Main

Chromatin structure regulates all DNA transactions, including transcription, replication and repair. The nucleosome, composed of 146 base pairs (bp) of DNA wrapped around a histone octamer, constitutes the repeated chromatin unit. Histone variants that deviate in sequence from the canonical ones and post-translational modification of histones can either directly alter chromatin structure or recruit proteins to achieve that effect10. Human TIP60-C is a 1.8-MDa complex, composed of at least 20 subunits, that has two chromatin-modifying enzymatic activities: acetylation of histones H4 and H2A; and incorporation, driven by an ATPase situated in subunit EP400, of the histone variant H2A.Z instead of the canonical H2A2,11. Histone acetylation by TIP60-C is associated with active gene expression12 as well as the repair of DNA double-strand breaks13. H2A.Z is an essential, conserved histone variant enriched in nucleosomes flanking the transcription start site, but its role is perplexing because it seems to foster transcriptional activation of some genes but repression of others14,15. The molecular mechanism that underlies the contradictory effects of H2A.Z is poorly understood.

In yeast, these chromatin modifications are performed by two separate multiprotein assemblies: the SWR1 complex introduces the yeast homologue of H2A.Z, Htz1, into nucleosomes by ATP-dependent exchange of H2A–H2B histone dimers for Htz1–H2B; and the NuA4 complex carries out the histone acetyltransferase activity (Extended Data Fig. 1). The yeast complexes are believed to act sequentially in a coordinated manner, and their human analogues are both incorporated into TIP60-C. Structures of the yeast NuA4 and SWR1 complexes have been resolved by single-particle cryo-electron microscopy (cryo-EM)4,5,6,7,8, but how these two activities physically merge in human TIP60-C and what this association entails are unknown. Here we present the 2.4–3.3 Å resolution cryo-EM structure of endogenous TIP60-C purified from human cells. We complement the structure with a set of biochemical tests to reveal its implications for the functional properties, mechanism of action and recruitment to chromatin of TIP60-C. On the basis of these results, we propose new perspectives for the outcome of H2A.Z deposition. We thus provide a molecular framework for understanding the activity of TIP60-C in DNA repair and transcription regulation as well as its function in human diseases. Indeed, synthetic lethality between TIP60-C and the SWI/SNF family of chromatin remodellers in human cancer cell lines suggests that inhibiting TIP60-C is an attractive therapeutic approach for the many cancer types in which SWI/SNF is perturbed16.

Overall structure of TIP60-C

To isolate endogenous TIP60-C, we used CRISPR–Cas9 technology to fuse an SBP-3×Flag affinity tag at the C terminus of subunit EPC1 in human erythroleukaemic K562 cells. The purified complex contained all expected subunits (Extended Data Fig. 2a,b) and included an H2A.Z–H2B dimer in near stoichiometric amounts. The complex was subjected to single-particle cryo-EM analysis (Extended Data Figs. 2c, 3a,d,f, 4, 5 and 6a and Extended Data Table 1). Notably, we find that the density map of TIP60-C is composed of three large subcomplexes instead of the expected two (Fig. 1a). The SWR1L subcomplex is dominated by a hetero-hexameric core of the RuvBL1 and RuvBL2 subunits as found in yeast8 (Fig. 1b). The C-lobe of the EP400 ATPase domain is bound to the hexamer, and projects a large extended insertion into it. The Swc6–Arp6 dimer that appears in yeast at the hexamer face opposing the ATPase C-lobe is absent in TIP60-C, whereas an isolated ARP (actin-related protein) is located beneath the hexamer base, where no protein occurs in yeast. The second subcomplex, reminiscent of the NuA4 core, is markedly diminished and rearranged with respect to the yeast version (Fig. 1b). In particular, a large robust neck domain in yeast serves to strongly couple the TRRAP homologue, Tra1, to the actin module at the core of NuA4. In TIP60-C, the actin module, which comprises two ARPs (actin and actin-like protein 6A (ACL6A)), the DMAP1 subunit and the N-terminal part of the helicase-SANT-associated (HSA) helix, sits on top of a highly dwindled small neck that shows very little resemblance to its yeast counterpart. Furthermore, as a result of the loss of a massive neck domain in TIP60-C, the 400-kDa TRRAP subunit is not tightly associated with NuA4L, but instead constitutes a separate semi-autonomous third subcomplex located on the opposite side of NuA4L when compared to yeast. Owing to this flexibility, our map of TIP60-C does not resolve TRRAP well, and it appears as a large cloud of density of roughly the expected size, next to the degenerate neck domain (Extended Data Fig. 3d). To investigate our interpretation of this cloud as a highly mobile TRRAP, we reconstituted a three-dimensional (3D) map of the holo-complex from negatively stained molecules. Three-dimensional classification clearly shows a very large density in different orientations around the neck (Extended Data Fig. 3b,c). The more populated of these classes show clear structural features of TRRAP, which could be fitted well into the maps (Extended Data Fig. 3e). Moreover, by focusing our image analysis on TRRAP, we could solve the structure of this subcomplex by itself (Fig. 1). We find that the SANT (Swi3–Ada–NCoR–TFIIIB) domain of subunit EP400 strongly associates with TRRAP, thus establishing that TRRAP is part of the holo-complex and is connected to the NuA4L core through an 83-residue, unfolded loop that lies between the SANT domain and the part of EP400 embedded in the NuA4L core. Next to the SANT domain we find a poorly resolved additional large density, the shape of which fits that of the KAT (lysine acetyltransferase) module (Extended Data Fig. 6b–e). The maps of TRRAP and holo-TIP60 do not reveal a density that can unambiguously be attributed to the KAT module. This is probably a direct result of the module’s high mobility, which is afforded by its tethering to NuA4L through a long, unstructured EPC1 linker of around 100 amino acids, as was found in NuA4 in yeast4. Conformational analysis of the TRRAP module identified a fraction of the particle population with a density next to the EP400 SANT domain, into which the KAT module could be fitted (Extended Data Fig. 6b,c). We thus hypothesized that this density corresponds to the KAT module, which, in a fraction of TIP60-C particles, physically associates with TRRAP or the SANT domain. Such an interaction has been detected previously and has been shown to occur and to regulate KAT activity in vivo17. As for yeast NuA4, a few subunits are not detected in the density maps because they are flexibly bound to the core. These are YEATS4 and the TINTIN module, which ‘read’ different histone post-translational modifications. TRRAP, or its yeast homologue Tra1, serves as a docking platform for activators that recruit complexes to specific chromatin loci18. In SAGA and yeast NuA4, in which TRRAP (Tra1) is firmly coupled to the core, the enzymatic modules have been endowed with inherent flexibility to reach their targets4. By contrast, the TIP60-C ATPase lacks such dynamics, and a rigid TRRAP would entail a fixed distance between the activator platform subunit and the active site of the ATPase. This would leave no leeway for the ATPase to search for its target nucleosome, apart from the flexibility afforded by the activation domain of the recruiting transcription factor, which can be minimal. It stands to reason, therefore, that TRRAP flexibility allows for target searching by SWR1L.

