Structure of a RSC-nucleosome complex and insights into chromatin remodeling.

Statistical analysis of RSC–nucleosome cryo-EM images For EM analysis of the RSC–nucleosome complex, we mixed RSC complex with a four-fold molar excess of nucleosomes. Quantitative analysis of the RSC–nucleosome interaction, performed under the conditions used for EM sample preparation, gave a K d value of ~7 nM ( Supplementary Fig. 6a online). Previous analysis in stain had detected density in the central RSC cavity after incubation of the complex with nucleosomes ( Supplementary Fig. 1 ). However, the detection of partially disordered structural elements near the cavity made it crucial to establish whether any changes observed after incubation with the nucleosomes were directly related to the presence of a nucleosome. First, it was important to determine whether incubation of RSC with nucleosomes led to an overall rearrangement of the RSC structure. To this end, we separated images in the RSC–nucleosome data set (~37,000) roughly according to their orientation by aligning them to reprojections of the RSC reconstruction. Comparison of the outcomes from repeated rounds of reference-free alignment within the resulting RSC–nucleosome classes indicated that RSC–nucleosome images had been successfully segregated into homogeneous groups by alignment to projections of the RSC structure and established an overall similarity between images of the RSC–nucleosome complex and RSC alone. Having established that incubation with nucleosomes did not cause large-scale changes in RSC structure, it was important to determine whether smaller differences between images of RSC alone and the RSC–nucleosome complex were directly related to the presence of a nucleosome. To this end, we combined the RSC and RSC–nucleosome data sets to use the RSC images as an internal control during image analysis. Statistical analysis was focused on projections related to the front view of the RSC structure, in which the central cavity where the nucleosome was likely to bind was most apparent. We used correspondence analysis and classification 19 to separate images according to density variations in and around the central cavity. The particle images were clearly separated between groups with and without density in the cavity, with the first eigenvector (corresponding to the most statistically significant source of variability in the images) accounting for most of the observed differences and apparently relating directly to the presence of additional density ( Supplementary Fig. 6b ). Plotting the distribution of images classified along the first eigenvector indicated a single class of particles with an empty cavity for the RSC data set, whereas images from the RSC–nucleosome data set were divided into two classes, with ~65% of the particles classified as having an occupied cavity and ~35% assigned to a group with an empty cavity ( Supplementary Fig. 6c ). Final classification of the mixed RSC and RSC–nucleosome data set resulted in a clear distinction between particles with and without density in the central cavity ( Supplementary Fig. 6d , first and second rows, respectively). Differences between occupied and empty averages for a given orientation revealed the shape of the density in the RSC cavity ( Supplementary Fig. 6d , below).

Refinement of RSC–nucleosome cryo-EM data

Having obtained direct evidence that the RSC–nucleosome images contained additional density in the cavity and that incubation of RSC with nucleosomes did not cause an appreciable change in RSC structure, we subjected the RSC–nucleosome data to three-dimensional reconstruction by projection matching, using a low-pass filtered RSC volume as initial reference. Density in the cavity was immediately apparent. Because the concentration of RSC used to prepare EM samples (~20 nM) was close to the estimated K d for formation of the RSC–nucleosome complex ( Supplementary Fig. 6a ), and because two-dimensional image analysis indicated that a fraction (about one-third) of images in the RSC–nucleosome data set corresponded to RSC alone ( Supplementary Fig. 6c ), we analyzed images in the RSC–nucleosome data set using a competitive projection-matching protocol 20 , 21 in which they were compared to projections of the initial RSC–nucleosome complex and RSC alone reconstructions. In agreement with the results obtained from two-dimensional classification, 65 ± 10% of the RSC–nucleosome images were matched to projections of the RSC–nucleosome structure, with the rest showing higher cross-correlation to projections of RSC alone. Finally, we subjected the images corresponding to actual RSC–nucleosome complexes (~20,000) to refinement by projection matching using the RSC reconstruction as an initial reference. The resulting RSC–nucleosome reconstruction was close in overall appearance to the RSC reconstruction. However, when contoured at a threshold level at which the RSC portion of the reconstruction matched the reconstruction of RSC alone, the RSC–nucleosome structure showed an occupied central cavity ( Fig. 3a ). Difference mapping revealed negligible local changes throughout the structure ( Supplementary Movie 3 online) and, in agreement with the results from two-dimensional image analysis, the presence of additional density in the central cavity and its immediate vicinity ( Fig. 3b ).

