Single-molecule Approaches to Probe the Structure, Kinetics, and Thermodynamics of Nucleoprotein Complexes That Regulate Transcription*

Single-molecule experimentation has contributed significantly to our understanding of the mechanics of nucleoprotein complexes that regulate epigenetic switches. In this minireview, we will discuss the application of the tethered-particle motion technique, magnetic tweezers, and atomic force microscopy to (i) directly visualize and thermodynamically characterize DNA loops induced by the lac, gal, and λ repressors and (ii) understand the mechanistic role of DNA-supercoiling and DNA-bending cofactors in both prokaryotic and eukaryotic systems.

Single-molecule experimentation has contributed significantly to our understanding of the mechanics of nucleoprotein complexes that regulate epigenetic switches. In this minireview, we will discuss the application of the tethered-particle motion technique, magnetic tweezers, and atomic force microscopy to (i) directly visualize and thermodynamically characterize DNA loops induced by the lac, gal, and repressors and (ii) understand the mechanistic role of DNA-supercoiling and DNAbending cofactors in both prokaryotic and eukaryotic systems.
Transcriptional regulation involving protein-mediated DNA bending, wrapping, and looping occurs ubiquitously in all organisms. By enabling long-distance interaction between transcription factors, limiting the accessibility to, and/or mechanically deforming promoters, nucleoprotein complexes can tune transcription effectively. In other words, protein-mediated bends, wraps, or loops in DNA constitute elements for transcriptional regulation. Some of the best known examples of DNA bends are induced by the IHF 2 protein in prokaryotes (1) and by the HMG1 protein in eukaryotes (2). The wrapping of DNA around histone octamers is a fundamental structure in eukaryotic chromatin (3), but most, if not all, DNA-binding proteins are hypothesized to be able to wrap and thereby organize flanking regions of DNA (4). Examples of DNA looping include the large chromatin loops proposed to explain insulator domains in eukaryotes (5) and the loops induced by prokaryotic repressors, such as the lac and gal repressors (see below).
Although these examples pertain to transcriptional regulation, the same conformational changes are common in the regulation of most DNA transactions, such as replication, recombination, etc. Therefore, characterizing the kinetics and thermody-namics of the formation of nucleoprotein complexes involving DNA bending, wrapping, or looping, as well as their structure and stoichiometry, is paramount to understand their functions. Enhanced understanding of epigenetic regulation also might enable the design of mechanistic mutations for biomedical applications. Single-molecule experiments are incisive means with which to explore protein-induced conformational changes in DNA because they avoid ensemble averaging and reveal heterogeneities that might go undetected in bulk measurements. The following is a review of recent TPM, MT, and AFM experiments on nucleoprotein complexes that regulate transcription.

TPM, MT, AFM, and Advantages of Single-molecule Experiments
Single-molecule assays yield distributions of individual measurements instead of averages for large ensembles of molecules. This provides a wealth of information and often reveals heterogeneous behaviors obscured in bulk experiments. Such discriminatory power has produced new insight even into "well understood" paradigms, such as prokaryotic transcriptional repressors (see below). Single-molecule microscopy and spectroscopy can be used to distinguish protein-induced DNA bending, looping, and wrapping, and they permit facile control of DNA supercoiling if required. Especially powerful is the combination of several techniques to relate structural and dynamic information on broad spatial and temporal ranges. In the study of transcriptional regulation, TPM, MT, and AFM have been combined effectively and will be described briefly below.
In TPM, a submicron diameter bead is tethered by a single DNA molecule to a glass surface. The bead labels the mobile end of the unlabeled DNA and exhibits Brownian motion restricted by the DNA tether. If the DNA tether undergoes topological changes that alter the effective length, the Brownian excursions of the bead, measured as the projected displacement vector, Ќ (Fig. 1a), will change correspondingly (6). Stochastic, protein-mediated DNA loop formation and breakdown will therefore generate a telegraph-like TPM signal over time (7,8). The experimental setup and analysis of TPM measurements have recently been the object of careful studies. TPM and MT, which also rely on the observation of the Brownian motion of microscopic beads, can reveal tether shortenings of 10 -20 nm and events lasting 0.1 s (9). In addition, careful characterization of the effect of the bead size, DNA tether length, and entropic forces on the measured signal (10 -13) has led to novel and improved methods of analysis of TPM data (14,15). Finally, small beads may be used with contrast methods based on dark field illumination (16 -18) to reduce distortions in the conformational distributions of the DNA tether due to the hydrodynamics and surface effects of larger beads.
