Molecular Architecture of the ATP-Dependent Chromatin-Remodeling Complex SWR1

1Inside the nucleus, eukaryotic DNA condenses into chromatin by associating with evolutionarily conserved histone proteins H2A, H2B, H3, and H4. About 150 DNA base pairs wrap around a histone octamer, which comprises one (H3/H4)2 tetramer and two H2A/H2B dimers, to form the nucleosome. Essential nuclear activities are regulated by processes that target the nucleosome. These processes are best characterized at gene promoters, where the biophysical properties, position, and composition of nucleosomes are strictly regulated. This results in a stereotypical chromatin structure that includes a histone-depleted region flanked by labile yet well-positioned nucleosomes containing the evolutionarily conserved histone variant H2A.Z. ATP-dependent chromatin remodeling complexes (remodelers) play a significant role in the regulation of promoter chromatin.

2Remodelers are conserved multisubunit complexes that can directly alter nucleosomal position and composition. All remodelers contain an ATPase domain—a member of the superfamily 2 (SF2) of translocases—within their core subunits. They also harbor domains located in cis to the ATPase that can regulate its ATPase activity and bind accessory subunits and/or histone modifications. These auxiliary domains are the basis for the common classification of remodelers into four subfamilies: SWI/SNF, ISWI, CHD, and INO80. Many remodelers collaborate at gene promoters to regulate transcriptional competency. Complexes of the SWI/SNF and ISWI subfamilies establish a nucleosome-depleted region around the promoter, thus exposing it to the transcriptional machinery. SWR1, a member of the INO80 subfamily, is targeted to this region to deposit H2A.Z at flanking nucleosomes. H2A.Z has been shown to affect the stability of its host nucleosome, higher-order chromatin folding, and recruitment of transcriptional factors.

3Most remodelers can reposition the nucleosome by “sliding” the histone octamer along the DNA. Although this activity depends on ATP-dependent DNA translocation by the core ATPase, remodelers function as multisubunit complexes. This highlights the importance of understanding how functional components assemble together into a remodeling machine and how this machine engages the nucleosome substrate. Structural approaches aimed at answering these questions are limited in number and resolution due to the complex compositions (2–15 subunits) and relatively large sizes (200–1,400 kDa) of remodelers. Nevertheless, they have provided some significant mechanistic insights. For example, a three-dimensional electron microscopy (3D EM) structure of the RSC complex showed it enveloping the nucleosome within a central cavity. The structure of nucleosome-bound ACF revealed that two remodelers bind to one nucleosome, which may underlie its ability to “measure” linker DNA and generate arrays of evenly spaced nucleosomes. In contrast to these “sliders,” SWR1 has evolved a mechanism for dimer exchange, which involves ejecting a resident H2A/H2B dimer from the substrate nucleosome and inserting a H2A.Z/H2B dimer in its place (Figure 1B).

4SWR1 functions as an ∼1 MDa complex containing 14 different polypeptides. Detailed dissection of its composition revealed three multisubunit modules that assemble on the core, catalytic subunit Swr1 (Figure 1A). The N-terminal half of the Swr1 polypeptide contains the helicase-SANT-associated (HSA) domain, which interacts with nuclear actin-related proteins (Arps). The Bdf1-Arp4-Act1-Swc4-Yaf9-Swc7 module—referred to as the N-Module here—is recruited to this region. Arp4 has been shown to interact directly with canonical nucleosomes and histones, Swc4 contains a SANT domain implicated in binding unmodified histone tails, and Bdf1 contains tandem bromodomains with affinity for acetylated histone tails. Therefore, the N-Module is likely involved in the targeting and binding of SWR1 to hyperacetylated nucleosomes. The core ATPase domain resides in the C-terminal half of Swr1 and harbors within it a long insert domain that is characteristic of the INO80, or “split-ATPase,” subfamily. This insert facilitates association of the two remaining modules, Swc3-Swc2-Arp6-Swc6 (termed here the C-Module) and the putative hexameric helicases Rvb1 and Rvb2, the latter module being another distinguishing characteristic of the INO80 subfamily. The C-Module binds the H2A.Z/H2B dimer, which is to be incorporated into the nucleosome, via the Swc2 subunit. Finally, Rvb1 and Rvb2 are AAA+ ATPases that can form homohexameric, heterohexameric, or dodecameric ring structures in isolation. Their oligomeric state in large complexes is not known structurally, although it has been suggested that they exist as a dodecamer, or double rings, in the SWR1-related INO80 complex.

5To address how functional modules of a dimer exchanger assemble as a complex, we have undertaken a multipronged approach to characterize the molecular architecture of SWR1 from Saccharomyces cerevisiae. Using EM, we have determined the 3D structure of SWR1 and mapped the locations of all functional modules. Our results show that the substrate-handling N- and C-Modules are arranged side-by-side and bracketed by the Swr1 ATPase and a single hexameric Rvb1/Rvb2 ring. Furthermore, neighboring relationships within SWR1, determined by chemical crosslinking and mass spectrometric (CX-MS) analysis, show that its components are highly interconnected. Finally, our reconstruction of a SWR1-nucleosome cocomplex reveals a large conformational change in the enzyme, which only forms limited contacts with the nucleosome substrate. Our data provide a structural framework for understanding SWR1’s unique dimer-exchange activity.

6We used 3D EM to determine the structure of SWR1 obtained from S. cerevisiae. Although the affinity-purified sample appeared biochemically pure, with all 14 SWR1 components and no visible contaminants (Figure 1C), it was structurally heterogeneous when observed under the electron microscope (Figure S1B available online). To overcome this, we adapted the GraFix (Gradient with Fixation) technique, which uses a combined glycerol and crosslinker gradient (Figure S1A) to both stabilize and purify macromolecular assemblies. We chose formaldehyde as the crosslinker because reversibility of the crosslinks would allow us to verify where in the gradient all SWR1 components were present (Figure S1A). We then imaged individual fractions to identify the most homogeneous sample for data collection (Figure 1D).

7We obtained initial, low-resolution models of SWR1 using the Orthogonal Tilt Reconstruction (OTR) approach and negatively stained samples (Figures S1C and S1D). To refine the models, we used cryo-negative stain (cryo-NS) data, which benefit from the high contrast provided by the heavy-atom stain and the structural integrity of frozen-hydrated samples. We performed projection-matching refinement, first against reference-free class averages and then against single particles (Figure S2A). The resulting 3D structure (Figure 1E; Movie S1) has an estimated resolution of 28 Å (0.5 Fourier Shell Correlation [FSC]) (Figure S2B). Two-dimensional (2D) projections calculated from this structure show a good match to reference-free class averages obtained from the cryo-NS data (Figure S2D).

