The structure of the NuA4–Tip60 complex reveals the mechanism and importance of long-range chromatin modification

Histone acetylation regulates most DNA transactions and is dynamically controlled by highly conserved enzymes. The only essential histone acetyltransferase (HAT) in yeast, Esa1, is part of the 1-MDa NuA4 complex, which plays pivotal roles in both transcription and DNA-damage repair. NuA4 has the unique capacity to acetylate histone targets located several nucleosomes away from its recruitment site. Neither the molecular mechanism of this activity nor its physiological importance are known. Here we report the structure of the Pichia pastoris NuA4 complex, with its core resolved at 3.4-Å resolution. Three subunits, Epl1, Eaf1 and Swc4, intertwine to form a stable platform that coordinates all other modules. The HAT module is firmly anchored into the core while retaining the ability to stretch out over a long distance. We provide structural, biochemical and genetic evidence that an unfolded linker region of the Epl1 subunit is critical for this long-range activity. Specifically, shortening the Epl1 linker causes severe growth defects and reduced H4 acetylation levels over broad chromatin regions in fission yeast. Our work lays the foundations for a mechanistic understanding of NuA4’s regulatory role and elucidates how its essential long-range activity is attained. Here, the authors use cryo-EM, biochemical and yeast assays of the HAT NuA4–Tip60 to reveal its mechanism of acetylating distant nucleosomes through the Epl1 linker establishing long-range chromatin interactions.

Histone acetylation regulates most DNA transactions and is dynamically controlled by highly conserved enzymes. The only essential histone acetyltransferase (HAT) in yeast, Esa1, is part of the 1-MDa NuA4 complex, which plays pivotal roles in both transcription and DNA-damage repair. NuA4 has the unique capacity to acetylate histone targets located several nucleosomes away from its recruitment site. Neither the molecular mechanism of this activity nor its physiological importance are known. Here we report the structure of the Pichia pastoris NuA4 complex, with its core resolved at 3.4-Å resolution. Three subunits, Epl1, Eaf1 and Swc4, intertwine to form a stable platform that coordinates all other modules. The HAT module is firmly anchored into the core while retaining the ability to stretch out over a long distance. We provide structural, biochemical and genetic evidence that an unfolded linker region of the Epl1 subunit is critical for this long-range activity. Specifically, shortening the Epl1 linker causes severe growth defects and reduced H4 acetylation levels over broad chromatin regions in fission yeast. Our work lays the foundations for a mechanistic understanding of NuA4's regulatory role and elucidates how its essential long-range activity is attained.
The major histone H4 acetylase NuA4 comprises at least 12 subunits organized into four modules 1,2 and plays crucial roles in transcription regulation and DNA repair 3,4 . Esa1 forms a tetrameric HAT module with the Eaf6, Epl1 and Yng2 subunits 1,2,5 , whose structure has been resolved by X-ray crystallography 6 . Esa1 promotes both targeted and untargeted histone H4 acetylation 2,7 . Targeted acetylation involves the NuA4 complex and occurs on specific promoters. Local recruitment of NuA4 can mediate long-range H4 acetylation patterns extending over several nucleosomes 7,8 . In Saccharomyces cerevisiae, the HAT module exists in isolation, separated from the other NuA4 subunits, to form the Piccolo complex. Piccolo, but not NuA4, promotes global, untargeted H4 acetylation 9,10 . In NuA4, the TINTIN module, composed of Eaf5, Eaf7 and Eaf3, which reads methylated histone H3, forms a sub-complex that regulates transcription elongation 11,12 . The 430kDa Tra1 protein interacts with sequence-specific transcription activators and forms a module involved in NuA4 recruitment to gene promoters 1,13,14 . Tra1 is a member of the phosphatidylinositol-3-kinase-related kinase (PIKK) family and is shared with the SAGA complex 15 . Although NuA4 and SAGA Article https://doi.org/10.1038/s41594-023-01056-x site within the promoter 7 . Fundamental questions therefore remain unanswered, including how HAT activity is embedded within the complex, and how the complex targets remote nucleosomes. More generally, other chromatin-modifying complexes are suspected to possess long-range activity; however, the mechanism and functional relevance of this activity remain elusive 24,25 .
Here we report the structure of NuA4 at 3.4-Å resolution. Amino acid densities were clear, allowing us to fully trace the core of the complex. We reveal how a scaffold composed of extended and interwoven Eaf1, Epl1 and Swc4 subunits tightly binds Tra1, as well as the actin module, coordinates several flexible histone-tail readers and incorporates the HAT module. Most importantly, a long, unfolded region of Epl1 connects an α-helix embedded in the core of NuA4 with the amino terminus of Epl1, which is an integral part of the HAT module. We show that this region endows the HAT module with its long-range action relative to the NuA4 core, and that this capacity is critical for fission yeast proliferation and bulk H4 acetylation in vivo. Finally, using ChIP-qPCR, we observed a reduction in H4 acetylation levels over extended genomic regions, spanning several kilobases, when the Epl1 linker was shortened, demonstrating its importance in establishing broad domains of H4 acetylation in vivo.

