The mechanism of variability in transcription start site selection

During transcription initiation, RNA polymerase (RNAP) binds to promoter DNA, unwinds promoter DNA to form an RNAP-promoter open complex (RPo) containing a single-stranded ‘transcription bubble,’ and selects a transcription start site (TSS). TSS selection occurs at different positions within the promoter region, depending on promoter sequence and initiating-substrate concentration. Variability in TSS selection has been proposed to involve DNA ‘scrunching’ and ‘anti-scrunching,’ the hallmarks of which are: (i) forward and reverse movement of the RNAP leading edge, but not trailing edge, relative to DNA, and (ii) expansion and contraction of the transcription bubble. Here, using in vitro and in vivo protein-DNA photocrosslinking and single-molecule nanomanipulation, we show bacterial TSS selection exhibits both hallmarks of scrunching and anti-scrunching, and we define energetics of scrunching and anti-scrunching. The results establish the mechanism of TSS selection by bacterial RNAP and suggest a general mechanism for TSS selection by bacterial, archaeal, and eukaryotic RNAP.


Introduction
During transcription initiation, RNA polymerase (RNAP) and one or more transcription initiation factor bind to promoter DNA through sequence-specific interactions with core promoter elements, unwind a turn of promoter DNA to form an RNAP-promoter open complex (RPo) containing an unwound 'transcription bubble,' and select a transcription start site (TSS). The distance between core promoter elements and the TSS can vary. TSS selection is a multi-factor process, in which the outcome reflects the contributions of promoter sequence and reaction conditions. TSS selection by bacterial RNAP and the bacterial transcription initiation factor s involves four promoter-sequence determinants: (i) distance relative to the promoter À10 element (preference for TSS selection at the position 7 bp downstream of the promoter À10 element; Aoyama and Takanami, 1985;Sørensen et al., 1993;Jeong and Kang, 1994;Liu and Turnbough, 1994;Walker and Osuna, 2002;Lewis and Adhya, 2004;Vvedenskaya et al., 2015;Winkelman et al., 2016a;Winkelman et al., 2016b); (ii) identities of the template-strand nucleotide at the TSS and the template-strand nucleotide immediately upstream of the TSS (strong preference for a template-strand pyrimidine at the TSS and preference for a template-strand purine immediately upstream of the TSS; Aoyama and Takanami, 1985;Sørensen et al., 1993;Jeong and Kang, 1994;Liu and Turnbough, 1994;Walker and Osuna, 2002;Lewis and Adhya, 2004;Vvedenskaya et al., 2015;Winkelman et al., 2016a;Winkelman et al., 2016b); (iii) the promoter 'core recognition element,' a segment of nontemplate-strand sequence spanning the TSS that interacts sequence-specifically with RNAP (preference for nontemplate-strand G immediately downstream of the TSS; Vvedenskaya et al., 2016), and (iv) the promoter 'discriminator element,' a nontemplate-strand sequence immediately downstream of the promoter À10 element that interacts sequence-specifically with s (preference for TSS selection at upstream positions for purine-rich discriminator sequences, and preference for TSS selection at downstream positions for pyrimidine-rich discriminator sequences; Winkelman et al., 2016aWinkelman et al., , 2016b. In addition to these four promoter-sequence determinants, the concentrations of initiating NTPs (Sørensen et al., 1993;Liu and Turnbough, 1994;Walker and Osuna, 2002;Vvedenskaya et al., 2015;Wilson et al., 1992;Qi and Turnbough, 1995;Tu and Turnbough, 1997;Walker et al., 2004;Turnbough, 2008;Turnbough and Switzer, 2008) and DNA topology (Vvedenskaya et al., 2015) also influence TSS selection.
