Main<\/h2>\n\nIntrinsically disordered regions (IDRs) in proteins lack a fixed tertiary structure under physiological conditions and have become increasingly recognized as having important functions1<\/a><\/sup>. IDRs can establish connections with complex protein\u2013protein interaction networks2<\/a><\/sup> and have been shown to contribute to essential cellular processes including signaling, transcription, RNA processing and cell cycle control3<\/a>,4<\/a>,5<\/a>,6<\/a><\/sup>. IDRs can also promote cellular compartmentalization through the formation of biomolecular condensates7<\/a>,8<\/a><\/sup> and have been implicated in human disease, including neurodegeneration3<\/a>,9<\/a><\/sup> and cancer10<\/a>,11<\/a>,12<\/a><\/sup>.<\/p>\nRecent studies have linked IDR-containing proteins, including some proteins with low complexity domains (LCDs) composed of a limited subset of amino acids, to different types of gene regulatory functions. For example, some transcription factors (TFs) can mediate condensate formation to recruit transcriptional machinery to enhancers and promoters5<\/a>,6<\/a>,10<\/a>,13<\/a>,14<\/a>,15<\/a><\/sup>. Components of repressive complexes can also drive phase separation to achieve chromatin compaction16<\/a><\/sup> and linker histones can participate in phase separation and chromatin organization17<\/a><\/sup>. However, less is known about the coordinated role of these proteins in orchestrating genome-wide gene regulation programs. This is in large part due to the lack of methods to profile DNA-associated disordered proteins simultaneously across the genome. Current epigenomic methodologies designed to map DNA-associated proteins genome-wide, such as ChIP\u2013seq, ORGANIC and CUT&RUN, require the use of antibodies and thus are limited to profiling previously defined proteins individually18<\/a>,19<\/a><\/sup>. We sought to overcome this limitation by taking advantage of the physical and chemical properties of DNA-associated disordered proteins to produce genome-wide profiles in an antibody-independent manner. In particular, starting with pioneering studies by the McKnight laboratory, biotinylated isoxazole (b-isox; Extended Data Fig. 1a<\/a>) has been shown to precipitate a large set of disordered proteins by forming microcrystals that nucleate the formation of \u03b2-strands in IDRs and LCDs10<\/a>,11<\/a>,20<\/a>,21<\/a>,22<\/a><\/sup>. These precipitates contain a complex mixture of disordered proteins and we reasoned that sequencing associated DNA fragments would produce genome-wide maps of the location of these proteins and enable us to study their involvement in gene regulation programs. Moreover, analysis of these profiles may also be used to identify disordered proteins linked to specific DNA-binding motifs, such as TFs with prominent IDRs, even when they are present in low amounts.<\/p>\nWe now show that the combination of b-isox precipitation and next-generation sequencing (DisP-seq, disordered protein precipitation followed by DNA sequencing) can produce genome-wide profiles of endogenous DNA-associated disordered proteins simultaneously. Using this approach, we find that DNA-associated disordered proteins are distributed widely across the genome in complex and cell-type-specific patterns and can undergo large-scale reorganization with changes in cellular states. Moreover, prominent features in these landscapes, such as large clusters of DNA-associated disordered proteins (DisP islands, disordered protein-associated islands), are shaped by interactions between IDRs and have critical IDR-dependent regulatory functions in cellular differentiation and disease.<\/p>\n<\/div>\n<\/div>\n
\nResults<\/h2>\n\nDisP-seq signals are widely distributed in the genome<\/h3>\n
We developed an assay capable of detecting endogenous DNA-associated disordered proteins through b-isox precipitation and DNA sequencing (DisP-seq; Fig. 1a<\/a>). The assay involves (1) isolation of nuclei from unfixed cells, (2) digestion by micrococcal nuclease (MNase), (3) incubation with b-isox to precipitate disordered protein-DNA complexes and (4) DNA extraction and library preparation for next-generation sequencing. Our initial experiments were optimized by selecting the conditions that produced the strongest signal intensities and highest number of peaks in the Ewing sarcoma cell line SKNMC using different concentrations of MNase and b-isox (Extended Data Fig. 1b\u2013e<\/a>). We selected Ewing sarcoma, the second most common pediatric bone cancer, for these experiments because it is driven by the disordered oncogenic fusion protein EWS-FLI1 and serves as a paradigm for the role of IDRs in gene regulation23<\/a>,24<\/a><\/sup>. In particular, the addition of the EWSR1 IDR to the ETS TF FLI1 allows the fusion protein to operate as a pioneer factor and induce active enhancers at GGAA microsatellite repeats in addition to binding canonical nonrepeat GGAA ETS binding sites10<\/a>,25<\/a><\/sup>.<\/p>\n\nFig. 1: Genome-wide analysis of DisP-seq signals in Ewing sarcoma cells.<\/b><\/figcaption>\n![\"Science](\"http:\/\/media.springernature.com\/lw685\/springer-static\/image\/art%3A10.1038%2Fs41587-023-01737-4\/MediaObjects\/41587_2023_1737_Fig1_HTML.png\")
<\/picture><\/a><\/div>\na<\/b>, Schematic of the DisP-seq assay. Nuclear extracts are treated by MNase digestion to release native endogenous proteins bound to DNA. Disordered proteins are precipitated by b-isox and the associated DNA is used to prepare libraries for sequencing. DMSO is used for controls. b<\/b>, Pie chart depicting genome-wide distribution of all DisP-seq peaks (22,633, left) and the fraction of DisP-seq sites with EWS-FLI1 signals (right) in SKNMC cells. c<\/b>, Representative examples of DisP-seq peaks and their chromatin context. Left, DisP-seq signals and DMSO controls at an EWS-FLI1 bound enhancer (highlighted in light gray) associated with CCND1<\/i>. ChIP\u2013seq signals for EWS-FLI1, H3K27ac and ATAC-seq signals are also shown. Right, DisP-seq signals and DMSO controls at a site without EWS-FLI1, H3K27ac ChIP\u2013seq or ATAC-seq signals (PPFIA1<\/i> locus, highlighted in light gray). d<\/b>\u2013f<\/b>, DisP-seq peaks are classified into six categories by ChromHMM (d<\/b>). The numbers of DisP-seq peaks and peaks per Mb in each category are shown in e<\/b>. DisP-seq signal intensities for each category are shown in f<\/b>, n<\/i>\u2009=\u20092 biologically independent experiments. Median value is shown as a line within the boxplot, which spans from the 25th to 75th percentiles. Whiskers indicate a 1.5\u00d7 interquartile range. g<\/b>, Identification of DisP islands in SKNMC cells. DisP islands are defined as the population of merged signals above the inflection point of the curve (slope\u2009=\u20091). DisP-seq signals within 20\u2009kb were merged for this analysis. h<\/b>, Representative example of a DisP island with H3K27ac ChIP\u2013seq and ATAC-seq signals (highlighted in light gray). i<\/b>,j<\/b>, Composite plots showing DisP-seq and H3K27ac or H3K4me1 signals for DisP peaks in active enhancers (i<\/b>) and weak enhancers (j<\/b>). DisP-seq peaks are divided into peaks inside or outside DisP islands. For each plot, \u00b15\u2009kb regions centered on the DisP-seq peaks are shown (x<\/i> axis). ***P<\/i>\u2009<\u20090.001 (paired two-sided t<\/i>-test). P<\/i>active enhancer DisP-seq<\/sub>\u2009=\u20091.01\u2009\u00d7\u200910\u221225<\/sup>, P<\/i>active enhancer H3K27ac<\/sub>\u2009=\u20095.51\u2009\u00d7\u200910\u221234<\/sup>, P<\/i>weak enhancer DisP-seq<\/sub>\u2009=\u20095.72\u2009\u00d7\u200910\u221239<\/sup>, P<\/i>weak enhancer H3K4me1<\/sub>\u2009=\u20091.53\u2009\u00d7\u200910\u221253<\/sup>.