Overall structure
NuA4 complex was purified from a strain of the budding yeast P. pastoris (Pp) carrying a streptavidin-binding protein (SBP) affinity tag fused to the endogenous Eaf1 subunit. The complex contained 12 subunits, with a composition that differs from that of the complex in S. cerevisiae in only the TINTIN module 12 , which harbors a 'reader' domain of histone-tail modifications (Extended Data Fig. 1). In P. pastoris, TINTIN lacks subunits Eaf5 and Eaf3, which recognize methylated histones, but contains instead a homolog of the human BRD8 component of TINTIN, use the same activator-docking module, genome-wide approaches have revealed that they regulate both overlapping and distinct sets of genes, suggesting that activators can distinguish between SAGA and NuA4 (refs. [16][17][18]. It is not known how the assembly and context of Tra1 within the two complexes affect activator binding. The NuA4 subunits Swc4, Yaf9 (which reads acetylated histones), Arp4 and Act1 form a module also found in the SWR1 chromatin-remodeling complex [19][20][21] . Arp4 and Act1 are also part of the INO80 complex, an ATP-dependent chromatin remodeler. The four modules are thought to be connected by a single scaffolding component, Eaf1, the only NuA4 subunit that is not shared with another chromatin modification and remodeling complex, or with Piccolo 10,11,22 .
Structures of isolated subunits or modules have been reported 6 , but structural information regarding holo-NuA4 is limited to intermediate resolution EM reconstructions 11,23 . These revealed an L-shape assembly in which the Actin and Tra1 modules each occupy a different domain while the HAT and TINTIN modules are poorly resolved. Moreover, the scaffold that orchestrates the binding and activity of the other modules within NuA4 could not be revealed. Importantly, and in contrast to the major H3 acetyltransferase SAGA, NuA4 is able to act more than 1,000 base pairs (kbp) away from its recruitment site 7,8 . First, the Workman group showed a difference in acetylation range between NuA4 and SAGA in vitro 8 . Using purified complexes targeting a discrete site on a chromatin template through GAL4-binding sites and a Gal4-VP16 activator, they found that NuA4 generates a broader acetylation profile than does SAGA, extending over 1.5 kbp from the site of recruitment, whereas SAGA can only acetylate adjacent nucleosomes. Second, the Struhl group used chromatin immunoprecipitation followed by quantitative PCR (ChIP-qPCR) to measure H4 acetylation levels at ribosomal protein (RP) genes in vivo and detected NuA4-dependent H4 acetylation more than 800 bp upstream and downstream of the NuA4-binding  Fig. 2). These variations may reflect the plasticity of the NuA4 reader domains, which allows it to adjust to specific epigenetic mechanisms of regulation. The structure of holo-NuA4 at 3.4-Å resolution was determined by single-particle cryo-electron microscopy (cryo-EM), and focused refinement and classification improved the resolution of the core of the complex to 3.3 Å (Fig. 1a, Table 1 and Extended Data Figs. 3 and 4). The high quality of the map showed clear density for the majority of side chains in the core and enabled construction of a nearly complete atomic model for this part of the complex (Fig. 1b and Extended Data Fig. 5). The core is composed of the Tra1 module, on which the actin Arp4-Act1 hetero-dimer, bound to the Swc4 SANT-like domain, is grafted by means of a scaffolding neck formed by intertwining Eaf1, Swc4 and the carboxy-terminal half of Epl1, which connects to the HAT module (Fig. 1c). The clear delineation of the C-terminal half of the Epl1 subunit in the core, guided by amino acid side chain densities, indicates that Epl1 is present in the vast majority of NuA4 particles. However, the N-terminal part of Epl1, and the HAT module organized around it 9 , are not readily detected in our cryo-EM map, indicating that the HAT module is inherently mobile relative to the core, as described below. Such movements probably also explain why the nucleosome-binding Yaf9 subunit, as well as the TINTIN module, are not discernible in the map. In sharp contrast, the Epl1 and Swc4 SANT domains, which can also bind nucleosome core particles or unmodified histone tails, are very stable. Hence, NuA4 has both flexible and static modes of interaction with a nucleosomal substrate.

