The conserved transcription coactivator SAGA is comprised of several modules that are involved in activator binding, TBP binding, histone acetylation (HAT) and deubiquitination (DUB). Crosslinking and mass spectrometry, together with genetic and biochemical analyses, were used to determine the molecular architecture of the SAGA‐TBP complex. We find that the SAGA Taf and Taf‐like subunits form a TFIID‐like core complex at the center of SAGA that makes extensive interactions with all other SAGA modules. SAGA‐TBP binding involves a network of interactions between subunits Spt3, Spt8, Spt20, and Spt7. The HAT and DUB modules are in close proximity, and the DUB module modestly stimulates HAT function. The large activator‐binding subunit Tra1 primarily connects to the TFIID‐like core via its FAT domain. These combined results were used to derive a model for the arrangement of the SAGA subunits and its interactions with TBP. Our results provide new insight into SAGA function in gene regulation, its structural similarity with TFIID, and functional interactions between the SAGA modules.
The molecular interactions between SAGA subunits and its modules are determined using a combination of crosslinking and mass spectrometry together with genetic and biochemical analysis.
The TAF subunits of SAGA form a central core module, resembling that of TFIID.
Spt7 and Spt20, two important structural subunits of SAGA, connect to distinct subunits and modules within the complex.
The DUB and HAT modules are positioned close to each other within SAGA and functionally interact.
Spt8 and Spt3 contact TBP at different locations and may simultaneously bind TBP.
Yeast SAGA and the closely related mammalian orthologs STAGA, PCAF, and TFTC are broadly conserved coactivator complexes that regulate the transcription of many inducible and developmentally regulated genes (Rodriguez‐Navarro, 2009; Samara & Wolberger, 2011; Helmlinger, 2012; Weake & Workman, 2012). In yeast, SAGA regulates ~ 10% of genes, with this gene set being highly enriched for genes induced by environmental stress and for genes with promoters containing a match to the consensus TATA element (Basehoar et al, 2004; Huisinga & Pugh, 2004). SAGA and related complexes are multifunctional, containing at least five separate activities: nucleosomal histone H3 acetyltransferase (HAT), histone H2B deubiquitinase (DUB), TBP binding, nucleosome binding, and activator binding. However, not all SAGA activities are required for transcription of every SAGA‐dependent gene. For example, less than half of SAGA‐dependent genes require the HAT activity for normal expression (Lee et al, 2000; Huisinga & Pugh, 2004). Similarly, the TBP‐binding function of SAGA is essential for TBP recruitment at some genes, while at other genes, SAGA functions at steps after formation of the complete RNA Pol II preinitiation complex (Larschan & Winston, 2001; Bhaumik & Green, 2002; Mohibullah & Hahn, 2008; Chen et al, 2012). These diverse activities allow SAGA to modulate transcription at many promoters with different coactivator requirements by regulating steps encompassing chromatin modification, stimulation of PIC assembly, and enhancing steps in early transcription elongation.
SAGA from Saccharomyces cerevisiae is a 1.8 MDa complex comprised of 19–20 subunits with its various activities separated into distinct structural modules. The activity and specificity of the HAT and DUB catalytic functions are dictated by non‐catalytic subunits within each module. For example, Ubp8, the subunit containing DUB activity, is inactive unless associated with the three other subunits of the DUB module (Köhler et al, 2008; Lang et al, 2011). The X‐ray structure of this module shows a complex intertwining of the DUB module subunits and provides an explanation for the interdependence of the subunits in DUB function (Köhler et al, 2010; Samara et al, 2010). Similarly, the substrate specificity of Gcn5 HAT activity is altered by its associated subunits, Ada2, Ada3, and Sgf29, to acetylate multiple residues within the nucleosome histone H3 tail (Grant et al, 1999; Balasubramanian et al, 2002; Bian et al, 2011).
Other groups of subunits also contribute to the disparate functions of SAGA. Both genetic and biochemical experiments have shown that the SAGA subunits Spt3 and Spt8 contact TBP (Eisenmann et al, 1992; Warfield et al, 2004; Sermwittayawong & Tan, 2006; Laprade et al, 2007; Mohibullah & Hahn, 2008); however, the mechanism of these interactions and whether both subunits can simultaneously bind TBP are unclear. SAGA shares a set of subunits with the coactivator TFIID, which likely play a structural role in SAGA architecture (Grant et al, 1998). Within TFIID, two copies each of the histone fold pair‐containing subunits Taf6‐Taf9 and Taf4‐Taf12, together with two copies of Taf5, form a symmetric complex (Bieniossek et al, 2013). This complex is bound by the Taf8‐Taf10 heterodimer, generating an asymmetric core structure upon which the other TFIID‐specific Tafs assemble. Likewise, SAGA contains a similar set of Tafs but with Taf4 and Taf8 replaced by the histone fold domain containing SAGA‐specific subunits Ada1 and Spt7 (Gangloff et al, 2000, 2001).
Because of its complex composition, we know relatively little about the molecular interactions between SAGA subunits, the overall arrangement of the different modules, and whether the modules functionally interact. For example, purification of SAGA via a tag on subunit Spt7 in strains with deletions of other SAGA subunits revealed complex interactions among the SAGA subunits (Wu & Winston, 2002). Similarly, purification of SAGA from strains containing TAF12 conditional mutations showed defects in association of subunits from different SAGA modules (Durso et al, 2001). A systematic mass spectrometry analysis of SAGA purified from strains lacking individual non‐essential SAGA subunits led to a model for the composition of each module and organization of the complex (Lee et al, 2011). However, this model differs in the proposed relative locations of the Tafs, Spt, and Tra1 subunits derived from an earlier model based on EM and immune localization of several subunits (Wu et al, 2004). Together, these initial findings suggest a series of complex interactions between different SAGA subunits and modules, perhaps somewhat analogous to the complex interactions between subunits of the DUB module.
