Heterotypic interactions can drive selective co-condensation of prion-like low-complexity domains of FET proteins and mammalian SWI/SNF complex

Abstract

Prion-like domains (PLDs) are low-complexity protein sequences enriched within nucleic acid-binding proteins including those involved in transcription and RNA processing. PLDs of FUS and EWSR1 play key roles in recruiting chromatin remodeler mammalian SWI/SNF (mSWI/SNF) complex to oncogenic FET fusion protein condensates. Here, we show that disordered low-complexity domains of multiple SWI/SNF subunits are prion-like with a strong propensity to undergo intracellular phase separation. These PLDs engage in sequence-specific heterotypic interactions with the PLD of FUS in the dilute phase at sub-saturation conditions, leading to the formation of PLD co-condensates. In the dense phase, homotypic and heterotypic PLD interactions are highly cooperative, resulting in the co-mixing of individual PLD phases and forming spatially homogeneous condensates. Heterotypic PLD-mediated positive cooperativity in protein-protein interaction networks is likely to play key roles in the co-phase separation of mSWI/SNF complex with transcription factors containing homologous low-complexity domains.

Introduction

Biomolecular condensates such as stress granules and transcription factories are membrane-less subcellular bodies that form and are regulated via the phase separation of multivalent proteins and nucleic acids^1,2,3. The physiological functions of biomolecular condensates range from signaling hubs under normal conditions to storage depots in response to cellular stress³. Sequence analyses of the proteins enriched in intracellular condensates both in the nucleus and cytoplasm have previously revealed an abundance of proteins containing long stretches of intrinsically disordered prion-like domains^4,5. Prion-like domains (PLDs) are typically characterized by their low complexity sequence features with an overrepresentation of aromatic (Y/F) and polar amino acids (G/S/Q/N) and depletion of charged residues^6,7. Proteins with PLDs have been identified in all life forms^8,9,10,11,12. Prion proteins were initially discovered as proteinaceous infectious agents in bovine spongiform encephalopathy and other neurodegenerative diseases^4,13,14,15, but are increasingly recognized with key functional roles in driving phase separation of RNA-binding proteins in the cell, and in the formation of functional amyloids^16,17.

What roles do PLDs play in the context of protein phase separation? Multivalent cohesive interactions between PLD chains as well as the chain-solvent interactions, which are encoded by the PLD primary sequence composition and patterning^18,19, have been recognized to be a key feature driving the phase separation of isolated PLD chains and PLD-containing proteins^{20,21,22,23,24,25}. Previous studies using hnRNP A1 PLD have demonstrated that the distributed aromatic (Y/F) amino acids act as “stickers” that mediate PLD-PLD interactions^18,26, whereas the polar amino acids (S/G/Q) can be described as “spacers”, which regulate chain solvation and cooperativity of sticker-sticker interactions. The importance of tyrosine residues in driving PLD phase separation has been further demonstrated experimentally for FUS²⁶ and EWSR1²⁷, and computationally for a large number of PLD sequence variants²⁸. In addition to aromatic residues, charged residues such as arginine (R) and polar amino acids such as glutamine (Q) also play smaller but important roles in PLD phase separation²⁸. Together, the sticker and spacer residues regulate PLD phase separation in a context-dependent manner¹⁹. In a broader context, however, the sequence grammar encoding LCD phase separation can be more complex and additional factors beyond aromaticity, such as the net charge and hydrophobicity of LCDs are likely to play equally dominant roles in driving their phase separation^29,30. When part of a multi-domain protein, π-π and cation-π interactions mediated by the aromatic and arginine residues in PLDs have been shown to drive phase separation of many full-length RNA and DNA binding proteins including FUS, EWSR1, TAF15, hnRNP A1, and EBF1^{21,31,32,33,34}. Further, debilitating point mutations in PLDs have been reported to promote the pathological transformation of protein condensates from a liquid-like state to solid aggregates^20,35,36. Thus, PLDs play important roles in the context of functional protein phase separation as well as disease processes associated with the formation of aberrant biomolecular condensates.

