A “Goldilocks zone” for recruiting BET proteins with bromodomain-1 selective ligands

Abstract

Synthetic genome readers/regulators (SynGRs) are bi-functional molecules, rationally designed to bind specific genomic sequences and engage cellular machinery that regulates the expression of targeted genes. The prototypical SynGR1 targets GAA trinucleotide repeats and recruits the BET (Bromodomain and Extended Terminal domain) family of transcriptional regulatory proteins via a flexibly tethered ligand, JQ1. This pan-BET ligand binds both tandem bromodomains of BET proteins (BD1 and BD2). Second-generation SynGRs, which substituted JQ1 with bromodomain-selective ligands, unexpectedly revealed that BD1-selective ligands failed to functionally engage BET proteins in living cells despite displaying the ability to bind BD1 in vitro. Mechanistically, recruiting a BET protein via BD1- or BD2-selective SynGRs should have resulted in indistinguishable functional outcomes. Here, we report the conversion of inactive BD1-targeting SynGRs into functional gene regulators by structure-guided redesign of the chemical linker that bridges the DNA-binding molecule to the highly selective BD1 ligand GSK778. The results point to an optimal zone for positioning the BD1-selective ligand for functional engagement of BET proteins on chromatin, consistent with the preferred binding of BD1 domains to distal acetyl-lysine residues on histone tails. The results not only resolve the mechanistic conundrum, but they also provide insight into domain-selective targeting and nuanced design of chemo probes and therapeutics.

Graphical Abstract:

Selective control of gene expression with rationally designed small molecules is a long-standing goal at the interface of chemistry and medicine. One approach to regulating genes and genomic functions is through DNA-binding molecules that target the genome in a sequence-specific manner, such as the pyrrole-imidazole polyamides^{1, 2}. Appending ligands that recruit distinct components of the transcriptional machinery to DNA-binding polyamides results in the creation of artificial transcription factors, a class of synthetic gene readers and regulators (SynGRs)^3–5. Based on a GAA-targeting polyamide^{6, 7}, SynGR1 (formerly SynTEF1) was designed to recruit enablers of transcription elongation to GAA trinucleotide repeat expansions that cause Friedreich’s ataxia (FRDA)⁸. These GAA repeat expansions occur in the first intron of the frataxin (FXN) gene, and stall productive transcription elongation by RNA polymerase (Pol II)^9–11. Reduction in transcription through these GAA repeats results in diminished mRNA levels and severe depletion of the FXN protein. Due to the key role played by FXN in the biogenesis of iron-sulfur cluster proteins in the mitochondria, its depletion leads to dysfunction and the onset of FRDA, a progressive neurodegenerative disease^12–14.

To overcome the GAA repeat-mediated block to FXN transcription elongation, SynGR1 was designed to recruit the BET family of master elongation factors via a polyamide-tethered BET ligand^{15, 16}. True to its design, SynGR1 recruits BET proteins to the FXN locus and licenses transcription elongation through the repressive GAA repeats, thereby restoring FXN protein levels in a wide array of patient-derived cells (Fig. 1A). JQ1¹⁷, the first-generation BET ligand used in the design of SynGR1, is a pan-BET ligand that binds, nearly indistinguishably, to the tandem bromodomains (BD1 and BD2) of all BET family members (BRDT, BRD2, BRD3, and BRD4). As newer domain-specific BET ligands were disclosed, we explored the efficacy of SynGRs that selectively engage one of the two bromodomains¹⁶. To our surprise, while SynGRs1–7, which bind BD2, consistently restored FXN expression, while SynGR8, which selectively binds BD1 (via GSK778) was unable to do so (Fig. 1A). This was particularly puzzling because all eight SynGRs only differ in the identity of their BET ligand and each bound their respective BET bromodomains with similar affinities, thus their ability to engage their target in vitro did not mirror their ability to restore FXN transcription in cells. Mechanistically, recruiting BET proteins and the associated elongation machinery to GAA repeats via SynGRs should have been domain-agnostic.

