Single-molecule imaging reveals DNA shape read-out by the INO80 chromatin remodeler

Abstract

ATP-dependent chromatin remodeling enzymes act together to construct complex chromatin architectures around functionally important regulatory regions like promoters and origins of replication (ORI). The INO80 complex accurately positions nucleosomes bordering these regions, which is crucial for proper transcription start site selection and efficient replication. How INO80 localizes these nucleosomes and regulates its remodeling activity to produce specific positions mechanistically is unclear, although recent findings suggest a role for DNA shape features and barrier factors. Here, we use single-molecule DNA curtains to directly observe interactions of INO80 with DNA, nucleosomes and barrier factor Reb1. We show that DNA shape features that are enriched in promoters and ORIs strongly regulate INO80 DNA binding and remodeling activity. Moreover, we find that INO80 performs 1D searches and intersegmental transfers to dynamically interact with nucleosomes and Reb1, but cannot bypass them, indicating that they could confine INO80 within promoters and thereby increase engagement with flanking nucleosomes. Our findings reveal how INO80’s target-site search and remodeling activity are influenced by DNA shape recognition and molecular architecture and elucidate how remodelers can integrate different kinds of information to transform the chromatin landscape.

How chromatin remodelers regulate nucleosome positioning remains poorly understood. Here, the authors use single-molecule fluorescence microscopy to reveal how read-out of DNA shape features regulates INO80 DNA interactions and remodeling activity.

Introduction

Manipulation of nucleosomes, the core unit of chromatin, constitutes a major share of eukaryotic genome regulation and results in the formation of diverse chromatin states^1–3. While histone modifications are modulated by different epigenetic modulators, the positioning and histone composition of nucleosomes is mainly directed by four families of ATP-driven chromatin remodeling complexes: SWI/SNF, ISWI, CHD, and INO80⁴. Significant research efforts conducted in the last two decades in S. cerevisiae have revealed that these protein families work in concert to generate specific chromatin architectures around important regulatory regions like promoters and origins of replication (ORI)^5–7. The architectures produced are characterized by a nucleosome-depleted region (NDR) flanked by two well-positioned nucleosomes, which, in the context of promoters, were coined the  +1 (over the transcription start site) and  −1 (opposite side) nucleosomes. Nucleosomes further upstream and downstream show regular spacing and are phased relative to the  +1/−1 nucleosomes^3,8. Maintenance of this architecture seems to be critical, as disruption has been shown to promote cryptic transcription and reduce replication efficiency^7,9. The only chromatin remodeler of the four families to accurately reproduce in vivo-like  +1/−1 nucleosome positions in an in vitro positioning assay was INO80¹⁰, a remodeler with nucleosome and hexasome sliding activity and proposed functions in multiple other pathways, such as replication and DNA repair^11–13. INO80 is present almost exclusively on nucleosomes flanking NDRs¹⁴, and its removal enhances cryptic transcription^5,9,15, making it a prime candidate for genome-wide positioning of  +1/−1 nucleosomes. Results from selective and simultaneous depletion of different remodelers in vivo support this notion^5,16.

In S. cerevisiae, INO80 is comprised of 15 subunits that can be grouped into three structural modules, hereinafter referred to as the Core-, A-, and N-module¹⁷ (Fig. 1a). All three modules are flexibly connected by the Ino80 subunit, which also contains the Snf2-like ATPase motor. During repositioning of nucleosomes, INO80 forms an extended configuration, where the Core-module binds to the nucleosome, while the A- and N-modules sit in front of the ATPase motor and interact with the extranucleosomal DNA, which is translocated towards the motor^18,19. Using this architecture, INO80 positions nucleosomes by pulling them into the NDR⁵.

Fig. 1. Single-molecule imaging of INO80 diffusion.

A Overview of DNA curtain setup and INO80 submodule organization¹⁸. B Kymograms of INO80 labeled with LD555 diffusing along DNA (top) and dCas9 labeled with Atto643 binding statically to DNA (bottom). C Aligned trajectories of all analyzed INO80 (cyan) and dCas9 (black) translocation events. n_INO80 = 677 molecules. n_dCas9 = 21. D Histogram of diffusion coefficients for INO80 and dCas9. Diffusion coefficients were calculated by linear fits of MSDs using the optimal number of fit points, determined by ref.⁷¹. E Left: Histogram of measured fluorescence intensities for all localized particles, showing two fitted peaks for monomers and dimers. Right: Relative fraction of monomeric and dimeric localizations. Source data are provided as a Source Data file.

The ability of INO80 to accurately slide nucleosomes to certain positions necessarily implies negative regulation of the sliding process through external cues. INO80 has been shown to position and phase nucleosomes relative to other proteins bound to DNA, e.g., general regulatory factors (GRF) Reb1 and Abf1 at promoters and ORC at ORIs, most likely through a ruler-like readout^7,20. How this is facilitated mechanistically is unknown. Interestingly, INO80 is also able to position  +1/−1 nucleosomes in the absence of such ‘barrier proteins’, solely based on DNA sequence inputs. Analysis of final +1 nucleosome positions remodeled in vitro by INO80 revealed distinct DNA shape profiles, which changed upon mutation of different subunits¹⁰. Another study observed that INO80 remodels nucleosomes with rigid linkers more slowly²¹. This suggests that INO80 can regulate its remodeling activity through recognition of DNA shape features. DNA shape sensing of proteins is mostly facilitated by interactions with the phosphate backbone of the DNA and reads out characteristics of the DNA’s three-dimensional structure like minor groove width, propeller twist, or DNA curvature²². Although structural and biochemical data have shown that INO80 can recognize these structures^10,19, it is not known how they influence INO80 binding dynamics.

Additionally, while the target-site search of sequence-specific transcription factors (ssTF) and other small DNA-binding proteins (DBP) has been studied for decades^23–25, a comprehensive model for how chromatin remodelers, including INO80, locate their target nucleosomes in a busy chromatin landscape is still missing. This is in part due to the limitations of classical biochemical bulk assays, which are limited in their ability to observe interaction dynamics while bound to the substrate and therefore cannot resolve molecular details of target site search. Recent single-molecule studies of RSC, ISW2, and SWR1 in vitro have presented the first insights by showing that these remodelers, similar to most DNA-binding proteins, use 1D diffusion to locate target proteins like nucleosomes on DNA^26,27. Comparable in-vitro studies are still missing for INO80.

In this study, we used single-molecule DNA curtains to address INO80’s target site search, DNA binding preferences, and interaction kinetics with Reb1 and H2A.Z-containing nucleosomes. Additionally, we performed sliding assays to reveal the effect of DNA shape recognition on nucleosome remodeling. We demonstrate that INO80 uses 1D diffusion by helical sliding to move along DNA and can jump between DNAs using intersegmental transfers. We observe highly DNA shape-dependent binding and diffusion kinetics, with stronger binding and slower diffusion speeds on DNA sequences with high propeller twist angles, which are frequent in promoter NDRs²⁸. Using truncation mutants, we identify the A-module as a sensor for these sequences and show that their recognition within nucleosomal linker DNA strongly inhibits INO80 remodeling. Additionally, we present dynamic interactions of INO80 with nucleosomes and Reb1 and find that both act as barriers to INO80 diffusion, indicating that they could trap INO80 in NDRs and increase engagement with the flanking  +1/−1 nucleosomes. Together, this study reveals detailed insights into how DNA shape, molecular composition, and interaction partners govern INO80’s kinetics on DNA, laying the foundation for a comprehensive target site search model and providing a framework for DNA shape-dependent nucleosome positioning.

Results

INO80 diffuses along DNA

We expressed and purified the S. cerevisiae multi-subunit INO80 complex endogenously and labeled it fluorescently with the help of a C-terminal ybbR-tag at the Ino80 subunit (Fig. 1a). To investigate INO80’s DNA binding and translocation, we used single-molecule DNA curtains, which allow the visualization of individual INO80 complexes on DNA in real time and with high throughput. We used flow cells with nano-fabricated chrome barriers that were passivated with a lipid bilayer, and assembled a double-tethered DNA curtain using end-modified λ-DNA molecules (48.5 kbp), as previously described²⁹ (Fig. 1a). A microfluidic system in combination with total internal reflection (TIRF) microscopy enabled us to apply fluorescently labeled INO80 complexes to the flow cell and to record association and dissociation events of individual molecules. All experimental buffers, unless stated otherwise, contained 1mM ATP.

INO80 bound stably to λ-DNA with a mean lifetime of 113  ± 9 s and slowly moved along the DNA in a non-directed fashion resembling a random walk (Fig. 1b, c). As a control, we applied a catalytically dead Cas9 mutant (dCas9) to the DNA, which, as expected, bound to its target sites and showed no movement along the DNA (Fig. 1b, c, Supplementary Fig. S1a). The analysis of single-molecule diffusion coefficients revealed a wide distribution for INO80 with a median diffusion coefficient over 30 times higher than for dCas9 (0.017 ± 0.001 kbp²/s for INO80 vs. (5.5 ± 3.8) × 10⁻⁴ kbp²/s for dCas9, Fig. 1d). We therefore conclude that INO80 translocates along DNA by 1D diffusion.

A small subset of complexes diffusing on DNA showed noticeably higher fluorescence intensities, and photobleaching analysis revealed that these complexes bleached in two steps (Supplementary Fig. S1b). Analysis of all localized intensities displayed two populations, indicating that  ≈ 11% of DNA-bound INO80 complexes were dimers (Fig. 1e). However, the diffusion coefficients and lifetimes of dimeric complexes showed no significant differences compared to monomers (Supplementary Fig. S1c, d and see Supplementary Note 2), implying that these complexes are functionally indistinguishable from monomers with respect to their DNA binding.

INO80 slides along DNA using a rotation-coupled mechanism

Having established that INO80 uses 1D diffusion, we sought to uncover how INO80 moves along DNA mechanistically. DBPs have been shown to mostly slide, hop, or use a mix of both mechanisms^27,30. This profoundly impacts the way the proteins interact with obstacles and interaction partners along the DNA and is therefore crucial for understanding their target site search mechanism³¹. To test whether INO80 preferably hops or slides, we compared INO80 diffusion at 120 mM KGlu with diffusion at 40 mM KGlu. As hopping proteins have to dissociate and re-associate continuously while moving, their diffusion coefficient is positively correlated to their off-rate and hence salt-dependent³⁰. As expected, with reduced salt concentration, INO80’s median lifetime increased fourfold from 15.1 to 61.9 s. Its median diffusion coefficient decreased slightly, but the change was not significant (0.012 ± 0.002 kbp²/s to 0.017 ± 0.001 kbp²/s; p = 0.055, Mann-Whitney U test, Fig. 2a). This implies that INO80 movement is mostly salt-independent and therefore largely based on sliding over the tested ionic strengths under our experimental conditions.

Fig. 2. INO80 slides along the helical axis and strongly binds to DNA with high negative propeller twist |Ψ|.