Beyond simple measurement of the DNA tether length, MT allow control of DNA supercoiling and tension to restrict more narrowly the conformations of the DNA substrate available to the protein. MT most commonly consist of a pair of magnets mounted above the microscope stage that can be both trans-lated along and rotated around the optical axis of the microscope. The magnetic field both attracts and rotationally constrains a DNA-tethered paramagnetic bead to enable stretching and twisting of the DNA tether ( Fig. 1b) (19). Gentle tension applied to the DNA tether using MT allows manipulation of thermodynamic equilibria using force. New configurations of optical tweezers also allow the application of subpiconewton tension in a constant force regime to short, axially oriented DNA (20).
Neither of these tethered-particle methods can discern features of the interacting proteins. Instead, AFM has been extremely useful in understanding DNA-protein complexes through visualization of the site of interaction, the conformational change induced, and quantification of the extent of DNA bending and/or looping (Fig. 1c). For AFM (21), specimens are deposited on an atomically flat surface, usually mica. Then, under liquids or in ambient pressure gas environments, the surface is scanned with a sharp stylus to reveal topography. For biological samples, the average radius of curvature of an AFM tip is 5-10 nm, so the spatial resolution can be 3-5 nm. Images are usually acquired at a scan speed of 1 m/s, although a few images at video rates have been reported (22). The MT, AFM, and optical trapping techniques have been reviewed recently (23).

Specific DNA Looping and Flexibility of the Closure Protein
Proteins that regulate transcription secure DNA loops in many different ways of varying complexity. The simplest and best known loop-based epigenetic switch is the repression of the lac operon in Escherichia coli (24 -26). It is mediated by a single tetrameric lac repressor (LacI, M r(tetramer) Ӎ 150,000), which has two DNA-binding domains. LacI can recognize any of three binding sites (operators) centered at ϩ11 (O 1 ), ϩ400 (O 2 ), and Ϫ82 (O 3 ) relative to the transcription start site of the lac operon (Fig. 2). Although LacI binding at O 1 sterically blocks RNA polymerase transcription of lactose-metabolizing enzymes (26), repressor binding and repression efficiency increase upon the formation of a ternary complex that includes a DNA loop (27,28). Attempts to demonstrate the existence of such a ternary complex using conventional biochemistry assays were not definitive because interactions and motions are not synchronized in the ensembles of molecules.
Instead, TPM and AFM experiments have rendered unequivocal evidence of DNA loop formation by LacI and greatly refined our knowledge of this paradigmatic epigenetic switch. TPM was used first to monitor directly in vitro repressor-mediated DNA loop formation and breakdown between two copies of the O 1 operator separated by 305 bp (Fig. 2). The excursions of beads attached to DNA tethers were monitored over time, and the probabilities and average lifetimes of the two DNA conformations were determined after establishing a . If a protein (green complex) induces looping, the tether shortens and reduces the amplitude of the motion of the bead. When the DNA molecule changes conformation due to protein association or dissociation (right), the amplitude of the Brownian motion, Ќ , will fluctuate between two levels, resembling a telegraphic signal (left). b, two magnets (N and S) above the sample can be translated vertically along the microscope optical axis (orange double-headed arrow) and/or rotated (orange dotted arrow). Because the paramagnetic bead aligns rigidly in the magnetic field (gray dashed lines), one turn of the magnets causes one turn of the bead, which twists the DNA tether (black curved line). Torsional stress in the DNA induces supercoiling, which can be detected as a change in DNA end-to-end distance (L). The white patches on the DNA indicate putative operators. Spatial resolution varies depending on DNA tether length, loop size, and external tension. c, left, a pair of octamers of CI protein secure a loop in 1555-bp DNA with operators separated by 400 bp (scale bar ϭ 100 nm). Middle, topography in the image is color-coded as shown. Right, shown