Condensates in fusion oncoprotein-driven leukemia: new biology and therapeutic opportunities

Abstract

Chromosomal rearrangements leading to the formation of fusion oncoproteins (FOs) are drivers of leukemia and other cancers. Many such FOs are now known to localize in nuclear or cytoplasmic condensates formed through the process of phase separation. FO-associated condensates facilitate spatiotemporal organization of key macromolecules required for oncogenic transformation. NUP98 fusions provided the first example of FO-associated condensates in leukemia, and other leukemic drivers subsequently illustrated the shared and distinct mechanisms by which phase separation contributes to leukemogenesis. Improved understanding of FO-associated condensates has in turn revealed new opportunities for therapeutic targeting. Pharmacologic approaches including dissolution of condensates, inhibition of key condensate components, or modification of condensate material properties may translate to more effective treatments for patients with FO-driven cancer.

Fusion oncoproteins and phase separation play key roles in leukemia

Since the discovery of the BCR::ABL1 fusion over 40 years ago [1], cancer biologists and clinicians alike have recognized the important role of fusion oncoproteins (FOs) in leukemia. Additional gene fusions were identified throughout the 1990s (including RUNX1::RUNX1T1 (AML1::ETO) [2,3], PML::RARA [4], ETV6::RUNX1 (TEL::AML1) [5], and KMT2A (MLL) to diverse partners [6,7] and Nucleoporin (NUP) fusions [8–10]), with an increasing number of fusions and affected patients recognized in the early 2000s. The advent of increasingly accessible next generation sequencing approaches led to the identification of FOs in up to 40% of leukemias, with higher rates in certain disease lineages and in younger patients [11,12]. Screening for several FOs is now regularly used to aid in the diagnosis of leukemia, and our understanding of the impact of FOs on leukemia biology, prognosis, and treatment has also grown in recent years.

Numerous studies have uncovered the effects of individual FOs on gene expression, chromatin state, and protein interactions, providing mechanistic insight into the processes underlying FO-driven transformation. Among the most recent mechanisms implicated in FO-driven leukemia is phase separation (PS) [13–15], which mediates the formation of macromolecular assemblies via often transient, multivalent interactions [16,17]. These macromolecular assemblies, termed biomolecular condensates, are essentially membraneless organelles and can form in the nucleus and/or cytoplasm [16,17]. Condensate formation is dependent on macromolecule concentration(s), protein posttranslational modifications, and environmental conditions, among other factors [16,17]. Furthermore, condensate formation is in part promoted by intrinsically disordered regions (IDRs) within proteins, which lack secondary and tertiary protein structure and often exhibit key sequence-related physiochemical properties that promote multivalent interactions and PS [16–18]. The partitioning of high concentrations of key macromolecules (proteins, as well as DNA and RNA for nuclear condensates) provides precise spatiotemporal control that is important in both physiologic processes, such as gene regulation [19–23], cell signaling [24,25], DNA damage repair [26,27], RNA processing [28,29], and stress response [30] and disease states. PS was first linked to FO-driven cancer through fusions involving FET family RNA-binding proteins, which are frequently seen in mesenchymal cancers [21,31] (Box 1). Subsequent studies have extended and strengthened the connection between FO biology and PS. Here the emerging role of PS in FO-driven leukemia and avenues for therapeutic intervention are discussed.

Box 1. Phase separation in other FO-driven cancers.

FET FOs are drivers of various cancers and were the first fusions to be linked to PS. FET family proteins (including FUS, EWS, and TAF15) share a conserved low complexity domain, which is intrinsically disordered [93]. FET FOs involve the DNA binding domain of a transcription factor (TF), such as those in the ETS family of TFs. FET::ETS FOs bind to GGAA motifs found in repetitive genomic elements, or microsatellites, near the transcription start sites of FO-bound genes, ultimately causing upregulation of these FO-bound target genes [94]. Chong and coauthors first showed that FET FOs formed nuclear assemblies termed hubs via IDR-mediated interactions and that these condensate-like structures were important for gene regulation, although they only demonstrated that these structures formed through “apparent PS” under overexpression conditions and not at endogenous FO levels [21]. Subsequent work using single-molecule approaches demonstrated that FET FOs form biomolecular condensates with liquid-like properties and that these FO condensates interact with the C-terminal domain of RNA polymerase II [31]. These condensates in turn induce gene expression, particularly from GGAA microsatellite sequences that reach a threshold number of repeats [31]. Related work examined FUS::CHOP, a FO observed in myxoid liposarcoma, and showed that this FO forms phase-separated condensates in vitro and in cells [95]. FUS::CHOP condensates colocalize with BRD4 [95], suggesting that they may promote formation of super-enhancers and activate transcription [19,20]. FUS FO condensates also recruit members of SWI/SNF chromatin remodeling complexes [61,62], again suggesting their implications in gene regulation. These studies, and those of NUP98 FOs, established a precedent for PS in FO-driven cancers. In the past five years, accumulating evidence has supported the role of PS in the context of FOs across tumor types, which has been reviewed in more detail elsewhere [96]. Most of these FOs involve IDRs and DNA-binding or chromatin-modifying domains, such as TFE3 FOs in renal cell carcinoma [97,98], ZFTA and YAP FOs in ependymoma [99,100], SS18::SSX in synovial sarcoma [101], and BRD4::NUT FOs in NUT carcinoma [102]. While these FO condensates localize in the nucleus and modulate gene regulation, other examples of cytoplasmic FO condensates, including AML4::ALK [103] and CCDC6::RET [104] in lung and other cancers, have been reported and involve alternate mechanisms to rewire kinase signaling. Expanding work in this field will reveal likely additional roles for PS in FO-driven cancers, uncovering new shared and unique features for FO-associated condensates in leukemia and other cancer types.