To understand how the diverse functions of SAGA are coordinated to modulate gene expression, it is important to determine the overall architecture of SAGA, the arrangement of the different modules, and whether these modules interact. To approach this problem, we have used a combination of crosslinking and mass spectrometry, together with genetic and biochemical analysis, to determine molecular interactions between the SAGA subunits and its modules. Based on this work, we derive a new model for SAGA architecture with the TFIID‐like Taf complex at the center of SAGA. We also find functional interactions between the two enzymatic modules and new insight into the mechanisms of SAGA‐TBP binding.
Formation of the SAGA‐TBP complex
To determine the molecular architecture of the SAGA complex and its interactions with TBP, we first purified yeast SAGA using a TAP‐tag at the C‐terminus of subunit Spt7. This approach selectively purifies SAGA and not the related SLIK complex, which lacks the C‐terminus of Spt7 (Wu & Winston, 2002). Purified SAGA was visualized on a silver‐stained protein gel (Fig 1A, lane 3). Mass spectrometry analysis of this preparation identified all known SAGA subunits with the exception of Chd1 (Supplementary Table S1), a chromatin remodeler that reportedly associates with SAGA (Pray‐Grant et al, 2005). In agreement with most previous findings, TBP did not copurify with SAGA, indicating a weak association. As a first attempt to generate the SAGA‐TBP complex, we purified SAGA from a strain with the spt3‐401 allele, containing the Spt3 mutation E240K, which was suggested to bind TBP more stably compared to wild‐type SAGA (Laprade et al, 2007). Under our purification conditions, however, this SAGA derivative was not stably associated with TBP (Fig 1A, lane 5). As an alternative approach for finding conditions to generate the SAGA‐TBP complex, purified SAGA was immobilized to calmodulin beads via the calmodulin‐binding tag on Spt7. Immobilized SAGA was incubated with recombinant TBP for 90 min, washed, and then eluted from the beads using EGTA. Both wild‐type and the Spt3‐401 SAGA derivative specifically bound TBP (Fig 1A, lanes 4, 6, 7). Quantitation of the TBP/Ada1 ratio for the two preparations showed that the Spt3‐401 SAGA bound about fourfold more TBP compared to wild‐type SAGA (Fig 1B). Since wild‐type SAGA bound TBP under these conditions, TBP and wild‐type SAGA were combined in a 1:1.2 molar ratio and used as a substrate for subsequent crosslinking experiments.
Chemical crosslinking and mass spectrometry of SAGA
We next used a chemical crosslinking and mass spectrometry approach (Chen et al, 2010; Müller & Sinz, 2012) to map protein–protein interactions within SAGA. The SAGA‐TBP complex was crosslinked using the bifunctional crosslinking reagent BS3, digested with trypsin, and the peptides analyzed by mass spectrometry. The BS3 crosslinker reacts with accessible primary amine groups found in lysine side chains and at the N‐termini of polypeptides (Supplementary Fig S1A). Each SAGA subunit, except Sgf11, contains many lysine residues that are distributed throughout the polypeptides (Fig 2, Supplementary Fig S2), making SAGA a good substrate for this crosslinking strategy. A database search algorithm pLink (Yang et al, 2012) was used to identify the crosslinked peptides, and the identifications were then used to assemble site‐specific linkage maps of all the crosslinked residues between and within the SAGA subunits (Fig 2, Supplementary Fig S2).
SAGA‐TBP was crosslinked with either 2 or 5 mM BS3, and the reactions were independently analyzed by mass spectrometry for protein crosslinks. Results from both experiments were combined. Altogether, we identified 199 intermolecular crosslinks (between two different subunits) and 240 intramolecular crosslinks (within a single subunit) (Fig 2, Supplementary Figs S1C and S2, and Supplementary Table S2). To validate our crosslinking approach, we mapped the crosslinked lysine pairs within known structured domains of SAGA and TBP. BS3 has a theoretical maximum Cα‐Cα crosslinking distance of ~30 Å between two lysine residues (Kalkhof & Sinz, 2008; Mädler et al, 2009; Chen et al, 2010; Müller & Sinz, 2012; Merkley et al, 2014). Within domains for which high‐resolution structural information is available, 17 out of the 19 crosslinks were within the theoretical crosslinking distance for BS3 (Supplementary Table S3).
