The oncogenic fusion protein EWS-FLI1 promotes premature ageing of biomolecular condensates by catalyzing fibril formation

Ewing sarcoma (EwS) is an aggressive pediatric cancer of bone and soft tissue. A chromosomal translocation that joins the low-complexity domain of EWS (EWSLCD) with the DNA-binding domain of FLI1 (FLI1DBD) creates EWS-FLI1, a fusion oncoprotein essential for EwS development and accounts for 85% of all EwS cases. EWS-FLI1 acts as an aberrant transcription factor and interferes with the normal functions of nucleic acid-binding proteins via multivalent interactions and biomolecular condensation. The FLI1DBD was found to directly interact with the EWSLCD causing enhanced phase separation and induced hardening of EWSLCD condensates. Three related ETS DBDs (ERG, ETV1 and PU.1) also induced EWSLCD condensate hardening. DNA binding blocked the interaction with the EWSLCD, and NMR spectroscopy confirmed that ETS DBDs interact with EWSLCD via the DNA-binding interface. Our results provide a physical basis for the dominant-negative effect EWS-FLI1 exerts on EWS and highlight the need for further investigations of the FLI1DBD-EWSLCD interaction in vivo.


Introduction
The oncogenic EWS-FLI1 fusion protein is the archetypical example of a related group of fusion proteins characteristic of Ewing sarcoma (EwS), an aggressive pediatric bone and soft tissue cancer 1 .
Arising from the t(11;22)(q12;q24) chromosomal translocation that fuses the N-terminal low-complexity domain (LCD) of the RNA-binding protein EWS (EWS) in frame with the DNA-binding domain (DBD) of the E-twenty-six transformation-specific (ETS) family transcription factor Friend leukemia integration 1 (FLI1), the resultant EWS-FLI1 fusion is responsible for approximately 85% of all EwS tumors 1 (Fig. 1a). EWS-FLI1 acts as a pioneering factor aiding in chromatin opening, yet these observations do not fully explain its oncogenic role in EwS. There is mounting evidence that EWS-FLI1 exerts a dominant negative effect on the normal roles of EWS in transcriptional regulation and splicing [2][3][4][5] . Indeed, it was recently noted that the presence of EWS-FLI1 at transcriptionally active sites prevents the release of Breast cancer type 1 susceptibility protein (BRCA1) from DNA-directed RNA polymerase II subunit RPB1 (RNA Pol II), resulting in elevated transcriptional stress and subsequent accumulation of unresolved R-loops 2 .
The FET family of RNA-binding proteins is named after its three members: fused-in-sarcoma (FUS), EWS, and TATA-binding protein associated factor 2N (TAF15). All FET proteins are implicated in neurodegenerative diseases amyotrophic lateral sclerosis and frontotemporal dementia, and various cancers including prostate, leukemia and sarcoma [6][7][8][9] . FET proteins contain an N-terminal LCD (also known as the transactivation domain) and a C-terminal RNA binding domain (Fig. 1a, Supplementary   Fig. 1). The LCD has low sequence complexity, is characterized by a degenerate repeated motif, SYGQ, and lacks stable secondary structure. These properties confer a propensity for self-association and liquidliquid phase separation (LLPS) [10][11][12] .
The ETS transcription factor family has 28 members characterized by a highly conserved DBD with a winged helix-turn-helix fold that recognizes an eight nucleotide consensus with a core GGAA/T sequence 13 . ETS transcription factors regulate genes involved in processes that can be tumorigenic when dysregulated such as cell cycle control, proliferation, migration, invasion, apoptosis, and angiogenesis [14][15][16][17] . For reasons that are not entirely clear, FLI1, transcriptional regulator ERG (ERG), and protein FEV (FEV) most commonly participate in the FET-ETS fusions characteristic of EwS. However, rare chromosomal rearrangements involving FET proteins with other ETS transcription factors have been identified in EwS and in the related family of primitive neuroectodermal tumors (PNETs) 18 .
Transcriptional activation is achieved through the recruitment of RNA Pol II via its C-terminal domain (CTD) in a phosphorylation-dependent manner 11,12,20 .