Epl1, Swc4 and Eaf1 subunits form a stable scaffold
An intricate path of the Eaf1 subunit could be traced unambiguously in our map, because this subunit traverses the neck of NuA4 multiple times (blue subunit, Fig. 1b,d). With the exception of the SANT domain, Eaf1 is mostly organized into loops, stretches of unfolded domains and isolated helices. Epl1 and Swc4 are both elongated and extended proteins that interact with Eaf1 extensively along the neck, from Tra1 to the actin module, and support the highly intricate topology of Eaf1 (Fig. 1b). Following a disordered N-terminal stretch that anchors the TINTIN module 23 , Eaf1 makes critical contacts with Tra1, then continues towards the neck base, where it forms a β-sheet stabilized by three strands contributed by Epl1 (Eaf1 1 in Fig. 1b,d). Eaf1 continues by providing two long helices to the bundle forming the bulk of the neck base, and the loop between the two helices binds again to Tra1 (Eaf1 2 ). Eaf1 then folds into the helicase-SANT-associated (Eaf1 HSA ) helix, which firmly binds the actin module and is followed by a 12-residue-long loop that partially encircles Act1 (Eaf1 3 , Fig. 1e). The engagement of the actin module is further bolstered by the Swc4 SANT-like domain as well as two strands and a helix from Epl1 that associate with the HSA helix and Arp4, respectively. Eaf1 traverses the neck domain a fourth time, completing the Eaf1-Epl1 β-sheet with an additional β-strand (Eaf1 1 ). The C-terminal part of Eaf1 connects to three distinct sites on Tra1 through its SANT domain (Eaf1 SANT ), as well as two additional helices, virtually encircling the FAT domain of Tra1 (Eaf1 4 in Fig. 1b,d). These two helices form an important hub of inter-module contacts. In addition to Tra1, they bind a helix of Swc4 that immediately precedes the domain anchoring Yaf9 and, most importantly, they also connect to the first helix of Epl1 in the core (helix H1). This Epl1 helix is preceded by a long, unfolded linker and the N-terminal half of Epl1 that is part of the HAT module. Hence, a major site of Tra1-module binding on the neck also associates with the domains of Swc4 and Epl1 that coordinate Yaf9 and the HAT module, respectively. The most prominent structured part of the neck is a helix bundle that forms strong contacts with Tra1 and in which all three scaffolding subunits intertwine. Epl1 contributes two helices that envelop two Eaf1 helices (Eaf1 2 ), and Swc4 completes the bundle with a long, kinked helix. Multiple strong interactions, including hydrophilic and hydrophobic bonds, stabilize this helix bundle, which provides stability and stiffness to the scaffolding neck. Altogether, Eaf1 contacts six proteins from all modules of NuA4 in accordance with its proposed scaffolding role 22 . However, Eaf1 is highly extended and contains a large proportion of unfolded loops that would not be sufficient to construct a stable interaction platform. Our results highlight the essential contributions of Epl1 and Swc4 to shaping the architecture of NuA4, because only an extensive interaction network between these components and Eaf1 provides a robust platform to stably anchor the functional modules.

The unique environment of the actin module
An actin hetero-dimer is found in several chromatin remodelers, in which it binds a long HSA helix and forms a mobile module that is poorly resolved in EM maps. Conversely, in the head part of NuA4, the Eaf1HSA-Act1-Arp4 module is strongly coupled to the neck and is well resolved. This stability stems from multiple additional interactions between this module and the SANT-like domain of Swc4, an Epl1 loop that is intercalated between the two actin proteins, as well as a loop from Eaf1 (Fig. 2). This complementary interaction network completely envelops and shields the Eaf1 HSA helix, which is unlikely to have a role similar to that of its analog in the INO80 complex, in which it serves as a long-range sensor for extra-nucleosomal DNA 28 . The nearly neutral charge of the Eaf1 HSA helix also supports this conclusion. Hence the role of the actin module in NuA4 remains enigmatic, but it seems that, at least in part, the actin module and Tra1 serve an important structural role by stabilizing and constraining the three scaffolding subunits at each extremity of the neck.

Ensuring selective use of Tra1
The large ~400-kDa Tra1 subunit is the only component that is shared between NuA4 and the SAGA H3 acetyltransferase, thus raising questions about the mechanism governing incorporation of specific subunits into either complex while preventing formation of chimeric assemblies between SAGA and NuA4 (ref. 27). The central module of SAGA associates with Tra1 through three main protein bridges 24,29 . We found that Eaf1 interactions with Tra1 engage all sites occupied by these bridges (Fig. 3a). In one case, the bridge formed by Eaf1 and Tra1 structurally mimics the analogous binding surface in SAGA: an aromatic residue (F523) of an Eaf1 loop protrudes into the Tra1 ring, as observed for Spt3 within SAGA. Thus, binding of Eaf1 is sufficient for nucleating Tra1 in the NuA4 complex (Fig. 3a). Tra1 serves as the major docking platform for activators that recruit SAGA and NuA4 to specific promoters 13 . However, many genes are predominantly regulated by either NuA4 or SAGA in yeast 17,27 , and the affinity between each complex and certain activators differs in vitro 13 . Our structure offers some clues that help us understand these functional differences. NuA4 forms more extensive interactions with Tra1 involving not only Eaf1, which practically envelops the Tra1 FAT domain, but also contributions from Swc4 and Epl1, mainly from the helix bundle at the neck base (Fig. 3b). Moreover, the NuA4 neck is tilted by 90° relative to the SAGA-specific subunits, resulting in distinct Tra1 surfaces being accessible to activators. Additionally, bridge-forming subunits might play auxiliary roles in activator binding 30 and further differentiate the two complexes. Lastly, we observe the flexibility of some internal domains of Tra1 differs between the two complexes ( Fig. 3c and Extended Data Fig. 6). The LBE-FRB module of the Tra1 pseudokinase is better ordered in NuA4, in which it may be stabilized by the N-terminal Eaf1 helix (residues 167-173) that interacts next to the FRB domain. Conversely, 9 helices forming the Tra1 ring solenoid (residues 2110-2322) show better local resolution in SAGA than in NuA4, indicating a higher mobility of the corresponding NuA4 HEAT repeats. Collectively, these effects may form specific activator-binding interfaces within each complex.