A Comparison of lifetimes and diffusion coefficients of INO80 on λ-DNA for different salt concentrations. Significance was determined using a two-tailed Mann-Whitney U test. n_40KGlu = 167 molecules. n_120KGlu = 677. B Example energy landscape felt by a protein sliding along DNA. Energy minima represent binding in register to a particular base pair, while maxima represent the transition state of sliding from one base pair register to the next, i.e., the free-energy barrier for this specific transition. Due to a heterogeneous base sequence, the free energy and barrier heights vary from position to position and can be represented by the root-mean-squared variation ε. C Comparison of ε for different DBPs moving along the helical pitch of the DNA, obtained from refs. ^32,36. Dark gray shading indicates proteins that strongly bend DNA. D Model for INO80 diffusion by sliding along its helical pitch. E Top: Kymogram showing a diffusing INO80 complex. Bottom: Respective trajectory time trace with instantaneous diffusion coefficient (window size: 5 s). Shaded areas indicate highly constrained movement (D_Instant < 0.01 kbp²/s), dots show localizations, black line shows filtered positions (Kalman filter). Left: Apparent diffusion coefficient of the respective molecule and average propeller twist of the DNA depending on position. F Nhp10-Motifs, average propeller twist (1 kbp sliding window), and different binding parameters along λ-DNA. Shaded areas: Unreliable data close to chrome barriers (gray), accumulation of Nhp10 motifs (red), high propeller twist (blue). n = 677 molecules. Right boxes: correlation of binding parameters with average propeller twist and Pearson’s correlation coefficient. Data are presented as mean values +/- 68% intervals from bootstrapping. G Correlation of probability density of localizations and average propeller twist for INO80 without ATP. H Survival plot of INO80 on λ-DNA for experiments with and without ATP. Rates were compared with log-rank test. I Comparison of diffusion coefficients of INO80 on λ-DNA for experiments with and without ATP. Significance was determined using a two-tailed Mann-Whitney U test. n_noATP = 719 molecules. n_ATP = 677. n.s. = Not significant; **** = p < 0.0001. All box plots show median, 25% and 75% quartiles and 10th and 90th percentiles. Source data are provided as a Source Data file.

Almost all sliding proteins diffuse by moving along the helical axis in a rotational fashion³². The exception are DBPs like PCNA or TALEs, which, unlike INO80, encircle the DNA through their molecular structure and can also slide linearly along the DNA^33,34. It follows that INO80 should also diffuse by rotation-coupled sliding. To test this hypothesis, we calculated the root-mean-squared variation of the free-energy barriers ε for INO80, which describes the energetic hurdles INO80 has to overcome while diffusing along DNA and provides insights into the characteristics of INO80-DNA interactions (see Supplementary Note 1).

Using the median diffusion coefficient measured at 120 mM KGlu and a radius of  ≈ 100 Å for INO80 (estimated from published cryo-EM structures), we inferred a free-energy barrier of  ≈ 1.75 k_BT for rotation-coupled diffusion, while linear diffusion would predict a much higher free-energy barrier of  ≈ 3.15 k_BT. For efficient target site search, models predict ε values from 0.5 to 2 k_BT³⁵, which has been confirmed experimentally for a wide variety of DBPs^32,36. Therefore, the ε values calculated for INO80 are consistent with rotational movement around the DNA, but incompatible with linear movement. Because ε is entirely dependent on the structural mechanism through which proteins bind DNA, proteins with similar DNA-binding domains should produce similar ε values (Fig. 2b). Indeed, the ε value calculated for rotation-coupled movement by INO80 closely fits the one determined for Nhp6, a protein closely related to the DNA binding subunit Nhp10 within the INO80 N-module that has also been shown to slide rotationally³⁷ (Fig. 2c). Additionally, high ε values of >1.4 k_BT have been associated with large conformational DNA changes upon binding of the corresponding proteins, e.g. strong bending of the DNA helical axis³⁶. Nhp6 bears a similar HMGB-type box domain as Nhp10, which induces high bending angles into the DNA during binding³⁸. We conclude that INO80 moves by rotation-coupled sliding along DNA and likely bends the DNA in the process (Fig. 2d).

INO80 transfers between DNA strands

In addition to sliding motion, INO80 frequently showed inter-DNA transfers and long-range jumps incompatible with sliding (Supplementary Fig. S2a–c). These included cis-jumps along the same DNA, which are characterized by a big step along y, without a change in x (Supplementary Fig. S2a), and trans-jumps to a different DNA, characterized by significant changes in both x and y (Supplementary Fig. S2c). The characteristics of these jumps are compatible with INO80 completely dissociating from DNA, followed by short 3D diffusion in solution and DNA re-association (Supplementary Fig. S2d). Additionally, we observed frequent events of INO80 intersegmental transfers between juxtaposed DNA molecules tethered within the same well (Supplementary Fig. S2b, c). Thermal fluctuations bring these DNA molecules transiently into close proximity and enable DBPs with multiple flexibly connected DNA-binding domains, such as INO80, to transfer between them by bridging, without the need for 3D search (Supplementary Fig. S2d)³⁹. These events were characterized by sudden but persistent changes in x, without changes in y.

To quantify the different types of jumping, we identified putative jump events within INO80 trajectories by an autoregressive Hidden-Markov model (arHMM), identifying sudden but persistent changes in x (Supplementary Fig. S2e). As a control, to discriminate between jumps and naturally occurring thermal fluctuations, we tracked DNA-bound dCas9, yielding the expected transversal fluctuation amplitude and timescale of proteins in their DNA-bound state (Supplementary Fig. S2f). Additionally, we identified putative INO80 long-range jumps by identifying all DNA-association events (trajectory starts) within a 200ms window after each dissociation event (trajectory end). The collected displacements in x and y at identified jumps were then combined in a scatter plot, where the origin represents the position of INO80 before a jump/transfer and the data points represent the relative position after the jump/transfer (Supplementary Fig. S2g, h). If INO80 were only able to jump, we would expect displacements to be isotropically scattered around the origin, following a near-Gaussian distribution, reflecting INO80’s 3D diffusion speed. While these displacements are present, the vast majority of displacements (83%) are found in two tight clusters in the positive and negative x-direction around the origin (Supplementary Fig. S2g, see right). Strikingly, the distribution of these displacements fits the expected restrictions set by the intersegmental transfer mechanism (Supplementary Fig. S2h): specifically, the displacement is limited in x by the transverse fluctuation amplitudes of the DNAs, and in y by the 1D diffusion speed. Crucially, dCas9 showed neither jumps nor transfers. These findings support that what we observe are indeed transfer events, and that intersegmental transfers are INO80’s preferred mechanism for switching between DNAs in our setup.

INO80 recognizes AT-rich DNA with high negative propeller twist

During closer analysis of INO80 trajectories, we noticed that diffusing INO80 complexes showed varying mobility on λ-DNA over time. This non-random movement was represented by transitions between highly diffusive and highly constrained segments within 1D trajectories of single complexes (Fig. 2e), an unexplained behavior also observed for RSC and ISW2²⁷. Given that INO80 is hypothesized to recognize DNA shape features and that our setup allowed us to analyze the DNA sequence background of diffusing complexes, we tested whether INO80 binding dynamics are influenced by DNA shape. We chose propeller twist as it was singled out as one of the most relevant DNA shape parameters to explain nucleosome placement by INO80 in a previous study¹⁰. The propeller twist Ψ measures the negative angle between the two bases of a base-pair and is correlated with the electrostatic potential (EP) of the minor groove, which is emerging as a major determinant of DNA shape read-out⁴⁰. Bacteriophage λ-DNA shows great variations in propeller twist over long segments (Fig. 2f), allowing us to investigate the relationship between INO80 binding parameters and Ψ values. We separated the DNA into 1 kbp bins and determined the localization probability density, the initial binding position, the off-rate, and the apparent position-dependent diffusion coefficient D_App of INO80 complexes dependent on the DNA position and correlated it with the respective Ψ at this position (Fig. 2f). Remarkably, INO80 was strongly enriched on DNA sequences with high |Ψ| and showed greater initial binding and longer lifetimes in these regions. The diffusion coefficient D_App showed much weaker correlations. All effects were visible independently of bin-size (Supplementary Fig. S3a).

Next, we asked whether other shape features⁴¹ or AT nucleotide content could explain INO80-DNA kinetics even better than Ψ. No features outperformed propeller twist, but AT-content, EP, stretch, opening, slide, and rise performed equally well (Supplementary Fig. S3b–d). Notably, all features that showed very strong correlations with INO80 binding parameters also correlated extremely well with each other (r = 0.99 for Ψ, EP, and AT-content; Supplementary Fig. S3b–d), preventing us from separating their influence on INO80 binding at our resolution.

Multiple subunits of INO80 can bind, but do not hydrolyze ATP¹⁹. Thus, we tested whether INO80 shape readout was influenced by ATP binding. Removal of ATP had no impact on INO80 binding preferences (Fig. 2g), but molecules showed significantly lower lifetimes and faster diffusion speeds (Fig. 2h, i), indicating that ATP-binding induces conformational changes that increase INO80’s DNA binding strength.

INO80 also showed lower off-rates in a segment of the λ-DNA between 12 and 18 kbp, which exhibits a comparably low |Ψ| but contains multiple binding motifs found for INO80-subunit Nhp10⁴² (Fig. 2f, red shaded area and datapoints marked in red). This suggests that the N-module might additionally weakly recognize sequence motifs during 1D diffusion, which may be relevant for different cellular functions of the complex besides remodeling.

Together, these observations demonstrate that INO80’s recognition of DNA shape features determines its mobility and influences its lifetime on DNA, enhancing its localization to high-|Ψ| sites.

Removing INO80’s N-Module speeds up kinetics but retains binding preferences

Having revealed that INO80 displays DNA shape-dependent binding kinetics, we wondered how DNA shape recognition by INO80 is achieved. Given that the N-module is highly species-dependent and that its removal keeps INO80’s ability to reposition nucleosomes largely unaffected^10,19, we hypothesized that a mutant, missing the N-module and the N-terminal end of the Ino80 subunit, would still retain its DNA shape readout capabilities. We chose an established ΔN-INO80 mutant from C. thermophilum that has been shown to be highly conserved in structure to S. cerevisiae INO80 and which produces similar +1 nucleosome positions^10,19 and applied it to DNA curtains. This mutant will be referred to from here on as Ct.ΔN-INO80. If no species identifier is given, the respective variant mentioned is an S. cerevisiae protein or mutant.