In recent years, biomolecular condensation, a process in which proteins and nucleic acids demix from the aqueous phase via LLPS to form membraneless organelles, has emerged as a sub-cellular organizational paradigm [28][29][30] . These membraneless compartments regulate essential cellular processes such as transcription 31,32 , splicing 33 , DNA damage repair 34 and the stress response 35 . Intrinsically disordered proteins or proteins with intrinsically disordered regions (IDRs), including FET family proteins, are crucial components of biomolecular condensates 10 . Indeed, oncogenic fusions of EWS-FLI1 appear to form dynamic yet specific assemblies with other LCD-containing proteins in cells such as EWS 12,36 and the CTD of RNA Pol II via multivalent interactions mediated by its LCD 37 .
Furthermore, self-association stabilizes the binding of EWS-FLI1 to GGAA microsatellites and is required transcriptional activation, but the exact nature of this self-association remains contentious 11,12,19 .
Since the oncogenic function of EWS-FLI1 requires features of both the EWS LCD and the FLI1 DBD , the interplay between these two domains was investigated. The properties of biomolecular condensates of EWS, EWS LCD in the presence and absence of EWS-FLI1, FLI1 DBD and ETS DBDs were investigated with biophysical approaches. Fluorescent recovery after photo bleaching (FRAP) and Thioflavin T (ThT) assays revealed that the dynamics of intermolecular interactions are altered in EWS-FLI1 condensates relative to EWS and EWS LCD condensates. FLI1 DBD catalyzed rapid hardening of EWS LCD condensates, an effect that was inhibited by DNA binding, which inhibited colocalization of FLI1 DBD to EWS LCD condensates. The rapid ageing effect was a general feature of ETS domains. NMR confirmed that ETS DBDs transiently interact with EWS LCD via two loops involved in DNA binding. Both the charge density and the relative conformation of the loops was inferred to be important for the EWS LCD interaction. These findings provide a structural explanation for the observed dominant-negative effect of EWS-FLI1 as well as help to explain the apparent toxicity of the fusion, even to EwS tumor cells.
Considering these data, further structural investigations as to how the interaction between EWS LCD and ETS DBDs might affect intermolecular interactions of FET-ETS fusions are warranted.

Results
EWS-FLI1 has altered condensation properties. Phase diagrams for full-length FUS, full-length EWS and full-length EWS-FLI1 were constructed from turbidity measurements. Phase separation of EWS and FUS was promoted at increased protein and low NaCl concentrations, consistent with previous reports for FUS, where it was determined that phase separation is driven by electrostatic interactions between Arg residues in the Arg-Gly-Gly (RGG) motifs and Tyr residues in the LCD 38,39 (Fig. 1b). Conversely, phase separation of EWS-FLI1 was promoted by high protein and NaCl concentrations (Fig. 1b), suggesting hydrophobic interactions drive condensation. EWS and EWS-FLI1 are known to interact in vivo, likely via multivalent LCD-LCD interactions in biomolecular condensates 36 . As a way of examining these LCD-LCD interactions, biomolecular condensates were prepared by mixing 25 µM EWS LCD , which is common to both EWS and EWS-FLI1 (EWS 1-264, Fig. 1a, Table 1) with 9µM wildtype (WT) EWS, 5 µM EWS-FLI1, or 25 µM EWS LCD (Fig. 1c). WT EWS and EWS-FLI1 readily form dual-component condensates with EWS LCD , similar to in vivo observations in a EwS cell line A673 40 .
Condensate dynamics were assessed by monitoring fluorescence recovery after photobleaching (FRAP) of the fluorescently labelled molecules within the condensates (Fig. 1c). Condensates containing WT EWS and EWS LCD alone showed rapid FRAP, while condensates containing EWS-FLI1 recovered slower (Fig. 1c), indicating that EWS-FLI1 alters the dynamics of EWS LCD molecules within condensates and that EWS-FLI1 condensates are less fluid and more gel-like. FLI1 DBD alters the physical properties of EWS LCD condensates. The FRAP experiments revealed that FLI1 DBD may adversely affect EWS condensate properties. The effect of FLI1 DBD on the phase separation propensity of EWS LCD was tested using turbidity measurements and microscopy (Fig. 2a).