It has been hypothesized that variability in the distance between core promoter elements and the TSS is accommodated by DNA 'scrunching' and 'anti-scrunching,' the defining hallmarks of which are: (i) forward and reverse movements of the RNAP leading edge, but not the RNAP trailing edge, relative to DNA, and (ii) expansion and contraction of the transcription bubble (Vvedenskaya et al., 2015;Winkelman et al., 2016aWinkelman et al., , 2016bVvedenskaya et al., 2016;Robb et al., 2013). In previous work, we showed that TSS selection exhibits the first hallmark of scrunching in vitro (Winkelman et al., 2016a). Here, we show that TSS selection also exhibits the first hallmark of scrunching in vivo, show that TSS selection exhibits the second hallmark of scrunching and antiscrunching, and define the energetics of scrunching and anti-scrunching.
eLife digest Genes store the information needed to build and repair cells. This information is written in a chemical code within the structure of DNA molecules. To make use of the information, cells copy sections of a gene into a DNA-like molecule called RNA. An enzyme called RNA polymerase makes RNA molecules from DNA templates in a process called transcription. RNA polymerase can only make RNA by attaching to DNA and separating the two strands of the DNA double helix. This creates a short region of single-stranded DNA known as a "transcription bubble".
RNA polymerase can start transcription at different distances from the sites where it initially attaches to DNA, depending on the DNA sequence and the cell's environment. It had not been known how RNA polymerase selects different transcription start sites in different cases. One hypothesis had been that differences in the size of the transcription bubble -the amount of unwound single-stranded DNA -could be responsible for differences in transcription start sites. For example, RNA polymerase could increase the size of the bubble through a process called "DNA scrunching", in which RNA polymerase pulls in and unwinds extra DNA from further along the gene.
Yu, Winkelman et al. looked for indicators of DNA scrunching to see whether it contributes to the selection of transcription start sites. By mapping the positions of the two edges of RNA polymerase relative to DNA, they saw that RNA polymerase pulls in extra DNA when selecting a transcription start site further from its initial attachment site. Next, by measuring the amount of DNA unwinding, they saw that RNA polymerase unwinds extra DNA when it selects a transcription start site further from its initial attachment site. This was the case for both RNA polymerase in a test tube and RNA polymerase in living bacterial cells. The results showed that DNA scrunching accounts for known patterns of selection of transcription start sites.
The findings hint at a common theory for the selection of transcription start sites across all life by DNA scrunching. Understanding these basic principles of biology reveals more about how cells work and how cells adapt to changing conditions. The experimental methods developed for mapping the positions of proteins on DNA and for measuring DNA unwinding will help scientists to learn more about other aspects of how DNA is stored, copied, read, and controlled.

Results and discussion
TSS selection exhibits first hallmark of scrunching-movements of RNAP leading edge but not RNAP trailing edge-both in vitro and in vivo In our prior work, we demonstrated that bacterial TSS selection in vitro exhibits the first hallmark of scrunching by defining, simultaneously, the TSS, the RNAP leading-edge position, and RNAP trailing-edge position for transcription complexes formed on a library of 10 6 promoter sequences (Winkelman et al., 2016a). We used RNA-seq to define the TSS, and we used unnatural-amino-acidmutagenesis, incorporating the photoactivatable amino acid p-benzoyl-L-phenylalanine (Bpa), and protein-DNA photocrosslinking to define RNAP-leading-edge and trailing-edge positions (Winkelman et al., 2016a). The results showed that the discriminator element (Haugen et al., 2006;Feklistov et al., 2006) influences TSS selection and does so through effects on sequence-specific s-DNA interaction that select between two alternative paths of the DNA nontemplate strand (Winkelman et al., 2016a). The results further showed that, as the TSS changes for different discriminator sequences, the RNAP-leading-edge position changes, but the RNAP-trailing-edge position does not change (Winkelman et al., 2016a). For example, replacing a GGG discriminator by a CCT discriminator causes a 2 bp downstream change in TSS (from the position 7 bp downstream of the À10 element to the position 9 bp downstream of the À10 element, due to differences in sequencespecific s-DNA interaction that result in different paths of the DNA nontemplate strand), causes a 2 bp downstream change in RNAP leading-edge position, but does not cause a change in RNAP trailing-edge position ( Figure 1A).
