Structural insights into the π-π-π stacking mechanism and DNA-binding activity of the YEATS domain

The YEATS domain has been identified as a reader of histone acylation and more recently emerged as a promising anti-cancer therapeutic target. Here, we detail the structural mechanisms for π-π-π stacking involving the YEATS domains of yeast Taf14 and human AF9 and acylated histone H3 peptides and explore DNA-binding activities of these domains. Taf14-YEATS selects for crotonyllysine, forming π stacking with both the crotonyl amide and the alkene moiety, whereas AF9-YEATS exhibits comparable affinities to saturated and unsaturated acyllysines, engaging them through π stacking with the acyl amide. Importantly, AF9-YEATS is capable of binding to DNA, whereas Taf14-YEATS is not. Using a structure-guided approach, we engineered a mutant of Taf14-YEATS that engages crotonyllysine through the aromatic-aliphatic-aromatic π stacking and shows high selectivity for the crotonyl H3K9 modification. Our findings shed light on the molecular principles underlying recognition of acyllysine marks and reveal a previously unidentified DNA-binding activity of AF9-YEATS.

A large number of epigenetic marks or posttranslational modifications (PTMs) in histones have been discovered over the past few years 1,2 . One of the major and widespread PTMs is acylation of the ε-amino group of lysine residues. Acylation neutralizes the positive charge and increases the hydrophobic character of the lysine side chain and alters chromatin structure utilizing two fundamental mechanisms. It weakens nonspecific electrostatic interactions between histones and DNA, leading to a more open and transcriptionally active chromatin. It also recruits acyllysine readers and their host proteins and complexes to specific genomic loci to facilitate diverse epigenetic-driven nuclear programs essential in chromatin remodeling, gene transcription, and DNA replication, recombination and repair 3,4 . At least nine acyllysine modifications in histones have been identified, including crotonylation.
Since the discovery of lysine crotonylation in mammalian cells, this PTM has drawn much attention becoming an increasingly important epigenetic mark 5,6 . Crotonylation is enriched around active gene promoters and potentially enhancers and was found to stimulate gene transcription to a higher degree than the corresponding acetylation 6 . Studies of genomic distribution of crotonyllysine and acetyllysine reveal some differences, which suggest that these modifications are not redundant and can be associated with distinct biological outcomes 5,6 . Nevertheless, a substantial overlap in genomic localization and similarities in chemical properties present a challenge in studying and distinguishing the biological roles of these PTMs. A strategy of employing separate readers as probes for targeting individual acyllysine modifications has not been well developed, in part because of the promiscuous nature of currently known readers. Canonical acyllysine readers, bromodomain and double PHD finger (DPF), display comparable binding capabilities toward various short acyl chain modifications [7][8][9][10][11][12] , and while the YEATS domain exhibits preference for crotonyllysine, it still associates albeit more weakly with other acyllysine marks [13][14][15][16][17][18][19][20][21][22] . We have previously shown that the YEATS domains of Taf14 (Taf14-YEATS) and AF9 (AF9-YEATS) recognize crotonyllysine through a non-canonical π-π-π stacking mechanism 16,17 , and in the case of Taf14-YEATS it involves aromatic-amide-aromatic π stacking and aromatic-aliphatic-aromatic π stacking 16 (Fig. 1a). In this study, we elucidate the significance of π stacking components in targeting of acetyllysines by the YEATS domains and report a previously uncharacterized DNA-binding activity of the human AF9-YEATS domain. We also show how the unique nature of the conjugated π system of the crotonyl modification could aid in the design of effectors indifferent to saturated acyllysines.

Results and discussion
Structural insight into the selectivity of Taf14-YEATS. To determine the contribution of π stacking to the binding energetics, we measured binding affinities of Taf14-YEATS to the histone H3K9acyl peptides containing unsaturated and saturated four-carbon acyl modifications. Tryptophan fluorescence measurements showed that binding of Taf14-YEATS was reduced~3fold when the crotonylated modification was replaced with a similar in length but saturated butyryl modification (Figs. 1b, c and Supplementary Figure 1). The binding was further decreased to a negatively charged succinyllysine, and no interaction was detected with a branched hydroxyisobutyryllysine in 1 H, 15 N heteronuclear single quantum coherence (HSQC) NMR titration experiments (Supplementary Figure 2). The selectivity of the YEATS domain for the crotonyllysine modification was substantiated by pulldown assays with biotinylated histone peptides (Fig. 1d). GST-fused Taf14-YEATS recognized the H3K9cr peptide but associated weaker with H3K9bu and H3K9ac peptides.
