The Dynamic Landscape of Transcription Initiation in Yeast Mitochondria

Controlling efficiency and fidelity in the early stage of mitochondrial DNA transcription is crucial for regulating cellular energy metabolism. Studies of bacteriophage and bacterial systems have revealed that transcription occurs through a series of conformational transitions during the initiation and elongation stages; however, how the conformational dynamics progress throughout these stages remains unknown. Here, we used single-molecule fluorescence resonance energy transfer techniques to examine the conformational dynamics of the two-component transcription system of yeast mitochondria with single-base resolution. We show that, unlike its single-component homologue in bacteriophages, the yeast mitochondrial transcription initiation complex dynamically transitions between closed, open, and scrunched conformations throughout the initiation stage, and then makes a sharp irreversible transition to an unbent conformation by promoter release at position +8. Remarkably, stalling the initiation complex revealed unscrunching dynamics without dissociating the RNA transcript, manifesting the existence of backtracking transitions with possible regulatory roles. The dynamic landscape of transcription initiation revealed here suggests a kinetically driven regulation of mitochondrial transcription.


5
Recent studies of transcription machinery have found that transcription initiation is not a 6 unidirectional process that invariably leads to elongation, but is rather a stochastic process 7 that involves divergent pathways such as abortive initiation, backtracking, and pausing in 8 addition to the progression to elongation 1-6 . These findings suggest that clearing initiation and 9 progressing to elongation might be rate-determining steps that control transcription 10 efficiency, and they have been suggested to be targets of transcription regulation 3,7-9 . 11 Transcription systems across lineages with different levels of complexity show highly 12 variable initiation efficiencies that depend on specific DNA elements 10,11 . The initiation stage 13 and transition to elongation are regulatory targets for many proteins and small molecules 8,12 . 14 Thus, investigating conformational dynamics during the early stage of transcription is crucial 15 to understanding the regulation of transcription efficiency and fidelity. 16 While multi-subunit RNAPs have been studied in great detail, we lack such deep 17 understanding of the mitochondrial transcription systems that share homology with 18 bacteriophage transcription systems. T7 RNA polymerase (RNAP), the single-subunit 19 transcription machinery of bacteriophages, is homologous to the yeast mitochondrial RNAP 20 (Rpo41) and the human mitochondrial RNAP (POLRMT) [13][14][15] . Transcription systems in 21 bacteriophages and mitochondria also share common promoter recognition mechanisms 15 . 22 While T7 RNAP does not require additional proteins to initiate transcription, yeast 23 mitochondrial transcription requires the initiation factor Mtf1, which is responsible for 24 stabilizing open promoter regions 6,[16][17][18] . Similarly, human mitochondrial transcription 1 requires the initiation factors TFAM and TFB2M [19][20][21] . Mtf1 is structurally and functionally 2 homologous to TFB2M 19,22 , and is functionally similar to the bacterial sigma factor [23][24][25] . To 3 regulate transcription efficiency at various promoters, these transcription systems have 4 developed a set of molecular mechanisms, in which they share several key features. Both 5 bacteriophage and bacterial RNAPs show scrunching of the downstream DNA into the active 6 site as the RNA-DNA hybrid and the transcription bubble grow during initiation. After a 7 stable full-length RNA-DNA hybrid is formed, the RNAP releases upstream promoter 8 contacts and the initiation bubble collapses to drive the transition into elongation [26][27][28][29][30][31][32] . A 9 similar mechanism is also thought to exist in higher organisms 33,34 . More recently, branching 10 between competing pathways and pausing during initiation have been observed in both 11 bacteriophages and bacteria and may play crucial roles in regulating transcription 12 activity 1,3,4,35 . 