Figure 1. Design and synthesis of BET-targeting SynGRs with variable linker lengths.

(A) BD2-binding SynGRs engage with BET proteins and license transcription across GAA repeats in the FXN gene; however, BD1-selective SynGRs (such as SynGR8) are inactive and do not license transcription. (B, C and D) Design and synthesis of GSK778-based SynGRs with variable (C6 and C6-(PEG)_n linkers.

While puzzling, these results were reminiscent of the domain-selective modes of BET protein engagement by different nuclear proteins^18–24. Transcription factors that selectively regulate signal-responsive genes appear to favor or preferentially engage BD2²⁰, whereas BD1 is thought to broadly associate with actively transcribed chromatin through interactions with acetylated lysine residues on histones. On closer examination, BD1 preferentially associates with acetylated N-terminal histone tail residues, such as lysines 5 and 8 on subunit 4 (H4K5ac, H4K8ac) that are distal from the surface of the nucleosome.

We reasoned that linkers which span the distance from the surface of the nucleosome to the acetyl lysine binding pocket on BD1 would enable BD1-selective ligands to effectively engage that domain in the context of a nucleosome-bound SynGR. Our structurally intuited linker optimization now resolves the mechanistic conundrum raised by the inactivity of BD1-selective SynGR8 by enabling GSK778, and likely other BD1-selective ligands, to engage BET proteins and restore FXN expression in patient-derived FRDA lymphoblasts and fibroblasts.

To test our hypothesis, we designed the next generation of BD1-selective SynGRs by systematically varying the linkers that bridge the BD1-selective ligand GSK778²⁵ and the GAA repeat-binding polyamide PA1 (Fig. 1C). In a key distinction from SynGR8, we first attached a C6 alkyl chain linker to GSK778. As this C6 linker alone is likely insufficient to span the distance between the BD1 pocket and the nucleosomal DNA, we incrementally appended PEG3 units (P3/P6/P9/P12/P16) to afford greater reach for bifunctional SynGRs to bind BD1 as well as the cognate DNA sites on the nucleosome simultaneously.

To generate BD1-selective SynGRs bearing varying linkers, polyamide PA1 was first synthesized via standard solid-phase synthesis, followed by coupling with suitable Fmoc-PEG-NHS ester linkers to yield Fmoc-protected PEG-linked PA1¹⁶. Subsequent deprotection of Fmoc with piperidine produced a free amine group on the linker²⁶. The synthesis of GSK778-C6-COOH intermediate is outlined in Figure 1C. A documented procedure was followed to yield GSK778-C6-COOtBu, which was then hydrolyzed to GSK778-C6-COOH using formic acid. To complete the synthesis of SynGR-C6 series (C6 to C6-P16), GSK778-C6-COOH was conjugated with the appropriate PEG-linked PA1 using PyBOP as previously described¹⁶. Characterization of this set of SynGRs is presented in Supplementary Figures S1–S18.

Surface plasmon resonance (SPR) binding studies^{16, 27} were used to measure the binding of BD1 or BD2 to immobilized SynGR-DNA (GAAGAAGAA) complex (Fig. 2A and Supplementary Fig. S19). The results indicate that neither the length of the PEG linker nor the presence of an aliphatic linker substantially affected either the binding affinity or specificity of GSK778-based SynGRs in vitro, with binding to DNA comparable to that of SynGR8. Notably, all GSK778-based SynGRs demonstrated robust affinity for BD1, while no binding was observed with BD2, consistent with the highly BD1-selective nature of this BET ligand.

Figure 2. Binding and restoration of FXN expression by BD1-selective SynGRs.