Uncovering the HAT module
We next sought to resolve how the HAT module is connected to the core. It has previously been shown that the N-terminal half of Epl1 is part of the HAT module, whereas the C-terminal half, which we resolved in the core, is crucial for its inclusion into NuA4 (ref. 9). Focused classification within a sphere expanding from the first Epl1 residue detected in our map revealed a fuzzy density that was apparent only at a low density threshold, suggesting that the HAT module is flexibly attached to NuA4. We posited that mobility is likely to be conferred by a stretch of amino acids predicted to be disordered (residues 359-418). This linker is conserved between yeasts and vertebrates (Extended Data Fig. 7) and connects the last Epl1 residue resolved in the crystal structure of the HAT module (K358) with the first residue that is discernible in our cryo-EM map (M419). This residue is closely followed by the first Epl1 helix in the core (helix H1, S431). We introduced a TEV protease cleavage site into the linker to investigate its role in connecting the HAT module to the core. Using an SBP affinity tag on Eaf1, holo-NuA4 was bound to streptavidin beads and then incubated with TEV protease. We observed that all HAT subunits were efficiently and specifically  Article https://doi.org/10.1038/s41594-023-01056-x released from holo-NuA4 upon cleavage, demonstrating that the HAT module is tethered to the core mainly by the long unstructured Epl1 linker (Extended Data Fig. 8). This observation agrees with cross-linking coupled to mass spectrometry analyses, which identified only a few interactions between the HAT module and the core 23 .
To better visualize the HAT module connected to NuA4, deletion mutants were designed in which the linker was gradually shortened (Epl1-SL1, Epl1-SL3, Epl1-SL5), bringing the HAT module closer to the core (Fig. 4a). The fuzzy density indeed grew bigger and drew nearer to the core as the loop got shorter. The density corresponding to deleted residues was no longer observed in the core, further supporting our Epl1 density assignment and residue register (Fig. 4b). Furthermore, neural-network-based heterogeneity analysis (cryoDRGN) of the wild type (WT) and the Epl1-SL1 and Epl1-SL3 mutants showed clearly that a density with the size of the HAT module appears close to the core in 90% of the Epl1-SL3 images, whereas such an adjacent density was observed in only 10% of the WT images (Fig. 4c). This analysis shows that shortening the Epl1 linker attracts a protein density that is in close proximity to the NuA4 core. Although residual flexibility precluded the resolution of molecular details, the appended domain is connected to the Epl1 H1 helix in representative classes (Fig. 4d). To gain insights into its molecular shape, particles were re-centered on the attached density and the signal of core NuA4 was subtracted. Ab initio three-dimensional reconstruction and heterogeneous refinement of the residual domain resulted in a cryo-EM map with a resolution of 11.5 Å that was highly similar in shape to the crystal structure of the HAT module 6 (Fig. 4e). An extra protein density was detected in the cryo-EM map that may correspond to parts that were not included in the crystal structure. Altogether, these experiments demonstrate that the HAT module is tethered to the Epl1 H1 helix within the core of NuA4 via a long unstructured Epl1 linker that determines its orientation and maximal distance from the core (Fig. 4f).
To test whether Esa1 enzymatic activity is regulated by the Epl1 linker, mutant NuA4 complexes containing short linkers were affinity-purified and their ability to acetylate nucleosome core particles was assayed by western blot analysis (Fig. 4g). These results show that the Epl1 mutants with short linker regions are enzymatically as active as WT NuA4 and are able to modify a diffusing mono-nucleosomal substrate in vitro. Hence, reducing the Epl1 linker length does not affect the intrinsic enzymatic activity of Esa1.