8Our 3D map of SWR1 shows a clear ring-shaped density with hexameric features (Figure 1E, left, and Figure 2), which fits well the crystal structure of a hexamer of RuvBL1, the human ortholog of Rvb1 (Figures 2D–2H). The orientation in our docking indicates that the Rvb insert domains (Figure 2D) mediate the ring’s interaction with the core of the SWR1 complex. Docking the RuvBL1 in the opposite orientation, with the insert domains facing away from the core of SWR1, results in a much poorer fit (Figure S3). The flexible Rvb insert domains protrude from the EM map (Figures 2G and 2H), suggesting that they undergo a rearrangement in SWR1. We also attempted to dock a 13 Å cryo-EM structure of a Rvb1/Rvb2 dodecamer from S. cerevisiae into our structure, but only one of its two rings fits the SWR1 density well (Figures 2A–2C). We further validated our structural data by performing a quantitative analysis of the Swr1:Rvb1:Rvb2 stoichiometry, which we determined to be ∼1:3:3 in the affinity-purified sample (Figure 3) and GraFix-treated fractions (Figure S1A), consistent with the presence of a single hexameric ring in SWR1. This stoichiometry does not appear to be a feature unique to SWR1 because we obtained similar results for the related INO80 complex (Figure S4).

9Although both Rvb1 and Rvb2 copurify with Swr1, it was formally possible that two populations of SWR1 coexist, each containing a homohexameric ring of either Rvb1 or Rvb2. Using CX-MS, we identified a crosslink between Rvb1 and Rvb2 (Figure 5; Table S2). All homotypic (Rvb1-Rvb1 or Rvb2-Rvb2) crosslinks we identified corresponded to distances most compatible with intramolecular crosslinks (Table S2), whereas the Rvb1-Rvb2 crosslink agrees with the expected intersubunit interface based on a homology model of the S. cerevisiae Rvb1 and Rvb2 heterohexamer (Figures 5B and 5C). Thus, we conclude that SWR1 contains a single Rvb1/Rvb2 heterohexamer.

10To determine the locations of the N- and C-Modules in the SWR1 structure, we characterized stable subcomplexes containing Swr1 and Rvb1/Rvb2, and either the N- or C-Module (Figures S5A and S5C; Table S1). We named these ∼700 kDa subcomplexes SWR1-ΔC-Mod and SWR1-ΔN-Mod to indicate the missing module (Figures 4A and 4B).

11We aligned and classified cryo-NS images to obtain 2D class averages, or views, of the subcomplexes. A comparison of these views with those from the full complex and 2D reprojections of the SWR1 structure led to two major observations. First, in the absence of either ∼300 kDa module, the remaining SWR1 components assembled into a structure very similar to that of the corresponding portion of the full complex (Figures 4C, S4B, and S4D). Notably, an ∼400 amino acid truncation in Swr1 did not significantly impact the assembly of the remaining subunits in SWR1-ΔN-Mod, suggesting that the N- and C-terminal halves of Swr1 fold independently of each other. Second, two major features were retained in both subcomplexes: the Rvb1/Rvb2 ring, and a prominent density distal to it (Figures 4C, S4B, and S4D). Because the subcomplexes only share Rvb1, Rvb2, and the catalytic bulk of Swr1, we conclude that the latter occupies the large density distal to the Rvb1/Rvb2 ring (Figure 4D).

12Next, we performed difference mapping between class averages of each subcomplex and those of full SWR1 for three different views of (Figure 4C). This analysis identified the locations of the C- and N-Modules. They form structurally discrete entities arranged side-by-side and bracketed by the Rvb1/Rvb2 ring and Swr1 (Figures 4C and 4D). These results allowed us to generate a low-resolution annotation of the 3D map (Figure 4D). Further support for this annotated map was provided by the general agreement of the theoretical molecular weights of the modules with those calculated from the corresponding densities in the EM map (Figure S5E).

13We used isotopic CX-MS to determine the spatial organization of SWR1 subunits in more detail. Our results confirmed a number of the previously determined interactions (Figure 5; Table S2). Specifically, we observed crosslinks connecting the N-Module with the N-terminal half of Swr1 (Swc4-Swr1), the C-Module with the C-terminal half of Swr1 (Swc3-Swr1), and Arp4 with Swc4 within the N-Module (Figure 5A; Table S2). Additionally, we obtained crosslinks connecting the small subunit Swc5 with both the N- and C-terminal halves of Swr1, in agreement with data showing that Swc5 requires the full SWR1 polypeptide to be present in the complex.

14We also observed additional crosslinks that suggest a high degree of interconnectedness among SWR1’s functional modules. The Rvb1/Rvb2 ring, previously shown to require the long insert in the ATPase domain of Swr1 to assemble into the full complex, crosslinked to the N-terminal half of Swr1 (its HSA domain) via Rvb1, the N-Module (Rvb1-Arp4), and the C-Module (Rvb2-Swc2) (Figure 5; Table S2). The N- and C-Modules also crosslinked to each other, through Bdf1-Swc2 and Arp4-Swc3 (Figure 5A; Table S2). Finally, Swc5 crosslinked to the N-Module via two interactions with Act1 and Yaf9 (Figure 5A; Table S2). Understanding the functional significance of these novel spatial relationships will require future work combining biochemistry and finer subunit mapping in SWR1.

15Our 3D reconstruction of SWR1 does not show a central cavity that could accommodate a nucleosome, as has been observed for the RSC complex. To explore the possibility that SWR1 interacts with its substrate nucleosome in a different manner, we obtained the structure of SWR1 bound to a nucleosome. We used a nucleosome with a single 43 bp linker based on a characterization of the effect of linker length on nucleosome binding by SWR1. Preliminary gel-shift studies showed similar binding of SWR1 to the nucleosome in the presence or absence of ATP. Therefore, we carried out an in vitro nucleosome-binding reaction in the absence of nucleotide (Figure S6A) and purified the sample using GraFix (Figure S6B). Western blotting confirmed the cosedimentation of histones with SWR1 in the glycerol gradient (Figure S6C), and the sample exhibited retarded electrophoretic mobility on a native gel relative to apo-SWR1 that had been similarly purified (Figure 6A). This suggested that a majority of the SWR1-nucleosome sample contained nucleosome-bound complexes. We imaged this sample under cryo-NS conditions.