(A) Structures of PEG linkers and binding of PA1 and SynGRs to DNA and bromodomains as measured by SPR in vitro. ND, not determined. (B-D) Relative expression of FXN mRNA, normalized to 0.1% DMSO control, after the treatment of patient-derived GM15850 lymphoblast cells with 1, 2 or 5μM SynGRs for 24 h (B) or 48 h (C) and treatment of GM04078 fibroblast cells with 1, 2 or 5μM SynGRs for 24 h (D). P-values in (B-D) were calculated using one-way ANOVA followed by Dunnett’s test, comparing SynGR1 and BD1-SynGRs to untreated DMSO controls. Data points are shown as independent dots. Significance is indicated as follows: ns (not significant) p > 0.05, *p ≤ 0.05, **p ≤ 0.005, ***p ≤ 0.001, ****p ≤ 0.0001.

Having validated the retention of selective BD1 and GAA-DNA binding by this set of bifunctional SynGRs, we evaluated their ability to restore FXN transcription in FRDA patient-derived GM15850 lymphoblast cells, which harbor ~650 and ~1030 GAA repeats on the two alleles of FXN (Fig. 2B, C) and in GM04078 fibroblast cells which harbor ~420 repeats on both FXN alleles (Fig. 2D). Despite exhibiting comparable binding affinities for BD1 and DNA, only SynGR-C6-P3, SynGR-C6-P9, and SynGR-C6-P12 elicited functionally relevant increases in FXN gene expression relative to the DMSO control, whereas SynGR-C6, SynGR-C6-P6, and SynGR-C6-P16 were comparable to the inactive SynGR8. As these SynGRs are all structurally identical except for variations in linker composition and length, we attribute the differences in their ability to activate FXN expression to the structural and physical properties of the linkers²⁸. Interestingly, linker length alone does not fully explain the activity of the molecules²⁹, as the PEG6 linker of the inactive SynGR8 has the maximal reach of ~22Å which is comparable to the ~21Å C6-PEG3 linker of the active SynGR-C6-P3. Further highlighting the importance of physical-chemical properties of the linker, the C6-PEG6 linker with its ~32Å reach displays no transcriptional functionality in cells despite SynGR-C6-P6 retaining the ability to robustly bind DNA and BD1 in vitro. Intriguingly, the ~3.7nM Kd measured for SynGR-C6-P6 binding to cognate DNA belies the consistently lower occupancy of the binding sites in SPR studies (Supplementary Fig. S20). Whether this lower fractional occupancy of available binding sites is indicative of unproductive conformational states (intramolecular or oligomeric states) adopted by the molecule remains to be determined. Finally, extending linkers beyond ~50Å also precipitously reduces functionality in cells without adversely impacting DNA or protein binding in vitro. Strikingly, while the C6-P3, C6-P9 and C6-P12 linkers generate SynGRs that are equally active at 24 hours, C6-P3 emerges as the most effective BD1-selective SynGR at 48 hours, suggestive of continued productive engagement of BET-associated elongation machinery in patient-derived cells. Taken together, the results suggest a “Goldilocks zone” of optimality in length and physical properties of the linker to effectively place the ligand in the BD1 domain of BET proteins in the context of nucleosome-bound genomic DNA^{29, 30}.

To investigate whether these BD1-selective SynGRs license FXN expression through engagement of the BD1 domain rather than adventitious binding to the tandem BD2 domain of BET proteins, we tested the impact of co-treatment of active SynGRs with unconjugated pan-BD1/BD2 BET ligand (JQ1) or domain-specific BD1 (GSK778) or BD2 (ABBV744²⁰ or GSK046¹⁹) ligands. We initiated the experiment by administering each SynGR to GM15850 cells for one hour, followed by assessing the impact of competitive inhibition through subsequent treatment with each of the BET-binding ligands (Fig. 3A).

Figure 3. Co-treatment of SynGRs with distinct classes of BET-binding ligands.

