cy3-/MS2-Cy5 probes. (B and C) Field size is 20 × 20 μm. Bar, 6.7 μm.

3′-end formation occurs at various rates, whereas release of HIV-1 mRNAs occurs seconds after the onset of polyadenylation

To further characterize RNA species present at transcription sites, we used a probe specific for polyadenylated HIV-1 mRNAs. This probe yielded robust signals in the cytoplasm but only faint signals at transcription sites (11% of E2 signal; Fig. 5 B). This was unlikely the result of a failure of the probe to hybridize to its target because the nuclear polyA-binding protein PABN1 was also not detected there (Fig. 5 B). This indicated that RNAs were released rapidly once polyadenylation started. Indeed, with a total 3′-end processing time of 63.5 s (Table I), we calculated that cleavage/polyadenylation took 54.7 s and mRNA release took 8.8 s (see Materials and methods).

The CTD of the large subunit of RNAPII has been proposed to connect the polymerase with the 3′-end processing machinery (for review see Meinhart et al., 2005). In addition, RNAPII lacking the CTD have defects in transcription initiation and in elongating through nucleosomal templates (Meininghaus et al., 2000). Remarkably, transcription sites generated by this RNAPII mutant recovered very slowly after photobleaching (half-recovery of 550 s vs. 75 s for WT^res; Fig. 3). A defect in initiation should result in lower amounts of fluorescent signal at the transcription site, but it should not affect the time that polymerases spend on the gene. Thus, the defects we observed arose from a reduced ability to synthesize or process nascent RNAs, which is in agreement with biochemical data.

To confirm that alteration of 3′-end processing could slow down the release of HIV-1 RNA, we made a reporter containing mutations near the polyA site. Besides the canonical AAUAAA and G/U-rich element, HIV-1 contains upstream elements that are required for efficient 3′-end formation. These elements are present in the U3 region of the LTR, partly explaining why the upstream polyA site in the 5′ LTR is not used (Ashe et al., 1995; Gilmartin et al., 1995). The removal of these activating sequences does not prevent 3′-end formation but renders the process inefficient (Gilmartin et al., 1995). For instance, polyA sites inserted downstream of the HIV-1 3′ LTR are normally not used, but they become active when the U3 region of the 3′ LTR is removed (Ashe et al., 1995). Thus, we created cell lines expressing HIV-1 reporters lacking the U3 region in the 3′ LTR (clone pTRIP_1_13). Remarkably, when MS2-GFP was bleached at these transcription sites, the signal recovered much more slowly than for WT HIV-1 reporters: half-recovery took 400 s instead of 65 s (Fig. 4 C). Thus, slowing down the rate of 3′-end formation correspondingly increased the time that nascent RNAs remained at their transcription site.

Computer-simulated mRNA synthesis

To better understand HIV-1 RNA biogenesis, we simulated the behavior of individual polymerases in silico. For each time interval, each polymerase could initiate, elongate by one nucleotide, cleave and polyadenylate, and eventually release their associated RNA. The probabilities to perform these steps were the means of the measured rate constants (see Materials and methods). Predicted FRAP curves approximated those obtained experimentally with both the short (Fig. 2 I) and long reporters (Fig. 4 B). Simulated SDs were also in range with experimental ones. Close examination of simulations indicated that a large part of the SD was caused by stochastic variation in the total number of polymerases on the gene, which affected the prebleach values and the extent of recovery.

Interestingly, the parameters measured with the short reporter were slightly too rapid to accurately predict the recovery curve of the long one (Table I and Fig. 4 B). An explanation for this observation might reside in the fact that RNAPII can pause because pausing is expected to be more frequent with longer sequences. Another interesting possibility would be that the site of integration of the reporter in the genome generates some variation in the kinetics of mRNA synthesis and processing.