The addition of equimolar concentrations of FLI1 DBD (25µM) significantly increased the turbidity of the sample, indicating an increase in phase separation, and dramatically altered the morphology of the condensates as visualized by bright-field microscopy (Fig. 2a). These condensates no longer coalesce and instead appear fusion-defective (Supplementary Movies 1 and 2). To determine whether this effect was specific to FLI1 DBD , equimolar concentrations of constructs including the RNA-recognition motif (RRM) of EWS (EWS RRM or EWS RRM-RGG2 , 25µM) were also tested (Fig. 2a, Table 1, Supplementary   Fig. 1). The EWS RRM construct induced a slight increase in turbidity and resulted in EWS LCD condensates that appeared more spherical than those formed in the absence of RRM (Fig. 2a). The EWS RRM-RGG2 construct induced an increase in turbidity that was less than that observed for FLI1 DBD , and induced spherical EWS LCD condensates (Fig. 2a). In contrast to FLI1, EWS LCD condensates formed with the EWS RRM and EWS RRM-RGG2 constructs retained the ability to fuse (Supplementary Movies 3 and 4). Therefore, FLI1 DBD increases the phase separation propensity of EWS LCD , has a different effect on the morphology of the condensates compared to the RNA-binding domains that are naturally found in WT EWS, and changes the material properties of the condensates such that they are fusion-defective. Biomolecular condensates formed by LCD-containing proteins similar in composition to EWS LCD "ripen" or "age" spontaneously over time becoming more gel-like 41 . Cross-b structure has been shown to stabilize the condensed state of the closely related FET protein, FUS 42 , and condensate ageing is concurrent with the formation of fibrillar structures that sometimes protrude from the condensates 41 . To investigate the effect of the FLI1 DBD on the formation of cross-b structure in EWS LCD condensates, ThioflavinT (ThT) fluorescence assays were developed. Under non-phase-separating conditions (no NaCl), ThT fluorescence was not observed for EWS LCD alone, or in the presence of FLI1 DBD (Fig. 2b).
Bright-field microscopy revealed that EWS LCD remains in the dilute phase, but the addition of FLI1 DBD was associated with the appearance of several small condensates (Fig. 2b), consistent with FLI1 DBD increasing the phase separation propensity of EWS LCD (Fig. 2a), however no measurable increase in turbidity was observed under these conditions. Under phase-separating conditions (150 mM NaCl), EWS LCD condensates age slowly over approximately 10 hours, with a small concomitant increase in ThT fluorescence (Fig. 2c), an effect similar to that reported for FUS 43,44 . The addition of FLI1 DBD , even at sub-stoichiometric concentrations (1:10 molar ratio), to EWS LCD condensates significantly accelerated condensate ageing in a manner dependent on the concentration of FLI1 DBD (Fig. 2c). FLI1 DBD also increased ThT fluorescence of WT EWS under phase-separating conditions ( Supplementary Fig. 2a). An EWS-FLI1 truncation mutant corresponding to the minimal region of the fusion required for oncogenic transformation 19 (Table 1 Aged, ThT positive samples of EWS LCD and EWS LCD with FLI1 DBD were analyzed via transmission electron microscopy (TEM) to assess amyloid fibril formation. Although fibrils formed by EWS have been observed 45,46 and fibril formation within FET protein condensates has also been reported 10,41 , the observed structures did not display the typical morphological features of amyloid fibrils. Instead, heterogeneously branched structures that appeared intertwined were observed that are consistent with a pre-fibrillar or proto-fibril state 45 (Fig. 2c, Supplementary Fig. 2c). Therefore, the observed increase in ThT fluorescence arises from the formation of cross-b structure that has not had time or is incapable of organizing into bona fide amyloid fibrils. As amyloid fibrils are not known to be associated with EwS, this line of investigation was not pursued and instead ThT fluorescence provided a convenient assay for probing the equilibrium between dilute and condensed phase EWS LCD . DNA binding to FLI1 inhibits EWS LCD condensate aging. The ability of FLI1 DBD to alter EWS LCD condensate properties suggests that the two proteins directly interact. To determine if the interaction site involves the FLI1 DNA-binding site, ThT assays were conducted in the presence a double stranded high-affinity consensus sequence for ETS DBDs 47 (HA DNA, Table 2). HA DNA inhibited FLI1 DBDinduced aging of EWS LCD condensates in a concentration-dependent manner (Fig. 3a). At the highest concentrations of HA DNA tested (50 µM corresponding to a 10:1 DNA:FLI1 DBD ratio), the effect of FLI1 on EWS LCD condensates was almost entirely abolished ( Supplementary Fig. 3a). A double stranded DNA oligonucleotide containing 10 tandem GGAA repeats, which are known to bind FLI1 DBD in vitro 27,48 , also inhibited the effect FLI1 DBD exerts on EWS LCD condensates (Fig. 3b, Table 2). A scrambled dsDNA control sequence had no inhibitory effect, demonstrating that the inhibitory effect of DNA in these ThT assays is due to DNA binding by FLI1 (Fig. 3b, Table 2).
