Eukaryotic DNA is packaged into nucleosome arrays, which are repositioned by chromatin remodeling complexes to control DNA accessibility. The Saccharomyces cerevisiae RSC (Remodeling the Structure of Chromatin) complex, a member of the SWI/SNF chromatin remodeler family, plays critical roles in genome maintenance, transcription, and DNA repair. Here, we report cryo-electron microscopy (cryo-EM) and crosslinking mass spectrometry (CLMS) studies of yeast RSC complex and show that RSC is composed of a rigid tripartite core and two flexible lobes. The core structure is scaffolded by an asymmetric Rsc8 dimer and built with the evolutionarily conserved subunits Sfh1, Rsc6, Rsc9 and Sth1. The flexible ATPase lobe, composed of helicase subunit Sth1, Arp7, Arp9 and Rtt102, is anchored to this core by the N-terminus of Sth1. Our cryo-EM analysis of RSC bound to a nucleosome core particle shows that in addition to the expected nucleosome-Sth1 interactions, RSC engages histones and nucleosomal DNA through one arm of the core structure, composed of the Rsc8 SWIRM domains, Sfh1 and Npl6. Our findings provide structural insights into the conserved assembly process for all members of the SWI/SNF family of remodelers, and illustrate how RSC selects, engages, and remodels nucleosomes.
Eukaryotes have four major families of chromatin remodelers: SWI/SNF, ISWI, CHD, and INO80 (Clapier and Cairns, 2009). Each of these remodelers plays distinct roles based on how they select and affect target nucleosomes. Together, these remodelers give rise to the distinct chromatin landscapes observed in eukaryotic cells and determine how genetic information is organized, replicated, transcribed, and repaired (Yen et al., 2012). In S. cerevisiae there are two members of the SWI/SNF family of chromatin remodelers: RSC and SWI/SNF (Côté et al., 1994; Cairns et al., 1996). SWI/SNF chromatin remodelers reposition nucleosomes by translocating DNA around the histone octamer, and in vitro assays have shown that they move nucleosomes to the ends of linear DNA fragments before evicting the histones from the DNA (Clapier et al., 2016). RSC is essential for yeast viability and is ten times more abundant than SWI/SNF (Cairns et al., 1996). In the context of transcription, RSC is responsible for maintaining nucleosome free regions (NFR), while SWI/SNF plays a role in remodeling nucleosomes during transcription initiation (Nagai et al., 2017; Krietenstein et al., 2016; Klein-Brill et al., 2019). Additionally, RSC is also involved in many transcription-independent processes, such as mitotic division, double stranded break repair, and telomere maintenance (Krietenstein et al., 2016; Erkina et al., 2010; Kuryan et al., 2012; Campsteijn et al., 2007; Shim et al., 2007; Ungar et al., 2009; Ng et al., 2002).
RSC has a molecular weight of ~1.1 MDa and is composed of 17 proteins, with two copies of Rsc8 and one copy of either Rsc1 or Rsc2 (Cairns et al., 1996; Cairns et al., 1999). Seven of the proteins (Arp7, Arp9, Rsc6, Rsc8, Sfh1, Sth1 and Rsc9) are conserved in all eukaryotic SWI/SNF chromatin remodeling complexes, with Sth1 containing the ATPase domain that is responsible for the remodeling activity (Cairns et al., 1996; Saha et al., 2002). Two additional subunits are conserved in yeast (Npl6 and Rtt102), and the rest are complex specific (Htl1, Ldb7, Rsc1/2, Rsc3, Rsc30, Rsc4 and Rsc58). Three subunits, Arp7, Arp9 and Rtt102, are shared between RSC and SWI/SNF, and the two Arps have been found to be important for efficient remodeling activity of RSC (Clapier et al., 2016). The SWI/SNF paralog of Rsc8, Swi3, has been shown to be important for complex assembly and to likely play a scaffolding role (Yang et al., 2007), while the paralog of Sfh1, Snf5, has been found to be important for remodeling activity (Sen et al., 2017). RSC contains six bromodomains (BrDs), two in RSC1/2 and Rsc4, and one in Rsc58 and Sth1. Additionally, Rsc1/2 also contain a BAH domain. The BrD and BAH domains of RSC have all been shown to interact with histones H3 (Chambers et al., 2013), with the BrDs favoring acetylated H3 tails (Zhang et al., 2010), and in vitro binding assays have shown that the affinity of RSC for acetylated H3 nucleosomes is indeed greater than for unmodified nucleosomes (Chatterjee et al., 2011). In vivo deletions or mutations in these domains have been shown to decrease cell survival (Cairns et al., 1999; Kasten et al., 2004). The fungal-specific subunits Npl6, Htl1, Ldb7, Rsc3 and Rsc30, have been proposed to form a structural module (Wilson et al., 2006), with Npl6 and Htl1 interacting with Rsc8, and Ldb7 with Sth1 and Arp9. Rsc3 and Rsc30, which dimerize with each other, share a very similar architecture (Angus-Hill et al., 2001), with an N-terminal Zinc cluster that has been shown to bind to the DNA sequence CGCG that preferentially occurs at gene promoters (Badis et al., 2008). Deletion of Rsc3 has been shown to weaken the ability of RSC to maintain NFR in vivo (Badis et al., 2008).
High-resolution structural studies of the SWI/SNF family of remodelers had until recently been limited to fragments, such as the Arp module (Arp7, Arp9 and some combination of Rtt102 and HSA helix of Snf2 [paralog to Sth1]) (Schubert et al., 2013), the SwiB domain of the human SMARCD (homolog of Rsc6), the WH domain of SMARCB (homolog of Sfh1) (Allen et al., 2015), the SWIRM domain of Swi3 (paralog of Rsc8) (Da et al., 2006), the RPT domain of SMARCB (homolog of Sfh1) (Sammak et al., 2018), the SWIRM-RPT complex of SMARCC-SMARCB (homologs of Rsc8 and Sfh1) (Yan et al., 2017), the SANT domain of SMARCC (homolog of Rsc8), and the ATPase of Snf2 (homolog of Sth1) on its own (Dürr et al., 2005; Xia et al., 2016) and in complex with a nucleosome with different ATP analogs (Liu et al., 2017; Li et al., 2019). Early structural studies of full SWI/SNF remodeling complexes by negative stain electron microscopy (EM) were limited by low resolution and/or reconstruction artifacts (Chaban et al., 2008; Dechassa et al., 2008; Asturias et al., 2002; Leschziner et al., 2007; Leschziner et al., 2005; Skiniotis et al., 2007; Smith et al., 2003), with only one study able to map the location of some subunits within the complex (for the yeast SWI/SNF complex using subunit deletion) (Zhang et al., 2018).
Here, we have used cryo-EM to determine the structure RSC. The core of the complex was resolved to 3 Å, which allowed us the de novo building of this entire region. The complex is scaffolded around a central Rsc8 dimer from which other evolutionarily conserved subunits assemble, leading to a model for the biogenesis of the complex that agrees with that previously proposed for human SWI-SNF complexes based on biochemical data (Mashtalir et al., 2018). We were also able to determine the structure of RSC bound to a nucleosome at 19 Å resolution, which, together with our structure of the core of RSC, and previous ones for the Arp module and nucleosome, allowed us to reveal how RSC engages nucleosomes.
We have used cryo-EM to determine the structure of the chromatin remodeler RSC from S. cerevisiae purified using the TAP-tag (on Sth1) method (Figure 1—figure supplement 1). We found that RSC is composed of five main lobes, three that form a relatively rigid core (head, body and arm) and two that are flexibly attached (leg and tail) (Figure 1A; Figure 1—figure supplement 2). Our negative stain analysis of the yeast SWI/SNF complex shows that, like RSC, it has a rigid core sharing some similar features, and a flexible leg that occupies similar overall positions (Figure 1—figure supplement 2). However, SWI/SNF lacks a tail lobe, indicating that while RSC and SWI/SNF share a conserved core architecture, RSC features additional regulatory domains (Figure 1—figure supplement 2). Our assignment of most of the tail region to the RSC-specific subunits Rsc3 and Rsc30 (see below) is in further agreement with this observation.
We determined the structure of the core to ~3.0 Å and mapped 14 proteins within this region: Rsc1/2, Rsc3, Rsc4, Rsc6, Rsc8 (two copies), Rsc9, Rsc30, Rsc58, Ldb7, Npl6, Htl1, Sfh1, and Sth1 (Figure 1B,C,F; Figure 1—figure supplements 3, 4, 5, 6 and 7). It was not possible to distinguish, based on our cryo-EM data, whether the structure contained Rsc1 or Rsc2. A region in our map could be mapped either to Rsc1 (776–809 and 892–924) or Rsc2 (787–820 and 849–881). These regions are highly similar between the two proteins (43% identical plus 22% similar) and the density we observe likely correspond to an average of these two proteins (Figure 1—figure supplement 6).