13 In yeast mitochondria, transcription initiates with the assembly of Rpo41 and Mtf1 at 14 has not been observed directly, but its single-subunit relative, T7 TIC, exhibits a concerted 23 motion of DNA scrunching and rotation during initiation, and the promoter unbends upon 24 transition to the elongation complex 30,32 . During transcription initiation by T7 RNAP, the 25 population map from flow-in for progression to position +8 revealed an earlier transition 1 from the high FRET state back to the mid or low FRET state, as demonstrated by the 2 coexistence of these states after time zero (Fig. 2b). The FRET population map from flow-in 3 for run-off revealed a similar behavior to that seen in the 0 to +8 measurements, but with the 4 low FRET state dominating more at the end, possibly due to differences in the efficiency of 5 incorporating 3'dCTP and CTP. 6 The dominant FRET population at position +8 and under run-off conditions was 7 indistinguishable from that of bare DNA ( Fig. 1c and 2b), possibly because DNA templates 8 I/II happened to have similar RD-A values for bare DNA and the elongation complex. This 9 finding made it difficult to judge if the observed high-to-low FRET transition represented 10 initiation-to-elongation transition or abortive initiation. To answer this question, we designed 11 another DNA template +11/−11 that had the same sequence as DNA template II but different 12 labeling positions, i.e., at positions −11 of the non-template strand and +11 of the template 13 strand to enable better resolution in the low FRET range (Fig. 2c). Using this template, bare 14 DNA and the open promoter (IC0) showed FRET levels of 0.32 and 0.75, respectively (Fig.  15 2d). IC7 showed a major FRET level of 0.64, which was lower than that of the open 16 promoter; this finding is reasonable if we consider twisting of the downstream DNA arm, 17 which would position the dye away from the transcription site at this stalling position. In 18 addition, IC7 showed frequent transitions to a low FRET state, whose level matches that of 19 the closed state in IC0 (Fig. 2d, Supplementary Fig. 3). The TIC stalled at position +8 20 showed a dominant FRET level of 0.44, which was different from that of bare DNA, and 21 stayed at this level for a long time, just as the TIC at position +8 on DNA template II 22 displayed a prolonged low FRET state (Fig. 2d, Supplementary Fig. 3). These results confirm 23 that the low FRET state observed at position +8 with DNA template II is distinct from the 24 low FRET state of bare DNA. Thus, we conclude that the TIC switches to the elongation 25 complex at position +8 (EC8), with a large and abrupt change of conformation. As expected, 1 under run-off conditions, DNA template +11/−11 displayed recovery of the FRET level to 2 that of bare DNA, with a small remaining FRET population representing the TIC at 3 intermediate steps (Fig. 2d). 4 5

The transcription initiation bubble collapses upon transition to elongation 6
To obtain further support for our proposal that the transcription complex at +8 represents the 7 elongation complex, we measured the kinetics of initiation bubble collapse. Figure 3a shows 8 a model in which the initiation region from −4 to −1 remains melted up to position +7, and 9 the initiation bubble collapses upon transition to elongation at position +8, reannealing the 10 −4 to −1 region into duplex DNA (Fig. 3a). To monitor the transition from initiation to 11 elongation, we substituted an adenosine at position −4 of the non-template strand with a 2AP 12 residue, which emits fluorescence at 370 nm that is quenched when it pairs with a thymine 13 residue (see online Methods) 36 . The fluorescence intensity of 2AP at position −4 was high in 14 the open complex as the −4 basepair was melted, which was expected to decrease upon 15 collapse of the initiation bubble. Thus, real-time monitoring of the fluorescence intensity 16 during an in vitro transcription assay on two walking templates ( Fig. 3b) allowed us to 17 measure how the kinetics of initiation bubble collapse depends on the transcription position. 18 Upon walking to position +7, the fluorescence of 2AP remained constant for the total 19 observation time (Fig. 3c). However, upon walking to position +8, the fluorescence of 2AP 20 decreased to a lower level at a rate of 0.004 s -1 . Similarly, upon walking to positions +9 and 21 +10, the fluorescence intensity decreased to even lower levels at higher rates of 0.0165 s -1 22 and 0.027 s -1 , respectively ( Fig. 3c and 3d). These results are consistent with our 23 interpretation of the single-molecule data that the complex at position +7 is an initiation 24 complex with an open upstream bubble, and complexes from position +8 onward represent 1 elongation complexes. The results also imply that the rate of transition to elongation 2 increases with increasing length of the RNA/DNA hybrid from +8 to +10. 3 The progression of the major TIC conformation observed here is consistent with the known 4 crystal structures of T7 and the human mitochondrial transcription machinery (Fig. 3e). The 5 crystal structure of the yeast mitochondrial transcription machinery has not been determined; 6 therefore, to represent IC0, we modeled the TIC structure of human mitochondrial RNAP 7 (POLRMT) in complex with transcription factors (TFAM and TFB2M) (PDB: 6ERP) 41 . The 8 promoter DNA in the TIC shows a sharply bent conformation, leading to a decrease in 9 distance between the FRET dye pair at positions +16 and −16. For IC7, we modeled the 10 known structure of the bacteriophage T7 RNAP in complex with the promoter DNA (PDB: 11 3E2E) 37 , which exhibits a further bent conformation, consistent with our smFRET data. For 12 the elongation complex, we modeled EC9 of the human mitochondrial transcription complex 13 with DNA (PDB: 4BOC) 38 , which shows a relaxed conformation of DNA. 14 15

