To detect RNAPII activity in living cells, we used a green fluorescent protein (GFP)–labeling approach adapted from the MS2 technique (5, 6). A cassette coding for 24 binding sites for the PP7 bacteriophage coat protein was inserted into the 5′ untranslated region (5′UTR) of a target gene (7). A PP7-GFP fusion protein is constitutively co-expressed and binds the stem loops after they have been transcribed, resulting in a GFP-labeled transcript (Fig. 1A) (8–10). Thus, one can directly measure both the earliest steps in pre-mRNA synthesis and the total dwell time of RNA at the site of transcription. The cell cycle–regulated POL1 promoter (POL1 pro) and PP7 binding site cassette (24 PBS) were knocked in upstream of the endogenous GLT1 gene, which was then found to transcribe similarly to the endogenous POL1 gene (fig. S1). When monitored by using wide-field microscopy, nascent transcription sites (TSs) are detected as diffraction-limited fluorescent puncta in the nucleus, the boundary of which is determined by using a red nuclear pore marker (Fig. 1B). The POL1 promoter was active during late G₁/S phase, but activity periods varied from cell to cell and even for different alleles within the same nucleus, as shown in a time series in two different cells (Fig. 1B and movie S1).

To analyze transcription kinetics, we followed the TS in the nucleus over time. The position and integrated intensity of an individual TS was determined by using a tracking algorithm (11) to generate the spatial trajectory and intensity time trace of the site (Fig. 1, C and D). Three distinct phases in the time trace are evident: (i) the increase in fluorescence signal corresponding to polymerization through the PP7 cassette as PP7-GFP binds recently transcribed RNA stem loops, (ii) a stable intensity of fluorescence corresponding to transcription of the GLT1 gene, and (iii) a sharp drop in fluorescence that corresponds to termination and release of the nascent RNA, which then rapidly diffuses away from the site (movies S2 and S3).

For an actively transcribing gene, many such events will be superimposed, with RNAPIIs in various positions along the gene, necessitating an analysis approach that takes into account these nonequilibrium kinetic processes (Fig. 2, A and B, and movie S4). To extract kinetic parameters from these transcription time traces, we developed an analysis approach that takes advantage of the fluctuations inherent in the measurement (12, 13). An idealized fluorescence time trace for transcription of a single pre-mRNA is shown in Fig. 2C. This process can be described by an autocorrelation function G(τ), which is a discrete autocorrelation over all the transition probabilities for RNAPII (6). The variables are the total dwell time of a transcript (T), which includes elongation and termination, and the transcript initiation rate (c). For long genes, the solution takes the form

where τ is the autocorrelation delay and H denotes the Heaviside step function, which is unity when τ ≤ T and zero when τ > T (Fig. 2D). Therefore, by steady-state observations of TS fluctuations it is possible to measure both the transcript initiation rate and the total time a transcript spends at the TS.

The autocorrelation for the POL1p-GLT1 construct reflects deterministic elongation times and no correlation between individual transcripts (Fig. 2E). The autocorrelation shows a linear decay with a sharp transition at the time that corresponds to the dwell time of a transcript at the TS (T = 290 ± 30 s) (SEM). The processive nature of pre-mRNA synthesis, consisting of thousands of sequential enzymatic cycles, results in a central-limit process where the total dwell time is distributed around a nonzero mean. Alternatively, a comparable number of micropauses interspersed with polymerization cycles would also produce a similar autocorrelation (14) (fig. S2). In addition, the autocorrelation vanishes at times greater than the dwell time, indicating that there is no transcriptional memory between initiation events. Thus, although the POL1 promoter is only active during a certain stage of the cell cycle, initiation events during this active period are stochastic and uncorrelated and occur with a frequency c = 1.3 ± 0.72 min⁻¹.

We then sought to separate the kinetic processes of initiation, elongation, and termination. By placing the PP7 cassette in either the 5′UTR or 3′UTR of a long gene such as MDN1 (15 kb), one can measure dwell times in which termination plays a lesser or greater role in comparison to elongation (Fig. 3, A to C) (15, 16). With the cassette in the 5′UTR, the dwell time is dominated by the downstream portion of the gene (PP7-MDN1, Fig. 3D, red curve, and movies S5 and S6), where the fluorescence is not changing, leading to the autocorrelation in Eq. 1. With the cassette in the 3′UTR (MDN1-PP7, Fig. 3D, blue curve, and movie S7), the autocorrelation reflects the transitions between N stem loops, which occur with frequency k:

In this case, the total dwell time T equals N/k. By using this complementary approach to selectively emphasize early and late events in the lifetime of a nascent RNA, we can separate elongation from termination.