16To obtain the 3D reconstruction of the SWR1-nucleosome complex, we first refined our apo-SWR1 model, obtained by refining the OTR model against 2D class averages of apo-SWR1, against 2D class averages generated from the SWR1-nucleosome data (Figure S7A). Using the resulting 3D map as a starting model, we performed maximum likelihood-based 3D classification of the entire single-particle SWR1-nucleosome data set. We obtained five classes that displayed overall structural similarity (Figure S7B) and proceeded to further refine the one in which all modules could be most easily identified. The resulting 3D structure had a resolution of 34 Å (Figures 6C and 6E).

17The SWR1-nucleosome structure is elongated relative to apo-SWR1 along an axis perpendicular to the Rvb1/Rvb2 ring (Figures 6B–6E). This elongation appears to be the result of an extension, away from the Rvb1/Rvb2 ring, of Swr1 and the C-Module (Figures 6B and 6D). Difference maps calculated between the apo-SWR1 and SWR1-nucleosome structures support this conformational change (Figures 6F and 6G), as does a comparison between reprojections of the apo-SWR1 3D map with reference-free class averages of the SWR1-nucleosome data (Figures 6I and S7D). At the reported resolution, we did not observe significant changes in the Rvb1/Rvb2 ring; Swr1 and the C-Module are the major densities that differ in positions between the two structures (Figures 6F and 6G). We note that the sample used in this analysis was purified and imaged under identical conditions to nucleosome-free SWR1. Therefore, experimental and computational variations are unlikely to have contributed to the observed conformational difference.

18The SWR1-nucleosome reconstruction showed a new density that is contiguous with that of the core subunit Swr1 and extends toward the Rvb1/Rvb2 ring (Figure 6C and 6E). This protrusion from Swr1 is also observed in 2D class averages (Figures 6I and S7D) and coincides with a peak in the difference map calculated by subtracting apo-SWR1 from SWR1-nucleosome (Figures 6G and 6H). We docked a 3D map generated from the yeast nucleosome crystal structure, filtered to the resolution of the SWR1-nucleosome map (34 Å), into the EM density. The extra density in our 3D map could accommodate the bulk of the nucleosome (Figures 6J and S7C). However, this density was not fully resolved, likely due to heterogeneity, both conformational and biochemical, in the data. The 3D map also indicated that the ATPase-containing portion of Swr1 mediates the most significant contact between SWR1 and the nucleosome core particle (Figures 6K and S7C).

19To confirm the orientation of the bound nucleosome suggested by our data, we generated a model by computationally adding the density of a nucleosome into its putative density in our SWR1-nucleosome reconstruction (Figure 6J). We then compared reprojections from this composite model against experimental 2D class averages. This analysis indicated that the location and orientation of the modeled nucleosome were in general agreement with the experimental data (Figures 6I and S7D). In this orientation, the nucleosome appears to bind over a central depression formed between the Swr1 ATPase and the Rvb1/Rvb2 ring. One side of the octamer faces the complex, whereas the other side is completely exposed (Figure S7C).

20Our study revealed the structure of SWR1 to be composed of structurally discrete domains, with the substrate-binding N- and C-Modules arranged side-by-side and bracketed by the catalytic Swr1 ATPase and the Rvb1/Rvb2 ring (Figure 4D). We mapped the ATPase-containing bulk of the core subunit Swr1 to a location distal to the ring, where its position appears to be supported mainly by the N-Module (Figure 4D). Consistent with this arrangement, the SWR1-ΔN-Mod subcomplex exhibited more widespread structural changes than SWR1-ΔC-Mod, with changes in the Rvb1/Rvb2 ring detected in the 2D difference maps (Figure 4C). Our CX-MS data indicate that Arp4 mediates the N-module’s interaction with the Rvb1/Rvb2 ring (Figures 5 and 7A). ATP binding by Arp4 has been shown to regulate its association with macromolecular complexes in vivo and may play a regulatory role in the assembly of SWR1.

21Our structural analysis of the subcomplexes provided direct visualization of the independent association of the N- and C-Modules with SWR1, as suggested by previous biochemical studies. In the absence of each ∼300 kDa module, the remaining components still associated in a manner similar to that in the full complex (Figures 4, S5B, and S5D). Therefore, SWR1 may assemble in a modular fashion that involves stable, preformed functional subcomplexes, as suggested for INO80 and the histone acetyltransferase SAGA. We speculate that modular assembly allows for efficient sharing of stable modules among functionally related complexes, thus regulating their recruitment and collective activity. This phenomenon may occur in SWR1, INO80, and the histone acetyltransferase NuA4—three complexes that are recruited to promoters and converge functionally at H2A.Z and compositionally at the Arp4-Act1 dimer.

22We determined structurally and biochemically that Rvb1 and Rvb2 associate with the SWR1 complex as a single heterohexameric ring (Figure 2) and established the orientation of the ring relative to the rest of the complex (Figures 2D, 2H, and S3). The Rvb inserts, known to be involved in ring-ring interactions in dodecameric structures, face the core of SWR1, making them unavailable to interact with a second ring. Our biochemical quantitation of the Rvb1:Rvb2:Swr1 stoichiometry is in agreement with the 3:3:1 ratios indicated by the structural and proteomic data (Figure 3). Although ∼6:6:1 ratios have previously been reported for the highly related INO80 complex, suggesting the presence of two hexameric rings, our quantitation for INO80 also resulted in ∼3:3:1 stoichiometry (Figure S4). Therefore, we conclude that remodelers in the INO80 subfamily are characterized by the presence of a single hexameric Rvb1/Rvb2 ring.

23Our results modify the previous assembly map of SWR1 (Figure 1A), which placed Swr1 at the center of the complex, bringing together individual modules via separate domains. We now show that Swr1 adopts a peripheral position in the complex, with the catalytic ATPase spatially separated from the Rvb1/Rvb2 ring (Figure 4D). Their interaction across the complex may be supported by the long insert within the Swr1 ATPase. We found that the N-terminal HSA domain in Swr1 and the HSA-associated Arp4 subunit directly crosslinked to Rvb1 (Figures 5A and 5B). This indicates that the N-terminal half of Swr1 extends across the complex and may mediate interactions between the N-Module and the Rvb1/Rvb2 ring. We propose that the hexameric ring plays an important structural role as an assembly platform for the independently assembled, substrate-interacting N- and C-Modules. Their side-by-side arrangement may in turn be important for proper spatial coordination of the nucleosome and the H2A.Z/H2B dimer during dimer exchange.