Kinetics of mRNA biogenesis and the transcription cycle of RNAPII

In this study, we were able to visualize mRNA transcription and processing in real time and single cells by fluorescence tagging of HIV-1 RNAs. By performing FRAP analysis on RNAPII and nascent RNAs through MS2-GFP, we could obtain a view of the entire HIV-1 transcription cycle. For this, the polymerase recovery curves were fitted with a diffusion- binding model and compared with that of nascent RNAs. This indicated that during the 333 s that the polymerase resided at the HIV-1 transcription site, 114 s could be attributed to elongation, and 63 s could be attributed to 3′-end processing and transcript release. The remaining 156 s could be the result of initiation or termination. In an attempt to discriminate these possibilities, we investigated the localization of Xrn2, a 5′ → 3′ exonuclease involved in transcription termination (West et al., 2004). Xrn2 is loaded cotranscriptionally at the end of genes (Luo et al., 2006), and it degrades the 3′ cleavage product generated by the 3′-end maturation of mRNAs (West et al., 2004). Interestingly, Xrn2 was present only in a minute amount at the HIV-1 transcription site (unpublished data), suggesting that termination and polymerase release may be rapid. Thus, a large fraction of the missing 156 s could be the result of initiation. In this view, as many as 47% of the polymerases could be initiating transcription at the promoter, 34% would be undergoing processive elongation, and 19% would be processing the RNA at the 3′-end (Fig. 7 B). However, it is equally possible that the steps between pre-mRNA cleavage and the entry of Xrn2 account for some of the missing 156 s. Likewise, an alternative hypothesis to explain the long residency time of the polymerase might be that a fraction of the polymerase would reengage transcription on the same gene by a looping mechanism, as described in yeast (O'Sullivan et al., 2004; Ansari and Hampsey, 2005). Finally, we also cannot exclude that some polymerases present at the HIV-1 transcription sites are stalled or are involved in as yet uncharacterized but slow processes.

Transcription initiation

We found that although the residency time of the polymerase at the HIV-1 gene was 333 s, the diffusion time between two binding events was 10 min. Previous FRAP experiments of the large subunit of RNAPII at random nucleoplasmic sites have indicated that for cellular genes, polymerases should spend only a third of their time engaged in transcription (Kimura et al., 2002). Because the half-life of transcription was estimated as 20 min, it was deduced that polymerases spend as much as 90 min diffusing between two transcriptional cycles (Kimura et al., 2002). Thus, initiation at the HIV-1 promoter is nearly 10 times more frequent than at cellular genes. This could be caused by a high efficiency of transcription initiation at the HIV-1 site or, alternatively, by a much higher density of active genes at the HIV-1 transcription site. Interestingly, we did not detect a rapid component in the recovery curves of RNAPII (Fig. 6 A), which corresponds to polymerases undergoing rapid binding and dissociation from the promoter (Darzacq et al., 2007). In contrast, the vast majority of polymerases appeared transiently immobilized on the HIV-1 gene and engaged in productive transcription. This suggested that the initiation of HIV-1 transcription was indeed very efficient. A high efficiency of initiation would be consistent with biochemical studies that have shown that the HIV-1 promoter is constantly occupied by the polymerase (Jeang et al., 1999; Marcello et al., 2001). In addition, it is well established that a major mode of trans-activating the HIV-1 promoter is at the level of elongation through the binding of Tat to nascent RNAs, and this requires that the promoter can efficiently initiate transcription (for reviews see Jeang et al., 1999; Marcello et al., 2001).

Transcription elongation

By analyzing the turnover of nascent RNAs, we found that nascent transcripts were elongated at a rate of 1.89 kb/min. Four important controls support these numbers. First, increasing the length of the mRNA increased the total synthesis time but yielded a similar elongation rate. Second, quantitative in situ hybridization indicated that a substantial fraction of nascent RNAs were at the level of 3′-end processing or transcript release, and the fraction of incompletely transcribed RNAs was consistent with the calculated elongation rate. Third, a slow mutant of RNAPII induced a reduced transcription rate. Fourth, treating cells with camptothecin, an inhibitor of elongation, also reduced the rate of RNA synthesis.