Analysis of the RSC–nucleosome cryo-EM reconstruction

Image analysis conclusively established that density in the center of the RSC–nucleosome reconstruction resulted from incubation of RSC with nucleosomes. However, unexpectedly, the shape of the density in the cavity of the RSC–nucleosome reconstruction did not match the shape of a nucleosome. The central portion of this density seemed to correspond in size and shape to a histone octamer. This was particularly apparent in a top view, where the outline of the histones and a characteristic dimple at the center of the octamer could be easily identified, immediately surrounded by either RSC, or what seemed to be poorly ordered DNA density ( Fig. 4a ). Docking of a resolution-matched model of the histone octamer derived from its X-ray structure confirmed the match and indicated the orientation of the octamer in the cavity ( Fig. 4b and Supplementary Movie 4 online). We were only partially able to resolve the density for one of the H2A-H2B dimers (marked by a red asterisk in Fig. 4b ). This could be due to partial dissociation or to inadequate segmentation of the dimer and RSC density resulting from the limited resolution of the cryo-EM reconstruction compounded with close association of the dimer with the mobile bottom domain of RSC. Finally, the fit of the histones in the central density is not perfect, and the possibility that interaction with RSC might induce changes in the structure of the octamer cannot be excluded. A determination of the distribution of DNA density surrounding the octamer was complicated by limited resolution and the close proximity of the RSC domains. Density adjacent to the histones and probably corresponding to nucleosomal DNA was readily apparent in two areas that have no RSC density nearby, but not in areas where RSC comes into close contact with the nucleosome or near the dyad ( Fig. 4a ). In contrast, reconstructions of nucleosomes on their own, preserved in stain (Y.C. and F.J.A., unpublished data) or in the frozen, hydrated state 22 , show clear DNA density surrounding the histone octamer. Because the RSC–nucleosome images were recorded under conditions where no DNA translocation was possible (no ATP was present), we surmise that our detection of DNA density was limited by a combination of mobility and limited segmentation from the closely associated RSC density. Interaction of the nucleosome with RSC might alter the balance of histone-DNA contacts that underlie nucleosome stability and result in changes in nucleosome structure. The histone octamer fits tightly into the RSC cavity, leading to an extensive interface between RSC and the free (top and bottom) surfaces of the octamer ( Supplementary Movie 4 ). RSC–nucleosome interactions along the periphery of the nucleosome are more localized, with RSC density coming into contact with the sides of the nucleosome at a site near the dyad ( Fig. 4c , protein density 1) and also interacting with it three to four helical turns to the right and left of the dyad ( Fig. 4c , protein densities 2 and 3, respectively). These positions must correspond to some of the multiple sites of RSC-DNA interaction previously revealed by cross-linking 23 , and one of them should correspond to the location of Sth1. No high-resolution structure of Sth1 is available, but the X-ray structure of the core domain of the Sulfolobus solfataricus Swi2/Snf2 ATPase complexed with DNA 24 (30% identity and 32% similarity to the core domain of Sth1) could be docked into protein densities 1 or 2 in our model, resulting in a reasonable alignment of the DNA of the X-ray structure with the expected position of nucleosomal DNA. However, only docking into density 1 is consistent with the position for interaction of Sth1 with the nucleosome suggested by functional studies. Interaction of RSC with the nucleosome results in marked sensitivity to DNase I digestion at a site approximately two turns from the nucleosomal dyad (that is, the center of the 146-bp DNA fragment), which seems to result directly from binding of the ATPase portion of Sth1 (ref. 10 ) at that position. Features in the RSC–nucleosome reconstruction are consistent with this observation. The outline of the H3-H4 tetramer is apparent in the RSC–nucleosome reconstruction, consistent with the idea that RSC binding leads to separation of DNA from the histones and formation of a bulge around the dyad. Such disruption of DNA-histone contacts near the dyad may also contribute to the apparent disordering of nucleosomal DNA near the ends of the nucleosome ( Fig. 4c ).

METHODS Electron microscopy sample preparation and data collection

We purified RSC as described 4 . Nucleosomes were prepared from rat liver histone octamers and a 160-bp DNA fragment containing the nucleosome-positioning sequence of the Xenopus laevis 5S ribosomal RNA (rRNA gene) as described 8 , and they were stored in buffer containing 10mM Tris, 10mM NaHSO 3 , 0.01 mM EDTA, pH 7.5. To prepare RSC EM samples, we diluted RSC aliquots (~420 μg ml −1 in 100 mM potassium acetate, 20mM HEPES, 10% (v/v) glycerol, 1 mM DTT, 1 mM EDTA, 0.01% (v/v) Nonidet P-40 and 1× protease inhibitors, pH 7.5) to a final concentration of ~25 μg ml −1 with a buffer containing 15 mM HEPES, 3 mM MgCl2 and 100 mM potassium acetate, pH 7.5. About 3 μl of protein was applied to a freshly glow-discharged (in the presence of amyl amine) carbon-coated Maxtaform, 300-mesh Cu/Rh grids (Ted Pella, Inc.) and preserved by flash freezing in amorphous ice 31 . We assessed nucleosome binding by RSC by gel electrophoresis as described 5 . The K d value (~7 nM) was determined from the slope of a double reciprocal plot of the intensity of the free-nucleosome band in the gel as a function of RSC concentration. To prepare the RSC–nucleosome complex, nucleosomes were mixed with RSC at a molar ratio of ~4:1 and diluted (15 mM HEPES, 3 mM MgCl 2 , 50 mM potassium acetate, pH 7.5) to obtain a final RSC concentration of ~25 μg ml −1 . The mixture was incubated at ~25 °C for 15 min. We prepared EM samples as described for RSC alone. We collected EM images under low-dose conditions using either a CM200 FEG or a Tecnai F20 microscope (Philips/FEI) equipped with field emission gun, operating at an accelerating voltage of 120 kV. Images were recorded on Kodak SO-163 film, at a magnification of ×66,000 or ×62,000 and with underfocus values between 1.3 and 3.2 μm. Micrographs were digitized on a Zeiss/SCAI flat bed densitometer (ZI/Zeiss) using a step size of 7 μm. Digitized images were two-fold pixel-averaged, resulting in a final pixel size corresponding to 2.06 Å. A total of 159 micrographs of the RSC complex and 198 micrographs of the RSC–nucleosome complex were digitized, yielding ~26,000 and ~37,000 particle images, respectively. The data sets were divided into defocus groups according to defocus values calculated independently for 12 distinct sections of every micrograph. All image analysis was carried out using the SPIDER software package 32 .