We confirm our model by mass spectrometry analysis of RSC chemically crosslinked using bis(sulfosuccinimidyl)suberate (BS3) (Figure 1—figure supplement 8). We identified 780 unique inter-molecular links between different subunits and 617 unique intra-molecular links within the same subunits. About 90% (151/168) of mappable crosslinks in our model of the RSC core structure are within 38 Å distance.
In order to shed light on the interaction of RSC with its substrate, we also obtained a 19 Å resolution map of RSC bound to a nucleosome core particle (NCP) modified with H3K4me3 and H3K(9/14/18)ac (Figure 1D,E,F; Figure 1—figure supplement 9). We used an acetylated nucleosome due to the higher affinity of RSC for nucleosomes containing acetylated H3 (Chatterjee et al., 2011). We were able to unambiguously fit our structure of the RSC core within this map, along with previously published models of the Arp module (Arp7, Arp9, Rtt102 and Sth1-HSA) and the NCP (Figure 1E) (Schubert et al., 2013; Li et al., 2019). We do not observe any density for the ATPase domain of Sth1; however, based on the structure of the Sth1 homolog Snf2 bound to the NCP, we expect the catalytic subunit to bind super-helical location-2 (SHL2) (Li et al., 2019).
Our data show two distinct modes of NCP binding by RSC, one where the whole tail is swiveled towards the nucleosome (termed swiveled), and one where only a small region emanating from the tail reaches towards the NCP (termed locked) (Figure 1D). Altogether, the core of RSC appears to bind the nucleosome in a well-defined orientation that in our docking-based model places the catalytic domain of Sth1 in a position that would allow it to interact with the SHL2. The fact that we observe nucleosome-bound RSC, even without a stable binding of the catalytic domain of Sth1 to the NCP, indicates that the contacts through the RSC core are sufficient for nucleosome engagement. It also makes our structure a likely intermediate in a pathway to full engagement of a nucleosome by RSC. We propose that the contacts made by the RSC core with the NCP contribute to the processivity of the nucleosome remodeling function of the catalytic domain of Sth1 (see below).
The RSC core is critically defined by subunits Rsc8, Rsc6, Rsc9, Rsc58, Sth1, and Sfh1. Five of these proteins (Rsc8, Rsc6, Rsc9, Sth1, and Sfh1) are evolutionarily conserved throughout the eukaryotic SWI/SNF family, and comprise 72% of the mass of the core density (Figure 2A; Figure 2—source data 1) (Wang et al., 1996; Kadoch and Crabtree, 2015). An asymmetric Rsc8 dimer defines the backbone for the complex, scaffolding the three core lobes and contacting all core proteins except Rsc3 and Rsc30 (Figure 2A,B). Rsc8 has four conserved domains, which are, from N to C terminus: SWIRM, ZZ, SANT, and coiled coil (CC) (Figure 2B). The SWIRM domains are in the arm, the ZZ and SANT domains in the head and the CC domains in the body. Previous work has shown that Rsc8 homologs are critical for the integrity of their respective SWI/SNF chromatin remodelers, supporting the idea that Rsc8 and its homologs are the structural backbone for all SWI/SNF family of remodelers (Mashtalir et al., 2018).
Apart from Rsc8, the only other subunits to span multiple lobes in our model of the core are Rsc6, Rsc9, Rsc58 and Sth1 (Figure 2C). Most of the structured regions of Rsc6 are in the body, except for a small region at the very C-terminus of the protein (residues 445–483), which occupies the space in between the three lobes. Rsc9 also primarily located in the body, but for a small region at the very N-terminus terminus (residues 28–80) interacts with the surface of the arm and head lobes. The N-terminal BrD and C-terminal Scaffold II domain of Rsc58 are in the head lobe, while the central Scaffold I domain is in the body lobe. Sth1 spans three lobes, the head, body and leg. The Scaffold I domain is in the head while the Scaffold II domain goes between the body to the head and back to the body. From the RSC-NCP model, we observe that Sth1 then continues from the Scaffold II domain to the HSA helix that, along with Arp7, Arp9 and Rtt102, form the leg lobe (Figure 2D). From the HSA helix, Sth1 continues into the ATPase domain that would putatively bind the nucleosome at position SHL2 (Figure 2D). This model of RSC would have the N- and C- termini of adjacent Sth1 domains at the right distance to connect through the short linkers between them (Figure 2D).
The body lobe of RSC contains a helical bundle backbone, an α-solenoid belly and a β-sandwich hook (Figure 3A). The helical bundle is made up of the two CC domains of Rsc8, along with the CC domains of Rsc6 and Htl1. This helical bundle interacts with the α-solenoid belly made up of the armadillo repeats of Rsc9 and the SwiB domain of Rsc6. Finally, the hook region off the belly contains β-strands from Rsc9 and Rsc6 that come to together to form a b-sandwich. Given the position of the Rsc9 sequences model in our structure, it is likely that the RFX-type WH of Rsc9 is part of the tail. In addition to these major structural regions there are some smaller elements that contribute to the body lobe. These include a segment of Rsc58 that binds between the belly and the backbone, a small segment of Sth1 that binds to the surface of the helical bundle, and a small segment of Sth1 that binds the surface of the armadillo repeat (Figure 3A). The last two components of the body are the Rsc3 and Rsc30 dimer that binds the surface of the Rsc9 armadillo repeat on top of Sth1 (Figure 3A). The Rsc3 and Rsc30 dimer appears to be absent in 75% of the RSC particles that were used in the high-resolution analysis (Figure 3B), in agreement with previous findings showing that some RSC complexes lack Rsc3 and Rsc30 (Angus-Hill et al., 2001). During our image analysis, we found that the presence or absence of the Rsc3 and Rsc30 dimer correlates well with the presence or absence of most of the tail lobe, making the Rsc3 and Rsc30 subunits the most likely constituents of this very flexible lobe (Figure 3C). The fact that there are no equivalent subunits in the yeast SWI-SNF complex, for which the tail domain is missing, further supports this assignment (Figure 1—figure supplement 2).
Four proteins contribute to most of the arm lobe, which is made of the two SWIRM domains of Rsc8 held together by the Scaffold domain of Npl6 and the RPT1 and RPT2 domains of Sfh1 (Figure 4A). Within our RSC-NCP structure, the arm lobe makes major interactions with the nucleosome, contacting both the acidic patch and nucleosomal DNA (Figure 4B). The region of the arm that contacts the acidic patch appears to emanate from the C-terminal end of the RPT2 domain of Sfh1 (Figure 4B). This region of Sfh1 is highly conserved and contains nine lysine and arginine residues that are predicted to form a helix (Figure 4C). The involvement of this region in nucleosome binding is supported by the fact that deletion of Snf5 (SWI/SNF homolog of Sfh1) in yeast results in less efficient chromatin remodeling activity of the SWI/SNF complex (Sen et al., 2017).
The region of the arm lobe contacting the nucleosomal DNA appears to emanate from either the C-terminus of the SWIRM domain of Rsc8, which is predicted to be disordered, or the N-terminus of the Scaffold domain of Npl6 (Figure 4B), which includes a CRC domain which shares homology to the WH domain of SMARCB (Allen et al., 2015; Söding et al., 2005). The WH domain of SMARCB has been shown to bind DNA and would be small enough to be accommodated within the extra density observed in our RSC-NCP structure, making the CRC domain of Npl6 the likely candidate for binding the nucleosomal DNA (Figure 4B) (Allen et al., 2015).
Six RSC subunits contribute majorly to the head lobe of RSC, which is made of the ZZ and SANT domains of the two copies of Rsc8, the BrD and Scaffold II domain of Rsc58, the Scaffold I and part of Scaffold II domain of Sth1, the β-barrel domain of Rsc4 and the anchor domain of Rsc1/2 (Figure 5A). Of these proteins, only Rsc8 and Sth1 are evolutionarily conserved, suggesting that this region probably contributes to specific functions of the RSC complex. Rsc58, Rsc4 and Rsc1/2 are notable for containing histone reader domains: BrD and BAH domains. Only one other protein within RSC, Sth1, has a histone reader domain (a BrD). These domains have been shown to interact with acetylated lysines on histone tails, particularly H3 (Chambers et al., 2013; Chatterjee et al., 2011; Duan and Smerdon, 2014). We could only model the BrD of Rsc58, while the remaining BrDs and the BAH domain extend from flexible linkers and are not visible in our structure (Figure 5A,B). Based on the position of the head relative to the arm lobe, the BrD-containing subunits in our RSC-nucleosome model are positioned so that these domains could reach the H3 tails of the engaged nucleosome (Figure 5B). Alternatively, it is possible that the BrDs of one RSC complex could interact with different, adjacent nucleosomes. Microarray data have shown that different bromodomains within RSC can recognize different acetylation sites, which could allow the complex to be targeted to a wide variety of genomic loci (Zhang et al., 2010; Filippakopoulos et al., 2012). The yeast SWI/SNF and its human homolog BAF have just one BrD – on the ATPase subunit – while human PBAF has eight, indicating that these domains may contribute to functional specificity of different classes within the SWI/SNF family.
