Recent studies have implicated the nucleolytic activity of RNA polymerase as part of a proofreading mechanism2–4
, similar to that found in DNA polymerases5
. A key feature of this proofreading mechanism is a short backtracking motion of the enzyme along the DNA template (directed upstream, opposite to the normal direction of transcriptional elongation). Similar rearward movements are thought to accompany the processes of transcriptional pausing6–8
, and transcription-coupled DNA repair11
. During backtracking, the transcription bubble shifts and the DNA–RNA hybrid duplex remains in register, while the 3′ end of the RNA transcript moves away from the active site, and may even protrude into the secondary channel (nucleotide entrance pore) of the enzyme6,7,9
, blocking the arrival of ribonucleoside triphosphates (NTPs). In its backtracked state, RNAP is able to cleave off and discard the most recently added base(s) by endonucleolysis, generating a fresh 3′ end at the active site for subsequent polymerization onto the nascent RNA chain. In this fashion, short RNA segments carrying misincorporated bases can be replaced, leading to the correction of transcriptional errors (). Accessory proteins have been identified that increase transcriptional fidelity by preferentially stimulating the cleavage of misincorporated nucleotides: GreA and GreB for E. coli
and SII/TFIIS for eukaryotic RNA polymerase II2,3
Figure 1 RNA polymerase transcription and proofreading studied by optical trapping. a, During normal elongation, RNAP (green) moves forward (downstream) on the DNA (blue) as it elongates the nascent RNA (red). At each position along the template, RNAP may slide (more ...)
We studied transcription by RNAP at physiological nucleotide concentrations using a new single-molecule assay together with improved optical trapping instrumentation. In combination, these achieve subnanometre resolution along with extremely low positional drift. Our current system is capable of near-base-pair resolution in individual records of RNAP displacements, and achieves base-pair resolution (<0.3 nm) in averages of multiple records. During an experiment, two beads are optically trapped in buffer above a microscope coverglass by independently steered laser traps. A recombinant derivative of E. coli RNAP is bound specifically via a biotin–avidin linkage to the smaller of two polystyrene beads, while the transcriptionally downstream end of the DNA template (or the upstream end, in the case of assisting forces) is bound to the larger bead via a digoxygenin-antibody linkage, forming a bead–RNAP–DNA–bead ‘dumbbell’ ().
The tension in the DNA was kept nearly constant (8.4 ± 0.8 pN), for loads both opposing and assisting transcription, by feedback control of the position of the optical trap holding the larger bead. A force of this magnitude has a negligible effect on transcription rates, and is well below the stall force for RNAP12
. An opposing load was applied in all experiments, except where noted. Transcriptional elongation was observed by measuring the position of the smaller bead as the polymerase moved (). We chose to make the trap holding the larger bead an order of magnitude stiffer than that holding the smaller bead so that all motion appeared in the latter (see Methods). None of the components of the assay were attached to the coverglass surface: this isolates the system from drift of the microscope stage relative to the objective and other optics, which represented a major source of low-frequency noise in previous single-molecule studies12–17
. Measured drift rates during our experiments were typically below 5 nm h−1
(data not shown). We recorded the transcriptional motion of over 150 individual RNAP molecules at 1 mM NTPs moving on a DNA template derived from the E. coli rpoB
gene sequence. As previously noted12,13,15,17
, RNAP activity consists of periods of continuous motion interrupted by distinct pauses of variable duration (). The velocity during the continuous-motion phase averaged ~15 bp s−1
, but varied among molecules, consistent with earlier reports12,13,15
Figure 2 Backtracking occurs upon entry into long, but not short, pauses. a, Transcription records of two individual RNAP molecules are shown, each moving over the same template sequence. Both traces contain multiple short pauses (most are too short to be seen (more ...)