24Our data provide a structural view of a dimer exchanger as a compact and interconnected assembly of discrete functional modules. In such assembly, potential nucleotide- or substrate-dependent conformational changes in the catalytic ATPase and/or the hexameric ring could efficiently propagate and thus mediate global structural dynamics involved in dimer exchange.

25Conformational changes have been reported for nucleosome-bound remodelers; however, they have yet to be visualized. Our 3D structure of a SWR1-nucleosome cocomplex, together with the annotated 3D map of SWR1 alone, allowed us to characterize the extension of Swr1 and the C-Module, away from the Rvb1/Rvb2 ring, that occurs upon substrate binding (Figure 6). This extension was observed in the absence of nucleotides, suggesting that recognition of nucleosomal features, such as linker DNA, nucleosomal DNA, and/or unmodified histone tails, by various components of SWR1 is sufficient to promote this conformational change.

26We showed that SWR1 makes only limited contact with the nucleosome, mediated primarily by the Swr1 ATPase, in contrast to other large remodelers that interact extensively with their substrate. The RSC complex, for example, envelops the nucleosome within a central cavity, and the SWI/SNF complex contacts ∼50 bp, or nearly one gyre, of nucleosomal DNA. We speculate that relatively extensive substrate interactions may be a feature of remodelers that slide the histone octamer. Specifically, as nucleosomal DNA is unraveled from the octamer, these interactions may serve to prevent substrate disassembly. SWR1, however, is the only known remodeler that does not slide the octamer (A.R. and C.W, unpublished data) and shows very limited ATP-dependent mobilization of nucleosomal DNA. Therefore, chromatin remodeling by SWR1 may not require large-scale disruptions of histone-DNA contacts, and its limited interaction with the nucleosome may reflect this mechanistic distinction.

27Our results do not address contacts that the complex may make with linker DNA, as observed for SWI/SNF, ISWIs, and suggested for the SWR1-related INO80 complex. Unmodified histone tails may also interact with components of the N-Module, such as Swc4 and Arp4. All of these potential interactions, which would occur outside of the nucleosome core particle, could stabilize the substrate-bound conformation revealed by our 3D map. Furthermore, because the 3D map of nucleosome-bound SWR1 was obtained in the absence of nucleotides, it remains to be seen how ATP binding and/or hydrolysis by Swr1 and perhaps Rvb1/Rvb2 may further affect the overall structure of the substrate-bound complex. It has been shown that, whereas the ATPase activity of SWR1 is enhanced upon binding to a canonical nucleosome, which we used in our study, it achieves the highest level of stimulation when the second substrate, the H2A.Z/H2B dimer, is also bound. Structural characterization of this fully liganded and highly activated cocomplex should provide important new insights into SWR1 function.

28In conclusion, our study presents the 3D structure of SWR1, revealing an interconnected assembly of discrete functional modules. This is also among the first structural characterizations of the functionally diverse AAA+ proteins Rvb1 and Rvb2 in the context of a larger complex. In SWR1, they form a single heterohexameric ring and serve as an assembly platform that connects all functional modules within the complex. Finally, we showed that SWR1 undergoes a large conformational change upon nucleosome binding (Figure 7B). The limited interaction between the complex and the nucleosome is distinct from other remodelers and may reflect unique mechanistic aspects of the dimer-exchange reaction.

29SWR1 was affinity purified from S. cerevisiae as described by (see Table S1 for strain information). Nucleosome-bound SWR1 was obtained by incubating 40 pmol recombinant nucleosomes with 10 pmol SWR1 in a 120 μl reaction for 30 min at room temperature. Apo- and nucleosome-bound SWR1 samples were further purified using the GraFix method, substituting glutaraldehyde with formaldehyde.

30For initial model reconstruction, SWR1 was applied to glow-discharged holey-carbon grids coated with a continuous-carbon support for ∼30 min at 4°C. The sample was stained with a 2% uranyl formate solution and dried under N2 gas. For refinement, apo-SWR1 and nucleosome-bound samples were absorbed onto glow-discharged Cu 200-mesh Quantifoil grids with a continuous-carbon support. Staining was carried out as described above, but stained samples were frozen in liquid N2 immediately after blotting, producing cryo-NS samples.

31Images were collected at +45° and −45° using a Tecnai G2 Spirit microscope (FEI, Hillsboro) operating at 120 keV and, equipped with a US4000 4 × 4k CCD camera (Gatan, Pleasanton) at a nominal magnification of 68,000×. The pixel size at the sample level was 1.65 Å.

32Cryo-NS data were collected at liquid nitrogen temperature using a field-emission gun (FEG) Tecnai F20 transmission electron microscope (FEI) operating at 120 keV and equipped with a Gatan 4 × 4k CCD. Images were collected at a nominal magnification of 62,000× and an electron dose of ∼20 electrons/Å2. The pixel size at the sample level was 1.73 Å.

33Single particles were manually selected from micrographs using the Boxer interface in EMAN1. We identified tilt mates using a custom-built SPIDER script. Tilted particles were CTF corrected using the CTFTILT program. Untilted, cryo-NS particles were phase flipped in EMAN2. Reference-free alignment and classification were carried out in IMAGIC. Initial models were obtained using the OTR method.

34We performed initial projection-matching refinement of the OTR models using class averages obtained from cryo-NS apo-SWR1 data. Refinement was carried out in SPIDER using the AP SH and BP 32F functions. The resulting 3D maps were further refined against single cryo-NS particles of apo-SWR1 to obtain the final SWR1 reconstruction, and against reference-free class averages generated from cryo-NS data of SWR1-nucleosome. The latter 3D map was used as an initial model for 3D classification in the RELION program. Five 3D classes were generated, and one was further refined against single particles assigned only to that class using the “Autorefine” function in RELION.

35We crosslinked ∼45 μg of sample with 1 mM disuccinimidyl suberate d0/d12 (DSS; Creative Molecules) at 37°C for 30 min and subsequently quenched the reaction by adding ammonium bicarbonate. Trypsin digestion was carried out at 37°C overnight. We enriched for crosslinked peptides using size exclusion chromatography, followed by LC-MS/MS analysis on an Orbitrap Elite mass spectrometer (Thermo Scientific, San Jose). Data analysis was performed using XQuest.