Here, to determine whether bacterial TSS selection in vivo also exhibits the first hallmark of scrunching, we adapted the above unnatural-amino-acid-mutagenesis and protein-DNA-photocrosslinking procedures to define RNAP leading-edge and trailing-edge positions in TSS selection in living cells (Figure 1, Figure 1-figure supplements 1-2). We developed approaches to assemble, trap, and UV-irradiate RPo formed by a Bpa-labeled RNAP derivative in living cells, to extract crosslinked material from cells, and and to map crosslinks at single-nucleotide resolution (Figure 1-figure supplement 1). In order to assemble, trap, and UV-irradiate RPo in living cells, despite the presence of high concentrations of initiating substrates that rapidly convert RPo into transcribing complexes, we used a mutationally inactivated RNAP derivative, b'D460A, that lacks a residue required for binding of the RNAP-active-center catalytic metal ion and initiating substrates (Zaychikov et al., 1996) (Figure 1-figure supplements 1-2). Control experiments confirm that, in vitro, in both the absence and presence of initiating substrates, the mutationally inactivated RNAP derivative remains trapped in RPo, exhibiting the same pattern of leading-edge and trailing-edge crosslinks as for wild-type RNAP in the absence of initiating substrates (Figure 1-figure supplement 2). In order to introduce Bpa at the leading-edge and trailing-edge of RPo in living cells, we co-produced, in Escherichia coli, a Bpa-labeled, decahistidine-tagged, mutationally inactivated RNAP derivative in the presence of unlabeled, untagged, wild-type RNAP, using a three-plasmid system comprising (i) a plasmid carrying a gene for RNAP b' subunit that contained a nonsense codon at the site for incorporation of Bpa, the b'D460A mutation, and a decahistidine coding sequence; (ii) a plasmid carrying genes for an engineered Bpa-specific nonsense-suppressor tRNA and an engineered Bpa-specific aminoacyl-tRNA synthase (Chin et al., 2002); and (iii) a plasmid containing a promoter of interest ( Figure 1-figure supplement 1A). (Using this merodiploid system, with both a plasmid-borne mutant gene for b' subunit and a chromosomal wild-type gene for b' subunit, enabled analysis of the mutationally inactivated RNAP derivative without loss of viability.) In order to perform RNAP-DNA crosslinking and to map resulting RNAP-DNA crosslinks, we then grew cells in medium containing Bpa, UV-irradiated cells, lysed cells, purified crosslinked material using immobilized metal-ion-affinity chromatography targeting the decahistidine tag on the Bpa-labeled, decahistidinetagged, mutationally inactivated RNAP derivative, and mapped crosslinks using primer extension ( Figure 1-figure supplement 1B). The results showed an exact correspondence of crosslinking patterns in vitro and in vivo ( Figure 1B, 'in vitro' vs. 'in vivo' lanes). The RNAP leading edge crosslinked 2 bp further downstream on CCT than on GGG, whereas the RNAP trailing edge crosslinked at the same positions on CCT and GGG ( Figure 1B). We conclude that TSS selection in vivo shows the first hallmark of scrunching.  The following figure supplements are available for figure 1: The results in Figure 1 establish that TSS selection in vitro and in vivo exhibits the first hallmark of scrunching. However, definitive demonstration that TSS selection involves scrunching also requires demonstration of the second hallmark of scrunching: that is, changes in transcription-bubble size. To determine whether bacterial TSS selection exhibits the second hallmark of scrunching we used a magnetic-tweezers single-molecule DNA-nanomanipulation assay that enables detection of RNAPdependent DNA unwinding with near-single-base-pair spatial resolution and sub-second temporal resolution (Revyakin et al., 2004(Revyakin et al., , 2005(Revyakin et al., , 2006 to assess whether TSS selection correlates with transcription-bubble size for GGG and CCT promoters ( Figure 2). The results indicate that transition amplitudes for RNAP-dependent DNA unwinding upon formation of RPo with CCT are larger than those for formation of RPo with GGG, on both positively and negatively supercoiled templates ( Figure 2B, left and center). Transition-amplitude histograms confirm that transition amplitudes with CCT are larger than with GGG, on both positively and negatively supercoiled templates ( Figure 2B, right). By combining the results with positively and negatively supercoiled templates to deconvolve effects of RNAP-dependent DNA unwinding and RNAP-dependent compaction (Revyakin et al., 2004(Revyakin et al., , 2005(Revyakin et al., , 2006, we find a 2 bp difference in RNAP-dependent DNA unwinding for CCT vs. GGG ( Figure 2C), corresponding exactly to the 2 bp difference in TSS selection ( Figure 1B). We conclude that TSS selection shows the second hallmark of scrunching.