(A) Schematic of co-treatment experiment and structures of the tested BET ligands JQ1, GSK778, and ABBV744. (B) Relative expression of FXN mRNA, normalized to 0.1% DMSO control, after the treatment of GM15850 cells with SynGR1 (B), SynGR8 (C), SynGR-C6-P3 (D), SynGR-C6-P12 (E) or SynGR-C6-P16 (F) at increasing concentrations for 1 h, followed by co-treatment with BET-binding ligands JQ1, GSK778, or ABBV744 at 0.5 μM for 23 h. Error bars are the SD of three replicates. P-values for FXN fold change in the co-treatment groups were calculated using one-way ANOVA followed by Tukey’s test, comparing each condition to its respective SynGR control based on the concentrations. Data points are shown as individual dots. Significance is denoted as follows: ns (not significant) p > 0.05, *p ≤ 0.05, **p ≤ 0.005, ***p ≤ 0.001, ****p ≤ 0.0001.

The administration of JQ1 significantly reduced the FXN expression activity of all BD1-selective SynGRs (Fig. 3B–F). Since JQ1 targets both bromodomains of BET proteins, this finding suggests that blocking the interaction of these SynGRs with either BD1, BD2 or both, diminishes the capacity of all SynGRs to recruit BET proteins. Meanwhile, the co-administration of the domain-specific ligands GSK778 or ABBV744 (or GSK046) exhibited a dose-responsive effect on the activity of the tested SynGRs (Fig. 3B–F and Supplementary Fig. S21). Consistent with previous observations, co-treatment with the BD1-selective ligand GSK778 further stimulated the activity of SynGR1 by facilitating binding to the BD2 domain (Fig. 3B)¹⁶. In contrast, likely due to competition for BD1-binding, co-treatment with GSK778 attenuated activity across the entire set of BD1-selective SynGRs (Fig. 3C–F and Supplementary Fig. S21D). More compellingly, converse to what is observed with BD2-binding SynGRs, such as SynGR1–7 (Fig. 3B)¹⁶, co-treatment with BD2-selective ligands ABBV744 or GSK046 further stimulated the activity of BD1-selective SynGRs (Fig. 3 and Supplementary Fig. S21A). We attribute these domain-selective effects to (i) the ability of the ligands to competitively displace a subset of BET proteins that rely on ligand-targeted bromodomain to bind chromatin or partner proteins, thereby increasing the available pool of BET proteins that can be recruited by SynGRs that engage the unoccupied bromodomain, (ii) increased efficiency in docking to the cognate domain, and (iii) potential interplay between the two bromodomains wherein docking into one domain enhances binding of ligands to the unoccupied domain. While we have not directly investigated these mechanisms, our results demonstrate that the C6-series of SynGRs functionally engage BET proteins via their BD1 domains in patient-derived cells.

To better understand the critical role of the linker in converting an inactive SynGR8 to active BD1-selective SynGRs, we examined prior reports of BD1 engagement with nucleosomes. Interestingly, BD1 of certain BET proteins was reported to bind DNA with modest affinity (18–118 μM)^31–33. In one report, the positively charged surface distal to the ligand binding pocket of BD1 engages with DNA³². However, a subsequent report provides evidence for the ligand-binding domain directly interacting with DNA with little dependence on the positive charge³⁴. In this latter mode, DNA binding sterically occludes docking of acetyl-lysine residues, and by extension other ligands, into the docking site on BD1. We therefore examined the affinity of BD1 and BD2 for GAA repeat-bearing DNA using SPR (Fig. 4A, B). The data unambiguously demonstrate that BRD2-BD1 interacts with the GAA-repeat DNA with a K_d of ~9 μM (see Methods for details). This measured affinity lies within the range reported by prior studies³². In contrast, BRD2-BD2 does not interact with GAA-repeat DNA under identical binding conditions.

Figure 4. Direct interactions of BD1 and BD2 domains with DNA.

(A-B) In vitro binding of purified BD1 and BD2 from human BRD2 to GAA-repeat DNA, measured by SPR, shows that BD1 interacts directly with DNA (A) while BD2 does not (B). To determine the dissociation constant, the binding profiles (in black) are fitted using standard models (in red). (C) Top-ranked AlphaFold3 models of BRD2-BD1 (PDB: 1X0J) and BRD2-BD2 (PDB:2DVV) bound to the nucleosome (PDB:1AOI) show the probable placement and orientation of the individual domains and their ligand-binding pockets.