Many intracellular biomolecular condensates, such as stress granules and transcriptional hubs, are known to contain a multitude of proteins with ^4,5,37. Despite being broadly classified as prion-like based on the frequencies of certain amino acids in a protein sequence, as noted above^6,7, individual PLD chains typically feature distinct sequence composition, amino acid patterning, and chain length^4,22,38. Do PLDs from distinct yet functionally related proteins interact with one another and undergo co-phase separation? Previous studies have reported that the PLDs in transcription factors, including the FET family of fusion oncoproteins, not only drive their phase separation but also facilitate the recruitment of essential coactivators, such as the catalytic subunit of the mammalian SWI/SNF (mSWI/SNF) complex, BRG1, in transcriptional condensates^{32,36,39,40,41}. Interestingly, BRG1 contains an N-terminal LCD that is prion-like, which can engage with the PLDs of FET fusion proteins via heterotypic interactions³². Although homotypic phase separation of some PLDs, such as FUS and hnRNP A1, are well characterized^18,19,31, little is known about how heterotypic interactions regulate the co-phase separation of PLD mixtures⁵ and how the sequence features of respective PLD chains contribute to this process. In general, in multi-component mixtures of multivalent LCDs, homotypic and heterotypic interactions between LCD chains can either positively cooperate, negatively cooperate, or form coexisting phases, resulting in a diverse phase behavior and dense phase co-partitioning. The interplay between the specificity and strengths of homotypic and heterotypic interactions is expected to dictate the co-condensation versus discrete condensate formation in an LCD sequence-specific manner⁴². Further, PLD-containing proteins such as FUS have recently been reported to form a heterogeneous pool of homo-oligomeric complexes below their saturation concentration for phase separation⁴³, which are thought to represent distinct functional states of the protein than the condensates that form at higher concentrations⁴⁴. However, a key unanswered question is whether heterotypic PLD interactions occur at sub-saturation conditions, which may provide a mechanism for LCD-mediated functional protein networking, such as interactions between FET oncofusions and mSWI/SNF complex, in the absence of phase separation.

Motivated by these open questions, here we systematically investigate the phase behavior of PLD mixtures encompassing FUS^PLD and the PLDs from the chromatin remodeler mSWI/SNF complex that aberrantly interact with FUS fusion oncoproteins in transcriptional reprogramming⁴⁵. Our study incorporates PLDs from four mSWI/SNF complex subunits: ARID1A, ARID1B, SS18, and BRG1, which are key components for spatiotemporal transcriptional regulation and chromatin remodeling^46,47,48. Employing in vitro experiments in conjunction with mammalian cell culture models, we show that there exists a broad range of saturation concentrations (�sat) of PLD chains in vitro that directly correlate with their ability to form phase-separated condensates in live cells. We find that, except BRG1, mSWI/SNF subunit PLDs undergo phase separation with �sat values substantially lower than the known PLDs of RNA-binding proteins such as TAF15, EWSR1, and FUS^18,31, suggesting a greater degree of homotypic interactions. Similar to FET PLDs, the phase separation propensity of mSWI/SNF PLDs is primarily dependent on aromatic residues, specifically tyrosine residues, and to a smaller extent on polar amino acids such as glutamine. Despite strong homotypic interactions, mSWI/SNF PLDs engage in heterotypic interactions with FUS^PLD, resulting in co-partitioning in the dense phase with partition coefficients that show a positive correlation with the number of aromatic residues. In mixtures of PLD condensates, individual PLD phases undergo complete mixing, and together, they form spatially homogeneous PLD co-condensates. These findings indicate that homotypic and heterotypic PLD interactions act cooperatively in the dense phase despite substantially different saturation concentrations of individual PLD chains, which we posit to be a direct manifestation of similarity in PLD sequence grammars. Importantly, heterotypic PLD-PLD interactions between FUS and mSWI/SNF subunits are detectable at sub-saturation concentrations in vitro and in live cells, indicating strong affinities between these low-complexity domains in the absence of phase separation. The observed specificity in interactions among PLDs is further highlighted by a lack of interactions between these PLDs with a functionally distinct non-prion-like LCD. We conjecture that PLD-mediated selective co-condensation of multiple subunits of the mSWI/SNF chromatin remodeling complex with FET fusion proteins may constitute an important step in establishing transcriptionally relevant protein interaction networks.