TSS selection downstream and upstream of modal TSS involves scrunching and anti-scrunching, respectively: forward and reverse movements of RNAP leading edge According to the hypothesis that TSS selection involves scrunching or anti-scrunching, TSS selection at the most frequently observed, modal TSS position (7 bp downstream of À10 element for majority of discriminator sequences, including GGG) involves neither scrunching nor anti-scrunching, TSS selection downstream of the modal position involves scrunching (transcription-bubble expansion), and TSS selection upstream of the modal position involves anti-scrunching (Vvedenskaya et al., 2015;Winkelman et al., 2016aWinkelman et al., , 2016bVvedenskaya et al., 2016;Robb et al., 2013). The results in . We analyzed a consensus bacterial promoter, lac-CONS, and used four ribotrinucleotide primers, UGG, GGA, GAA, and AAU, to program TSS selection at positions 6, 7, 8, and 9 bp downstream of the À10 element ( Figure 3A). Experiments analogous to those in Figure 1 show a one-for-one, bp-for-bp correlation between primer-programmed changes in TSS and changes in RNAP-leading-edge position. The leading-edge crosslink positions with the four primers differed in single-nucleotide increments, but the trailing-edge crosslink positions were the same ( Figure 3B). With the primer GGA, which programs TSS selection at the modal position (7 bp downstream of À10 element for this discriminator sequence), the leading-edge crosslinks were exactly as in experiments with no primer (Figure 3-figure supplement 1). With primers GAA and AAU, which program TSS selection 1 and 2 bp downstream (positions 8 and 9), leading-edge crosslinks were 1 and 2 bp downstream of crosslinks with GGA ( Figure 3B). With primer UGG, which programs TSS selection 1 bp upstream (position 6), leading-edge crosslinks were 1 bp upstream of crosslinks with GGA ( Figure 3B). The results show that successive single-base-pair changes in TSS selection are matched by successive single-base-pair changes in RNAP leading-edge Ribotrinucleotide primers program TSS selection at positions 6, 7, 8, and 9 bp downstream of À10 element (UGG, GGA, GAA, and AAU). Cyan, green, orange, and red denote primers UGG, GGA, GAA, and AAU, respectively. Rectangle with rounded corners highlights case of primer GGA, which programs TSS selection at same position as in absence of primer (7 bp downstream of À10 element). Other colors as in Figure 1A. (B) Use of protein- Figure 3 continued on next page TSS selection downstream and upstream of modal TSS involves scrunching and anti-scrunching, respectively: increases and decreases in RNAP-dependent DNA unwinding We next used magnetic-tweezers single-molecule DNA-nanomanipulation to analyze primer-programmed TSS selection. To enable single-base-pair resolution, we reduced the DNA-tether length from 2.0 kb to 1.3 kb, thereby reducing noise due to compliance (Figure 4-figure supplement 1; see Revyakin et al., 2005). The resulting transition amplitudes, transition-amplitude histograms, and RNAP-dependent DNA unwinding values for TSS selection with saturating concentrations of the four primers show a one-for-one, base-pair-for-base-pair correlation between primer-programmed changes in TSS and changes in RNAP-dependent DNA unwinding ( Figure 4). With primer GGA, which programs TSS selection at the modal position (7 bp downstream of the À10 element for this discriminator sequence), DNA unwinding was exactly as in experiments with no primer (Figure 4figure supplement 2). With primers GAA and AAU, which program TSS selection 1 and 2 bp further downstream (positions 8 and 9), DNA unwinding was~1 and~2 bp greater than with GGA (Figure 4). With primer UGG, which programs TSS selection 1 bp upstream, DNA unwinding was~1 bp less than that in experiments with GGA ( Figure 4). The results show that successive single-base-pair changes in TSS selection are matched by successive single-base-pair changes in DNA unwinding for a full range of TSS positions including, importantly, a position upstream of the modal TSS expected to involve anti-scrunching. Taken

Energetic costs of scrunching and anti-scrunching
To quantify the energetic costs of scrunching and anti-scrunching, we measured primer-concentration dependences of lifetimes of unwound states (Figures 5-6, Figure 6-figure supplement 1). For each primer, increasing the primer concentration increases the lifetime of the unwound state (t unwound ), as expected for coupled equilibria of promoter unwinding, promoter scrunching, and primer binding ( Figures 5-6). The results in Figure 6D show that the slopes of plots of mean t unwound ( t unwound ) vs. primer concentration differ for different primers. Fitting the results to the equation describing the coupled equilibria ( Figure 6C) yields values of K NpNpN , DG NpNpN , K scrunch , and DG scrunch for the four primers ( Figure 6E, Figure 6-figure supplement 1). The results indicate that scrunching by 1 bp requires 0.7 kcal/mol, scrunching by 2 bp requires 1.7 kcal/mol, and anti-scrunching by 1 bp requires 1.8 kcal/mol ( Figure 6E, Figure 6-figure supplement 1).