To quantify if FLI1 DBD associates with EWS LCD condensates or remains in the dilute phase, aged samples (24 hrs old) from ThT assays were pelleted. Though condensates are readily visible via microscopy, the majority (> 95 %) of EWS LCD remains in the dilute phase (supernatant) in the absence of FLI1 DBD (Fig. 3c). However, in the presence of FLI1 DBD , the partitioning of EWS LCD into the pellet fraction was noticeably higher (~ 50 % of total EWS LCD , Fig. 3c), consistent with FLI1 DBD enhancing EWS LCD phase separation (Fig. 2a). FLI1 DBD also partitioned into the pellet fraction (~ 55 % of total FLI1 DBD ), indicating direct association with EWS LCD . Addition of HA DNA and FLI1 DBD reduced the partitioning of both EWS LCD and FLI1 DBD (~ 90 % of each protein remains in the dilute phase) into the pellet further indicating that DNA inhibits the EWS LCD -FLI1 DBD interaction. HA DNA alone had no effect on EWS LCD phase separation and FLI1 DBD incubated alone remained in the dilute phase (Fig. 3c).
Furthermore, the partitioning of FLI1 DBD with EWS LCD in the pellet was a specific phenomenon as EWS RRM-RGG2 remained mostly in the supernatant fraction when incubated with EWS LCD condensates under identical conditions ( Supplementary Fig. 3b).
Fluorescence microscopy was used to characterize multicomponent EWS LCD condensates.
EWS LCD,650 (labeled with DyLight 650) was mixed with FLI1 DBD,488 (labeled with DyLight 488) under phase-separating conditions. Fluorescence from FLI1 DBD,488 spatially overlapped with fluorescence arising from EWS LCD,650 condensates, consistent with colocalization of FLI1 DBD in EWS LCD in condensates (Fig. 3d). FLI1 DBD,488 does not phase separate under these conditions. This colocalization is specific and further reinforces the notion that EWS LCD and FLI1 DBD interact since unrelated proteins such as green fluorescent protein (GFP) do not colocalize to EWS LCD condensates unless fused to EWS LCD (Supplementary Fig. 4a). Surprisingly, mixing EWS LCD,488 condensates with unlabeled FLI1 DBD and HA DNA 650 revealed that HA DNA was mostly excluded from the condensates (Fig. 3d).
Exclusion of DNA from EWS LCD condensates may be due to a lack of charge neutralization of the DNA phosphate backbone within the condensates since out of 264 residues the EWS LCD contains only two positively charged residues in contrast with six negatively charged and 27 tyrosines with a delocalized electron in their sidechains. DNA exclusion from the condensate indicated that the inhibitory effect of DNA in the ThT assays may arise from decreased partitioning of FLI1 DBD into EWS LCD condensates due DNA binding. To test this, fluorescence microscopy experiments using EWS LCD,650 , FLI1 DBD,488 , and increasing concentrations of unlabeled HA DNA were conducted (Fig. 3e). A fluorescence intensity ratio (Iin/Iout) was calculated to quantify partitioning of FLI1 DBD in the condensates (Fig. 3e). In the absence of DNA, Iin/Iout was ~ 5.8 ± 2.0. As the concentration of HA DNA increased, the intensity of FLI1 DBD,488 in condensates became noticeably dimmer, and Iin/Iout reduced to 1.6 ± 0.2 for EWS LCD,650 with FLI1 DBD,488 and 50 µM HA DNA (Fig. 3e). Therefore, FLI1 DNA binding out-competes the interactions between EWS LCD and FLI1 DBD that drive condensate colocalization.