Fig. 1: Architecture of the human TIP60–EP400 complex (TIP60-C).
figure 1

a, Two views of the composite cryo-EM map of endogenous human TIP60-C. b, Corresponding views of the atomic model of TIP60-C. Colours of subunits in maps and models match the colours of their written names. Dashed lines represent physical connections that are not visible in the maps. ACT, actin.

Full size image

Connecting the subcomplexes

The structure of TIP60-C reveals that the two enzymatic parts, SWR1L and NuA4L, are firmly connected through several bridging elements. The N-terminal domain of the YL1/VPS72 subunit holds the H2A.Z–H2B dimer before its deposition into the nucleosome (Fig. 2a). More to its C terminus, a long and extended region presenting a β-sheet interacts with and intercalates between two protomers of the RuvBL1–RuvBL2 hexamer (Fig. 2b,c). This extended region then crosses the entire SWR1L part, to be followed by an additional compact domain (residues 290–355) that is not observed in yeast SWR1 and comprises a helix, a β-sheet with two strands and several structured loops. This compact YL1 domain at the periphery of the supercomplex acts as a glue that connects the SWR1L and NuA4L parts. One face of this YL1 domain packs against a β-sheet from the base of a protomer in the RuvBL1–RuvBL2 hexamer, whereas its opposite face binds to a loop and a helix from the DMAP1 subunit, which is a component of NuA4L where it envelopes the actin module.

Fig. 2: YL1 glues SWR1L to NuA4L.
figure 2

a, Domain organization of YL1. The domain boundaries are marked with residue numbers. The magenta parts represent the modelled domains. b, Location of YL1 within TIP60-C. c, Enlarged view of the boxed area showing the interaction of YL1 with both SWR1L and NuA4L.

Full size image

By contributing to all three observed subcomplexes, the EP400 subunit serves as a scaffold that connects and coordinates them into one supercomplex (Fig. 3a). The huge 400-kDa EP400 subunit is a uniquely elongated, extended and spread-out protein. The N-terminal part of EP400 is not resolved in our map but was shown to bind to the highly flexible TINTIN module that includes several ‘readers’ of histone marks19. The first residues of EP400 that are observed in our map form a short α-helix that, together with another short helix donated by the DMAP1 subunit, constitutes the dwindled neck of NuA4L (Fig. 3b). The helix is followed by a loop and then a 90-residue-long HSA helix (residues 810–900) that emanates from the back of NuA4L, binds to the actin and the ACTL6A subunits to form the actin module, traverses the gap between the NuA4L and SWR1L parts and continues deep into the SWR1 part, where it binds to a third ARP (Fig. 3c). EP400 then folds into the ATPase in a position identical to that found in the yeast SWR1 complex8 (Fig. 3d). Following the long insertion that protrudes into the RuvBL1–RuvBL2 hexamer, EP400 completes the ATPase C-lobe. It then traverses again the gap that separates NuA4L from the ATPase as a second long helix, hereafter named the counter-helix (Fig. 3c). The counter-helix is nearly parallel to the HSA and, in its path beneath the RuvBL1–RuvBL2 hexamer, binds to one of the hexamer subunits as well as to the YL1 subunit, thus contributing to the stability of the supercomplex. EP400 is then embedded again in the NuA4L core, forming interactions with a YL1 loop (residues 210–260) and several elements of EPC1 and DMAP1 to construct a new interaction hub (Fig. 3e). Finally, EP400 extends out of NuA4L as a long loop that reaches the TRRAP subunit, where EP400 folds into the SANT domain (residues 2367–2479); this is followed by a loop and two helices (C-HLX, residues 2493–2523) that envelope the FAT (helix repeats named after FRAP, ATM and TRRAP) region of TRRAP (Fig. 3f).