Based on our structure and existing biochemical data, we propose a 4-step mechanism of RSC-nucleosome engagement (Figure 6). Initial recruitment of RSC to a nucleosome likely occurs through interaction with H3 acetylation marks, given that RSC has an increased affinity for these nucleosomes (Kasten et al., 2004) and that most of them as flexible, facilitating an initial encounter. RSC could then engage the nucleosome through its arm region and place the SHL2 site of the nucleosome in position to bind the flexibly tethered catalytic domain of Sth1. The ATPase domain would then be able to translocate the DNA around the histone octamer, while the core of RSC holds onto the histone core through its direct interactions with histones. This process is very likely to apply to all SWI/SNF chromatin remodelers, as they share the same architecture and contain the conserved subunits/domains involved in the steps just proposed. Specifically, all complexes have bromodomains to bind acetylated histones (minimally in the C-terminus of their catalytic subunit), a Sfh1 homolog to bind the acidic patch of the nucleosome, and a catalytic DNA translocase domain to remodel nucleosomes (Figure 2—source data 1) (Clapier and Cairns, 2009).
Our structure of the RSC core allows us to visually represent the architecture of assembly intermediates of human SWI/SNF complexes that were previously identified based by biochemical and mass-spectrometry experiments (Figure 7) (Mashtalir et al., 2018) and to rationalize the assembly pathway based on the molecular contacts observed in our structure. According to this analysis, dimerization of the Rsc8 CC domains is most likely the first step in RSC complex assembly, followed by the binding of Rsc6 to this CC dimer. The RPT1 and RPT2 motifs within Sfh1 would then bring the two SWIRM domains of Rsc8 together. This initial order of assembly is based on the assembly intermediates identified for the mammalian SWI/SNF complexes BAF and PBAF (Mashtalir et al., 2018) where initial dimerization of SMARCC (the mammalian Rsc8 homolog) is followed by the sequential binding of SMARCD (the mammalian Rsc6 homolog), SMARCB (the mammalian Sfh1 homolog), and SMARCE (no known RSC/yeast homolog) to form the BAF core.
Subsequent steps of RSC assembly likely involve binding of Rsc9 and Sth1. The armadillo repeat domain of Rsc9 binds the Rsc8 CC and Rsc6 to form most of the body lobe, while the scaffolding domains (I and II) of Sth1 contribute to both the head and body regions and exit towards the flexible leg lobe. The integration of Sth1 into the complex would also lead to recruitment of the components of the ARP module (Arp7, Arp9 and Rtt102). This order of events is derived from the corresponding phase in the assembly of mammalian SWI/SNF complexes (Mashtalir et al., 2018), which involves association of either ARID1 or ARID2 with the BAF core. ARID1 and ARID2 are paralogs, and both are orthologs to Rsc9, while the rest of the BAF complex-specific factors have no clear yeast orthologs. The incorporation of the ARID subunits is followed by the addition of the ATPase module, which includes the conserved SMARCA, ACTB, and ACTL6A subunits, which are the homologs of yeast Sth1, Arp7, and Arp9, respectively.
It is important to note that this initial assembly process primarily involves evolutionarily conserved subunits and domains and is thus likely conserved for all SWI/SNF chromatin remodelers (Kadoch and Crabtree, 2015). This model is supported by the structural similarities we observe between the yeast SWI/SNF and RSC complexes (and in agreement with a recently reported structure of yeast SWI/SNF bound to a nucleosome by Han et al., 2019). In the case of RSC, assembly would then continue with the addition of the yeast specific factor Npl6, which completes the arm lobe, and RSC-specific subunits Htl1, Rsc58, Rsc1/2, Rsc4, and Ldb7. These subunits either contribute to the scaffold of the complex (Htl1 and Ldb7), serve to anchor bromodomains to the core (Rsc1/2 and Rsc4), or both (Rsc58). The last two subunits to be recruited to the complex are likely the RSC-specific subunits Rsc3 and Rsc30, which form the tail module.
Our studies provide the detailed structure for the conserved core of SWI/SNF complexes and lead to a model of assembly for this family of remodelers that agrees with previous biochemical data on mammalian complexes. Our work also provides a model of how RSC stably contacts the core histones in a nucleosome via its body module, while engaging nucleosomal DNA via the Sth1 ATPase. Both the structural integrity of the core and the interaction with nucleosomes rely on evolutionarily conserved subunits, while the non-conserved proteins in RSC are likely to play a role in recruitment and regulation of the complex. Future studies will be needed to understand the mechanisms by which individual members of the SWI/SNF family of chromatin remodelers are brought to different genome loci and are distinctly regulated.
SWI/SNF was purified from Saccharomyces cerivisiae using a modified TAP purification as described in Nagai et al. (2017); Puig et al. (2001). A strain modified with a TAP tag on the C terminus of SNF2 was obtained from GE Dharmacon and grown at 30°C in YPD. 20 L of cells were harvested at OD six and lysed using a cryo-mill. The ground cells were resuspended in a lysis buffer (50 mM HEPES pH 7.9, 25 mM ammonium sulfate, 0.5 mM EDTA, 100 μM zinc sulfate, 5% glycerol, 5 mM DTT, 10 μM leupeptin, protease inhibitor,. 01% NP-40) and dounced to ensure homogenization. The lysate was spun at 11,000 g for 20 min. The supernatant was removed and brought to 200 mM ammonium sulfate, followed by an addition of polyethyleneimine (0.2% final concentration) to precipitate DNA. The sample was spun again at 17,000 g for 40 min. Supernatant was removed and brought up to 2.2 M ammonium sulfate, then spun again at 17,000 g for 40 min. The protein pellet was resuspended in buffer (lysis buffer with 0 M ammonium sulfate and 2 mM DTT) and the ammonium sulfate concentration was brought up to 400 mM. IgG resin (1 ml packed) was equilibrated and incubated for 4 hr. Resin was washed with 500 mM ammonium sulfate buffer, then 1x with 100 mM ammonium sulfate buffer. Protein was released via TEV cleavage by overnight incubation with 20 μg TEV (MacroLab) in AS-100. Following cleavage, samples were spun to collect supernatant. The CaCl2 concentration of the supernatant was brought up to 2 mM and loaded onto a 100 μl equilibrated CBP resin. After 4 hr of incubation, supernatant was removed and the resin was washed 4x (20 mM HEPES pH 7.9, 2 mM MgCl2, 10% glycerol, 250 mM KCl, 50 μM ZnCl2, 2 mM CaCl2, 10 μM leupeptin, 1 mM TCEP, 0.01% NP-40). The sample was eluted with one volume elution buffer (20 mM HEPES pH 7.9, 2 mM MgCl2, 10% glycerol, 250 mM KCl, 50 μM ZnCl2, 2 mM EGTA, 10 μM leupeptin, 1 mM TCEP, 0.01% NP-40) for 30 min, then 15 min for each subsequent elution. Samples were aliquoted, frozen in liquid nitrogen and stored at −80°C.
RSC was purified from Saccharomyces cerivisiae using a TAP-tag method as described (Puig et al., 2001; Dutta et al., 2017). A strain modified with a TAP tag at the C terminus of STH1 was obtained from GE Dharmacon and grown at 30°C in YPD. 10 L of cells were harvested at OD seven and lysed using a cryo-mill. The ground cells were resuspended in a lysis buffer (50 mM HEPES pH 7.9, 250 mM KCl, 0.5 mM EDTA, 100 μM ZnSO4, 10% glycerol, 2 mM DTT, 10 μM leupeptin, protease inhibitor, 0.05% NP-40) and dounced to ensure homogenization, followed by an addition of 18 μL of benzonase (Sigma) while spinning on ice. After 10 min, another 18 μL of benzonase were added. Heparin (Sigma) was then added slowly to a final concentration of 0.5 mg/mL and incubated for 10 min. The lysate was then spun for 90 min at 17,000 g. Supernatant was collected and clarified through a column frit, then bound to 2.5 mL packed IgG resin and incubated for 4 hr at 4°C. After incubation, the supernatant was removed. TCEP was added to the lysis buffer for a final concentration of 0.5 mM and washed 5x over the resin. The sample was released via overnight incubation with 25 μg TEV protease in one volume buffer. Following cleavage, samples were spun to collect supernatant. The CaCl2 concentration of the supernatant was brought up to 2 mM and loaded onto 100 μL CBP resin equilibrated five times with wash buffer (20 mM HEPES pH 7.9, 10% glycerol, 150 mM KCl, 50 μM ZnCl2, 2 mM CaCl2, 10 μM leupeptin, 1 mM TCEP, 0.01% NP-40). After 4 hr of incubation, the supernatant was removed and the resin was washed 2x with wash buffer with 750 mM KCl, then 3x with wash buffer with 250 mM KCl buffer, then 1x with wash buffer without CaCl2. The sample was eluted with one volume (20 mM HEPES pH 7.9, 2 mM MgCl2, 10% glycerol, 150 mM KCl, 50 μM ZnCl, 2 mM EGTA, 10 μM leupeptin, 1 mM TCEP, 0.01% NP-40) for 30 min, then 15 min for each subsequent elution. Samples were aliquoted, frozen in liquid nitrogen and stored at −80°C.