is a schematic view of the AFM stylus deflecting to reveal a height, h, during a trace (white dotted arrow in the image to the left) across a DNA double helix. The stylus is not drawn to scale. threshold value to separate the two observed states. The dwell times of the LacI-mediated DNA loops were exponentially distributed, with an average lifetime of 58 s, whereas the distribution of lifetimes of the unlooped DNA conformation in the presence of repressor decayed as a double exponential, with time constants of 77 and 5 s, respectively (7). The double exponential indicated that there are two routes to forming the looped DNA conformation. This early study also indicated that the primary mechanism for loop breakdown was not tetramer dissociation into two bound dimers, but tetramer dissociation from operator DNA (7). Further TPM experimentation (11,29) accompanied by AFM imaging (30) showed that the loop is preferentially antiparallel and that LacI may adopt two structures in dynamic equilibrium: the "V" conformation observed in crystallographic studies and a more extended conformation that considerably relieves the strain due to DNA bending. Protein flexure may facilitate the ability to switch between different structures in simple short loops. This may sustain transcriptional regulation even as the mechanical work required to form a loop changes with metabolic conditions (30) because slight tension significantly reduces the propensity of LacI-mediated looping (29).
These single-molecule studies on LacI-mediated DNA looping showed that (i) protein-mediated looping can be detected directly and its probability measured; (ii) TPM data expose the relatively simple kinetics of loop formation and breakdown by one protein; (iii) this kinetic information, together with AFM imaging, yields insight into a paradigmatic mechanism of transcriptional regulation; and (iv) TPM data provide insight into how DNA stiffness affects the formation of bent topologies. However, a protein molecule rarely acts alone, and a variety of synergistic and cooperative interactions are known among transcriptional regulators. In particular, protein-induced DNA bending may either enhance or obstruct DNA looping by other proteins. Thus, understanding the concerted actions of multiple proteins on DNA topology is fundamental.

DNA Bending in Specific Loops of Different Sizes: Experiments with IHF
IHF is a nucleoid-associated protein involved in several cellular processes, including transcription (31)(32)(33)(34)(35)(36). IHF has been called the "master bender" (37) and bends DNA by 150 -180°. IHF-induced bending may play a role in the regulation of a variety of promoters by facilitating protein-protein interactions between RNA polymerase and upstream activators (38,39) or between proteins that secure regulatory DNA loops. Complementary TPM and AFM measurements can be used to probe the effect of the bend introduced by IHF on the formation of loops, as was demonstrated in a recent proof-of-principle study (40). There, a binding site for IHF was placed between two lac operators, and the probability of DNA looping mediated by LacI was characterized as a function of operator separation. The IHF-induced bends changed the probability of LacI-mediated DNA looping to different degrees according to the position of the bend and the size of the loop. Instead, compaction of DNA by high concentrations of IHF enhanced or attenuated the formation of short or long loops, respectively. Although further experimentation is needed to determine in detail the effect of a bender on protein-mediated DNA regulatory loops of various lengths, the role of the HU (heat-unstable) protein in the naturally occurring gal repressor-mediated DNA loop was well characterized (41).

Bending DNA in a Small Specific Supercoiled Loop: HU and the gal Repressor
The bacterial HU protein (a nucleoid-associated protein) bends DNA, prefers single-to double-stranded DNA, and is an accessory factor in the repression of the gal operon in E. coli (42). Transcription of this operon is repressed by the gal repressor protein (GalR), which binds as a dimer to two sites, O E and O I . These operators encompass two partially overlapping promoters and are separated by only 113 bp (Fig. 2). Efficient repression is achieved only with both GalR and HU and requires DNA supercoiling (43).