A TFIID‐like core complex at the center of SAGA
SAGA and TFIID share a subset of Tafs that are important for the function of both complexes. A recent study reported the cryo‐EM structure of a human TFIID subcomplex, comprised of two copies each of Tafs 4, 5, 6, 9, and 12 (Bieniossek et al, 2013). Addition of the Taf8‐Taf10 heterodimer to this complex generates an asymmetric TFIID core complex that acts as a scaffold to assemble the remaining TFIID subunits. All of the above‐mentioned Tafs are also subunits of SAGA, with the exceptions of Taf4 and Taf8. The SAGA subunit Ada1 has a histone fold domain homologous to that of Taf4 and forms a histone fold pair with Taf12 (Gangloff et al, 2000), while Spt7 has a histone fold domain homologous to Taf8 and pairs with Taf10 (Gangloff et al, 2001). Indeed, we observed crosslinks between Taf12 and Ada1, and between Taf10 and Spt7, close to the histone fold domains (Fig 2), indicating that these proteins likely form heterodimers through histone fold pairing within the SAGA complex. We also observed crosslinks between all of the other Taf subunits except for Taf6 and Taf9, which are known to form a histone fold pair within TFIID (Bieniossek et al, 2013). Together with previous findings on the TFIID core complex, our combined results suggest that SAGA contains a TFIID‐like core. As described in more detail below, this TFIID‐like core complex crosslinks to all other SAGA modules (Fig 3A). Thus, similar to its role in TFIID, this TFIID‐like core is likely positioned at the center of SAGA where it functions to nucleate assembly of other SAGA modules.
Subunits Spt7 and Spt20 crosslink to distinct SAGA modules
The SAGA subunits Spt7, Spt20, and Ada1 were initially proposed to be components of a core module (Wu & Winston, 2002), as mutation of any of these subunits disrupted conventional purification of SAGA (Sterner et al, 1999). This view was supported by systematic deletion analysis, suggesting that all Spt subunits were part of a common module (Lee et al, 2011). In contrast, mapping Spt7 and Spt20 within the SAGA EM structure led to an alternative view, suggesting that these two subunits are positioned at different ends of a large structural module (Wu et al, 2004). Consistent with this latter view, our crosslinking results suggest distinct roles for these two subunits in SAGA assembly as Spt7 and Spt20 crosslink to different sets of subunits but do not crosslink to each other.
We found that Spt7, part of the TFIID‐like module, crosslinks to Spt8, one of the TBP‐binding subunits (Fig 3B). This interaction occurs primarily between the C‐terminus of Spt7 and WD40 repeats at the center of Spt8. This is in perfect agreement with previous biochemical results, which found that the 152 C‐terminal residues of Spt7 are required for Spt8 binding to SAGA (Wu & Winston, 2002). Spt7 also crosslinks to Gcn5 and Ada3, showing close positioning of Spt7 to the HAT module. In contrast, Spt20 crosslinks to neither Spt7, Spt8, nor the HAT module (Fig 3C). Instead, we observe extensive crosslinking between Spt20 and the TFIID‐like core domain, the DUB module, and the TBP‐binding subunit Spt3.
Tra1, an activator‐binding subunit of both SAGA and the coactivator NuA4, crosslinks to Spt20 at several sites within its FAT domain. The FAT domain also crosslinks to TAF12, in a region of Taf12 that also crosslinks to Spt20. Thus, Tra1 seems to primarily associate with SAGA via interactions between its FAT domain and segments of Taf12 and Spt20 that are in close proximity (Figs 2 and 3). While the sparse Tra1‐SAGA crosslinking outside of the FAT domain does not rule out other Tra1 interactions, our observation that the major Tra1‐SAGA crosslinking is localized within the relatively small Tra1 FAT domain, together with our findings that the FAT domain crosslinks with two closely positioned SAGA subunits, suggests a peripheral position for the large Tra1 subunit within the SAGA complex, in agreement with a previous model (Wu et al, 2004).
A network of interactions between TBP and four SAGA subunits
Crosslinking of the SAGA‐TBP complex revealed four subunits positioned close to TBP: Spt3, 8, 7, and 20 (Figs 2 and 4A). While both genetic and biochemical studies have implicated Spt3 and Spt8 in TBP binding, Spt8 is the only subunit known to bind to TBP in the absence of other SAGA components (Warfield et al, 2004; Sermwittayawong & Tan, 2006). While the crosslinks we observe between Spt3, Spt8, and TBP are consistent with previous findings, the additional crosslinks of TBP with Spt20 and Spt7 suggest that these subunits may function to facilitate TBP binding. The lysine residues of TBP that crosslink to Spt3 and Spt8 are located on opposite ends of the TBP surface (Fig 4B). Previous studies using the photocrosslinker BPA (ρ‐benzoyl‐phenylalanine) positioned on TBP identified two TBP surfaces that are close to Spt3 and Spt8 in the preinitiation complex (Mohibullah & Hahn, 2008). Consistent with this, we found that the TBP lysine residues that crosslink to Spt3 or Spt8 are close to the previously identified BPA crosslinks, further supporting the existence of two separate SAGA binding interfaces on TBP. The region of TBP that crosslinks to Spt3 also agrees with the location of mutations in TBP that suppress Spt3 mutations (Fig 4B) (Eisenmann et al, 1992; Laprade et al, 2007).
Spt3 contains both halves of an interacting histone fold pair, located at the N and C‐terminal ends of the protein (Birck et al, 1998). The spt3‐401 allele (E240K), a suppressor of a TBP mutation in the Spt3‐binding surface, lies within the C‐terminal histone fold and implicates this region in TBP binding (Fig 4A) (Laprade et al, 2007). Consistent with this, we find that Spt3 lysine residue 190, located at one end of the C‐terminal histone fold, crosslinks to the C‐terminal half of TBP. Spt3 K190 also crosslinks to two residues within the disordered and non‐conserved TBP N‐terminal region. This latter interaction is probably not functionally important, as the N‐terminal region of TBP can be deleted with no obvious phenotype (Reddy & Hahn, 1991). Interestingly, the C‐terminal region of Spt20 crosslinks to both ends of the Spt3 C‐terminal histone fold and to the opposite side of TBP near the Spt8‐interacting surface (Fig 4A and B). Since Spt3 alone cannot stably bind TBP, our findings suggest that Spt20 may play a complex role in SAGA‐TBP binding by modulating the Spt3‐TBP interaction and possibly by direct interactions with TBP.