Landscape of conformational dynamics during transcription initiation and elongation 16
Different FRET states representing unbent, bent, and scrunched TIC conformations were well 17 distinguished by our FRET histograms and traces ( Fig. 1c and 1d). We used hidden Markov 18 modeling (HMM) to extract kinetic information from the smFRET data (see online 19 Methods) 43 . Figure 4A shows representative smFRET traces of IC2 and IC7 along with 20 hidden state dynamics from the HMM analysis, assuming three hidden states. The high FRET 21 state (scrunched conformation) frequently switched to the mid FRET state (unscrunched 22 conformation) or the low FRET state (closed promoter), but mainly transitioned through the 23 mid FRET state to reach the low FRET state, consistent with what we observed in 24 synchronized FRET population maps (Fig. 2b). It is also worth noting that the unscrunching 1 events were rare in IC7 compared with IC2, reflecting higher stability of the scrunched TIC 2 at a later stage of transcription. The unscrunching rate revealed by the HMM analysis was 3 0.31 at position +2, and gradually decreased to 0.065 at position +7 during the initiation 4 phase (Fig. 4b). 5 Next, we constructed transition density plots (TDPs) from the HMM analysis at each stalling 6 position (Fig. 4c). The TDP at position 0 clearly shows the dynamics of promoter opening 7 and closing (transition between low and mid FRET states). In IC2, the dominant events are 8 the scrunching-unscrunching events (transition between mid and high FRET states) with rare 9 promoter opening-closing events. After proceeding to positions +3, +5, +6, and +7, the level 10 of the high FRET population gradually increased, consistent with the corresponding FRET 11 histograms. At positions +5, +6, and +7, the relative density of the low-mid FRET transition 12 became higher, which was due to a reduction in mid-high FRET transition events at later The stalled initiation complex makes unscrunching transitions without dissociating 20

RNA transcript 21
A straightforward interpretation of the frequent drops in the FRET level for stalled initiation 22 complexes at positions +2 to +7 is that they represent dissociation of the RNA transcript, 23 followed by unscrunching of the DNA (high-to-mid FRET transition), i.e., abortive initiation, 24 and then the re-initiation of transcription using fresh NTP molecules (mid-to-high FRET 1 transition), occasionally intervened by full closing and re-opening of the promoter 2 (transitions to low FRET level). As we constantly supplied NTP mix to keep the reaction at 3 equilibrium in the experiments described above, it was not clear if the drops and rises in the 4 FRET level truly represented the abortion and re-initiation of transcription. Thus, we 5 performed a different set of experiments in which we washed out the NTP mix, Rpo41, and 6 Mtf1, and tracked changes in the FRET distribution over time. At position +2, in just 1 min 7 after NTP wash-out, the scrunched population (high FRET) disappeared almost completely, 8 suggesting rapid dissociation of the dinucleotide transcript ( Fig. 5a and 5b). By contrast, at 9 position +7, the scrunched population decreased much slower, with a half-life of 10 approximately 5 min, reflecting the higher stability of IC7 compared with IC2 ( Fig. 5a and  11 5c). This finding is consistent with the fact that exit of RNAP from the late initiation stage is 12 markedly slower than that from earlier stages during bacterial transcription 35 . 13 DNA template II could not distinguish EC8 from the bare DNA at position +8, so we used 14 DNA template +11/−11 to examine conformational changes at this position. Upon NTP 15 wash-out, the scrunched population at position +8 diminished even slower than it did at 16 position +7, demonstrating the high stability of EC8, which is typical of the elongation Next, we examined whether the conformational dynamics of the TIC disappeared in the 1 absence of NTP mix to re-initiate transcription with. The NTP mix was washed out from the 2 TIC equilibrated at position +7 and, in most traces, the FRET level dropped a little while 3 after the wash-out, possibly representing an abortive initiation event. However, to our 4 surprise, many traces showing a drop in the FRET level subsequently returned to the high 5 FRET level (Fig. 6a, Supplementary Fig. 5), which must have happened without dissociating 6 RNA strands and synthesizing new ones because there was no NTP substrate remaining. This 7 observation demonstrates that the TIC makes conformational transitions not necessarily by 8 aborting RNA synthesis. As such dynamics occurred with bound RNA strands, the observed 9 mid or low FRET level should not represent the conformation of the open promoter or the 10 bare DNA template. 