Fig. 3: EP400 scaffolds the three TIP60-C lobes.
figure 3

a, Domain organization of EP400. The domain boundaries are marked with residue numbers, and the coloured parts correspond to the modelled domains. C-HLX, α-helix C-terminal of the SANT domain. b, Representation of the EP400 domains within TIP60-C. Colours of EP400 domains as in a. c–f, Enlarged views of key EP400 domains (c, HSA, ATPase N-lobe and counter-helix; d, ATPase C-lobe RUVB insert; e, interaction hub; and f, TRRAP connection) and their interactions with TIP60-C subunits.

Full size image

In all other members of the SWR1/INO80 family of chromatin remodellers, the extended HSA helix, which houses three ARPs, forms a highly mobile module placed at the periphery of the main remodeller body and is poorly resolved in the cryo-EM maps8,9 (Extended Data Fig. 7). This mobile module was shown in INO80 to act as a sensor for extranucleosomal DNA, mainly through interaction with the HSA helix itself, which is crucial for the activity of these remodellers and regulates their recruitment to chromatin20. Although the TIP60-C HSA module shares some structural principles with its counterparts in other INO80/SWR1 remodellers, it also shows some key differences (Extended Data Fig. 7b–d). In TIP60-C, this module is located at the centre of the holo-complex, and it is fixed and well resolved in our maps. Most importantly, the TIP60-C HSA is occluded all along its trajectory and probably cannot serve as a DNA sensor (Fig. 3c and Extended Data Fig. 7c,d). The TIP60-C HSA module also deviates from its counterparts in its architecture. In other remodellers, the three ARPs are placed one next to another21, whereas in TIP60-C there is a gap of 50 Å between the second and third (isolated) ARP, so that the first two are located in NuA4L whereas the third is hosted in the SWR1L part. These differences must be reflected in how these remodellers are recruited to chromatin, and we discuss this later.

The rearranged neck acts as an interaction hub

The neck is a large region that coordinates all yeast NuA4 modules4. This part is completely rearranged in the human complex and bears little similarity to its yeast counterpart (Fig. 4a). The most prominent feature in the yeast neck is a bundle of four long helices donated by three intertwined subunits that serves to strongly couple the TRRAP homologue, Tra1, to the actin module at the core of NuA4 (Fig. 4d,e). In the human version, we observe a highly degenerate design of this bundle, formed by the weak interaction between two short EP400 and DMAP1 helices (Fig. 4a,b). Furthermore, whereas the rest of the yeast neck is characterized by an intricate network of unfolded threads, our structure reveals that in TIP60-C, the neck is dominated by a unique construction made of eleven β-strands, with some similarity to a β-barrel (Fig. 4c). This barrel-like module is composed of strands contributed by all three neck subunits—namely, EP400, EPC1 and DMAP1—and forms a tight interaction between these proteins. Moreover, this construction acts as a hub that coordinates not only other modules of NuA4L, but also the contact between NuA4L and SWR1L, as well as TRRAP. The hub is the structured formation immediately preceding the unfolded EPC1 loop that extends towards the KAT module, and it also contains the EP400 loop that leads to the SANT domain connecting to TRRAP. The hub further interacts with the HSA helix and the ARPs. Finally, this construction immediately succeeds the EP400 counter-helix that returns from SWR1L, embedding EP400 back into NuA4L. The fully rearranged neck domain is thus designed to interconnect NuA4L with SWR1L as well as with TRRAP, and to conserve its role in coordinating the functional TIP60-C modules.

Fig. 4: Comparison between TIP60-C and NuA4.
figure 4

a, Side view of TIP60-C showing in colour the NuA4L lobe and the scaffolding EP400 subunit. b, Enlarged view of the neck domain of human TIP60-C boxed in a. c, View of the β-barrel interaction hub highlighted by dotted lines in b, and rotated by 45°. d, View of yeast NuA4 in which the actin module is aligned on the ACL6A–ACT module of NuA4L. e, Enlarged view of the neck domain of yeast NuA4 boxed in d.