<\/p>\n<\/div>\nFull size image<\/span><\/a><\/p>\n<\/figure>\n<\/div>\nWe also verified that the conditions used in our assay lead to the precipitation of disordered proteins as previously described10<\/a>,11<\/a>,26<\/a><\/sup>. We used mass spectrometry to analyze our b-isox precipitates and found that the median IDR annotation length in the MobiDB IDR database27<\/a><\/sup> is markedly longer for precipitated nuclear proteins compared to a random size-matched subsampling of the human proteome (138 amino acids versus 31 amino acids; Extended Data Fig. 1f<\/a>; P<\/i>\u2009=<\/i>\u20091.96\u2009\u00d7\u200910\u2212229<\/sup>). We also observed that most of the precipitated nuclear proteins contain large IDRs (greater than 100 amino acids). In contrast, large IDRs are found in 20% of the human proteome (Extended Data Fig. 1g<\/a>). We expect that a subset of these proteins as well as other IDR-containing proteins that may be below the threshold of detection by mass spectrometry will be associated with DNA and will be visible in DisP-seq profiles.<\/p>\nDisP-seq of SKNMC cells showed strong peaks at 22,633 sites that were highly concordant between experiments (Pearson correlation 0.9, p<\/i>-value\u2009<\u200910-5<\/sup>; Extended Data Fig. 1h<\/a>). Annotation of these sites showed that 90% of DisP-seq peaks were associated with distal regions and 10% with gene promoters (Fig. 1b<\/a>). We next analyzed signals at known binding sites for endogenous EWS-FLI1 in SKNMC cells as a positive control. This showed strong DisP-seq signals centered on EWS-FLI1 binding sites genome-wide, together with corresponding H3K27ac and ATAC signals indicating active and open chromatin (Extended Data Fig. 1i<\/a>). DisP-seq is thus capable of yielding robust and specific peaks for a well-known disordered TF through antibody-independent chemical precipitation. Notably, EWS-FLI1 binding sites accounted for only a minority of DisP-seq peaks (9%; Fig. 1b<\/a>), pointing to the detection of many other DNA-associated disordered proteins by this assay. Figure 1c<\/a> shows examples of DisP-seq peaks associated with EWS-FLI1 (active enhancer with H3K27ac and ATAC-seq signals in the vicinity of CCND1<\/i>) or without the presence of the fusion protein (a location without EWS-FLI1, H3K27ac or ATAC-seq signals).<\/p>\nA comparison with genome-wide chromatin accessibility ATAC-seq signals shows that DisP-seq peaks can occur at locations with either open (76%) or closed chromatin (24%). Moreover, DisP-seq peaks overlap only a minority of ATAC-seq peaks (Extended Data Fig. 1j<\/a>), showing that DNA-associated disordered proteins are not evenly distributed in open chromatin. We also determined the distribution of DisP-seq signals in terms of large-scale 3D genomic compartments using ENCODE Hi-C maps for SKNMC cells. Fifty-six percent of DisP-seq peaks were present in the open A compartment and 44% in the closed B compartment, which is comparable to the ratio of compartment assignments in the genomic background of SKNMC cells (51% versus 49%; Extended Data Fig. 1k<\/a>). We also observed the similar signal intensity levels in A and B compartments (Extended Data Fig. 1l<\/a>). To more precisely determine which chromatin states were associated with DisP-seq signals, we assigned DisP-seq peaks to one of six states defined by ChromHMM using chromatin profiles for SKNMC cells25<\/a>,28<\/a><\/sup> (Fig. 1d<\/a> and Extended Data Fig. 2a\u2013f<\/a>). In contrast to the genomic background (Extended Data Fig. 1m,n<\/a>), DisP-seq peaks were most frequently found in weak enhancer regions, followed by active enhancer regions (Fig. 1e<\/a>). The median levels of DisP-seq signals in different chromatin states were similar, with active enhancers having moderately higher levels overall (Fig. 1f<\/a>). Because most DisP-seq peaks occur at enhancers, we also compared enhancers with or without DisP-seq peaks based on levels of different histone modifications. Active enhancers with DisP-seq peaks are associated with moderately higher H3K27ac and H3K4me1 signals (Extended Data Fig. 2g<\/a>). Weak enhancers followed a similar pattern (Extended Data Fig. 2h<\/a>). DisP-seq signals are thus associated with increased activation signals at enhancers, suggesting a potential relationship with increased regulatory activity. Taken together, our results show that DisP-seq peaks can be detected in the context of different types of genomic elements and are present in a subset of well-defined regulatory sites such as enhancers and promoters.<\/p>\nDisP-seq reveals many large clusters of disordered proteins<\/h3>\n
Given that disordered proteins can engage in weak multivalent interactions through their IDRs29<\/a><\/sup>, we considered whether DisP-seq signals may occur as clusters in the genome. While the majority of DisP-seq signals occur as single peaks, we noted that certain locations contained clusters of DisP-seq signals spanning large genomic regions. To identify these regions, we grouped nearby DisP-seq peaks and ranked these groups by the total content of DisP-seq signals. This analysis showed a set of large DisP-seq clusters with the highest total DisP-seq signals, which we named DisP islands (Fig. 1g,h<\/a>). DisP islands comprise 32.7% of the total DisP-seq signals in the genome and their median length is 21,975\u2009bp compared to 981\u2009bp for single DisP-seq peaks. Remarkably, almost half of DisP islands identified in SKNMC cells contain EWS-FLI1 peaks (Extended Data Fig. 2i<\/a>), indicating that the presence of EWS-FLI1 is associated with neighboring signals for other disordered proteins.<\/p>\nTo provide a detailed view of the chromatin landscape of DisP islands, we classified them according to our ChromHMM model for SKNMC cells. DisP islands are associated with various chromatin states and most are in enhancer regions (active enhancer and weak enhancer, Extended Data Fig. 2j<\/a>). However, only a small subset of DisP islands overlapped superenhancers in SKNMC cells30<\/a><\/sup> (Extended Data Fig. 2k<\/a>). Because most DisP-seq peaks occur at enhancers, we next compared DisP-seq and histone mark signals at the enhancers associated DisP-seq peaks within or outside of DisP islands. At active enhancers, DisP island peaks are associated with higher DisP-seq signals and H3K27ac active enhancer marks compared to non-DisP island peaks (Fig. 1i<\/a>). Similarly, we also observed higher DisP-seq signals and H3K4me1 basal enhancer marks in DisP island peaks of weak enhancers (Fig. 1j<\/a>). These results suggest that DisP islands can provide an environment with high local concentrations of disordered proteins and increased chromatin marks linked to regulatory potential.<\/p>\nDisP-seq peaks are enriched for specific TF motifs<\/h3>\n