Exclusion of HA DNA from EWS LCD condensates was also observed when full-length EWS-FLI1 was substituted for FLI1 DBD (Supplementary Fig. 4b).
FRAP was used to determine whether colocalization of FLI1 DBD to EWS LCD condensates results in a reduction in the dynamics of EWS LCD within the condensates, as observed for EWS-FLI1 (Fig. 1c).
Freshly prepared EWS LCD condensates underwent rapid recovery after photobleaching, indicating that the condensates are liquid-like (Fig. 3f). In contrast, freshly prepared condensates formed in the presence of FLI1 DBD displayed slower fluorescence recovery, indicating that the dynamics of EWS LCD molecules within the condensates change in the presence of FLI1 DBD (Fig. 3f). When EWS LCD condensates were prepared with FLI1 DBD and increasing concentrations of HA DNA, fluorescence recovered more rapidly in a DNA concentration-dependent manner, consistent with DNA binding inhibiting the effect of FLI1 DBD on EWS LCD (Figs. 3e and f).

The FLI1 DNA-binding and dimer interfaces are not involved in the interaction with EWS LCD . The
ThT assays and fluorescent colocalization data indicate that FLI1 DBD interacts directly with EWS LCD .
The DNA recognition helix (a3) harbors two highly conserved arginine residues that contact core bases of the ETS consensus sequence in all solved crystal structures of ETS DBDs complexed with DNA [49][50][51][52][53] ( Fig. 4A, Supplementary Fig. 5a). To test if these arginines participate in Arg-Tyr stacking interactions with EWS LCD , both arginines were mutated to leucines (R2L2). This mutant is incapable of binding HA DNA ( Supplementary Fig. 6a) and its effect on EWS LCD condensates was assessed using ThT assays (Fig. 4b). Surprisingly, the R2L2 mutant retained the ability to enhance the rate of EWS LCD condensate ageing (Fig. 4b). Together these results suggest that the positively charged DNA recognition helix of FLI1 may not be involved in the EWS LCD -FLI1 DBD interaction.
The structures of ETS DBDs solved to date share the same fold ( Supplementary Fig. 5b), however the precise margins of the ETS domain vary in the existing literature. In some studies an 85 amino acid construct comprising only three a-helices is used 4,54-56 , while other studies used a longer construct with a fourth a-helix 4,22,50 . A recent study crystalized FLI1 DBD as a dimer with the fourth a-helix participating in the dimer interface 50 . The FLI1 DBD construct used in the ThT assays includes this fourth a-helix (residues 362 -369) as well as ~30 C-terminal disordered residues. To determine whether this fourth ahelix with exposed hydrophobic residues interacted with EWS LCD , a FLI1 DBD construct truncated at residue 361was subjected to our ThT assays (Fig. 4a, Table 1, Supplementary Fig. 1). As anticipated, this construct retained DNA binding activity as judged by electrophoretic mobility shift assays ( Supplementary Fig. 6b). Further, the shorter FLI1 DBD construct exerted the same effect as the longer construct on EWS LCD condensates in ThT assays (Fig. 4c). Together, these observations further delineate the EWS LCD interaction site that is responsible for inducing EWS LCD ageing in condensates to the folded core of FLI1, between residues 276 and 361 but excluding the DNA recognition helix (a3, Fig. 4a).  Fig. 7). Bright-field microscopy images were acquired for each sample at the end of the ThT assay, (T > 10 hrs) (Fig. 5). The control EWS RRM and EWS RRM-RGG2 constructs had minimal effects on the rate at which ThT fluorescence increased for EWS LCD condensates and the condensates remained well dispersed and predominantly spherical (Fig. 5). In contrast, all four ETS DBDs significantly enhanced the rate at which the condensates aged (Fig. 5). Furthermore, EWS LCD condensates formed with ETS DBDs were irregularly shaped, most notably in the presence of PU.1 DBD and ETV1 DBD (Fig. 5). At higher concentrations of FLI1 DBD (> 50 µM), EWS LCD condensates were highly irregularly shaped, even at T ~ 0 hours ( Supplementary Fig. 8).

