Direct visualization of the effect of DNA structure and ionic conditions on HU–DNA interactions

Architectural DNA–binding proteins are involved in many important DNA transactions by virtue of their ability to change DNA conformation. Histone-like protein from E. coli strain U93, HU, is one of the most studied bacterial architectural DNA–binding proteins. Nevertheless, there is still a limited understanding of how the interactions between HU and DNA are affected by ionic conditions and the structure of DNA. Here, using optical tweezers in combination with fluorescent confocal imaging, we investigated how ionic conditions affect the interaction between HU and DNA. We directly visualized the binding and the diffusion of fluorescently labelled HU dimers on DNA. HU binds with high affinity and exhibits low mobility on the DNA in the absence of Mg2+; it moves 30-times faster and stays shorter on the DNA with 8 mM Mg2+ in solution. Additionally, we investigated the effect of DNA tension on HU–DNA complexes. On the one hand, our studies show that binding of HU enhances DNA helix stability. On the other hand, we note that the binding affinity of HU for DNA in the presence of Mg2+ increases at tensions above 50 pN, which we attribute to force-induced structural changes in the DNA. The observation that HU diffuses faster along DNA in presence of Mg2+ compared to without Mg2+ suggests that the free energy barrier for rotational diffusion along DNA is reduced, which can be interpreted in terms of reduced electrostatic interaction between HU and DNA, possibly coinciding with reduced DNA bending.

. However, whether changes in physico-chemical conditions affect the type of complexes formed and whether this is due to changes in protein conformation or 'simply' due to effects on electrostatic protein-DNA interactions is unclear for most architectural DNA-binding proteins.
In this study, we investigate HU, a conserved DNA-binding protein found in most bacteria. This protein exists as a heterodimer or homodimer in Escherichia coli (E. coli), with two genes encoding the two different HU subunits ( HU α and HU β ) 25 . In other bacterial species, HU often exists as a homodimer. As outlined above HU binding is known to occur in two distinct modes as a function of protein concentration 19,22,26 . The DNA binding properties of HU do not depend on specific DNA sequences, but rather on DNA bending/intrinsic curvature, evidenced as a preference for AT-rich DNA [27][28][29] . Binding of HU to DNA is tension-dependent: application of force induces protein dissociation 29,30 . Moreover, binding of HU to DNA is sensitive to ionic strength: increasing the concentration of NaCl or K-glutamate reduces the affinity of HU for DNA 30 .
The binding of HU to DNA and its motion along DNA have been investigated in single-molecule fluorescence imaging studies both in vitro 31 and in vivo 32 , yielding information on binding and diffusion kinetics. The diffusion coefficients measured in these studies are not very similar: the diffusion coefficient is five times higher in vitro than in vivo (0.5 μm s −1 in vitro vs 0.1 μm s −1 in vivo). This discrepancy is attributed to different ionic conditions and the complexity of the in vivo medium with the numerous macromolecules present in the cytoplasm resulting in crowding, which likely influences the DNA binding properties of HU 33 . However, both studies underline the dynamic nature of the bound HU on DNA.
To elucidate the dynamics of HU-DNA interactions we here investigate (1) the effect of Mg 2+ on the DNA binding behavior of HU and (2) the effect of different structural states of DNA using applied tension as a parameter. Using confocal microscopy combined with optical tweezers in a microfluidic system 34,35 , we visualize and characterize the binding of individual HU dimers on DNA. Our studies reveal that the binding of HU is tensiondependent and that HU stabilizes double stranded DNA during overstretching. The motion of HU along DNA is dramatically different depending on whether Mg 2+ is present or not. With 8 mM Mg 2+ , the diffusion coefficient is in the order of 0.1 μm s −1 , but the diffusion coefficient is more than an order of magnitude smaller in the absence of Mg 2+ . The impact of magnesium ions on HU sliding along DNA is significant because Mg 2+ is also present in variable amounts in vivo, which might be implied in regulatory roles of HU.