Full size image

Implications for function and mechanism

As we have shown, TIP60-C exhibits fundamental structural differences with respect to its individual yeast counterparts. A completely altered neck architecture and a mobile docking platform for activators sets it apart from NuA4. A static actin module placed close to the core and pointing towards the EP400 ATPase, an occluded HSA with a specific spread of actins along the helix, as well as the absence of the Swc6–Arp6 dimer, distinguish it from SWR1. To gain insight into how these differences and the unique connections between the parts of the supercomplex might affect the TIP60-C mechanism, we first modelled a nucleosome into the SWR1L part of TIP60-C, guided by the structure of yeast SWR1–nucleosome8 (Fig. 5a). The modelled nucleosome is clasped between the ATPase, the last turns of the HSA helix and the third, isolated, ARP that is carried by the HSA (mass-spectrometry data suggest that it is an additional copy of either ACTL6 or actin). We therefore believe that the tip of the HSA helix, and possibly the third ARP, have the role of a grip, analogous to a remodeller with a mechanical motor20. Sequence alignment shows that the tip of the HSA, also known as the post-HSA, is the most conserved part of this helix among all remodellers, thus suggesting a common function (Extended Data Fig. 7a). It was proposed in the INO80 and SWR1 remodellers that the grip maintains the DNA and octamer in place as the ATPase motor pumps DNA into the nucleosome, yielding a gradually growing strain in the DNA and causing a disruption of DNA–histone contacts that could drive histone exchange21. In the INO80 and SWR1 remodellers, the grip and ATPase are located at opposite sides of the nucleosome disc, and are thus separated by a long stretch of DNA (Extended Data Fig. 7b). By contrast, the grip and motor are separated by less than one DNA turn in TIP60-C, allowing the development of considerable strain in the DNA with fewer ATPase strokes. Furthermore, unlike INO80 or SWR1, the TIP60-C grip is able to bind to two DNA segments as the HSA tip contacts nucleosomal DNA close to its dyad, and the third ARP associates with the DNA linker (Fig. 5a). Notably, this position of the third ARP is similar to that of the linker histone H1 in the chromatosome22. We suggest that disruptions of DNA–histone contacts are accumulated and propagated differently in TIP60-C, perhaps reflecting its wider range of activities, which also include the incorporation of the H3.3 histone variant.

Fig. 5: Function of TIP60-C.
figure 5

a, Nucleosome modelled into TIP60-C according to the yeast SWR1 structure8. Distances, and their approximate equivalent lengths of linear DNA base pairs, from the exit site of DNA to the ARP subunits are marked. b, Incorporation of endogenous TIP60-C-bound H2A.Z–H2B into 60N34 nucleosomes (Nuc) in an ATP-dependent manner. H2A.Z incorporated into the nucleosome and H3 (control) are detected by western blots. Bars represent the proportion of H2A.Z incorporated out of the total H2A.Z bound to purified TIP60-C. c, The efficiency of H2A.Z incorporation increases with DNA linker length until 40 bp. ‘Widom’ denotes 147-bp nucleosome core particles, and ‘xNy’ denotes nucleosomes with linker lengths x and y. Dotted line indicates background levels of H2A.Z deposition. d, Incorporation of exogenous H2A.Z–H2B present in excess is stimulated by acetyl-CoA and is not enhanced by nucleosome DNA linker longer than 41 bp. Bars represent the proportion of H2A.Z incorporated into nucleosome out of the total H2A.Z present in the reaction. NAP1L1 is a histone chaperone. e, TIP60-C acetylation of nucleosomes with either H2A or H2A.Z. Acetylation was detected by an antibody for acetyl-lysine. Graphs in b–e represent the average measurement of three independent replicates (bars) with s.d. (error bars). Vertical lines in d,e represent the position where the original blots have been spliced.

Full size image

Other domains of TIP60-C that possibly bind to the nucleosome DNA linker include the NuA4L actin module, which might contact DNA located 95 Å or around 30 bp from the nucleosome core (Fig. 5a). To lend support to our placement of the nucleosome, we compared the H2A.Z deposition activity of TIP60-C on nucleosomes with varied DNA-linker length. We first show that endogenously TIP60-C-bound H2A.Z is incorporated into nucleosomes with two long linkers in an ATP-dependent manner (Fig. 5b). By contrast, nucleosomes with no linker (Widom) yielded a H2A.Z deposition activity that was only slightly higher than background (Fig. 5c). A short 16-bp linker (16N0) already increased the activity significantly, in line with our model, in which such a linker would form considerable interactions with the third ARP. Of note, it has been shown that the conserved actin core of ARPs can indeed bind to DNA9. Increasing the linker length to 41 bp (41N0) resulted in a further substantial increase of exchange efficiency, supporting the likelihood of additional DNA interactions with the NuA4L actin module. The YL1 homologue Swc2 has been shown to sense extranucleosomal DNA in yeast SWR1 (refs. 23,24). We cannot therefore exclude the possibility that the increased activity is partially due to enhanced YL1–DNA interactions, but it seems that Swc2 is mainly implicated in binding longer free DNA. Also, consistent with the third ARP binding to only one of the DNA linkers, addition of a second linker (60N34) did not enhance activity.