Similar to the Sc. wildtype, Ct.ΔN-INO80 readily bound to λ-DNA and moved along the DNA (Fig. 3a), but displayed much faster diffusion dynamics with a median diffusion coefficient over 10 times higher than the Sc. wildtype complex (0.2 ± 0.009 kbp²/s vs. 0.017 ± 0.001 kbp²/s; Fig. 3b). Compared to the Sc. wildtype the calculated free-energy barrier for Ct.ΔN-INO80 was halved (0.86 ± 0.15 k_BT vs. 1.84 ± 0.07 k_BT) and consistent with values for DBPs without strong DNA bending (Supplementary Fig. S4a). This validates our earlier observations that diffusion speed in the wild-type complex is mainly dominated by the high free-energy barrier of Nhp10’s non-sequence-specific HMG-boxes. Remarkably, the diffusion speed of Ct.ΔN-INO80 complexes also showed a much stronger correlation with the mean propeller twist angles of the encountered DNA than the Sc. wild-type complex (Supplementary Fig. S4b, bottom). Analysis of single trajectories revealed that steep transitions from high to low |Ψ| on the DNA acted as strong barriers for Ct.ΔN-INO80, which ‘trapped’ the protein in DNA regions with high |Ψ| (Fig. 3c). Similar to the Sc. wild-type complex, other binding parameters of Ct.ΔN-INO80 also correlated strongly with Ψ (Supplementary Fig. S4b), indicating that recognition of DNA shape is independent of the Ct. N-module and most likely an evolutionarily conserved property of the combined Core- and A-module. As expected, the subset of long-living complexes near Nhp10 binding motifs also disappeared.

Fig. 3. The A-module is a DNA shape sensor.

A Top: Schematic of Ct.ΔN-INO80 and DNA curtains assay. Bottom: Kymogram of LD555-labeled Ct.ΔN-INO80 diffusing along DNA. BDiffusion coefficients of all INO80 variants. Box plot shows the median, 25 and 75% quartiles and 10 and 90th percentiles n_wt = 627 molecules. n_ΔN = 461. n_A-Mod = 705. C Top: Single trajectory of Ct.ΔN-INO80 complex. Arrows indicate where the complex is blocked at transitions from high to low propeller twist. Bottom: Position-dependent propeller twist, INO80 localization probability density and D_App for the trajectory shown above. Asterisks indicate DNA positions with high propeller twist occupied by the complex. D Top: Schematic of INO80 A-module and DNA curtains assay. Bottom: Kymogram of LD555-labeled A-module diffusing slowly (S) or fast (F) along DNA. E Correlation of probability density of localizations and apparent diffusion coefficient with average propeller twist and Pearson’s correlation coefficient for Ct.ΔN-INO80 (magenta, n = 1180 molecules) and INO80 A-module (green, n = 851). Source data are provided as a Source Data file.

We again observed switches between DNA strands without a change in y-position similar to those of the Sc. wild-type, suggesting that the Ct.ΔN-INO80 complex architecture still enables intersegmental transfers (Supplementary Fig. S4c).

The A-module is a sensor for DNA shape

After establishing that the N-module was redundant for DNA shape sensing, we turned our attention to the conserved A-module. The position of the A-module directly in front of the ATPase motor and their shared connection via the Ino80 HSA/post-HSA domain suggest that it might play a crucial role in recognizing DNA shape and regulating nucleosome sliding activity allosterically¹⁹. We therefore purified and labeled the S. cerevisiae A-module, using a truncated version of Ino80 (440–598), which binds all relevant subunits and forms a structurally fully cohesive A-module¹⁹.

A-module complexes on DNA curtains bound strongly to λ-DNA and showed very diverse kinetics (Fig. 3d). The majority of the complexes were almost static (S), while some complexes diffused very rapidly (F). Surprisingly, on average, A-module complexes diffused slightly slower than the Sc. wild-type complex and much slower than Ct.ΔN-INO80 (Fig. 3b), with a median diffusion coefficient of (9.7 ± 0.8) × 10⁻³ kbp²/s. Nevertheless, analysis of apparent diffusion speeds at specific positions revealed that kinetics were again dependent on DNA shape, with slow diffusion at positions with high |Ψ| (Fig. 3e, bottom; Supplementary Fig. S4d). Accordingly, the probability density, initial binding positions, and off-rates were highly dependent on the propeller twist of the DNA (Fig. 3e, top and Supplementary Fig. S4d). The fact that diffusion kinetics were very heterogeneous, highly DNA-shape dependent, but on average almost static compared to Ct.ΔN-INO80, suggests the existence of multiple DNA binding configurations, which in the full complex are additionally modulated by the associated Core-module. In line with this, structures of the Ct. A-module show that it can bind to DNA in at least two different states, with or without additional protein-DNA contacts at the N-terminal end of the module^19,43. Taken together, our data are consistent with the model that the A-module acts as a DNA shape sensor that confers strong binding preferences for DNA with high Ψ to the INO80 complex.

Recognition of consecutive poly(dA:dT)-tracts inhibits INO80 remodeling

With a clear picture of shape-dependent DNA kinetics by INO80 emerging, we wondered whether DNA shape recognition by the A-module might also influence remodeling kinetics. In gel-based sliding assays using nucleosomes with sufficiently long linker DNA, INO80 translocates end-positioned nucleosomes into the center of the DNA⁴⁴. We designed a nucleosome construct with 80 bp linker DNA (0N80), whose average Ψ is randomized across the whole linker sequence (‘No A-tract’). We then inserted poly(dA:dT)-tracts (A-tracts), which exhibit extremely high |Ψ|, at positions where they are contacted only by the A- and N-module during the remodeling process^19,43,45,46, (Fig. 4a, b and Supplementary Fig. S5a).

Fig. 4. Effect of DNA shape sensing on nucleosome remodeling.

A Left: Schematic of 0N80 nucleosome design for gel-based sliding assays. Sequences exchanged are shown. Right: Predicted propeller twist for linker DNA of all nucleosome constructs. Areas highlighted indicate changes in shape compared to the control sequence. B Schematic showing the positioning of exchanged sequences in respect to INO80 ATPase (light cyan) and A-module (cyan) before and after remodeling, based on structural data and biochemical mapping^19,43,45,46. C Native gel showing remodeling of 0N80 nucleosomes over time depending on linker DNA. Bottom band represents unremodeled end-positioned nucleosome, top band represents fully remodeled center-positioned nucleosome. D Comparison of fully remodeled nucleosome fraction for all nucleosomal substrates. Significance was determined using an unpaired two-tailed t-test for all comparisons. n = 4 independent experiments. n.s. = Not significant; * = p < 0.05; ** = p < 0.01; *** = p < 0.001; **** = p < 0.0001. E Schematic showing the architecture of S. cerevisiae promoters classified in ref. ⁴⁸ (ssTF = sequence-specific transcription factor). F Bottom: Average shape profiles around +1-nucleosomes of different S. cerevisiae promoter classes oriented according to the direction of transcription. The region of high propeller twist is highlighted. n_Con = 2390 promoters. n_Ins = 1737. n_Ind = 920. Top: Schematic showing the superimposed positioning of the INO80 ATPase and A-module bound to the +1 nucleosome in the two possible orientations. INO80 pulls nucleosomes in the direction of its bound A-module. G Average shape profiles around subsets of +1-nucleosomes of S. cerevisiae UNB promoters with different NDR sizes, oriented according to the direction of transcription. The region of high propeller twist is highlighted. The average NDR midpoint for each subset is indicated in red. H Average shape profiles around NDR midpoints for all uni- and bidirectional promoters in S. cerevisiae. n_Uni = 4955 promoters. n_Bi = 587. Source data are provided as a Source Data file.

The insertion of a single long A-tract (‘single’) led to a reduction of fully remodeled product for all time points, but the decrease was very modest (Fig. 4c, d). However, strikingly, when we introduced three short A-tracts in succession, phased by 10 bp, such that they face the same direction (‘phased’), INO80 sliding was severely inhibited. To test whether this reduction in sliding activity originated through recognition of shape or AT-content, we exchanged the phased A-tracts with (AT)_n-sequences (‘phased control’), which display only moderate |Ψ|, but provide the same AT-content as A-tracts (Supplementary Fig. S5b). This led to a significant speed-up of the remodeling process, although the sliding was still slightly slower than for the single A-tract substrate (Fig. 4c, d). These results indicate that, rather than AT-content, the A-module specifically recognizes sequences with high |Ψ| or EP. In agreement with this finding, a previous study had found reduced remodeling speeds for the nucleosomes with native rigid linker sequences²¹, which showed higher |Ψ| than their flexible counterparts (Supplementary Fig. S5c). In addition, our results suggest that recognition of these shape features has to happen at multiple interfaces across the A-module to significantly repress INO80’s remodeling activity.

Identification of a high-Ψ stop signal at open promoters in S. cerevisiae

Prompted by the discovery that multiple high-|Ψ|-motifs in succession can act as a stop signal for INO80 remodeling, we asked whether this signal is present in S. cerevisiae genome sequences. To test this, we used published in vivo nucleosome dyad distributions⁴⁷ and analyzed the DNA shape around +1 nucleosomes, stratified by different promoter classes (Fig. 4e)⁴⁸. Additionally, we analyzed promoters with and without a consensus TATA-box.

Remarkably, the profiles revealed a pronounced peak of high |Ψ| over  ≈ 25 bp within the NDR of constitutive, insulated, and TATA-less promoters (Fig. 4f and Supplementary Fig. S5d, e). When INO80 is bound to the  +1 nucleosome while sliding into the NDR, this peak corresponds positionally to the N-terminal end of the A-module, similar to the ‘phased’ nucleosome substrate, which inhibited INO80 sliding activity (Fig. 4a–d). The position of this peak is not dependent on NDR size and instead is fixed to the  +1 nucleosome position (Fig. 4g). Further, analysis of uni- and bidirectional promoters shows that the signal is weaker for  −1 nucleosomes (Fig. 4h). Altogether, these findings show that the identified stop signal is present in vivo and is associated with  +1 nucleosome positions.

INO80 dynamically interacts with nucleosomes

Next, we focused our attention on the role of interaction partners of INO80 within the context of target-site search and remodeling. In vivo, INO80 preferentially interacts with and remodels +1 nucleosomes, which are enriched for the histone variant H2A.Z⁴⁹. We therefore assembled labeled H2A.Z-containing nucleosomes on λ-DNA and generated sparsely chromatinized DNA curtains to which we added INO80 (Fig. 5a).

Fig. 5. INO80 interactions with nucleosomes and Reb1.

A Schematic of chromatinized DNA curtains assay. B Top: Kymogram of an INO80 complex (cyan) co-localizing with a nucleosome (orange). Bottom: Corresponding time traces with co-localizations indicated by shaded areas. C Comparison of INO80 co-localization times with nucleosomes (n = 142 molecules) and Reb1 (n = 34). Box plot shows the median, 25% and 75% quartiles, and 10th and 90th percentiles. D Left: Schematic for eviction assay. Right: Survival plot of nucleosome lifetimes alone or while bound by INO80. Rates were compared with the log-rank test. E Schematic of DNA curtains assay to test interaction between Reb1 and INO80 on DNA. F Same as B, but for INO80 (cyan) co-localizing with Reb1 (yellow). G Top: Schematic showing possible outcomes of co-localization events. Bottom: Pie charts indicate the fraction of co-localization events that result in a block for nucleosomes and Reb1. Source data are provided as a Source Data file.