Together these results show that Taf14-YEATS selects for unbranched acyl modifications, and the alkene group in the modification enhances the protein-binding capability.
To gain insight into the molecular basis for the enhancement, we obtained the crystal structure of Taf14-YEATS in complex with H3K9bu peptide and compared it with the previously reported structure of this domain in complex with H3K9cr 16 . The structures were superimposed with a root mean square deviation of 0.1 Å, indicating overall conservation of the binding mode. As in the case of crotonyllysine, butyryllysine transversed a narrow tunnel and was sandwiched between two aromatic residues of the protein, W81 and F62 (Fig. 1e, Supplementary Figure 3a and Supplementary Table 1). Amide nitrogen and carbonyl oxygen of butyryllysine were restrained through hydrogen bonds with the hydroxyl group of T61 and backbone amide of W81 and through a water-mediated hydrogen bond with G82. A similar set of polar interactions stabilized the complex with crotonyllysine, however, substantial differences were observed in the position of W81. It adopted a single conformation, rotamer 1 (r1), involved in the π stacking interaction with butyryl amide in the Taf14 YEATS-H3K9bu complex. In contrast, when complexed with H3K9cr, W81 adopted two conformations, with r1 and r2 providing maximum π stacking with the crotonyl amide and the crotonyl alkene moiety, respectively ( Supplementary Figures 3b  and 4). Neither the complex with propionylated H3K9pr peptide nor the complex with acetylated H3K9ac peptide had the r2 conformation of W81, indicating that it is a unique feature available to the unsaturated acyl modification but not to saturated acyllysine modifications (Fig. 1f).
Design of the crotonyllysine-specific Taf14-YEATS. Comparison of the H3K9bu-and H3K9cr-bound Taf14-YEATS structures revealed that the side chain of W81 is located within 4 Å of the α carbon atom of G82. Modeling in G82A mutation suggested that the presence of an additional methyl group at position 82 would hinder the ability of W81 to adopt the r1 conformation required for the π amide stacking. To test this hypothesis, we generated the G82A mutant of Taf14-YEATS and determined its crystal structure (Fig. 2a). In support of our idea, the indole moiety of W81 in this mutant was in a tilted conformation incompatible with r1 ( Fig. 2b). We next assessed the acyllysinebinding activity of Taf14-YEATS G82A using NMR titration experiments (Fig. 2c). Large chemical shift perturbations (CSPs) in 15 N-labeled Taf14-YEATS G82A upon addition of H3K9cr peptide indicated that the mutant retains its ability to recognize the crotonyl modification. However, association of Taf14-YEATS with the H3K9ac peptide was almost negligible, as very small CSPs were observed ( Fig. 2c and Supplementary Figure 5b). Tryptophan fluorescence measurements yielded a K d of 124 μM for the interaction of Taf14-YEATS G82A with H3K9cr (Fig. 2d). Although binding affinity of the G82A mutant to H3K9cr was reduced compared with the binding affinity of the wild-type protein, it remains in the range of affinities exhibited by the wellestablished readers of acetyllysine, bromodomains 7,9 . We next characterized the Taf14-YEATS G82A mutant in the cellular context. We have previously shown that abrogating the Taf14 association with H3K9acyl through mutating W81 to alanine impacts the transcript levels of a variety of yeast genes both positively and negatively 14 . To determine the effect of blocking Taf14 from selectively associating with H3K9ac but not with H3K9cr, we carried out a real-time quantitative PCR (qPCR) analysis for a set of Taf14-regulated genes in the TAF14 deletion strain (taf14Δ) rescued with a vector only, wild-type TAF14, or with a form carrying either the W81A or G82A mutation in the YEATS domain (taf14 W81A and taf14 G82A) (Fig. 2e). As expected, taf14Δ deletion impacted the expression of multiple genes that were previously reported to be regulated by Taf14 interaction with H3K9acyl, including YBL041, YPR145, YPR158, YER145C, YKL150W, and YKR093W 14 . For control, we also included several genes, such as YLR290, YAL017,and MRPL10, which were shown not to be regulated by Taf14 14 . Intriguingly, although the effect of the entirely loss-of-function taf14 W81A mutant was largely consistent with the effect seen with the loss of Taf14, the taf14 G82A mutant showed more variable results (Fig. 2e). In some cases, the taf14 G82A mutant impacted the expression of genes similar to that of the taf14 W81A mutant (YBL041, YPR158, YKL150W, and YKR093W), whereas in other cases, the taf14 G82A mutant showed distinct effects on gene transcripts not seen in either the taf14 W81A mutant or the TAF14 deletion (YPR145 and YER145C). Given the taf14 G82A mutation impacts H3K9ac binding to a greater degree than H3K9cr binding, these findings suggest a differential requirement of H3K9ac and H3K9cr for the proper expression of Taf14regulated genes.