Results
For our studies of HU-DNA interaction, we use an integrated optical tweezers/confocal microscopy instrument to permit concurrent manipulation of DNA and to visualize DNA-bound fluorescently labelled proteins (Fig. 1A). The microfluidic sample chamber permits efficient repeated stepwise capture and assembly of protein-DNA complexes between beads (Fig. 1B), followed by mechanical characterization and fluorescent microscopy visualization 34,35 . HU protein binds DNA under high tension. In order to determine the tension dependent DNA binding characteristics of HU on DNA and to characterize the impact of Mg 2+ on this, we determined the force-extension behavior of individual dsDNA molecules as a function of HU concentration, both in presence and absence of 8 mM Mg 2+ . First, we performed a control experiment on bare DNA without and with 8 mM Mg 2+ (Fig. 2A). The force-distance (FD) curves of bare dsDNA molecules both without and with 8 mM Mg 2+ exhibit a sawtooth shape at high tension indicative of dsDNA unpeeling during overstretching. The overstretching occurs at 58.6 ± 0.5 pN without Mg 2+ (N = 98, containing 50 mM NaCl, 10 mM Tris pH 7.5, 1 mM DTT) and 67.8 ± 0.4 pN with 8 mM Mg 2+ (N = 92, containing 50 mM NaCl, 10 mM Tris pH 7.5) ( Table 1). A similar effect of Mg 2+ on force-extension behavior of bare DNA has been reported before 36 . It is thought that this effect can be attributed to neutralization of the backbone charge of DNA by Mg 2+ resulting in reduced intramolecular repulsion 36 . Hence it effectively stabilizes the helical twist of the dsDNA and consequently a higher force is required to reach overstretching.
In contrast, HU-DNA complexes exhibit distinct features depending on the HU concentration (Fig. 2B,C). Smooth overstretching becomes gradually more prevalent when the HU concentration is increased. Moreover, Figure 1. Schematic illustrating the experimental layout of an experiment using confocal optical tweezers. (A) Alexa555-labelled HU proteins (pink spheres) are excited at 532 nm. Fluorescence emission is measured at 555 nm. (B) A top view of a microfluidic sample chamber with laminar flow. DNA was captured between beads, and HU-DNA complexes were assembled inside the protein channel and in the presence of flow. The measurements were carried out inside one of the two top channels without flow.  Table 1). Lc is independent of the HU concentration both without and with 8 mM Mg 2+ (Fig. 3, top). In contrast, it was found that the Lp, is dependent on the HU concentration in both experimental conditions. Lp is reduced compared to bare DNA at low HU concentration due to HU binding as expected 22,26 (Fig. 3, bottom). Lp increases compared to bare DNA at high HU concentration. The dashed line in this figure indicates the Lp of bare DNA. The decrease in Lp is due to HU-induced bending of DNA resulting in compaction, whereas the increase in Lp corresponds to HU-DNA filament formation. However, a lower magnitude of the increase in Lp was observed in the condition with 8 mM Mg 2+ . This is in agreement with earlier observations that increasing ionic strength reduces HU binding affinity 30 . Overall our results agree with earlier observations 22,26,33 and indicate switching between two modes of binding depending on HU concentration. were acquired in a protein-free channel, that of (B) and (C) in a channel containing HU protein.  Clusters of fluorescent signals of proteins appear along the timeline in the condition without Mg 2+ and at tension above 10 pN, while no evidence for clusters is found with 8 mM Mg 2+ . In our earlier studies 33 we showed that HU dimers bind DNA with high cooperativity (both in the absence and presence of 8 mM Mg 2+ ) when there is no tension on the DNA. Similar binding behaviour of HU is expected at tension under 10 pN since the extension of the DNA is shorter than Lc below that force and thus somewhat comparable to the condition in the TPM experiment. Protein cluster formation is pronounced at higher tension without Mg 2+ , while protein cluster formation is suppressed by Mg 2+ . At high tension, the straight DNA molecule acts against the bending force of HU proteins 22,29 and thus the bending mode of the bound HU proteins is suppressed. Multiple HU dimers together can form filaments along DNA 22,26,38 and such filaments appear as fluorescent clusters along the DNA. As the stretching force increases, the distance between the sites at which HU proteins are bound to DNA, also increases. This may reduce the interaction between adjacent HU proteins, mediated by the opposite surface charge of HU 'arms' and 'body' 39 (Supplementary SI Fig. 1a). These dimer-dimer interactions, which occur in addition to the HU-DNA interaction, would cause cooperativity in filament formation. Without Mg 2+ a part of the 'body' of the dimer could interact stronger with parts of the 'arms' of the other HU dimers promoting filamentation. Mg 2+ could neutralize the negative surface of the HU dimer at the 'body' (Supplementary SI Fig. 1a), hence reducing the attraction between individual HU dimers. Some dark regions are visible at the center of the HU-DNA complex at tensions above 60 pN. These are AT-rich areas which melt forming bubbles during overstretching 40 (Fig. 4B,D). Whereas HU can bind to singlestranded DNA (ssDNA), the binding affinity is threefold lower than that for binding double-stranded DNA 39,41,42 . Most importantly anti-cooperativity (ω ~ 0.4) is observed 43 , explaining the lack of protein binding. The high fluorescent signal outside the dark region of the DNA suggests that during overstretching, DNA forms bubbles but does not peel from the ends, in a HU-DNA complex.