We next performed an unbiased motif analysis for all DisP-seq peaks to identify enrichment for specific TFs at these sites in addition to EWS-FLI1. The top four motifs identified correspond to the TFs AP-2\u03b1, NFIB and EWS-FLI1 (single GGAA and GGAA repeat; Fig. 2a<\/a>). To validate these results, we used western blotting to test precipitation of these TFs by b-isox. As expected, these experiments showed strong signals for AP-2\u03b1, NFIB and EWS-FLI1 (Fig. 2b<\/a>). In contrast, signals were not observed for the wild-type endogenous ETS TF GABP\u03b1. These results match the prediction of IDRs by PONDR31<\/a><\/sup>. AP-2\u03b1, NFIB and EWS-FLI1 all have large IDRs compared to GABP\u03b1 (Fig. 2c,d<\/a> and Extended Data Fig. 3a,b<\/a>). Given that several prediction methods for IDRs are available and may have different limitations27<\/a>,32<\/a>,33<\/a>,34<\/a><\/sup>, we validated our results with a second approach (metapredict V2)32<\/a><\/sup> and obtained similar results (Extended Data Fig. 3c<\/a>).<\/p>\n\nFig. 2: Identification of disordered TFs by DisP-seq in SKNMC cells.<\/b><\/figcaption>\n![\"Science](\"http:\/\/media.springernature.com\/lw685\/springer-static\/image\/art%3A10.1038%2Fs41587-023-01737-4\/MediaObjects\/41587_2023_1737_Fig2_HTML.png\")
<\/picture><\/a><\/div>\n\na<\/b>, Motif enrichment analysis for DisP-seq signals in SKNMC cells. The top four motifs identified by HOMER are shown. b<\/b>, Western blot after b-isox precipitation for the TFs identified in a<\/b>. An ETS TF without a prominent disordered domain (GABP\u03b1) is shown as a control. c<\/b>,d<\/b>, Graphs showing intrinsic disorder for AP-2\u03b1 (c<\/b>) and NFIB (d<\/b>). Intrinsic disorder scores were calculated by PONDR. IDRs were defined regions with PONDR scores equal to or higher than 0.5 for at least 50 amino acids. e<\/b>,f<\/b>, Heatmaps showing AP-2\u03b1 ChIP\u2013seq (e<\/b>) and NFIB ChIP\u2013seq (f<\/b>) signals at DisP-seq sites in SKNMC cells (8,578 and 11,369 sites, respectively). For each heatmap, \u00b15\u2009kb regions centered on the TF ChIP\u2013seq peaks are shown. g<\/b>, Percentage of peaks of different disordered TFs in each ChromHMM category.<\/p>\n
Source data<\/a><\/p>\n<\/div>\n<\/div>\nFull size image<\/span><\/a><\/p>\n<\/figure>\n<\/div>\nTo directly match DisP-seq signals with the TFs identified, we performed ChIP\u2013seq for AP-2\u03b1 and NFIB in SKNMC cells. Similar to our results for EWS-FLI1, AP-2\u03b1 and NFIB sites were associated with strong DisP-seq signals centered on the corresponding TF-binding sites and with varying levels of H3K27ac and ATAC-seq (Fig. 2e,f<\/a>). These signals were substantially higher than those at nonoverlapping GABP\u03b1 sites despite strong GABP\u03b1 ChIP\u2013seq signals (Extended Data Fig. 3d<\/a>). Examples of DisP-seq signals associated with AP-2\u03b1 and NFIB peaks are shown in Extended Data Fig. 3e,f<\/a>.<\/p>\nTo explore the distribution of AP-2\u03b1 and NFIB peaks identified by DisP-seq, we assigned these disordered TF peaks to six categories that we defined using ChromHMM (Fig. 2g<\/a>). We found that most AP-2\u03b1 and NFIB peaks are located at active and weak enhancers (73% and 75%, respectively). This is a similar distribution to EWS-FLI1 (87% at enhancers). In sum, DisP-seq signals are closely associated with TFs that contain prominent IDRs and can be used to identify these proteins in an unbiased antibody-independent manner.<\/p>\nDisP signals are reorganized by changes in cellular states<\/h3>\n
We next sought to test whether changes in cellular states can lead to major differences in the distribution of DisP-seq signals. For this purpose, we measured DisP-seq signals in EWS-FLI1 depletion experiments, where the loss of the fusion protein is known to result in widespread changes in chromatin25<\/a><\/sup>. This process involves not only the loss of active signals at EWS-FLI1 binding sites but also the reactivation of normal mesenchymal differentiation programs that are typical of mesenchymal precursors, the putative cells of origin of Ewing sarcoma10<\/a>,25<\/a>,35<\/a>,36<\/a>,37<\/a><\/sup>. Mesenchymal differentiation in Ewing sarcoma cells with reduced levels of EWS-FLI1 has also been linked to increased migration, invasion and metastatic potential23<\/a>,24<\/a>,38<\/a><\/sup>.<\/p>\nDisP-seq profiling showed 1,730 sites with decreased DisP-seq signals and 13,500 sites with increased signals after EWS-FLI1 depletion (Extended Data Fig. 4a\u2013c<\/a>). As expected, the top motif associated with decreased DisP-seq signals corresponds to EWS-FLI1 (Fig. 3a<\/a>, GGAA repeats, peak set 1). Increasing peaks were strongly enriched for the disordered TF NFIB (P<\/i>\u2009=\u200910\u22121118<\/sup>; Fig. 3a<\/a>, peak set 2), which we initially identified in DisP-seq profiles of wild-type SKNMC cells and contains a large IDR (Fig. 2a,d<\/a>). To verify the changes observed in DisP-seq peaks, we performed NFIB ChIP\u2013seq in EWS-FLI1 depletion experiments and compared these results to DisP-seq signals, and ChIP\u2013seq data for EWS-FLI1 and H3K27ac25<\/a><\/sup>. In agreement with our motif analysis, we found strong EWS-FLI1 ChIP\u2013seq signals in peak set 1 and marked increases in NFIB ChIP\u2013seq signals in peak set 2 (Fig. 3b,c<\/a>). These peak sets were matched with corresponding decreases or increases in H3K27ac and ATAC-seq signals (Fig. 3b,c<\/a>). Similar results were observed for DisP islands. A total of 486 DisP islands were lost and 1,306 DisP islands were gained upon EWS-FLI1 depletion. More than half of lost DisP islands were associated with EWS-FLI1 (Fig. 3d<\/a>, pattern A), while 82% of gained DisP islands contain NFIB (Fig. 3g<\/a>, pattern C). Our data thus show that DisP-seq signals can undergo substantial reorganization with changes in cellular states and that these profiles can be used to identify disordered TFs linked to these processes.<\/p>\n\nFig. 3: EWS-FLI1 depletion leads to reorganization of DisP islands in SKNMC cells.<\/b><\/figcaption>\n![\"Science](\"http:\/\/media.springernature.com\/lw685\/springer-static\/image\/art%3A10.1038%2Fs41587-023-01737-4\/MediaObjects\/41587_2023_1737_Fig3_HTML.png\")