A long linker is essential for viability in S. pombe
We next examined the function of the unstructured Epl1 linker in vivo. For this, we switched to S. cerevisiae because of its amenability to genetic manipulation. As EPL1 deletion is lethal 31 , we created epl1 linker mutants in a sds3Δ deletion background, which suppresses NuA4-null   32 . Tetrad analysis of heterozygous diploid mutants demonstrated that the Epl1-SL3 and Epl1-SL5 mutations affect neither growth nor bulk histone H4 acetylation levels (Extended Data Fig. 9). Epl1-SL5 mutants, which have the longest linker deletion, display mild sensitivity to high temperature, suggesting defects in targeted, activator-driven H4 acetylation and transcriptional induction. However, biochemical and genetic evidence indicates that incorporation of the HAT module into NuA4 is not essential in S. cerevisiae because of the existence of an unbound version of the HAT module, named Piccolo 9,10,32 . Indeed, deletions of either the scaffolding subunit Eaf1 or the C-terminal half of Epl1 are viable in S. cerevisiae. Our result supports this model and suggests that Piccolo masks the putative effects of Epl1 linker shortening. By marked contrast, in the distantly related fission yeast Schizosaccharomyces pombe, the Eaf1 homolog Vid21 is essential for viability, despite the conserved NuA4 subunit composition 27,33 (Extended Data Fig. 1). We thus reasoned that, rather than Piccolo, the holo-NuA4 complex is critical for proliferation in fission yeast. Accordingly, we observed that deleting the C-terminal end of S. pombe Epl1 causes lethality (Extended Data Fig. 9), whereas an analogous truncation mutation detaches the HAT module from holo-NuA4 and is viable in S. cerevisiae 9 . Thus, unlike in S. cerevisiae, incorporation of the HAT module into NuA4 is essential in S. pombe.
This finding prompted us to test the importance of the distance between the HAT module and the core-NuA4 in this yeast. To introduce conditional mutations at the endogenous epl1 locus, we implemented a novel strategy based on a Cre recombinase-dependent 'flip-excision' switch (FLEx), initially developed in mice 34 (Fig. 5a). Briefly, FLEx takes advantage of the β-estradiol inducibility of a CreER fusion, the ability of Cre to either invert or excise a DNA fragment depending on the orientation of the flanking lox sites, and the existence of lox variants that are not cross-compatible. We obtained three Epl1-FLEx mutants, Epl1-SL1, Epl1-SL5 and Epl1-DCt (Fig. 5b). PCR of genomic DNA and marker analyses revealed that addition of β-estradiol induces a rapid, efficient and irreversible inversion of the epl1 locus, demonstrating the utility of this approach for conditional genetic manipulation in S. pombe (Fig. 5c,d).
Remarkably, we observed β-estradiol-induced proliferation defects leading to strong reduction in S. pombe viability in both the epl1-SL1 and epl1-SL5 FLEx mutants (Fig. 5d,e). As expected from our tetrad analysis, we observed a similar phenotype in the epl1-DCt-FLEx strain, validating the FLEx approach. These results demonstrate that the length of the Epl1 linker has critical functions in S. pombe proliferation. Finally, western blot analyses of total protein extracts showed that the Epl1-DCt, Epl1-SL1 and Epl1-SL5 mutants are stably expressed and confirmed that β-estradiol induces a robust switch from wild-type MYC-tagged Epl1 to the three HA-tagged mutant forms (Fig. 5f).

The Epl1 linker is required for long-range H4 acetylation
We then tested the effect of shortening the S. pombe Epl1 linker on histone H4 acetylation. Western blot analysis of total protein extracts revealed a modest, but reproducible, reduction of normalized H4   Article https://doi.org/10.1038/s41594-023-01056-x acetylation levels in both the epl1-SL1-FLEx and epl1-SL5-FLEx mutants, as compared with uninduced conditions and with WT control strains, suggesting that shortening the Epl1 linker affects NuA4 activity in vivo (Fig. 6a). Semi-quantitative analyses of signal intensities indicated that the longer truncation mutant, Epl1-SL5, has a stronger effect than does the shorter truncation mutant, Epl1-SL1, as compared with the complete loss of the Epl1 C-terminal moiety (Fig. 6b). Together with our finding that the Epl1 linker does not regulate the intrinsic enzymatic activity of Esa1 in vitro, these results indicate that the distance between the HAT module and the core NuA4 is important for bulk H4 acetylation in vivo. We next used ChIP-qPCR to measure H4 acetylation at chromatin in conditional epl1-SL5 mutants. Specifically, we examined three distinct loci, including the adh1 gene, which has a 2-kb-long upstream intergenic region, the highly expressed mmf1 and rpl1603 genes, and part of the subtelomeric region of chromosome 2's right arm, which contains only pseudogenes and the repressed tlh2 gene (Fig. 6c-e).
Overall, at all loci, we observed a reproducible decrease of acetylated H4 levels normalized to total H3, as compared with levels in uninduced conditions. Notably, at the adh1 locus, this reduction is apparent both in the coding region and over 1 kb upstream of the transcription start site (TSS, +1), but not within 500 bp upstream of the TSS (Fig. 6c), which is where NuA4 presumably binds to chromatin through activator recruitment. Likewise, H4 acetylation levels decrease where NuA4 likely does not bind. First, we observed reduced H4 acetylation levels over the coding region of the mmf1 and rpl1603 genes, including towards their 3′ ends (Fig. 6d). Second, we show that shortening the Epl1 linker reduces basal H4 acetylation levels over an extended region of a subtelomeric region, spanning several kilobases and containing only the repressed genes and pseudogenes (Fig. 6e). Overall, these observations are consistent with a model in which, in S. pombe, the Epl1 linker endows NuA4 with the ability to acetylate histone H4 over large nucleosomal domains.

