Acetylation-independent mechanism for binding of ENL to intrinsically disordered regions in TAF2/6 and SNF5

Abstract

Eleven-nineteen-leukemia (ENL) is a well-established epigenetic regulator essential for transcription in development and implicated in cancer. Here, we report the molecular mechanisms and functional consequences of the ENL association with components of transcriptional and chromatin remodeling complexes, TAF2, TAF6, and SNF5, and describe ENL intramolecular contacts and autoregulation. We show that the extra-terminal domain of ENL binds in an acetylation-independent manner to the hydrophobic motifs identified in intrinsically disordered regions of TAF2, TAF6, and SNF5. Genomic analysis reveals genome-wide co-occupancy of ENL and TAF2 or SNF5 on active promoters, and proteomic and transcriptomic analyses demonstrate increased tumor-associated correlations of ENL with TAF2, TAF6, and SNF5 and their coordinated co-expression patterns in several types of cancer. Our findings shed light on the mechanistic details by which ENL associates with chromatin cofactors and is autoregulated, paving the way for the development of novel ENL targeted antitumor interventions.

Graphical Abstract

Graphical Abstract.

Introduction

Eleven-nineteen-leukemia (ENL), also known as MLLT1, is a transcriptional co-activator and an oncoprotein implicated in blood and kidney cancers. It was identified in the early 1990s as a fusion partner of the histone methyltransferase MLL in chromosomal translocations associated with acute leukemia [1–3] and has rapidly emerged as a promising anticancer drug target [4–7]. In acute myeloid and acute lymphoid leukemias, MLL-ENL fusions constitutively tether ENL (along with components of the Super Elongation Complex or the DOT1L methyltransferase complex) to MLL target loci, including the clustered homeobox (HOX) genes. Such aberrant recruitment drives leukemogenesis through sustained overexpression of these genes [8]. ENL itself has a role in maintaining expression of proleukemic genes, and this activity depends on its YEATS domain (ENL_YEATS) [4]. Gain-of-function hotspot mutations within ENL_YEATS are found in Wilms tumor, the most common pediatric kidney cancer [9]. These mutations act as an oncogenic driver, promoting ENL self-association into biomolecular condensates and leading to aberrant activation of developmental transcriptional programs, characterized by highly active HOX and WNT signaling genes [6, 10, 11].

Structurally, ENL comprises two folded modules, ENL_YEATS and the extra-terminal domain (ENL_ET), connected by a long intrinsically disordered linker region. ENL_YEATS binds to the acyllysine marks, selecting for acetylated and crotonylated lysine 18 and 27 in histone H3 [4, 12], but also recognizes acyllysines in nonhistone proteins, such as the acetyltransferase MOZ [13]. ENL_ET associates with MOZ through binding to the hydrophobic motif of MOZ [13]. Although ENL_ET itself remains poorly characterized, interaction of the ET domain of homologous AF9 (AF9_ET) with components of several chromatin-associated complexes harboring the hydrophobic hxhxh motif is well documented [8, 14, 15]. This interaction recruits or stabilizes subunits of these complexes, including AF4, DOT1L, CBX8, and BCOR, at specific genomic sites, facilitating transcriptional activation or repression [2, 14, 16–20].

Here, we describe an acetylation-independent association of ENL with coregulators and report the ENL autoregulation. We show that ENL binds to intrinsically disordered regions (IDRs) of the components of the transcriptional TFIID complex TAF2 and TAF6, and to IDR of the component of the chromatin remodeling SWI/SNF complex SNF5 and colocalizes with the coregulators on promoters of active genes. Strong tumor-associated expression correlations identified between ENL and the coregulators at both transcriptomic and proteomic levels across multiple cancer types suggests a potential strategy for the development of ENL targeted antitumor therapies.

Materials and methods

Protein purification

Unlabeled and ¹⁵N-labeled ENL_YEATS (aa 1–148, pNIC-CH vector) and ENL_ET (aa 483–559, pGEX-6P-1) were expressed in Escherichia coli Rosetta-2 (DE3) pLysS cells grown in either Luria Broth, Terrific Broth, or M9 minimal media supplemented with ¹⁵NH₄Cl (Sigma–Aldrich). For tryptophan fluorescence and MST experiments, tryptophan and a His-tag, respectively, were introduced to the C-terminus of ENL_ET using a QuikChange Lightning site-directed mutagenesis kit (Agilent). The cells were induced with 0.2 mM isopropyl β-D-1-thiogalactopyranoside (IPTG), incubated overnight at 17°C for 16 h, harvested by centrifugation, and lysed by sonication in a buffer consisting of 25–50 mM Tris (pH 7.0–7.5) buffer, 500 mM NaCl, 2 mM β mercaptoethanol or 5 mM DTT, 0%–1% Triton X-100, phenylmethanesulfonylfluoride (PMSF), and DNase I. His-tag fusion proteins were purified on a HisTrap HP affinity column and eluted by a gradient of 0.5 M imidazole. GST-fusion proteins were purified on glutathione agarose 4B beads (Thermo Fisher Scientific). The GST-tag was cleaved on-resin overnight at 4°C with tobacco etch virus (TEV) protease. Proteins were further purified by size exclusion chromatography (SEC) using HiPrep 16/600 Superdex 75 columns (GE Healthcare) in buffer 50 mM Tris (pH 7.5), 150 mM NaCl, and 2 mM BME or 5 mM DTT. The proteins were concentrated in Millipore concentrators and stored at −80°C.