Crucially, a linker longer than 41 bp (Fig. 5c,d) increased the remodelling activity only very moderately (60N0, 100N0). This is in sharp contrast to SWR1, in which a marked increase in activity is observed when the linker DNA exceeds 40 bp (ref. 23), and to INO80, in which only a long linker allows the flexible actin module to bind to extranucleosomal DNA and allosterically regulate the ATPase motor21. This points to fundamental differences between TIP60-C, INO80 and SWR1 in their mode of recruitment to chromatin and their regulation. SWR1 and its direct human homologue SRCAP, which lacks a NuA4L part, are thought to be enlisted to chromatin through their high affinity to the long stretches of free DNA (more than 40 bp) that occur at gene promoters23,24. In INO80, and possibly SWR1, the mobile actin module senses such long stretches to couple the ATPase motor to productive cycles20,21,25. Our biochemical tests—which did not find any increase in activity with long DNA—as well as the static actin module and the position of the obstructed HSA helix show that this is not the case for TIP60-C. Instead, the supercomplex probably uses the TRRAP docking platform of transcription activators for recruitment to specific DNA sites. TRRAP, or its yeast homologue Tra1, serve this role in NuA4 and SAGA. However, in these complexes, TRRAP is firmly bound to their cores, and flexibility is endowed to their enzymatic modules4,26. A highly dynamic TRRAP is probably necessary in TIP60-C to allow the relatively rigid and fixed SWR1L part to search for a suitable target. Furthermore, we observe much higher H2A.Z exchange activity when acetyl-CoA is present in the reaction mixture (Fig. 5d). This suggests that acetylation of H2A and H4 on the nucleosome by TIP60-C itself stimulates exchange, in accordance with previous findings in Drosophila and yeast27,28. We suggest therefore the following order of events: TIP60-C is first recruited to chromatin by activators binding to TRRAP; nucleosomes at long range around the site of recruitment are acetylated on H2A and H4, as was observed4,29 for yeast NuA4; and TIP60-C recognizes acetylated nucleosomes through its BRD8 or YEATS4 subunits and exchanges H2A for H2A.Z. Notably, we find that TIP60-C acetylates nucleosomes that already bear H2A.Z much less efficiently, thus preserving its acetylation potential for nucleosomes that could undergo H2A to H2A.Z exchange (Fig. 5e).

SRCAP and TIP60-C are therefore wired to target different sites on chromatin. Indeed, there is little overlap between the in vivo activities of the complexes, because mutations in one cannot be compensated by the other30. We speculate that the contradictory roles of H2A.Z in promoters can be traced, at least in part, to the different machinery that was used to deposit it. Indeed, H2A.Z exchange by TIP60-C entails further acetylation of histones, as well as non-histone targets and the recruitment of additional factors, possibly leading to a different outcome as compared with exchange performed by SRCAP. Merging exchange and acetylation seems to be linked to cell-fate decisions and differentiation. In certain circumstances, SWR1 and NuA4 are joined even in unicellular organisms; for example, in polymorphic Candida albicans strains during the reversible yeast–hyphae transition31. Although TIP60-C is essential for embryonic development in mice32, conditional knockout of Tip60 in differentiated liver cells has only moderate effects33.

Methods

Establishment of the K562 cell line with a SBP-3×Flag-tagged EPC1

To create a K562 EPC1-SBP-3×Flag cell line, 2 × 106 K562 cells (obtained from ATCC) were transfected by electroporation (Amaxa2D, program X001; Lonza solution V) with 2 µg of pX458 plasmid expressing Cas9, eGFP and guide RNA against the EPC1 locus (targeting DNA cleavage immediately 3′ to the EPC1 STOP codon), and 9 µg of EPC1-SBP-3×Flag donor plasmid. Directly after, transfected cells were treated with 2 µM M3814 (Selleckchem), a DNA-PK inhibitor34. Three days after transfection, GFP-positive cells were sorted (BD FACS sorter Melody) and single-cell-derived clones were screened by PCR and Sanger sequencing for precise integration of the SBP-3×Flag tag at the EPC1 locus. Expression of EPC1-SBP-3×Flag was confirmed by western blot using anti-Flag antibody (M2, Sigma Aldrich) and one of the K562 EPC1-SBP-3×Flag clones, named K562 EPC1-SBP-3×Flag #1D1, was expanded for purification of TIP60-C.