Discussion
In vitro experiments show that NuA4 has the capacity to find its targets over long distances, in contrast with SAGA, in which the HAT module is directly coupled to the core via a structured domain 8 . Specifically, distinct histone acetylation patterns were observed upon using an activator to recruit either SAGA or NuA4 to an artificial promoter. SAGA acetylates H3 from nucleosomes immediately adjacent to the activator-binding site; NuA4 generates a broader H4 acetylation profile, extending up to 1.5 kbp, which corresponds to 7-8 nucleosomes.
In vivo observations further support the long-range activity of NuA4. First, ChIP-qPCR analyses show that, in S. cerevisiae, Esa1-dependent H4 acetylation spans an entire gene encoding a ribosomal protein, including the promoter and coding sequences, although NuA4 is specifically recruited to the activator-binding sites within the promoter 7 . Second, NuA4 is recruited to the upstream region of the PHO5 gene promoter, through interactions with Pho2, but its chromatin-modifying activity was shown to span four nucleosomes 35 .  We show here that the distance between the HAT module and the rest of the complex is important for yeast proliferation and contributes to maintain basal H4 acetylation levels in S. pombe. Furthermore, our ChIP assays show that shortening the Epl1 linker does not affect H4 acetylation levels proximal to the NuA4 recruitment sites within the promoter, but decreases distal H4 acetylation in the gene body. Altogether, our work describes the mechanism that endows a chromatin-modifying enzyme with long-range activity and provides evidence for the importance of this function in vivo. We note that the length of the Epl1 linker varies between species (Extended Data Fig. 7c), ranging from 50 residues in S. pombe to 109 residues in S. cerevisiae. It is thus likely that the theoretical physical reach changes between species. One possible explanation for these variations might relate to differences in nucleosome spacing. Notably, the linker length is shorter in S. pombe than in S. cerevisiae, suggesting that, in the latter, NuA4 can reach longer distances to modify a given number of nucleosomes 36,37 . Finally, other chromatin-modifying enzymes, including the histone deacetylase Sin3L-Rpd3L co-repressor complex 25 , have unstructured protein domains linking the enzymatic modules to the core of the complex. We suggest that, for these enzymatic complexes, a similar mechanism allows histone modifications to be deposited at long distances from the recruitment site.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41594-023-01056-x. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Purification of NuA4
A strain of the budding yeast Komagataella phaffii (also known as P. pastoris) with the 38-amino-acid streptavidin-binding peptide (SBP) affinity tag fused to the C-terminus of the endogenous Eaf1 subunit was produced following standard yeast molecular biology techniques. Two liters of yeast cells were grown at 24 °C with glycerol as a carbon source and were collected when the OD 600 nm reached 16-18. Cells were washed in ice-cold water and then treated with 10 mM DTT in 1.1 M sorbitol. The cell wall was digested by addition of lyticase at 30 °C and spheroplasts were pelleted at 5,500g for 24 min. All further steps were performed at 0-4 °C. Protease inhibitors were added to all buffers. Spheroplasts were washed twice in 1.4 M sorbitol and were then disrupted by suspension in a hypotonic buffer (15-18% Ficoll 400, 0.6 mM MgCl 2 , 20 mM K-phosphate buffer pH 6.6) using a ULTRA-TURRAX disperser. Sucrose as an additional 5 mM MgCl 2 , was added in order to precipitate some remaining organelles and membrane parts by a short centrifugation (33,000g for 10 min). The PEG 20000 concentration was then increased to 5.8%, and NUA4 was precipitated in a second short centrifugation step. The pellet was resuspended in a minimal volume, and avidin was added to block endogenously biotinylated proteins. The suspension was incubated with streptavidin beads for 5 h in buffer A (20 mM HEPES pH 8.0, 250 mM sodium chloride, 10% sucrose, 2 mM MgCl 2 , 2 mM DTT), washed 5 times with buffer A supplemented with 0.05% Tween-20 and finally eluted with buffer A containing 10 mM biotin. The eluate was concentrated with Millipore Amicon-Ultra (50 kDa cut-off) and loaded on a sucrose gradient generated by a Gradient Master 108 (BioComp) from a light solution (10% sucrose, 20 mM HEPES pH 8.0, 150 mM potassium acetate, 2 mM TCEP, 5 mM MgCl 2 , 0.0045% dodecyl-maltoside) and a heavy solution (30% sucrose instead of 10%) that contained 0.1% glutaraldehyde 38 . Following centrifugation in a rotor SW60 (180,000g for 14 h) the gradient was fractionated by puncturing the bottom of the tube with a needle. Glutaraldehyde was quenched by addition of 75 mM ammonium acetate. Peak fractions were concentrated (Millipore Amicon-Ultra) and dialyzed against a buffer containing 20 mM Hepes pH 8, 150 mM potassium acetate, 5 mM magnesium acetate, 2 mM TCEP, 5 mM ammonium acetate and 0.0025% dodecyl-maltoside. The quality of the sample and its concentration after dialysis were estimated by electron microscopy of negatively stained samples.

Yeast procedures and growth conditions
Standard culture medium and genetic manipulations were used. S. cerevisiae strains were grown in YPD at 30 °C to mid-log phase (~1 × 10 7 cells ml -1 ). S. pombe strains were grown in either rich (YES) or minimal (EMM) medium at 32 °C to mid-log phase (~0.5 × 10 7 cells ml -1 ). Proliferation assays were performed by inoculating single colonies in liquid medium and measuring the optical density at 595 nm at different time points, and by spotting tenfold serial dilutions of liquid cultures on rich medium and incubated for 3 d at 32 °C. For the longer time course, cultures were diluted to keep cells in constant exponential growth. For CreER-loxP-mediated recombination, cells were treated with either 1 µM β-estradiol (E2758, Sigma) or DMSO alone for various time points, as indicated.