Peptides

The following peptides were synthesized by Synpeptide and used in NMR, tryptophan fluorescence, and MST: ENL_EBM: Ac-GVMVMPEGADTVSR-NH2 (aa 140–153), TAF2: Ac-ELGLVLNLKE-NH2 (aa 992–1000), SNF5: Ac-ENPLPTVEIAIR-NH2 (aa 330–341), TAF6_EBM1: Ac-LPRVPLDVCLKA-NH2 (aa 122–133), and its mutants containing V125D, L127D, V129D, L131D, and TAF6_EBM2: Ac-KYIVVSLPPTGEGK-NH2 (aa 605–616).

ChIP-seq data analysis

All raw ChIP-seq data were obtained from the Gene Expression Omnibus. TAF2 ChIP-seq data were obtained from GSE232675, SNF5 ChIP-seq data were obtained from GSE36134, and ENL, H3K4me3, H3K27ac, and H3K9ac ChIP-seq data were obtained from GSE82116. Briefly, raw fastq reads were preprocessed with fastp and mapped to hg38 using Bowtie2. Mapped reads were de-duplicated and then filtered using samtools and used for peak calling with MACS2. Peak genomic distributions were quantified with ChIPseeker and peak-to-gene enrichments were performed with ChIP-Enrich and ENRICHR. Heat maps were generated using deepTools, and motif enrichments were performed with HOMER. ChIP-seq tracks were visualized using Integrated Genome Browser (IGV). Default settings were used in all cases unless otherwise noted.

NMR experiments

NMR experiments were performed at 298 K on a Bruker 600 MHz spectrometer equipped with a cryoprobe. ¹H,¹⁵N HSQC spectra were acquired using 0.1 mM uniformly ¹⁵N-labeled proteins in 25 mM Tris-HCl (pH 6.8), 150 mM NaCl, and 0–5 mM DTT, supplemented with 8%–10% D₂O. Binding was characterized by monitoring chemical shift changes in ¹H,¹⁵N HSQC spectra of the proteins induced by addition of peptides or ENL_YEATS. NMR data were processed and analyzed with NMRPipe and NMRDraw as previously described [21].

Tryptophan fluorescence

Spectra were recorded at 25°C on a Fluoromax-3 spectrofluorometer (HORIBA) as described [22] with the following modifications. The samples containing 1 µM tryptophan-containing ENL_ET (in 50 mM Tris (pH 6.8), 150 mM NaCl, and 2.5 mM DTT) and progressively increasing concentrations of peptides were excited at 295 nm. Emission spectra were recorded between 320 and 380 nm with a 0.5-nm step size and a 0.6-s integration time, averaged over three scans. The dissociation constant (K_d) was determined using nonlinear least-squares analysis according to the following equation:

where [L] is concentration of the peptide, [P] is 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 is the average of three separate experiments with error reported as SEM.

MST

Microscale thermophoresis (MST) experiments were carried out on a Monolith NT.115 instrument (NanoTemper). Experiments were performed using ENL_ET-His protein in a 25 mM Tris (pH 7.0) buffer, containing 150 mM NaCl and 2.5 mM DTT. ENL_ET-His was labeled using a His-Tag Labeling Kit RED-tris-NTA (2nd Generation, NanoTemper) and kept constant at 10 nM. Dissociation constant was determined using a direct binding assay in which TAF6_EBM V125D mutant was varied in concentration by serial dilution of discrete samples. The measurement was performed at 50% LED and medium MST power with 3 s prelaser time, 20 s laser on-time, and 1s off-time. The K_d value was calculated using MO Affinity Analysis software (NanoTemper) using a 1:1 stoichiometry and averaged over three separate experiments with error reported as SEM. The plot was generated in GraphPad PRISM.

Protein level correlations and DEPs of pan cancer proteomic datasets

Public cancer proteomics data were obtained from the Clinical Proteomic Tumor Analysis Consortium (CPTAC) through the Proteomic Data Commons (PDC). The Proteome_BCM_GENCODE_v34_harmonized_v1 version was used. Protein abundance values were retrieved and analyzed as log₂-normalized protein intensities. The analyzed cohorts included BRCA (breast cancer) tumor samples (n = 122), CCRCC (clear cell renal cell carcinoma) (normal (n = 80) and tumor (n = 103), COAD (colon adenocarcinoma) normal (n = 100) and tumor (n = 97), GBM (glioblastoma) tumor samples (n = 99), HNSCC (head and neck squamous cell carcinoma) normal (n = 62) and tumor (n = 108), LSCC (lung squamous cell carcinoma) normal (n = 99) and tumor (n = 108), LUAD (lung adenocarcinoma) normal (n = 101) and tumor (n = 106), OV (ovarian cancer) normal (n = 19) and tumor (n = 83), PDAC (pancreatic ductal adenocarcinoma) (normal (n = 44) and tumor (n = 105), and UCEC (uterine corpus endometrial carcinoma) (normal (n = 18) and tumor (n = 95). For each cohort, pairwise Pearson correlations were calculated between ENL (ENSG00000130382) and SNF5 (ENSG00000099956), TAF2 (ENSG00000064313), or TAF6 (ENSG00000106290) abundance using log₂-normalized protein intensities. Correlations were calculated separately in normal and tumor samples when both sample types were available. Tumor-associated protein abundance changes were calculated as the median log₂ normalized protein intensity in tumor samples minus the median log₂ normalized protein intensity in normal samples. Statistical significance for tumor versus normal abundance comparisons was assessed using Wilcoxon tests. Differentially enriched proteins (DEPs) were identified by comparing samples in the upper quartile versus the lower quartile (top 25% versus bottom 25%) of ENL, SNF5, or TAF2 protein abundance. For each protein of interest, DEPs were defined using a Benjamini–Hochberg-adjusted Wilcoxon P-value ≤ .05 and a median log₂ normalized protein intensity difference ≥ ±0.5 (fold change of protein intensity at log₂ scale: log₂FC) between high and low abundance groups.

