To monitor rDNA transcription in single-yeast cells, a synthetic construct containing 32 tandem copies of a 43–base-pair probe-binding sequence was inserted downstream of the 35S promoter into one of the repeats in the rDNA array (). These cells were then fixed and hybridized with a single-stranded oligodeoxyribonucleotide probe, labeled with five tetramethylrhodamine (TMR) fluorescent molecules, which was complementary to the tandem repeated sequences. The binding of many fluorophores to each RNA molecule allows its detection as a diffraction-limited spot in a conventional wide-field fluorescence microscope (Femino et al, 1998
; Raj et al, 2006
). We observed cells with no or few RNA molecules and also cells with many RNAs and bright transcription sites (, top-right panel). These correspond to cells with silenced and active tagged repeats, respectively. The bright transcription sites observed are consistent with EM Miller spreads that show many polymerases transcribing from an active gene producing many RNAs.
Figure 1 Genetic manipulation, methods and modeling. (A) A synthetic construct is inserted behind the 35S promoter in one of the repeats in the array to report the transcriptional activity of that repeat. Probes labeled with TMR fluorphores are designed to bind (more ...)
For cells with few RNAs, the RNAs are spatially resolved as individual spots. However, for cells with RNA counts above 40 or cells with bright transcription sites, the RNA molecules are too close to each other to be resolved. Using the cells with low RNA counts, we obtained a linear relationship between the total fluorescence and number of RNAs (Supplementary Figure S1
) and estimated the number of RNA molecules in bright cells by linear extrapolation from their total fluorescence. This process is automated by a custom-designed MATLAB image processing code. The RNA distribution obtained for wild-type cells is shown in . To determine whether genome position within the rDNA locus has an important function in regulating transcription activity, five different colonies obtained after plasmid integration were examined. As the construct integrated randomly along the array, these colonies have the construct integrated at different positions. The distributions of the five different integrants were similar (Supplementary Figure S2
), indicating that position effect does not affect transcription significantly. Colony 1 was selected for further studies.
Figure 2 Transcriptional changes during diauxic shift. During diauxic shift, mean RNA production decreases. This decrease is due to a decrease in burst size. The gene activation rate remains constant. (A) RNA expression profiles of a single-rDNA repeat at t=0, (more ...)
The experimentally determined distributions enable us to test dynamical models of transcription regulation and to determine their quantitative parameters. We assume that the burst model (Raj et al, 2006
) holds for rDNA transcription. Each repeat can be either in the active state of high transcription or in the inactive rate of no transcription. This model is supported by Miller spread (French et al, 2003
) data that shows that each rDNA gene is either silent or transcribing many RNA. This model is characterized by four parameters (). The activation rate α is the rate of switching from the inactive state to the active state. The inactivation rate γ is the rate of switching from the active state back to the inactive state and transcription rate μ refers to the rate of RNA synthesis in the active state. δ is the first-order rate constant of RNA decay. The three normalized rates, α/δ, γ/δ and μ/δ, characterize the steady-state distributions. displays calculated distributions obtained at different switching rates.
The model depicted in can be solved analytically (Raj et al, 2006
) and was fit to the experimental steady-state distribution. Because of the long tail of the distributions, the normalized deactivation rate γ/δ and the normalized transcription rate μ/δ cannot be obtained independently. However, the normalized activation rate α/δ and the burst size μ/γ, which is the ratio of transcription rate over gene inactivation rate can be obtained accurately. The burst size μ/γ corresponds to the average number of RNA molecules that are transcribed in each gene activation event. We obtained an average of α/δ=(0.42±0.07) and μ/γ=(190±20). We determined the decay rate δ of the RNA directly in an independent experiment by using thiolutin, an inhibitor of RNA synthesis, and measuring the amount of RNA left at various times after transcriptional inhibition (Supplementary Figure S4
). From these measurements, we obtained δ=(0.28±0.03) per min. This yields α=(0.12±0.02) per min, corresponding to an average time of (8±2) min for an inactive repeat to transit to the active state. Earlier, psoralens cross-linking results have shown that at log-phase in minimal medium, about 40% of the repeats are active (Dammann et al, 1995
). Hence, an active repeat transcribes RNA for about 5 min before transiting back to the inactive state.
Next, we explored how rDNA synthesis responds to perturbations in external conditions. It is known that rRNA synthesis decreases in response to nutrient downshift (Sandmeier et al, 2002
). To determine whether the decrease is due to a decrease in normalized activation rate α/δ or burst size μ/γ, a culture starting at OD600
=0.4 was monitored for 180 min. We measured the RNA distribution of cells at t
=0, 60, 90, 120, 150 and 180 min and fitted the distributions to the steady-state solution of the burst model. Because of fast decay rate of the RNA as compared with the rate of change of the other parameters, the distribution of RNA closely reflects that of the distribution at steady state. Hence, fitting the RNA distributions to the steady-state solution is a good approximation (see Materials and methods; Supplementary Figure S6
). As expected, the mean RNA counts decreases as the cell density increases (). Using the model, we infer that the normalized activation rate α/δ remains fairly constant, whereas the burst size, μ/γ decreases (; Supplementary Table I
). We measured the RNA half-life to be (3.4±0.8) min for the cells at t
=180 min (Supplementary Figure S4
). This decay rate is similar to the half-life (3.6±0.5) min measured at early log-phase, suggesting that activation rate α is constant throughout the 180 min of the experiment. Hence, the decrease in rRNA synthesis is due to the decrease in burst size. Earlier work has shown that the fraction of open repeats decreases during diauxic shift and that Rpd3, a histone deacetylase, is responsible for this decrease (Sandmeier et al, 2002
). As our measurements show that the activation rate α remains constant, we deduced that Rpd3 decreases the fraction of open repeats by affecting the deactivation rate γ.