Consequences and composition of leukemic FO condensates

Phase separation of NUP98 fusion oncoproteins facilitates leukemic transformation and gene deregulation

NUP98::HOXA9 was the first example of PS in the context of FO-driven leukemia [13–15] (Figure 1A, Key Figure). NUP98::HOXA9, as well as NUP98 FOs, are observed across a spectrum of hematologic malignancies, although they are most common in acute myeloid leukemia (AML). Together, NUP98 FOs are observed in ~5% of pediatric and ~3% of adult patients [11,32–34]. Importantly, NUP98-rearranged leukemia is associated with poor outcome, including chemoresistance and high rates of relapse [11,32,34,35].

Figure 1. Known condensates and corresponding mechanisms in FO-driven leukemia.

A) NUP98 FO condensates serve as transcriptional hubs to drive leukemic cell transformation and gene deregulation. Recent work shows that NPM1c and KMT2A FO condensates are biophysically indistinguishable from NUP98 FO condensates; collectively, these condensates have been referred to as C-bodies. B) As part of PML::RARA FO microspeckles, FO and BRD4 bind to super-enhancers and broad peaks to activate leukemic gene expression. C) FUS::ERG FO undergoes PS in sarcoma as well as leukemia, likely through similar mechanisms. The figure was created with Biorender.com. IDR, intrinsically disordered region; FG, phenylalanine-glycine; Y, tyrosine

NUP98 FOs universally retain the N-terminal, phenylalanine-glycine (FG)-rich IDR of NUP98 [32]. The normal role of NUP98 is within the nuclear pore complex, where it mediates the facilitated, selective transport of DNA, RNA, and proteins between the nucleus and the cytoplasm [36]. However, most NUP98 FO C-terminal partners contain a DNA-binding domain (as in the case of HOXA9) or one or more chromatin-binding or - modifying domains (e.g. plant homedomain (PHD), su(var)3–9, enhancer-of-zeste and thorax (SET) domain, and bromodomain (BRD)) [32]. While some of these domains possess enzymatic activity to write, read, or erase histone modifications (e.g. in NUP98:NSD1, ::KMT2A, ::PHF23 and ::KAT7), many do not; thus, the primary function of the C-terminal partners is generally as a transcriptional scaffold that enables interaction with chromatin. The resulting changes in gene regulation, particularly at developmental gene loci such as HOX gene clusters, lead to de-differentiation and self-renewal phenotypes and are a major mechanism by which NUP98 FOs drive leukemogenesis [37,38].

The punctate, nuclear localization pattern of NUP98::HOXA9 was recently linked to PS through various lines of evidence [13–15]. NUP98::HOXA9 condensates are sensitive to treatment with 1,6-hexanediol [13,14], a chemical known to disrupt phase-separated condensates formed through multivalent hydrophobic interactions [45]. Moreover, NUP98::HOXA9 forms liquid-like condensates at concentrations as low as 10 nM in vitro [15]. Additionally, in cells, fluorescence recovery after photobleaching (FRAP) experiments showed that NUP98::HOXA9 is mobile within condensates [15].