We used the ~250 μg RSC2 TAP-tag purified RSC complex and crosslinked by 4 mM bis(sulfosuccinimidyl)suberate (BS3) at RT for 2 hr. Sample processing and mass spectrometry data analyses were done as described before (Mashtalir et al., 2018). After plink2 and Nexus database searches against the RSC subunit sequences, about 6.6% of interlinked spectra and 3.1% intralinked spectra were removed after manually spectrum checking. All the crosslinked spectra can be viewed at https://www.yeastrc.org/proxl_public/viewProject.do?project_id=234.
For negative stain, the samples were cross-linked at room temperature for 5 min using 1 mM final concentration of BS3. After cross-linking, 4 μL were applied to a glow discharged continuous carbon grid for 5 min, then stained with uranyl formate. A tilted negative stain data set was collected on a Tecnai F20 microscope (FEI) operated at 120 keV and equipped with an Ultrascan 4000 camera (Gatan). Data were collected using Leginon data acquisition software (Suloway et al., 2005). The CTF parameters were estimated using Gctf (version 1.16) and particles were picked using Gautomatch (version 0.50, from K Zhang, MRC-LMB, Cambridge) using gaussian blob templates (Zhang, 2016). Data processing was done using Relion (version 3.0) (Zivanov et al., 2018). The negative stain structure (EMD-6834) from Zhang et al. (2018) was used as an initial model. Extracted particles were subjected to 2D classification and 3D classification to obtain a homogenous population. Particles that went into the best classes were then refined.
For cryo-EM sample preparation we used a Vitrobot Mark IV (FEI). RSC was crosslinked on ice using 1 mM BS3 (Thermo Fisher Scientific) for 15 min before 4 μL of sample was applied to either a 2/2 holey carbon or 1.2/1.3 UltrAuFoil grids (Quantifoil) at 4°C under 100% humidity. Grids were cleaned using Toguro plasma cleaner, using the 3 nm carbon setting run twice. The sample was immediately blotted away using Whatman #1 for 2–3 s at 0 N force and then immediately plunge frozen in liquid ethane cooled by liquid nitrogen. For the RSC-nucleosome complex sample preparation, 4 μL of RSC (2 pmol) and 1 μL of nucleosome (H3K4me3 and H3K(9/14/18)ac) (1 pmol) (Epicypher) were incubated for 10 min at 30 °C followed by the addition of 0.5 μL of AMPPNP (0.5 nmol) and an additional incubation for 10 min at 30°C. The samples were then placed on ice and crosslinked with 1 mM BS3 for 15 min. RSC-nucleosome samples were prepared in the same way as RSC.
For the RSC sample, frozen grids were clipped and transferred to the autoloader of a Titan Krios electron microscope (Thermo Fischer Scientific) operating at 300 keV (PNCC). Images were recorded with a K3 direct electron detector (Gatan) operating in super-resolution mode at a calibrated magnification of 46,339 (1.079 Å/pixel) and a mean defocus of −1.04 μm with a 0.24 μm standard deviation, using the SerialEM data collection software (Schorb et al., 2019). 50-frame exposures were taken at 0.06 s per frame, using a dose rate of 11.455 e-/pixel/s (0.6 e- Å−2 per frame), corresponding to a total dose of 40 e-Å−2 per micrograph (Figure 1—figure supplement 3). A total of 8122 movies were collected from a total of 3 grids.
For the RSC-nucleosome sample, frozen grids were clipped and transferred to the autoloader of a Talos Arctica electron microscope (Thermo Fischer Scientific) operating at 200 keV acceleration voltage (UCB). Images were recorded with a K3 direct electron detector (Gatan) operating in super-resolution mode at a calibrated magnification of 43,859 (1.14 Å/pixel) and a mean defocus of −1.66 μm with a 0.41 μm standard deviation, using the SerialEM data collection software (Schorb et al., 2019). 50-frame exposures were taken at 0.065 s per frame, using a dose rate of 11.838 e-/pixel/s (1 e- Å (Yen et al., 2012) per frame), corresponding to a total dose of 50 e-Å−2 per micrograph (Figure 1—figure supplement 9). A total of 9190 movie were collected from a single grid.
All data processing was performed using Relion3 (version 3.0) (Zivanov et al., 2018). For the RSC dataset, whole movie frames were aligned and binned by 2 (1.079 Å/pixel) with MotionCor2 to correct for specimen motion The CTF parameters were estimated using Gctf (Zhang, 2016; Zheng et al., 2017). 1,245,498 particles were picked with LoG picker. Particles were extracted binned by 4 (4.316 Å/pixel) and subjected to two- and three-dimensional classification to remove ice and empty picks, which resulted in 1,074,750 particles. The negative stain reconstruction was used as an initial model for 3D classification. The particles were then centered and reextracted bin 1.33 (1.4386 Å/pixel) and refined. The refinement was performed without a mask and resulted in a reconstruction were only the core of the complex was well resolved. The refinement was then continued with a mask around the core. The masked refinement resulted in a 3.8 Å-resolution map of the core. Masked local search 3D classification was performed to select for the best particles. The best class, containing 252,918 particles, was selected and refined, resulting in a 3.4 Å map. Three iterations of CTF refinement, particle polishing and 3D refinement were performed, which resulted in reconstructions at 3.26, 3.21 and 3.18 Å (Zivanov et al., 2019). Local search 3D classification was performed to select the best particles. The best class, containing 192,066 particle images, was selected and further refined, resulting in a 3.14 Å-resolution map. One last iteration of CTF refinement, particle polishing and refinement was performed and led to a reconstructions at 3.07 Å resolution. The core was then subjected to multibody refinement by masking each of the three lobes separately (Nakane et al., 2018). The arm, head and body refined to 3.23, 3.14, and 2.96 Å respectively. 3D classification of the partially signal subtracted particles was then performed. For the arm lobe a single good class was identified, which when refined resulted in a 3.16 Å map. For the head and body lobes two good classes were found, with one containing several extra helices. The classes containing the extra density were selected and refined. The head lobe refined to 3.07 Å, and the body refined to 3.48 Å. To characterize the two flexible lobes of the complex 3D classification was performed for the 192,066 subset of particles. The classification with four classes resulted in a continuum of states for the leg lobe. The complete tail lobe was only present in two of the four classes, with the tail in two different conformations.
For the RSC dataset, whole movie frames were aligned and binned by 2 (1.14 Å/pixel) with MotionCor2 to correct for specimen motion The CTF parameters were estimated using Gctf (Zhang, 2016; Zheng et al., 2017). 2,327,957 particles were picked with LoG picker. Particles were extracted binned by 4 (4.56 Å/pixel) and subjected to two- and three-dimensional classification to remove ice, empty picks, free nucleosomes which resulted in 48,222 particles. The cryo-EM reconstruction of RSC alone was used as an initial model for 3D classification. The particles were refined to 19 Å and was subjected to another round to 3D classification. Further refinement from 3D classification did not improve reconstruction quality (not shown).
Some of the software packages mentioned above were configured by SBgrid (Morin et al., 2013).
The model for the RSC core was generated by manually building a poly-alanine trace through the final global refinement and multibody maps in COOT (Emsley et al., 2010). Each of the chains was then identified with the help of blobMapper.py (Zukin, 2020; copy archived at https://github.com/elifesciences-publications/blobMapper) and secondary structure predication (Figure 1—figure supplement 4) (Jones, 1999; Buchan et al., 2013; Jones and Cozzetto, 2015). The resulting coordinate model was iteratively refined using the real space refinement algorithm implemented in PHENIX (Afonine, 2018a). Ramachandran, secondary structure, Cβ, and rotamer restraints, as well as bond length and bond angle restraints for the Zn2+ ion in the Rsc8 ZZ domain, were used throughout to ensure good model geometry. The final round of refinement comprised 5 rounds of global minimization as well as b-factor refinement and used a resolution limit of 3.1 Å, according to the average resolution of cryo-EM maps used (Figure 1—figure supplement 7F) in order to avoid overfitting of the model. The refinement weight was automatically determined by PHENIX, which monitors the bond length and bond angle deviations to maintain good model geometry and avoids over-fitting of the model to the map (Afonine et al., 2018b). The model was validated using MTRIAGE and MOLPROBITY within PHENIX (Afonine et al., 2018b). The refinement statistics are given in (Figure 1—figure supplement 7H) and show values typical for structures in this resolution range (MOLPROBITY score = 2.1) (Duan and Smerdon, 2014). The FSC curve between the model and the map shows good correlation up to 3.2 Å resolution according to the FSC = 0.5 criterion (Figure 1—figure supplement 7G) (Afonine et al., 2018b).