To gain structural insights into how the two domains separated by an intervening flexible linker would align on the nucleosome, we used AlphaFold3³⁵ to model the crystal structures of the two human BRD2 bromodomains with the nucleosome. Figure 4C displays the highest ranked model predicted by AlphaFold3, with the positively charged surface of BD1 proximal to the DNA. This positioning of the two bromodomains was obtained without imposing any structural constraints imparted by the peptide linker that tethers the two domains. In this arrangement, BD2 is positioned with its ligand-docking site proximal to the DNA surface, whereas the ligand-docking site of BD1 is distal from the nucleosomal DNA. The arrangement of the two bromodomains in an antiparallel state, where the acetyl lysine binding pocket is proximal to the nucleosome surface in BD2 and distal from the nucleosome surface in BD1, is also consistent with the accumulated knowledge from biological studies of how these domains engage different cellular proteins. The BD2 docking site is thought to be more promiscuous and more accepting of acetyl lysines on transcription factors and other domains that bind close to DNA, whereas BD1 is thought to engage with histone tails, especially at the N-terminal end, such as H4K5ac and H4K8ac, which are distal from the DNA surface.

It is important to note that AlphaFold3 also revealed additional potential arrangements of the two domains with respect to each other as subsequent lower ranked structures (Supplementary Fig. S22). This is consistent with the flexibility afforded by the linker that tethers BD1 and BD2 and with structures of homodimers of BD2 bound to homo-bivalent ligands (e.g. two molecules of JQ1 separated by a PEG linker. In that latter arrangement, the two ligand-binding sites are in parallel conformation, with both docking sites adjacent to each other. The range of potential interdomain orientations put forth by modeling implies that the two domains could readily accommodate different arrangements of docked ligands. This insight opens the path to designing molecules that might engage the two domains in distinct “preferred” orientations, thereby leveraging the different interdomain organizations to gain specificity. In essence, it is likely that at different binding sites, the unique arrangement of acetyl-lysines on transcription factors and histone tails will engage BET proteins in a distinguishable manner from other sites of BET association with chromatin. This insight could be leveraged to achieve selective gene regulation or bromodomain function at a desired set of genomic targets.

In conclusion, our data resolve the puzzling observation that ligand SynGRs bearing a for BD1 interact robustly with cognate DNA and target proteins in vitro but does not activate in cells, which can be explained by the positioning of the BET BD1 ligand by the SynGR on the nucleosomal surface rather than on naked DNA. From this perspective, our results explain the greater constraints on linker properties for ligands that dock into BD1 versus BD2, and they also point to new design principles for leveraging structure to selectively interfere with BET family function at subsets of genes.

Materials and Methods

Details of reagents and instruments.

Reagents and solvents were purchased from standard suppliers and used without further purification unless otherwise stated. HPLC purification was performed using a Waters HPLC pump, a Waters 2545 binary gradient modular, Waters 2998 photodiode array detector and a Discovery BIO wide pore C18–10 reverse phase column (25 cm × 21.1 mm, 10 μm) in 0.1% TFA in water with CH₃CN as eluent at a flow rate of 15 mL/min, and a linear gradient elution of 5–60% CH₃CN over 30 min with detection at 254 nm. ESI-TOFMS was recorded on a BRUKER microflex LC/MS mass spectrometer using a positive ionization mode. All reactions as well as compound purity were monitored by UPLC-MS (Acquity PDA detector, Acquity SQ detector and Acquity UPLC BEH-C18 column 1.7 μm, 2.1 x 50 mm [Waters Corp.]) with mobile phase of 0.1% formic acid in water and CH₃CN. The mass spectrometer was operated in positive-ion and negative-ion modes with electrospray ionization and data were acquired using MassLynx, version 4.1. All reactions were carried out using conventional glassware. Reagents were of 95% or greater purity, and solvents were of HPLC grade unless otherwise stated. Anhydrous solvents used were purchased as “anhydrous” grade and used without further drying.