RNA level correlations and DEGs of LSCC and GBM cohorts

RNA level analyses for LSCC (n = 503), GBM (n = 611), AML (n = 2766), and Wilms tumor (n = 132) were performed for TCGA/TARGET GDC datasets using cBioPortal platform for Cancer Genomics (https://www.cbioportal.org/). RNA expression was analyzed as log₂ (TPM + 1). Pairwise Pearson correlations were calculated between ENL and SNF5, TAF2, or TAF6. Differentially expressed genes (DEGs) were obtained using cBioPortal module, and high versus low quartile comparisons were performed for ENL, SNF5, or TAF2 expression and ranked by q-value. In LSCC, upregulated DEGs were selected as the top 5000 significant genes with log₂FC ≥ 0.5, whereas the top 500 significant genes with log₂FC ≤ − 0.5 were selected as downregulated genes. In GBM, upregulated DEGs were selected as the top 5000 significant genes with log₂FC ≥ 1, whereas downregulated DEGs the top 1000 significant genes with log₂FC ≤ −1. Similarly, in Wilms tumor RNA dataset, the top 5000 significantly genes (q-value) in the high expression group were retained as DEGs up with ENL, SNF5, or TAF2.

Gene ontology analysis of DEPs and DEGs

Pairwise intersections were compiled between ENL associated upregulated DEP or DEG sets and the corresponding SNF5 or TAF2 associated upregulated sets. Functional enrichment analyses were performed on the ENL/SNF5 and ENL/TAF2 pairwise intersections using public R package (clusterProfiler). Gene Ontology of Biological Process and KEGG pathways enrichment: for each intersection, enriched terms were ranked using a composite score integrating gene ratio, gene count, fold enrichment, and statistical significance, and the top 10 terms per pairwise intersection were visualized. Gene ratio was defined as the number of genes from the intersection assigned to a given term divided by the total number of genes in that intersection. In ontology bubble plots, bubble fill represents gene ratio, bubble size represents gene count, and bubble stroke thickness represents enrichment p-value. Term-correlation analysis of Wilms tumor DEG sets: the ENL DEG set was intersected with the SNF5 set and, separately, with that of TAF2. Genes within each intersection were subjected to over-representation analysis against GO Biological Process and KEGG using Shiny GO (version 0.85), yielding an FDR and a fold-enrichment value per term. Terms that are significant (FDR ≤ 0.05) in either intersection were compared between the two analyses by plotting their—Log₁₀FDR values. Agreement was quantified by Pearson and Spearman correlation, and each term was colored by its mean fold enrichment across the two intersects.

Results and discussion

ENL occupies active promoters bound by TAF2 and SNF5

We have previously shown that ENL associates with the acetyltransferase complex MOZ/MORF [13], and studies from other groups demonstrated the presence of ENL or homologous AF9 in TFIID and SWI/SNF complexes. Particularly, the association of human ENL and the yeast counterpart of ENL, Taf14, with the SNF5 subunit of SWI/SNF [23, 24] and AF9 with the TAF6 subunit of TFIID [16, 25] was detected in pulldown assays, and binding of Taf14 to another subunit of TFIID, Taf2, was biochemically and structurally validated [26, 27]. To explore whether ENL functionally cooperates with TAF2 or SNF5 in cells, we examined their genomic occupancies using chromatin immunoprecipitation coupled with deep sequencing (ChIP-seq) datasets previously reported for ENL, TAF2, and SNF5 in MV4;11, HCT116, and NCCIT human cell lines, respectively. As expected, the genomic binding sites of ENL (81%), TAF2 (95%), and SNF5 (56%) were mapped predominantly to promoters (Fig. 1a). A strong signal, observed in the heat map of ENL at TAF2-bound regions, suggested high genetic correlation, and comparative analysis of the peak overlap revealed that 1215 (56%) of TAF2 binding sites were co-occupied by ENL (n = 8869) (Fig. 1b and c). Likewise, a substantial overlap in binding sites was observed for ENL and SNF5, as 2984 genomic regions (34%) were co-occupied by both ENL and SNF5 (n = 15 457). The TAF2⁺ ENL⁺ and SNF5⁺ ENL⁺ cobound regions showed significant enrichment in H3K4me3, H3K27ac, and H3K9ac, indicating that co-localization occurred on active promoters (Fig. 1b). Co-localization of ENL with either TAF2 or SNF5, but not both, was also observed on active promoters of individual genes, such as ANP32E and FOXP1 (Fig. 1d).