Purification of TIP60-C

The K562 cell line, in which the endogenous subunit EPC1 was fused at the C terminus to an SBP-3×Flag affinity tag, was cultured in PRMI medium supplemented with 10% calf serum and 50 μg ml−1 gentamycin. All of the following steps were performed at 0–4 °C. Six-litre cultures were centrifuged (1,000g for 10 min). Nuclear extraction then followed a previously published protocol35 with several modifications. Cell pellets were first washed in cold phosphate-buffered saline (PBS) and then in buffer K75 (10 mM HEPES, pH 7.9, 75 mM KCl, 1.5 mM MgCl2, 1 mM DTT, 0.5 mM PMSF, 5 μg ml−1 pepstaine A and 3 μg ml−1 E64). Cell pellets were resuspended in hypotonic K0 buffer (10 mM HEPES, pH 7.9, 1.5 mM MgCl2, 1 mM DTT, 0.5 mM PMSF, 5 μg ml−1 pepstaine A, 3 μg ml−1 E64 and ROCHE protease inhibitor cocktail) and homogenized in a 100-ml Dounce homogenizer. Sucrose was added (final 10% w/v) and nuclei were pelleted for 25 min at 5,000g. Nuclear pellet was washed once with sucrose buffer (10 mM HEPES, pH 7.9, 1.5 mM MgCl2, 10 mM KCl, 1 mM DTT, 0.5 mM PMSF, 5 μg ml−1 pepstaine A, 3 μg ml−1 E64 and 10% sucrose w/v). Nuclear pellets were resuspended in no-salt buffer (20 mM HEPES, pH 7.9, 25% glycerol, 1.5 mM MgCl2, 2 mM DTT, 0.5 mM PMSF, 5 μg ml−1 pepstaine A, 3 μg ml−1 E64 and ROCHE protease inhibitor cocktail) and homogenized in a 40-ml Dounce homogenizer. NaCl was added to a final concentration of 0.3 M before a 30-min incubation with rotation. The nuclear extract was cleared by centrifugation (30,000g for 30 min) and frozen in liquid nitrogen. Low concentrations of PEG 20,000 as well as 5 mM MgCl2 were added to precipitate some remaining organelles and membrane parts by a short centrifugation (33,000g for 10 min). The PEG 20,000 concentration was then increased to 5–6% and TIP60-C was precipitated in a second short centrifugation step. The pellet was resuspended in a minimal volume and was incubated with anti-flag M2 agarose beads overnight in buffer A (20 mM HEPES pH 8.0, 250 mM sodium chloride, 10% sucrose, 2 mM MgCl2, 0.5 mM DTT, 0.5 mM PMSF, 5 µg ml−1 pepstaine A and 3 µg ml−1 E64). The beads were washed four times with buffer A supplemented with 0.05% Tween-20 and finally eluted with buffer A containing 150 μg ml−1 3×Flag peptide. The eluate was either directly used for histone exchange assays or further purified for cryo-EM. Elution from the Flag beads was mixed with streptavidin beads and incubated for four hours. The beads were washed three times with buffer B (20 mM HEPES pH 8.0, 250 mM sodium chloride, 10% sucrose, 2 mM MgCl2, 1 mM DTT and 0.05% Tween-20) and finally eluted with either buffer C (20 mM HEPES pH 8.0, 250 mM sodium chloride, 10% sucrose, 2 mM MgCl2, 1 mM DTT, 20 mM biotin and 0.05% Tween-20) for the in vitro acetylation activity test, or buffer D (20 mM HEPES pH 8.0, 250 mM sodium chloride, 1% trehalose, 2 mM MgCl2, 1 mM DTT, 20 mM biotin and 0.0025% dodecyl-maltoside) for cryo-EM analysis. TIP60-C was flash-frozen in liquid nitrogen and stored at −80 °C.

Reconstitution of histone octamers and preparation of nucleosome DNA

Octamers were reconstituted from individual Xenopus laevis (canonical) or human (H2A.Z) histones expressed as inclusion bodies according to the standard protocol36,37.

The biotinylated nucleosomal DNAs Widom, 16N0, 41N0, 60N0, 60N34 and 100N0 were generated by PCR (‘xNy’ corresponds to nucleosomes with linker lengths x and y). In-house purified Phusion polymerase was used for the PCR reaction on the following template:

CGACGGCCAGTGAACCACGATTCGGTACTCGGGTTCTAGACCATGATTACGCCAAGCTTTTCCTATGACTCATCCAGTTCTGCAGGCGATCACTACATGCACAGGATGGCTAGCTCTGACACGTGCCTGGAGACTAGGGAGTAATCCCCTTGGCGGTTAAAACGCGGGGGACAGCGCGTACGTGCGTTTAAGCGGTGCTAGAGCTGTCTACGACCAATTGAGCGGCCTCGGCACCGGGATTCTCGCATCTAACTCCCGGGTGTGCCTATAAAAGGAT (The Widom-601 positioning sequence is in bold and the PstI restriction site is underlined). The primers used for the PCR reactions were: biotin–GGCGATCACTACATGCACAGGATGGCTAGC for the 16-bp linker, biotin–AGCCGGAGGACAGTCCTCCGCTGCAGGCGATCACTACATGCACAGGATG for the 41-bp linker, biotin–GATTACGCCAAGCTTAGCCGGAGGACAGTCCTCCGCTGCAGGCGATCACTACATGCACAGGATG for the 60-bp linker, biotin–CGACGGCCAGTGAACCACGATTCGG for the 100-bp linker, TGCGAGAATCCCGGTGCCGAGGCCG for the 0-bp linker reverse primer and ATCCTTTTATAGGCACACCCGGGAGTTAGATGCG for the 34-bp linker reverse primer. The PCR products were purified on Macherey-Nagel nucleospin plasmid columns (ref. 740588.50; 1 column per 0.5 ml PCR mixture) using solutions from Macherey-Nagel Gel and PCR Clean-up kit (ref. 740609.50) according to PCR clean-up protocol. The eluted DNA was precipitated with 0.3 M sodium acetate and 70% ethanol. DNA pellet was dissolved in 1× TE buffer (10 mM Tris-HCl pH 7.5, 1 mM EDTA).

Reconstitution of nucleosomes

Nucleosomes with a 147-bp Widom-601 positioning sequence with linkers of different length were prepared according to the NEB Dilution Assembly Protocol (E5350) (https://international.neb.com/protocols/2012/06/02/dilution-assembly-protocol-e5350) with the following modifications: 2.75 µM DNA was mixed with 2.75 µM canonical or H2A.Z histone octamers in a solution containing 2 M NaCl, 1 mM EDTA and 5 mM β-mercaptoethanol. The solution was incubated for 30 min at room temperature and then underwent serial dilutions down to 1.48 M, 1 M, 0.6 M and 0.25 M NaCl with low-salt buffer (10 mM HEPES-KOH pH 8.0 and 2.5 mM β-mercaptoethanol). After each dilution, the solution was incubated at room temperature for 30 min. To reduce the final NaCl concentration, nucleosomes were concentrated in 0.5-ml 100-kDa cut-off Amicon Ultra filter up to 100 µl, then diluted five times with low-salt buffer. This step was repeated one more time. Finally, nucleosomes were concentrated to 3–4 µM and analysed in a 5% native 0.2× TBE polyacrylamide gel to ascertain the quality of the sample and the absence of free DNA.