After injection, we observed INO80 co-localizing with nucleosomes over extended periods, with a median interaction time of 15.3 ± 4.6 s (Fig. 5b, c and Supplementary Fig. S6a). Closer analysis of co-localization times revealed three types of interactions (Supplementary Fig. S6b). First, extremely short co-localizations (0.23 ± 0.1 s, 24.5%), which most likely constitute collisions without nucleosome binding. Second, short associations (16.6 ± 3.2 s, 41.3%) that were represented by complexes binding dynamically to nucleosomes in-between 1D searches. These interaction times are in line with previously reported nucleosome occupancy times of other remodelers²⁷. Third, we also detected a cohort of extremely long-lived co-localizations (1200 ± 660 s, 34.1%) that were mostly characterized by complexes already bound to a nucleosome at the start of recording and that remained statically bound until they dissociated or photobleached (Supplementary Fig. S6a). 85.3% of INO80 molecules (29/34), for which we could determine the method of association, associated with the nucleosome through 1D searches, as opposed to 3D searches (Supplementary Fig. S6c).

Contrasting a previous study of RSC and ISWI²⁷, INO80 did not induce long-range nucleosome translocation, most likely because remodeling distances were below our resolution limit. To ensure that the used INO80 complex was active even in its labeled form, we confirmed the remodeling capability with gel-based sliding assays (Supplementary Fig. S6d).

No indication of H2A.Z/H2B-dimer eviction by INO80

Specific eviction of H2A.Z-histones by INO80 in vivo and in vitro has been a controversial topic in INO80 research, with some studies detecting radical changes in H2A.Z distribution across the genome upon INO80 deletion⁵⁰, while others observe no differences⁵¹. Labeling of the H2B histone allowed us to observe whether INO80 evicts H2A.Z/H2B-dimers from λ-DNA during our assay. While we could not distinguish whether steps in fluorescent signals were caused by photobleaching or H2A.Z/H2B eviction, we were able to compare the time until fluorophore signal disappearance of nucleosomes bound by INO80, with those of nucleosomes imaged in the absence of INO80. Both were well-described by a single exponential distribution with a mean lifetime of 193 ± 23.1 s and 244 ± 28.6 s, respectively (Fig. 5d). The disappearance rate between both categories showed no significant difference (p = 0.29, log-rank test), indicating that INO80 did not substantially enhance the eviction rate for H2A.Z/H2B-dimers in our assay.

INO80 does not stably bind to DNA-bound Reb1

Apart from DNA sequence, nucleosome positioning by INO80 is influenced by general regulatory factors (GRFs), which bind DNA at defined motif sites and have been proposed to act as specific barriers to the remodeling process and thereby shape nucleosome positioning^7,20. Whether this type of regulation requires INO80 binding to GRFs or whether these proteins just act as passive roadblocks for remodeling is unclear. To answer these questions, we purified and labeled the GRF Reb1 from S. cerevisiae and assessed its integrity on DNA curtains. Purified Reb1 was functional and bound to known strong binding sites on λ-DNA (Supplementary Fig. S6e). To test whether INO80 and Reb1 interact on DNA, we loaded both proteins in consecutive steps onto DNA curtains (Fig. 5e). We did not observe binding of INO80 from solution to DNA-bound Reb1. Additionally, we did not observe any simultaneous binding of both proteins. This was unchanged when we reversed the order of injections or co-incubated INO80 and Reb1 for five minutes prior to injection. We then analyzed interactions between INO80 and Reb1 arising through encounters during INO80 1D search (Fig. 5f). Co-localization times were extremely short, with only 13% of associations being longer than 2 s and a median of 0.80 ± 0.19 s, suggesting that INO80 does not specifically interact with DNA-bound Reb1 for extended periods of time (Fig. 5c). Hence, we conclude that Reb1 functions as a passive barrier for INO80 during nucleosome positioning.

Nucleosomes and DNA-bound Reb1 act as barriers for INO80 diffusion

Finally, to understand how INO80 locates the promoter and the +1 nucleosome within a crowded chromatin environment, it is crucial to uncover how INO80’s path along DNA during 1D searches is affected by collision or interaction with other proteins. We therefore asked whether nucleosomes or Reb1 might act as roadblocks for INO80, confining the complex within subsegments of the promoter. Accordingly, we divided all interactions of INO80 with nucleosomes and Reb1 that showed 1D diffusion before and after into block and bypass events (Fig. 5g). Blocks were defined as events where INO80 stays on the same side of the interaction partner; all remaining interactions were defined as bypasses. INO80 was clearly confined by both proteins, with 90% of nucleosome (81/90) and 98% of Reb1 interactions (51/52) resulting in a block. We then explicitly looked at longer co-localizations with nucleosomes (>5 s), as in these cases INO80 interacts unequivocally with nucleosomes and passive collisions are excluded (Supplementary Fig. S6f). These events showed a similar picture, with 86% (25/30) of interactions categorized as blocks. This indicates that nucleosome-bound INO80 preferably disengages in the same direction hitbound from and does not hop over the nucleosome. Together, these results suggest that INO80 diffusion is blocked by both collisions and interactions with nucleosomes and Reb1. Additionally, these observations validate our earlier findings that INO80 mostly uses a sliding mechanism to diffuse along DNA.

Discussion

DNA binding and promoter targeting

Depending on the measurement method, each S. cerevisiae cell nucleus harbors 57,000 to 60,000 nucleosomes⁵². Of these, only 4500 to 5500 (7.5 to 9.6%) constitute possible  +1 nucleosomes^47,51, the main target of INO80¹⁴. A unique trait highlighting  +1 nucleosomes is the unilateral presence of long stretches of free neighboring promoter DNA. Our experiments clearly demonstrate that INO80 binds strongly to DNA and furthermore recognizes/prefers certain DNA shape properties, e.g., high propeller twist |Ψ| and high electrostatic potential (EP) (Figs. 2f, 4a, d). Here, INO80 exhibits a notably increased on- and reduced off-rate, resulting in accumulation of the complex on these sequences. Intriguingly, in S. cerevisiae promoters and ORIs are greatly enriched in poly(dA:dT) sequence tracts (A-tracts) that are characterized by especially high |Ψ| and high EP^28,53. Closer inspection of A-tracts in these promoter sequences revealed further that they are positioned next to each bordering nucleosome and that their strength positively correlates with direction of transcription^10,28 (Fig. 4f and Supplementary Fig. S5e and Fig. 6a). Given the strong shape preferences we observed by INO80, these asymmetrically positioned A-tracts could guide the complex towards promoters in general and towards  +1 nucleosomes in particular and explain how differential recognition of  +1 and  −1 nucleosomes is achieved (Fig. 6b). In general, INO80 shows a high preference for long linkers in nucleosome binding and remodeling assays^18,43,44,54, which can be explained by its structure. With the N- and A-modules, the complex possesses two DNA-binding modules, which bind extranucleosomal DNA in tandem with a combined footprint of  ≈ 70 bp¹⁴. Linkers within gene bodies, on the other hand, are short in S. cerevisiae, with a length of only  ≈ 18 bp⁵⁵.

Fig. 6. Model for INO80 target site-search and DNA-shape-dependent remodeling regulation.

A Scheme of promoter architecture of constitutively expressed, TATA-less genes in S. cerevisiae. If present, Reb1 motifs (yellow) are found in the center of the NDR, while A-tracts (blue) are positioned  ≈ 95 to 120 bp away from the bordering nucleosome dyads. A-tracts are characterized by high |Ψ| and EP and show stronger enrichment next to  +1 nucleosomes. B INO80 is recruited to promoters through its strong preference for long free DNA stretches and read-out of DNA shape properties. Linker DNA in gene bodies on average is too short for proper binding of INO80s DNA-binding modules (dark teal). The positioning of preferred DNA shapes guides INO80 into the vicinity of the +1 nucleosome. C Once bound to the promoter, INO80 performs 1D searches to associate with the  +1 nucleosome. Due to its sliding mechanism, Reb1 (yellow) and nucleosomes act as roadblocks to INO80 diffusion and confine it within the promoter. Placement of its DNA-binding modules strongly disfavors dissociation from the nucleosome into the gene body. D During 1D searches, INO80 can transfer between DNA strands. This might be possible due to its multiple DNA-binding domains which are flexibly connected and would allow bridging of DNA strands in dense chromatin through a ‘monkey-bar’ mechanism. E During remodeling, DNA is translocated in a corkscrew motion toward the nucleosome by the Snf2-type motor, generating overwound DNA in front of the motor, which is relieved by free rotation of the DNA. Simultaneously, the extranucleosomal DNA is probed by the upstream A- and N-module. Strong binding of the A-module to consecutive A-tracts or physical collision of the N-module with Reb1 lead to fixation of the DNA. As the motor continues to translocate DNA, positive torsional strain is built up, which ultimately might cause eviction of the complex.

Combined, these findings are consistent with a model where steric restrictions prevent INO80 from proper binding within the gene body (Fig. 6b). In consequence, INO80 is targeted to NDRs through recognition of A-tract signals within promoters and ORIs and thereby guided into proximity of flanking nucleosomes.

1D search at a crowded promoter

Once INO80 is bound at an NDR, it needs to associate with the neighboring +1 nucleosome. Here, we observed that INO80 performs 1D diffusion along DNA (Fig. 1a–d) and presented examples of dynamic interactions with H2A.Z-containing nucleosomes (Fig. 5b). It is well-established that nuclear proteins, which act on DNA or DNA-bound targets, combine diffusion in 3D space and 1D along DNA to reduce search times for their targets in a process called ‘facilitated diffusion’^{23,34,56–58}. The majority of interactions we recorded were initiated directly through 1D searches (Supplementary Fig. S6c), showing that INO80 interactions with +1 nucleosomes through prior 1D searches represent a possible pathway of association in vivo. Our observations therefore reinforce the notion that chromatin remodeling complexes use facilitated diffusion, which is supported by other recent single-molecule studies with chromatin remodelers RSC, IWS2 and SWR1^26,27. Note that we cannot exclude that the ratio of 1D to 3D association might be biased towards 1D association, due to the scarcity of nucleosomes on our DNA templates.

By closer analysis of 1D diffusion kinetics, we found that INO80 diffuses mostly by sliding along the helical pitch of the DNA (Fig. 2a, c, d). In line with this, nucleosomes and the GRF Reb1 acted as roadblocks for INO80 movement (Fig. 5b, f, g). Additionally, we observed that INO80 continued 1D searches after interacting with nucleosomes and preferably dissociated back into the same direction it bound from (Supplementary Fig. S6f). This preference can be explained by the tandem placement of its DNA-binding modules on the entry site of the extranucleosomal DNA. Dissociation into the opposite direction would require big conformational changes, i.e. un- and rebinding of both modules onto the DNA on the exit side of the nucleosome. Not being able to bypass nucleosomes, even by dissociation after remodeling, could capture INO80 within the NDR and prevent ‘spilling’ into intragenic regions (Fig. 6c). Coupled with A-tract-mediated DNA binding, this mechanism might drive differential  +1 nucleosome remodeling by INO80. In addition, it promotes repositioning of nucleosomes into the NDR, by assuring that INO80 binds them in the correct orientation.