Figure 1.

ENL occupies active promoters bound by TAF2 and SNF5. (a) Genomic distributions of TAF2, ENL and SNF5 ChIP-seq peaks. (b) Heat maps of TAF2, ENL, SNF5, H3K4me3, H3K27ac, and H3K9ac ChIP-seq data centered on TAF2 peaks (left) and SNF5 peaks (right). (c) Overlap of ENL and TAF2 ChIP-seq peaks (top), and ENL and SNF5 ChIP-seq peaks (bottom). (d) Representative tracks from ChIP-seq of ENL, TAF2, SNF5, H3K4me3, H3K27ac, and H3K9ac at two genomic loci.

The genomic regions cobound by TAF2 and ENL (99%) or SNF5 and ENL (94%) were mapped even more strongly to promoters (Fig. 2a). The TAF2⁺ ENL⁺ cobound promoters were characterized by a significant enrichment of the YY1 binding motif (p = 1E-132), whereas SNF5⁺ ENL⁺ cobound regions were enriched for the SOX motifs (Fig. 2b). TAF2⁺ ENL⁺ promoters were associated with genes related to MLL1 (KAT2A) binding, RELA binding, MYC signaling, and translation, whereas SNF5⁺ ENL⁺ proximal peaks were enriched for CHD1 binding, RELA binding, G2M checkpoint regulation and NOTCH signaling (Fig. 2c and d). This analysis suggested that TAF2⁺ ENL⁺ regions are largely distinct from SNF5⁺ ENL⁺ regions, though a small but noticeable degree of commonality was observed in their putative regulatory programs (discussed below). Altogether, these results demonstrated that ENL and components of TFIID and SWI/SNF complexes co-occupy transcriptionally active promoters.

Figure 2.

ENL, TAF2 and SNF5 regulatory target analysis. (a) Genomic distribution of TAF2⁺ ENL⁺ cobound regions (left) and SNF5⁺ and ENL⁺ cobound regions (right). (b) Motif enrichments of TAF2⁺ ENL⁺ cobound regions (left) and SNF5⁺ and ENL⁺ co-bound regions (right). (c) ENCODE ChIP-seq enrichments of TAF2⁺ ENL⁺ cobound regions (left) and SNF5⁺ and ENL⁺ co-bound regions (right). (d) Hallmark (top) and Reactome (bottom) enrichments of genes with proximal overlapping TAF2 and ENL peaks (left) and overlapping SNF5 and ENL peaks (right).

ENL_ET binds to TAF2

We have reported acetylation-dependent and acetylation-independent mechanisms for the association of ENL with MOZ [13]. The acetylation-dependent mechanism involves binding of ENL_YEATS to acetylation sites of MOZ present in the IDR of MOZ, whereas the acetylation-independent mechanism involves binding of ENL_ET to the hydrophobic motif, also located in IDR (Fig. 3a). The high level of H3K9ac/H3K27ac at the ENL/TAF2 and ENL/SNF5 cobound genomic regions (Fig. 1b) might indicate that ENL_YEATS at least in part forms complexes with these histone marks and therefore is less available for the association with acetylated TAF2 or SNF5. Although we cannot exclude the contribution of the acetylation-dependent mechanism in the case of binding of ENL to TAF2, TAF6, or SNF5, in this study we explored the acetylation-independent mechanism. AF9_ET has been shown to recognize the hydrophobic hxhxh motif [8, 14], and the high degree of sequence similarity between AF9 and ENL suggested that ENL has this activity as well. We searched the entire amino acid sequence of TAF2 and found 10 regions containing the hxhxh motif (Fig. 3b). Mapping these regions onto the cryo-EM structure of TAF2 revealed that nine regions are located in the structured domain of TAF2, composing either α-helices or β-strands and thus are not accessible for binding of other proteins without a substantial conformational rearrangement within the domain (Fig. 3c). In addition to the folded domain, the AlphaFold modeling of TAF2 revealed a long unstructured C-terminal tail, encompassing amino acids 989–1199 of TAF2 that contains the tenth sequence LVLNL (aa 994–998) (Fig. 3d). To assess whether ENL_ET binds to this sequence, we carried out NMR titration experiments. We collected ¹H,¹⁵N heteronuclear single quantum coherence (HSQC) spectra of ¹⁵N-labeled ENL_ET, while unlabeled TAF2 (aa 989–1000) peptide (TAF2_EBM; EBM: ET-binding motif) was added stepwise (Fig. 3e). Substantial chemical shift perturbations (CSPs) in the slow exchange regime on the NMR timescale induced by TAF2_EBM indicated the formation of a tight complex (Fig. 3e), and binding affinity of ENL_ET to TAF2_EBM measured by tryptophan fluorescence confirmed this interaction (Fig. 3f).

Figure 3.