36SWR1 was affinity-purified from S. cerevisiae as previously described. We prepared a glycerol-formaldehyde gradient in a 4 ml ultracentrifuge tube (Beckman Ultra-Clear™) by layering 2 ml of “SWR1 buffer” (25 mM HEPES-KOH, pH 7.6, 1 mM EDTA, 2 mM MgCl2, 0.01% NP-40, 1 mM DTT, 100 mM KCl) containing 10% glycerol and 0.2% formaldehyde over 2 ml of SWR1 buffer containing 60% glycerol and 1.5% formaldehyde. The gradient was formed using a GradientMaster instrument (Model 107ip, BioComp). After applying 4-10 pmoles of SWR1 to the top of the gradient, we centrifuged the sample in a SW 60 Ti rotor (Beckman) at 35,000 rpm and 4°C for 20 hr. We then manually fractionated the gradient from bottom to top into 100 μl fractions and quenched the formaldehyde in each fraction using 80 mM glycine. We stored the fractions at 4°C.To determine the best fraction to image, we reversed the crosslinks for 15 μl of each fraction in 0.3 M Tris, 0.1% SDS, and β-mercaptoethanol for 12 hr at 65°C and 30 min at 95°C in a thermocycler. We screened for fractions containing the full complement of 14 subunits by running the treated samples in a 10% SDS-PAGE gel and silver-staining it. Having identified candidate fractions after crosslink reversal and SDS-PAGE analysis, we proceeded to dialyze the original crosslinked samples in those fractions against SWR1 buffer containing no glycerol. The dialyzed samples were stored at 4°C, ready to be stained and imaged.We prepared mono-nucleosomes containing recombinant fly H3-H4, yeast H2A-H2B and 190 bp DNA bearing the 601 nucleosome-positioning sequence and a 43 bp asymmetric linker. To obtain nucleosome-bound SWR1, we carried out an in vitro binding reaction with 40 pmoles reconstituted nucleosomes and 10 pmoles SWR1 in a 120 μl reaction for 30 min at room temperature. We then proceeded to purify SWR1-Nucl as described above for SWR1.To qualitatively assess nucleosome occupancy in the GraFix-treated sample, we compared the electrophoretic mobility of this sample against that of apo-SWR1, which was similarly purified. After dialysis, crosslinked samples from single fractions were electrophoresed in a NuPAGE Novex 3%–8% Tris-Acetate, 1.0 mm gel (Invitrogen) at 4°C. Sample loading buffer contained 10% glycerol, 0.01% w.v. bromophenol blue, 43 mM imidazole, 35 mM HEPES, pH 7.4. Electrophoresis buffer contained 43 mM imidazole and 35 mM HEPES, pH 7.4. The gel was silver-stained.To confirm cosedimentation of nucleosomes and SWR1 in the glycerol gradient, we quenched and reversed the formaldehyde crosslinking as described above. Then, an aliquot of each fraction was electrophoresed in a precast PROTEAN 4%–20% polyacrylamide gel (Bio-Rad). Proteins were transferred onto a nitrocellulose membrane and blotted for,H3 using rabbit α-H3 (AB1791) at 1:3,000 dilution and GαRb 2° at 1:5,000 dilutionSwr1-3xFLAG using mouse α-FLAG (Bio-Rad) at 1:3,000 dilution and rabbit anti-mouse 2° at the same dilutionSwc2 using chicken α-Swc2 at 1:10,000 dilution and α-chicken HRP at 1:5,000 dilutionMembranes were exposed to Kodak chemiluminescence films.We obtained the initial models of SWR1 using negatively stained samples. We applied 5-10 μl of the dialyzed sample to homemade holey grids coated with a thin layer of carbon, let the sample absorb for 15-30 min at 4°C, stained with a 2% uranyl formate solution and then floated a second layer of thin carbon (“sandwich”) before drying the grid under a gentle flow of N2 gas.The initial models were refined against cryo-negative stain (cryo-NS) data. To prepare cryo-NS samples, we stained as described above but, instead of drying, we froze the stained grids in liquid nitrogen.To build the initial 3D models for SWR1, we collected images under low-dose conditions at +45° and −45° in a Tecnai G2 Spirit microscope (FEI, Hillsboro, OR), operating at 120 keV and equipped with a US4000 4k × 4k CCD camera (Gatan, Inc, Pleasanton, CA) at a nominal magnification of 68,000 and with a dose of ∼20 electrons/Å2. The pixel size at the sample level was 1.65Å.To obtain apo- and nucleosome-bound SWR1 data for refinement, we collected untilted cryo-NS data at liquid-nitrogen temperature and under low-dose conditions. We used a field-emission gun (FEG) Tecnai G2 F20 transmission electron microscope (FEI) operating at 120 keV and equipped with a Gatan 4k × 4k CCD. Images were collected at a nominal magnification of 62,000x and an electron dose of ∼20 electrons/Å2. The pixel size at the sample level was 1.73 Å.In order to extract single particles from the micrographs we windowed out particles in one set of micrographs using the Boxer interface in EMAN1 and used custom-built SPIDER scripts to calculate alignment parameters between the +45° and −45° micrographs and extract the tilt mates in the other set. We estimated and corrected for the Contrast Transfer Function (CTF) using the program CTFTILT and the SPIDER command TF CT. Single particles were binned by 2, resulting in a pixel size of 3.3 Å. We combined the +45° and −45° data sets into a stack of ∼32,000 particles and performed reference-free 2D alignment and classification in IMAGIC. We computed initial models from classes containing 100–200 members using the Orthogonal Tilt Reconstruction approach as described.We initially refined the OTR models against class averages of cryo-NS data. To generate the class averages, we extracted particles from the micrographs as described above and performed CTF estimation and phase flipping using the EMAN2 workflow. Then, the particles were binned by 2, resulting in a pixel size of 3.45 Å. We subjected ∼32,000 particles to reference-free 2D alignment and classification in IMAGIC. In order to minimize heterogeneity, we generated classes with relatively few (15-20) particles.We filtered the OTR models to 80 Å resolution and performed 15-23 iterations of projection matching refinement using angular steps of 25°, 20°, 15°, 10°, and 8°-5° (every degree) against 2D class averages in SPIDER using the AP SH and BP 32F commands. To minimize build up of noise in the reconstructions, we applied a threshold mask calculated for 500% to 150% the theoretical molecular weight of the sample (1.0 MDa). The mask was gradually tightened during refinement and its filtration was determined by the resolution of the 3D map, computed according to the 0.5 FSC criterion. Refinement results were generally stable after 15 iterations, and the resolutions of the 3D maps were 50-60 Å. We then refined the resulting 3D maps (without additional filtration) against single cryo-NS particles. For this step, we carried out 15 iterations of projection-matching refinement at angular steps 25°, 