To study the mechanism of GalR-mediated repression, Lia et al. (41) used MT (Fig. 1b) to supercoil single DNA molecules. Observing individual fragments of gal DNA subjected to different degrees and handedness of supercoiling and comparing these results with in vitro transcription assays, it was possible to conclude the following. First, DNA loops occur despite the weak affinity between the two GalR dimers and the short separation between the operator sites. This is possible through the concerted action of DNA unwinding and HU, which have mechanistic and thermodynamic roles, respectively. Second, GalR/HU-mediated DNA looping represses transcription. DNA looping was observed in the presence of GalR and HU only if DNA was unwound beyond a threshold of ϳ1.5%. This prompted the idea that unwinding had denatured the AT-rich HU-binding site near the apex of the loop, which became an easy target for HU bending to bring the operator-bound GalR dimers into juxtaposition and secure a loop. This mechanism was validated by observing that SSB (sϪingle-stranded DNAbinding protein) abrogated GalR/HU-mediated looping (41).
MT also allow stretching of the DNA tether with different forces by varying the separation between the magnets and the bead. Analysis of the loop lifetimes under different tensions showed the GalR/HU-mediated loop forms two possible antiparallel conformations, A1 and A2 (44,45), with similar probability (46). Thus, the unique ability of A1 (and not A2) to repress transcription must reflect the different loop topologies that alter the interaction of RNA polymerase with the nucleoprotein complex. This subtlety would have been difficult to detect without single-molecule experimentation and warrants further investigation.
In the lac and gal operons, there are three and two operators, respectively. Other systems have higher numbers of operators, and therefore, the numbers of possible looped and unlooped conformations increase. One example is the repressor, which has six operators and is described below.

Long Specific Loops between Tandem Operator Sequences: Repressor
Transcription of the repressor (CI) is autoregulated to generate high enough concentrations to repress lytic genes but not so high as to compromise the switch to lysis. This simple and yet finely tuned epigenetic switch is based on a modular arrange- whereas O R3 is a lower affinity site (Fig. 2). Occupancy of O L1 and O R1 represses transcription of lytic genes from promoters P L and P R , respectively. Cooperativity between adjacently bound CI dimer molecules at O L1 /O L2 or O R1 /O R2 strengthens this repression and favors occupancy of O R2 , which stimulates transcription of CI from P RM . Although O R3 partially overlaps this promoter, CI binding is too weak at physiological concentrations. Thus, one might imagine that CI transcription could rise to levels that would prevent switching to lysis.
However, TPM experiments have provided direct evidence that CI mediates a DNA loop that stimulates CI binding to O R3 as originally hypothesized by Dodd et al. (48). Also, performing a "looping titration" at different physiological CI concentrations on DNA fragments that contained either the entire wildtype regulatory region or double-point mutant O L3 and O R3 operators showed that these sites play a pivotal role in the thermodynamic stability of the loop (49). These results strongly support the idea that a DNA loop may first form due to the long-range interaction between CI dimers bound at O L1 and O L2 on one side and at O R1 and O R2 on the other side, a CI octamer-mediated loop (48). Once a stable octamer has formed, a CI dimer bound at O L3 can stabilize, via proteinprotein interaction, a juxtaposed dimer at O R3 . Thus, DNA looping seems to enable occupancy of weak O R3 and P RM repression without extremely high levels of CI that would interfere with efficient switching to lysis.
Analysis of the TPM data showed that the loop is most thermodynamically stable only when all six CI sites are occupied and contribute to its closure. The probability of occurrence of this loop mediated by an "octamer plus tetramer" of CI was shown to increase with the amount of protein available (49). This model was corroborated by in vitro transcription assays performed in the presence of the same CI concentrations used in the TPM measurements. 3 Indeed, transcription from the lytic promoters, P L and P R , was repressed, whereas that of P RM was activated above basal levels to a maximum value as the concentration of CI increased. Further increase in CI concentration, which increases looping probability, progressively repressed P RM .