To understand the connection between NUP98’s IDR, PS, and disease, mutant versions of NUP98::HOXA9 were generated in which the number of FG repeats (or FG valence) was altered by mutation and/or deletion [13,15]. Alteration of all FG repeats prevented the formation of FO-associated condensates; instead of forming condensates, FG-poor versions of NUP98::HOXA9 exhibited diffuse localization throughout the nucleus [13,15]. In contrast, both in vitro and in HEK293T cells, more modest changes in FG valence were associated with lower condensate numbers and reduced partitioning into condensates compared to native NUP98::HOXA9 [13,15]. Furthermore, expression of mutant NUP98::HOXA9 constructs in hematopoietic stem and progenitor cells led to altered PS [13,15]. Only the expression of NUP98::HOXA9 FOs capable of forming condensates (native FO sequence and modest FG mutants) led to cell proliferation, self-renewal, and in vivo leukemogenesis. This result demonstrated both the association of condensate formation with leukemic phenotypes and the importance of homotypic and heterotypic interactions mediated by the NUP98 IDR to leukemic transformation [13,15]. Further experiments illustrated the critical role of additional heterotypic interactions between FO and DNA: when NUP98::HOXA9 was mutated to abrogate DNA binding, FO condensates were larger in size and fewer in number, indicating the influence of heterotypic interactions with chromatin on condensate morphology. Importantly, the DNA-binding mutant also did not transform hematopoietic stem and progenitor cells [15].

PS contributes to the formation of transcription centers with key roles in gene regulation [19–23,46]. Consistent with this, PS mediated by the IDR of NUP98::HOXA9 led to changes in gene expression and chromatin remodeling. Compared to native NUP98::HOXA9, expression of FG-poor, condensate-negative NUP98::HOXA9 mutants led to downregulation of development- and leukemia-associated cell genes, including HOXA9 and PBX3 [13,15]. These PS-defective FO mutants also exhibited decreased genomic occupancy at many of the same loci and no longer exhibited characteristics of super-enhancers, including profound loss of histone 3 lysine 27 acetylation (H3K27ac) [13]. Furthermore, chromosome conformation capture experiments demonstrated that NUP98::HOXA9 forms CTCF-independent loops between enhancers and target genes, but FG mutants do not [13]. NUP98 FO condensates remodel local chromatin at FO-bound sites, such that gene-activating histone modifications replace repressive ones [47]. These changes in chromatin state allow co-binding of interacting proteins, creating a feedforward mechanism to increase target gene transcription [47]. Together, these findings indicate that IDR-mediated PS of NUP98::HOXA9 rewires gene expression by targeting the FO to super-enhancer-like sites.

NUP98 FOs interact with proteins involved in gene activation and repression. Interactions between NUP98 FOs and CREBBP (CBP)/EP300 (p300) [39,48], HDAC1 [41,48], and WDR-SET-COMPASS complex members [49] have been demonstrated using immunoprecipitation followed by western blot analysis. NUP98 FOs and nuclear export factor XPO1 (also known as CRM1) also interact and share chromatin binding sites, including HOX gene clusters [40,50]. Unbiased proteomic approaches, such as liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS), identified NUP98 FO interactors involved in RNA splicing, ribosome biogenesis, and transcriptional control [14]. These studies also allow comparisons of the interactome for NUP98 FOs with various partner genes displaying HOX, PHD, or other domains. Most interactors were shared, including those with roles in biomolecular condensation, such as FUS, HNRNPA1, and GAR1 [14]. The overlapping interactome of NUP98 FOs with diverse partner proteins suggests that i.) interactions are mediated by the shared IDR of NUP98 and ii.) NUP98 FOs utilize molecular mechanisms shared by fusions with varied partner proteins.

To determine whether NUP98 FO interactors are present in condensates, biotinylated isoxazole-mediated condensome mass spectrometry (biCon-MS) [51] was used to enrich for PS-related factors [14]. Performing biCon-MS in HL-60 cells overexpressing NUP98 FOs confirmed that FOs were present in condensates and showed that FO expression enriches transcriptional machinery and leukemia factors in the condensome [14]. NUP98 FOs enriched in condensates were also identified by cross-linking mass spectrometry, which detected interactions on chromatin in HEK293T cells expressing empty vector, NUP98, or eight NUP98 FOs. This analysis revealed known and novel FO interactors, including members of MYST family histone acetyltransferase (HAT) complexes, discussed further below [52].

Thus, the NUP98 FO interactome includes proteins with key roles in gene regulation, along with many other interactors that remain uncharacterized in the context of NUP98-rearranged leukemia or FO-associated condensates. Moreover, previous studies have not fully elucidated which FO interactors localize within condensates, the extent of variability in condensate composition between different members of a FO family, or how individual interactors contribute to PS and oncogenicity and whether they represent therapeutic targets.