<\/picture><\/a><\/div>\na<\/b>, Top motifs for increasing and decreasing DisP-seq peaks (peak sets 1 and 2). b<\/b>,c<\/b>, Top\u2014heatmaps depicting sites from peaks sets 1 (b<\/b>) and 2 (c<\/b>). DisP-seq, ChIP\u2013seq and ATAC-seq signal intensities are shown. Bottom\u2014composite plots for the same sites. d<\/b>, Schematic of different patterns of DisP islands lost upon EWS-FLI1 depletion. DisP islands overlap with at least one (pattern A) or no EWS-FLI1 sites (pattern B). e<\/b>, Representative example of DisP island classified as pattern A. f<\/b>, Boxplot showing changes of NFIB signals after EWS-FLI1 depletion. Signals are grouped by their distance to EWS-FLI1 peaks and by whether they are in DisP islands associated with EWS-FLI1 (n<\/i>\u2009=\u20092 biologically independent experiments). *P<\/i>\u2009<\u20090.05, **P<\/i>\u2009<\u20090.01 and ***P<\/i>\u2009<\u20090.001 (two-sided t<\/i>-test). Median value is shown as a line within the boxplot, which spans from the 25th to 75th percentiles. Whiskers indicate a 1.5\u00d7 interquartile range. P<\/i>0\u2013150\u2009bp<\/sub>\u2009=\u20093.40\u2009\u00d7\u200910\u221218<\/sup>, P<\/i>150\u20131\u2009kb<\/sub>\u2009=\u20090.04, P<\/i>1k\u20135\u2009kb<\/sub>\u2009=\u20090.004 and P<\/i>\u22655\u2009kb<\/sub>\u2009=\u20090.04. g<\/b>, Schematic of different patterns of gained DisP islands upon EWS-FLI1 depletion. DisP islands that overlap with at least one (pattern C) or no increased NFIB peaks (pattern D). h<\/b>, Representative example of DisP island classified as pattern C. i<\/b>, Boxplot showing changes in expression for NFIB target genes after EWS-FLI1 depletion. Upregulated genes associated with NFIB sites are classified according to whether they are inside or outside of DisP islands (n<\/i>\u2009=\u20092 biologically independent experiments). ***P<\/i>\u2009<\u20090.001 (two-sided t<\/i>-test). Median value is shown as a line within the boxplot, which spans from the 25th to 75th percentiles. Whiskers indicate a 1.5\u00d7 interquartile range. P<\/i>\u2009=\u20091.69\u2009\u00d7\u200910\u221207<\/sup> j<\/b>, GO analysis of upregulated genes associated with gained DisP islands after EWS-FLI1 depletion. k<\/b>, Schematic of DisP island reorganization after EWS-FLI1 depletion. In Ewing sarcoma cells, NFIB partially colocalizes with EWS-FLI1 in pathologic DisP islands. After EWS-FLI1 knockdown, NFIB signals are lost at GGAA repeats and gained at previously unoccupied NFIB sites to form physiologic DisP islands. The reorganization of DisP islands is associated with reactivation of normal mesenchymal differentiation programs in tumor cells. COF, cofactor.<\/p>\n<\/div>\nFull size image<\/span><\/a><\/p>\n<\/figure>\n<\/div>\nDisP islands can enable sequestration of disordered TFs<\/h3>\n
In addition to these gained NFIB peaks, we also noted the presence of strong NFIB ChIP\u2013seq signals near EWS-FLI1 sites before EWS-FLI1 depletion (Fig. 3b<\/a>), including 270 of 272 EWS-FLI1 dependent DisP islands. This may be partly explained by the fact the half-site NFIB motif (TGGCA) is similar to GGAA and can be found within imperfect segments of GGAA microsatellites (64% of EWS-FLI1 bound GGAA repeat sites; Extended Data Fig. 4d<\/a>) and by immunoprecipitation experiments that show an interaction between EWS-FLI1 and NFIB (Extended Data Fig. 4e<\/a>). However, our data also shows that NFIB peaks within EWS-FLI1 DisP islands (pattern A) can be separated from the fusion protein by long distances (Fig. 3e<\/a>) and prompted us to consider the relationship between EWS-FLI1 and nearby NFIB peaks in more detail. Remarkably, our data show that depletion of EWS-FLI1 strongly decreases NFIB binding (Fig. 3e,f<\/a>) at both short and long distances between peaks (less than 150\u2009bp and also between 150\u2009bp and 1\u2009kb). Moreover, NFIB peaks that are contained in EWS-FLI1-associated DisP islands are affected more strongly by the depletion of the fusion protein. This effect is less pronounced but still noticeable beyond 1\u2009kb (Fig. 3f<\/a>). Our data thus show effects of EWS-FLI1 on NFIB binding that are beyond the 50\u2009bp range typically observed for cooperative binding39<\/a><\/sup>, suggesting that DisP islands can facilitate the coordinated binding of disordered TFs over large genomic regions.<\/p>\nGiven the strong connection between NFIB and new DisP islands that appear after EWS-FLI1 depletion, we also characterized gene regulation programs associated with these events. NFIB is part of a family of TFs with roles in the development of various tissue types, including the differentiation of mesenchymal precursors40<\/a><\/sup>. Previous studies have also shown that NFIB can increase chromatin accessibility in small-cell lung cancer (SCLC) cells and stem cells41<\/a>,42<\/a><\/sup>. Our results show that increases in NFIB binding mainly occur in distal regions (Extended Data Fig. 4f<\/a>) and lead to the establishment of DisP islands (Fig. 3g,h<\/a>). To explore the role of gained DisP islands in gene regulation, we examined upregulated genes near increasing NFIB signals after EWS-FLI1 KD. This analysis shows that increases in expression levels were more pronounced for genes associated with NFIB-containing DisP islands compared to NFIB peaks outside of DisP islands (Fig. 3i<\/a>). Moreover, the signals of DisP-seq, NFIB and H3K27ac ChIP\u2013seq are also higher in peaks located in gained DisP islands (Extended Data Fig. 4g<\/a>). These results show that incorporation of NFIB into an environment with high local concentration of disordered proteins in DisP islands leads to enhanced effects on transcription programs.<\/p>\nGene ontology (GO) analysis of potential target genes of these gained DisP islands shows a strong association with mesenchymal programs induced after EWS-FLI1 depletion (for example locomotion, adhesion and migration; Fig. 3j<\/a>), pointing to a role for the reorganization of DisP islands in these processes. Moreover, sites with DisP-seq peaks in these DisP islands display high average DNase I signals in mesenchymal cell types profiled by ENCODE (113 human cell types, GSE29692<\/a>), suggesting that they correspond to regulatory sites in this lineage43<\/a><\/sup> (Extended Data Fig. 4h<\/a>). In contrast, NFIB peak locations in DisP islands associated with EWS-FLI1 in SKNMC cells show low signals in all ENCODE cell types, suggesting that they are associated primarily with the Ewing sarcoma pathologic state (Extended Data Fig. 4h<\/a>).<\/p>\nTaken together, our results show the reorganization of the DisP island landscape upon EWS-FLI1 depletion (Fig. 3k<\/a>). In wild-type SKNMC cells, NFIB is partially sequestered by EWS-FLI1 in pathologic DisP islands, which are exclusively observed in the context of Ewing sarcoma. After EWS-FLI1 depletion, NFIB is released from these sites and relocates to NFIB binding sites linked to mesenchymal differentiation to establish physiologic DisP islands. In addition, the small increase in the level of NFIB protein after EWS-FLI1 KD may also contribute to the formation of these physiological DisP islands (Extended Data Fig. 4b<\/a>). These changes are linked to the reactivation of latent mesenchymal differentiation programs in tumor cells through increased chromatin accessibility and enhancer activation.<\/p>\nIDRs can mediate the incorporation of TFs into DisP islands<\/h3>\n