21°, 18°, 15°, 13°, 11° and 10°–4° (every degree). Threshold masks computed for 500% to 100% the MW were also utilized as described above.To compare 3D maps against experimental data, we generated reprojections of the filtered maps at defined theta values using the PJ 3Q command in SPIDER. The reprojections and experimental class averages were aligned to each other using the AP SH and RT SQ commands in SPIDER.We performed 3D structure analysis and image rendering using the UCSF Chimera package. EM-like 3D maps were generated from published crystal coordinates using the CP FROM PDB command in SPIDER. To dock 3D maps or crystal structures into our EM densities, we roughly placed the former into the latter and used Chimera’s “Fit in Map” function for the final fitting.To generate the composite nucleosome-bound SWR1 map, we generated a 3D map of the yeast nucleosome from the published crystal coordinates (PDB: 1IDB,) as described above. Then, we placed this map into our experimental map and converted the former’s 3D coordinate system to the latter via resampling (vop resample command in Chimera). The two maps were then normalized, added, and filtered in SPIDER.We segmented the SWR1 3D map using the Eraser tool in Chimera. We used the results from our module mapping (Figure 4) to guide the segmentation. Once we had obtained a segmented module using the eraser, we calculated the molecular weight of the enclosed volume by obtaining the number of voxels in Chimera.We performed 2D difference mapping between class averages by first aligning, rotating, and shifting them in IMAGIC using the MSA-ALIGN command. Then, we normalized and subtracted the aligned images in SPIDER.The difference maps were normalized (mean = 0, standard deviation = 1). Threshold masks were calculated for each of them, setting values below 2σ to 0 and these masks were then applied to the normalized difference maps. The resulting images were colored in Photoshop by turning them into RGBs and converting the grayscale values to a rainbow gradient. The gradient begins at 2σ (as determined by the thresholding) and ends at around 6σ, based on the maximum pixel values present in each of the difference maps. These maxima changed slightly between difference maps, ranging from ∼6σ to ∼7σ. We used a single color scale for the figure for simplicity, even though it ignores small differences in the heights of the peaks among the images shown.To obtain the 3D structure of nucleosome-bound SWR1, we used the 3D map obtained after refinement of the OTR model against apo-SWR1 class averages. This model was low-pass filtered to 60 Å and refined against 2D class averages generated from cryo-NS data of the nucleosome-bound sample. The SWR1-nucleosome class averages were obtained as described above for apo-SWR1. Using the resulting 3D map as a starting model, we performed maximum-likelihood-based 3D classification using the RELION program. We generated 5 classes from ∼45,000 single particles. Single particles were phase-flipped in EMAN2, binned by 3 for a resulting pixel size of 5.17 Å, and normalized in XMIPP. After 3D classification, we performed single-model refinement by using the “Autorefine” option in RELION. Because a majority of the resulting 3D maps exhibited missing densities, we selected a model in which densities for all modules identified in the apo-SWR1 structure were accounted for. To minimize heterogeneity, we refined it against single particles assigned only to that class by 3D classification, which constituted ∼20% of the data set.The 3D homology model of S. cerevisiae Rvb1/2 heterohexamer was computed using the SWISS-MODEL interface. We used the crystal structure of the human Rvb1 homolog (2C9O,) as the template to align sequences for Saccharomyces cerevisiae Rvb1 (YDR190C) and Rvb2 (YPL235W). With the 3D homology models for the Rvb1 and Rvb2 monomers, we generated the hetero-hexameric Rvb1/2 model in PyMOL (The PyMOL Molecular Graphics System, Version 1.5.0.4 Schrödinger, LLC) by computationally aligning each monomer to the corresponding homolog in the crystal structure of the truncated human Rvb1/2 hexameric ring (2XSZ,). The composite final structure was saved as a PDB file. Subsequent analyses involving this homology model were carried out in UCSF Chimera.Purified SWR1 and INO80 complexes were resolved on 4%–12% Bis-Tris gel (Novex) with MOPS running buffer. Protein gels were stained with Coomassie or Sypro Orange dye and imaged on Fuji Image Quant LAS 3000. Intensity of relevant protein bands was measured using ImageQuant TL software (GE). The ratios of Swr1 or Ino80 to Rvb1 and Rvb2 were calculated after normalizing the band intensities for protein size.We approximated Rvbs:Swr1 stoichiometry in the GraFix-treated sample for each fraction. After crosslink reversal and SDS-PAGE, as described above, the gel was silver-stained and imaged on the GE (Amersham) Typhoon Trio Imager. Band intensities were measured using the ImageQuant TL software (GE) after background subtraction (rolling-ball method).We crosslinked roughly 45 μg of sample with 1 mM disuccinimidyl suberate d0/d12 (DSS, Creative Molecules Inc.) directly in SWR1 buffer at 37°C for 30 min and subsequently quenched the reaction by adding ammonium bicarbonate to a final concentration of 50 mM for 10 min at 37°C.We reduced the sample with 2.5 mM Tris (2-Carboxyethyl) phosphine hydrochloride (TCEP, Pierce) in 8 M urea at 37°C for 30 min and subsequently alkylated it with 5 mM iodoacetamide (Sigma-Aldrich) for 30 min at room temperature in the dark. For digestion, we diluted the sample with ammonium bicarbonate to a 1 M final concentration of urea and added 2% w/w trypsin (Promega). Digestion was carried out at 37°C over night and stopped by acidification to 2% (w/v) trifluoroacetic acid (TFA). We purified peptides with Sep-Pak C18 MicroSpin columns (Waters, Milford, MA), according to the manufacturer’s protocol, followed by enrichment of crosslinked peptides using size exclusion chromatography. We carried out LC-MS/MS analysis on an Orbitrap Elite mass spectrometer (Thermo Scientific, San Jose, CA).We searched the data using XQuest in iontag mode against a database containing the protein sequences of all 14 previously identified SWR1 proteins with a precursor mass tolerance of up to 20 ppm. For matching of fragment ions tolerances of 0.2 Da for common-ions and 0.3 Da for crosslink ions were used. We identified crosslinked peptides with a linear discriminant (ld) score > 25 and further analyzed them by visual inspection in order to ensure good matches of ion series on both crosslinked peptide chains for the most abundant peaks.