ENL_ET binds to TAF2_EBM. (a) Domain architecture of ENL. The YEATS and ET domains connected by an IDR linker and their ligands are shown. (b) The hydrophobic hxhxh sequences found in TAF2. (c) The ribbon diagram of the cryo-EM structure of TAF2 (PDB: 5FUR) colored gray with the hydrophobic hxhxh sequences highlighted in color. (d) An AlphaFold model of TAF2 (UniProt, Q6P1 × 5) showing the C-terminal IDR tail of TAF2. TAF2_EBM is colored red and labeled. (e) Overlaid ¹H,¹⁵N HSQC spectra of ¹⁵N-labeled ENL_ET recorded in the presence of increasing amounts of TAF2_EBM peptide. Spectra are color coded according to the protein:peptide molar ratio. (f) Representative binding curve used to determine binding affinity of ENL_ET to TAF2_EBM by tryptophan fluorescence. K_d is represented as the average ± SEM of three independent experiments.

ENL_ET binds to TAF6

Full length TAF6 contains four hydrophobic hxhxh motif sequences and, based on the cryo-EM structure of TAF6, two motifs are present in α-helices and not readily accessible (Fig. 4a and b). The other two motifs are solvent exposed and might serve as ligands for ENL_ET. One (TAF6_EBM1) is found in the IDR loop, which is only partially observed in the cryo-EM structure (Fig. 4b), and another (TAF6_EBM2) is in the C-terminal tail of TAF6 that is not observed in the cryo-EM structure but can be modeled in AlphaFold (Fig. 4b and c). Titration of either TAF6_EBM1 or TAF6_EBM2 caused CSPs in ¹⁵N-labeled ENL_ET, indicating binding, and measurements of binding affinities of ENL_ET to TAF6_EBM1 and TAF6_EBM2 yielded K_ds of 2.4 and 3.3 μM (Fig. 4e and f). Much like TAF2_EBM, TAF6_EBM1 has a hydrophobic two-residue sequence Val–Pro (Leu–Gly in TAF2) prior to the motif that may contribute to the interaction. We tested binding of ENL_ET to TAF6 peptides in which L127, V129, and L131, as well as V125, were replaced with an aspartate in NMR experiments (Fig. 4g, h, j and k). No substantial CSPs were observed in ENL_ET upon addition of L127D TAF6_EBM1 and V129D TAF6_EBM1 and small CSPs were observed upon addition of L131D TAF6_EBM1, revealing that L127, V129, and L131 are necessary for the tight binding (Fig. 4h, j and k). The V125D TAF6_EBM1 peptide however induced CSPs and retained its binding activity albeit weaker than that of WT TAF6_EBM1 peptide (K_d of 33 μM) (Fig. 4g and i). The latter data delineated the EBM to the residues 127–131 of TAF6.

Figure 4.

ENL_ET binds to TAF6_EBM. (a) The hydrophobic hxhxh sequences found in TAF6. (b) The cryo-EM structure of TAF6 (PDB: 7EGB) colored gray with the hydrophobic hxhxh sequences highlighted in color. (c) An AlphaFold model of TAF6 (UniProt, P49848) showing the IDR loop and the C-terminal IDR tail of TAF6. Two indicated TAF6_EBMs are colored red and magenta and labeled. (d) Overlaid ¹H,¹⁵N HSQC spectra of ¹⁵N-labeled ENL_ET recorded in the presence of increasing amounts of TAF6_EBM1 peptide. Spectra are color coded according to the protein:peptide molar ratio. (e, f) Binding curves used to determine binding affinities of ENL_ET to TAF6_EBMs by tryptophan fluorescence. K_d is represented as the average ± SEM of three independent experiments. (g, h) Overlaid ¹H,¹⁵N HSQC spectra of ¹⁵N-labeled ENL_ET recorded in the presence of increasing amounts of indicated TAF6_EBM1 peptides. Spectra are color coded according to the protein:peptide molar ratio. (i) Binding curve used to determine binding affinity of ENL_ET to V125D TAF6_EBM1 by MST. K_d​ value is represented as mean ± SEM from three independent measurements. (j, k) Overlaid ¹H,¹⁵N HSQC spectra of ¹⁵N-labeled ENL_ET recorded in the presence of increasing amounts of indicated TAF6_EBM1 peptides. Spectra are color coded according to the protein:peptide molar ratio.

ENL_ET binds to SNF5

A sole accessible hxhxh motif sequence (aa 336–340) in SNF5 is located in the IDR loop (Fig. 5a and b). Both NMR titration experiments and measurements of binding affinity confirmed recognition of this sequence (SNF5_EBM) by ENL_ET. The pattern of CSPs upon titration of SNF5_EBM into ¹⁵N-labeled ENL_ET was generally similar to the pattern of CSPs observed upon titration of TAF2_EBM or TAF6_EBMs, implying that EBMs tested occupy the same binding site in ENL_ET (Fig. 5c). Furthermore, a K_d of 4.7 μM for the association of ENL_ET with SNF5_EBM was in the range of binding affinities of ENL_ET toward TAF2_EBM and TAF6_EBMs, indicating that the interactions involving ENL_ET are mutually exclusive (Fig. 5d). Yet, ChIP-seq analysis revealed colocalization of ENL, TAF2, and SNF5 at some active promoters, which suggested that the acetylation-dependent mechanism may play a role (Fig. 5e and f). Indeed, acetylation sites at K422, K1075, and K1100 in IDRs of TAF2 and at K77 in IDR of SNF5 were identified by proteomic analyses (refs. UniProt, PTM databases, and [28]) and could be targeted by ENL_YEATS independently or concurrently to the interaction of ENL_ET. This would allow for tethering a nonacylated cofactor via ENL_ET while engaging with an acylated cofactor via ENL_YEATS. In addition, colocalization of these cofactors could be driven through interactions involving other subunits of their respective complexes and binding to specific DNA sequences [29, 30].