Given the strong association between NFIB and changes in DisP-seq signals, we next considered whether the IDR of NFIB has a role in the binding and function of this TF. NFIB contains a large IDR in its C-terminal region (Fig. 2d<\/a>) and we generated a mutant lacking this domain (NFIB\u0394IDR<\/sup>; Fig. 4a<\/a>). We first compared wild-type NFIB (NFIBWT<\/sup>) and NFIB\u0394IDR<\/sup> by b-isox precipitation and, as expected, only NFIBWT<\/sup> signals were detected by Western blot in the precipitated fraction (Fig. 4a<\/a>).<\/p>\n
Intrinsically disordered regions (IDRs) in proteins lack a fixed tertiary structure under physiological conditions and have become increasingly recognized as having important functions1<\/a><\/sup>. IDRs can establish connections with complex protein\u2013protein interaction networks2<\/a><\/sup> and have been shown to contribute to essential cellular processes including signaling, transcription, RNA processing and cell cycle control3<\/a>,4<\/a>,5<\/a>,6<\/a><\/sup>. IDRs can also promote cellular compartmentalization through the formation of biomolecular condensates7<\/a>,8<\/a><\/sup> and have been implicated in human disease, including neurodegeneration3<\/a>,9<\/a><\/sup> and cancer10<\/a>,11<\/a>,12<\/a><\/sup>.<\/p>\n Recent studies have linked IDR-containing proteins, including some proteins with low complexity domains (LCDs) composed of a limited subset of amino acids, to different types of gene regulatory functions. For example, some transcription factors (TFs) can mediate condensate formation to recruit transcriptional machinery to enhancers and promoters5<\/a>,6<\/a>,10<\/a>,13<\/a>,14<\/a>,15<\/a><\/sup>. Components of repressive complexes can also drive phase separation to achieve chromatin compaction16<\/a><\/sup> and linker histones can participate in phase separation and chromatin organization17<\/a><\/sup>. However, less is known about the coordinated role of these proteins in orchestrating genome-wide gene regulation programs. This is in large part due to the lack of methods to profile DNA-associated disordered proteins simultaneously across the genome. Current epigenomic methodologies designed to map DNA-associated proteins genome-wide, such as ChIP\u2013seq, ORGANIC and CUT&RUN, require the use of antibodies and thus are limited to profiling previously defined proteins individually18<\/a>,19<\/a><\/sup>. We sought to overcome this limitation by taking advantage of the physical and chemical properties of DNA-associated disordered proteins to produce genome-wide profiles in an antibody-independent manner. In particular, starting with pioneering studies by the McKnight laboratory, biotinylated isoxazole (b-isox; Extended Data Fig. 1a<\/a>) has been shown to precipitate a large set of disordered proteins by forming microcrystals that nucleate the formation of \u03b2-strands in IDRs and LCDs10<\/a>,11<\/a>,20<\/a>,21<\/a>,22<\/a><\/sup>. These precipitates contain a complex mixture of disordered proteins and we reasoned that sequencing associated DNA fragments would produce genome-wide maps of the location of these proteins and enable us to study their involvement in gene regulation programs. Moreover, analysis of these profiles may also be used to identify disordered proteins linked to specific DNA-binding motifs, such as TFs with prominent IDRs, even when they are present in low amounts.<\/p>\n We now show that the combination of b-isox precipitation and next-generation sequencing (DisP-seq, disordered protein precipitation followed by DNA sequencing) can produce genome-wide profiles of endogenous DNA-associated disordered proteins simultaneously. Using this approach, we find that DNA-associated disordered proteins are distributed widely across the genome in complex and cell-type-specific patterns and can undergo large-scale reorganization with changes in cellular states. Moreover, prominent features in these landscapes, such as large clusters of DNA-associated disordered proteins (DisP islands, disordered protein-associated islands), are shaped by interactions between IDRs and have critical IDR-dependent regulatory functions in cellular differentiation and disease.<\/p>\n<\/div>\n<\/div>\n We developed an assay capable of detecting endogenous DNA-associated disordered proteins through b-isox precipitation and DNA sequencing (DisP-seq; Fig. 1a<\/a>). The assay involves (1) isolation of nuclei from unfixed cells, (2) digestion by micrococcal nuclease (MNase), (3) incubation with b-isox to precipitate disordered protein-DNA complexes and (4) DNA extraction and library preparation for next-generation sequencing. Our initial experiments were optimized by selecting the conditions that produced the strongest signal intensities and highest number of peaks in the Ewing sarcoma cell line SKNMC using different concentrations of MNase and b-isox (Extended Data Fig. 1b\u2013e<\/a>). We selected Ewing sarcoma, the second most common pediatric bone cancer, for these experiments because it is driven by the disordered oncogenic fusion protein EWS-FLI1 and serves as a paradigm for the role of IDRs in gene regulation23<\/a>,24<\/a><\/sup>. In particular, the addition of the EWSR1 IDR to the ETS TF FLI1 allows the fusion protein to operate as a pioneer factor and induce active enhancers at GGAA microsatellite repeats in addition to binding canonical nonrepeat GGAA ETS binding sites10<\/a>,25<\/a><\/sup>.<\/p>\n a<\/b>, Schematic of the DisP-seq assay. Nuclear extracts are treated by MNase digestion to release native endogenous proteins bound to DNA. Disordered proteins are precipitated by b-isox and the associated DNA is used to prepare libraries for sequencing. DMSO is used for controls. b<\/b>, Pie chart depicting genome-wide distribution of all DisP-seq peaks (22,633, left) and the fraction of DisP-seq sites with EWS-FLI1 signals (right) in SKNMC cells. c<\/b>, Representative examples of DisP-seq peaks and their chromatin context. Left, DisP-seq signals and DMSO controls at an EWS-FLI1 bound enhancer (highlighted in light gray) associated with CCND1<\/i>. ChIP\u2013seq signals for EWS-FLI1, H3K27ac and ATAC-seq signals are also shown. Right, DisP-seq signals and DMSO controls at a site without EWS-FLI1, H3K27ac ChIP\u2013seq or ATAC-seq signals (PPFIA1<\/i> locus, highlighted in light gray). d<\/b>\u2013f<\/b>, DisP-seq peaks are classified into six categories by ChromHMM (d<\/b>). The numbers of DisP-seq peaks and peaks per Mb in each category are shown in e<\/b>. DisP-seq signal intensities for each category are shown in f<\/b>, n<\/i>\u2009=\u20092 biologically independent experiments. Median value is shown as a line within the boxplot, which spans from the 25th to 75th percentiles. Whiskers indicate a 1.5\u00d7 interquartile range. g<\/b>, Identification of DisP islands in SKNMC cells. DisP islands are defined as the population of merged signals above the inflection point of the curve (slope\u2009=\u20091). DisP-seq signals within 20\u2009kb were merged for this analysis. h<\/b>, Representative example of a DisP island with H3K27ac ChIP\u2013seq and ATAC-seq signals (highlighted in light gray). i<\/b>,j<\/b>, Composite plots showing DisP-seq and H3K27ac or H3K4me1 signals for DisP peaks in active enhancers (i<\/b>) and weak enhancers (j<\/b>). DisP-seq peaks are divided into peaks inside or outside DisP islands. For each plot, \u00b15\u2009kb regions centered on the DisP-seq peaks are shown (x<\/i> axis). ***P<\/i>\u2009<\u20090.001 (paired two-sided t<\/i>-test). P<\/i>active enhancer DisP-seq<\/sub>\u2009=\u20091.01\u2009\u00d7\u200910\u221225<\/sup>, P<\/i>active enhancer H3K27ac<\/sub>\u2009=\u20095.51\u2009\u00d7\u200910\u221234<\/sup>, P<\/i>weak enhancer DisP-seq<\/sub>\u2009=\u20095.72\u2009\u00d7\u200910\u221239<\/sup>, P<\/i>weak enhancer H3K4me1<\/sub>\u2009=\u20091.53\u2009\u00d7\u200910\u221253<\/sup>.