Intersegmental transfers

In areas with high DNA density, we repeatedly observed that INO80 and Ct.ΔN-INO80 complexes, but not the A-module mutant, switched to a neighboring DNA strand without changes in y-position (Supplementary Figs. S2b, c, g, h, S4c), indicating the occurrence of intersegmental transfers. Intersegmental transfers have been shown to increase efficient target site search in theoretical models and in single-molecule experiments for small DBPs by reducing local oversampling, especially in environments with high DNA density, where many transfer opportunities arise^24,59,60. It is tempting to speculate that to achieve this, INO80 might—like ISWI—use a ‘monkey-bar’ mechanism to transfer to the neighboring DNA-strand, a bridging mechanism that would be available to protein complexes with multiple DNA-binding domains/subunits that are flexibly connected^39,61 (Fig. 6d). In addition, all INO80 variants were able to jump in cis and trans, most likely through dissociation, 3D search and reassociation (Supplementary Figs. S2a, c, S4e). Enhancement in mobility through intersegmental transfers could increase overall  +1 nucleosome targeting, especially in euchromatic regions or in the context of transcriptional condensates, where NDRs are thought to be in close proximity⁶².

Implications for nucleosome remodeling

After association with the  +1 nucleosome, INO80 pulls it towards the NDR⁵. For accurate positioning, INO80 remodeling has to be stopped through negative regulation by external cues. So far, two such cues have been identified indirectly, through analysis of nucleosome positions in vitro and in vivo. The first cue includes blocking of DNA translocation by a barrier protein, bound to DNA (e.g., Reb1), through an unknown mechanism^7,10, while the second involves DNA shape features within the nucleosomal and extranucleosomal DNA, which impede the remodeling process^10,21. Note that promoters with the Reb1 motif also contain positioned DNA shape features, and current data indicate a synergistic effect of both cues.

Inhibition through a physical barrier might include specific interactions between the barrier and INO80 through protein-protein contacts with the N-module, which inhibit INO80’s remodeling capabilities allosterically. Alternatively, a passive barrier bound strongly to DNA could be enough to prevent INO80 from reeling in further DNA, and specific contacts would not be required. We showed that Reb1, bound to its motif, is a potent barrier for INO80 diffusion, but strikingly, we did not observe long-lived interactions that imply specific binding (Fig. 5e–g). Therefore, our observations are consistent with the passive barrier model. In line with this, Reb1 can act as a regulator for both  +1 nucleosome positions at bidirectional promoters in vivo, and both ORC and Reb1 can perform this function irrespective of motif orientation^7,10. Furthermore, to our knowledge, no common interaction motifs between INO80 and any of the proposed barrier proteins have been found so far.

Inhibition through DNA shape features, on the other hand, requires tracking and read-out of DNA shape by INO80 during nucleosome remodeling. Our experiments show that the A-module can individually recognize high |Ψ| and high EP (Fig. 3e and Supplementary Fig. S4d), reinforcing its proposed role as a sensor for structural DNA features in the INO80 complex that regulates nucleosome sliding. In agreement with this, the placement of three phased A-tracts at the N-terminal end of the A-module was enough to nearly abrogate INO80 remodeling (Fig. 4a–d). In contrast, introducing a single A-tract reduced the remodeling speed only modestly. This indicates that for efficient inhibition, multiple interfaces of the A-module need to bind high |Ψ| DNA in parallel. Consecutive A-tracts, as they are found in promoters, might therefore act as a stop signal for nucleosome positioning²⁸. Consistent with this theory, in vivo shape profiles near +1 nucleosome positions revealed a broad peak of high |Ψ| at the same position -95 to -120 bp from the nucleosome dyad (Fig. 4f), similar to shape profiles near  +1 nucleosomes remodeled by INO80 in vitro¹⁰. Intriguingly, this peak is largely absent in inducible promoters and promoters with a TATA-box (Supplementary Fig. S5e), suggesting a heightened importance of +1 nucleosome positioning by INO80 at TATA-less promoters. Indeed, TATA-less promoters show more uniform +1 nucleosome positioning, and recent studies indicate that in the absence of a canonical TATA-box, effective assembly of the pre-initiation complex is reliant on the position of the +1 nucleosome⁶³.

While the molecular details of DNA shape recognition remain unclear, the set of shape features preferred by INO80 implies a readout of the negative electrostatic potential within the minor groove through arginines and lysines (Supplementary Fig. S3b, d)⁴⁰. This is supported by a recent study, which showed that mutation of four positively charged amino acids at the N-terminal end of the A-module led to a severe speed-up of remodeling and ATP hydrolysis, while drastically reducing the DNA binding capabilities of the A-module¹⁹.

Taken together, our findings suggest that inhibition of remodeling by INO80 through DNA shape is connected to stronger binding by the A-module. Interestingly, this indicates that remodeling inhibition through a barrier or through DNA shape features may act by a similar mechanism, namely by fixing the DNA at a certain position in front of the ATPase domain. Although anchoring of the DNA provides a mechanism to prevent effective DNA translocation, it seems counterproductive for dissociation of the complex. Intriguingly, current models predict overwinding of the DNA in front of the motor during remodeling⁴⁵. Strong binding of the A-Module might enhance this effect: as the motor goes through additional ATP cycles, the deformation stress of the DNA cannot be resolved by rotation anymore (Fig. 6e). This would cause additional positive torsional strain, which is associated with minor groove narrowing⁶⁴ and could therefore stabilize A-module-DNA interactions even further. However, whether these twist deformations could directly cause eviction of the complex by disrupting the binding of the ATPase motor or whether they might induce conformational changes that lead to eviction has to be investigated by future research.

Our results demonstrate that both the target-site search of INO80 and its capacity to accurately position nucleosomes are closely connected to its ability to dynamically read out DNA shape features. Intriguingly, the remodeling activities of other remodelers like ISW1a or RSC are also modulated by these features⁶⁵. In light of these findings, DNA shape patterns, originating through A-tracts, emerge as central cues, around which remodeler recruitment, activity, and therefore promoter architecture are organized, opposing previous theories of intrinsic nucleosome exclusion mechanisms. Additionally, our observations underpin recent studies, which show that the concept of facilitated diffusion extends to chromatin remodelers. In the case of INO80, 1D and 3D diffusion are combined with characteristics of its molecular architecture to exploit traits that highlight target nucleosomes and thereby maximize engagement. Taken together, our study supports the notion that remodelers have evolved into complex machines that are able to simultaneously integrate multiple forms of information to construct diverse chromatin landscapes.

Methods

Protein expression and purification

S.cerevisiae wild-type INO80

The yeast strain for S. cerevisiae Wild-type INO80 originated from the S288c genetic background and was generated as previously described⁶⁶. In brief, 15 subunits of INO80 were cloned in pairs into one of four versions of a yeast expression vector containing a bidirectional GAL promoter pJF2-5. The exception was Taf14, which was paired with Gal4. Ino80 was tagged at the C-terminus with a 3x-FLAG-tag and a ybbR-tag connected by a short linker. For each of the four markers, two expression plasmids were created, which were transformed sequentially either into a MAT alpha or MAT a version of yJF1. The two final haploid strains, each expressing eight proteins, were mated to produce the diploid INO80 expression strain.

Wild-type INO80 was expressed endogenously from S. cerevisiae and purified similar to a previously described protocol⁶⁶. In short, W303-1a cells were grown in 8L YPD for 24h at 30 °C and were then collected by centrifugation (6000 g, 10 min, 4 °C). The pellet was resuspended in an equal volume of 2x lysis buffer (1x = 25 mM HEPES-KOH pH 7.6, 500 mM KCl, 10% glycerol, 0.05% NP-40, 1 mM EDTA, 1 mM DTT, 4 mM MgCl₂) with protease inhibitors 0.2 mM PMSF (Roth), 0.2 μM pepstatin A (Genaxxon), 0.2 μg/mL aprotinin (Genaxxon), and 2 μM leupeptin (Genaxxon). Cells were then frozen in liquid nitrogen in a dropwise manner and the frozen cells were crushed using a Freezer Mill (SPEX SamplePrep 6875 Freezer/Mill) (6 cycles for 2 min, crushing rate 15). For purification of the complex, the frozen cell powder was thawed, resuspended in 50 mL 1x lysis buffer with protease inhibitors and cleared by ultracentrifugation (235,000 g, 1 h, 4 °C). The supernatant was incubated with 1.5mL pre-washed anti-FLAG M2 affinity gel resin (Sigma) for 1 h at 4 °C and applied to a gravity flow column. The resin was washed with 4 column volumes (CV) of lysis buffer and 1 CV of wash buffer 25 mM Tris-HCl pH 7.2, 200 mM KCl, 10% glycerol, 0.05% NP-40, 1 mM EDTA, 1mM DTT, 4 mM MgCl₂). INO80 was eluted by 1mL of the same buffer with 0.5 mg/mL 3xFLAG peptide (Sigma), followed by 2x 1mL of buffer with 0.25 mg/mL 3xFLAG peptide. Eluted fractions were analyzed by SDS-PAGE, and pooled.

For fluorescent labeling, Sfp (made in-house) was mixed with INO80 in a molar ratio of 2:1 and a 10x excess of LD555-CoA dye (Lumidyne) as well as 15 mM MgCl₂. The reaction was incubated for 30 min at 25 °C and further purified with a Mono Q 5/50 GL column using a gradient from 100 to 600 mM KCl (25 mM Tris-HCl pH 7.2, 25% glycerol, 0.05% NP-40, 1 mM EDTA, 1 mM DTT, 4 mM MgCl₂). The protein was stored at -70 °C.

The unlabeled S.cerevisiae wild-type INO80 complex used for sliding assays was purified as previously described¹⁰. In brief, coding sequences for S. cerevisiae Ino80 carrying a C-terminal 2xFlag-tag, Rvb1, Rvb2, Arp5-His, Ies6 (pFBDM1) and Actin, Arp4, Arp8, Taf14, Ies2, Ies4, Ies1, Ies3, Ies5, and Nhp10 (pFBDM2) were subcloned into pFBDM vectors. Bacmids of both vectors were generated, and protein was expressed using Trichoplusia ni High Five cells (Invitrogen, B85502). Expressed protein was harvested by centrifugation, and protein complexes were purified using ANTI-FLAG M2 Affinity Gel (Sigma-Aldrich, A2220) and subsequent anion exchange chromatography (MonoQ 5/50 GL, GE Healthcare).

C. thermophilum Ct. ΔN-INO80

The Ct.ΔN-INO80 mutant from C. thermophilum was cloned using the MultiBac technology. The gene coding for Ino80^718-1848-2xFLAG-ybbR was cloned in pACEBac1 (Addgene); genes for Rvb1 and Rvb2 were cloned in pIDC; and genes coding for Arp5, Ies6, and Ies2 were cloned in the pIDK vector. Together, they were combined in one bacmid. Genes coding for Ies4 and Taf14 were cloned in pACEBac1, and genes coding for Arp8, actin, and Arp4 were cloned in a pIDK vector and combined on a separate bacmid. PirHC (Geneva Biotech) and E. coli XL1-Blue (Stratagene) cells were used for all recombination steps by the addition of Cre recombinase (NEB). Both baculoviruses (1:200) were used to infect 1L of Trichoplusia ni High Five cells (Invitrogen, B85502) for expression.