Other leukemic drivers also localize in condensates

NUP98 FOs share upregulation of HOXA genes and MEIS1 with other leukemia subtypes, including those harboring NPM1 mutations and KMT2A rearrangements [35]. NPM1 mutations occur in approximately one third of adult AML, involving insertions in a C-terminal exon of NPM1 that generate a nuclear export sequence (NES) to shuttle NPM1 to the cytoplasm [53,54]. Regardless, a portion of NPM1c binds to chromatin to activate leukemogenic genes, including many of the same targets upregulated by NUP98 FOs [55,56]. The chromatin-binding fraction of NPM1c localizes in condensates, which have been termed coordinating bodies (C-bodies), and NPM1c expression is both necessary and sufficient for condensate formation [57]. Truncation mutants of NPM1c capable of forming condensates maintain stem-like cell states and gene expression patterns [57], consistent with observations described above in the context of NUP98-rearranged leukemia [13,15]. Condensates formed by NPM1c, like those described for NUP98 and NUP214 FOs [52,58], colocalize with XPO1, NUP98, KMT2A, and MENIN, and with the HOXA9 locus [57]. Finally, when co-expressed in cells, NPM1c and other condensate forming FOs (NUP98::NSD1, SET::NUP214, and KMT2A::AFF1) form indistinguishable condensates, thus demonstrating that NUP/KMT2A FO condensates are equivalent to C-bodies (Figure 1A) [57].

Emerging data also supports the importance of PS in the context of other genetic and transcriptomic leukemia subtypes. The PML::RARA FO, a defining feature of acute promyelocytic leukemia, was long thought to exert its effects primarily through the destruction of promyelocytic leukemia (PML) nuclear bodies, which have tumor suppressive functions. Recent work, however, demonstrates the role of PML::RARA FO microspeckles in transcriptional activation [59]. The PML::RARA FO microspeckles that replace PML nuclear bodies in both APL cell lines and patient samples are formed through PS via IDR-mediated interactions (Figure 1B) [59]. Within these FO-associated microspeckles, PML::RARA binds to super-enhancers and broad-promoters, along with BRD4 and likely other chromatin modifiers, to increase the expression of target genes including GFI1 and MYC and prevent differentiation [59].

EWS::FLI1 and similar FOs, defining features of Ewing’s sarcoma, have been extensively studied in the context of PS (Box 1). A related FO, FUS::ERG, which has been identified in Ewing’s sarcoma as well as AML, also undergoes PS both in vitro and in cells (Figure 1C) [60]. While leukemic FUS::ERG FO condensates likely utilize similar biological mechanisms as their counterparts in sarcoma [31,46,61,62] (Box 1), a formal comparison of the lineage-specific features of these FOs has not been performed.

Mechanisms underlying other non-FO leukemic driver events have also been linked to biomolecular condensates. MN1 overexpression is observed across various cytogenetic subgroups of AML [63,64], and in some cases, AMLs involve translocations of MN1 to ETV6, STAT3, or FLI1 [65–67]. MN1 overexpression is associated with poor prognosis in AML [63], and overexpression of MN1 induces leukemia in mice [68,69]. Nevertheless, relatively little is known about the function of MN1, or how it contributes to leukemogenesis. MN1 overexpression leads to the localization of MN1 in nuclear condensates [64], where MN1 also colocalizes with BAF complex members on chromatin [64]. Notably, MN1 includes one of the longest polyQ stretches in the human proteome. MN1’s polyQ region stabilizes its interaction with the BAF complex and is required for leukemic transformation [64], although how this region of MN1 contributes to condensate formation was not reported.

ENL has also been studied in leukemia, both as a fusion partner of KMT2A and in the maintenance of oncogenic gene expression [70]. ENL is a transcriptional coactivator important for transcriptional elongation via association with the super elongation complex and DOT1L. While the KMT2A::ENL FO occurs in ~4% of KMT2A-rearranged AML [71], mutations in ENL’s YEATS domain have also been identified in other AML patients. These ENL mutants lead to the formation of biomolecular condensates, as well as enhanced transcriptional activation [72]. Moreover, ENL condensates colocalize with HOX genes as well as with CDK9 and AFF4, ultimately functioning as “pathologic elongation condensates” [73]. Various domains within mutant ENL affect condensate formation and cofactor interactions [73], illustrating important charge-based and residue-specific determinants of PS [74] (Box 2).

Box 2. Sequence determinants of FO condensates.