11 Compared with that in the presence of NTP, the dwell time of the high FRET state in the 12 absence of NTP was markedly longer and displayed a non-exponential distribution, further 13 supporting the proposal that the dynamics of conformational changes in the absence of NTP 14 are distinct from those of abortive initiation (Fig. 6b). The duration that the TIC spent out of 15 the high FRET state was also much longer in the absence of NTP than in the presence of 0.5 16 mM NTP (Fig. 6c). As the majority of NTP-independent transitions occurred to and from a 17 mid FRET level ( Supplementary Fig. 5), we measured the high-to-mid and mid-to-high 18 transition rates as if they represented unscrunching and scrunching kinetics, respectively, 19 occurring with bound RNA strands. In the absence of NTP, both transition rates were much 20 lower than those in the presence of NTP, indicating that in the absence of NTP, the TIC goes 21 through slower transitions that are distinct from those of abortive transcription and re-22 initiation (Fig. 6d). 23 After stalling the TIC at position +7 without NTP for longer than 2 min, we flowed in 0.5 1 mM 3'dCTP to see if the TIC could still progress to the elongation stage. Soon after the flow-2 in, the FRET level dropped to that of EC8 (Fig. 6a, Supplementary Fig. 5). This occurred 3 consistently in most traces and a FRET population map generated from the traces 4 synchronized at the moment of 3'dCTP flow-in clearly showed the high-to-low FRET 5 transition occurring in 5 seconds (Fig. 6e). As DNA template II could not distinguish the 6 conformation of EC8 from that of bare DNA, we repeated the same measurement using DNA 7 template +11/−11, and the majority of the traces displayed transitions to the FRET level of 8 EC8 (Fig. 6f). These results show that the TIC can be stalled at a late stage of initiation, make 9 multi-step conformational transitions to unbent or unscrunched DNA conformations without 10 completely dissociating RNA strands, and then resume transcription to progress to the 11 elongation stage. 12 As shown above, the transient drop of FRET level in stalled IC7 should not represent 13 unscrunching of IC7 by abortive dissociation of RNA. Another possibility is that it represents 14 temporary transition forward to an EC-like structure where the upstream promoter is released 15 prior to the addition of the 8 th nucleotide and the promoter is unbent. FRET histogram was 16 built from low-FRET events in IC7 with DNA template +11/−11 after NTP was washed out. 17 Then it was compared to the FRET histogram of EC8 with DNA template +11/−11 (Fig. 6g). 18 Major FRET level of the transient drops was distinguishably lower than that of EC8, 19 suggesting that it represents a structure distinct from EC8. Thus, we conclude that IC7 20 branches into a conformation, which represents neither unscrunching by RNA dissociation 21 nor an EC-like structure. 30% of smFRET traces from stalled IC7 showed transient drops of 22 FRET level while 64% showed an irreversible drop (Sup. Fig. 6). Branching ratio between 23 them is in rough agreement with the ratio found from the decay rates of high FRET state 24 before and after NTP wash-out (0.026 vs (0.067−0.026); Fig. 6b). The only possibility is that 25 this distinct transient conformation represents an unscrunched downstream DNA that still has 1 an RNA bound, which can be elongated further. Consequently, downstream DNA 2 unscrunching would destabilize the RNA-DNA hybrid and fray the 3'-end of the RNA, 3 resembling a backtracked complex. In this study, we combined single-molecule techniques with ensemble biochemical assays to 7 dissect the conformational dynamics of the mitochondrial TIC throughout the initiation and 8 elongation stages. At each nucleotide step, the TIC did not adopt a stationary conformation, 9 but rather exhibited dynamic transitions between closed, open, and scrunched conformations. This finding is reminiscent of the NTP-independent scrunching-unscrunching dynamics of 20 bacterial transcription systems 3 . The unscrunching motion in bacterial transcription 21 mechanistically resembles backtracking that leads to long-lived catalytically inactive 22 initiation complexes; this process is thought to play a regulatory role by maintaining an 23 "elongation-ready" initiation complex that can be conditionally triggered. Backtracking 24 during transcription is known to occur by extruding the 3' end of nascent RNA through the 1 secondary channel 45 . There is yet to be any evidence that mitochondrial RNAP has a 2 secondary channel like multi-subunit RNAPs. The absence of a distinct secondary channel in 3 mitochondrial RNAP might explain why the lifespan of the unscrunched state at position +7 4 in the Rpo41-Mtf1 complex was relatively short compared to what was found in the bacterial 5 system. However, the structure of human mitochondrial RNAP suggests that there is room for 6 3'-end extrusion near the active site. Thus, the NTP-independent unscrunching motion may 7 represent transient backtracking during initiation that could regulate transcription efficiency. 8 Future experiments on the basepairing state of RNA and the unscrunched structure of TIC 9 would reveal the identity of the transiently unscrunched state found here. 10 Hidden Markov analysis of single-molecule traces allowed us to quantify the conformational 11 transition rates. The unscrunching rate decreased more than 4-fold as transcription initiation 12 proceeded from position +2 to +7, implying that the conformational stability of scrunched 13 DNA is highly sensitive to changes in the TIC structure. The sharp dependence of the 14 unscrunching rate on the transcript length might stem from differences in the stability of 15 RNA-DNA hybrid, which may provide a mechanism to prevent incorrectly transcribed RNA 16 primers entering the elongation stage due to lower stability of the RNA-DNA hybrid. It 17 further relates to the observation that mitochondrial transcription efficiency varies up to ~ 18 100-fold between different initiating nucleotide sequences at positions +1 and +2 10 , 19 suggesting that the scrunched TIC conformation is differentially stabilized by different 20 initiating sequences. Taken together, these results suggest that the initiation stage may 21 function as a regulatory step to control transcription kinetics depending on promoter 22 sequences and biochemical circumstances. In bacteriophage and bacterial transcription 23 systems, the progression through transcription initiation serves as a rate-determining step in 24 It is intriguing that the conformational progression of the TIC within the initiation stage was 1 more reversible in this two-protein transcription system of yeast mitochondria than in the 2 more primitive, single-protein transcription system of T7 bacteriophage. While transcription 3 initiation by T7 RNAP shows a single dominant FRET population at each step representing 4 one stable conformation 30 , Rpo41 and Mtf1 showed highly reversible dynamics between 5 distinct conformations. Rpo41 is homologous to T7 RNAP, but requires Mtf1 to initiate 6 transcription 13,14,46 . Mtf1 is thought to stabilize the open promoter complex 18,24,36 , but our 7 results suggest that Mtf1 in complex with Rpo41 allows dynamic exchange between 8 scrunched and unscrunched conformations, which might result in enhanced production of 9 abortive transcripts. This proposal resonates with the observation that the addition of Mtf1 10 increases the production of abortive transcripts by Rpo41 on pre-melted DNA 18 . Thus, our 11 observations suggest novel roles of initiation factors in facilitating conformational transitions 12 of the TIC, thereby providing proofreading steps or conditional gates that determine 13 progression to the elongation stage. 14 Transition to the elongation stage was found from smFRET measurements to occur in a 15 single step precisely at position +8, as indicated by a large change in the FRET level and the 16 disappearance of conformational fluctuations, setting the TIC in a stable and less bent 17 conformation. This contrasts with the smFRET observation on the single-subunit T7 18 transcription system, in which the transition occurs gradually over positions +8 to +12 30,39 . 19 Ensemble-level biochemical studies and crystal structure analyses have also suggested that 20 transition of T7 RNAP to the elongation stage occurs in multiple steps with a series of 21 conformational transitions, producing abortive transcripts of at least up to 20 bases 37,39,40,47 . 