Crotonyllysine recognition by Taf14-YEATS G82A. To elucidate the mechanism by which Taf14-YEATS G82A distinguishes crotonyllysine and discriminates against saturated counterparts, we obtained the crystal structure of the H3K9cr-bound G82A mutant (Fig. 3a). The structure of the complex showed that the aromatic side chains of W81 and F62 lay parallel to each other and at equal distance of 3.4-3.6 Å from the crotonyl alkene group (Fig. 3b). Remarkably, only a single r2 conformation of W81 was observed in the structure of the complex. Analysis of molecular orbitals (MOs) for the W81, Kcr, and F62 assembly at the B3LYP/ 6-311 + G(d,p) level of density functional theory (DFT) revealed that the bonding MOs transcending all three π systems are deeply buried, indicating considerable synergism of the π-π-π stacking interaction (Supplementary Figure 6). A representative MO, the highest occupied MO HOMO-8, is shown in Fig. 3c. We concluded that the ability of the W81 r2 rotamer to engage the crotonyl alkene moiety most likely accounts for the selectivity of Taf14-YEATS G82A.
AF9-YEATS is a more versatile acyllysine reader. Alignment of the YEATS domain sequences derived from the family of human and yeast proteins shows conservation of the WG motif (W81 and G82 in Taf14) in all YEATS members but AF9 and ENL, which instead contain a YA motif (Fig. 4a)  the presence of an alanine following the aromatic Y78 residue in AF9, the YEATS domain of AF9 did not differentiate between acylated modifications on H3K9 and bound almost equally well to H3K9cr, H3K9bu, and H3K9ac, supporting the original findings 17 ( Fig. 4b and Supplementary Figure 7). To compare the binding modes of the YEATS domains harboring the WG and YA motifs, we determined the structure of AF9-YEATS in complex with H3K9bu peptide. Structural overlay of the AF9-YEATS: H3K9bu and previously reported AF9-YEATS:H3K9cr 17 complexes showed that both saturated and unsaturated four-carbon acyl chains of K9 are bound by the three aromatic residues Y78, F59, and F28 in a very similar way (Fig. 4c). Short (<4 Å) distances observed between C γ , C δ , C ε , and C η atoms of Y78, the amide group of acyllysine, and the C ε and C ζ atoms of F59 indicated that Y78 and F59 of AF9 are involved in the aromaticamide-aromatic π-π-π stacking in either complex (Fig. 4c, yellow dashed lines). However, neither Y78 nor F28 appear to form π-π-π stacking with the alkene group of crotonyllysine as only a single carbon atom in Y78 (C ε ) or F28 (C ε ) was close enough to the C = C carbon atoms to fully engage the double bond (Fig. 4c, red dashed lines). This inability to form the additional energetically favorable contact is likely the reason behind the indifference of the YEATS domain of AF9 toward crotonyl-, butyryl-, and acetyl-H3K9acyl modifications.
We next explored the possibility of engineering an AF9-based acyllysine-specific reader through mutating Y78 in AF9-YEATS to a tryptophan and therefore re-creating the WA motif seen in the Taf14 Fig. 4 Structural insight into the selectivity of AF9-YEATS. a Alignment of the YEATS domain sequences: absolutely, moderately, and weakly conserved residues are colored green, orange, and pink, respectively. b Binding affinities of AF9-YEATS to indicated histone peptides. Values represent the average of three separate experiments with error calculated as the SD between the runs. c Structural overlay of the acyllysine-binding sites in the AF9-YEATS:H3K9bu (pink) and AF9-YEATS:H3K9cr 17 (light gray) complexes. Yellow dashed lines represent short distances indicative of the aromatic-amide-aromatic π stacking interaction in AF9-YEATS:H3K9cr 17 . Red dashed lines represent short, <4 Å, distances between the crotonyl alkene group and the aromatic residues. d Superimposed 1 H, 15  AF9-YEATS Y78W, we obtained the crystal structure of AF9-YEATS Y78W in complex with H3K9cr peptide (Fig. 4f). In this complex, W78 was in a single r1 conformation, which was engaged in π stacking with the crotonyl amide. Notably, A79 imposed steric hindrance on W78 to a higher degree than on wild-type Y78 at this position, resulting in a tilted conformation of W78 (Supplementary Figure 8b). W78 is incapable of adopting the r2 conformation necessary for π stacking with the alkene group most likely because of steric hindrance from F28 (Supplementary Figure 8c). Additional mutation of A79 to a glycine in AF9, to recapitulate the Taf14' WG motif, failed to induce the selectivity as the AF9-YEATS Y78WA79G mutant bound equally well to either H3K9cr or H3K9ac peptide (Fig. 5a). By contrast, the W81Y mutant of Taf14-YEATS retained its capability to select for H3K9cr, suggesting that position of the tyrosine residue in this mutant allowed for π stacking with the alkene group ( Fig. 5a and Supplementary Figure 9).