The results provide the first experimental determination of the energetic costs of scrunching and anti-scrunching in any context. We hypothesize that energetic costs on the same scale,~0.7-1.8 kcal/mol per scrunched bp, also apply in the structurally and mechanistically related scrunching that occurs during initial transcription by RNAP (Revyakin et al., 2006;Kapanidis et al., 2006). We note that, according to this hypothesis, the scrunching by~10 bp that occurs during initial transcription (Revyakin et al., 2006) results in an increase in the state energy of the transcription initiation complex by a total of~7-18 kcal/mol (~10 x~0.7-1.8 kcal/mol). This is an increase in state energy potentially sufficient to yield a 'stressed intermediate' (Revyakin et al., 2006;Straney and Crothers, 1987) having scrunching-dependent 'stress' comparable to the free energies of RNAP-promoter and RNAP-initiation-factor interactions that anchor RNAP at a promoter (~7-9 kcal/mol for sequencespecific component of RNAP-promoter interaction and~13 kcal/mol for RNAP-initiation-factor position 7 (0-1.8 kcal/mol) all are less than or comparable to 3k B T (~2 kcal/mol), where k B is the Bolztmann constant and T is temperature in˚K, indicating that TSS selection at these positions requires no energy beyond energy available in the thermal bath. Indeed, the probabilities of TSS selection at positions 6, 7, 8, and 9 as observed in a comprehensive analysis of TSS-region sequences (8%, 55%, 29%, 7%; Vvedenskaya et al., 2015) can be predicted from the Boltzmann-distribution probabilities for the DG scrunch values for TSS selection at these positions (3%, 70%, 23%, 4%; Figure 6F). The finding that values of DG scrunch for scrunching and anti-scrunching in TSS selection are~1 kcal mol À1 bp À1 and~2 kcal mol À1 bp À1 , respectively, implies that TSS selection at positions >2 bp downstream or >1 bp upstream of the modal position would exceed the energy fluctuations available to 99% of molecules at 20-37˚C, and therefore explains the observation that TSS selection >2 bp downstream or >1 bp upstream of the modal position occurs rarely (Vvedenskaya et al., 2015).