TBP residues K133 and K138 crosslink to the Spt8 WD40 repeats and are located in a conserved and highly positively charged groove on the convex surface of TBP (Fig 4B). These two lysine residues, together with K145 on TBP helix 2, are essential for interactions within Pol I, Pol II, and Pol III preinitiation complexes. For example, Mot1, the Taf1 TAND2 domain, and TFIIA all interact with this positively charged groove (Buratowski & Zhou, 1992; Kim & Roeder, 1994; Cui et al, 2011; Anandapadamanaban et al, 2013). Both the Taf1 TAND2 domain and Spt8 compete with TFIIA for binding TBP (Kokubo et al, 1998; Bagby et al, 2000; Warfield et al, 2004), suggesting that Spt8 might also bind TBP via the positively charged groove. To test this model, we introduced a double leucine substitution to TBP (K133, 138L) that is defective in forming the TBP‐Mot1‐DNA complex (Auble & Hahn, 1993) and in interacting with TFIIA (Buratowski & Zhou, 1992). These two lysine residues are both crosslinked to Spt8 by BS3 (Fig 4A and B). To test the effect of these TBP mutations on Spt8 binding, we performed a pull‐down assay where His‐tagged wild‐type or K133,138L TBP were immobilized on Ni Sepharose resin and then incubated with GST or GST‐tagged Spt8 (Fig 4C). Our results showed that the K133,138L mutation reduced TBP‐Spt8 binding by twofold (Fig 4C). Taken together, our crosslinking and biochemical analyses suggest that Spt8 physically contacts the positively charged groove on TBP and that this interaction contributes to TBP‐Spt8 binding.
Finally, we found that the C‐terminal region of Spt7 crosslinks to both Spt8 and TBP (Fig 4A), suggesting that this region of Spt7 may be directly involved in TBP binding. Although recombinant Spt8 alone can bind TBP (Fig 4C), Spt7 may modulate this interaction. Interestingly, both Spt7 and Spt20 crosslink to nearby positions on the TBP N‐terminal stirrup where they could potentially interact (Fig 4B). Thus, our crosslinking analysis suggests that Spt3, Spt20, Spt8, and Spt7 have a complex network of interactions with TBP.
A network of interactions linking SAGA with the HAT and DUB modules
Both the HAT and DUB modules function to enhance transcription (Weake & Workman, 2010). Recruitment of SAGA to promoters allows targeted acetylation of the H3 tail by Gcn5, promoting the binding of other regulatory factors and generating a permissive environment for chromatin remodeling. The DUB module removes ubiquitin from H2B, a histone modification that transiently increases at the 5′ end of transcribed regions upon gene induction and is linked to transcription elongation. Ubiquitin removal allows subsequent recruitment of the CTD Ser2 kinase CTK1 to coding regions (Wyce et al, 2007), promoting efficient transcription elongation and the placement of additional downstream chromatin marks. A mutant Poly Q‐containing form of the DUB module subunit Ataxin7 (the human ortholog of yeast Sgf73) was shown to inhibit HAT function (McMahon et al, 2005; Palhan et al, 2005), and Gcn5 can enhance the stability of the DUB module in the related mammalian complex STAGA (Atanassov et al, 2009).
Our crosslinking data showed a surprising network of interactions between the HAT, DUB and TFIID‐like core modules (Figs 2 and 3D, and Supplementary Fig S3). Within the DUB module, the C‐terminal region of Sgf73 makes extensive crosslinks with the TFIID‐like core, Spt20, and the HAT module subunits Ada2 and Ada3. This agrees with previous results showing that Sgf73 is responsible for linking the DUB module to SAGA (Köhler et al, 2008; Lee et al, 2009). The HAT module subunit Ada3 crosslinks to the TFIID‐like core through Taf5, Taf6, Spt7, and Ada1. Together, our results demonstrate that these two enzymatic modules are positioned near each other and raise the possibility that they functionally interact.
We next determined the regions of Sgf73, Ada2, and Ada3 required for integrity of each module and their association with SAGA using the crosslinks as a guide to test specific regions of these subunits. Internal deletions were designed within subunits Sgf73, Ada2, and Ada3, and complex integrity was monitored by immunoprecipitation (IP) experiments (Fig 5). Within Sgf73, mutations were generated between residues 196–575 in segments that crosslink to Spt20, Ada2, Ada3, or the TFIID‐like core (Fig 5A and D, and Supplementary Fig S3A and B). An N‐terminal Sgf73 deletion (Δ2–104) that was shown to block association of the other three DUB subunits (Köhler et al, 2008) was also included in the analysis. Whole cell extracts containing the Sgf73 derivatives were assayed by IP using either a 3× Flag tag at the C‐terminus of Sgf73 or via the Spt7 TAP‐tag (Fig 5A and D). Our results showed that deletion of Sgf73 residues 196–314 (Δ1; encompassing the nucleosome‐binding SCA7 domain) or residues at the C‐terminus (401–479 and 535–575; Δ2B and Δ3) had no effect on integrity of SAGA or the DUB module (Fig 5D, lanes 3, 6, 7). In contrast, deletion of the Sgf73 N‐terminus (Δ4) caused dissociation of Ubp8 as expected (Fig 5D, lane 10). Deletion of Sgf73 residues 350–400 (Δ2A) had the most severe effect, resulting in release of Sgf73 and the DUB module from SAGA, as Sgf73 failed to copurify with the TFIID‐like core and the HAT modules (Fig 5D, lane 5). In a yeast growth assay, this latter mutant had a slow growth rate similar to a strain with a complete deletion of SGF73 (Supplementary Fig S4A). While this important segment overlaps with the region of Sgf73 that crosslinks to the HAT module, Spt20, and the TFIID‐like core, deletion of adjacent regions in Sgf73 that also crosslink to these subunits had no effect on DUB‐SAGA association. We speculate that these adjacent regions of Sgf73 are in close proximity to other SAGA subunits but do not make direct protein–protein contacts.