Figure 5.

ENL_ET binds to SNF5_EBM. (a) The hydrophobic hxhxh sequences found in SNF5. (b) The cryo-EM structure of SNF5 (PDB: 6LTH) colored gray with the hydrophobic hxhxh sequences highlighted in color. SNF5_EBM in the IDR loop is colored red and labeled. (c) Overlaid ¹H,¹⁵N HSQC spectra of ¹⁵N-labeled ENL_ET recorded in the presence of increasing amounts of SNF5_EBM peptide. Spectra are color coded according to the protein:peptide molar ratio. (d) Binding curve used to determine binding affinity of ENL_ET to SNF5_EBM peptide by tryptophan fluorescence. K_d is represented as the average ± SEM of three independent experiments. (e) Heat maps of TAF2, ENL, and SNF5 centered on TAF2 peaks. (f) ChIP-seq tracks at two genomic loci.

ENL autoregulation

Despite the 350-residue IDR linker in human ENL is much longer than the 40-residue linker in the yeast counterpart Taf14, the AlphaFold modeling predicts a direct contact between the YEATS domain and the ET domain in both cases (Fig. 6a and b). Particularly in ENL, the β8 strand of ENL_YEATS appears to pair with the β1 strand of ENL_ET in an antiparallel manner, forming a contiguous β sheet. This β sheet connects one of the four-stranded planes of the ENL_YEATS’s β-sandwich with the double-stranded β-sheet of ENL_ET (Fig. 6a). Furthermore, the cryo-EM structure of Taf14 within the yeast TFIIF complex shows that Taf14_YEATS and Taf14_ET are in a similar orientation [31], and structural studies from our group showed that β-sheets of Taf14_ET and Taf14_YEATS could be linked through concomitantly engaging with an extra β strand of Taf2 [27]. We therefore tested whether the isolated ENL_YEATS and ENL_ET are in contact using NMR titration experiments. CSPs, observed in ¹⁵N-labeled ENL_ET upon addition of unlabeled ENL_YEATS, confirmed that the two domains interact (Fig. 6c). We note that all gain-of-function Wilms tumor mutations are identified in the ENL_YEATS’s β8 strand (colored yellow in Fig. 6a), and it’s tempting to propose that such mutations may disrupt a fine-tuned equilibrium of the “open-closed” state of ENL, facilitating tumorigenesis.

Figure 6.

ENL autoregulation. (a) An AlphaFold model of ENL (UniProt, Q03111). A large portion of the IDR linker connecting ENL_YEATS (green) and ENL_ET (wheat) is not shown for clarity. The β8 strand sequence of ENL_YEATS predicted to be in contact with ENL_ET and mutated in Wilms tumor is yellow. Internal ENL_EBM is colored orange and labeled. (b) A schematic of ENL autoregulation via the “open-closed” states. (Overlayed ¹H,¹⁵N HSQC spectra of ENL_ET recorded in the presence of increasing amounts of ENL_YEATS (c) or the ENL_EBM peptide (d), respectively. The spectra are color coded according to the protein:ligand molar ratio. (e) Binding curve used to determine binding affinity of ENL_ET to ENL_EBM by tryptophan fluorescence. K_d is represented as the average ± SEM of three independent experiments. (f) Overlay of the AlphaFold model ENL as in panel (a) with the solution NMR structure of AF9_ET (gray) in complex with DOT1L (red) (PDB: 2MV7).

The ENL “open-closed” equilibrium might be autoregulated. The ENL sequence itself contains a hydrophobic motif (VMVMP) in the IDR linker in close proximity to ENL_YEATS. The ENL peptide bearing this sequence caused CSPs in ¹⁵N-labeled ENL_ET, which suggested that ENL has an internal ENL_EBM (Fig. 6d). The pattern of CSPs was similar to the pattern of CSPs induced by external EBMs (Figs. 3e, 4d, 5c, and 6d), and a ∼20-fold weaker binding affinity of internal ENL_EBM indicated an autoregulatory mechanism (Fig. 6e). Interestingly, CSPs in ENL_ET due to binding of EBMs and due to binding of ENL_YEATS were somewhat dissimilar (compare Fig. 6c and d). Because EBM sequences have been shown to engage with the β2 strand of AF9_ET [8, 14], the difference may indicate that ENL_YEATS and the EBM ligands associate with opposed sides of the same ENL_ET’s β-sheet (Fig. 6f). It will be essential in future studies to explore such a possible assembly and assess how ENL autoregulation could mediate binding of EBM-containing ligands, for example based on accessibility of the EBM motifs in the context of full length proteins and intact protein complexes.