Unified mechanism of TSS selection by multisubunit RNAP
TSS selection by archaeal RNAP, eukaryotic RNAP I, eukaryotic RNAP II from most species, and eukaryotic RNAP III involves the same range of TSS positions as TSS selection by bacterial RNAP (positions ± 2 bp from the modal TSS; Learned and Tjian, 1982;Samuels et al., 1984;Thomm and Wich, 1988;Reiter et al., 1990;Fruscoloni et al., 1995;Zecherle et al., 1996). We propose that TSS selection by all of these enzymes is mediated by scrunching and anti-scrunching driven by energy available in the thermal bath. In contrast, TSS selection by S. cerevisiae RNAP II involves a range of TSS positions of 10 s to 100 s of bp (long-range TSS scanning; Giardina and Lis, 1993;Kuehner and Brow, 2006). We propose that TSS scanning by S. cerevisiae RNAP II also is mediated by scrunching and anti-scrunching, but, in this case, involves not only energy from the thermal bath, but also energy from the ATPase activity of RNAP II transcription factor TFIIH (Sainsbury et al., 2015). This proposal could account for the ATP-dependent, TFIIH-dependent cycles of DNA compaction and de-compaction of 10 s to 100 s of bp observed in single-molecule optical-tweezer analyses of TSS scanning by S. cerevisiae RNAP II (Fazal et al., 2015).  (Svetlov and Artsimovitch, 2015) as described (Winkelman et al., 2015). Wild-type RNAP for single-molecule DNA-nanomanipulation experiments was prepared from E. coli strain BL21(DE3) (New England Biolabs) transformed with plasmid pVS10 (Artsimovitch et al., 2003) as described (Artsimovitch et al., 2003). Bpa-containing RNAP core-enzyme derivatives for in vitro protein-DNA photocrosslinking (b'R1148Bpa for analysis of RNAP leading-edge positions; b'T48Bpa for analysis of RNAP trailingedge positions) were prepared from E. coli strain NiCo21(DE3) (New England BioLabs) co-transformed with plasmid pEVOL-pBpF (Chin et al., 2002; Addgene) and plasmid pIA900-b'R1148Bpa (Winkelman et al., 2015) or pIA900-b'T48Bpa (Winkelman et al., 2015), as in Winkelman et al. (2015).

Determination of TSS position in vivo
E. coli strain NiCo21(DE3) (New England BioLabs) transformed with plasmid pCDF-CP-lacCONS-GGG or pCDF-CP-lacCONS-CCT was plated on LB agar (Sambrook and Russell, 2001) containing 50 mg/ml spectinomycin and 50 mg/ml streptomycin, single colonies were inoculated into 25 ml LB broth (Sambrook and Russell, 2001) containing 50 mg/ml spectinomycin and 50 mg/ml streptomycin in 125 ml Bellco flasks, and cultures were shaken (220 rpm) at 37˚C. When cell densities reached OD 600 = 0.6, 2 ml aliquots were centrifuged 2 min at 4˚C at 23,000xg, and resulting cell pellets were frozen at À80˚C. Cell pellets were thawed in 1 ml TRI Reagent (Molecular Research Center) at 25˚C for 5 min, completely re-suspended by pipetting up and down, incubated 10 min at 70˚C, and centrifuged 2 min at 25˚C at 23,000 x g. Supernatants were transferred to fresh 1.7 ml microfuge tubes, 200 ml chloroform (Ambion) was added, vortexed, and samples were centrifuged 1 min at 25˚C at 23,000 x g. Aqueous phases were transferred to a fresh tube and nucleic acids were extracted with acid phenol:chloroform (Sambrook and Russell, 2001). Nucleic acids were recovered by ethanol precipitation (Sambrook and Russell, 2001), and re-suspended in 20 ml 10 mM Tris-Cl, pH 8.0. Primer extension was performed as described in the preceding section.
Determination of RNAP leading-edge and trailing-edge positions in vivo: protein-DNA photocrosslinking in vivo Experiments in Figure 1B were performed using a three-plasmid merodiploid system that enabled production of a Bpa-containing, mutationally inactivated, decahistidine-tagged RNAP holoenzyme derivative in vivo and enabled trapping of RPo consisting of the Bpa-containing, mutationally inactivated, decahistidine-tagged RNAP holoenzyme derivative and a lacCONS promoter with GGG or CCT discriminator in vivo, and UV-irradiation of cells (Figure 1-figure supplement 1).