After addition of viruses, cells were cultured for 60 h at 27 °C and harvested by centrifugation at 4 °C. Cells were disrupted in lysis buffer (30 mM HEPES pH 7.8, 400 mM NaCl, 10% glycerol, 0.25 mM dithiothreitol (DTT), 0.28 μg/mL pepstatin A, 0.17 mg/mL phenylmethylsulfonyl fluoride (PMSF), 0.33 mg/mL benzamidine, 2 mM MgCl₂) for complex purification and gently sonicated. Raw lysate was cleared by centrifugation at 20,500 g at 4 °C for 30 min. Supernatant was incubated with 2 mL of ANTI-FLAG M2 Affinity Gel (Sigma-Aldrich, A2220) for 1h and washed with 50 mL of lysis buffer and 75 mL of wash buffer (30 mM HEPES pH 7.8, 150 mM NaCl, 5% glycerol, 1 mM DTT). The protein was eluted by incubation with 4.5 mL of wash buffer (supplemented with 0.2 mg/mL FLAG peptide) in three incubation steps of 20 min each and the elution fractions were concentrated.

For ybbR-tag labeling, Sfp (made in-house) was mixed with Ct.ΔN-INO80 in a molar ratio of 4:1 and a 5x excess of LD555-CoA dye (Lumidyne) as well as 10mM MgCl₂. The reaction was performed for 40 min at 25 °C. The labeling reaction was loaded onto a Mono Q 5/50 GL column (GE Healthcare) and eluted by an increasing salt gradient (0.2 to 1 M NaCl), resulting in highly pure labeled Ct.ΔN-INO80.

S. cerevisiae INO80 A-module

The S.cerevisiae INO80 A-module mutant was cloned using the MultiBac technology. The gene coding for Ino80^440-598-2xFLAG-ybbR was cloned into pLIB by Gibson assembly of PCR products containing the ybbR-pLIB backbone and Ino80^440-598-2xFLAG. The coding sequences of the INO80 subunits comprising the A-module (Arp4, Arp8, N-Actin, Ies2, Ies4, Taf14) were cloned into pFBDM vectors and then combined into a single vector. All cloning was carried out in E. coli XL1-Blue (Stratagene). From each bacmid (generated in E. coli DH10 MultiBac cells), baculoviruses were generated in Spodoptera frugiperda (Sf21) insect cells (Thermo Fisher Scientific, 11497013). Both baculoviruses (1:200) were used to infect 100mL Trichoplusia ni High Five cells (Invitrogen, B85502) for expression.

After addition of the virus, cells were grown for 70 h and harvested by centrifugation at 4 °C. All further steps were performed at 4 °C. After resuspension in lysis buffer (50 mM Tris-HCl pH 7.9, 500 mM NaCl, 10% glycerol, 2 mM DTT), cells were supplemented with half a protease inhibitor tablet, lysed by sonication (duty cycle 40%, Intensity 5) and centrifuged for 30 min at 30,000 g. Lysate was incubated with 100 μL anti-FLAG M2 affinity gel resin (Merck, A2220) for 1h while rotating, before supernatant was removed by centrifugation (5 min, 1000 g). Anti-FLAG beads were then washed twice with 1 mL Wash buffer A (25 mM HEPES pH 8.0, 500 mM KCl, 10% glycerol, 0.05% IGEPAL CA630, 1 mM DTT, 4 mM MgCl₂) and twice with Wash buffer B (25 mM HEPES pH 8.0, 200 mM KCl, 10% glycerol, 0.05% IGEPAL CA630, 1 mM DTT, 4 mM MgCl₂). The protein was eluted in two steps by addition of 200 μL Wash buffer B (supplemented with 0.2 mg/mL FLAG peptide) for 20 min.

For labeling, the eluate was supplemented with MgCl₂ to a final concentration of 15 mM and mixed with Sfp (made in-house) in a molar ratio of 2:1 and a 10x excess of LD555-CoA dye (Lumidyne). The reaction was incubated for 30 min at 25 °C and further purified by size exclusion chromatography using a Superdex S200 10/300 GL (Cytiva) in Wash buffer B. The protein was stored at −70 °C.

S. cerevisiae Reb1

The pET21-Reb1 construct with an N-terminal 10x-Histidine-tag and a ybbR tag connected by a short linker was generated by Gibson assembly of three PCR-generated fragments containing the pET21 backbone, both tags, and the gene, respectively. All cloning for the construct was carried out in E. coli XL1-Blue (Agilent Technologies).

For expression, 500 mL culture of E. coli Rosetta(DE3) (Merck) was grown in LB medium, induced with 200 μM IPTG, cultivated for 4h at 37 °C and harvested by centrifugation at 4 °C. All further purification steps were performed at 4 °C. Cells were then resuspended in lysis buffer (50 mM NaH₂PO₄ pH 8.0, 100 mM NaCl, 20 mM imidazole, 10% glycerol, 2 mM MgCl₂) supplemented with a protease inhibitor tablet. Cells were then lysed by sonication (duty cycle 40%, Intensity 5) and 0.1 μL/mL Pierce universal nuclease was added. After 30 min the lysate was cleared by centrifugation (20 min, 20,000 g) then 5 mM beta-mercaptoethanol was added and salt concentration was adjusted to 500 mM NaCl. A gravity flow column was prepared with 2 mL of Ni-NTA beads (Macherey-Nagel) equilibrated with 10 CV of lysis buffer and the supernatant was added in three steps of 20min. Afterwards the beads were washed twice with 10 CV of wash buffer (50 mM NaH₂PO₄ pH 8.0, 1.5 M NaCl, 40 mM imidazole, 10% glycerol, 2 mM MgCl₂, 5 mM beta-mercaptoethanol). The protein was eluted from the column with 7.5 CV of elution buffer (50 mM NaH₂PO₄ pH 8.0, 150 mM NaCl, 250 mM imidazole, 10% glycerol, 2 mM MgCl₂, 5 mM beta-mercaptoethanol). Fractions containing Reb1 were pooled and dialyzed against Heparin buffer (25 mM HEPES pH 7.5, 150 mM NaCl, 5% glycerol, 1mM DTT) for 16 h before further purification on a Cytiva HiTrap Heparin HP column with a 150 to 1000 mM NaCl gradient.

For labeling, the eluate was supplemented with MgCl₂ to a final concentration of 15 mM and mixed with Sfp (made in-house) in a molar ratio of 2:1 and a 10x excess of LD655-CoA dye (Lumidyne). The reaction was incubated for 30 min at 25 °C and further purified by size exclusion chromatography using a Superdex S200 10/300 GL (Cytiva) in Reb1 storage buffer (25 mM HEPES pH 7.5, 200 mM NaCl, 10% glycerol, 1 mM DTT). The protein was stored at −70 °C.

SpyCatcher003

A pET28a-based plasmid containing a SpyCatcher003 construct with N-terminal MGGGC-extension and C-terminal 6xHis-tag was transformed in Rosetta (DE3). Cells were grown at 37 °C to an optical density 600. The expression was induced by 0.4 mM IPTG and carried out at 18 °C overnight. Harvested cells were resuspended in lysis buffer (25 mM Tris-HCl pH 7.5, 500 mM NaCl, 15 mM imidazole) supplemented with 1 mM PMSF and 0.1 mg/mL lysozyme, sonicated, and centrifuged. The supernatant was loaded onto a HisTrap FF-columns (GE Healthcare, Chicago, IL), washed with 10 CV lysis buffer, and eluted in lysis buffer containing 200 mM imidazole. Pooled fractions were gel-filtered using a Sephacryl S300 (GE Healthcare) or Superdex S200 (GE Healthcare) column. Eluted proteins were equilibrated in storage buffer (25 mM Tris-HCl pH 7.2, 150 mM NaCl) and stored at -70 °C.

S. pyogenes dCas9

To create the expression construct, the coding sequence for nuclease-deficient dCas9 was combined with an N-terminal Spytag003 and a C-terminal intein-CBD (chitin-binding-domain) cassette in a pET28a vector by Gibson assembly. All cloning for the construct was carried out in E. coli XL1-Blue.

S.pyogenes dCas9 was expressed in E. coli Rosetta(DE3) (Merck). 2 L culture was grown in LB medium, induced with 200 μM IPTG, cultivated for 24 h at 18 °C and harvested at 4 °C. Pellets were resuspended in 50 mL loading buffer (25 mM Tris pH 7.5, 150 mM NaCl) supplemented with 1 mM PMSF and 20 μg/mL lysozyme and cells were then lysed by sonication (duty cycle 40%, Intensity 5). The lysate was cleared by centrifugation (30 min, 2820 g, 4 °C) and added to 4 mL equilibrated chitin resin (NEB IMPACT-system) in a gravity flow column. The column was washed with 20 CV of wash buffer (25 mM Tris pH 7.5, 1M NaCl) and on-column cleavage was induced by incubation in 3 CV wash buffer supplemented with 50 mM DTT for 40 h. To elute the protein, flow was allowed to continue, eluted fractions containing the protein were dialyzed against storage buffer (25 mM Tris pH 7.5, 150 mM NaCl, 1 mM DTT) and concentrated using Amicon Ultra-15 filters (30 kDa MWCO, Merck). The protein was stored at -70 °C.

The protein was labeled using the Spytag/Spycatcher system⁶⁷. Spycatcher003 protein with an N-terminal MGGGC-extension was labeled by slow addition of a 4x molar excess of ATTO643-maleimide dye (ATTO-Tec) and rotation for 2 h at 25 °C. Precipitates were removed by centrifugation and excess dye by desalting using a HiTrap Desalting column (Cytiva) in Desalting buffer (100 mM Na₂HPO₄ pH 7.5, 150 mM NaCl). Labeled Spycatcher003 was then incubated with purified dCas9 for 30 min and the conjugation reaction was again cleaned up by size exclusion chromatography as described above, but in dCas9 storage buffer (10 mM Tris pH 7.5, 300 mM NaCl, 0.5 mM EDTA, 1 mM DTT, 10% glycerol) and stored at -70 °C.

Purified dCas9 was incubated with annealed gRNA (Metabion) in ratio of 1:1.2 for 20 min at 25 °C and assembled RNP complexes were stored at -70 °C.

H. sapiens Histone octamer

Recombinant H. sapiens Histones H2A.Z and H4 were purchased (The Histone Source). To enable site-specific cysteine labeling of histone H2B, two mutations were introduced into the constructs of H2B and H3.2 by PCR and ligation (H2B: T116C, H3: C111A). All cloning for the constructs was carried out in E. coli XL1-Blue (Agilent Technologies).