Expression of ENL and TAF2, TAF6, and SNF5 correlates in multiple tumor types

To explore whether ENL correlates with TAF2, TAF6, and SNF5 in tumorigenesis, we analyzed their protein levels using CPTAC/PDC proteomics cohorts across multiple tumor types (Fig. 7). Pearson correlation analysis showed that ENL abundance positively correlated with SNF5, TAF2, and TAF6 abundances in several cohorts, particularly in lung squamous cell carcinoma (LSCC) and glioblastoma (GBM). These correlations were generally stronger in tumor samples than in matched normal samples, indicating increased coordination between ENL and TAF2, TAF6, or SNF5 proteins in cancer contexts (Fig. 7a). Furthermore, tumor versus normal tissue comparisons showed a broad tendency toward increased ENL, SNF5, TAF2, and TAF6 protein levels in tumors (Fig. 7b). Several of these changes were statistically significant, and LSCC displayed significant tumor-associated upregulation of the four proteins (Supplementary Fig. S1). Transcriptomic analysis using TCGA RNA expression datasets for LSCC and GBM (that showed strongest correlations, Fig. 7a) confirmed that ENL expression is significantly correlated with SNF5, TAF2, and TAF6 expression (Fig. 7c). Together, these analyses suggested that ENL and TAF2, TAF6, and SNF5 not only correlate in tumors, but also have coordinated co-expression patterns in several types of cancer.

Figure 7.

ENL, SNF5, TAF2, and TAF6 show coordinated tumor-associated protein and RNA expression levels. (a) Per-cohort Pearson correlations of protein abundance between ENL and SNF5, TAF2, or TAF6 in CPTAC/PDC proteomics cohorts. Correlations were determined using log₂-normalized protein intensities and are shown separately for normal and tumor samples. (b) Per-cohort protein abundance differences between tumor and normal tissues for ENL, SNF5, TAF2, and TAF6. Values represent the median log₂ normalized protein intensity in tumor samples minus the median log₂-normalized protein intensity in normal samples. Statistical significance was assessed using Wilcoxon tests. (c) RNA level Pearson correlations between ENL and SNF5, TAF2, or TAF6 in LSCC and GBM obtained using TCGA/cBioPortal RNA expression data. (d) Volcano plots showing DEPs associated with high versus low ENL, SNF5, or TAF2 protein abundance in LSCC and GBM. High and low groups of patients were defined as the upper and lower quartiles (25% top versus 25% bottom) of abundance of the corresponding protein of interest. DEPs were defined using a Benjamini–Hochberg adjusted Wilcoxon P ≤ .05 and a median log₂-normalized protein intensity difference ≥ 0.5 (log₂FC). DEP counts are indicated on the plots. (e) Venn diagrams showing pairwise overlaps between ENL associated upregulated DEPs and SNF5 or TAF2 associated upregulated DEPs in LSCC and GBM protein abundance datasets. (f) Gene Ontology enrichment analysis of Biological Process (GO_BP) of the ENL/SNF5 and ENL/TAF2 upregulated DEP intersections in LSCC and GBM. The top ten enriched terms per intersection are shown. Bubble fill represents gene ratio, bubble size represents gene count, and bubble stroke thickness represents enrichment in P-value. (g) KEGG pathway enrichment analysis of the ENL/SNF5 and ENL/TAF2 upregulated DEP intersections in LSCC and GBM proteomic datasets. The top ten enriched terms per intersection are shown.

We then assessed whether high ENL and SNF5 or TAF2 abundances are associated with common or distinct molecular programs. We stratified LSCC and GBM patients into upper and lower quartile of ENL, SNF5, and TAF2 protein abundances and performed DEP analyses to identify proteins significantly upregulated or downregulated in the upper compared to lower quartiles of ENL, SNF5, and TAF2 abundances (Fig. 7d). We used the same approach with RNA-level analyses and similarly identified broad changes in gene expression associated with high ENL and SNF5 or TAF2 expression (Supplementary Fig. S2). Collectively, these results indicated that alterations in ENL expression in conjunction with SNF5 or TAF2 expression are associated with a substantial impact on proteomic and transcriptomic states in cancer.

ENL-associated proteome and transcriptome programs overlap with SNF5- and TAF2-associated programs

Given the significant changes in abundance and expression patterns of ENL, SNF5, and TAF2, we tested whether their associated molecular programs overlap. In both LSCC and GBM, upregulated DEPs associated with high ENL abundance strongly overlapped with upregulated DEPs associated with high SNF5 abundance or high TAF2 abundance (Fig. 7e). These shared DEP sets supported functional convergence of ENL and SNF5 or TAF2 associated molecular programs on the protein level. In agreement, a comparable pattern was observed at the RNA level, i.e. upregulated DEGs associated with high ENL expression overlapped with upregulated DEGs associated with high SNF5 expression or high TAF2 expression (Supplementary Fig. 3a).

GO Biological Process enrichment revealed highly overlapping terms between the ENL/SNF5 and ENL/TAF2 protein pairwise intersections in both LSCC and GBM (Fig. 7f). These shared terms included DNA damage response, DNA repair and replication, and chromatin remodeling and organization. KEGG enrichment further supported this convergence, highlighting pathways related to DNA replication, chromatin remodeling, and cell cycle programs (Fig. 7g). In agreement, RNA level ontology analyses of ENL/SNF5 and ENL/TAF2 DEG intersections also showed an overlapped pattern (Supplementary Fig. 3). A recurrent enrichment for cell cycle, chromatin remodeling, DNA damage response, and DNA replication programs was identified across both cohorts and molecular layers. Altogether these data suggested that ENL, SNF5, and TAF2 deregulation converge on shared cancer-associated chromatin and genome maintenance pathways.