DNA constructs for magnetic-tweezers single-molecule DNA-nanomanipulation were prepared from the above 2.0 kb and 1.3 kb DNA fragments by ligating, at the XbaI end, a 1.0 kb DNA fragment bearing multiple biotin residues on both strands [prepared by PCR amplification of plasmid pARTaqRPOC-lacCONS using primers 'XbaRPOC4050' and 'RPOC3140' Supplementary file 1) and conditions as described (Revyakin et al., 2004, 2005, Revyakin et al., 2003 Determination of RNAP-dependent DNA unwinding by single-molecule DNA-nanomanipulation: data collection Experiments were performed essentially as described (Revyakin et al., 2004, 2005, Revyakin et al., 2003 Experiments in Figure 2 (experiments addressing TSS selection for promoters with GGG or CCT discriminator sequence), were performed using standard reactions containing mechanically extended, torsionally constrained, 2.0 kb DNA molecule carrying GGG or CCT promoter (extension force = 0.3 pN; superhelical density = 0.021 for experiments with positively supercoiled DNA; superhelical density = À0.021 for experiments with negatively supercoiled DNA) and RNAP holoenzyme (10 nM for experiments with positively supercoiled DNA; 0.5 nM for experiments with negatively supercoiled DNA) in 25 mM Na-HEPES, pH 7.9, 75 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol, 0.1% Tween-20, 0.1 mg/ml bovine serum albumin) at 30˚C. Data from each of three single DNA molecules were pooled [differences in plectoneme size (at superhelical density ± 0.021) and D l obs 5%].
Experiments in Figure 4-figure supplement 1 (experiments demonstrating that reduction in DNA-fragment length from 2.0 to 1.3 kb enables single-bp resolution) were performed using standard reactions containing mechanically extended, torsionally constrained, 2.0 kb or 1.3 kb DNA molecule carrying lacCONS promoter (extension force = 0.3 pN; initial superhelical density = 0.021 or 0.024 for experiments with 2.0 kb or 1.3 kb positively supercoiled DNA; superhelical density = À0.021 or À0.024 for experiments with 2.0 kb or 1.3 kb negatively supercoiled DNA) and RNAP holoenzyme (10 nM for experiments with positively supercoiled DNA; 0.5 nM for experiments with negatively supercoiled DNA) in the buffer of the preceding paragraph at 30˚C. For each DNAfragment length, data were collected on one single DNA molecule. Experiments in Figure 4 and Figure 4-figure supplement 2 (experiments addressing primerprogrammed TSS selection with primers UGG, GGA, GAA, AAU) were performed using standard reactions containing mechanically extended, torsionally constrained 1.3 kb DNA molecule carrying lacCONS promoter (extension force = 0.3 pN; initial superhelical density = 0.024 for experiments with positively supercoiled DNA; superhelical density = À0.024 for experiments with negatively supercoiled DNA) and RNAP holoenzyme (10 nM for experiments with positively supercoiled DNA; 0.5 nM for experiments with negatively supercoiled DNA) in the buffer of the preceding paragraph at 30˚C. Primers UGG, GGA, GAA, and AAU were present at 0 or 1 mM, 0 or 1 mM, 0 or 2.5 mM, and 0 or 1 mM, respectively. For experiments with positively supercoiled DNA, data from each of seven single DNA molecules were normalized based on D l obs,pos in absence of primer and pooled; for experiments with negatively supercoiled DNA, data from each of two single DNA molecules were normalized based on D l obs,neg in absence of primer and pooled.
For experiments with negatively supercoiled DNA, for which t unwound >> 1 h (Revyakin et al., 2004, 2005, Revyakin et al., 2003 Determination of RNAP-dependent DNA unwinding by single-molecule DNA-nanomanipulation: data reduction for determination of DNA unwinding Raw time traces were analyzed to yield DNA extension (l) as described (Revyakin et al., 2004, 2005, Revyakin et al., 2003 Changes in l attributable to DNA unwinding (Dl u ) and changes in l attributable to DNA compaction (Dl c ) were calculated as: Dl u = (Dl obs,neg + Dl obs,pos )/2, and Dl c = (Dl obs,pos -Dl obs,neg )/2, where Dl obs,pos and Dl obs,neg are observed changes in l in experiments with positively supercoiled DNA and negatively supercoiled DNA, as described (Revyakin et al., 2004, 2005, Revyakin et al., 2003 Determination of RNAP-dependent DNA unwinding by single-molecule DNA-nanomanipulation: data reduction for determination of energetics of scrunching and anti-scrunching Lifetimes of unwound states (t unwound ) were extracted from single-molecule traces as described (Revyakin et al., 2004, 2005, Revyakin et al., 2003 For experiments in absence of primer (Qi and Turnbough, 1995