The normalized fluorescence signals collected for eight tension ranges (0-5, 5-10, 10-20, 20-30, 30-40, 40-50, 50-60, 60-70 pN) indicate the amount of bound proteins on DNA in each range. The amount of bound HU on DNA is affected by DNA tension and by the presence of 8 mM Mg 2+ (Fig. 4F). Without Mg 2+ , HU binds to the DNA in a tension independent manner at tensions between 10 and 60 pN (Fig. 4E). Our observations in the absence of Mg 2+ and under 10 pN agree with the results of Xiao et al. that HU proteins dissociate from the DNA as a consequence of the tension applied to the DNA 29 . The normalized fluorescence signals decreased upon increasing tension up to 50 pN in the presence of Mg 2+ . HU dimers tend to dissociate from the DNA and/ or are less likely to bind to DNA. However, under these conditions (i.e. with Mg 2+ ) the normalized fluorescence signals increase again at tensions above 50 pN indicating that the DNA binding affinity of HU increases. This   (Fig. 5A,B). To quantify these differences, we determined the diffusion coefficient (D) by fitting the data to 1-dimensional mean square displacement (MSD) curves for individual proteins (N > 14 for each tension) 45 . The D values at each experimental condition are obtained from a Gaussian fitting to a 1000-bootstrap resampling ("Materials and methods"). The diffusion coefficients are roughly 30-times higher in the buffer with 8 mM Mg 2+ on average (120 ± 20 nm 2 s −1 ) compared to that without Mg 2+ (5 ± 1 nm 2 s −1 ) (Fig. 5C). This analysis indicates that Mg 2+ enhances the mobility of HU on dsDNA. In a recent study, Leven and Levy 7 calculated the energy barrier of HU to move on DNA, illustrating that the electrostatic interaction between protein and DNA causes an energy barrier for protein motion on the DNA. Therefore, the higher mobility of HU with 8 mM Mg 2+ could be due to shielding of DNA from the positively charged interaction surface involving the arms and extending down the side of HU 39 in the presence   (Fig. 5C), a distinct difference in diffusion coefficient has been seen for HU bound to DNA at tensions above 30 pN (6 nm 2 s −1 ) and at tensions below 30 pN (3 nm 2 s −1 ). This difference could be explained by a weaker contact of the HU dimer with DNA since the ability of HU to bend is reduced by stretching DNA 31,44 . At 60 pN, DNA starts to melt (Table 1, 58 ± 1 pN) in Mg 2+ -free condition. A decreased diffusion rate (about 2 nm 2 s −1 ) suggests that HU proteins might encounter roadblocks consisting of melted DNA bubbles hindering diffusion as has been reported for other proteins 46 . Next, we obtained the residence time for each experimental condition applying the same method as used for determining the diffusion coefficient ("Materials and methods"). The residence time of HU exhibits no obvious dependence on tension both without and with 8 mM of Mg 2+ (Fig. 5D). However, HU resides shorter on the DNA in the presence of 8 mM Mg 2+ than without Mg 2+ . Our results are indicative of a higher off-rate, at forces above 10 pN, in the presence of 8 mM Mg 2+ , which is in line with the higher diffusion rate suggesting a looser interaction with the DNA.