MDN1 transcription dwell times and initiation rates varied over the cell cycle. The MDN1 gene was monitored at various stages of the cell cycle, resulting in fluorescence intensity traces during G₁ for PP7-MDN1 and MDN1-PP7 (Fig. 3, E and F). During G₁, the mean dwell time of PP7-MDN1 transcripts is T = 770 ± 260 s, and the mean dwell time of MDN1-PP7 is T = 140 ± 30 s (Fig. 3, G and H). From these data, it is possible to directly compute the elongation velocity according to the equation , where L is the length of the PP7 cassette plus the downstream sequence and T is the dwell time as measured above. The velocity of RNAPII on MDN1 is v = 20 ± 8 bases per second, and the termination time is 70 ± 41 s. The initiation rates are c = 0.16 ± 0.07 min⁻¹ for PP7-MDN1 and c = 0.24 ± 0.1 min⁻¹ for MDN1-PP7. The agreement between initiation rates follows from the processivity of RNAPIIs, which clear the promoter and proceed to termination. In late S/G₂ phase, PP7-MDN1 was observed to have a much shorter mean dwell time (T = 310 ± 42 s) and a faster mean initiation rate (c = 0.40 ± 0.18 min⁻¹) (Fig. 3I and fig. S3), resulting in a lower bound estimate of the velocity of 46 ± 6.2 bases per second. This variable dwell time of single RNAPIIs on the MDN1 gene is obscured by population averages of the single-molecule data and can be misinterpreted as RNAPII pausing (figs. S4 and S5).

Because POL1p-GLT1 and MDN1 each exhibit uncorrelated initiation events, gene firing is likely determined by a single rate-limiting step. Such a step might be the binding of a trans-acting factor to DNA, so we used two-photon fluorescence correlation spectroscopy (2PE-FCS) to determine the abundance and reaction-diffusion properties of the transcription factor Mbp1p (17, 18), which binds the POL1 promoter in vivo (2) and is necessary and sufficient for correct cell cycle–regulated expression of POL1 (19). The 2PE-FCS autocorrelation of Mbp1-GFP under endogenous regulation is shown along with a fit to a model where Mbp1-GFP alternates between bound and unbound states (Fig. 4 and fig. S6) (20). The time in the unbound state corresponds to three-dimensional (3D) diffusion within the nucleus, and the time in the bound state corresponds to 1D diffusion along DNA while the factor is nonspecifically bound (21). GFP targeted to the nucleus with a nuclear localization sequence (NLS-GFP) (18) shows simple 3D diffusion. The number of Mbp1-GFP molecules in the nucleus is 350 ± 18.

From these data, it is possible to directly compute the search time (t_s), according to the Berg and von Hippel model of facilitated diffusion (22): where M is the number of base pairs in the nucleus (12 × 10⁶), D_1D (5.1 × 10⁵ bp²/s) is the 1D diffusion coefficient measured in vitro (21), and N is the number of factors doing the search. The calculated search time is 52 ± 8 s, which compared to the measured time between initiation events of 46 ± 25 s (Fig. 2E). A modified description of facilitated diffusion that postulates that the search regions are determined by the nucleosome-free DNA results in a search time of 84 ± 11 s (fig. S7) (23). Both models of facilitated diffusion predict a search time that is in quantitative agreement with the direct measurement of transcription initiation frequency for the POL1 promoter. Thus, the initiation rate from this active state shows a rate-limiting step that is consistent with the search time for Mbp1p, suggesting that every encounter leads to a promoter clearance event.

The view of transcription in yeast that emerges from these data is consistent with a model where transcription activation is determined by recruitment of a rate-limiting factor, which then assembles the pre-initiation complex (fig. S7) (24, 25). Remarkably for the POL1 promoter, this recruitment process is highly efficient, but the Mbp1p-DNA complex is also dynamic, as is the basal transcription machinery (26), resulting in single uncorrelated transcription events instead of the bursts of transcription observed for other genes (8, 16, 27, 28). Moreover, transcription rates in yeast are consistent with a search process occurring in a functionally homogeneous nuclear compartment. The single-factor, single-target search time for Mbp1p in the yeast nucleus is ~5 hours. In contrast, the Lac repressor search time in bacteria is ~300 s (21). However, the effective search times are similar, because of the number of factors doing the search (for LacI, 20 molecules; for Mbp1, 350 molecules). It is unclear whether the biophysical aspects of the search process are modified in the metazoan nucleus, where many genes are not expressed and chromatin shows substantial compartmentalization. The live-cell approach described in this report, which uses fluctuation analysis to observe the activity of RNAPII and the reaction-diffusion behavior of transcription factors, will be critical in deciphering transcriptional control at single genes operating in their native context.

MOV1

MOV2

MOV3

MOV4

MOV5

MOV6

MOV7

PDF