Using the same strategy, we designed internal deletions surrounding the crosslinked lysine residues within HAT subunits Ada2 and Ada3 (Fig 5B and C). For Ada2, the IP assays of these mutant proteins agreed with the crosslinks and showed that the regions required for association with Ada3 and the TFIID‐like core precisely overlap. Ada2 deletion Δ1, removing residues 232–317, retained association with Gcn5 but was released from its interactions with Ada3 and the TFIID‐like core (Fig 5E, lane 3). This strain has a slow growth phenotype, equivalent to deletion of the ADA2 gene (Supplementary Fig S4B). Deletion of Ada2 residues 353–434 (Δ2) retained association with other HAT module subunits, but its binding to the TFIID‐like core was diminished (Fig 5E, lane 4). This mutation has only a slight slow growth phenotype (Supplementary Fig S4B), suggesting that the HAT‐SAGA interaction is not completely disrupted in vivo. Within Ada3, deletion of residues 334–425 (Δ1) and 483–534 (Δ2A) showed similar biochemical defects with an intact HAT module that was released from the TFIID‐like core (Fig 5F, lanes 3, 5). In contrast, deletion of Ada3 residues 535–616 (Δ2B) disrupted the entire HAT module (Fig 5F, lane 6). All these mutant strains showed a slow growth phenotype (Supplementary Fig S4C). Interestingly, Ada3 deletion Δ3 (residues 640–697) was not affected for SAGA or HAT stability (Fig 5F, lane 7) but still showed a slow growth phenotype equivalent to deletion of the ADA3 gene (Supplementary Fig S4C). This suggests that the Ada3 C‐terminus may play a role in modulating Gcn5 activity, perhaps related to its interactions with the TFIID‐like core and Spt7.
While the IP assays did not reveal SAGA‐HAT association defects resulting from mutations within the DUB module or vice versa, the fact that these two modules are in close proximity raised the possibility of a functional interaction. To examine whether one of these enzymatic modules could affect the other, we first measured subunit association in the mutant strains using a more stringent assay. Three mutations were selected based on the above IP experiments: (i) Sgf73 Δ4, which results in specific dissociation of DUB module subunits but leaves Sgf73‐SAGA association intact, (ii) Sgf73 Δ2A, which results in release of Sgf73 and its associated DUB module subunits from SAGA, and (iii) Ada3 Δ2A, which results in release of the SAGA‐HAT module.
SAGA was TAP‐tag purified from each of these mutant strains, and Fig 6A shows a representative silver‐stained gel of the preparations. The purified complexes were digested in‐solution with trypsin and analyzed by mass spectrometry. The iBAQ (intensity‐based absolute quantification) values (Schwanhäusser et al, 2011) were calculated using MaxQuant (Cox & Mann, 2008) to determine the relative abundance of each SAGA subunit in the purified complexes. The iBAQ value for each individual subunit was then normalized to the value of Spt7, because all complexes were purified through the tagged Spt7 subunit (Fig 6B, Supplementary Table S4A).
For Sgf73 Δ4, the MS analysis confirmed that SAGA lacked the DUB subunits Ubp8, Sgf11, and Sus1, while retaining wild‐type levels of Sgf73 (Fig 6B). As expected from the IP results, the more severe mutation Sgf73 Δ2A caused dissociation of all DUB associated subunits including Sgf73 (Fig 6B). In contrast, mutation of the HAT module subunit Ada3 (Δ2A) had no effect on the stability of the DUB module apart from a surprising 70% decrease in association of Sgf11 (Fig 6B). Finally, we found that both Sgf73 mutations affected the stability of the HAT module, consistent with crosslinks observed between Sgf73 and HAT subunits. SAGA purified from the mutants Sgf73 Δ2A and Δ4 showed a consistent 15–63% decrease in the amount of the associated HAT subunits (Fig 6B). We obtained similar results by analyzing the same data set and calculating the adjusted normalized spectral abundance factor (adjNSAF) using Abacus (Fermin et al, 2011) (Supplementary Table S4B). Together, our analysis of subunit composition in the HAT and DUB mutants suggests that the DUB module plays a modest role in stabilizing association of the HAT module with SAGA.