Enhancement of EWS
The addition of HA DNA along with ERG DBD and PU.1 DBD revealed that DNA binding had the same inhibitory effect on EWS LCD condensate ageing as was observed for FLI1 DBD , suggesting a common mechanism for ETS domains (Supplementary Fig. 9a). Additionally, the ETS DBD constructs, EWS RRM or EWS RRM-RGG2 constructs alone were incapable of forming ThT-positive structures ( Supplementary Fig. 9b). Potential contribution of the 8x His-tag used for purification of ETS DBDs, EWS RRM and EWS RRM-RGG2 proteins was assessed with a His-tag free version of PU.1 DBD (Supplementary Fig. 9c). Since the EWS RRM and EWS RRM-RGG2 constructs also retained His-tags but did not induce condensate ageing and the His-tag free PU.1 DBD induced condensate ageing at the same rate as His-tagged PU.1 DBD , the His-tag did not non-specifically induce ageing of EWS LCD condensates ( Supplementary Fig. 9c). These finding suggest that the effect FLI1 DBD exerts on EWS LCD condensates is conserved for members of the ETS TF family.  (Fig. 6a,   Supplementary Fig. 10). The signal intensity uniformly decreased between the initial and last titration points likely due to co-aggregation of PU.1 with EWS LCD condensates (Fig. 6b). However, a few peaks were differentially broadened, and coincided with or were located near residues with CSPs (Figs. 6a and b). Residues with CSPs greater than one standard deviation and residues with differential signal intensities less than one standard deviation were plotted onto the AlphaFold 58 structure of human PU.1 (Fig. 6c). Notably, these residues clustered to one face of the DBD, primarily incorporating residues in the ETS DBD "wings" (Loops 4 and 6) that contact the phosphate backbone of DNA 51 , however no shifts or differential broadening were observed for the DNA recognition helix, supporting our earlier hypothesis that it is not involved in the interaction with EWS LCD (Fig. 6d). Furthermore, no shifts or peak broadening were observed on the opposite face of the DBD (Fig. 6c). The CSPs and differential broadening observed for the disordered C-terminal tail were deemed non-specific since they are not conserved in other ETS DBDs and since C-terminally truncated ETS domains (FLI1 DBD, Da4 ) retain condensate ageing activity (Fig. 4c).

ETS DBDs interact with the EWS
Comparison of the sequences of PU.1 loops 4 and 6 to those of ERG, FLI1 and ETV1 revealed that the total charge varies between +6 to +1 between the DBDs (Fig. 6d). The loops in PU.1 are flexible and enriched in Lys residues and thus have a high net charge of +6, loops 4 and 6 of FLI1 and ERG are identical with a net charge of +3, and ETV1 has a net charge of +1 ( Supplementary Fig. 11). Each loop contains at least one highly conserved positively charged residue (Arg or Lys) at 220 and 247 (PU.1 sequence numbering, Fig. 6d). The ThT assays revealed that PU.1 DBD induced condensate ageing at the fastest rate, followed by FLI1 DBD and ERG DBD , and ETV1 DBD induced ageing at the slowest rate (Fig.   5). Therefore, the net charge of loops 4 and 6 is correlated to the rate at which the ETS DBD induces EWS LCD condensate ageing with higher overall charge inducing the fastest ageing. This correlation suggests that the interaction between ETS DBDs and EWS LCD that drives condensate ageing is influenced by electrostatic interactions. However, electrostatic interactions are clearly not the only factor that influences condensate ageing because the EWS RRM-RGG2 construct is enriched with positively charged Arg residues, yet this construct does not affect EWS LCD condensate ageing like the ETS DBDs ( Fig. 5). Therefore, it is likely that the relative positioning of the two loops in the ETS DBD structure is also an important factor contributing to EWS LCD condensate ageing.

Discussion
Though the roles of EWS are not completely defined, knockout of EWS is postnatal lethal in mice, indicating it functions in homologous recombination, the DNA damage response, and in splicing 2,59-61 .