<\/p>\n<\/div>\n Full size image<\/span><\/a><\/p>\n<\/figure>\n<\/div>\n We also verified that the conditions used in our assay lead to the precipitation of disordered proteins as previously described10<\/a>,11<\/a>,26<\/a><\/sup>. We used mass spectrometry to analyze our b-isox precipitates and found that the median IDR annotation length in the MobiDB IDR database27<\/a><\/sup> is markedly longer for precipitated nuclear proteins compared to a random size-matched subsampling of the human proteome (138 amino acids versus 31 amino acids; Extended Data Fig. 1f<\/a>; P<\/i>\u2009=<\/i>\u20091.96\u2009\u00d7\u200910\u2212229<\/sup>). We also observed that most of the precipitated nuclear proteins contain large IDRs (greater than 100 amino acids). In contrast, large IDRs are found in 20% of the human proteome (Extended Data Fig. 1g<\/a>). We expect that a subset of these proteins as well as other IDR-containing proteins that may be below the threshold of detection by mass spectrometry will be associated with DNA and will be visible in DisP-seq profiles.<\/p>\n DisP-seq of SKNMC cells showed strong peaks at 22,633 sites that were highly concordant between experiments (Pearson correlation 0.9, p<\/i>-value\u2009<\u200910-5<\/sup>; Extended Data Fig. 1h<\/a>). Annotation of these sites showed that 90% of DisP-seq peaks were associated with distal regions and 10% with gene promoters (Fig. 1b<\/a>). We next analyzed signals at known binding sites for endogenous EWS-FLI1 in SKNMC cells as a positive control. This showed strong DisP-seq signals centered on EWS-FLI1 binding sites genome-wide, together with corresponding H3K27ac and ATAC signals indicating active and open chromatin (Extended Data Fig. 1i<\/a>). DisP-seq is thus capable of yielding robust and specific peaks for a well-known disordered TF through antibody-independent chemical precipitation. Notably, EWS-FLI1 binding sites accounted for only a minority of DisP-seq peaks (9%; Fig. 1b<\/a>), pointing to the detection of many other DNA-associated disordered proteins by this assay. Figure 1c<\/a> shows examples of DisP-seq peaks associated with EWS-FLI1 (active enhancer with H3K27ac and ATAC-seq signals in the vicinity of CCND1<\/i>) or without the presence of the fusion protein (a location without EWS-FLI1, H3K27ac or ATAC-seq signals).<\/p>\n A comparison with genome-wide chromatin accessibility ATAC-seq signals shows that DisP-seq peaks can occur at locations with either open (76%) or closed chromatin (24%). Moreover, DisP-seq peaks overlap only a minority of ATAC-seq peaks (Extended Data Fig. 1j<\/a>), showing that DNA-associated disordered proteins are not evenly distributed in open chromatin. We also determined the distribution of DisP-seq signals in terms of large-scale 3D genomic compartments using ENCODE Hi-C maps for SKNMC cells. Fifty-six percent of DisP-seq peaks were present in the open A compartment and 44% in the closed B compartment, which is comparable to the ratio of compartment assignments in the genomic background of SKNMC cells (51% versus 49%; Extended Data Fig. 1k<\/a>). We also observed the similar signal intensity levels in A and B compartments (Extended Data Fig. 1l<\/a>). To more precisely determine which chromatin states were associated with DisP-seq signals, we assigned DisP-seq peaks to one of six states defined by ChromHMM using chromatin profiles for SKNMC cells25<\/a>,28<\/a><\/sup> (Fig. 1d<\/a> and Extended Data Fig. 2a\u2013f<\/a>). In contrast to the genomic background (Extended Data Fig. 1m,n<\/a>), DisP-seq peaks were most frequently found in weak enhancer regions, followed by active enhancer regions (Fig. 1e<\/a>). The median levels of DisP-seq signals in different chromatin states were similar, with active enhancers having moderately higher levels overall (Fig. 1f<\/a>). Because most DisP-seq peaks occur at enhancers, we also compared enhancers with or without DisP-seq peaks based on levels of different histone modifications. Active enhancers with DisP-seq peaks are associated with moderately higher H3K27ac and H3K4me1 signals (Extended Data Fig. 2g<\/a>). Weak enhancers followed a similar pattern (Extended Data Fig. 2h<\/a>). DisP-seq signals are thus associated with increased activation signals at enhancers, suggesting a potential relationship with increased regulatory activity. Taken together, our results show that DisP-seq peaks can be detected in the context of different types of genomic elements and are present in a subset of well-defined regulatory sites such as enhancers and promoters.<\/p>\n Given that disordered proteins can engage in weak multivalent interactions through their IDRs29<\/a><\/sup>, we considered whether DisP-seq signals may occur as clusters in the genome. While the majority of DisP-seq signals occur as single peaks, we noted that certain locations contained clusters of DisP-seq signals spanning large genomic regions. To identify these regions, we grouped nearby DisP-seq peaks and ranked these groups by the total content of DisP-seq signals. This analysis showed a set of large DisP-seq clusters with the highest total DisP-seq signals, which we named DisP islands (Fig. 1g,h<\/a>). DisP islands comprise 32.7% of the total DisP-seq signals in the genome and their median length is 21,975\u2009bp compared to 981\u2009bp for single DisP-seq peaks. Remarkably, almost half of DisP islands identified in SKNMC cells contain EWS-FLI1 peaks (Extended Data Fig. 2i<\/a>), indicating that the presence of EWS-FLI1 is associated with neighboring signals for other disordered proteins.<\/p>\n To provide a detailed view of the chromatin landscape of DisP islands, we classified them according to our ChromHMM model for SKNMC cells. DisP islands are associated with various chromatin states and most are in enhancer regions (active enhancer and weak enhancer, Extended Data Fig. 2j<\/a>). However, only a small subset of DisP islands overlapped superenhancers in SKNMC cells30<\/a><\/sup> (Extended Data Fig. 2k<\/a>). Because most DisP-seq peaks occur at enhancers, we next compared DisP-seq and histone mark signals at the enhancers associated DisP-seq peaks within or outside of DisP islands. At active enhancers, DisP island peaks are associated with higher DisP-seq signals and H3K27ac active enhancer marks compared to non-DisP island peaks (Fig. 1i<\/a>). Similarly, we also observed higher DisP-seq signals and H3K4me1 basal enhancer marks in DisP island peaks of weak enhancers (Fig. 1j<\/a>). These results suggest that DisP islands can provide an environment with high local concentrations of disordered proteins and increased chromatin marks linked to regulatory potential.