Calculation of the RSC cryo–electron microscopy reconstruction

We used a published reconstruction of RSC obtained by using the Random Conical Tilt (RCT) method 33 and images of RSC particles preserved in stain 6 as a starting point for determination of an initial cryo-EM RSC reconstruction using a hybrid strategy for particle-orientation determination. In this hybrid strategy, supervised and unsupervised alignment and classification were used in combination with a modified version of the common-lines algorithm implemented in the SPIDER image analysis package 32 , to progressively compensate for distortion of the initial RCT RSC volume caused by particle dehydration and flattening and missing cone artifacts ( Supplementary Fig. 3 ). We further refined the resulting RSC cryo-EM reconstruction using an iterative projection-matching algorithm 15 with angular separation between reference projections progressively decreasing from 15° to 2°. Consistency of the reconstructed volume with the cyro-EM data was monitored by comparing reprojections of the volume to averages obtained directly from the original images by multiple rounds of the reference-free alignment 34 performed in corresponding particle classes defined by supervised classification.

RSC structure-flexibility analysis

For domain-flexibility analysis, the effect of the contrast transfer function (CTF) was minimized by phase flipping, and RSC images were low-pass filtered to a spatial frequency corresponding to one-fourteenth of an angstrom. Images were divided into 195 reference groups according to alignment parameters determined by matching to reference projections calculated from the three-dimensional RSC reconstruction using an angular separation of 10°. We generated three-dimensional masks covering regions of interest and, from them, obtained two-dimensional masks for each specific projection. Correspondence analysis 35 of the densities under the two-dimensional masks, followed by Hierarchical Ascendant Classification using the Ward’s criterion 36 , were used to generate class averages. Two-dimensional occupancy analysis in RSC–nucleosome images Images of RSC–nucleosome particles were CTF corrected, low-pass filtered and divided into groups by alignment to projections of the RSC cryo-EM reconstruction, as described above for the RSC data. RSC and RSC–nucleosome images corresponding to the same projection directions were combined, keeping track of the data set from which each image originated. We created a cylindrical three-dimensional mask including the central RSC cavity and surrounding regions and used it to generate twp-dimensional masks for each projection direction. Multivariate statistical analysis and classification were performed as described above. Analysis of eigenimages, as well as the distribution of particle images along eigenvectors, immediately indicated sharp differences between RSC and RSC–nucleosome data sets. Final image classification using 12 factors yielded class averages showing well-defined differences in the central cavity, as highlighted by difference mapping.

Calculation of the RSC–nucleosome cryo–electron microscopy reconstruction

We subjected RSC–nucleosome data to ten rounds of refinement by projection matching using as the initial reference a low-pass filtered RSC volume. Additional density in the RSC central cavity became immediately apparent. However, because the two-dimensional image analysis suggested heterogeneity of the RSC–nucleosome data set, we used a competitive-refinement strategy 20 , 21 to identify images in the RSC–nucleosome data corresponding to full complexes. Images in the RSC–nucleosome data set were iteratively cross-correlated to projections of the RSC and RSC–nucleosome volumes and classified as occupied (nucleosome bound) or empty (without a nucleosome). The portion of the RSC–nucleosome data classified as occupied was then refined using the structure of RSC alone as the initial reference. As part of the nucleosome density in the RSC–nucleosome reconstruction is disordered, the appropriate threshold for rendering the RSC–nucleosome reconstruction could not be established simply from the anticipated mass of the complex and was instead determined by matching the RSC portion of the two volumes. Difference mapping and comparison with the nucleosome X-ray structure We carried out volume inspection and comparison with the X-ray structure of the nucleosome using Chimera 37 , which was also used to generate figures. Before difference mapping, all volume densities below the threshold value chosen for surface representation were set to zero. Volume densities were then normalized, and the RSC volume was subtracted from the RSC–nucleosome volume. The only substantial feature in the difference map was a large central density (several small, disconnected peripheral densities representing minor changes in the position or surface appearance of poorly ordered domains were removed from subsequent analysis by flood filling the main density of the difference map using the Situs software package 38 ). The structure of the nucleosome 39 , rendered at a resolution comparable to the resolution of the RSC–nucleosome model was docked into the difference volume using the program COAN 40 .

Supplementary Material pdf Note: Supplementary information is available on the Nature Structural & Molecular Biology website.