22 The interplay between RNAP and initiation factors in yeast mitochondria appears to make an 23 abrupt release of the promoter at a sharply defined position, which is not reversible. This 24 process prevents the elongation complex from reversing to the initiation stage or escaping 25 through an abortive pathway. Irreversible dissociation of Mtf1 in the elongation stage 1 possibly functions as a kinetic latch to prevent such non-productive pathways. Though Mtf1 2 is known to dissociate after transcribing 13 nucleotides 48 , it is not clear whether it dissociates 3 at or after the unbending transition at position +8. On the other hand, 2AP measurements 4 showed the collapse of the initiation bubble starting at position +8, but the bubble appeared to 5 gradually collapse over positions +8 to +10 (Fig. 3c). In case of T7 RNAP, promoter release 6 and bubble collapse occur synchronously. This does not appear to be the case in 7 mitochondrial transcription, where the TIC takes further steps until the initiation bubble fully 8 zips up. Our results suggest that promoter release occurs first to form an unbent EC-like state 9 and then bubble collapse follows to complete transition to elongation. 10 Integrating the conformational transitions identified in this study, we constructed a kinetic 11 model of mitochondrial transcription initiation (Fig. 7). Initial bound complex of 12

DNA/Rpo41/Mtf1 transitions between closed and open promoter states (IC0closed and IC0open). 13
Upon incorporation of the initiating nucleotides, IC0open starts scrunching in stepwise manner 14 (ICnscrunched), but it often reverts to IC0open and even to IC0closed by dissociating RNA to re-15 initiate transcription. The rate of abortive transcription greatly decreases with increasing 16 length of RNA, making the scrunched TIC increasingly stable. In addition to the abortive 17 transcription, IC7scrunched makes slower, reversible transitions to an unscrunched conformation 18 (IC7unscrunched), which may also occur at earlier positions. In these states, the transcript 19 remains bound to the complex, possibly with its 3'-end extruding from the complex. Then, 20 IC7scrunched makes a large irreversible conformational transition at position +8 by releasing the 21 upstream DNA that might be accompanied by dissociation of Mtf1 (EC8). It is followed by 22 gradual zipping of the upstream bubble in the following steps. 23 In conclusion, our study reveals the highly dynamic and reversible nature of mitochondrial 1 transcription initiation. We propose that the dynamic initiation stage and irreversible 2 transition to the elongation stage serve as key features that distinguish multi-subunit 3 transcription machineries from simpler single-subunit systems. Our findings provide novel 4 mechanistic perspectives on how transcription machinery regulates the progression of 5 transcription initiation, and form a basis for studies of more sophisticated transcription 6 machineries of higher organisms. were purified by HPLC (Integrated DNA Technologies, USA). The oligonucleotides were 12 fluorescently labeled at the amine groups by standard assays using Cy3 or Cy5 NHS esters 13 (Lumiprobe, USA), and unreacted dyes were removed by ethanol precipitation. To generate 14 duplex DNA molecules, single-stranded DNAs were mixed in a 1:1 ratio, annealed at 95°C 15 for 1 min, and then slowly cooled to room temperature for 1 h. 16

Single-molecule data analysis 10
Movies obtained using the TIRF microscope were analyzed using custom software to extract 11 single-molecule fluorescence traces, as described previously 6 . The FRET efficiency was 12 calculated with background and leakage correction as EFRET = (IA − 0.08 × ID)/(ID + IA), 13 where ID and IA are the background-subtracted intensities of the donor and acceptor dyes, 14 respectively. The acceptor dyes were briefly excited at the beginning and end of each movie 15 to exclude traces lacking acceptor dyes from further analysis. Each FRET histogram was 16 built from more than 50 movies by selecting traces with a single pair of Cy3 and Cy5 dyes 17 and representing each trace by EFRET averaged over five frames. Hidden Markov analysis was 18 performed using ebFRET software developed by the Gonzalez group 43 . 19

2-Aminopurine fluorescence assay 20
Steady-state fluorescence measurements were carried out at 25°C using a Fluoro-Max-2 21 spectrofluorometer (Jobin Yvon Spex Instruments S.A., Inc., USA) in buffer containing 50 22 mM Tris-acetate (pH 7.5), 100 mM potassium glutamate, and 10 mM magnesium acetate. 23 The fluorescence spectra of 200 nM 2AP-incorporated duplex promoters were collected from 24  template complexed with Rpo41 and Mtf1 was observed using a total internal reflection 31 fluorescence microscope. b, DNA templates used in basepair-wise measurements of initiation 32 complex dynamics. DNA template I could be stalled at positions +2, +3, +5, and +6, while 33 DNA template II, which differed from DNA template I by four basepairs (blue), could be 34 stalled at positions +7 and +8. Both templates were labeled with Cy5 at position −16 of the 35 non-template strand (magenta) and Cy3 at position +16 of template strand (green). The 36 transcription promoter (underscored) and start site (arrow) are indicated. c, FRET histograms 1 from single-molecule traces with colocalized Cy3 and Cy5 signals at each stalling position. 