The intrinsically disordered and folded regions of FOs define their PS ability and effects on gene regulation, cell signaling, and other outputs [13,15,60]. While models have been established to predict PS across the broader human proteome [105–107], these often rely on data from the literature rather than on results generated under standardized experimental conditions. Rigorous and systematic testing of large sets of FOs have provided a more useful starting point to predict condensate formation and identify the key sequence characteristics contributing to PS [74]. Indeed, screening of 166 FOs demonstrated that their condensate formation is driven by distinct patterns of physicochemical features, which were leveraged to accurately predict condensate formation in other FOs using machine learning approaches [74]. Since PS is often substantially influenced by IDRs, subsequent studies have assessed FO-derived IDRs [18]. In a screen examining condensate formation of 215 FO-derived IDRs, 19% formed condensates by themselves in cells [18]. Like condensate-prone FOs, condensate-prone IDRs clustered into distinct groups on the basis of their physicochemical features, which could then be used to train an IDR-specific machine learning model to predict the behavior of IDRs across the human proteome [18]. This analysis revealed that condensate-prone IDRs function in RNA processing and transcriptional regulation and thus suggests that FOs containing such IDRs may both form condensates and regulate RNA and/or transcription as well [18].

Tools including the SAK3.0 computational pipeline [18,74] are important resources for condensate research in FO-driven cancer and other diseases. The SAK3.0 pipeline analyzes multiple properties of protein sequences, including sequence complexity, presence of conserved domains, disorder content, potential for pi-pi, pi-cation, and prion-like interactions, charge patterning, and physicochemical feature and amino acid enrichments. Although first developed for FOs, SAK3.0 also predicts the condensation likelihood for whole protein sequences, as well as for all IDRs identified within a sequence. This tool identifies sequence features associated with multivalent interactions that underlie PS, often within IDRs, and the associated enriched amino acids, enabling rational modulation of condensate formation through mutagenesis and tests of their roles in oncogenic cellular phenotypes [15,99]. While IDRs are often dominant mediators of PS by FO, they sometimes require synergy with flanking folded domains, additional IDRs and/or interaction partners to form condensates [18]. Tools like SAK3.0 and NARDINI+, a complementary platform from the Pappu lab [108], serve to guide elucidation of the features of FO and other protein sequences associated with condensate formation. Additionally, other recently developed artificial intelligence-based tools, such as ProtGPS [109] and FINCHES [110], may provide further insights to better predict the cellular localization of intrinsically disordered proteins and their interactions with other proteins.

Finally, truncating mutations in ASXL1, which confer poor prognosis [75] and occur in ~20% of AMLs as well as in chronic myelogenous leukemia (CML) and other hematopoietic malignancies [76,77], have also been linked to PS. ASXL1 is a component of the polycomb repressive deubiquitinase complex, which prevents gene silencing by removing ubiquitin from histone 2A lysine 119 (H2AK119), thus opposing the effect of polycomb repressive complex 1. The ASXL1 mutations observed in leukemia result in loss of the negatively-charged C-terminal IDR of ASXL1, while the positively-charged N-terminus is retained. Condensation of mutant ASXL1 leads to aberrant chromatin binding and transcriptional activation, at least in part through enhancing the concentration and activity of BAP1, another member of the polycomb repressive deubiquitinase complex [78,79].

Other examples of PS in fusion-positive leukemias are expected based on recent work [60,74]. An experimental screen of 166 FOs from diverse cancers revealed that 58% formed cellular condensates under overexpression conditions [74] (Box 2). One-half of these condensates were nuclear, one-third were cytoplasmic, and the remainder were observed in both the nucleus and cytoplasm. Of the leukemic FOs tested, ~75% exhibit punctate localization in the nucleus and/or cytoplasm, although in many cases the potential role of PS in FO-driven leukemogenesis is unknown (Table 1). Of these leukemic FOs, only the BCR::ABL1 FO displayed condensate localization solely in the cytoplasm. While this finding requires further investigation and has not been formally linked to PS, previous results suggest that the localization of BCR::ABL1 in stress granules (also formed via PS) is important for oncogenesis [80]. Building on these results, a machine-learning model based on sequence-derived physicochemical features was used to predict condensate formation in more than 3,000 previously uncharacterized FOs. (Box 2). 67% of these were predicted to form condensates and many displayed sequence features predictive of nuclear localization and involvement in regulation of gene expression [74]. Collectively, these findings warrant further investigation into how condensate formation and the associated potential for recruitment of functional biomolecular partners and altered cell physiology might apply more broadly in leukemia and cancer in general.

Table 1. Condensate-positive FOs in leukemia.