The model of the RSC-NCP complex used for visualization was generated by docking our model of the RSC core, the crystal structure of the Arp module (PDB 4I6M: Arp7, Arp9, Rtt102 and Snf2-HSA) (Schubert et al., 2013) and cryo-EM structure of Snf2-MMTV nucleosome complex bound at the SHL2 with ADP (PDB 6IY2: Snf2, nucleosome) (Li et al., 2019). The sequence for the Snf2-HSA helix in 4I6M and Snf2 helicase domain in 6IY2 were aligned to the Sth1 sequence and mutated in COOT (Emsley et al., 2010). For deposition of the coordinate model to the PDB, we replaced the model of the Snf2-MMTV nucleosome complex with the structure of a human nucleosome (PDB 2CV5) (Tsunaka et al., 2005), without bound ATPase domains, because the ATPase domain of Sth1 was not resolved in our cryo-EM map.
Some of the software packages mentioned above were configured by SBgrid (Morin et al., 2013).
Depiction of molecular models were generated using PyMOL (The PyMOL Molecular Graphics System, version 1.8, Schrödinger), the UCSF Chimera package from the Computer Graphics Laboratory, University of California, San Francisco (supported by National Institutes of Health P41 RR-01081) and UCSF ChimeraX developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco, with support from National Institutes of Health R01-GM129325 and the Office of Cyber Infrastructure and Computational Biology, National Institute of Allergy and Infectious Diseases (Pettersen et al., 2004; Goddard et al., 2018). Protein domains graphs (Figure 1—figure supplement 4; Figure 4) were generated using domainsGraph.py (Patel, 2020; copy archived at https://github.com/elifesciences-publications/domainsGraph).
Some of the software packages mentioned above were configured by SBgrid (Morin et al., 2013).
During the submission of this manuscript, another structure of RSC bound to a nucleosome was published by Ye et al. (2019) and Wagner et al. (2019). The structures by Ye et al and our own are very similar with few exceptions. These mainly stem from the apparent dissociation of parts of Ldb7 from the body and the binding of the ATPase domain to nucleosome in the reconstruction of Ye et al (Appendix 1—figure 1). The structure by Wagner et al has not yet been deposited so direct comparison could not be made but the overall structure appears very similar.
The region we attribute to be residues 59–127 of Ldb7 appears to be missing from the reconstruction of Ye et al (Appendix 1—figure 1A). Additionally, segments of some chains we observe around this site appear to have been more flexible in the Ye et al. structure. The region of Sth1 attributed to residues (2-34) by Ye et al is assigned by us to correspond to Ldb7 residues 6–36 (Appendix 1—figure 1 – Appendix 1—figure 1—figure supplement 1A). Additionally, two β-strands within the β-sandwich that we identified as being the ZZ domain of Rsc8 were proposed to correspond to Rsc4 residues 429–449 (Appendix 1—figure 1 – Appendix 1—figure 1—figure supplement 1B C, D). This resulted in a change in register for a third β-strand (residues 364–390) (Appendix 1—figure 1 – Appendix 1—figure 1—figure supplement 1E, F, G, H). Within our map, we see connecting density between the SANT and ZZ domains of Rsc8 that allowed us to trace this chain.
Our docking of the ARP module into the leg density and prediction that the ATPase module would bind at the SHL2 site of the nucleosome matched well with the structure of Ye et al. Within their structure they were able to observe the binding of the ATPase domain to the SHL2 site. This likely explains the shift of the ARP module between our two reconstructions, with the ARP module closer to the nucleosome in their reconstruction then in ours. This likely means that the ARP module remains flexible to some degree until the ATPase module binds the nucleosome.
One major difference between the three structure of RSC bound to the nucleosome was the length of the DNA used in generating the nucleosome. We had only the 147 bp WIdom 601 sequence with no extranucleosomal DNA, Ye et al used the 147 bp WIdom 601 sequence with an additional 20 bp on one end making the total length 174 bp and Wagner et al used the 145 bp WIdom 601 with an additional 58 bp on one end and 37 bp on the other making the total length 237 bp. Due to the extra length of the extranucleosomal DNA in their structure, Wagner et al were able to observe the extranucleosomal DNA extending to the tail lobe where it appears to bind. In our RSC-NCP sample we also observe the tail lobe binding DNA but, in that case, the entire tail lobe is swiveled down towards the nucleosome and is interacting with the nucleosomal DNA. Due to the lack of resolution in both structures for this region, it was not possible to determine structure of the tail lobe and therefore, future studies will be required to identify full function of the tail of RSC.
During the submission of this manuscript two studies by Wagner et al. (2019) and Valencia et al. (2019) showed that the the C-terminal region of Sfh1 and SMARCB1, respectively, interacts with the acidic patch of the nucleosome. In the study by Wagner et al, the structure of RSC-NCP was determined using cryo-EM and it was shown that a portion of the C-terminus of Sfh1 forms an α-helix and binds the acidic patch of the nucleosome. Valencia et al show that the C-terminal region of SMARCB1 (Sfh1 homolog) forms an α-helix by NMR. Additionally, these authors showed that this α-helix will crosslink to residues near the acidic patch of the nucleosome and by computational docking that this α-helix will bind the acidic patch of the nucleosome. Valencia et al also showed that deletions of mutations in this helix reduced remodeling efficiency in vitro but did not prevent nuclear localization in vivo. These last two findings support our proposal that that Sfh1 and the arm lobe plat a role in nucleosome engagement but not RSC recruitment.
During the submission of this manuscript the structure of SWI/SNF bound to a nucleosome was published by Han et al. (2019). While at the time of final submission of this manuscript, the structure by Han et al has not been deposited yet, and a direct comparison could not be made, several clear similarities and differences are apparent. While the overall structure of SWI/SNF and RSC appear to be similar with respect to organization of homologous subunits, the resulting shape of the complex differs. The core of RSC appears to have three distinct lobes (head, body and arm) but in the case of SWI/SNF the core appears to form a single triangular shaped lobe.
The biggest difference between the two remodelers appears to stem form the lack of many of the head lobe components of RSC in SWI/SNF (e.g. Rsc58, Rsc1/2, Rsc4 and Ldb7). Because of this, the remaining head lobe components of SWI/SNF, the SANT domains of Swi3 (Rsc8 homolog) and the anchor domain of Snf2 (homologous to the Scaffold II domain of Sth1), interact more closely with the body lobe.
The body lobe of SWI/SNF is organized in a similar way as in RSC, but the overall shape appears different. While Swi1, Swi3 and Swp73 are similar to their RSC counterparts Rsc9, Rsc8 and Rsc6, they are different enough so that the body appears to curve toward the arm lobe in SWI/SNF as opposed to away from it, as in RSC. Interestingly it appears that Swi6 may be homologous to Htl1 as they share similar binding sites in the helical bundle within the body.
The region most similar between RSC and SWI/SNF appears to be the arm lobe. This may be due to structural or functional constraints, considering that it is the arm lobe of both complexes that interacts the nucleosome. However, the arms of the two complexes bind over the nucleosome at different positions. The arm of RSC binds over the H2A/H2B dimer of the nucleosome while in SWI/SNF, the arm binds over the H3/H4 surface. So, it remains to be seen why the binding sites for the arm lobes is different between these two complexes despite the lobes themselves being so similar.
Real-space refinement in PHENIX for cryo-EM and crystallographyActa Crystallographica Section D Structural Biology 74:531–544.https://doi.org/10.1107/S2059798318006551
New tools for the analysis and validation of cryo-EM maps and atomic modelsActa Crystallographica Section D Structural Biology 74:814–840.https://doi.org/10.1107/S2059798318009324
Scalable web services for the PSIPRED protein analysis workbenchNucleic Acids Research 41:W349–W357.https://doi.org/10.1093/nar/gkt381
The PSIPRED protein analysis workbench: 20 years onNucleic Acids Research 47:W402–W407.https://doi.org/10.1093/nar/gkz297
Structure of a RSC-nucleosome complex and insights into chromatin remodelingNature Structural & Molecular Biology 15:1272–1277.https://doi.org/10.1038/nsmb.1524
The BAH domain of Rsc2 is a histone H3 binding domainNucleic Acids Research 41:9168–9182.https://doi.org/10.1093/nar/gkt662
Histone H3 tail acetylation modulates ATP-dependent remodeling through multiple mechanismsNucleic Acids Research 39:8378–8391.https://doi.org/10.1093/nar/gkr535
The biology of chromatin remodeling complexesAnnual Review of Biochemistry 78:273–304.https://doi.org/10.1146/annurev.biochem.77.062706.153223
Composition and function of mutant swi/Snf complexesCell Reports 18:2124–2134.https://doi.org/10.1016/j.celrep.2017.01.058
Protein secondary structure prediction based on position-specific scoring matricesJournal of Molecular Biology 292:195–202.https://doi.org/10.1006/jmbi.1999.3091
iMODFIT: efficient and robust flexible fitting based on vibrational analysis in internal coordinatesJournal of Structural Biology 184:261–270.https://doi.org/10.1016/j.jsb.2013.08.010
domainsGraph, version 0edfc05GitHub.