Although limited data available for AML and Wilms tumor precluded full evaluation, analysis of RNA expression showed the consistent positive co-expression of ENL, SNF5, TAF2, and TAF6 in both cancers, corroborating that these factors are coregulated at the transcriptional level (Fig. 8a). Moreover, the extensive overlap of their associated gene expression programs in Wilms tumor is consistent with those observed in solid tumors (Fig. 8b). The high correlation of GO terms between the ENL/SNF5 and ENL/TAF2 intersections further demonstrate the relevance of their interactions towards potentiating the same downstream cell processes, most notably chromatin organization and remodeling, cell-cycle control, and polycomb-related regulation pathways (Fig. 8c and d).

Figure 8.

Co-expression of ENL, SNF5, TAF2 and TAF6 in AML and Wilms tumor. (a) Pairwise co-expression of ENL with SNF5, TAF6, and TAF2 at the RNA level Log₂(TPM + 1) in acute myeloid leukemia (AML; n = 2766) and in Wilms tumor (n = 132), retrieved from cBioPortal (TARGET/GDC). Pearson correlation coefficients are indicated. All correlations were significant (P < .0001). (b) Venn diagrams of the top 5000 genes upregulated in the high- versus low-expression quartile of each factor in Wilms tumor. The intersecting gene set of each comparison was carried forward for functional enrichment analysis. Term-level correlation of functional enrichment between the ENL/SNF5 intersect and that of ENL/TAF2 for GO Biological Process (c; n = 1127 terms) and KEGG pathways (d; n = 119 terms). Each point is an enriched term plotted by its significance (−Log₁₀FDR) in the ENL/SNF5 intersection (x-axis) versus the ENL/TAF2 intersection (y-axis). Point color encodes the mean fold enrichment of each term across the two analyses. Pearson r, Spearman ρ and term counts are indicated. Representative top-ranked terms are labeled.

In conclusion, in this study we report the molecular mechanism by which ENL associates with the components of transcriptional and chromatin remodeling complexes, TAF2, TAF6, and SNF5. ENL_ET binds in an acetylation-independent manner to the hydrophobic EBMs identified in IDRs of TAF2, TAF6, and SNF5. In addition, ENL_ET interacts with the acetyllysine binding ENL_YEATS, likely leading to the formation of a “closed” state of the protein. ENL “open-closed” equilibrium can play a role in fine-tuning the recognition of specific external EBMs and can also be autoregulated, as ENL itself contains an internal EBM in the IDR linker. Furthermore, Wilms tumor mutations, characterized by small in-frame insertions or deletions in the ENL_YEATS’s β8 strand, may also perturb this equilibrium. The Wilms tumor mutations have been shown to induce structural changes in the β8 strand and promote the formation of phase separated biomolecular condensates that drive ENL recruitment to cancer-related genes, resulting in their aberrant activation [6, 10]. We found that ENL and TAF2 or SNF5 colocalize at promoters of active genes, which suggests a potential functional link between ENL and these components of the SWI/SNF and TFIID complexes in transcription.

Our analyses of proteomic and transcriptomic data reveal tumor-associated correlations of ENL with TAF2, TAF6, and SNF5 and their coordinated co-expression patterns in several cancers. The consistent co-expression of ENL with TAF2, TAF6, and SNF5 highlights a shared mechanism by which cancers use these genes clusters to force rapid cell growth. The strong overlap between ENL and SNF5 or TAF2 associated DEP/DEG programs, together with recurrent enrichment for DNA replication/repair, chromatin remodeling and cell cycle, is consistent with prior work placing SWI/SNF-Polycomb antagonism and TFIID mediated promoter engagement as key transcriptional determinants of cell proliferation [32, 33]. For instance, in LSCC, this convergence is consistent with the broader gene deregulation observed in squamous lung cancer, where genomic alterations coalesce around squamous cell lineage and proliferative circuitry [34]. The ENL/SNF5 and ENL/TAF2 intersection signatures identified in this study suggest additional transcriptional programs that contribute to tumorigenesis and necessitate comprehensive investigation in future research. Similarly to the ET domains of ENL and AF9, the ET domains of bromodomain containing (BRD) proteins mediate protein–protein interactions and drive formation of phase separated condensates at super enhancers to promote oncogenic programs [35]. Given that the ET domain serves as a hub for assembling protein complexes, targeting its binding and phase separation functions might represent a promising alternative to the inhibition of acetyllysine recognizing YEATS domains and bromodomains in these oncoproteins.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported in part by grants from the NIH: CA252707, GM157928, and AG067664 to T.G.K., CA279317 to M.A.B., and CA255506 to H.W., and from the Canadian Institutes of Health Research to E.B.A. M.K. is a Fonds de recherche du Québec PhD scholar.

Data availability

All relevant data supporting the key findings of this study are available within the article or from the corresponding author upon reasonable request.