Computer analysis of RNAP records identified transcriptional pauses ranging from 1 s (our detection threshold) to more than 30 min. Only intervals where transcriptional elongation ceased and subsequently recovered were scored as pauses. Pausing events could be broken up into two broad categories: 95% of events were ‘short,’ with lifetimes drawn from a double-exponential distribution with time constants of 1.5 s and 6.5 s, similar to our previous findings12
. The remaining 5% of events were ‘long,’ with lifetimes >20 s and a broad, non-exponential temporal distribution. Long pauses occurred at positions randomly distributed along the DNA template, rather than at stereotyped locations, and appeared to be sequence-independent within the resolution of these experiments. On average, long pauses occurred with a frequency of 0.95 ± 0.21 kb−1
, a value that corresponds closely to ribonucleotide misincorporation rates during RNA synthesis in vitro1
, suggesting a possible role for such pauses in proofreading.
Operationally, we defined the duration of a ‘pause’ as the interval between the cessation of forward transcriptional motion and its subsequent recovery. At high spatial resolution, however, long pauses were found to consist of three distinct phases of motion that could be discerned in some individual records (), as well as in averages of multiple records (see below). After abruptly stopping forward transcription, the enzyme underwent a slow rearward movement (phase 1, backtracking), typically lasting from 1–5 s, before stopping altogether for a variable interval (phase 2, pause). At the end of phase 2, rather than immediately resuming transcription at normal rates, RNAP moved forward gradually, typically for 3–10 s, transitioning to elongation mode only after a significant fraction of the initial backtracking distance had been retraced (phase 3, recovery). In contrast, neither the backtracking nor the recovery phases were evident in records of short pauses, where transitions both to and from normal elongation were abrupt ().
To analyse the mean behaviour during phases 1 and 3, we averaged records of long pauses obtained under opposing loads after placing these in register along their rising edges, immediately before the cessation or the resumption of elongation, respectively (). This procedure allows one to probe details of the motion that would otherwise be obscured by noise in individual records: similar trace-averaging techniques have been successfully employed to detect nanoscale steps in motor proteins such as myosin18
, as well as to look for fast transients within the 8-nm step of kinesin20
. The average backtracking displacement during phase 1 of long pauses was 4.7 ± 0.8 bp, and could be fitted by a decaying exponential with a time constant of 1.2 ± 0.1 s. Both the duration and frequency of backtracking pauses are expected to display a strong force dependence due to the underlying motions involved. We found that the frequency of long pauses decreased dramatically from 0.95 ± 0.21 kb−1
under an ~8 pN opposing load to below 0.03 kb−1
under an ~8 pN assisting load (). This finding is consistent with a previous report of force dependence in the duration of very long pauses (~90 s) using a low-resolution optical trapping assay13
Figure 3 Averages of aligned long-pause records reveal details of backtracking and transcript cleavage events. a, Long-pause average of records at 1 mM NTPs (N = 56) displays a backtracking motion of ~5 bp (phase 1). Recovery (phase 3) is gradual, lasting ~5 s, (more ...)
Long pause statistics under different experimental conditions
Averages of records during phase 3 displayed a gradual forward motion, at an average velocity of 0.29 ± 0.01 bp s−1, before the resumption of normal elongation at 13.2 ± 0.1 bp s−1 (). The average forward displacement during recovery was 2.5 ± 1.0 bp, that is, about half of the initial backtracking distance. This reduced distance may reflect a mixed population of records, some of which exited from the pause more abruptly than others. However, the difference might also reflect the trace-alignment procedure. The exit from phase 2 is far less distinct than the entry into phase 1, and is therefore harder to pinpoint: minor registration errors tend to alter the magnitude of motions in averaged traces.