2 Histograms were fit to single, double, or triple Gaussian peaks. The brown, green, and 3 magenta curves represent low, mid, and high FRET populations, respectively. d, 4 Representative smFRET traces at positions 0, +2, and +6 showing the Cy3 (green) and Cy5 5 (magenta) signals and the FRET efficiency traces (navy). e, The FRET level of the major 6 population in (c) shown for each stalling position as the center of the major Gaussian peak. 7 Error bars represent the error in Gaussian fitting. f, The Cy3-to-Cy5 distance at each stalling 8 position calculated from the average between major FRET levels from DNA templates I/II (e) 9 and I/II NT ( Supplementary Fig. 1c). Error bars represent propagation of the errors in FRET 10 levels. annealed to the template DNA from positions +1 to +7. The template bases from −4 to −1 are 3 single-stranded, resulting in a strong fluorescence signal of 2AP at position −4 of the non-4 template strand (green). At position +8, EC8 is shown where the initiation bubble has 5 collapsed and the −4 to −1 region is reannealed, resulting in the quenching of 2AP 6 fluorescence. b, Design of DNA templates used for 2AP fluorescence measurements, to be 7 stalled at positions +7, +8, +9, and +10. c, Changes in 2AP fluorescence measured along in 8 vitro transcription reactions to the indicated positions, normalized against the initial intensity. 9 The intensity traces were fit to a single exponential decay curve to determine the transition 10 rates. d, Fluorescence decay rates measured from (c) at different walking positions. e, Models 11 of the IC0, IC7, and EC9 structures generated using PyMOL (Schrödinger, USA). IC0 was 12 modeled using PDB 6erp (human mitochondrial RNA polymerase initiation complex), IC7 13 was modeled using PDB 3e2e (bacteriophage T7 RNA polymerase initiation complex with 7 14 bp RNA:DNA), and EC9 was modeled using PDB 4boc (human mitochondrial RNA 15 polymerase elongation complex with 9 bp RNA:DNA). The green and red balls represent the 16 a, FRET histograms obtained at equilibrium and after washing out the NTP mixture for 7 positions +2, +7, and +8, and run-off conditions. For position +8, results from DNA template 8 +11/−11 are included to distinguish between the populations of the elongation complex and 9 DNA only. Each histogram was obtained from 12 short movies taken during each minute 10 after washing out the NTP mixture. b,c, The relative population of the high FRET state 11 (scrunched complex) obtained from histograms at positions +2 (b) and +7 (c) shown relative 12 to time. Each graph was fit to a single exponential decay curve, and the half-life is shown. d, 13 The relative population of the mid FRET state (elongation complex) obtained from 14 histograms at position +8 on DNA template +11/−11 shown relative to time. template II equilibrated at position +7 (gray region; 0.5 mM each of ATP, GTP, and UTP) 22 after washing out the NTP mix (first arrow). Subsequently, 0.5 mM 3'dCTP was added to 23 promote progression to position +8 (second arrow). The abrupt drop in the FRET efficiency 24 indicates successful progression to position +8. b, Dwell time histogram of the high FRET 1 (scrunched) state in the presence of NTP mix (grey; 809 events) and after NTP wash-out 2 (magenta; 214 events). c, The time taken to recover the high FRET state after dropping to 3 lower FRET levels (blue arrow in (a)) in the presence of NTP mix (grey; 843 events) and 4 after NTP wash-out (magenta; 284 events). d, Comparison of unscrunching and scrunching 5 rates in the presence of NTP mix and after NTP wash-out, measured as the inverse of average 6 dwell times in (b) and (c). e,f, FRET evolution maps generated from the traces supplied with 7 0.5 mM 3'dCTP after long stalling at position +7 by washing out the NTP mix, synchronized 8 at the moment of flowing in 3'dCTP (0 seconds). 53 and 41 traces were used for the maps 9 generated for DNA templates II and +11/−11, respectively. g, Using DNA template 10 +11/−11, FRET histogram of low-FRET events after NTP wash-out at position +7 (magenta) 11 was compared to that of position +8 (grey, from Fig. 2d).  