37SWR1 was affinity-purified from S. cerevisiae as previously described. We prepared a glycerol-formaldehyde gradient in a 4 ml ultracentrifuge tube (Beckman Ultra-Clear™) by layering 2 ml of “SWR1 buffer” (25 mM HEPES-KOH, pH 7.6, 1 mM EDTA, 2 mM MgCl2, 0.01% NP-40, 1 mM DTT, 100 mM KCl) containing 10% glycerol and 0.2% formaldehyde over 2 ml of SWR1 buffer containing 60% glycerol and 1.5% formaldehyde. The gradient was formed using a GradientMaster instrument (Model 107ip, BioComp). After applying 4-10 pmoles of SWR1 to the top of the gradient, we centrifuged the sample in a SW 60 Ti rotor (Beckman) at 35,000 rpm and 4°C for 20 hr. We then manually fractionated the gradient from bottom to top into 100 μl fractions and quenched the formaldehyde in each fraction using 80 mM glycine. We stored the fractions at 4°C.

38To determine the best fraction to image, we reversed the crosslinks for 15 μl of each fraction in 0.3 M Tris, 0.1% SDS, and β-mercaptoethanol for 12 hr at 65°C and 30 min at 95°C in a thermocycler. We screened for fractions containing the full complement of 14 subunits by running the treated samples in a 10% SDS-PAGE gel and silver-staining it. Having identified candidate fractions after crosslink reversal and SDS-PAGE analysis, we proceeded to dialyze the original crosslinked samples in those fractions against SWR1 buffer containing no glycerol. The dialyzed samples were stored at 4°C, ready to be stained and imaged.

39We prepared mono-nucleosomes containing recombinant fly H3-H4, yeast H2A-H2B and 190 bp DNA bearing the 601 nucleosome-positioning sequence and a 43 bp asymmetric linker. To obtain nucleosome-bound SWR1, we carried out an in vitro binding reaction with 40 pmoles reconstituted nucleosomes and 10 pmoles SWR1 in a 120 μl reaction for 30 min at room temperature. We then proceeded to purify SWR1-Nucl as described above for SWR1.

40To qualitatively assess nucleosome occupancy in the GraFix-treated sample, we compared the electrophoretic mobility of this sample against that of apo-SWR1, which was similarly purified. After dialysis, crosslinked samples from single fractions were electrophoresed in a NuPAGE Novex 3%–8% Tris-Acetate, 1.0 mm gel (Invitrogen) at 4°C. Sample loading buffer contained 10% glycerol, 0.01% w.v. bromophenol blue, 43 mM imidazole, 35 mM HEPES, pH 7.4. Electrophoresis buffer contained 43 mM imidazole and 35 mM HEPES, pH 7.4. The gel was silver-stained.

41To confirm cosedimentation of nucleosomes and SWR1 in the glycerol gradient, we quenched and reversed the formaldehyde crosslinking as described above. Then, an aliquot of each fraction was electrophoresed in a precast PROTEAN 4%–20% polyacrylamide gel (Bio-Rad). Proteins were transferred onto a nitrocellulose membrane and blotted for,H3 using rabbit α-H3 (AB1791) at 1:3,000 dilution and GαRb 2° at 1:5,000 dilutionSwr1-3xFLAG using mouse α-FLAG (Bio-Rad) at 1:3,000 dilution and rabbit anti-mouse 2° at the same dilutionSwc2 using chicken α-Swc2 at 1:10,000 dilution and α-chicken HRP at 1:5,000 dilution

42H3 using rabbit α-H3 (AB1791) at 1:3,000 dilution and GαRb 2° at 1:5,000 dilution

43Swr1-3xFLAG using mouse α-FLAG (Bio-Rad) at 1:3,000 dilution and rabbit anti-mouse 2° at the same dilution

44Swc2 using chicken α-Swc2 at 1:10,000 dilution and α-chicken HRP at 1:5,000 dilution

45Membranes were exposed to Kodak chemiluminescence films.

46We obtained the initial models of SWR1 using negatively stained samples. We applied 5-10 μl of the dialyzed sample to homemade holey grids coated with a thin layer of carbon, let the sample absorb for 15-30 min at 4°C, stained with a 2% uranyl formate solution and then floated a second layer of thin carbon (“sandwich”) before drying the grid under a gentle flow of N2 gas.

47The initial models were refined against cryo-negative stain (cryo-NS) data. To prepare cryo-NS samples, we stained as described above but, instead of drying, we froze the stained grids in liquid nitrogen.

48To build the initial 3D models for SWR1, we collected images under low-dose conditions at +45° and −45° in a Tecnai G2 Spirit microscope (FEI, Hillsboro, OR), operating at 120 keV and equipped with a US4000 4k × 4k CCD camera (Gatan, Inc, Pleasanton, CA) at a nominal magnification of 68,000 and with a dose of ∼20 electrons/Å2. The pixel size at the sample level was 1.65Å.

49To obtain apo- and nucleosome-bound SWR1 data for refinement, we collected untilted cryo-NS data at liquid-nitrogen temperature and under low-dose conditions. We used a field-emission gun (FEG) Tecnai G2 F20 transmission electron microscope (FEI) operating at 120 keV and equipped with a Gatan 4k × 4k CCD. Images were collected at a nominal magnification of 62,000x and an electron dose of ∼20 electrons/Å2. The pixel size at the sample level was 1.73 Å.

50In order to extract single particles from the micrographs we windowed out particles in one set of micrographs using the Boxer interface in EMAN1 and used custom-built SPIDER scripts to calculate alignment parameters between the +45° and −45° micrographs and extract the tilt mates in the other set. We estimated and corrected for the Contrast Transfer Function (CTF) using the program CTFTILT and the SPIDER command TF CT. Single particles were binned by 2, resulting in a pixel size of 3.3 Å. We combined the +45° and −45° data sets into a stack of ∼32,000 particles and performed reference-free 2D alignment and classification in IMAGIC. We computed initial models from classes containing 100–200 members using the Orthogonal Tilt Reconstruction approach as described.

51We initially refined the OTR models against class averages of cryo-NS data. To generate the class averages, we extracted particles from the micrographs as described above and performed CTF estimation and phase flipping using the EMAN2 workflow. Then, the particles were binned by 2, resulting in a pixel size of 3.45 Å. We subjected ∼32,000 particles to reference-free 2D alignment and classification in IMAGIC. In order to minimize heterogeneity, we generated classes with relatively few (15-20) particles.

52We filtered the OTR models to 80 Å resolution and performed 15-23 iterations of projection matching refinement using angular steps of 25°, 20°, 15°, 10°, and 8°-5° (every degree) against 2D class averages in SPIDER using the AP SH and BP 32F commands. To minimize build up of noise in the reconstructions, we applied a threshold mask calculated for 500% to 150% the theoretical molecular weight of the sample (1.0 MDa). The mask was gradually tightened during refinement and its filtration was determined by the resolution of the 3D map, computed according to the 0.5 FSC criterion. Refinement results were generally stable after 15 iterations, and the resolutions of the 3D maps were 50-60 Å. We then refined the resulting 3D maps (without additional filtration) against single cryo-NS particles. For this step, we carried out 15 iterations of projection-matching refinement at angular steps 25°, 21°, 18°, 15°, 13°, 11° and 10°–4° (every degree). Threshold masks computed for 500% to 100% the MW were also utilized as described above.