Mutations in the DUB module subunit Sgf73 affect SAGA nucleosomal HAT activity
Since the above analysis showed an interaction between the DUB and HAT modules, we tested whether mutations in Sfg73 affect HAT function using both recombinant histone H3 and mono‐nucleosome substrates. The amounts of purified wild‐type and mutant SAGA added to the HAT assays were normalized based on the amounts of added Gcn5. Histone H3 acetylation was monitored by Western blot assay using an antibody against acetylated H3K9. Unexpectedly, we observed that the two Sgf73 mutations have opposite effects on HAT activity.
SAGA purified from the more severe Sgf73 Δ2A mutation had HAT activity equivalent to wild‐type SAGA using the histone H3 substrate, but showed about fourfold higher acetylation of the mono‐nucleosome substrate after 30 min of incubation (Fig 6C, Supplementary Fig S5C). In contrast, SAGA purified from the less severe Sgf73 Δ4 mutation shows about twofold lower activity compared to wild‐type SAGA on both the H3 and mono‐nucleosome substrates (Fig 6D, Supplementary Fig S5C). Since this latter mutation only affects stability of the DUB module, it suggests that the DUB modestly enhances enzymatic activity of the HAT module. Because Sgf73 makes numerous protein–protein interactions with other SAGA modules, complete dissociation of Sgf73 in the Δ2A mutant may indirectly affect SAGA‐HAT activity due to changes in other SAGA‐HAT interactions. Our combined analysis of SAGA indicates a close proximity between the DUB and HAT modules and that the DUB module modestly enhances the enzymatic activity of the HAT module.
SAGA is a multifunctional transcription coactivator that preferentially acts at inducible promoters where its multiple activities stimulate steps during transcription initiation and early elongation. The complexity of SAGA has led to several different models for the organization of its subunits and the architecture of the complex. To resolve these differences, we used an integrated approach to define molecular interactions between subunits and to determine how the different SAGA modules fit into the complete complex. Our work has led to a model for SAGA architecture, identification of molecular interactions between the different modules and new insight into the mechanisms and regulation of SAGA‐TBP binding.
Previous studies revealed that SAGA contains a subset of the Taf subunits from the coactivator TFIID and two SAGA‐specific subunits (Ada1 and Spt7) with Taf‐like histone fold domains. Within TFIID, this set of Taf proteins forms an asymmetric core complex at the center of TFIID, which acts as a scaffold for assembly of the other Tafs. Our results are consistent with an analogous structure at the center of SAGA. First, the SAGA Taf and Taf‐like subunits crosslink to each other as expected from an analogous complex within SAGA. Second, previous studies showed disruption of the SAGA complex upon inactivation of the SAGA Taf and Taf‐like proteins. Our crosslinking data suggest that these subunits are at the center of SAGA, as they are a unique group of subunits that crosslink to all other SAGA modules. Within the TFIID‐like core, SAGA‐specific regions within Spt7 and Ada1 outside of their histone fold domains play an important role in bridging other structural modules with the TFIID‐like core.
A Model for the architecture of SAGA
Figure 7 (see also Supplementary Movie S1) shows a model for the architecture of SAGA with the TFIID‐like structure at the center. Based on the proposed arrangement of the TFIID subunits, the central core of the SAGA model contains two copies each of the histone fold containing pairs Taf6‐Taf9, and Ada1‐Taf12 together with two copies of Taf5. This core structure contains extra SAGA‐specific density (not shown) since Ada1 has over 170 residues both N‐ and C‐terminal to its histone fold domain that are involved in interactions with other SAGA subunits. One copy of Spt7‐Taf10 is predicted to bind this complex, analogous to the binding of Taf8‐Taf10 in TFIID. Spt7 has a broad interface with this structure, as it crosslinks to the widely separated Taf5 N and C‐terminal domains. Taf10, which requires only a 91‐residue segment containing the histone fold domain for function (Kirschner et al, 2002), crosslinks to the histone fold domain of Taf12 in the lower half of the core.
Our crosslinking also reveals the interactions this TFIID‐like core makes with the other SAGA modules and more peripheral subunits. For example, in addition to its crosslinks with the Taf5 N‐terminal domain, the C‐terminus of Spt7 also crosslinks to Spt8 WD40 repeats, likely positioning Spt8 in the lower half of the TFIID‐like core. Sgf73 links the DUB module to the TFIID‐like core, demonstrated by its extensive crosslinks with the histone fold domains of Ada1 and Taf12 as well as the Taf5 N‐terminus. These crosslinks position the DUB module on the lower half of the TFIID‐like core. Spt20, which crosslinks to the DUB module, Tra1, and Spt3, displays numerous crosslinks with the histone fold domains of Ada1, Taf9, Taf10, Taf12, and the N‐terminal domain of Taf5. These results locate Spt20 in the center of the TFIID‐like complex and explain how Spt20 and Sgf73 can extensively crosslink. Ada3, the HAT module subunit that links this module to SAGA, crosslinks to the histone fold domains of Ada1, and Taf12, the WD40 repeats in Taf5 and near the HEAT repeat of Taf6. This suggests that the HAT module is situated in the central and upper half of the TFIID‐like core and in a position to crosslink with the C‐terminus of Sgf73. The TBP‐binding subunit, Spt3, crosslinks extensively to the same region of Spt20 that makes extensive crosslinks to the Taf12 histone fold domain and the Taf5 N‐terminus. This maps Spt3 to the lower half of the TFIID‐like complex, and potentially situated so that both Spt8 and Spt3 can simultaneously interact with TBP.