RNAPII is a fundamental enzyme in the cell. However, few studies were able to directly measure its activity in vivo. In vitro kinetic analyses of human RNAPII have yielded elongation rates between 0.9 and 1.8 kb/min and have suggested that the chemical step is rate limiting (Burton et al., 2005). Our in vivo data are in the upper range of these values, indicating that RNAPII may work close to its maximal speed. The giant gene of human dystrophin and a long yeast gene have also been used to evaluate elongation rates (Tennyson et al., 1995; Mason and Struhl, 2005). These studies yielded values of 2.4 and 2 kb/min, which is in agreement with our measures. The range of values obtained in different biological systems have observed elongation rates ranging from 1.5 kb/min for living bacteria to 5.7 kb/min for eukaryotic ribosomal RNA genes (Dundr et al., 2002; Golding et al., 2005). Our model system provides a tool of choice for the quantitative analysis of transcription in real time and living cells.

In vitro, single-molecule analyses of bacterial RNA polymerases have shown that the time taken to transcribe a DNA segment can vary substantially between individual molecules, and at least part of this variability can be attributed to polymerase pausing (Herbert et al., 2006). In higher eukaryotes, pausing is also a well-known phenomenon, and there is some evidence that the elongation rate can regulate gene expression (Batsche et al., 2006). We describe elongation as the repetition of an elementary step and model it as a straight line in the recovery curves. This does not take into account the polymerase heterogeneity that can be caused by pausing and may directly contribute to the exponential component of the curve. Although determining polymerase pausing time may not be easy in many cases, it seems that pausing occurs when cells were treated with camptothecin. Indeed, in this case, the MS2-GFP FRAP assay indicates a large increase in the exponential component of the recovery curve. Because camptothecin is believed to physically arrest the polymerase, it is tempting to speculate that this increase represents pausing.

Chromatin immunoprecipitation

Chromatin immunoprecipitation analysis was performed on Exo1 cells treated with GST-Tat essentially as previously described (Lusic et al., 2003; du Chene et al., 2007). Primer set A corresponds to primer sets for Nuc1 in Lusic et al. (2003) and map to the promoter-proximal region. Primer set B corresponds to a region of the vector proximal to the 3′ LTR (primer Bfw [5′-CATGGAGCAATCACAAGTAGC-3′] and primer Brv [5′-ATCTTGTCTTCGTTGGGAGTG-3′]). Primer set C maps 3′ of the 3′ LTR within the backbone of the vector (primer Cfw [5′-AGCATCTGGCTTACTGAAGCAG-3′] and primer Crv [5′-ATCGGTGATGTCGGCGATATAG-3′]). Primer set B13 corresponds to an unrelated genomic region, as described in Lusic et al. (2003). Quantification of immunoprecipitated material was performed by semiquantitative PCR and normalized for input DNA and for B13. The antibody against RNAPII was purchased from Santa Cruz Biotechnology, Inc. (N-20).

In situ hybridization

In situ hybridization was performed as previously described (Fusco et al., 2003). The formamide concentration was 50% in the hybridization and washing mixture except for the pA+ probe (hybridized and washed at 30% formamide) and the MS2 probe (10% formamide). The sequences of the probes were as follows (X stands for amino-allyl-T): HIV_I_E2 (5′-A X GGGTTGGGAGGTGGGTC X GAAACGATAATGGTGAAT X A); HIV_E1a (5′-A X GAGAGCTCCTCTGG X TTCCCTTTCGCTT X CAAGTCCCTGTTC X A); HIV_E1b (5′-A X TCTTGCCGTGCGCGCT X CAGCAAGCCGAGTCCTGCGT X A); HIV_intron1 (5′-A X TCTCGCACCCATC X CTCTCCTTCTAGCC X CCGCTAGTCAAAAT X A); HIV_intron2 (5′-A X AACTGCGAATCGT X CTAGCTCCCTGCT X GCCCATACTATATGTT X A); HIV_Exon2a (5′-A X GTGGCTAAGATC X ACAGCTGCCTTGTAAG X CATTGGTCTTAAA X A); HIV_Exon2b; (5′-A X ATCTTGTCTTCG X TGGGAGTGAATTAGCCC X TCCAGTCCCCC X A); HIV_pA+ (5′-A X TTTTTTTTTTTTTTTTTTTTTTTTTT X TGAAGCACTCAAGGCAAGC X A); MS2 (transcribed strand; 5′-A X GTCGACCTGCAGACA X GGGTGATCCTCA X GTTTTCTAGGCAAT X A); and MS2 (nontranscribed strand; 5′-A X AGTATTCCCGGG X TCATTAGATCC X AAGGTACCTAATTGC X A). The modified oligonucleotide probes for RNA FISH were synthesized by J-M. Escudier (Plateforme de synthèse d'Oligonucléotides modifiés de l'Interface Chimie Biologie de l'ITAV).