Synthesis of GSK778-C6-COOH.

To a stirred solution of GSK778 (25 mg, 0.04 mmol) and K₂CO₃ (16 mg, 0.12 mmol) in N,N-dimethylformamide (1 mL) was added tert-butyl 6-bromohexanoate (17 mg, 0.07 mmol). The reaction mixture was stirred at room temperature for 48 h. After the reaction was complete, it was diluted in 15% of CH₃CN in water and purified by using Prep-HPLC. Fractions that showed pure polyamide were frozen in liquid nitrogen and lyophilized to afford a white or off-white powder. For hydrolysis the GSK778-C6 ester was dissolved in 2 mL formic acid and the reaction was stirred for 24 h. The reaction mixture was concentrated and used for the next reaction without further purification.

Synthesis of SynGR-C6 to SynGR-C6-P16

Each of the SynGRs were synthesized using standard coupling protocols¹⁶.

Binding analysis of SynGRs and BRD2-BD1/BD2 by surface plasmon resonance

The surface plasmon resonance (SPR) experiments were carried out using Biacore T200 instrument (Cytiva), at 18°C. The biotinylated hDNA1 (10μM) was captured on flow cells (Fc) 1 and 2 or flow cells (Fc) 3 and 4 of series S SA sensor chip (Cytiva), to approximately 150 RUs, using manufacturer’s capture protocol. The reference surface (Fc1 or Fc3) was prepared by capturing PA1 to 180 RU. The SynGR compounds were captured onto Fc2 or Fc4, to approximately 55 RU. Each SynGR compound was captured for each BRD2 protein analysis individually, and the streptavidin/DNA surface was regenerated with 1 min injection of 1M NaCl in 50 mM NaOH to wash off DNA-bound SynGR at the end of the analysis cycle to allow for consecutive SynGR capture. The 3-fold serial dilutions of BRD2 proteins were prepared in running buffer (10 mM HEPES pH 7.4, 150 mM NaCl, 0.05% Tween 20) to yield a series of five concentrations with 100 nM top concentration. The sample compartment temperature was set to 12°C to store BDR2 proteins. The concentration series of BRD2 proteins were injected in single-cycle kinetics mode, at 30μL/min flow rate, with 60s association and 300s final dissociation phases. The blank sample was prepared using BRD2 storage buffer titrated in the same manner as BRD2 proteins. Double-referenced sensorgrams were first analyzed using 1:1 binding model, which did not fit data (Biacore Evaluation Software v.3.1). Since BRD2 proteins showed some extent of binding to streptavidin/PA1 surface and considering possibility of different conformations of DNA/SynGR complexes, we used heterogeneous ligand model for data analysis. This model assumes two different populations of immobilized ligand giving rise to two distinct interaction types. The model fitting outputs interaction parameters (kinetic rate constants and equilibrium dissociation constants) for each individual interaction type. Additionally, the fitting yields maximal binding response values (Rmax) for each of the two interaction types. These Rmax values were used as estimates of a fraction of each individual type of interaction in order to calculate weighted average KD using the equation:

Cell Culture

GM15850 and GM04078 cell lines were obtained from the National Institute of General Medical Sciences (NIGMS) Human Genetic Cell Repository at the Coriell Institute for Medical Research (Camden, NJ). Cells were cultured in RPMI (GIBCO) containing 15% fetal bovine serum (GIBCO). Small molecules were dissolved in DMSO and were added to fresh culture in media.

Co-treatment of SynGRs with BET ligands

GM15850 cells in 1 mL RPMI medium were treated with 1 μL of 1 mM SynGRs, mixed thoroughly, and incubated for 1 hour. Subsequently, 1 μL of 0.5 mM BET binding ligand was added to the SynGR-treated cells. After 23 hours, cells were collected by centrifugation at 14,000 rpm, lysed in 1% BME-containing RLT buffer, and total RNA was purified using the RNeasy Mini Kit (Qiagen, Valencia, CA). DNA contamination was removed by on-column DNase I treatment (ZYMO Research) or with the Ambion TURBO DNA-free Kit, according to the manufacturer’s instructions. The resulting cDNA was analyzed by qPCR, with primer pairs for TATA-binding protein (TBP) and FXN, as shown in Figure 3B–F.