Reconstitution of the H2A.Z–H2B dimer

Two milligrams of lyophilized monomers of H2A.Z and H2B were dissolved in 7 M guanidine hydrochloride, 20 mM Tris pH 7.5 and 10 mM DTT at room temperature, and mixed in stoichiometric amounts. The mixture was dialysed against three changes of refolding buffer (10 mM Tris pH 7.5, 2 M NaCl, 1 mM EDTA and 7 mM β-mercaptoethanol), and H2A.Z–H2B dimer was purified over a Superdex 200 column (GE Healthcare) pre-equilibrated with GF buffer (20 mM Tris pH 7.5, 2 M NaCl, 1 mM EDTA and 0.5 mM TCEP). Glycerol was added (final concentration 20% w/v) to H2A.Z–H2B dimers, which were then aliquoted, flash-frozen in liquid nitrogen and stored at −80 °C.

Histone acetyltransferase assay

Reactions were performed in 20 mM HEPES (pH 7.5), 50 mM NaCl, 50 μM acetyl-CoA, 0.1 mg ml−1 BSA, 2 mM MgCl2, 0.2 mM TCEP, 1 μM ZnSO4, 5 μM trichostatin A, 10% (v/v) glycerol and 1 μM nucleosome. The reactions were initiated by adding 10 nM TIP60-C. After an incubation at 30 °C for 30 min, the reactions were stopped by adding SDS loading buffer and heating at 95 °C for 3 min. Proteins were then resolved by 15% SDS–PAGE and analysed by western blot, applying a primary antibody against acetylated lysine (Cell Signaling Technology, 9441S). The levels of histone H3 revealed by primary antibodies (Cell Signaling Technology, 14269) were used as a control.

Histone exchange assay

To test whether the TIP60-C-bound H2A.Z–H2B (that is, endogenous histones that are part of the purified complex) can be incorporated by TIP60-C into nucleosomes, 200 nM of biotinylated mono-nucleosomes with the indicated DNA linker length were mixed with 20 nM TIP60-C in 20 μl histone exchange buffer (20 mM HEPES, pH 7.5, 50 mM NaCl, 5 mM MgCL2, 10% glycerol, 0.1 mg ml−1 BSA, 1 mM ATP or ATPγS and 0.05% Tween-20), and incubated at 30 °C for 2 h. Ionic strength was increased by the addition of 0.6 μl 5 M NaCl to prevent non-specific association of TIP60-C with beads and the reaction mix was then incubated (6 °C, 30 min, 1,250 rpm shaking in ThermoMixer C) with 6 μl streptavidin magnetic beads (Thermo Fisher Scientific, 88816) pre-equilibrated with washing buffer (20 mM HEPES, pH 7.5, 250 mM NaCl, 5 mM MgCl2, 10% glycerol, 0.1 mg ml−1 BSA and 0.05% Tween-20). Beads were then pelleted and resuspended in 50 μl washing buffer with 2 mM ATPγS for 5 min at room temperature to further eliminate any non-specific association of TIP60-C. After three times of washing as above, 10 μl of 1× SDS loading buffer was added and heated at 95 °C for 3 min. Proteins were then resolved by 15% SDS–PAGE that was followed by a western blot with a primary antibody against H2A.Z (Cell Signaling Technology, 2718). Western blot using primary antibody against H3 (Cell Signaling Technology, 14269) was done as control.

When testing H2A.Z exchange by TIP60-C in the presence of excess H2A.Z–H2B dimers, we encountered the difficulty presented by high-background due to non-specific binding of H2A.Z to nucleosome linker DNA as well as to salmon sperm DNA that associates with the beads28. To overcome this issue, we took advantage of a PstI site located on the long DNA linker close to the nucleosome core. Cutting with the restriction enzyme released the core nucleosome from the beads, carrying with it only traces of non-specifically bound H2A.Z. For each reaction, 200 nM of biotinylated mono-nucleosomes were mixed with 20 nM TIP60-C and 400 nM H2A.Z–H2B dimer in 20 μl histone exchange buffer (20 mM HEPES, pH 7.5, 50 mM NaCl, 5 mM MgCL2, 10% glycerol, 0.1 mg ml−1 BSA, 1 mM ATP and 0.05% Tween-20, with or without 50 μM acetyl-CoA), and incubated at 30 °C for 2 h. Two microlitres of quenching solution (4 mg ml−1 salmon sperm DNA, 100 mM EDTA and 10 mM ATPγS) was added, and incubated at 37 °C for 30 min. The reaction mixture was then incubated with 6 μl streptavidin magnetic beads for 30 min at 6 °C. Beads were pelleted and then resuspended in 50 μl washing buffer for 5 min at room temperature. After three washes, 20 μl of restriction enzyme solution (20 mM HEPES, pH 7.5, 50 mM NaCl, 5 mM MgCl2, 10% glycerol, 0.1 mg ml−1 BSA, 0.05% Tween-20 and 20 U PstI-HF) was added, and incubated at 37 °C for 1 h at 1,250 rpm in a ThermoMixer C (Eppendorf). The supernatant containing the released nucleosomes was collected, mixed with SDS loading buffer and heated at 95 °C for 3 min. The results were analysed by western blot as above. Quantification of signal intensity was performed on digitally acquired images (ImageQuant 800, Cytiva Life Sciences) by ImageJ. All statistical tests were performed using Prism.