Discussion and conclusion
Here we investigated the effect of magnesium ions and DNA tension on the DNA binding behavior of HU. In earlier studies the effect of salt conditions on HU binding has been addressed 7,29,38,45 . However, these studies did not provide direct information on the dynamics of the interaction between HU and DNA. Recently, HU diffusion behavior has been visually recorded and measured in vitro and in vivo 32,44 . However, the specific effects of divalent cations and distinct DNA structures, invoked by the application of force, were not addressed in these studies.
In our research, we systematically investigated the interaction between HU and DNA in the presence and absence of Mg 2+ , and as a function of tension on the DNA. With 8 mM Mg 2+ , HU suppresses the tension-induced melting presumably by increasing the stability of the DNA duplex. Hence, overstretching is smooth suggesting that either s-DNA or bubbles are formed. Our fluorescence data show that Mg 2+ increases the dissociation rate of HU protein to DNA. Moreover, Mg 2+ enhances the diffusion of HU on DNA. Increased access to the DNA www.nature.com/scientificreports/ by HU could explain the structural stabilization of DNA during overstretching. Moreover, we speculate that the high tension on DNA provides HU dimers with more opportunities to bind DNA in a configuration involving low bending and quick rebinding of proteins will replace proteins that dissociate from DNA thus increasing the total amount of HU on the DNA at high tension. We compared the diffusion behavior of HU on DNA reported by Kamagata et al. 31 with that of our studies. Since the DNA was extended in DNA arrays used by Kamagata et al. it is reasonable to use the diffusion coefficient at 40 pN, where DNA is fully extended (~ Lc), for comparison. Interestingly, a very large difference in the diffusion coefficient is found in our study (0.007 μm 2 s −1 , without Mg 2+ ) compared to that found earlier (0.39 μm 2 s −1 ) 31 . Note, however, that both the salt conditions and the site of fluorescent labelling on HU differed between the two studies. Instead of NaCl in our study, K-glutamate (K-Glu) was used by Kamagata and colleagues. However, as the affinity of HU for DNA in NaCl and in K-glutamate (K-Glu) are similar 30 , that does not explain the difference in diffusion. We labelled HupB-cys43 at the β-strand, close to the center of the 'body' , while Kamagata et al. labelled HupB-cys91, at the end of the C-terminal α helix at the 'body' (Supplementary SI Fig. 1B,C). To evaluate the possible influence of the labelling site on the binding and diffusion behavior of HU on DNA, we generated a simple model of a HU-DNA complex, based on the IHF-DNA complex ( Supplementary  SI Fig. 1b). In contrast to the bending induced by IHF-which is rigid-that induced by HU is flexible 22 . The IHF-DNA complex thus represents a lowly populated, extreme conformation. In this extreme conformation, position 43 is at a distance of 4 Å , and position 91 at a distance of 10 Å from DNA. However, in the case of less bent DNA this distance increases, with HupB-43 being further away from the DNA in an unbent conformation. Moreover, at the C-terminus, the label at position 91 has more positional freedom increasing the possibility to interact with DNA compared to a label at position 43. It is feasible that the label at HupB-91 thus influences and reduces the interaction between protein and DNA because it is closer to DNA compared to position HupB-43. As a consequence, HU labelled at position 91 could move more freely with a larger diffusion rate. Using position 43 for labelling might therefore provide more reliable measurement of HU diffusion along DNA.
In our experimental condition with 8 mM Mg 2+ , the maximal diffusion coefficient, ~ 0.2 μm s −1 , at 40 pN is still lower than that reported by Kamagata et al. in 50 mM KGlu 31 . At the lowest tension, 5 pN, a diffusion coefficient of 0.094 μm s −1 is found which is close to an in vivo result (0.14 μm s −1 ) reported earlier 32 . While the agreement is close, note that the in vivo environment is complex, with ionic conditions, interactions with other proteins and macromolecular crowding confounding the in vitro behavior in a simple buffer.