RNA isolation and qPCR

RNA was isolated using a Qiagen RNeasy isolation kit (74106) and eluted in 50 μL nuclease-free water. The resulting RNA concentration was determined using nanodrop and diluted to 400 ng/μL using nuclease-free water. RNA (400 ng) was used to generate cDNA using an iScript reverse transcription cDNA synthesis kit (Bio-Rad 1708891). The resulting cDNA concentration was determined using nanodrop and diluted to 100 ng/μL. Quantitative polymerase chain reaction (qPCR) was performed using SYBR Green SsoAdvanced (Bio-Rad 1725275) with 100 ng cDNA per sample. qPCR was performed on ABI7900.

AlphaFold3 modeling

Preparation of Input Sequence:

The following FASTA sequences of the target proteins from a reliable database such as the Protein Data Bank (PDB) were retrieved:

1AOI_PALINDROMIC 146 BP DNA REPEAT 8/9 FROM HUMAN X-CHROMOSOME ALPHA SATELLITE DNA

ATCAATATCCACCTGCAGATTCTACCAAAAGTGTATTTGGAAACTGCTCCATCAAAAGGCATGTTCAGCTGAATTCAGCTGAACATGCCTTTTGATGGAGCAGTTTCCAAATACACTTTTGGTAGAATCTGCAGGTGGATATTGAT

HISTONE H3

LATKAARKSAPATGGVKKPHRYRPGTVALREIRRYQKSTELLIRKLPFQRLVREIAQDFKTDLRFQSSAVMALQEASEAYLVALFEDTNLCAIHAKRVTIMPKDIQLARRIRGERA

HISTONE H4

KRHRKVLRDNIQGITKPAIRRLARRGGVKRISGLIYEETRGVLKVFLENVIRDAVTYTEHAKRKTVTAMDVVYALKRQGRTLYGFGG

HISTONE H2A

GKQGGKTRAKAKTRSSRAGLQFPVGRVHRLLRKGNYAERVGAGAPVYLAAVLEYLTAEILELAGNAARDNKKTRIIPRHLQLAVRNDEELNKLLGRVTIAQGGVLPNIQSVLLPKK

HISTONE H2B

KKRRKTRKESYAIYVYKVLKQVHPDTGISSKAMSIMNSFVNDVFERIAGEASRLAHYNKRSTITSREIQTAVRLLLPGELAKHAVSEGTKAVTKYTSAK

>X0J_1|BRD2-BD1

GRVTNQLQYLHKVVMKALWKHQFAWPFRQPVDAVKLGLPDYHKIIKQPMDMGTIKRRLENNYYWAASECMQDFNTMFTNCYIYNKPTDDIVLMAQTLEKIFLQKVASMPQEEQELVVTIPKN

>2DVV_1|BRD2-BD2

GSHMEQLKHCNGILKELLSKKHAAYAWPFYKPVDASALGLHDYHDIIKHPMDLSTVKRKMENRDYRDAQEFAADVRLMFSNCYKYNPPDHDVVAMARKLQDVFEFRYAKMPD

These FASTA sequences were uploaded to the AlphaFold3 online server and run locally. The prediction runs used standard AlphaFold options. The program performed multiple sequence alignments (MSAs) and homology searches to identify evolutionary relationships, followed by the generation of 3D structures. AlphaFold typically outputs several models with confidence scores (based on pLDDT), each ranked according to predicted accuracy. AlphaFold3 provided a ranking based on predicted accuracy, with the highest-ranked models generally being more reliable. The highest-ranked models were selected based on the provided scores and visualized in PyMOL or ChimeraX.