For comparison, we aligned and averaged an identical number of short pause records (). In contrast to the long pause average, the short pause average displayed sharp transitions both into and out of the pause (that is, no phase 1 or phase 3 motions), with no associated movement greater than a base pair. The average of a much larger population of short pauses (>250) showed the same behaviour (data not shown). The absence of backtracking in short pauses directly confirms and extends the conclusions of a recent study that examined the frequency and duration of short pauses, and found that these were independent of external load. This lack of force dependence implies an absence of backtracking motion, even as small as a single base pair12
We performed parallel experiments in the presence of the ribonucleotide analogue inosine triphosphate (ITP). Inosine mimics guanosine, forming a weak Watson–Crick pair that is slightly more stable than some measured mispairings21,22
. ITP incorporation inhibits next-nucleotide addition in human polymerase II by an amount similar to that of a mismatched base3
. In elongating complexes, inosine incorporation decreases the stability of the RNA–DNA hybrid, changing the relative stability of the backtracked and non-backtracked states, and also decreases the stability of secondary structures in the nascent RNA formed behind the complex. The addition of 200μM ITP to the standard transcription buffer (containing 1 mM levels of GTP, CTP, ATP and UTP) increased both the frequency and duration of long pauses (; ; Supplementary Information). Both phase 1 and phase 3 of long pauses were quantitatively similar to those observed in the absence of ITP (). However, ITP did not affect either the frequency or the duration of short pauses, nor the average transcriptional velocity between pauses.
To assay the effects of transcript cleavage on long pauses, we added the E. coli
transcription factors GreA and GreB. Addition of either 2μM GreA or 1μM GreB decreased the frequency of long pauses. Moreover, GreB significantly decreased the duration of long pauses, whereas the effect of GreA on duration was negligible (; ). The average distance backtracked in the presence of GreA increased to 6.6 ± 0.7 bp, whereas GreB appeared to abolish backtracking pauses altogether, yielding a mean backtracking distance close to zero for the few remaining long pauses, 0.5 ± 0.8 bp (Supplementary Information). These findings are consistent with the known properties of the two factors. GreA stimulates the cleavage of short, dinucleotide segments of backtracked RNA, and so its addition should relieve only those pauses associated with short backtracking motions, leaving pauses involving larger displacements. Conversely, GreB accelerates the removal of larger fragments to such a degree that the duration of backtracking pauses falls below the discrimination threshold of 20 s23
In the presence of both GreA and GreB, the duration (phase 2) of ITP-induced long pauses decreased dramatically, from 285 ± 83 s to 60 ± 13 s, while the average backtracking distance (phase 1) remained the same, 5.0 ± 1.0 bp (; ). Significantly, in the presence of both these transcription factors, the pause recovery (phase 3) became abrupt, and not gradual. We interpret this difference as being caused by Gre-stimulated cleavage of the RNA blocking the secondary channel after backtracking. Such cleavage would lead to the prompt removal of oligonucleotides containing a potential mismatch, thereby reducing the lifetime of the backtracked, paused state (phase 2), and restoring the new 3′ end of the nascent RNA to a position adjacent to the enzyme active site, ready for immediate polymerization.
In the absence of Gre-stimulated cleavage, RNAP must recover during phase 3 in a less direct fashion. In one mechanism, RNAP relies on its slow endogenous endonucleolytic activity to cleave the RNA at the enzyme active site, leading to long pause lifetimes but still allowing the possibility of error correction. In an alternative mechanism, thermal motions within the stalled complex may reverse the backtracking motion in a random walk process, carrying the original 3′ end of the RNA back to the enzyme active site, once again leading to longer lifetimes, but without concomitant error correction. The gradual recovery seen during phase 3 in the absence of Gre factors may reflect these processes. Random fluctuations of polymerase motion during phase 2 would not be apparent in averaged records, which only show ensemble behaviour.
Taken together, our high-resolution, single-molecule experiments are consistent with a proofreading mechanism in E. coli RNA polymerase involving entry into an initial backtracked state of the enzyme on the DNA template, followed by cleavage of the most recently polymerized RNA (1–10 bp) and enzymatic recovery. Under the conditions explored here (including an opposing load of ~8 pN), RNAP appeared to enter into long, backtracking pauses spontaneously at a rate of roughly once per kilobase: this rate was sensitive to transcription errors, and enhanced at least twofold by the addition of a nucleotide analogue. Incorporation of inosine leading to the backtracked state was relieved quantitatively by the action of transcription factors GreA and GreB, which are known to stimulate transcript cleavage. This simple editing mechanism may function, in principle, in many polymerase systems, including both prokaryotes and eukaryotes.