AF9-YEATS binds to DNA. Analysis of electrostatic surface potential of AF9-YEATS reveals that the part of the domain opposite to the H3K9cr-binding site is highly positively charged. Particularly, two apparent clusters of basic residues on the protein surface suggest that they could be involved in binding to negatively charged DNA (Fig. 5b). To determine whether AF9-YEATS is capable of interacting with DNA, we examined association of AF9-YEATS with 147-bp 601 Widom DNA using an electrophoretic mobility shift assay (EMSA). The 601 DNA was incubated with increasing amounts of AF9-YEATS and the reaction mixtures were resolved on native polyacrylamide gels (Fig. 5c). A gradual increase in the amount of added AF9-YEATS resulted in a shift and disappearance of the 601 DNA band and the appearance of the bands corresponding to multiple complexes formed between AF9-YEATS and multiple major/minor grooves 23 of 601 DNA (Fig. 5c). Direct binding to DNA was corroborated by NMR titration experiments, in which substantial CSPs were observed in 15 N-labeled AF9-YEATS upon gradual addition of 601 DNA (Fig. 5d and Supplementary Figure 10a, b).
To assess how AF9-YEATS interacts with both H3K9acyl and DNA, we reconstituted H3K9acyl-containing nucleosome core particles (H3K9cr-NCPs and H3K9ac-NCP) and examined binding of AF9-YEATS to these NCPs and to unmodified NCP in EMSA. Incubation of increasing amounts of AF9-YEATS with H3K9cr-NCP led to a disappearance of the H3K9cr-NCP band and of free 601 DNA present in the sample, indicating formation of the AF9-YEATS:H3K9cr-NCP and AF9-YEATS:601 DNA, respectively, complexes (Fig. 5e). We note that the free 601 DNA band disappeared faster than the H3K9cr-NCP band, implying that AF9-YEATS prefers a more accessible, free DNA to the DNA wrapped around the nucleosome, despite the fact that H3K9cr-NCP contains the additional binding partner of the YEATS domain-H3K9cr. Yet, the recognition of H3K9cr by the YEATS domain is required for strong interaction, as the AF9-YEATS association with unmodified NCP was substantially compromised (Fig. 5f) and its association with H3K9ac-NCP was slightly diminished (Supplementary Figure 10c). In further support, the H3K9cr-NCP binding capability of the AF9-YEATS F59A/Y78A mutant, which is defective in H3K9acyl binding 22 , was also markedly decreased (Fig. 5g).
DNA-and H3K9cr-binding sites in AF9-YEATS do not overlap. To map the DNA-binding site of AF9-YEATS, we plotted NMR CSPs induced in each amide of 15 N-labeled AF9-YEATS by 601 DNA and compared with CSPs induced in this protein by the H3K9cr peptide (Figs. 6a, b). Notably, residues located in and around the H3K9cr-binding site, including Y78, were most perturbed upon addition of the H3K9cr peptide (Figs. 6a, c). However, an entirely different set of residues was perturbed due to binding of 601 DNA (Figs. 6b, c). The most pronounced changes were observed in the residues located in three AF9-YEATS regions, i.e., the 60th, 90th, and 130th patches (Figs. 6c,  d). These patches contain the surface exposed positively charged R61, K63, R64, K67, K92, R96, K97, R133, K134, and K137 residues that can be directly or indirectly involved in electrostatic interactions with negatively charged 601 DNA.