Wu et al reported an EM model of the SAGA complex, in which they mapped the positions of several SAGA subunits through immune labeling and electron microscopic analysis (Wu et al, 2004). Most of these mapping studies are in good agreement with our crosslinking results, with TAFs in a central position, Spt20 connecting with Spt3, and Spt7 close to Gcn5. The only discrepancy is their mapping of Tra1 and Spt20 at opposite ends of SAGA, in contrast to our crosslinking, which shows Tra1 and Spt20 in close proximity. We reason that this difference is likely due to the extended structure of Tra1, as Wu et al used Myc‐antibody to probe the N‐terminus of Tra1, and we observed Spt20 crosslinks near the Tra1 C‐terminus (Fig 2). Our model is also distinct from another model, based on combinatorial depletion analysis, which linked all Spt subunits, Ada1 and Tra1 in a single group and positioned the Taf subunits in a more peripheral part of the protein–protein interaction network (Lee et al, 2011).
A surprisingly small interface between Tra1 and SAGA
Crosslinks between SAGA and the large activator‐binding subunit Tra1 are primarily between the Tra1 FAT domain and a region of Taf12 that is N‐terminal to the histone fold domain. While the absence of other crosslinks does not rule out additional Tra1‐SAGA interactions, this limited region of crosslinking suggests a peripheral position for the majority of Tra1, perhaps on the opposite side of the TFIID‐like core from the other modules. Interestingly, the FAT domain also crosslinks to a small segment of Sgf73, which in turn interacts with the HAT module subunits Ada2, Ada3, and Gcn5. It was previously shown that a small deletion within the Tra1 FAT domain caused inhibition of HAT activity (Knutson & Hahn, 2011). From our crosslinking results, this inhibition is likely an indirect effect. One explanation for this inhibitory effect is that the FAT domain mutation alters its interaction with Sgf73, causing an aberrant interaction between Sgf73 and the HAT module. It is also interesting that many small deletions and other mutations throughout Tra1 cause dissociation of Tra1 from other SAGA components (Knutson & Hahn, 2011). Given the domain topology derived from the crosslinking results, most of these mutations likely affect the FAT domain‐SAGA interaction. Intramolecular Tra1 crosslinks show interactions between the FAT and HEAT repeat domains as well as interactions between the Tra1 PI3K, FRB, and FAT domains (Supplementary Fig S2). We propose that disrupting these intramolecular FAT domain interactions alters the conformation of the FAT domain and disrupts Tra1‐SAGA contacts.
The two TBP‐binding subunits, Spt3 and Spt8, play complex roles in gene regulation and, depending on the regulated gene, can have either positive or negative effects on transcription (Bhaumik, 2011). For example, deletion of Spt8 causes an increase in basal transcription at some genes (Belotserkovskaya et al, 2000) and Spt8 can compete with TFIIA for binding to TBP in vitro (Warfield et al, 2004). Our crosslinking and biochemical results provide an explanation for this competition, as both the unstructured region of the TFIIA large subunit and Spt8 bind a positively charged groove on TBP. Since DNA was found to compete with Spt8 for TBP binding (Sermwittayawong & Tan, 2006), it is likely that the isolated Spt8 subunit additionally interacts with TBP near its DNA‐binding surface. Although Spt3‐TBP binding is genetically the most important SAGA‐TBP interaction for gene activation, Spt3 alone cannot bind TBP. The crosslinking data show that Spt20 is positioned either side of the Spt3 histone fold domain involved in TBP binding. This finding suggests that Spt20 may enhance Spt3‐TBP binding by promoting a conformation of Spt3 that is active for TBP binding. Our combined results show a complex network of interactions between TBP and Spt3, 7, 8, and 20.
SAGA DUB‐HAT interactions
The human DUB module subunit Ataxin7 (orthologous to yeast Sgf73) contains a region that can undergo poly‐glutamine expansion, causing the disease spinocerebellar ataxia type 7. Human Ataxin7 can partially complement a SGF73 deletion, but the Poly Q form inhibits HAT activity in vivo and in vitro (McMahon et al, 2005; Palhan et al, 2005; Burke et al, 2013). Here, we have shown that mutations in the DUB module subunit Sgf73 have modest effects on SAGA nucleosomal HAT activity and stability of the HAT module within SAGA. This relationship is consistent with our observed protein crosslinking as well as inhibition of HAT activity by the mutant form of Ataxin7.
Taken together, our results have led to a model for SAGA architecture and revealed new and unexpected interactions between the SAGA subunits and modules. Our model reveals insights into how the disparate functions of SAGA are coordinated to activate a wide range of genes with different coactivator requirements and suggest new directions for understanding the molecular interactions of SAGA with transcription regulators and components of the transcription machinery.
Materials and Methods
See Supplementary Methods for strains and plasmids used, methods of protein purification, HAT and IP assays and structure modeling methods.
Wild‐type SAGA complex was mixed with 20% molar excess of recombinant TBP. 50 μg of this complex were incubated with 2 mM or 5 mM BS3 (Thermo Scientific) in 200 μl total volume at room temperature for 2 h. Reactions were quenched with 10 μl 1 M Tris pH 7.5. Small‐scale titrations were performed with ~3 μg of SAGA complex in 20 μl total volume with various amount of BS3 as shown in Supplementary Fig S1B. Reactions were analyzed by SDS‐PAGE and silver stain (Invitrogen).