UCSF chimera-a visualization system for exploratory research and analysisJournal of Computational Chemistry 25:1605–1612.https://doi.org/10.1002/jcc.20084
Chromatin remodeling by RSC involves ATP-dependent DNA translocationGenes & Development 16:2120–2134.https://doi.org/10.1101/gad.995002
RSC mobilizes nucleosomes to improve accessibility of repair machinery to the damaged chromatinMolecular and Cellular Biology 27:1602–1613.https://doi.org/10.1128/MCB.01956-06
Acetylated histone tail peptides induce structural rearrangements in the RSC chromatin remodeling complexJournal of Biological Chemistry 282:20804–20808.https://doi.org/10.1074/jbc.C700081200
Structural analysis of the yeast SWI/SNF chromatin remodeling complexNature Structural Biology 10:141–145.https://doi.org/10.1038/nsb888
The HHpred interactive server for protein homology detection and structure predictionNucleic Acids Research 33:W244–W248.https://doi.org/10.1093/nar/gki408
Automated molecular microscopy: the new leginon systemJournal of Structural Biology 151:41–60.https://doi.org/10.1016/j.jsb.2005.03.010
Alteration of the nucleosomal DNA path in the crystal structure of a human nucleosome core particleNucleic Acids Research 33:3424–3434.https://doi.org/10.1093/nar/gki663
A genome-wide screen for essential yeast genes that affect telomere length maintenanceNucleic Acids Research 37:3840–3849.https://doi.org/10.1093/nar/gkp259
Diversity and specialization of mammalian SWI/SNF complexesGenes & Development 10:2117–2130.https://doi.org/10.1101/gad.10.17.2117
Structure of chromatin remodeler Swi2/Snf2 in the resting stateNature Structural & Molecular Biology 23:722–729.https://doi.org/10.1038/nsmb.3259
Structural insights into BAF47 and BAF155 complex formationJournal of Molecular Biology 429:1650–1660.https://doi.org/10.1016/j.jmb.2017.04.008
Swi3p controls SWI/SNF assembly and ATP-dependent H2A-H2B displacementNature Structural & Molecular Biology 14:540–547.https://doi.org/10.1038/nsmb1238
Architecture of SWI/SNF chromatin remodeling complexProtein & Cell 9:1045–1049.https://doi.org/10.1007/s13238-018-0524-9
BlobMapper, version 94182f5GitHub.
Cynthia WolbergerSenior and Reviewing Editor; Johns Hopkins University School of Medicine, United States
In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.
The RSC complex is nucleosome remodeler that maintains nucleosome-free regions in yeast and is the most abundant member of the SWI/SNF family of enzymes. This paper reports the cryo EM structure of the RSC complex core, containing 14 of the 17 RSC subunits. The structure reveals the architecture of the complex, and the similarities and differences as compared to other SWI/SNF remodelers. A low-resolution structure of RSC bound to a nucleosome core particle suggest a model for how RSC engages its substrate. Taken together, this work is an important advance in the chromatin and transcription fields, and provides a structural basis for interpreting previous biochemical studies in order to derive a possible model for RSC assembly and function.https://doi.org/10.7554/eLife.54449.sa1
[Editors' note: we include below the reviews that the authors received from another journal, along with the authors’ responses.]
As you will see the paper has been significantly changed. While all the original content is still present, we have restructured most of this. This was in part to fit the eLife format but also to accommodate the additional discussion on analyzing our structure. In addition to these changes and additions we have also added (as was asked) a comparison of our structure with the one published in Science by Ye et al. The comparison has been added as Appendix I. We also included two other Appendices (II and III) which cover three new papers: 1) [Valencia et al., Cell] on the analysis of SMARCB1 C-terminus and its ability to bind the acidic patch of the nucleosome, 2) [Wagner et al., bioRxiv] on the structure of RSC bound to a nucleosome, and 3) on the SWI/SNF bound to a nucleosome.
Overall, we felt that while some of the issues raised by the reviewers are understandable and will help us revise our manuscript, others are not.
One of the major areas of concern raised was that the assembly process could not be determined from our structure alone. This is indeed true, but it misses the fact that our model is based not only on our structure, but also on published literature (citation 16: Mashtalir, N. et al.) that provides biochemical data in support of our model. To emphasize this, we will change “The pattern of protein‐protein interactions observed in our structure provides insight into RSC assembly” to “Based on the observed protein-protein contacts in our structure of RSC and previous mass spectrometry analysis of the assembly of human BAF complexes, we propose a model for RSC assembly”. In this way we clarify that the model we propose originates from previous biochemical data and that our structure can now help to rationalize this model at the domain level. While we thought that this would be obvious, given that the literature describing the assembly process is highly cited in our work, it appears that both reviewers 1 and 2 did not recognize this and we will make it as clear as possible in the revised manuscript.
We have now included more extensive discussion on this by referencing the key intermediates that Mashtalir et al. have found.
The other major issue that was brought up (by reviewer 1 and 2) was that our cryo-EM structure of the RSC-NCP was too low of a resolution and that our claims about the structure were too speculative. While it is true that our structure of nucleosome-bound RSC is at low resolution (similar to previously published structures) nearly all previous reconstructions were of very low quality (e.g. Leschziner et al., 2007, Yuriy et al., NSMB 2008, Dechassa et al., 2008 and Zhang et al., 2018). It should be noted that resolution does not always accurately measure map quality. Therefore, while our map is at a similar resolution to previous reconstructions it does not suffer from the clear misalignment, anisotropy, and apparent flattening of previous works. Specifically, the EM map by Zhang et al., which was suggested by reviewer 1 to be of similar quality to ours, suffers from severe flattening and precludes fitting of the molecular components, as revealed by even the most cursory inspection. We therefore strongly disagree with the claim that our structure does not provide significant new insight into the architecture of nucleosome-bound RSC. Based on our map, we can unambiguously dock and assign all observable density, the core, the ARP module and the nucleosome. As far as not being able to observe any bromodomain density we would like to point out that one would not expect to see these domains because they are flexibly attached to the core of the complex and even in the case of the nucleosome bound complex, they would be interacting with histone tails, which are themselves also flexible.
The remaining issues raised by reviewer 1 are hard to comprehend.
Specifically, (i) the issue with Rsc3 and Rsc30 being sub-stoichiometric has been well documented and the corresponding literature has been cited in our paper; (ii) we do not think that measuring complex stoichiometry based on the gel or running a glycerol gradient to ascertain that we indeed purified an assembled complex is worthwhile, given that we have a 3-Å resolution structure of the core of complex; (iii) we cite 2 of the 3 key papers the reviewer mentions we “missed” (which includes one authored by some of the co-authors of our study), while the third is only tangentially relevant.
It is our assessment that the paper can be revised to clarify some of the issues raised by the reviewers. However, several of the issues raised are not applicable, as detailed above.
In addition to these general considerations, we provide point-by-point answers to the reviewers’ comments below.
In the first part of this paper, the authors use cryo-EM to determine the structure of free RSC, one of two SWI/SNF complexes present in yeast, and were able to resolve the core of this complex to 3.1 angstroms. The core has a more rigid region consisting of a head, body, and arm regions plus a more flexible leg region. The structure they observe is at a higher resolution than any previously reported structure for a SWI/SNF complex and provides detailed information about the subunit-subunit interactions within this complex. They complement their cryo-EM structure by also probing the structure using chemical crosslinking and mass spectrometry (CL-MS) in collaboration with the Ranish lab.
In support of their CL-MS results they cite similar data from the Kadoch lab, but missed the earlier work from Sen et al., 2017, using yeast SWI/SNF that is actually more similar to RSC.
We do not use the work of the Kadoch lab to support our CL-MS results. Their work is mainly cited for the assembly process that they propose. The work of Sen et al. is cited (number 22) for other findings.
From their data it was not clear how they handled the heterogeneity of RSC since their purified enzyme had both Rsc1 or Rsc2, which are mutually exclusively of each other in the complex, and raises the question why they didn’t tag Rsc1 or Rsc2 so they could purify one or the other form of the RSC complex.
In our cryo-EM data processing we did not find that the complexes could be distinguished based on whether they contain Rsc1 or Rsc2. However, we did find a region in our cryo-EM map where we could model a small region of Rsc1 and 2 that is highly similar between these two proteins. It is worth noting that the sequences of Rsc1 and Rsc2 are >40% identical and another 20% is highly homologous, suggesting that they form highly similar contacts within RSC.
We reasoned that if Rsc1 or 2 lead to two structurally distinct complexes we would be able to distinguish these computationally, while otherwise, the two can be treated as the same. The latter ended up being the case as the regions that anchor the two proteins into the core of the complex are highly similar.
We have included an additional figure (Figure 1 - figure supplement 6) showing an alignment of the regions of RSC1 and 2 we modeled and its fit into the density.
It would be helpful if in the paper the authors stated which portions of each of the 14 different subunits were visualized in the core and were well resolved and which portions were too flexible to be resolved.
Figure 1—figure supplement 4 contains this information.
Figure 1—figure supplement 4 was essentially indiscernible and needs to be remade so that it can be actually read.