In order to investigate the mechanistic basis of the effects of different types of ions we applied the diffusion rotation-coupled equation also used by Kamagata et al.: where k B , T, η, d, R oc , R and ε denote the Boltzmann constant, temperature, solvent viscosity, the distance between two base pairs of DNA, the distance between DNA and the protein center, the protein radius, and free-energy barrier, respectively. This allow us to calculate effects on the free energy barrier, but also to consider possible effects on the radius of gyration of the protein. Note that Kamagata et al. used R, the radius of gyration of the protein structures corresponding to the HU-DNA complexes (PDB 1P71), and R oc ∼ R to calculate the freeenergy barrier (2.1 Tk B ) for diffusion. We obtained the free-energy barrier in 2.3 Tk B using the same R and R oc values (PDB 1P71), close to the previously reported value. We also obtained a radius, R = 2.52 nm, based on the measurement at 40 pN with Mg 2+ and the free-energy barrier (ε) obtained from Kamagata's work. It is possible that a larger angle between the two arms of HU yields a larger gyration radius or that the distance between DNA and protein, R oc , is larger.
In the condition without Mg 2+ HU is almost immobile. With Mg 2+ diffusion might be enhanced due to an increase in the distance between positively charged DNA interaction surface of HU and DNA, hence reducing the friction diffusing over the DNA. Another possibility is that our observations can be explained based on the HU-DNA complex model of Hammel et al. They proposed that HU dimers can bind DNA using only one of the arms of the HU dimer to interact with the dsDNA helix 38 . This would effectively reduce the affinity of HU for DNA and promote sliding on DNA in line with our observations. From a biological point of view, Mg 2+ is a basic constituent of the cytoplasm. It is critical for many enzymatic processes. Bacterial pathogens are known to encounter variations in Mg 2+ upon infecting host tissue. In these cases, environmental Mg 2+ concentrations are low compared to the Mg 2+ concentrations in the bacterial cytoplasm 47 . It is possible that switching between high and low Mg 2+ concentration in such context has effects on HU affinity, DNA bending and filament formation, as seen in our experiments, which are translated in effects on expression of genes of the HU regulon. In summary, our study is a starting point to explain from a mechanistic point of view how DNA structure and Mg 2+ cations affect HU dynamics. Further investigation of the details of how structural changes in DNA (e.g. changes in a twist, extension, conformation) correlate with HU movement (e.g. sliding, pause, or hopping) are required to be able to better understand and ultimately adequately include HU behavior in whole nucleoid models 48,49 . Materials and methods DNA construct. Bacteriophage DNA (Spherotech) (47% GC content) was labelled at the cos sites using Klenow DNA polymerase exo-minus (Thermo Scientific) with 80 mM biotin-14-dATP and biotin-14-dCTP added to a mixture of 100 mM dTTP and dGTP non-modified deoxynucleotides (Thermo Scientific). The DNA product was purified using the GenElute PCR Clean-up kit (Sigma-Aldrich) and stored in TE buffer (10 mM Tris-HCl pH 8.0 and 1 mM EDTA) at 4 • C . Further details can be found in 50 .
Scientific Reports | (2021) 11:18492 | https://doi.org/10.1038/s41598-021-97763-w www.nature.com/scientificreports/ Protein samples. Purification of wild-type HU protein was carried out as described before 19,22,51 . HU A43C was expressed from a pRLM118 derivative 52 (a kind gift from George Chaconas) following heat shock induced expression. The purification procedure was identical to that used for wild-type HU. For fluorescent labeling, 0.2 mM HU A43C was incubated in buffer (25 mM Tris pH 7.5, 1 mM EDTA pH 7.5, 10% glycerol, 50 mM NaCl) containing 0.8 mM DTT for 2 h at room temperature (RT) to break down disulfide bonds between cysteines. DTT was removed using a spin column (Illustra AutoSeq G-50 columns, Thermo Scientific) before introducing the dye. Fluorescent labeling was carried out in oxygen-free environment. To achieve this N 2 gas was bubbled through the solution to deoxygenate the HU sample mixes with the dye, with tenfold excess of dye compared to protein for 4 h at RT. The labelled HU protein was purified using an SP column (1 ml). The protein was eluted at 200 mM NaCl. Alexa555-HU concentration was determined using BCA (Bicinchoninic Acid) protein assay (Thermo Scientific). The Alexa555 concentration was determined by NanoDrop. Using this information, the labeling ratio (maximally two fluorophores per HU dimer) was determined at 90%. The DNA bending ability of Alexa555-labelled HU was tested using optical tweezers and was found to be comparable to that of wild-type HU protein (Supplementary SI Fig. 2).