For quantitative measurements, 3D image stacks were collected and deconvolved with Hyugens (Bitplane AG). Background was removed, and the total light intensity at the transcription site was calculated and divided by the number of planes. The number of molecules was then computed from a calibration curve of the probes (Fusco et al., 2003), or, alternatively, transcription site signals were normalized to the ones of the cytoplasm. For each probe, 15–40 transcription sites were analyzed.

For Cy3/Cy5 quantification relative to the MS2-Cy5 probe (Fig. 4), cells were hybridized sequentially with the Cy3 and MS2-Cy5 probes (the MS2 probes hybridized at the lower stringency than the other probes). Images were taken in both channels, and the amount of light at the transcription site was calculated using the same mask for both colors. The Cy3/Cy5 ratios were then corrected for the specific activity of each probe by measuring the signals of an equimolar solution of the Cy3 and Cy5 probes under the microscope. The SDs were E1 (0.55), I (1.55), I-E2 (0.35), and E2 (0.16).

For the quantification of E1-Cy5 versus E2-Cy3 (Fig. S4), the two probes were hybridized simultaneously, and the signal of the two probes at the transcription site was normalized to the amount of signal in the cytoplasm. For this, each image stack was summed along the depth axis (z), and the projected signal was quantified along a line (six pixels wide) that passed across the transcription site. The two probes were normalized by first removing the background and then equalizing the cytoplasmic signals. The surface of the peak at the transcription sites was measured for both probes, and the values of exon 1 were divided by that of exon 2.

Immunofluorescence and image acquisition of fixed cells

Immunofluorescence was performed as previously described (Marcello et al., 2003). Anti–Pol II (all isoforms) was used at the following dilutions: 8WG16 at 1:100 and anti-SC35 (Sigma-Aldrich) at 1:100. Fluorescent images of fixed cells were captured on a 100× NA 1.4 wide-field microscope (DMRA; Leica) equipped with a camera (CoolSNAP HQ; Roper Scientific) and was controlled by MetaMorph software (Universal Imaging Corp.). Stacks of wide-field images were deconvolved with Huygens (Bitplane AG) and mounted with Photoshop (Adobe).

FRAP experiments

For live cell imaging, cells were maintained at 37°C in appropriate medium (Fusco et al., 2003). Two microscopic setting were used to perform FRAP. For analysis of rapid recoveries (Fig. 2, A and B), we used a confocal microscope (Meta LSM510; Carl Zeiss MicroImaging, Inc.) with a 100× NA 1.4 objective. MS2-GFP at transcription sites or in the nucleoplasm was bleached at 488 nm in a circle of 1.5-μm diameter at full laser power and for one passage (bleaching time of 100 ms). Recoveries were measured at a high frame rate (one image for ≤160 ms) and for a short time (up to 10 s) using ≤1% of the power of the 488-nm laser line. Images were analyzed as previously described (Phair and Misteli, 2000) by recording the fluorescence of the bleached region. Background was removed, intensities at each time point were corrected for bleaching by dividing them by the total cell fluorescence, and these values were finally normalized by dividing them with the fluorescent intensity before the bleach. In Figs. 2 (E–I), 3, and 6 B (right), postbleach values were additionally set to zero.