Data are taken from ref. [74]. PS, phase separation; AML, acute myeloid leukemia; APL, acute promyelocytic leukemia; B-ALL, B-cell acute lymphoblastic leukemia; MPAL, mixed phenotype acute leukemia; AMKL, acute megakaryocytic leukemia; CML, chronic myelogenous leukemia

FO (or FO family)	Primary disease indication	PS previously reported?	Condensate localization?
			Nuclear	Cytoplasmic
NUP98	AML	Yes [13–15]	Yes	No
KMT2A	B-ALL, AML	Yes [57]	Yes	No
PML::RARA	APL	Yes [59]	Yes	No
FUS	AML	Yes [60]	Yes	No
NUTM1	B-ALL	No	Yes	No
PAX5	B-ALL	No	Varies	Varies
ZNF384	B-ALL/MPAL	No	Yes	No
MEF2D	B-ALL	No	Varies	Varies
MLLT10	AML	No	Varies	Varies
TCF3	B-ALL	No	Yes	No
CBFA2T3::GLIS 2	AMKL	No	Yes	Yes
ETV6::RUNX1	B-ALL	No	Yes	Yes
BCR::ABL1	CML, B-ALL	No	No	Yes

Condensates provide opportunities for therapeutic targeting

Many FOs harbor IDRs which, by traditional standards, are considered undruggable. Further, while many leukemia-associated FOs contain folded domains, these often function in DNA or chromatin binding and generally lack small molecule inhibitors. Thus, the prevalence of IDRs coupled with the general lack of inhibitors against folded domains within FOs presents challenges for direct pharmacologic inhibition and will require identification of other avenues for therapeutic targeting. The key role of condensate formation in FO-driven leukemic transformation may offer new opportunities to identify and counteract key proteins and processes in currently intractable fusion-driven cancers.

Imaging-based screening approaches to identify compounds that dissolve condensates

Given the central mechanistic role of condensates in leukemia and other diseases, an important direction for therapeutic development is the pursuit of small molecules that disrupt condensate formation (Figure 2A). Indeed, imaging-based screening approaches have identified cytotoxic compounds in FO-expressing cells. A high-content, time-lapse imaging platform called DropScan was employed to evaluate the effects of 1,777 drugs on condensate formation by a DNA-binding deficient form of FUS::ERG [60]. Eight compounds were validated to dissolve condensates, including clinical CDK4/6 inhibitor LY2835219. LY2835219 likely acts through increased formation of lysosomes and leads to decreased expression of FO target genes [60]. Although work is needed to fully define the mechanism of LY2835219 and evaluate other validated hits, DropScan provides a starting point for similar high-throughput screening approaches in condensate-positive cancers. Alternatively, studies have utilized optogenetic approaches to dissolve condensates [81]. Notably, both this approach and DropScan depend on overexpression of tagged proteins rather than probing cancer-associated condensates formed by proteins at endogenous levels.

Figure 2. Potential mechanisms for therapeutic targeting of condensates.

A) Condensate dissolution, B) inhibition of FO interactors, and C) micropeptide killswitch-activated alteration of condensate material properties have shown promise as treatment strategies for condensate-positive, FO-driven leukemias. The figure was created with Biorender.com.

Emerging work seeks to better understand drugs capable of modifying FO and other condensates; these agents are termed c-mods. In addition to condensate-dissolving compounds, or “dissolvers”, such as LY2835219 described above, studies have identified small molecules and biologics that lead to the formation of new condensates or change the localization or material properties of existing condensates [82]. Furthermore, other research has examined how c-mods are partitioned into condensates. Antineoplastic drugs, such as commonly used anti-cancer agents cisplatin and tamoxifen, are highly concentrated in condensates, which influences their activity [83]. Furthermore, the enrichment and exclusion of small molecules in condensates has been investigated using machine learning approaches with large numbers of compounds, showing that the chemical environment of condensates and the physiochemical properties of the compounds (particularly solubility and hydrophobicity) are important contributors to their partitioning into condensates [84,85]. Importantly, continued efforts to identify compounds that dissolve condensates will inform artificial intelligence-based predictive modeling of FO and/or small molecule features that nominate effective condensate-disrupting therapies.