53To compare 3D maps against experimental data, we generated reprojections of the filtered maps at defined theta values using the PJ 3Q command in SPIDER. The reprojections and experimental class averages were aligned to each other using the AP SH and RT SQ commands in SPIDER.

54We performed 3D structure analysis and image rendering using the UCSF Chimera package. EM-like 3D maps were generated from published crystal coordinates using the CP FROM PDB command in SPIDER. To dock 3D maps or crystal structures into our EM densities, we roughly placed the former into the latter and used Chimera’s “Fit in Map” function for the final fitting.

55To generate the composite nucleosome-bound SWR1 map, we generated a 3D map of the yeast nucleosome from the published crystal coordinates (PDB: 1IDB,) as described above. Then, we placed this map into our experimental map and converted the former’s 3D coordinate system to the latter via resampling (vop resample command in Chimera). The two maps were then normalized, added, and filtered in SPIDER.

56We segmented the SWR1 3D map using the Eraser tool in Chimera. We used the results from our module mapping (Figure 4) to guide the segmentation. Once we had obtained a segmented module using the eraser, we calculated the molecular weight of the enclosed volume by obtaining the number of voxels in Chimera.

57We performed 2D difference mapping between class averages by first aligning, rotating, and shifting them in IMAGIC using the MSA-ALIGN command. Then, we normalized and subtracted the aligned images in SPIDER.

58The difference maps were normalized (mean = 0, standard deviation = 1). Threshold masks were calculated for each of them, setting values below 2σ to 0 and these masks were then applied to the normalized difference maps. The resulting images were colored in Photoshop by turning them into RGBs and converting the grayscale values to a rainbow gradient. The gradient begins at 2σ (as determined by the thresholding) and ends at around 6σ, based on the maximum pixel values present in each of the difference maps. These maxima changed slightly between difference maps, ranging from ∼6σ to ∼7σ. We used a single color scale for the figure for simplicity, even though it ignores small differences in the heights of the peaks among the images shown.

59To obtain the 3D structure of nucleosome-bound SWR1, we used the 3D map obtained after refinement of the OTR model against apo-SWR1 class averages. This model was low-pass filtered to 60 Å and refined against 2D class averages generated from cryo-NS data of the nucleosome-bound sample. The SWR1-nucleosome class averages were obtained as described above for apo-SWR1. Using the resulting 3D map as a starting model, we performed maximum-likelihood-based 3D classification using the RELION program. We generated 5 classes from ∼45,000 single particles. Single particles were phase-flipped in EMAN2, binned by 3 for a resulting pixel size of 5.17 Å, and normalized in XMIPP. After 3D classification, we performed single-model refinement by using the “Autorefine” option in RELION. Because a majority of the resulting 3D maps exhibited missing densities, we selected a model in which densities for all modules identified in the apo-SWR1 structure were accounted for. To minimize heterogeneity, we refined it against single particles assigned only to that class by 3D classification, which constituted ∼20% of the data set.

60The 3D homology model of S. cerevisiae Rvb1/2 heterohexamer was computed using the SWISS-MODEL interface. We used the crystal structure of the human Rvb1 homolog (2C9O,) as the template to align sequences for Saccharomyces cerevisiae Rvb1 (YDR190C) and Rvb2 (YPL235W). With the 3D homology models for the Rvb1 and Rvb2 monomers, we generated the hetero-hexameric Rvb1/2 model in PyMOL (The PyMOL Molecular Graphics System, Version 1.5.0.4 Schrödinger, LLC) by computationally aligning each monomer to the corresponding homolog in the crystal structure of the truncated human Rvb1/2 hexameric ring (2XSZ,). The composite final structure was saved as a PDB file. Subsequent analyses involving this homology model were carried out in UCSF Chimera.

61Purified SWR1 and INO80 complexes were resolved on 4%–12% Bis-Tris gel (Novex) with MOPS running buffer. Protein gels were stained with Coomassie or Sypro Orange dye and imaged on Fuji Image Quant LAS 3000. Intensity of relevant protein bands was measured using ImageQuant TL software (GE). The ratios of Swr1 or Ino80 to Rvb1 and Rvb2 were calculated after normalizing the band intensities for protein size.

62We approximated Rvbs:Swr1 stoichiometry in the GraFix-treated sample for each fraction. After crosslink reversal and SDS-PAGE, as described above, the gel was silver-stained and imaged on the GE (Amersham) Typhoon Trio Imager. Band intensities were measured using the ImageQuant TL software (GE) after background subtraction (rolling-ball method).

63We crosslinked roughly 45 μg of sample with 1 mM disuccinimidyl suberate d0/d12 (DSS, Creative Molecules Inc.) directly in SWR1 buffer at 37°C for 30 min and subsequently quenched the reaction by adding ammonium bicarbonate to a final concentration of 50 mM for 10 min at 37°C.

64We reduced the sample with 2.5 mM Tris (2-Carboxyethyl) phosphine hydrochloride (TCEP, Pierce) in 8 M urea at 37°C for 30 min and subsequently alkylated it with 5 mM iodoacetamide (Sigma-Aldrich) for 30 min at room temperature in the dark. For digestion, we diluted the sample with ammonium bicarbonate to a 1 M final concentration of urea and added 2% w/w trypsin (Promega). Digestion was carried out at 37°C over night and stopped by acidification to 2% (w/v) trifluoroacetic acid (TFA). We purified peptides with Sep-Pak C18 MicroSpin columns (Waters, Milford, MA), according to the manufacturer’s protocol, followed by enrichment of crosslinked peptides using size exclusion chromatography. We carried out LC-MS/MS analysis on an Orbitrap Elite mass spectrometer (Thermo Scientific, San Jose, CA).

65We searched the data using XQuest in iontag mode against a database containing the protein sequences of all 14 previously identified SWR1 proteins with a precursor mass tolerance of up to 20 ppm. For matching of fragment ions tolerances of 0.2 Da for common-ions and 0.3 Da for crosslink ions were used. We identified crosslinked peptides with a linear discriminant (ld) score > 25 and further analyzed them by visual inspection in order to ensure good matches of ion series on both crosslinked peptide chains for the most abundant peaks.