H. sapiens histones H2A, H2B, and H3.2 were expressed in E. coli LOBSTR (Kerafast). For each histone 1 L culture was grown in LB medium, induced with 400μM IPTG, cultivated for 4 h at 37 °C and harvested at 4 °C. Cells were then resuspended in 10 mL Tris-sucrose buffer (50 mM Tris-HCl pH 7.5, 10% (w/v) sucrose) and flash-frozen at -80 °C. Histones were then purified from inclusion bodies and reconstituted to nucleosomes as described before, but with small adjustments⁶⁸. 0.05% (w/v) Tween-20 was used instead of Triton X-100. For the nucleosome reconstitutions the histone mixture was dialyzed against refolding buffer with a total of five buffer changes and refolding buffer was supplemented with 0.2 mM TCEP instead of beta-mercaptoethanol from the first buffer change onward. After dialysis, aggregates were removed by centrifugation and the mix was concentrated to 1.5 mL using Amicon Ultra-15 filters (10 kDa MWCO, Merck). The labeling reaction was started by slow addition of 10x molar excess of ATTO643-maleimide dye (ATTO-Tec) to the mixture and rotation for 16 h at 4 °C. Labeled octamers were purified from labeled dimers by size exclusion chromatography as described in the original protocol⁶⁸. Purified canonical octamers were used for generation of mononucleosomes and H2A.Z-containing octamers for generation of chromatinized λ-DNA constructs.

H. sapiens mononucleosome

6-FAM-labeled DNA template for 0N80 nucleosomes was produced by PCR, using a plasmid carrying the Widom 601 sequence¹⁹, as well as primers with overhangs, which generated the different linker DNA sequences. 6-FAM labeled short primers, were then used to produce PCR products were purified by anion exchange chromatography (CaptoQ) with a salt gradient from 150 mM to 2 M NaCl in TE-buffer (25 mM Tris-HCl pH 7.5, 0.2 mM EDTA) over 15 CV, followed by size exclusion chromatography using a HiPrep 16/60 Sephacryl S-300 HR column. The DNA was eluted in TE-buffer supplemented with 150 mM NaCl and was concentrated by ethanol precipitation. Concentrated DNA was then combined with H. sapiens histone octamer at a 1.1-fold excess in reconstitution buffer (25 mM HEPES pH 7.5, 2 M NaCl, 10% glycerol, 0.25 mM DTT) in 7 kDa MWCO Slide-A-Lyzer Dialysis Units (Thermo Fischer). Using a pump the salt concentration was decreased to 50 mM NaCl over 18 h at 4 °C. After dialysis, the nucleosomes were purified by anion exchange chromatography using a SourceQ 1 mL column. Nucleosome species were eluted by a salt gradient from 450 to 800 mM NaCl over 25 CV in reconstitution buffer. Fractions containing nucleosomes were pooled and dialyzed against reconstitution buffer containing 50 mM NaCl at 4 °C overnight. Finally, nucleosomes were concentrated to 1 mg/mL using Amicon Ultra 0.5 mL filters (30 kDa MWCO, Merck) and stored at -70 °C.

Generation of λ-DNA constructs for DNA curtains

λ-DNA constructs for DNA curtains were prepared as previously described⁶⁹. Oligos containing biotin (5'-P-AGGTCGCCGCCC-bio-3') or digoxigenin (5'-P-GGGCGGCGACCT-dig-3') were hybridized to the cos sites of the lambda DNA, ligated overnight at room temperature using T4 ligase (NEB), and purified using a HiPrep 16/60 Sephacryl S-300 HR column (Cytiva) in TE150 buffer. For the assembly of nucleosomes, λ-DNA constructs were concentrated using Amicon Ultra-15 filters (30 kDa MWCO, Merck). 30 μL of a 30:1 reconstitution mix (10 mM Tris-HCl pH 7.5, 1 mM EDTA, 2 M NaCl, 0.25 mM DTT, 10% glycerol, 0.2 mg/mL BSA, 27 nM octamers^H2A.Z, 0.9 nM λ-DNA) was dialyzed against reconstitution buffer (25 mM HEPES pH 7.5, 2 M NaCl, 10% glycerol, 0.25 mM DTT) using 7 kDa MWCO Slide-A-Lyzer Dialysis Units (Thermo Fischer). The NaCl concentration was decreased to 50 mM over 16 h at 4 °C using a pump, and the mix was dialyzed for an additional 4 h against reconstitution buffer containing 20 mM NaCl. The resulting chromatinized λ-DNA was centrifuged to remove aggregates and stored at -20 °C in 50% glycerol.

DNA curtains

Flow cell preparation

Custom-made flow cells were assembled from silica-fused slides (UQG Optics) grafted with chromium barriers produced via E-beam lithography and cover slips with double-sided tape using adaptations to a protocol described previously²⁹. Slides were spin-coated with a layer of polymethylmethacrylate (PMMA, Allresist, AR-P 672.03) for 5 s at 800 rpm, 45 s at 4500 rpm and then baked for 1 min at 180 °C, followed by a layer of Electra 92 (Allresist, AR-PC 5090.02), spun at 5 s at 500 rpm, 60 s at 1000 rpm and baking at 90 °C for 2 min. Barrier patterns were written by Ebeam lithography using an RAITH e-LINE scanning electron microscope equipped with a pattern generator and lithography control system. After patterning, Electra92 was rinsed off with deionized water, and the resist was developed using a 3:1 solution of 2-propanol and methyl isobutyl ketone (MIBK). The substrate was then rinsed in 2-propanol and dried with N₂. A 20 nm layer of Chromium was deposited directly onto the fused silica using a Semicore electron beam evaporator, and PMMA was lift-off using acetone.

Flow cell loading and lipid master mix preparation were adapted from a previous protocol²⁹. In brief, 100 mg DOPC (Avanti 850375P-200 mg) dissolved in 1 mL chloroform, 1 mL DOPE-PEG (Avanti 880130C-25 mg) and 50 μL DOPE-biotin (Avanti 870273C-25 mg) were mixed and stored at -20 °C. 100 μL Master Mix was dried using N₂ followed by applying a vacuum for 16h and subsequently resolved in 2 mL lipid buffer (10 mM Tris pH 7.5, 200 mM NaCl). Lipids were sonicated five times for 1 min with 1 min pauses on ice (amplitude 20%, duty cycle 20%), filtered through 0.22 μm PVDF filters and stored at 4 °C for 3 to 4 weeks. Lipids were diluted 1:10 in lipid buffer containing 20 mM MgCl₂ and flow cells were incubated four times for 5 min with 200 μL diluted lipids. After washout, flow cells were incubated with 2 μL of 1 mg/mL anti-digoxigenin (produced in house) in 500 μL lipid buffer for 15 min, followed by 5 μL streptavidin (Carl Roth) in 1 mL BSA buffer (20 mM HEPES pH 8.0, 1 mg/mL BSA, 4 mM Mg(OAc)₂). After this, 20 pM naked or chromatinized λ-DNA in BSA buffer was added in four steps with 5 min incubation time.

Microscope setup

DNA curtain experiments were carried out on a prism-type TIRF microscope (Nikon Eclipse Ti2), equipped with three illumination lasers (488, 561 and 640 ms Coherent OBIS), an electron multiplying charged coupled camera (iXon Life, Andor) and a syringe-pump-driven microfluidics system supplying the sample chamber. All experiments were carried out using an OptoSplit II (Cairn Research) with an HC BS 640 beam splitter (AHF Analysetechnik) and additional filters (585/65 BP + 700/75 BP, Thorlabs). Videos were recorded in NIS-Elements (Nikon) and analyzed in Igor Pro 8 (Wavemetrics) using custom written code (see below). Videos were aquired with a frame rate and illumination time of 50ms for all experiments with the exception of Ct.ΔN-INO80, where 200 ms were used.

Single-protein diffusion

For single-protein diffusion assays on λ-DNA, the protein was diluted in 80 μL DNA curtains buffer and immediately injected into the flow cell with a constant flow of 0.3 mL/min. The flow was stopped when the protein concentration reached its peak and interactions were observed for up to 8 min per video. Protein was removed from λ-DNA with a 2 M NaCl wash before another injection. Single-molecule measurements were performed in DNA curtains buffer (20 mM HEPES pH 8.0, 1 mg/mL BSA, 4 mM Mg(OAc)₂, 1 mM ATP, 1 mM DTT, 80 mM KGlu if not indicated otherwise). Ct.ΔN-INO80 was measured at 60 mM NaCl and data shown for wt-INO80 alone was measured at 120 mM KGlu. Protein concentrations injected varied between 125 to 250 pM for INO80 variants, 2 nM for dCas9 and 75 pM for Reb1.

Nucleosome bleaching, INO80–Reb1 and INO80–nucleosome interaction

To determine nucleosome bleaching rates, flowcells were prepared using chromatinized λ-DNA (see above). The flowcell was washed with 350 mM NaCl to remove partially formed nucleosomes or labeled histones bound to DNA and remaining nucleosomes were imaged until they had all bleached. For remodeler interaction experiments, INO80 was then flushed in and the flow was stopped before live-imaging was initiated.

For Reb1 interaction assays, INO80 and Reb1 were injected either sequentially or incubated for 10 min at a 10x concentration before dilution with DNA curtains buffer and injection into the flowcell. Concentrations used were the same as in single protein diffusion experiments.

Mononucleosome sliding assays

To analyze INO80 sliding activity dependent on linker DNA shape, 50 nM S.cerevisiae INO80 was incubated with 100 nM of 6-FAM-labeled nucleosomes in sliding buffer (25 mM HEPES pH 8.0, 60 mM KCl, 7% glycerol, 0.1 mg/mL BSA, 0.25 mM DTT) at 25 °C. The reaction was started by addition of 1 mM ATP complexed with 2 mM MgCl₂ and stopped at different timepoints (1, 2, 5, 10, 20, and 60 min) by addition of Lambda DNA (0.15 mg/mL; NEB). An ATP-negative control was taken before the start of the reaction and treated like the last sample of the time series. Nucleosome species were separated by native polyacrylamide gel electrophoresis (PAGE) on a 3–12% acrylamide bis-tris gel (Invitrogen) and visualized using a Typhoon imaging system (GE Healthcare).

Sliding assays to control for sliding activity of labeled endogenous S.cerevisiae INO80 were conducted with 2 nM INO80 and 100 nM of 6-FAM-labeled nucleosomes. The rest of the assay was performed as described above.

Data analysis

Statistical analysis

Statistical analysis was performed using Igor Pro 8 (Wavemetrics). Respective type of test is stated in the main text, figure caption or methods section. Significance was set at p < 0.05 and indicated using the following nomenclature: p > 0.05 = ns.; p < 0.05= *; p < 0.01= **; p < 0.001= ***; p < 0.0001= ****. The results were reproducible and conducted with established internal controls. All samples that met proper experimental conditions were included in the analysis.