To determine the contribution of the perturbed patches to DNA binding, we generated four mutants of AF9-YEATS, including R61E/K63E/K67E, K92E/R96E/K97E, R96E/K97E, and R133E/K134E/K137E. Among them, only the R61E/K63E/K67E (thereafter referred to as RKK) mutant was soluble and folded. Binding of the AF9-YEATS RKK mutant to 601 DNA and H3K9cr-NCP was examined by EMSA. As shown in Figs. 6e-g, the DNA-binding and H3K9cr-NCP-binding abilities of this mutant were markedly reduced compared with the respective binding abilities of WT AF9-YEATS, pointing to a critical role of the 60th patch residues of AF9-YEATS in the interaction with DNA (Fig. 6c). Taken together, the EMSA results and NMR CSPs indicate that the 60th patch residues are part of the DNA-binding site of AF9-YEATS. A lack of mutually perturbed residues due to the interaction with H3K9cr peptide or 601 DNA further suggests that the H3K9cr-and DNA-binding sites of AF9-YEATS do not overlap.
Bivalent association of AF9 is not conserved in Taf14-YEATS. To identify the DNA fragments smaller than 147 bp of 601 DNA that can be bound by AF9-YEATS, we carried out EMSA assays with 20-bp DNA and 15-bp DNA (Fig. 7a and Supplementary  Figure 11a). Both 15-and 20-mer DNA bands disappeared in an AF9-YEATS concentration-dependent manner, and quantitative analysis of the band intensities yielded K d s of 47 μM and 57 μM for 15-mer DNA and 20-mer DNA, respectively (Fig. 7b), which is in the range of DNA-binding affinities of other transcriptional regulators 24 . Overall, these and the results described above suggest that AF9-YEATS interacts with chromatin in a bivalent manner, i.e., it binds both H3K9acyl and DNA (Fig. 7c).
Is the bivalent interaction conserved in other YEATS domains? Alignment of the YEATS sequences reveals that all three positively charged patches are conserved in the YEATS domain of another human protein, ENL, and thus ENL-YEATS is likely capable of binding to DNA (Fig. 7d). Although these patches are not strictly conserved in the YEATS domain of the human protein YEATS2, the "DNA-binding" surface in YEATS2-YEATS is also highly positively charged (Supplementary Figure 11b). Altogether, these observations suggest that the DNA-binding function may be conserved in the family of human YEATS readers. In contrast, the corresponding Taf14-YEATS surface is enriched in the negatively charged residues that can electrostatically repulse DNA (Fig. 7e). Indeed, EMSA experiment using 601 DNA showed that even a large, up to~10 3 -fold, excess of Taf14-YEATS failed to shift the DNA band, implying that this domain does not bind DNA (Fig. 7f). Interestingly, similar to Taf14-YEATS, the YEATS domain of another yeast protein, Yaf9, also has a negatively charged and therefore repulsive of DNA surface opposite to its H3K27acyl-binding site, which may indicate that the ability to interact with DNA is lost in yeast YEATS domains (Supplementary Figure 11b).

Methods
HKL3000 was used for indexing, scaling, and data reduction 31 . The phase solutions were solved by molecular replacement with Phaser using a modified Taf14 or AF9-YEATS domain as a search model (PDB 5D7E and 4TMP, respectively). Model building was carried out with Coot and refinement was performed with Phenix.Refine 32,33 . Coot and Molprobity were used to verify the model fit to density and model quality 32,34 . The crystallographic and refinement statistics are shown in Supplementary Table 1.
where [L] is concentration of free acetyllysine, [P] is concentration of the protein, Δδ is the observed chemical shift change, and Δδ max is the normalized chemical shift change at saturation. Normalized chemical shift changes were calculated using the equation where Δδ is the change in chemical shift in parts per million (ppm).
Fluorescence spectroscopy. Spectra were recorded at 25°C on a Fluoromax-3 spectrofluorometer (HORIBA). The samples containing 1-2 µM wild-type or mutated YEATS domains in PBS buffer pH 7.4 (Taf14) or 20 mM Tris pH 7.5, 500 mM NaCl, and 2 mM BME (AF9) and progressively increasing concentrations of H3 peptides, H3K9cr, H3K9ac, or H3K9bu (all aa 5-13) (Synpeptide) were excited at 295 or 280 nm. Emission spectra were recorded between 300 and 380 nm with a 1 nm step size and a 0.5 s integration time. K d values were determined using a nonlinear least-squares analysis and the equation: where [L] is the concentration of the histone peptide, [P] is the concentration of the protein, ΔI is the observed change of signal intensity, and ΔI max is the difference in signal intensity of the free and bound states of the protein. K d values were averaged over three separate experiments, and error was calculated as the standard deviation between the runs.