Optical tweezers experiment. The experiments were performed on a commercial dual-trap optical tweezers instrument with confocal fluorescence imaging (C-trap; Lumicks). A -dsDNA molecule was captured in laminar flow between a pair of streptavidin-coated polystyrene microspheres (4.65 µm in diameter, Kisker) held by optical traps. The single dsDNA attached in between the beads was then moved into a protein channel for protein incubation. Force-extension experiments on protein-DNA complexes were performed following 1 min incubation. Individual HU-DNA complexes were discarded after one force-extension stretching measurement. Tension was applied on the protein-DNA complex by increasing the end-to-end distance of the dsDNA molecule at a rate of 0.2 μm s −1 . All optical tweezer experiments were performed at RT.
Three HU protein concentrations were used: 40, 400 and 2000 nM. The protein was diluted in two types of buffers according to desired experimental conditions: buffer I (50 mM NaCl, 100 mM Tris-HCl pH 7.5, 1 mM EDTA) and buffer II (50 mM NaCl, 100 mM Tris-HCl pH 7.5, 8 mM MgCl 2 ).

Fluorescence experiment.
The experiments in which the mechanical properties of HU-DNA complexes were characterized and those in which HU binding patterns along DNA were mapped, were carried out in the protein channel. The experiments in which the mobility of HU on DNA was investigated, were carried in a protein-free channel. Within the protein channel, measurements were carried out following a 1 min incubation. For measurements in the protein-free channel, protein-bound DNA molecules were moved from the protein channel to the protein-free channel following a 1 min incubation at different tensions. Translocation of the protein-DNA complex occurred at a constant rate of 0.2 μm s −1 , with the end-to-end distance of DNA kept constant. A 532 nm excitation laser was used to excite Alexa555 labelled protein. An electronic multiplying charge-coupled device (EMCCD) camera was used for collecting signals.
For all fluorescence experiments 40 nM Alexa555-labelled HU protein was used. Each protein sample had been incubated in buffer I or II for 2 h on ice prior to the measurements. Data analysis. Lc and Lp were obtained from the force-extension curve of HU-DNA complexes by fitting to an extensible worm-like chain (eWLC) model using a custom-written program in Matlab 36 . Only data at forces below 30 pN were fitted.
The overstretched tension is defined as the tension corresponding to the first peak which appears at the beginning of the melting region on FD curve. In the overstretched region, the end-to-end distance of DNA increases by a large magnitude corresponding to a small amount of increase in tension. With 8 mM MgCl 2 , the bursts are suppressed in the melting region. Hence the overstretched tension is defined as the tension corresponding to a slope on the FD curve that is < 0.02.
Quantification of fluorescence intensity. Fluorescence data was analyzed using ImageJ (NIH) 53 . The fluorescence intensity at each range of tension was normalized by dividing the total image pixels as a unit of fluorescence signal per pixel. The fluorescence signals per pixel of one trace have further been normalized to 1 by dividing the values of the signal for the highest value of the signal per pixel, for each trace.
Determination of HU diffusion coefficient. The displacement of each protein dimer on the DNA was measured over time by recording fluorescence movies at a high frame rate (167 Hz). The corresponding trajectories (> 6 s) were analyzed with a custom-written MATLAB-based program 45 that tracked the position of each Alexa-555-HU dimer as a function of time. The diffusion constant (D) was determined for each individual trajectory with the same analysis program using 1-dimensional mean square displacement (MSD) 35 . Further, we obtained the diffusion coefficient of each experimental condition, from a 1000-bootstraps (using Matlab commercial script) histogram fitted by a Gaussian function (using OriginLab) ( Supplementary SI Figs. 3, 4). For each experimental condition at least 20 traces were used. Only traces spanning longer than 20 s and which did not cross other traces were considered. Errors are given as the standard deviation of the mean, which was divided by the square of the amount of raw traces of each condition.
Determination of HU residence time. First, the residence times of each individual protein trajectory is collected from Kymographs. Residence times of all the trajectories were then categorized according to their measurement conditions; with or without 8 mM Mg 2+ and at which tension. Due to the different number of