The phenotype induced by expression of EWS-FLI1 mimics EWS knockdown phenotype in HeLa and
EwS cells leading to hypothesis that EWS-FLI1 exerts a dominant-negative effect on the normal cellular functions of EWS 2,4,5,62 . This work demonstrated that FLI1 DBD enhanced the phase separation propensity of EWS LCD , localized specifically to EWS LCD condensates, and accelerated condensate ageing. This effect was conserved for three other ETS DBDs. Boulay et al. recently reported that b-isoxazole mediated precipitation of EWS-FLI1 in EwS cell lines was enhanced relative to WT EWS 19 , supporting the hypothesis that the interaction between EWS LCD and FLI1 DBD enhances phase separation. Previous studies in prostate cancer cell lines also identified interactions between EWS and ERG, ETV1, ETV4 and ETV5 and these were proposed to be necessary and sufficient for the tumor phenotype 6 .
Furthermore, co-immunoprecipitation experiments demonstrated a direct interaction between PU.1 and FUS that inhibits the normal functions of FUS in splicing 63,64 . Therefore, mis-localization of ETS domains either as fusions or due to aberrant regulation negatively impacts the function of EWS and possibly other nucleic acid-binding proteins. The interaction between ETS DBDs and EWS LCD appears to involve residues that are at least partially occluded by DNA binding. As a result, DNA binding reduces the interactions between ETS DBDs and EWS LCD that modulate phase separation and promote the formation of cross b-structure within EWS LCD condensates (Fig. 7). The in vivo situation is more complicated since other proteins, nucleic acids, and post-translational modifications will modulate the properties of the condensate, thus fibril formation is not a likely endpoint in EwS, rather the dominant negative effect is attributable to enhanced self-association of EWS and likely other, related proteins.
Self-association of EWS LCD stabilizes EWS-FLI1 binding to GGAA microsatellites, helps recruit RNA Pol II, and is required for transcriptional activation of GGAA-associated genes that drive While TEM did not conclusively demonstrate the formation of amyloid fibrils, fibril formation by EWS has been observed previously 45,46 , though unlike typical amyloid fibrils, these are labile to disassembly 45,66 . Furthermore, hydrogels of the LCD of FET family proteins have been shown to consist of amyloid-like fibrils 10,11 . There are no high resolution structural models for EWS fibrils, although several models exist for short segments of FUS 67,68 and for the entire LCD of FUS 42 . Assuming EWS LCD forms cross-b structures in a similar way, via the formation of parallel, in-register b-strands, a key structural feature of these fibrils would be "ladders" of identical sidechains stacked at an interval corresponding to the interstrand distance 69 . These stabilizing ladders could be formed by aromatic tyrosine residues that undergo p-p stacking or by glutamine/asparagine residues that form complimentary hydrogen bonds across the stacked b-strands 69 . Structural insight remains elusive for how ETS DBDs enhance the rate at which EWS LCD condensates become ThT positive, but tyrosine residues in the EWS LCD are important for functional self-association 19,70 . Therefore, tyrosine residues in EWS LCD may interact with the positive charges (cation-π) on ETS DBD loops in a conformation conducive for aromatic ladder formation which in turn may serve to catalyze the formation of cross-b structure. This interaction is expected to be extremely transient and thus the increased high-local concentration of the EWS LCD in condensates enhances the rate of cross-b formation, consistent with the observation that the rate of ThT-positivity is enhanced under phase-separating conditions. The CTD of RNA Pol II has been shown to bind to hydrogels comprised of fibrillar polymers formed by the LCD of FET family proteins, and the degree of CTD binding correlates with the degree of transcriptional activation 11 . It is therefore possible that the interaction between FLI1 DBD and EWS LCD in the fusion protein alter the biophysical properties of EWS-FLI1 condensates in a way that enhances recruitment of RNA Pol II via CTD interactions.