Cryo-EM sample preparation and data acquisition

Three microlitres of sample was applied onto a holey gold grid (UltrAuFoil R1.2/1.3 300 mesh) rendered hydrophilic by a 90-s treatment in a Fischione 1070 plasma cleaner operating at 34% power with a gas mixture of 80% argon:20% oxygen. The grid was blotted for 1.5 s at blot force 8 and flash-frozen in liquid ethane using a Vitrobot Mark IV (FEI) at 4 °C and 95–100% humidity. Two different datasets were collected. The first dataset was acquired on a Cs-corrected Titan Krios (Thermo Fisher Scientific) microscope operating at 300 kV in nanoprobe mode using serialEM for automated data collection38. Movie frames were recorded 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 acquired hardware-binned at a nominal magnification of 81,000×, which yielded a pixel size of 0.862 Å. Forty movie frames were recorded at a dose of 1.12 and 1.30 e− per Å2 per frame. To improve resolution on the SWR1L and NuA4L parts, a second image dataset was acquired on a Titan Krios G4 microscope operating at 300 kV in nanoprobe mode using SerialEM for automated data collection. Movie frames were recorded on a Flacon 4i direct electron detector after a Selectris X energy filter using a 10-eV slit width in zero-loss mode. Images were acquired at a nominal magnification of 165,000×, which yielded a pixel size of 0.73 Å.

Image processing

For the first dataset, WARP was used to perform the initial pre-processing steps, align movie frames, perform dose-weighting, correct the beam-induced specimen motion and estimate the contrast transfer function (CTF)39. After visual inspection, images with poor CTF, showing particle aggregation or abundant ice contamination were discarded. The second dataset was pre-processed using cryoSPARC40. Particle coordinates were determined using crYOLO41 for both datasets. The datasets were analysed in RELION-3.1 (ref. 42) and cryoSPARC according to standard protocols. In brief, three rounds of reference-free two-dimensional (2D) classification of the individual particle images were performed in cryoSPARC to remove images corresponding to contaminating or damaged particles and ice contaminations. References (3D models) were generated by the ab initio 3D reconstruction program of cryoSPARC. These structures were then used as references for 3D classification jobs in cryoSPARC and particles corresponding to high-resolution 3D classes were selected and used for non-uniform refinement. Refined particles were subjected to 3D classification in RELION without alignment using various regularization parameter (T) values. Particles corresponding to high-resolution classes were used in the subsequent non-uniform refinement in cryoSPARC. We performed a focused refinement of different parts of TIP60-C as follows. Particles were binned to a pixel size of 2 Å and were subjected to 3D classifications in RELION using masks. After visual inspection, particles corresponding to the best classes were refined in cryoSPARC locally with a mask covering the region of interest. Global resolution estimates were determined using the Fourier shell correlation (FSC) = 0.143 criterion after a gold-standard refinement. Local resolutions were estimated with cryoSPARC. To analyse the heterogeneity in the cryo-EM map owing to the presence of the flexible TRRAP module (430 kDa), we used the neural-network-based cryoDRGN43 and OPUS-DSD44 reconstruction to map the particles on two principal components. We partitioned the latent space into 20 regions and a density map was generated from the centre of each region.

Model building

RuvBL1–RuvBL2 hexamer was extracted from the human INO80–nucleosome structure (Protein Data Bank (PDB) 6HTS), placed into the map by rigid body fitting in Chimera45 and used as a starting point for manual editing in Coot46. The EP400 ATPase C-lobe was modelled in AlphaFold47, docked into the map by ADP_EM48 and manually adjusted according to density. The EP400 insertion into the hexamer was then manually built. This was facilitated by secondary structure prediction in RaptorX49. The ATPase N-lobe was modelled in AlphaFold and docked by ADP_EM into the map generated by focused refinement of this region in SWR1L. An actin molecule was docked into the same map as the third ARP. Mass spectrometry suggests that this protein is an additional copy of either actin or ACTL6A. The TRRAP structure was extracted from PDB 7KTR and fitted into the corresponding map. Manual adjustment of TRRAP and building of some additional loops, where density was clearly visible, were carried out. The structure of yeast NuA4 (PDB 7ZVW) served as a starting point for building the actin module. The rest of the NuA4L core and the novel interaction hub were manually built into a map generated by focused refinement on NuA4L. Manual model building was guided by secondary structure prediction and the density of bulky side chains (Lys, His, Arg, Phe, Tyr and Trp). The YL1– VPS71 subunit was also manually built, as were the HSA helix and the counter-helix. The atomic model was refined in PHENIX by real-space refinement with secondary structure restrains50 and in Isolde51. All display images were generated using ChimeraX52.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.