When recovery of transcription sites had to be recorded for periods exceeding 10 s, we used an adapted microscopic setting. A microscope (TE200; Nikon) equipped for both confocal and wide-field imaging was used with a 100× NA 1.45 objective. Transcription sites were bleached with the confocal port using a circular region of 2.5-μm diameter (bleaching time of 1 s). Recoveries were then recorded in the wide-field port using MetaMorph (Universal Imaging Corp.) with an excitatory light of low intensity. Images were recorded with an EM-CCD camera (Cascade 512K; Roper Scientific). Stacks of nine images 0.5 μm apart were collected every 3 s (one stack took 0.5–1 s). For image analysis, fluorescence intensities were measured in a small parallelepiped (1 × 1 × 1.5 μm) placed at the most intense area of the transcription site. This operation was performed automatically by a macro that was created in ImageJ software (National Institutes of Health). This automatic tracking of transcription sites in 3D allowed us to correct cell movements and to minimize signal from diffusing MS2-GFP. When a nucleoplasmic region devoid of transcription site was bleached, the cube was placed at the center of the bleached region. The values obtained were then treated and normalized as in the previous paragraph except that the postbleach value was taken at the 5-s time point. Indeed, at this time, the diffusing pool of MS2-GFP had come back very close to its equilibrium, allowing us to neglect the diffusion of MS2-GFP in the analysis (Fig. 2). In all figures except Fig. 2 C, the postbleach value was set to zero to facilitate comparison of the curves.

Diffusion coefficients

Diffusion coefficients were measured by FRAP on transfected HeLa cells. For 21 μm²/s of free GFP, we exploited the solution for free diffusion described by Soumpasis (1983). For 15 μm²/s MS2-GFPnls, we exploited the solution proposed for the reaction-diffusion model by Sprague et al. (2004).

Curve fitting

For the MS2-GFP FRAP experiment, the curve of 10–20 cells were averaged and fitted with a straight line followed by an exponential: f(t) = A + α × t, t ≤ t_elong; f(t) = A + α × t_elong + (B − A − α × t_elong) × (1 − exp[−α × (t − t_elong)/(B −A − α × t_elong)]), t > t_elong. t_elong is the time point at which the line converts into an exponential and corresponds to the length of the linear phase. A is the intensity at the zero time point, α is the slope of the initial linear part, and B is the immobile fraction. The minimization routine of the C++ GNU Scientific Library (http://www.gnu.org/software/gsl/) was used for finding the minimum of the chi square. Chi square values were 0.06, 0.06, 0.04, 0.01, and 0.04 for Exo1, Exo2, ExoLong, Wt^res, and hC4, respectively.

Recovery curves of RNA polymerase at the transcription site were fitted with the binding-only simplification of the diffusion/reaction model developed by Sprague et al. (2004): f(t) = C_eq × exp[−K_off × t], where K_off is the binding off rate and C_eq is the ratio of on and off rates. This model was fitted with the R (http://www.r-project.org/) nonlinear least squares function. This yielded a residency time of 333 s and a delay between two binding events of 660 s.

Computer simulations of mRNA biogenesis

The software was written in Metal, a BASIC emulator for Mac computers (http://www.iit.edu/~sarimar/GDS/metal.html). It simulated a small population of RNA polymerase molecules using a stochastic model: during an elementary time period, polymerases in a given state had a certain probability to perform the next reaction in the pathway. We used the following circular scheme: inactive → first nucleotide transcribed → second nucleotide transcribed…last nucleotide transcribed → pre-mRNA is cleaved → mRNA is released and polymerase is inactive.

The kinetic rates of each step were determined from the experimental data: elongation by one nucleotide, 31.5 s⁻¹ (1.89 kb/min); cleavage, 0.018 s⁻¹ (54.7 s); and mRNA release, 0.11 s⁻¹ (8.8 s). To estimate initiation rates, we measured the number of RNA molecules by quantitative in situ hybridization using the MS2 probe. We found that a mean of 105 RNA molecules was present at the transcription site of Exo1 cells (±50). Because FRAP experiments estimated that the MS2-tagged RNAs stayed a mean of 128 s at this site, the initiation rate was deduced to 0.99 s⁻¹ for the entire array. The total number of polymerases simulated was calculated such that 105 molecules of MS2-tagged RNA were present at the transcription site at equilibrium. At the start of the simulation, the population of polymerase was distributed according to the equilibrium. Then, at each time point of the simulation and for each polymerase, a random draw determined whether the next step of the pathway occurred or not. The probabilities of success were calculated to match the rate constants, and the time resolution was small enough to ensure that a single event per polymerase could occur. A variable fluorescent value was attributed to each polymerase, which corresponded to the number of nucleotides transcribed in the MS2 repeat. For the simulation of FRAP, all fluorescence was set to zero, and recovery was plotted as a function of time. The values obtained were treated as experimental data (i.e., normalized between the pre- and postbleach values).