While fluorescence microscopy revealed, unexpectedly, that HA DNA is mostly excluded from condensates formed by EWS LCD , the in vivo situation involving WT EWS and other proteins is likely to be different. Cation-p interactions between tyrosine residues in the LCD and arginine residues in the RGG motifs likely modulate phase separation of WT EWS, as has been observed for FUS 71  EWS-FLI1 drives oncogenesis through dysregulation of genes downstream of GGAA microsatellites and dysregulation of alternative splicing programs, a function dependent on an intact DBD 1,61,73 . As part of its oncogenic function, EWS-FLI1 also associates with a multitude of nucleic acid binding proteins, including the remaining copy of EWS 36 , FUS, and RNA Pol II 3,21,74 . These interactions remain poorly understood, likely due to their heterogenous and transient nature. The central finding presented here is that ETS DBDs colocalize to and interact transiently yet specifically with the EWS LCD , altering the intermolecular interactions that govern its LLPS and subsequent condensate ageing (Fig. 7).
These results provide a mechanism that may explain how EWS-FLI1 interferes with the normal functions of EWS and potentially other crucial nucleic acid binding proteins. The presence of the FLI1 DBD (or other ETS DBD) promotes self-association and aggregation of EWS altering the local dynamics and intermolecular interactions crucial for EWS function.

Methods
Protein expression and purification. Protein constructs used are listed in Table 1 Table 2) were dissolved in 20 mM sodium phosphate buffer pH 7.4, 150 mM NaCl (or 0 mM NaCl), 10 µM ThT and prepared in triplicate in 384-well flat bottom black plates (Greiner Bio-One, NC). Immediately prior to the start of the assay, 2 µL of a 625 µM stock of EWS LCD (or EWS-FLI1 or WT EWS) in 20 mM CAPS pH 11 was added and mixed thoroughly by pipetting to initiate phase separation and the plate was sealed (Thermo Scientific). ThT fluorescence was read at 10-minute intervals for 24 hours at 25°C using a Tecan Infinite M200 plate reader (Tecan Trading AG). Raw fluorescence intensities were normalized to a control sample.
Pelleting assays. Aged samples (T > 10 hours) from ThT assays were removed from 384-well plates and centrifuged at 21,300 x g for 15 minutes at ambient temperature. The supernatant was removed, and the pellet was resuspended in 8 M urea. The supernatant, pellet, and a sample prior to centrifugation ("total") were analyzed by SDS-PAGE.

Microscopy.
Freshly prepared (T ~ 0-1 hour) or aged samples (T ~ 24 hours) were imaged directly in sealed 384-well microwell plates with a BioTek Cytation Gen 5 imaging plate reader (Agilent) using a 20 x objective. Alternatively, sealed, chambered (50-well) coverslips (Grace Biolabs), coated with 1% Pluronic F-127 were imaged with an Olympus FV300 inverted confocal microscope using a 40x oilimmersion objective lens operating at 1% power using the 488 nm laser for transmitted light and the 488 nm and 650 nm lasers for fluorescence imaging. Images were acquired simultaneously in differential interference contrast (DIC) and fluorescent modes. Images were processed using Fiji 78  Electrophoretic mobility shift assays. 20 µL samples were loaded into wells of a 20% polyacrylamide gel prepared in 0.5 x TBE buffer. Electrophoresis was performed for 60 minutes at 150 V using 0.5 x TBE and stained with SYBRsafe (Thermo Fisher, MA, USA) according to the manufacturer's specifications for 30 minutes before imaging using UV transillumination. Circular dichroism. Circular dichroic spectra were recorded on 10 µM samples of all proteins in 50 mM sodium phosphate buffer pH 6.5 at 25°C in a 2 mm pathlength cuvette using a Jasco 810 spectropolarimeter (Jasco, OK) at a scan speed of 50 nm/min with 0.2 nm increments. Each sample was recorded in triplicate, the data was then averaged and converted to mean residue ellipticity. Data Availability: NMRPipe processing scripts are available upon reasonable request, expression plasmids containing the EWS, EWS-FLI1, and the ETS DBD constructs were deposited with Addgene (#####). The backbone resonance assignments for the PU.1 DBD were deposited in the BMRB (#####).
Author contributions: EES made the protein samples, developed the ThT assays, prepared reagents, collected, processed, and interpreted data, wrote manuscript, conceptualized the study, AKRM made protein samples, conducted phase separation assays, processed data, SA made protein samples, conducted phase separation assays, processed data, XX designed expression construct plasmids, purified protein, DSL collected and processed data, wrote manuscript, obtained funding, and conceptualized the study. All authors contributed to editing the manuscript and have read and approved the manuscript for publication.