Strain construction
All S. pombe and S. cerevisiae strains used are listed in Extended Data Table 1 and were constructed by standard procedures, using either yeast transformation or genetic crosses. Strains with truncations and C-terminally epitope-tagged proteins were constructed by PCR-based gene targeting of the respective open reading frame (ORF) with kanMX6, amplified from pFA6a backbone plasmids. Strains with N-terminally epitope-tagged proteins and internal deletions were constructed using CRISPR-Cas9-mediated genome editing. DNA fragments used for homologous recombination were generated by PCR, Gibson assembly cloning (kit E2611L, New England Biolabs) or gene synthesis. Transformants were screened for correct integration by PCR and, when appropriate, verified by Sanger sequencing or western blotting. For each transformation, 2-4 individual clones were purified and analyzed. For tetrad analyses in both S. cerevisiae and S. pombe (Extended Data Fig. 9), mutations were introduced in diploid strains to generate heterozygotes, which were then sporulated to analyze growth phenotypes. At least 20 distinct meiotic events were analyzed for growth and allele segregation by replica plating and PCR. The four progenies from each tetrad are labeled A-D (Extended Data Fig. 9e).

Construction of S. cerevisiae epl1 linker mutants
To construct seamless deletions of the linker region from endogenous S. cerevisiae Epl1, and to avoid possible issues with the genetic manipulation of an essential gene, we took advantage of the ability of sds3Δ deletion mutants to suppress the loss of viability observed in NuA4 HAT mutants, including epl1Δ 32 . In this background, we obtained an epl1-SL1 allele, which removes K395 to V501, and an epl1-SL5 allele, which removes I401 to H487. The sds3Δ epl1-SL1 and sds3Δ epl1-SL5 double mutants were then backcrossed to a WT strain to segregate each mutated allele independently.

Construction of endogenous conditional mutants in S. pombe
Using the FLEx strategy originally developed in mice 34 , we constructed S. pombe strains in which we could conditionally replace the WT epl1+ allele with a mutant copy of epl1 at its endogenous locus, thereby leaving all 5′ cis-regulatory elements intact. Specifically, we obtained two Epl1-FLEx mutants, epl1-SL1-FLEx and epl1-SL5-FLEx, which have seamless deletions of residues V323-A343 and K317-P357, respectively. We also designed an epl1-DCt-FLEx allele, in which Epl1 is conditionally replaced by the same C-terminal truncation mutant (Y405 to stop) analyzed in heterozygous diploids (Fig. 5a).

Generation of P. pastoris epl1 linker mutants
We used the NEBuilder enzyme mixture (NEB) to seamlessly assemble the following DNA fragments into a linearized pUC19 backbone: 1,000 bp upstream in the yeast genome relative to the position of the deletion in the linker; the yeast genome sequence from the deletion to the end of Epl1; the affinity tag sequence with a stop codon followed by the AOX gene transcription termination sequence; a Zeocin selection marker; and 1,000 bp in the yeast genome downstream of Epl1. PCR was used to amplify the assembled DNA. Pichia pastoris X33 WT cells were transformed with 180 µg of the PCR product by electroporation.

Statistical analysis
All statistical tests were performed using GraphPad Prism (version 9.2.0). t tests were used when comparing two means. One-and two-way analyses of variance (ANOVAs) were performed to compare more than two means, across one (for example 'genotype') and two variables (for example 'genotype' as a between-subject factor and 'treatment' as a within-subject factor), respectively. One-and two-way ANOVAs were followed by Tukey and Bonferroni post-hoc pairwise comparisons, respectively. A significance level (α) of 0.05 was used a priori for all statistical tests, except when otherwise indicated. Comparisons that are statistically significant (P ≤ 0.05) are marked with one asterisk. Statistical details of experiments can be found in the figure legends, including the statistical tests used, the minimum number of biological replicates shown (n, isogenic clones of each strain) and a description of the center and dispersion statistics. Quantitative values are represented as individual values (n) overlaid with the mean (black bar) and standard deviation (s.d.).

Chromatin immunoprecipitation
ChIP experiments were performed as previously described 40

Cryo-EM sample preparation and data acquisition
Three microliters of sample was applied onto a holey carbon grid (Quantifoil R2/2 300 mesh) rendered hydrophilic by 90 s of treatment in a Fischione 1070 plasma cleaner operating at 30% power with a gas mixture of 80% argon:20% oxygen. The grid was blotted for 1 s at blot force 8 and flash-frozen in liquid ethane using Vitrobot Mark IV (FEI) at 4 °C and 95-100% humidity. Images were acquired on a Cs-corrected Titan Krios (Thermo Fisher) microscope operating at 300 kV in nanoprobe mode using the serialEM software for automated data collection 41 . Movie frames were recorded either on a Gatan K2 Summit or on a Gatan K3 direct electron detector after a Quantum Ls 967 energy filter using a 20 eV slit width in zero-loss mode. Images were collected in super-resolution mode at a nominal magnification of ×105,000 (K2 summit) or ×81,000 (K3), which yielded a pixel size of 0.55 or 0.852 Å. For each image, 40 movie frames were recorded at a dose of 1.32 electrons per Å² per frame, corresponding to a total dose of 52.8 e/Å 2 , but only the last 38 frames were kept for further processing.