<\/p>\n We next performed an unbiased motif analysis for all DisP-seq peaks to identify enrichment for specific TFs at these sites in addition to EWS-FLI1. The top four motifs identified correspond to the TFs AP-2\u03b1, NFIB and EWS-FLI1 (single GGAA and GGAA repeat; Fig. 2a<\/a>). To validate these results, we used western blotting to test precipitation of these TFs by b-isox. As expected, these experiments showed strong signals for AP-2\u03b1, NFIB and EWS-FLI1 (Fig. 2b<\/a>). In contrast, signals were not observed for the wild-type endogenous ETS TF GABP\u03b1. These results match the prediction of IDRs by PONDR31<\/a><\/sup>. AP-2\u03b1, NFIB and EWS-FLI1 all have large IDRs compared to GABP\u03b1 (Fig. 2c,d<\/a> and Extended Data Fig. 3a,b<\/a>). Given that several prediction methods for IDRs are available and may have different limitations27<\/a>,32<\/a>,33<\/a>,34<\/a><\/sup>, we validated our results with a second approach (metapredict V2)32<\/a><\/sup> and obtained similar results (Extended Data Fig. 3c<\/a>).<\/p>\n a<\/b>, Motif enrichment analysis for DisP-seq signals in SKNMC cells. The top four motifs identified by HOMER are shown. b<\/b>, Western blot after b-isox precipitation for the TFs identified in a<\/b>. An ETS TF without a prominent disordered domain (GABP\u03b1) is shown as a control. c<\/b>,d<\/b>, Graphs showing intrinsic disorder for AP-2\u03b1 (c<\/b>) and NFIB (d<\/b>). Intrinsic disorder scores were calculated by PONDR. IDRs were defined regions with PONDR scores equal to or higher than 0.5 for at least 50 amino acids. e<\/b>,f<\/b>, Heatmaps showing AP-2\u03b1 ChIP\u2013seq (e<\/b>) and NFIB ChIP\u2013seq (f<\/b>) signals at DisP-seq sites in SKNMC cells (8,578 and 11,369 sites, respectively). For each heatmap, \u00b15\u2009kb regions centered on the TF ChIP\u2013seq peaks are shown. g<\/b>, Percentage of peaks of different disordered TFs in each ChromHMM category.<\/p>\n Source data<\/a><\/p>\n<\/div>\n<\/div>\n Full size image<\/span><\/a><\/p>\n<\/figure>\n<\/div>\n To directly match DisP-seq signals with the TFs identified, we performed ChIP\u2013seq for AP-2\u03b1 and NFIB in SKNMC cells. Similar to our results for EWS-FLI1, AP-2\u03b1 and NFIB sites were associated with strong DisP-seq signals centered on the corresponding TF-binding sites and with varying levels of H3K27ac and ATAC-seq (Fig. 2e,f<\/a>). These signals were substantially higher than those at nonoverlapping GABP\u03b1 sites despite strong GABP\u03b1 ChIP\u2013seq signals (Extended Data Fig. 3d<\/a>). Examples of DisP-seq signals associated with AP-2\u03b1 and NFIB peaks are shown in Extended Data Fig. 3e,f<\/a>.<\/p>\n To explore the distribution of AP-2\u03b1 and NFIB peaks identified by DisP-seq, we assigned these disordered TF peaks to six categories that we defined using ChromHMM (Fig. 2g<\/a>). We found that most AP-2\u03b1 and NFIB peaks are located at active and weak enhancers (73% and 75%, respectively). This is a similar distribution to EWS-FLI1 (87% at enhancers). In sum, DisP-seq signals are closely associated with TFs that contain prominent IDRs and can be used to identify these proteins in an unbiased antibody-independent manner.<\/p>\n We next sought to test whether changes in cellular states can lead to major differences in the distribution of DisP-seq signals. For this purpose, we measured DisP-seq signals in EWS-FLI1 depletion experiments, where the loss of the fusion protein is known to result in widespread changes in chromatin25<\/a><\/sup>. This process involves not only the loss of active signals at EWS-FLI1 binding sites but also the reactivation of normal mesenchymal differentiation programs that are typical of mesenchymal precursors, the putative cells of origin of Ewing sarcoma10<\/a>,25<\/a>,35<\/a>,36<\/a>,37<\/a><\/sup>. Mesenchymal differentiation in Ewing sarcoma cells with reduced levels of EWS-FLI1 has also been linked to increased migration, invasion and metastatic potential23<\/a>,24<\/a>,38<\/a><\/sup>.<\/p>\n DisP-seq profiling showed 1,730 sites with decreased DisP-seq signals and 13,500 sites with increased signals after EWS-FLI1 depletion (Extended Data Fig. 4a\u2013c<\/a>). As expected, the top motif associated with decreased DisP-seq signals corresponds to EWS-FLI1 (Fig. 3a<\/a>, GGAA repeats, peak set 1). Increasing peaks were strongly enriched for the disordered TF NFIB (P<\/i>\u2009=\u200910\u22121118<\/sup>; Fig. 3a<\/a>, peak set 2), which we initially identified in DisP-seq profiles of wild-type SKNMC cells and contains a large IDR (Fig. 2a,d<\/a>). To verify the changes observed in DisP-seq peaks, we performed NFIB ChIP\u2013seq in EWS-FLI1 depletion experiments and compared these results to DisP-seq signals, and ChIP\u2013seq data for EWS-FLI1 and H3K27ac25<\/a><\/sup>. In agreement with our motif analysis, we found strong EWS-FLI1 ChIP\u2013seq signals in peak set 1 and marked increases in NFIB ChIP\u2013seq signals in peak set 2 (Fig. 3b,c<\/a>). These peak sets were matched with corresponding decreases or increases in H3K27ac and ATAC-seq signals (Fig. 3b,c<\/a>). Similar results were observed for DisP islands. A total of 486 DisP islands were lost and 1,306 DisP islands were gained upon EWS-FLI1 depletion. More than half of lost DisP islands were associated with EWS-FLI1 (Fig. 3d<\/a>, pattern A), while 82% of gained DisP islands contain NFIB (Fig. 3g<\/a>, pattern C). Our data thus show that DisP-seq signals can undergo substantial reorganization with changes in cellular states and that these profiles can be used to identify disordered TFs linked to these processes.<\/p>\n a<\/b>, Top motifs for increasing and decreasing DisP-seq peaks (peak sets 1 and 2). b<\/b>,c<\/b>, Top\u2014heatmaps depicting sites from peaks sets 1 (b<\/b>) and 2 (c<\/b>). DisP-seq, ChIP\u2013seq and ATAC-seq signal intensities are shown. Bottom\u2014composite plots for the same sites. d<\/b>, Schematic of different patterns of DisP islands lost upon EWS-FLI1 depletion. DisP islands overlap with at least one (pattern A) or no EWS-FLI1 sites (pattern B). e<\/b>, Representative example of DisP island classified as pattern A. f<\/b>, Boxplot showing changes of NFIB signals after EWS-FLI1 depletion. Signals are grouped by their distance to EWS-FLI1 peaks and by whether they are in DisP islands associated with EWS-FLI1 (n<\/i>\u2009=\u20092 biologically independent experiments). *P<\/i>\u2009<\u20090.05, **P<\/i>\u2009<\u20090.01 and ***P<\/i>\u2009<\u20090.001 (two-sided t<\/i>-test). Median value is shown as a line within the boxplot, which spans from the 25th to 75th percentiles. Whiskers indicate a 1.5\u00d7 interquartile range. P<\/i>0\u2013150\u2009bp<\/sub>\u2009=\u20093.40\u2009\u00d7\u200910\u221218<\/sup>, P<\/i>150\u20131\u2009kb<\/sub>\u2009=\u20090.04, P<\/i>1k\u20135\u2009kb<\/sub>\u2009=\u20090.004 and P<\/i>\u22655\u2009kb<\/sub>\u2009=\u20090.04. g<\/b>, Schematic of different patterns of gained DisP islands upon EWS-FLI1 depletion. DisP