We have made a new version of the domain map figure for the main text (Figure 1F). It does not have the secondary structure prediction or sequence conservation mapped on to the domain map. Instead it the modeled region shown in much larger font. We chose to keep the original figure (we made some changes) in the supplement (Figure 1—figure supplement 4) because the secondary structure predication was used when building to assign proteins to chains. Additoinal we have included Figure 1 - figure supplement 5 to show the fit of various modeled regions into the map.
The purity and overall subunit stoichiometry in the purified RSC appear to be in question and was difficult to discern in the SDS-PAGE shown in Figure 1—figure supplement 1A. What type of gel was used, what was the percentage acrylamide, and what type of staining was used to visualize the proteins?
We used a BioRad 4-20% gel and stained with flamingo stain. We can rerun the gel to better separate the region that contains Rsc4, Rsc6, Rsc8, Rsc9, Npl6, Sfh1, Rsc58, Arp7, Arp9, but as the molecular weights differ only 20 kDa between the largest and smallest subunit of this group, complete separation of these bands is not likely.
Based on the in-solution tryptic digest and mass spectrometry data the sample is pure (Figure 1—figure supplement 2). The relative stoichiometry is difficult to determine for this complex using gel densitometry as many of the bands overlap in molecular weight.
If the gel was stained with Coomassie Blue then the authors need to give estimates of subunit stoichiometry based on staining intensity. Could the authors also comment on why the Rsc3 and Rsc30 subunits appear to be absent in 75% of their complexes and if this agrees with their stained gel?
The gel is not Coomassie Blue stained.
The fact that Rsc3 and 30 are substoichiometric has previously been reported, see citation 20 (El-Gebali et al., 2019).
Given the potential heterogeneity it is imperative to show how well assembled their complex is using such approaches as gel filtration or glycerol gradient at the minimum.
We can include negative stain micrographs and 2D class averages (Figure 1 -figure supplement 2) that show that we do not observe any partially assembled complexes. We also want to note that (i) we used complexes purified from a native source, making mis-assembly of the complex unlikely and (ii) we present in this paper a high-resolution structure of the core of this complex, which of course supports the notion that we are looking at a well-assembled and pure molecular assembly.
The negative stain comparison of RSC vs SWI/SNF was a good addition for showing the lack of a tail lobe that likely contains those proteins or domains not shared between the two complexes. Could the tail lobe contain the Rsc3 and Rsc30 subunits?
Yes, we do think this is the case and have stated “we found that the presence or absence of Rsc3 and Rsc30 correlates well with the presence or absence of most of the tail lobe, making the Rsc3 and Rsc30 subunits the most likely constituents of this very flexible lobe (now in Figure 3).” The reviewer apparently missed this part of our manuscript.
The authors also did not cite the yeast SWI/SNF work from Craig Peterson’s lab which showed the importance of Swi3 for scaffolding of the complex (Yang et al., Nat. Struct. Mol. Biol. 14:540-547 ), while only citing the later work with BAF155/170.
We have added this citation. However, this paper only looked at how the SANT domain of Swi3 effected the assembly of SWI/SNF, and so we felt that the paper did not strongly support the fact that all domains of Rsc8 were important in scaffolding RSC.
This paper makes significant progress in unraveling the subunit architecture of the free RSC complex, but I think it goes too far in suggesting the order of RSC assembly based solely on the structure. A fair amount of emphasis is placed on the order of assembly as it is discussed in two paragraphs and as stated is merely the authors’ proposal without any additional supporting evidence.
We can change “The pattern of protein‐protein interactions observed in our structure provides insight into RSC assembly” to “Based on the observable protein-protein contacts in our structure of RSC, along with previous mass spectrometry analysis of the assembly of human BAF complexes, we propose a model for RSC assembly.”
Our aim was to build off the model that was recently proposed in the Kadoch lab’s recent paper (citation 16: Mashtalir, N. et al.). What our work adds to theirs is moleuclar detail, as we can show which subunits are interacting, and suggest what the Kadoch model would look like in real molecular terms.
Like we stated above we have significantly expanded the discussion on this by referencing the key intermediates that Mashtalir et al. have found.
The authors then visualize the RSC-nucleosome structure at a resolution of 19 angstroms, which is not significantly better than a recent report of Zhang et al., 2018, for the SWI/SNF complex. At this resolution it is not possible to readily discern the structural features or subunit organization of how RSC interact with nucleosomes. The other serious issue with this structure is the lack of observable density of the ATPase domain where it should be engaging nucleosomes.
While our resolution is not significantly better than the work of Zhang et al., our reconstruction is far more homogeneous and isotropic, and thus allowed use to unambiguously dock the structure of the nucleosome. Additionally, we performed cryo-EM where we are able to observe the nucleic acid of the nucleosome far better then what is possible in negative stain, which was the method used by Zhang et al. We also want to emphasize that the map by Zhang et al. suffers from severe flattening while ours does not.
While we are not certain of why we do not observe clear density for the ATPase domain, one possibility is that the stability of the interaction is reduced in the absence of extranucleosomal DNA. However, we believe the fact that we do see engagement with the nucleosome by the core and tail of RSC has added value, and propose that our structure describes an intermediate in the process of nucleosome engagement.
These observations raise the question as to whether the RSC-nucleosome complex formed is active for remodeling, and there is unfortunately no biochemical data to show whether this is the case or not.
We do not claim that our complex is active for remodeling. Instead we are suggesting that what we are observing is an intermediate in the process of nucleosome binding by RSC (Figure 4B).
We have changed two key words to indicate the speculative nature of our model.
It also was not clear as to why these nucleosomes were modified with lysines acetylated at residues 9, 14 and 18 and methylated at residue 4 of histone H3. Should these modifications change the interactions of RSC with nucleosomes, and if so shouldn’t there be an accompanying control of unmodified nucleosomes?
We decided to use these nucleosome as the acetylation marks have been shown to interact with RSC and its domains in numerous studies (see citation 24, 25, and 26 – more citations can be added). Our complex is therefore more representative of a physiological substrate compared to an entirely unmodified nucleosome.
It seems there should be more dialog as to why these particular histone modifications and potential supporting experimental data.
We can add the following statement:
“In order to shed light on the interaction of RSC with its substrate, we obtained a 19Å resolution map of RSC bound to a nucleosome core particle (NCP) modified with H3K4me3 and H3K(9/14/18)ac (Figure 1D, E, F; Figure 1—figure supplement 9). We used acetylated nucleosome because is has been shown that RSC has a higher affinity for H3 acetylated nucleosomes” (citations 24, 25, and 26 can be used as well as others).
The assumption that the structure of free RSC remains the same when bound to nucleosomes as used in this paper is too much of an extrapolation and suffers from the lack of data. There seems to be too much prediction and modeling used in this part, all because of the lack of resolution of the nucleosome-bound structure. The authors are left to make correlations based on previous data and are unable to truly move the field forward in terms of how RSC interacts with nucleosomes.
We disagree with this statement. While we do not have high resolution for the core of RSC in our RSC-NCP structure, we can still see that the overall shape remains the same. We can fit the core of RSC, the ARP module and NCP unambiguously into our RSC-NCP map.
We did use existing structures of the ARP and NCP and our newly determined RSC core structure for docking, and, as we stated, we were able to unambiguously place them within the lower resolution structure. For this reason, our model cannot consider “too much of an extrapolation.”
Lastly while we do not have high resolution for the regions that interact with the nucleosome, through the unambiguous docking of high-resolution structures of the pieces we are able to indicate what regions are interacting with the nucleosome (Sfh1, Rtt102 and Npl6) and which are not (the SWIRM, ZZ and SANT domains of Rsc8 – all of which have been predicted to bind DNA or nucleosomes).
Although the nucleosome has four acetylated lysines, there doesn’t seem to be any clear discernible structure showing the six bromodomains of RSC interacting with their likely targets. The authors also missed yet another earlier work showing the importance of RSC recruitment with acetylated histone H3 tails (see Chatterjee et al., 2011). I also found Figure 3 hard to read as the text is too small.
As we explain in the paper, the bromodomains are likely flexible with respect to the core. This is part of the reason we do not see them. We are also unlikely to see the bromodomains in the RSC-NCP complex, as the flexible bromodomains would be interacting with flexible histone tails, making them very flexible relative to the rest of the RSC-NCP complex.
We have cited Chatterjee et al. (see citation number 24).
We can make Figure 3 larger by moving panel c below a and b, and we will adjust the font size to improve legibility.
We have remade/rearranged the figure entirely. The figure captions are hopefully large enough now.
A structure of the RSC complex is presented. This reveals a tripartite core made up of the Sth1, RSC8, RSC9, Sfh1, Rsc6 subunits, which are conserved in evolution. As a result, the findings are likely to inform structure of human complexes that play important roles in a range of human diseases.
The structure refers to lower resolution negative stain structure of the SWI/SNF complex. From this it is concluded that SWI/SNF is likely to lack the tail lobe. As the tail lobe is not well resolved in RSC, an alternative is that it is dynamic in SWI/SNF and not resolved.