TPM control measurements on DNA lacking both the O L and O R regions showed that the unlooped DNA shifted toward shorter tether lengths with increasing protein concentration. This suggested that nonspecific CI binding was significant and appreciably shortened the DNA tether (49). Further evidence of nonspecific CI-induced DNA bending was obtained by stretching and relaxing single DNA molecules with MT. Recording the measured DNA extension as a function of applied force in the absence and presence of CI produced markedly different curves. More force was necessary to extend the DNA in the presence of CI, perhaps indicating that nonspecifically bound CI kinks and bends DNA through transient interactions between nearby dimers. The number of nonspecifically bound proteins was quantified using the recent model proposed by Zhang and Marko (50). 4 Because CI mRNA is transcribed and translated close to P RM , the high local concentration of CI dimers likely saturates the specific operators (47,51), and nonspecific binding may be physiologically relevant, as shown for other proteins (40,52). Kinetic analysis of the TPM data showing formation and breakdown of the CI-mediated loop is particularly enlightening in this respect. 5 The pdf for the dwell times of the looped and unlooped DNA conformations (states) spans several orders of magnitude and decays nonexponentially. Indeed, it was found that a stretched exponential form satisfactorily fitted the probability distribution of the unlooped state dwell times for several tested CI concentrations. On the other hand, the first part of the dwell time distribution for the looped DNA state could not be fitted. Only the distribution of the long dwell times could be fitted with a power law decay (pdf pl ϰ t Ϫm ), which does not yield a mean lifetime (Fig. 3).
The nonexponential kinetic behavior of both formation and breakdown of the regulatory loop, combined with (i) the pivotal role of the O 3 sites in the thermodynamics of loop formation, (ii) the concentration dependence of both loop formation and breakdown, and (iii) the presence of significant nonspecific CI binding even at low protein concentrations revealed by the single-molecule experiments reviewed above, led to the formulation of the following kinetic mechanism.
Assuming that a CI octamer is the minimum requirement for loop formation (49,53), which is then stabilized by an additional CI tetramer, the dependence of looping on CI concentration could simply reflect an increase in the population of unlooped molecules having at least two pairs of adjacent, specifically bound CI dimers as the amount of CI increases. Kinetic complexity might then arise from nonspecific binding of CI to DNA. Nonspecifically bound CI dimers may shorten the DNA by bending or softening the double helix upon binding or through interaction of the C-terminal domain residues of nearby dimers. The ensuing DNA bending could facilitate loop formation by reducing the elastic energy barrier to loop closure, which in turn increases the encounter probability among the proteins bound at these two regions. These effects would depend on the number of nonspecifically bound proteins in the loop region, n NS . The variation in the number of nonspecifically bound dimers would generate a distribution of rate constants for loop formation (schematically represented by k L (n NS ) in Fig.  3) that could explain the observed stretched exponential pdf.
Nonspecific CI binding also may contribute to the observed power law decay of the loop dwell times. In particular, nonspecifically bound dimers within the loop may tetramerize, similar to specifically bound dimers, to fortify the specific nucleoprotein complex that secures the loop. In Fig. 3, k L (n NS ) and k U represent the rate constants of loop formation and breakdown. If no additional nonspecifically bound proteins are present, k L (n NS ϭ 0) is single-valued, and the kinetics of the system are exponential. In the presence of nonspecifically bound CI, the variation of n NS broadens the distribution of k L (n NS ). Any further tetramerization between these nonspecifically bound dimers would be dependent on their number, n NS, and on their relative separation, ᐉ, as diagrammed schematically in Fig. 3. There, k a (n NS ,ᐉ) represents a distribution of rate constants for the association between additional loop-stabilizing CI tetramers (red), and k d is the rate constant for their dissociation. These manifold looped states would give rise to a continuous distribution of waiting times for the breakdown of the CI-mediated loop, producing a power law-like decay. Note that only a narrow range of looping free energies separates activation and repression of the P RM promoter. Thus, nonspecific binding might function as a highly concentration-sensitive mechanism for the fine-tuning of the looping free energy. This kinetic model may be tested theoretically by deriving the distribution of rate constants produced by nonspecific binding and comparing it with the experimental pdfs. AFM also could be used to visualize the various looped and unlooped species and their dependence on CI concentration. AFM images could directly reveal the number of nonspecifically bound CI dimers and their interactions.