Single molecule tracking

Video files were analyzed using custom-written software in Igor Pro (Wavemetrics). Protein trajectories were localized using an algorithm inspired by DAOSTORM⁷⁰ and tracked using an in-house Markov chain Monte Carlo-based algorithm. Trajectories were manually screened for detections of particles that displayed exceedingly small (surface-bound) movement or unexpectedly bright or dim fluorescence (typically < 10%) and these trajectories were filtered out. Trajectories with less than 30 detections were also omitted from diffusion and binding analysis.

Diffusion analysis

Mean squared displacements (MSDs)

$$\text{MSD}\left(n\mathrm{\Delta }t\right)={⟨{\left(y_t-y_{t+n\mathrm{\Delta }t}\right)}^2⟩}_t,n=1,2,\dots$$ 1

were determined from tracking data. The uncertainty of the MSD values were estimated using a published procedure⁷¹ and used as weights for fits to the model

$$M\left(t\right)=2Dt+2{\sigma }^2,$$ 2

where σ is the localization noise. At short time delays, the MSDs are dominated by the relaxation time of the thermally fluctuating DNA, rather than movement of proteins along the DNA. Based on tracking of non-moving dCas9 molecules, the threshold where the DNA movement dominates was determined and MSD data with t < 200 ms were excluded from the fit. Fits were performed up to n = 100 ≡ 5 s for all proteins except Ct.ΔN-INO80, where n = 10 ≡ 2 s was used. D_Instant was identified by the same procedure applied to a sliding window of fixed length that was scanned along the trajectory. Errors for median diffusion coefficients are 68% intervals from bootstrapping.

Calculation of the free-energy barrier

The Stokes-friction-dominated diffusion coefficient of a protein diffusing along DNA was modeled by

$$D=\frac{k_{\mathrm{B}}T}{{\gamma }_{\mathrm{lin}}+{\gamma }_{\mathrm{rot}}}F\left(\epsilon \right)=\frac{k_{\mathrm{B}}T}{6\pi \eta R+{\left(\frac{2\pi }{b}\right)}^2\left(8\pi \eta R^3+6\pi \eta R{R}_{\text{OC}}^2\right)}{e}^{-{\left(\frac{\epsilon }{k_{\mathrm{B}}T}\right)}^2},$$ 3

where γ_lin and γ_rot represent linear and rotational drag, F(ε) is the retarding factor due to an energy barrier ε (Fig. 2b), η is the solvent viscosity, b is the distance between base pairs, R is the radius of the protein and R_OC is the distance of the protein from the DNA³². For calculation of the 3D diffusion coefficient γ_rot and ε were set to zero.

To determine ε for wild-type S. cerevisiae INO80, the median diffusion coefficient at 120mM KGlu was used, R was estimated from cryo-EM structures to lie between 100 to 120 Å and R_OC was assumed to be the same as R.

To compare ε between wild-type S. cerevisiae INO80 and Ch. thermophilumΔN-INO80, the median diffusion coefficient of both proteins at 40 mM KGlu and 60 mM NaCl respectively, was used, R was estimated from cryo-EM structures to lie between 100 to 120 Å for wild-type Sc. INO80 and 95 to 115 Å for Ct.ΔN-INO80 to account for the loss of the N-module. R_OC was assumed to be the same as R.

ε values of other DBPs were taken from previous single-molecule studies^32,36.

DNA switch analysis

Inter-DNA switches were determined by an autoregressive Hidden-Markov model (arHMM), previously implemented for identifying state switches in optical tweezer trajectories⁷². For each INO80 movement trajectory perpendicular to the direction of tethering, the likeliest average trajectory (i.e., jumping between neighboring DNAs) was determined by maximizing the trajectory likelihood, provided by the Viterbi algorithm^73,74, while optimizing the switching probability matrix, DNA fluctuation amplitude, and the state positions, using a More-Hebdon optimizer in Igor Pro 8. The fluctuation dynamics of tethered DNA molecules perpendicular to their tethering direction were determined by tracking DNA-bound dCas9 molecules and analyzing their mean-square displacement relaxation behavior (Supplementary Fig. S2f). MSD fits to the relaxation model $f\left(t\right)=f_{\infty }+\left(f_0-f_{\infty }\right)\exp \left(-t/{\tau }_{\mathrm{r}}\right)$ resulted in a transversal relaxation time of ≈ 68 ms, which was used as an input to the arHMM, and a transversal fluctuation amplitude of  ≈ 139 nm. The arHMM produced a list of switch frames. For each switch occurring between frames i − 1 and i, the displacement perpendicular (Δx = x_i − x_i−1) and parallel (Δy = y_i − y_i−1) to the extended DNA was recorded and compared with displacements occurring at random switch times. Additional long-range inter-DNA switches facilitated by jumps, as well as jumps on the same DNA, which are not captured by the arHMM were detected by identifying all INO80 DNA association events within a 200 ms window after a dissociation event. For each jump/transfer detected this way, the displacement perpendicular (Δx) and parallel (Δy) to the extended DNA was calculated by using the last localization before dissociation and the first localization after association. Note that this method can misclassify two separate complexes unbinding and binding in quick succession as a jump. To discriminate false positives, we calculated the Stokes-friction-dominated 3D diffusion coefficient of INO80 diffusing in solution using

$$D=\frac{k_{\mathrm{B}}T}{6\pi \eta R}$$ 4

where η is the solvent viscosity and R is the radius of the protein and determined the maximum diffusion distance r_max for a timeframe of 200 ms and a cutoff of 99.9%, meaning that only 0.1% of INO80 complexes will move further than r_max in the given timeframe. Displacements outside the border were marked as false positives and removed from further analysis.

DNA shape analysis and correlation with observables

The position of every localization of a molecule on the DNA was calculated in relation to the positions of the chromium anchors and barriers, where the DNA is tethered. The DNA propeller twist Ψ for each base pair was predicted using Deep DNAshape^41,75 and averaged in 1 kbp bins.

The protein localization probability density along the DNA was calculated by summing all localizations within a bin and normalizing by the total number of localizations. Errors are 68% intervals from bootstrapping. Binding events were binned according to the first detected localization of a trajectory, while the off-rate was calculated by dividing the number of dissociation events by the total time of protein occupancy within a bin. To determine the position-dependent apparent diffusion coefficient, D_App was calculated for all trajectories according to:

$$D_{\mathrm{App}}=\frac{\mathrm{\Delta }y}{2\mathrm{\Delta }t}$$ 5

where the Δt is a timeframe and Δy is the measured displacement within that timeframe. For Ct.ΔN-INO80, Δt was set to 2 s and for Sc. wildtype and the A-module to 5 s. Calculated D_App values were binned depending on the mean position of y along the DNA. Pearson correlation coefficients were calculated by correlating the binned parameter with the binned mean Ψ values. Bins 0 to 5 kbp and 44 to 49 kbp were excluded from the correlations due to the proximity to the chrome barriers.

Interaction analysis

Interactions of INO80 with nucleosomes and Reb1 were determined manually from Kalman-filtered tracks. Filtering was used to suppress the impact of fluctuations that are caused by movement of the DNA. To ensure that candidates for collisions are bound to the same DNA, the transversal movement perpendicular to the end-to-end vector of the DNA for both candidates was analyzed. Positive correlation and a cross-correlation decay compatible with the transversal relaxation time of the DNA were used as an indicator for binding to the same DNA. Interactions with 1D searches before and after the interaction were further separated into block if both 1D searches happened on the same side of the protein, or bypass otherwise. It should be noted that, using this definition, a bypass does not necessitate a jump of INO80 over the protein constituting a barrier, but can include specific binding to the protein followed by 1D diffusion on the other side.

Eviction assay — photobleaching

To assess H2A.Z/H2B dimer eviction by INO80, nucleosomes imaged with and without INO80 injection were tracked, and the time durations until fluorescence signal loss were determined manually from fluorescence trajectories. Lifetimes of nucleosome fluorophores imaged without injection of INO80 were calculated, and nucleosomes surviving beyond the end of the recording were marked as censored. For fluorophores of nucleosomes imaged with INO80, only the intervals during interaction with INO80 were considered, and nucleosomes without bleaching/eviction events during the interaction were also marked as censored. Collected lifetimes were then compared using log-rank tests, and survival plots were created using the Kaplan-Meier method⁷⁶. Error bars are 68% intervals from bootstrapping.

Sliding assay analysis

Band intensities were analyzed using ImageLab 6.1 (Bio-Rad) and IGOR Pro 8. Previous experiments have shown that hexasomes are remodeled by INO80 and that fully remodeled 0H80 hexasomes co-migrate with unremodeled 0N80 nucleosomes in native acrylamide bis-tris gels¹². Nucleosome purification can contain trace amounts of hexasomes, visible as slight bands below the 0N80 nucleosome. To prevent erroneous calculation of the unremodeled nucleosome band, it was corrected for the amount of remodeled hexasome present at each time point. For that, the fraction of all nucleosome species, including the hexasome, was determined at every time point. Then, to determine how much hexasome was remodeled, the fraction of the hexasome band was compared to the first measured time point (1 min) and the difference was subtracted from the unremodeled nucleosome band. Finally, to determine whether nucleosomes with sequence differences were remodeled differently, all nucleosome bands (unremodeled, intermediates, fully remodeled) were normalized to 100% and fully remodeled band fractions for different nucleosomes at each time point were compared using a two-tailed t-test.

Generation of in vivo shape profiles

DNA shape features for the whole yeast genome were predicted using Deep DNAshape^41,75. Genome feature annotations were taken from the Saccharomyces Genome Database (SGD). TATA-box annotations were taken from⁷⁷. Dyad distributions of three replicates, identified in ref. ⁴⁷, were combined. For all dyads within  ± 90 bp of annotated +1 nucleosomes positions, shape values in a 300 bp window around each dyad were extracted and averaged, resulting in a weighted-average shape profile for each +1 nucleosome in the yeast genome. Shape profiles belonging to each promoter class were then oriented according to the direction of transcription, combined and averaged again. For analysis of uni- or bidirectional promoters, shape values around annotated NDR midpoint positions were combined and averaged. Analysis and calculations were done in IGOR Pro 8 using custom-written code.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Investigation: G.L. and D.H.; methodology: G.L.; data analysis: G.L. and J.S.; data visualization: G.L. and S.Z.; programming: G.L. and J.S.; writing: G.L., S.Z., and J.S.; supervision: J.S.; funding aquisition: J.S.

Peer review

Peer review information

Nature Communications thanks Aakash Basu and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

G.L, D.H, S.Z, and J.H. were supported by the LMUexcellent investment fund, an ERC Starting Grant [758124], and funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) [577467251]. Open Access funding enabled and organized by Projekt DEAL.

Data availability

Publicly available nucleosome dyad data generated in⁴⁷ can be found in: GSE97290. Large unprocessed video data files are available upon request from the corresponding author. Source data are provided with this paper.

Code availability

All code generated for this study was implemented in IGOR Pro 8.04 and is deposited at Zenodo: 10.5281/zenodo.22081857.⁷⁸ All source code is provided under the GNU General Public License (GPL v3).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77906-1.