This is true. However, we can include 2D-class averages where we can see a tail density for RSC and not SWI/SNF. The reason we do not see the tail density in 3D is because we did not sort for it, and as it is sub-stoichiometric, it is only observable at much lower density thresholds. We can however do the sorting in 3D to show this.
The structure of the complex is used to infer the pathway by which it is assembled. It is incorrect to do this. No measurements have been made of the timing with which nascent proteins associate, and assembly intermediates predicted to arise in the absence of different subunits have not been identified. The structure defines interfaces that indicate how the complex is held together, not how it is assembled, and the manuscript should be re-written to reflect this.
We can change “The pattern of protein‐protein interactions observed in our structure provides insight into RSC assembly” to “Based on the observable protein-protein contacts in our structure of RSC, along with previous mass spectrometry analysis of the assembly of human BAF complexes, we propose a model for RSC assembly.”
We chose to build off the model that was proposed in a recent paper from the Kadoch lab (citation 16: Mashtalir, N. et al.). In their study, Mashtalir et al. did not investigate how nascent proteins associate, but they did identify “assembly intermediates predicted to arise in the absence of different subunits” by knocking out a subunit and analyzing the resulting complexes by mass spectrometry. What our work adds to theirs is subunit detail as we can see which subunits are interacting and suggest what their model would look like in atomic detail. This needs to be made clearer in the revised manuscript.
A low-resolution structure of RSC bound to a nucleosome is also provided. This should be submitted as a separate EMDB submission with statistics. Given the low resolution, it is highly speculative to describe interaction surfaces for components of the complex, such as bromodomains and the protein surfaces that contact nucleosomes. The nucleosome complex is used to propose a mechanism for remodelling involving, for example, interactions with histone acetylation marks. However, the manuscript adds nothing new in this respect as none of the interactions with acetylated histones are resolved.
We will be depositing the RSC-NCP map to the EMDB and can include an FSC curve to Figure 1—figure supplement 8. We do not state that we observed bromodomain density nor do we fit any bromodomains. What we do see is the ARP module and NCP, which we can dock unambiguously.
We are proposing a model of nucleosome engagement, not remodeling. What we are proposing is that RSC bind the nucleosome first through the acetylated tails and then through contacts with the NCP (such as the acidic patch).
We have added two key words to indicate the speculative nature of our model (in italics).
RSC could then engage the nucleosome…
The ATPase domain would then be able to translocate the DNA around…
The penultimate four paragraphs (prior to the final summary paragraph) are highly speculative. The two paragraphs before these incorrectly infer a pathway for the assembly of the complex.
As we have stated above, we can modify the opening statement of the paragraph discussing assembly to clarify where the model of assembly comes from (citation 16: Mashtalir, N. et al.).
The manuscript includes important new data describing how the core subunits of the RSC complex are organised, but a manuscript has been prepared with little attention to this strength.
We have significantly changed/expanded the content of our manuscript to include more discussion on the analysis of the structure. We now have dedicated sections for describing the head body and arm lobes of RSC and the roles each plays in RSC function.
Chromatin remodelers are essential protein complexes that modify the position of nucleosomes in chromatin. The SWI/SNF family of remodelers have been studied extensively over the past several decades and are of high biological importance. While low resolution reconstructions of SWI/SNF modelers have been determined, the only high-resolution data that exists for SWI/SNF remodelers are of the ATPase domain alone and bound to the nucleosome.
Here, Patel et al. used cryo-electron microscopy to determine the structure of apo RSC to near atomic resolution, as well as a low-resolution structure of RSC bound to the nucleosome. The overall structure can be subdivided into 5 distinct modules that include portions of 14 of its 17 subunits. Subunit assignments were verified by mass spectrometry crosslinking experiments. The main body of the structure is constructed around a Rsc8 dimer, and the authors put forth a model for how RSC assembles into a complex. Surprisingly, the ATPase domain is not resolved in either structure, suggesting that the ATPase-containing module is extremely flexible. The structure of RSC is a seminal achievement and will not only shed new insights on decades of past biochemical research but be informative for future studies on SWI/SNF remodeling.
Overall, there are no major issues that need to be addressed. I would recommend several minor changes to the manuscript.
SWRIM should be SWIRM.
We thank the reviewer for pointing out this typographic error, which will be corrected.
The authors introduce RSC as an important complex with broad functional applications, but do not mention that its specific function is to translocate DNA around the nucleosome until later in the text. For the sake of the general audience, it would be helpful if the authors briefly described what RSC does and what sets it apart from the other chromatin remodeling families.
A brief description of the function of RSC will be added.
There is no Figure 1E.
The reference to figures at this point should just state Figure 1D. We thank the reviewer for spotting this error.
Subsection “Structure of RSC and RSC-NCP”: “Our negative stain analysis of the yeast SWI/SNF complex shows that, like RSC, it has…. RSC features additional regulatory domains (Figure 1—figure supplement 2).” This point should be moved to where Rsc3 and Rsc30 are discussed so that it can be developed further there.
This can be done.
Figure 3D doesn’t exist but is referred to multiple times throughout the text.
This will be corrected, to Figure 3C.
Snf5 should be referred to as Sfh1 in the RSC complex.
We are referring to a domain of Sfh1 called Snf5. We see this can be confusing and can change it to clarify the nomenclature.
“Figure RSC-NCP2” Properly label.
This should be removed.
Epicypher nucleosomes often only have 147 bp of DNA wrapped around them. Do your nucleosomes have sufficient extranucleosomal DNA that this interaction would be possible?
While other studies have used nucleosome with extranucleosomal DNA, the ATPase domain binds to SHL+2 which would be within the 147 bp-region. However, we cannot fully exclude that the reason we do not observe the ATPase is because we are lacking this extra DNA.
Figure 3C: This figure needs to be reworked so that the RSC:NCP interactions are highlighted clearly. In the legend, the four contacts points should appear in the same order as they appear in the text. It would also be useful if they were numbered (i, ii, iii, etc) on the structure to easily locate the area being referred to.
These changed can be made.
Figure 1—figure supplement 3: Purple and grey volumes should be differentiated in the figure legend text.
A sentence stating that the purple classes indicate the classed that were selected for further processing will be added.
Figure 1—figure supplement 7: Figure needs a title.
The title “RSC core model validation” will be added.
Figure S5: 2D class averages should be included.
This will be added. Along with a micrograph.https://doi.org/10.7554/eLife.54449.sa2
- Eva Nogales
- Eva Nogales
- Eva Nogales
- Jeff Ranish
- Jeff Ranish
- Eva Nogales
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We thank S Baskaranm, H Asahara, and R Lesch for assistance with yeast work, A Iavarone for performing in-gel mass spectrometry data collection and analysis, P Grob, and D Toso for electron microscopy support, A Chintangal and P Tobias for computing support, and C Yoshioka and the OSHU Cryo-EM Facility for help with data collection. A portion of this research was supported by NIH grant U24GM129547 and performed at the PNCC at OHSU and accessed through EMSL (grid.436923.9), a DOE Office of Science User Facility sponsored by the Office of Biological and Environmental Research. This work was funded through NIGMS grants R01-GM63072 and R35-GM127018 to EN, and R01-GM110064 and R01-HL133678 to JR. EN is a Howard Hughes Medical Institute Investigator.
- Cynthia Wolberger, Johns Hopkins University School of Medicine, United States
© 2019, Patel et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Downloads (link to download the article as PDF)
Download citations (links to download the citations from this article in formats compatible with various reference manager tools)
Open citations (links to open the citations from this article in various online reference manager services)
Nanodiscs are membrane mimetics that consist of a protein belt surrounding a lipid bilayer, and are broadly used for characterization of membrane proteins. Here, we investigate the structure, dynamics and biophysical properties of two small nanodiscs, MSP1D1ΔH5 and ΔH4H5. We combine our SAXS and SANS experiments with molecular dynamics simulations and previously obtained NMR and EPR data to derive and validate a conformational ensemble that represents the structure and dynamics of the nanodisc. We find that it displays conformational heterogeneity with various elliptical shapes, and with substantial differences in lipid ordering in the centre and rim of the discs. Together, our results reconcile previous apparently conflicting observations about the shape of nanodiscs, and paves the way for future integrative studies of larger complex systems such as membrane proteins embedded in nanodiscs.
The molecular bases of heteromeric assembly and link between Na+ self-inhibition and protease-sensitivity in epithelial sodium channels (ENaCs) are not fully understood. Previously, we demonstrated that ENaC subunits – α, β, and γ – assemble in a counterclockwise configuration when viewed from outside the cell with the protease-sensitive GRIP domains in the periphery (Noreng et al., 2018). Here we describe the structure of ENaC resolved by cryo-electron microscopy at 3 Å. We find that a combination of precise domain arrangement and complementary hydrogen bonding network defines the subunit arrangement. Furthermore, we determined that the α subunit has a primary functional module consisting of the finger and GRIP domains. The module is bifurcated by the α2 helix dividing two distinct regulatory sites: Na+ and the inhibitory peptide. Removal of the inhibitory peptide perturbs the Na+ site via the α2 helix highlighting the critical role of the α2 helix in regulating ENaC function.