Real-time qPCR. Real-time qPCR was performed essentially as described 35 . Briefly, wild-type or taf14 mutant yeast cells were cultured in YPD overnight at 30°C and cell pellets corresponding to 10 OD 600 equivalents were collected. Total RNA was extracted using the phenol/chloroform method as described 35 and RNA was treated with DNase I to eliminate residual DNA. In all, 2 mg of RNA was used to produce complementary DNA (cDNA) with random hexamer primers and reverse transcriptase IScript (BioRad). cDNA was diluted to 0.1 mg/ml and SYBR green master mix was used for real-time qPCR. ACT1 was also examined and used as a normalization control. The expression of all Taf14 proteins were equal in abundance, indicating that the expression effects observed are due to Taf14 acyllysinebinding defects. The data shown are from three biological replicates.
Expression and purification of H3 mutants H3K9ac and H3K9cr. To incorporate N-ε-acetyl-L-lysine (AcK) into the K9 position of histone H3, pETDuet-1 vector encoding human histone H3 with amber stop codon (pETDuet-1-H3K9TAG) introduced at H3K9 was used to co-transform E. coli BL21 (DE3)-ΔcobB strain with pEVOL-MmAcKRS. Single colony was picked and inoculated in 2YT medium supplemented with 100 mg/L ampicillin and 34 mg/L chloramphenicol. When OD reached to 0.6, H3 expression was induced at 37°C by adding 0.5 mM IPTG, 0.2% (w/v) L-arabinose and 5 mM AcK into cell culture. Cells were harvested 6 h after induction, and purified in the same steps as previously reported 36 . The whole procedure of CrK (N-ε-crotonyl-L-lysine) incorporation into the K9 position of histone H3 was identical with AcK incorporation, except that pEVOL-MmPylRS-384W vector was used, and 1 mM of CrK (N-εcrotonyl-L-lysine) was added into 2YT medium after induction.
Assembly of nucleosomes. Recombinant histone His-TEV-H2A, His-TEV-H2B, and His-SUMO-TEV-H4 were purified in the previously reported steps 36  . Nucleosomes formed when salt concentration was reduced to around 150 mM (measured by EX170 salinity meter), and the DNA histone mixture solution was further dialyzed into low salt Tris buffer (20 mM Tris, 20 mM NaCl, 0.5 mM EDTA, pH 7.8). Precipitates were removed by centrifuge, and the concentration of the nucleosomes was measured by A260 reading using the Biotek synergy H1 plate reader. His-TEV protease was added to nucleosome solutions (TEV:nucleosome 1:30, w-w) to remove all the histone tags after incubation for 1 h at 37°C, and finally all the His tagged impurities were removed from nucleosome solution by Ni 2+ -NTA resin (Thermo Fisher #88221).
Peptide pulldown assays. One microgram of biotinylated histone peptides with different modifications were incubated with 1-2 μg of GST-fused proteins in binding buffer (50 mM Tris-HCl 7.5, 150 mM NaCl, 0.1% NP-40, 1 mM phenylmethylsulfonyl fluoride (PMSF)) overnight. Streptavidin beads (Amersham) were added to the mixture, and the mixture was incubated for 1 h with rotation. The beads were then washed three times and analyzed using sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and western blotting. Anti-GST (sc-459, 1:1000) antibody was from Santa Cruz (Supplementary Figure 13).
MOs computations. All computations were performed on a Linux HPC cluster with the Gaussian 09 quantum chemical software package (Gaussian 09, Revision A.02, M. J. Frisch et al., Gaussian, Inc., Wallingford CT, 2016). X-ray coordinates of the carbon, oxygen, and nitrogen atoms belonging to the triple-stacked truncated trp-crotonamide-phe fragment were frozen, whereas the positions of the attached hydrogen atoms were fully optimized at the B3LYP/6-311+G(d,p) level of DFT theory producing the truncated structure (Supplementary Figure 6). XYZ coordinates for the truncated triple-stack structure with fully optimized hydrogens are available upon request.

EMSA.
A total of 32 repeats of the 601 Widom DNA sequence were cloned into the pJ201 plasmid and transformed into DH5α cells. The plasmid was purified as previously described 25

Data availability
The atomic coordinates and structure factors of Taf14-YEATS and AF9-YEATS have been deposited in the Protein Data Bank under the accession codes 6MIQ, 6MIP, 6MIO, 6MIN, 6MIM, and 6MIL. The data that support this work are available from the corresponding author upon reasonable request.