Condensate components as molecular dependencies and/or drug targets

Improved understanding of the composition of FO-associated condensates has also led to the identification of FO interacting proteins that are important for the fitness of leukemic cells and may serve as promising targets for therapeutic development (Figure 2B). NUP98-rearranged leukemia again offers some of the best examples of this approach. In particular, inhibitors of the transcriptional regulator Menin, a known interaction partner of NUP98 FOs [49], are the most clinically advanced and were Food and Drug Administration (FDA)-approved for the treatment of a subset of AMLs in 2024. Menin is a targetable dependency in NUP98-rearranged AML, with inhibitors showing efficacy in multiple preclinical models [86]. Menin inhibition is also effective in NPM1c and KMT2A-rearranged AML [87,88], presumably due to the comparable composition of condensates associated with these leukemic drivers [57]. While Menin inhibition does not dissolve condensates, it does deplete them of Menin, a key cofactor driving leukemic gene expression changes [57]. Similarly, mutation of the FG-rich IDR or the PHD finger of NUP98::PHF23 prevents colocalization of FO and Menin in condensates [47]. Moreover, both inhibition of Menin and degradation of WDR5 (a member of KMT2/MLL complexes) can displace NUP98 FOs from chromatin at key pro-leukemogenic sites [47,52,86].

Earlier work also highlighted Nucleosome Remodeling Factor (NURF) complex member SMARCA5 as a NUP98 FO interaction partner and functional dependency. SMARCA5 interacts not only with NUP98::NSD1 but also with the N-terminus of NUP98 and wildtype NSD1. Furthermore, in NUP98::NSD1/FLT3-ITD immortalized hematopoietic cells, shRNA knockdown of Smarca5 decreased colony formation in methylcellulose and expression of Hoxa9 and Meis1, although condensate formation was intact [89].

XPO1 interacts with both NUP98 and NUP214 FOs in condensates and shares HOX gene binding sites with these FOs [40,50,58]. The XPO1 inhibitor/degrader KPT-330 displaces SET::NUP214 from chromatin and disrupts FO condensates in LOUCY (SET::NUP214-positive) cells, although condensate formation is altered at drug concentrations below those required for cytotoxicity or apoptosis [58]. Consistent with this, treatment with the XPO1 inhibitors Eltanexor or Selinexor alters the formation of condensates, localizing NPM1c to the nucleolus, depleting condensates of XPO1, NUP98, and KMT2A, and leading to myeloid differentiation [57].

Key interacting proteins in NUP98-rearranged leukemia, including MYST family HATs, KAT6A and KAT7 have also been identified. Genetic knockout of each HAT decreased NUP98-rearranged cell fitness, and pharmacologic inhibition of KAT6A/7 using the small molecule PF9363 led to myeloid cell differentiation in cell models and in vivo [52]. Moreover, the combination of PF9363 and the Menin inhibitor SNDX-5613 was synergistic and overcame Menin inhibitor resistance in a highly refractory NUP98::NSD1 patient-derived xenograft model system [52]. Similar results were reported in KMT2A-rearranged leukemia [90], potentially via shared mechanisms involving condensates [57]. However, despite the effects of co-targeting KAT6A/7 and Menin on FO occupancy on chromatin, chromatin modeling, gene expression and leukemic cell growth, neither PF9363, SNDX-5613, nor the combination, significantly altered condensate integrity, at least in the overexpression model used [52].

Inhibition of condensate interactors has also been considered in the context of PML::RARA FO condensates [59]. BRD4 inhibition using the small molecule JQ1 did not dissolve FO condensates but prevented BRD4 colocalization with condensates, displaced BRD4 from chromatin, and reversed FO-induced gene expression changes [59]. These results, like observations above, highlight the importance of targeting key functional interacting partners of FOs but not necessarily a requirement for condensate dissolution. Additional studies will be needed to further dissect how drug treatment impacts condensate function and to determine which inhibitor or inhibitor combinations targeting the condensate interactome are most effective.

A micropeptide kill switch alters the material properties of condensates

Another promising strategy to target condensate-positive leukemias (and other diseases linked to PS) involves disrupting the material properties of condensates (Figure 2C). Recruitment of a micropeptide “killswitch”—composed of 17 amino acids including 3 key phenylalanine residues—significantly alters the material properties but not formation of condensates induced by fusion-recipient proteins, including NUP98 and others [91]. The killswitch inhibits the partitioning and dynamics of key condensate components, as demonstrated by depletion of RNA polymerase II in BRD4::NUT condensates and delayed recovery after XPO1 FRAP in NUP98::DDX10 condensates [91]. The killswitch also rendered condensates insensitive to dissolution after 1,6-hexandiol treatment [91]. In NUP98::KDM5A mouse leukemia cells, the presence of the killswitch prevented cell transformation and FO target gene expression [91]. Killswitch expression also reduced both the number of FO condensates as well as the FO protein level in hematopoietic cells; this finding highlights the delicate balance of material properties required to maintain oncogenic condensates in leukemia and likely other cancer cells [91]. Altogether, use of a micropeptide killswitch can alter condensate composition, component dynamics, and oncogenic function. Since the killswitch requires introduction using protein engineering and exhibits varying effects in the context of diverse physiologic and pathologic condensates, further studies will be necessary to learn how to most effectively disrupt the material properties of condensates. Importantly, recent computational work has successfully designed proteins capable of binding to arbitrary intrinsically disordered peptides or proteins [92]. This may provide a starting point in nominating peptides capable of altering condensate properties. Such peptides might then be refined via rigorous experimental testing to inform the development of novel therapeutics.