MS sample preparation and data analysis
Protein samples were denatured in 50% trifluoroethanol (TFE) at 60°C for 30 min and reduced by addition of 5 mM TCEP at 37°C for 30 min. The samples were then alkylated in the dark at room temperature with 10 mM iodoacetamide for 30 min. The samples were diluted tenfold with 100 mM ammonium bicarbonate and digested with trypsin (3:50 w/w) overnight at 37°C. Tryptic peptides were then purified by C18 chromatography (Waters) and dried in a speedvac. For crosslinked samples, dried peptides were resuspended in 30 μl buffer A (25 mM ammonium formate, 20% acetonitrile [ACN], pH 2.8) and then fractionated by microcapillary partisphere strong cation exchange (SCX, 5 μm, 200 Å, GE Healthcare). For separation of crosslinked peptides by HPLC, peptides were loaded onto the capillary column equilibrated in 5% ACN/0.1% trifluoroacetic acid (TFA) and washed with 20% ACN/0.1% formic acid (FA). Bound peptides were eluted with 50 μl of buffer A with 10, 30, 50, 70 and 90% buffer B (800 mM ammonium formate, 20% ACN, pH 2.8), followed by 50 μl 100% buffer B and 50 μl buffer C (1 M ammonium acetate, 10% ACN, 0.1% FA, pH 8). All fractions were then dried in a speedvac. Most crosslinked peptides were eluted in fractions with 50–90% buffer B.
All dried peptides were analyzed on a Thermo Scientific Orbitrap Elite at the Proteomics facility at the Fred Hutchinson Cancer Research Center (FHCRC, Seattle, Washington, USA) with HCD fragmentation and serial MS events that included one FTMS1 event at 30,000 resolution followed by 10 FTMS2 events at 15,000 resolution. Other instrument settings included: Charge state rejection: +1, +2, +3 and unassigned charges; Monoisotopic precursor selection enabled; Dynamic exclusion enabled: repeat count 1, exclusion list size 500, exclusion duration 30 s; HCD normalized collision energy 35%, isolation width 3 Da, minimum signal count 5,000; FTMS MSn AGC target 50,000, FTMS MSn Max ion time 250 ms. HPLC uses a 90 min gradient from 5% ACN to 40% ACN for crosslinked fractions and a 60 min gradient from 5% ACN to 35% ACN for normal non‐crosslinked peptides.
Crosslinking data were analyzed using pLink (Yang et al, 2012) with default settings (precursor monoisotopic mass tolerance: ± 10 ppm; fragment mass tolerance: ± 20 ppm; up to 4 isotopic peaks; max evalue 1; static modification on cysteines; 57.0215 Da; differential oxidation modification on methionines; 15.9949 Da) against a database containing 22 proteins (21 SAGA subunits and TBP), and the search engine outputs results with 5% FPR. Each spectrum was manually evaluated for the quality of the match to each peptide using the COMET/Lorikeet Spectrum Viewer (Trans‐Proteomic Pipeline, TPP) (Supplementary Fig S6). The crosslinked peptides were considered confidently identified if at least 3 consecutive b or y ions for each peptide were observed with minimum peptide length ≥ 4 and the majority of the observed ions are accounted for. Fourteen percent of spectra were removed after manual inspection. For iBAQ analysis, MS raw files were loaded to MaxQuant (Cox & Mann, 2008) (version 126.96.36.199) and searched against the yeast whole genome database with deletions in Sgf73 or Ada3 under default settings with the following exceptions: group‐specific parameters tab, multiplicity = 1, enzyme = trypsin; identification & quantification tab, Min. ration count = 1; Misc. tab, check re‐quantify and keep low‐scoring versions of identified peptides only within parameter groups, check match between runs, check label‐free quantification, LFQ min. ratio count = 1, uncheck Fast LFQ, check iBAQ and log fit, check calculate peak properties. Alternatively, MS data were first searched against the yeast whole genome database using X!Tandem on the computer cluster at the Institute for Systems Biology (ISB, Seattle, Washington, USA). The search results were then analyzed using PeptideProphet and ProteinProphet in TPP on the computer cluster at ISB. Adjusted NSAF (adjNSAF) values were calculated using Abacus (Fermin et al, 2011) with default parameters.
All authors designed the research, analyzed the data, and wrote the paper. YH and JL performed the research.
Conflict of interest
The authors declare that they have no conflict of interest.
Supplementary Figure S1
Supplementary Figure S2
Supplementary Figure S3
Supplementary Figure S4
Supplementary Figure S5
Supplementary Figure S6
Supplementary Table S1
Supplementary Table S2
Supplementary Table S3
Supplementary Table S4
Supplementary Table S5
Supplementary Movie S1
Source Data for Supplementary Figure S1B
Source Data for Supplementary Figure S5B
Source Data for Supplementary Figure S5C
We thank Steve Henikoff (FHCRC) for mono‐nucleosomes, Toshio Tsukiyama (FHCRC) for H3K14, K23 acetylation antibodies, Patrick Schultz (IGBMC, Strasbourg, France) for sharing the coordinates of the human TFIID core model and Phil Gafken and the FHCRC proteomics resource for sample analysis. We also thank the members of the Hahn Lab (FHCRC) and the Ranish Lab (ISB) for help throughout the course of this project and B. Knutson and S. Grünberg for comments on the manuscript. This work was supported by NIH grants 2RO1GM057114 to SH and 2P50GM076547 and R21CA175849 to JR.
FundingNIH 2RO1GM057114 2P50GM076547 R21CA175849
- © 2014 The Authors