To further explore the role of Rpd3 in controlling rDNA transcription, we examined the effect of deleting Rpd3. Despite Rpd3's role in diauxic shift, little is known about its effect during log-phase. Indeed, it seems like Rpd3 does not affect rDNA transcription in the log-phase as both the fraction of open repeats and average rRNA concentration in wild-type and rpd3
Δ strains are similar (Sandmeier et al, 2002
). To determine its effect during the log-phase, rpd3
Δ strains were grown to OD600
=0.4 and fixed for hybridization. Consistent with earlier data, we observed that the mean of the rpd3
Δ strain is similar to wild type (). However, the RNA distribution is different (). After analyzing the experimental distribution with the burst model, we find that the activation rate α in the rpd3
Δ cells is twice that of wild type, and its burst size μ/γ is reduced by half (). The change in the activation rate α and burst size μ/γ in the rpd3
Δ cells could not have been revealed by traditional biochemical methods, which measure population averages. We further measured the RNA distributions for the rpd3
Δ strain at t
=90, 150 and 180 min after it reached OD600
=0.4. Unlike the increase in activation rate α, which is constant in the rpd3
Δ strain at the different time points, the burst size μ/γ becomes more similar to that of wild type as cell density increases (Supplementary Figure S9A
; Supplementary Table II
Figure 3 Effect of Rpd3 deletion. Although the rpd3Δ strain has similar mean RNA expression as wild type, its activation rate is double and its burst size is half that of wild type. (A) RNA expression profiles of a single-rDNA repeat in the rpd3Δ (more ...)
Here, we used transcript counting to study the kinetics of rDNA transcriptional regulation. Our measured switching rate of the rDNA repeat is faster than reported switching rates of markers inserted into other regions that show gene silencing like the telomeres (Gottschling et al, 1990
) and the mating locus (Xu et al, 2006
). Our results suggest that each repeat spends about 5 min in the active state producing about 190 RNA molecules before switching off. This yields a Pol I reinitation interval of 300/190=1.6 s, which is comparable to the 1.2–2.2 s estimate obtained using EM Miller spreading (French et al, 2003
). It has been suggested that active chromatin structure cannot be directly inherited as replication of transcriptionally active rDNA gene results in two newly replicated inactive coding regions packaged into nucleosomes (Lucchini and Sogo, 1995
). Here, we suggest that not only the subset of active chromatin is not inherited but the set of active repeats is changing within one cell cycle. We ran a simulation to obtain the probability density for the amount of time that each repeat is open per cell cycle using different switching rates (Supplementary Figure S7
). With the inferred switching rates, almost all the repeats will be active in each cell cycle. It has been shown that repair of active genes occurs faster than silenced genes (Conconi et al, 2002
). Hence, one advantage of fast switching may be to allow damaged repeats to be repaired before it is inherited.
Population measurements performed on rpd3
Δ and wild-type strains showed that the two strains have similar fractions of active genes and mean transcriptional activity during log-phase. However, our single-cell measurement shows that the activation rate doubles and the burst size is reduced by half in the rpd3
Δ strain. As the fraction of open repeats in the rpd3
Δ strain is reported to be similar to that of wild type (Sandmeier et al, 2002
), we suspect that the deactivation rate would have also doubled. This would explain the reduction in burst size if transcription rate is unaffected by Rpd3. This suggests that Rpd3 functions to decrease both the activation and deactivation rates in log-phase. There are evidences showing that Rpd3 can function both as a transcriptional activator and deactivator (Deckert and Struhl, 2002
; Sertil et al, 2007
). If we assume that the transcription rate is unaffected by Rpd3, we can normalize the burst size in the wild-type strain to that of the rpd3
Δ to obtain the ratio of their deactivation rates γrpd3Δ
at different time points. This ratio becomes closer to 1 at later time points. This suggests that unlike the effect of Rpd3 on the activation rate α, which is independent of cell density, the effect of Rpd3 on the deactivation rate γ decreases with cell density. One hypothesis for how Rpd3 can regulate activation and deactivation rates independently is that Rpd3 uses different mechanisms to control them. For example, Rpd3 may decrease the activation rate α by deacetylating the H3 or H4 subunits of UAF, a transcription factor for Polymerase I recruitment (Keys et al, 1996
) and increase the deactivation rate γ by affecting the nucleosomes surrounding the rDNA promoter. This suggests that in wild-type cells, Rpd3 does not inactivate ON repeats as cell density increases. Rather, Rpd3 actively acts to repress the deactivation rate γ at low cell density and stops repressing at high cell density, thus enabling the deactivation rate γ to increase (Supplementary Figure S9B
). Future work is needed to elucidate the detailed molecular mechanism through which Rpd3 affects these rates.