Estimation of transcription elongation versus processing/release time by direct comparison of the FRAP curves of the short and long reporter

Curves normalized for their initiation rate (the initial slope of the curve) are shown in Fig. 4 B and indicate that 1.5 times more RNAs accumulated at the transcription site of the long reporter than for the short one. If initiation rates are constant, the amount (i) of fluorescent RNA at the transcription site is proportional to the time interval (t) between the moment polymerase reaches the MS2 site and the moment RNA leaves the transcription site. Thus, t = k × i, where k is a constant. However, t can be decomposed in two components: a variable one as a result of elongation (t1), which is proportional the length (l) of RNA that remains to be transcribed, and a constant time (t2), representing the time required for 3′-end processing and release. Thus, k × i = a × l + t2, where a is a constant. Because l is 2,180 for the short reporter and 4,580 for the long one, it follows that t1 is 1.2 times t2 for the short reporter and 2.5 times t2 for the long one.

Estimation of elongation versus processing time by quantitative in situ hybridization in Exo1 cells

The ratio between 5′ and 3′ probes depends on the relative amount of time that the polymerase spends on the transcription site once it reaches these hybridization sites. The same hypothesis was made as above by assuming that the time that a nascent RNA remains at the transcription site is decomposed into a variable part related to elongation and a constant part related to all other processes. In this case, the intensity (i) at a given hybridization site can be written as k × i = a × l + t2, where l is the length separating the hybridization site from the polyA site and k and a are constants. Assuming an elongation rate of 2.03 kb/min and a 3′-end processing time of 63.5 s (in Exo1 cells), one can calculate the values given in Fig. 5.

Estimation of cleavage and release rates in Exo1 cells

The FRAP curves yield a total time for 3′-end formation and release of 67.3 s on average. 3′-end formation can be decomposed in cleavage, polyadenylation, and transcript release. This latter rate can be estimated from the amount of polyadenylated mRNA present at the transcription site of Exo1 cells. The polyadenylated species represents 11% of the exon 2 signal, meaning that when a polymerase reaches exon 2, the nascent RNA then spends 89% of its time completing transcription and 3′-end processing and 11% as a polyadenylated species. Because the exon 2 probe is 560 nucleotides away from the polyA site, it should take polymerases 16.6 s to go from there to the polyA site (at 2.03 kb/min; value from Exo1 cells) and then a further 63.5 s to process the RNA (value from Exo1 cells). Thus, polyadenylated RNAs should remain at transcription sites 11% of 80.1 s (i.e., 8.8 s, yielding a rate of 0.11 s⁻¹). Because a total time of 63.5 s is required for 3′ processing, the time required for cleavage/polyadenylation was then estimated at 54.7 s (0.018 s⁻¹).

We thank R. Bordonné, G. Pegoraro, and T. Vasselon for critical readings of the manuscript. We are particularly indebted to X. Darzacq and R.H. Singer for their help and advice as well as for the scientific exchanges that occurred during the course of this project.

This work was supported by the Association pour la recherche sur le cancer (grant 3109), the European Community Systems Biology of RNA Metabolism in Yeast project (grant LSHG-CT-2005-518280), and the Network of Excellence European Alternative Splicing Network. A. Marcello was supported by the European Community Specific Targeted Research Projects consortium (grant 012182), a Human Frontier Science Program Young Investigators grant, and the AIDS project of the Istituto Superiore di Sanità of Italy. A. Kornblihtt is a Howard Hughes Medical Institute International Research Scholar and a career investigator of the Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET). M. de la Mata is a recipient of a CONICET fellowship. E. Basyuk was supported by a fellowship from Agence nationale de recherche sur le sida.