Image processing
WARP was used to perform the first pre-processing steps, aligning movie frames, dose-weighting, correcting the beam-induced specimen motion and contrast transfer function (CTF) estimation 42 . After visual inspection, images with poor CTF and those showing particle aggregation or abundant ice contamination were discarded. Particle coordinates were determined using crYOLO 43 . The datasets were analyzed in Relion 3.1 (ref. 44) and cryoSPARC 45 according to standard protocols. Briefly, three rounds of reference free two-dimensional classification of the individual particle images were performed in cryoSPARC to remove images corresponding to contaminating or damaged particles and ice contamination. Four references (three-dimensional (3D) models) were generated by the ab initio 3D reconstruction program of cryoSPARC. These structures were then used as references for 3D classification in cryoSPARC, and particles corresponding to high-resolution 3D classes were selected and used for non-uniform refinement. We carried out focused refinements on three different parts of NuA4 using masked volumes including Tra1-tail, Tra1-ring and Core, which comprise the actin, neck, Tra1 FAT and Tra1 kinase domains, respectively. Global resolution estimates were determined using the FSC = 0.143 criterion after a gold-standard refinement. Local resolutions were estimated with ResMap 46 and cryoSPARC. To analyze the heterogeneity in the cryo-EM map due to the presence of the flexible Piccolo module (MW, 144 kDa), we used the neural-network-based cryoDRGN reconstruction to map the particles on two principal components 47 . For each WT and short linker mutant data set, we partitioned the latent space into 20 regions, and a density map was generated from the center of each region. The maps were thresholded to a value that corresponds to the expected volume of NuA4 (MW: 1,030,447 Da), assuming an average protein density of 0.825 Da A -3 . Upon visual inspection, the maps were grouped into three classes according to the absence (class 1) or presence of an additional mass that was disconnected (class 2) or connected (class 3) to the core of NuA4. To determine the low-resolution map of Piccolo, the density of the stable core-NuA4 was subtracted from the original images, and the images were re-centered according to the expected position of Piccolo. A two-dimensional classification step was performed to assure the correct subtraction of the core of NuA4 and the presence of a residual density. Ab initio reconstruction followed by 3D classification steps in cryoSPARC were used to reconstruct the map of Piccolo at 12 Å resolution. Heterogeneous 3D refinement of the dataset that was low-pass filtered to 6 Å produced a reconstruction at 11.5 Å resolution.

Model building
Homology models for Act1, Arp4 and the SANT domains of Eaf1 and Swc4 were generated using iTASSER 48 or Swiss-model 2019 and docked into the maps using the chimera docking tool. Eaf1, Swc4 and Epl1 were built de novo using the Coot software 49 . To facilitate de novo model building, RaptorX 50 and PSIPRED 51 were used to predict secondary structure and disordered parts. A large proportion (called density) of the coulombic potential difference map associated with each side chain was clear, and practically all densities in the core of NuA4 could be assigned with confidence. For Eaf1 residues 367-389 and Tra1 residues 4-300, 2019-2392 and 2969-3051, the local resolution of the structure did not allow us to place the side chain but was sufficient for placing

Nature Structural & Molecular Biology
Article https://doi.org/10.1038/s41594-023-01056-x of helices and tracing of the Cα backbone. The atomic model was refined in Phenix by real-space refinement with secondary structure restrains 52 . All display images were generated using UCSF Chimera 53 and ChimeraX 54 .

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Article https://doi.org/10.1038/s41594-023-01056-x Extended Data Fig. 9 | The Epl1 linker is dispensable for NuA4 activity and functions in budding yeast. a, Schematic illustration of the S. cerevisiae Epl1 linker mutants that were constructed. The epl1-SL1 allele corresponds to a seamless deletion of residues 401-487. The epl1-SL5 allele corresponds to a longer, seamless deletion of residues 395-501 from endogenous Epl1. b, S. cerevisiae EPL1+/epl1-SL1 SDS3+/sds3Δ and EPL1+/epl1-SL5 SDS3+/sds3Δ double heterozygous diploids were sporulated, dissected, and germinated to show the growth phenotype of all four possible genotypes (top panel). Single epl1-SL1 and epl1-SL5 mutants were isolated, along with an isogenic wild-type (WT) controls, grown to exponential phase and spotted on rich media in ten-fold serial dilutions at grown at different temperatures, as indicated (bottom panel). Data are representative of four independent experiments performed with distinct mutant clones. c, Acetylated histone H4 levels in exponentially growing S. cerevisiae epl1 mutants. The control epl1-SL1 and epl1-SL5 single mutants were obtained from the tetrad analysis shown in C. The epl1-DNt sds3Δ and epl1-CNt sds3Δ double mutants were obtained from 32 and corresponds to a deletion of residues 1-485 and 485-833, respectively. Western blots of total protein extracts were probed with an anti-pan-acetyl-H4 antibody. An anti-H3 antibody and Ponceau red staining were used as controls for equal loading between lanes. Data are representative of two independent experiments. Source data for c are provided as a Source Data File. d, Illustration of S. pombe Epl1 mutants. The epl1-DCt allele has a seamless deletion from residues 405 of Epl1 until its C-terminal end (residue 557). e, Tetrad analysis of a heterozygous epl1+/epl1-DCt diploid strain following sporulation and germination to show 2:2 segregation of a lethal phenotype.