Concluding remarks

Growing evidence supports the prevalence and function of biomolecular condensates in FO-driven leukemia, as well as other subtypes of leukemia and other diseases. Work from our group and others has provided evidence of the importance of PS in NUP98-rearranged leukemic transformation [13–15]. Additionally, increasing work now offers promising opportunities to leverage our growing knowledge of condensate biology for therapeutic benefit. While many questions remain regarding the specific composition of condensates across different classes of leukemia FO, as well as which features are critical for their functions in gene regulation and leukemogenesis (see Outstanding Questions), the insights gained from successful examples of condensate targeting usher in a new realm of pharmacology. Exciting possibilities currently under exploration in PS-dependent leukemia may extend to condensates in other cancers and disease types.

Outstanding Questions Box.

Which homotypic and heterotypic interactions are important for FO-associated condensate formation? Which sequence features contribute to these interactions and how do they facilitate recruitment of functional partners, including the transcriptional machinery?

What additional proteins, regions of chromatin and RNA molecules partition into FO condensates?

What are the best approaches to therapeutically target FO condensates?

How do effective treatments impact the composition, dynamics, and function (including epigenetic and transcriptional outputs) of FO condensates?

How are drugs partitioned into FO condensates, and does this influence how they alter condensate features or lead to cell death/differentiation?

Which FO-associated condensate components are functionally important and therapeutically targetable? Does targeting multiple FO interactors provide synergistic benefit?

Which aspects of condensate biology are FO- and/or cancer type-specific, and which others are generalizable across condensate-positive cancers and even other condensate-related diseases?

Highlights.

• Many fusion oncoproteins (and other genetic drivers) in leukemia localize in condensates formed via PS

• Fusion oncoprotein-associated condensates play key roles in leukemic transformation, often via their impact on chromatin state and transcriptional regulation

• Agents that alter the integrity, composition and/or material properties of condensates represent potential treatments for fusion-driven cancers

Glossary

1,6-hexanediol

aliphatic alcohol often used to dissolve condensates, which acts by weakening hydrophobic interactions

Biomolecular condensates

membraneless organelles formed through PS of biomolecules via weak, multivalent interactions, which perform diverse functions in physiology and disease

C-mods

condensate modifying drugs, which lead to condensate formation (inducers) or dissolution (dissolvers), alter condensate localization (localizers), or change condensate material properties (morphers)

Chromosome conformation capture

epigenomic technique involving DNA cross-linking, digestion using restriction enzyme(s), ligation, and sequencing, which is used to identify how DNA is arranged in 3D space within the nucleus

Fluorescence recovery after photobleaching (FRAP)

microscopy technique used to measure the diffusion of fluorescently-tagged biomolecules into and within condensates in living cells

Fusion oncoprotein (FO)

chimeric protein resulting from chromosomal rearrangement, which contributes to the development and/or progression of cancer

Homotypic interactions

interactions between two or more of the same macromolecules (for example, interactions between multiple NUP98 FOs)

Heterotypic interactions

interactions between different macromolecules (for example, interactions between NUP98 FO and XPO1 or NUP98 FO and chromatin)

Intrinsically disordered regions (IDRs)

amino acid sequences that lack secondary and tertiary structure and instead adopt flexible 3D conformations

Multivalent interactions

many weak, noncovalent bonds that together contribute to dynamic, multisite interactions between macromolecules

Liquid-liquid phase separation (LLPS)

PS yielding two liquid phases (one macromolecule-rich dense phase and another macromolecule-poor dilute phase), which is driven by more energetically favorable interactions of macromolecules with one another versus with the solvent

Phase separation (PS)

reversible transition mediating the formation of two or more separate phases, e.g. a light phase and dense phase(s) with a range of possible material states (e.g., liquid-like, viscoelastic, or solid-like)

Transcription factor

protein that binds to DNA sites within chromatin to regulate gene expression

Super-enhancer

cluster of enhancers bound by multiple transcription factors (such as BRD4 and/or Mediator) that have dramatic effects on gene regulation and key roles in establishing cell identity, marked by high levels of H3K27ac and frequently dysregulated in cancer and other diseases