Given the crowded cellular environment, nonspecific binding is likely to play a role in the regulation of several transcriptional systems, of which the epigenetic switch is just one example. In the loops described, proteins recognize and juxtapose two particular segments of DNA to establish a functional assembly. More dynamic loops, in which at least one end inter-acts nonspecifically, might be more commonly associated with nucleosome-remodeling proteins that translocate along DNA powered by ATP hydrolysis.

Dynamic Nonspecific Loops: Chromatin-remodeling Factors RSC and ISWI
Many Snf2 proteins alter chromatin structure in vitro in an ATP-dependent manner, and some modify the chromatin structure in vivo (for a review, see Ref. 54). However, Snf2 proteins can utilize substrates other than nucleosomes (see, for example, Ref. 55). Indeed, it appears that a common feature of the Snf2 family is the ability to alter DNA structure and thereby regulate DNA-protein interactions. MT and AFM were used to investigate the ATPase activity of RSC, a member of the Snf2 family (56). Detection and manipulation of DNA supercoiling were instrumental to show the reversible translocase activity of this protein, which leads to the formation of DNA loops. It was also observed that RSC-mediated loop size and lifetime depended on ATP concentration. The ability to supercoil the DNA with MT revealed additional details of the DNA translocation by RSC. Because the linking number of the DNA in the tweezers remains constant unless a topoisomerase is present or the magnets are rotated, enzymatically created twist and writhe in one part of the DNA must be compensated in another part of the molecule. In single-molecule MT experiments, when RSC induced a negatively supercoiled loop in the DNA, the adjacent segments of the double helix adopted compensatory positive supercoiling. Thus, in molecules shortened by supercoiling induced by winding of the DNA with the MT, RSC-induced, negatively supercoiled loops triggered compensatory positive supercoiling, which decreased the overall extension of the molecule. Instead, in molecules shortened by supercoils induced by unwinding of the DNA, RSC-induced, negatively supercoiled loops triggered compensatory positive supercoiling in the adjacent segments, which increased the overall extension of the molecule. Facile manipulation of the DNA supercoiling using MT revealed not only DNA translocation into loops but also generation of torsion by the enzyme that could enhance nucleosome-remodeling activity.
TPM and AFM were also used to show that ISWI, another member of the Snf2 family, can translocate along DNA by wrapping first and subsequently looping the double helix in an ATP-dependent manner (Fig. 2) (57). Like RSC, this nucleosome-remodeling ATPase shortened DNA tethers as measured using TPM. Whereas the size of RSC loops seemed to be limited by torsional strain, ISWI-induced loops were limited in duration by the available ATP. From data relating the duration of shortening to the concentration of ATP, a value for the Michaelis-Menten constant was obtained, which was consistent with the micromolar values observed for other ATPases. This suggested that ISWI requires ATP to maintain traction on DNA to secure loops. In AFM experiments on positively supercoiled plasmids, ISWI decreased the number of DNA crossovers. This was interpreted as positive superhelical tension from the activity of the protein that induced compensatory negative supercoiling in the remainder of the plasmid and decreased writhe. Characterization of the topology of DNA loops generated by chromatin remodelers is straightforward using MT and can indicate whether the DNA in the looped region is adequately supercoiled for other biochemical reactions, assuming that it contains promoters, replication origins, etc.

Conclusions
The studies summarized above highlight the sensitivity of single-molecule techniques for the characterization of the kinetics, thermodynamics, topology, and mechanisms of transcriptional regulators. TPM and MT allow direct observation of molecular dynamics that are obscured by the average behavior of the asynchronous ensemble in bulk. Furthermore, MT enable exquisite control of the supercoiling level of DNA molecules over a range that is inaccessible biochemically using intercalating agents or topoisomerases. Such control is a great advantage in studies of transcriptional regulation because DNA supercoiling itself is a regulatory parameter. AFM is a powerful complementary technique for visualizing and roughly quantifying the stoichiometry of the nucleoprotein complexes that constitute epigenetic switches. AFM has the advantage over electron microscopy of being less expensive and more user friendly. These single-molecule tools have exposed topological details that were crucial for more completely understanding transcriptional regulators. The rapidly growing single-molecule toolbox will be incisive for investigations of a wide array of DNA-binding proteins that alter DNA topology.