islands that overlap with at least one (pattern C) or no increased NFIB peaks (pattern D). h<\/b>, Representative example of DisP island classified as pattern C. i<\/b>, Boxplot showing changes in expression for NFIB target genes after EWS-FLI1 depletion. Upregulated genes associated with NFIB sites are classified according to whether they are inside or outside of DisP islands (n<\/i>\u2009=\u20092 biologically independent experiments). ***P<\/i>\u2009<\u20090.001 (two-sided t<\/i>-test). Median value is shown as a line within the boxplot, which spans from the 25th to 75th percentiles. Whiskers indicate a 1.5\u00d7 interquartile range. P<\/i>\u2009=\u20091.69\u2009\u00d7\u200910\u221207<\/sup> j<\/b>, GO analysis of upregulated genes associated with gained DisP islands after EWS-FLI1 depletion. k<\/b>, Schematic of DisP island reorganization after EWS-FLI1 depletion. In Ewing sarcoma cells, NFIB partially colocalizes with EWS-FLI1 in pathologic DisP islands. After EWS-FLI1 knockdown, NFIB signals are lost at GGAA repeats and gained at previously unoccupied NFIB sites to form physiologic DisP islands. The reorganization of DisP islands is associated with reactivation of normal mesenchymal differentiation programs in tumor cells. COF, cofactor.<\/p>\n<\/div>\n Full size image<\/span><\/a><\/p>\n<\/figure>\n<\/div>\n In addition to these gained NFIB peaks, we also noted the presence of strong NFIB ChIP\u2013seq signals near EWS-FLI1 sites before EWS-FLI1 depletion (Fig. 3b<\/a>), including 270 of 272 EWS-FLI1 dependent DisP islands. This may be partly explained by the fact the half-site NFIB motif (TGGCA) is similar to GGAA and can be found within imperfect segments of GGAA microsatellites (64% of EWS-FLI1 bound GGAA repeat sites; Extended Data Fig. 4d<\/a>) and by immunoprecipitation experiments that show an interaction between EWS-FLI1 and NFIB (Extended Data Fig. 4e<\/a>). However, our data also shows that NFIB peaks within EWS-FLI1 DisP islands (pattern A) can be separated from the fusion protein by long distances (Fig. 3e<\/a>) and prompted us to consider the relationship between EWS-FLI1 and nearby NFIB peaks in more detail. Remarkably, our data show that depletion of EWS-FLI1 strongly decreases NFIB binding (Fig. 3e,f<\/a>) at both short and long distances between peaks (less than 150\u2009bp and also between 150\u2009bp and 1\u2009kb). Moreover, NFIB peaks that are contained in EWS-FLI1-associated DisP islands are affected more strongly by the depletion of the fusion protein. This effect is less pronounced but still noticeable beyond 1\u2009kb (Fig. 3f<\/a>). Our data thus show effects of EWS-FLI1 on NFIB binding that are beyond the 50\u2009bp range typically observed for cooperative binding39<\/a><\/sup>, suggesting that DisP islands can facilitate the coordinated binding of disordered TFs over large genomic regions.<\/p>\n Given the strong connection between NFIB and new DisP islands that appear after EWS-FLI1 depletion, we also characterized gene regulation programs associated with these events. NFIB is part of a family of TFs with roles in the development of various tissue types, including the differentiation of mesenchymal precursors40<\/a><\/sup>. Previous studies have also shown that NFIB can increase chromatin accessibility in small-cell lung cancer (SCLC) cells and stem cells41<\/a>,42<\/a><\/sup>. Our results show that increases in NFIB binding mainly occur in distal regions (Extended Data Fig. 4f<\/a>) and lead to the establishment of DisP islands (Fig. 3g,h<\/a>). To explore the role of gained DisP islands in gene regulation, we examined upregulated genes near increasing NFIB signals after EWS-FLI1 KD. This analysis shows that increases in expression levels were more pronounced for genes associated with NFIB-containing DisP islands compared to NFIB peaks outside of DisP islands (Fig. 3i<\/a>). Moreover, the signals of DisP-seq, NFIB and H3K27ac ChIP\u2013seq are also higher in peaks located in gained DisP islands (Extended Data Fig. 4g<\/a>). These results show that incorporation of NFIB into an environment with high local concentration of disordered proteins in DisP islands leads to enhanced effects on transcription programs.<\/p>\n Gene ontology (GO) analysis of potential target genes of these gained DisP islands shows a strong association with mesenchymal programs induced after EWS-FLI1 depletion (for example locomotion, adhesion and migration; Fig. 3j<\/a>), pointing to a role for the reorganization of DisP islands in these processes. Moreover, sites with DisP-seq peaks in these DisP islands display high average DNase I signals in mesenchymal cell types profiled by ENCODE (113 human cell types, GSE29692<\/a>), suggesting that they correspond to regulatory sites in this lineage43<\/a><\/sup> (Extended Data Fig. 4h<\/a>). In contrast, NFIB peak locations in DisP islands associated with EWS-FLI1 in SKNMC cells show low signals in all ENCODE cell types, suggesting that they are associated primarily with the Ewing sarcoma pathologic state (Extended Data Fig. 4h<\/a>).<\/p>\n Taken together, our results show the reorganization of the DisP island landscape upon EWS-FLI1 depletion (Fig. 3k<\/a>). In wild-type SKNMC cells, NFIB is partially sequestered by EWS-FLI1 in pathologic DisP islands, which are exclusively observed in the context of Ewing sarcoma. After EWS-FLI1 depletion, NFIB is released from these sites and relocates to NFIB binding sites linked to mesenchymal differentiation to establish physiologic DisP islands. In addition, the small increase in the level of NFIB protein after EWS-FLI1 KD may also contribute to the formation of these physiological DisP islands (Extended Data Fig. 4b<\/a>). These changes are linked to the reactivation of latent mesenchymal differentiation programs in tumor cells through increased chromatin accessibility and enhancer activation.<\/p>\n Given the strong association between NFIB and changes in DisP-seq signals, we next considered whether the IDR of NFIB has a role in the binding and function of this TF. NFIB contains a large IDR in its C-terminal region (Fig. 2d<\/a>) and we generated a mutant lacking this domain (NFIB\u0394IDR<\/sup>; Fig. 4a<\/a>). We first compared wild-type NFIB (NFIBWT<\/sup>) and NFIB\u0394IDR<\/sup> by b-isox precipitation and, as expected, only NFIBWT<\/sup> signals were detected by Western blot in the precipitated fraction (Fig. 4a<\/a>).<\/p>\nResults<\/h2>\n
DisP-seq signals are widely distributed in the genome<\/h3>\n
<\/picture><\/a><\/div>\nDisP-seq reveals many large clusters of disordered proteins<\/h3>\n
DisP-seq peaks are enriched for specific TF motifs<\/h3>\n
<\/picture><\/a><\/div>\nDisP signals are reorganized by changes in cellular states<\/h3>\n
<\/picture><\/a><\/div>\nDisP islands can enable sequestration of disordered TFs<\/h3>\n
IDRs can mediate the incorporation of TFs into DisP islands<\/h3>\n
![\"Science](\"http:\/\/media.springernature.com\/lw685\/springer-static\/image\/art%3A10.1038%2Fs41587-023-01737-4\/MediaObjects\/41587_2023_1737_Fig4_HTML.png\")
<\/picture><\/a><\/div>\n