a, Single-molecule measurements of transcription initiation dynamics. The dual-labeled DNA template complexed with Rpo41 and Mtf1 was observed using a total internal reflection fluorescence microscope. b, DNA templates used in basepair-wise measurements of initiation complex dynamics. DNA template I could be stalled at positions +2, +3, +5, and +6, while DNA template II, which differed from DNA template I by four basepairs (blue), could be stalled at positions +7 and +8. Both templates were labeled with Cy5 at position −16 of the non-template strand (magenta) and Cy3 at position +16 of template strand (green). The transcription promoter (underscored) and start site (arrow) are indicated. c, FRET histograms from singlemolecule traces with colocalized Cy3 and Cy5 signals at each stalling position. Histograms were fit to single, double, or triple Gaussian peaks. The brown, green, and magenta curves represent low, mid, and high FRET populations, respectively. d, Representative smFRET traces at positions 0, +2, and +6 showing the Cy3 (green) and Cy5 (magenta) signals and the FRET efficiency traces (navy). e, The FRET level of the major population in (c) shown for each stalling position as the center of the major Gaussian peak. Error bars represent the error in Gaussian fitting. f, The Cy3-to-Cy5 distance at each stalling position calculated from the average between major FRET levels from DNA templates I/II (e) and I/II NT (Supplementary Fig. 1c). Error bars represent propagation of the errors in FRET levels.     shown where the initiation bubble has collapsed and the −4 to −1 region is reannealed, resulting in the quenching of 2AP fluorescence. b, Design of DNA templates used for 2AP fluorescence measurements, to be stalled at positions +7, +8, +9, and +10. c, Changes in 2AP fluorescence measured along in vitro transcription reactions to the indicated positions, normalized against the initial intensity. The intensity traces were fit to a single exponential decay curve to determine the transition rates. d, Fluorescence decay rates measured from (c) at different walking positions. e, Models of the IC0, IC7, and EC9 structures generated using PyMOL (Schrödinger, USA). IC0 was modeled using PDB 6erp (human mitochondrial RNA polymerase initiation complex), IC7 was modeled using PDB 3e2e (bacteriophage T7 RNA polymerase initiation complex with 7 bp RNA:DNA), and EC9 was modeled using PDB 4boc (human mitochondrial RNA polymerase elongation complex with 9 bp RNA:DNA). The green and red balls represent the Cy3 and Cy5 fluorophores at positions +11 and −11, respectively. The double-stranded DNA and RNA:DNA hybrid (RNA in green) is highlighted as bound to the protein in the background.   a, FRET histograms obtained at equilibrium and after washing out the NTP mixture for positions +2, +7, and +8, and run-off conditions. For position +8, results from DNA template +11/−11 are included to distinguish between the populations of the elongation complex and DNA only. Each histogram was obtained from 12 short movies taken during each minute after washing out the NTP mixture. b,c, The relative population of the high FRET state (scrunched complex) obtained from histograms at positions +2 (b) and +7 (c) shown relative to time. Each graph was fit to a single exponential decay curve, and the half-life is shown. d, The relative population of the mid FRET state (elongation complex) obtained from histograms at position +8 on DNA template +11/−11 shown relative to time. e, The relative population of the mid FRET state (open complex) obtained from histograms under run-off conditions shown relative to time.  a, Representative smFRET trace showing the conformational dynamics of the TIC of DNA template II equilibrated at position +7 (gray region; 0.5 mM each of ATP, GTP, and UTP) after washing out the NTP mix (first arrow). Subsequently, 0.5 mM 3'dCTP was added to promote progression to position +8 (second arrow). The abrupt drop in the FRET efficiency indicates successful progression to position +8. b, Dwell time histogram of the high FRET (scrunched) state in the presence of NTP mix (grey; 809 events) and after NTP wash-out (magenta; 214 events). c, The time taken to recover the high FRET state after dropping to lower FRET levels (blue arrow in (a)) in the presence of NTP mix (grey; 843 events) and after NTP wash-out (magenta; 284 events). d, Comparison of unscrunching and scrunching rates in the presence of NTP mix and after NTP wash-out, measured as the inverse of average dwell times in (b) and (c). e,f, FRET evolution maps generated from the traces supplied with 0.5 mM 3'dCTP after long stalling at position +7 by washing out the NTP mix, synchronized at the moment of flowing in 3'dCTP (0 seconds). 53 and 41 traces were used for the maps generated for DNA templates II and +11/−11, respectively. g, Using DNA template +11/−11, FRET histogram of low-FRET events after NTP wash-out at position +7 (magenta) was compared to that of position +8 (grey, from Fig. 2d).  Conformational states and their transition kinetics identified in this study were integrated into the model. Measured transition rates were marked which were reflected in the thickness of arrows. Blank arrows represent indistinguishable transitions or transitions whose rates depend on NTP concentration. Unidentified states (IC6 unscrunched ) or molecules (Mtf1 in EC8) were expressed semi-transparent. DNA/RNA oligos and proteins were drawn as in Fig. 1a