Results

Prion-like domains of mSWI/SNF subunits form dynamic phase-separated condensates in live cells

mSWI/SNF chromatin remodeler complex is enriched in subunits that have large disordered low complexity regions with unknown functions⁴⁹. Many of these disordered regions have prion-like sequences (Fig. S1)³². Since PLDs of RNA and DNA binding proteins can drive phase separation and contribute to the formation of biomolecular condensates in cells^4,21,31, we investigated whether mSWI/SNF subunit PLDs are phase separation competent. We selected the top four PLDs in the complex based on their length, functional and disease relevance, which correspond to the following subunits – BRG1 [catalytic subunit], ARID1A, and ARID1B [among most mutated proteins in cancer⁴⁸], and SS18 [relevant to fusion oncoprotein SS18-SSX⁵⁰] (Figs. 1a, b; S1). We noted that although the prion prediction algorithm PLAAC⁶ categorizes these low complexity domains as prion-like, these PLDs have varying sequence composition and their lengths are significantly higher than the PLDs from RNA binding proteins (Fig. 1b; Tables S1–3). To determine if they were phase separation competent, we titrated concentrations of recombinant PLDs in vitro (buffer: 125 mM NaCl, 25 mM Tris.HCl pH 7.5) and observed that apart from BRG1^PLD, all other PLDs form spherical condensates in a concentration-dependent manner (Figs. 1c, d; S2a). Further, ARID1B^PLD condensates showed cluster-like morphologies upon phase separation, suggesting a percolation-type network formation⁵¹ (Fig. 1c). Based on the optical microscopy data, we quantified the saturation concentrations (�sat) for the PLDs as ≤2.5 μM for ARID1A^PLD and SS18^PLD, and ≤ 5.0 μM for ARID1B^PLD (Figs. 1d; S2a). Under similar experimental conditions, FUS^PLD undergoes phase separation with a �sat of ≤ 200 μM (Fig. S2a)⁵², which is almost two orders of magnitude higher than ARID1A^PLD and SS18^PLD. Although BRG1^PLD did not phase separate under these conditions (Fig. S2a) it can be induced to form spherical condensates in the presence of a macromolecular crowder (20% Ficoll PM70; Fig. S2b)³². These data suggest that except for BRG1, mSWI/SNF subunit PLDs are highly phase separation competent. We next probed whether these condensates are dynamic using fluorescence recovery after photobleaching (FRAP) experiments. FRAP recovery traces indicate that all PLD condensates have liquid-like properties with varying diffusivity dynamics. (Fig. 1e). Based on the FRAP traces, we find that ARID1A^PLD forms the most dynamic condensates with more than 80% recovery, SS18^PLD is intermediate with ~60% recovery, and ARID1B^PLD is the least dynamic with less than 40% recovery within the same observational timeframe. The reduced dynamicity of ARID1B^PLD condensates is consistent with the percolation-driven network formation observed for these condensates (Fig. 1c).

Although PLDs have emerged as a driver of many ribonucleoprotein phase separation under physiological and pathological conditions, expression of these domains alone typically does not lead to the formation of condensates in live cells^20,32,42,53. This is consistent with their known �sat values in vitro, which range from 100 to 200 μM and are typically much higher than their intracellular concentrations^18,31,52,54. Since mSWI/SNF PLDs show low micromolar �sat values in vitro, we posited that they may form condensates in live cells at relatively low expression levels compared to FUS^PLD. To test this idea, we transiently transfected HEK293T cells with GFP-PLD plasmids. Upon expression, ARID1A^PLD, ARID1B^PLD, and SS18^PLD readily formed spherical nuclear foci, whereas BRG1^PLD and FUS^PLD remained diffused at all expression levels (Fig. 1f). To estimate the relative �sat of mSWI/SNF PLDs within the nucleus, we used GFP fluorescence intensity as a proxy for concentration and leveraged the stochastic nature of intracellular PLD expression that spanned over two orders of magnitude. We observed that SS18^PLD has the lowest �sat followed by ARID1A^PLD and ARID1B^PLD (Figs. 1g, S3). This rank order of cellular saturation concentrations is similar to their in vitro phase behavior (Fig. 1d). FRAP experiments revealed that the nuclear condensates of ARID1A^PLD, ARID1B^PLD, and SS18^PLD are dynamic (Fig. 1h). Interestingly, the morphology of the PLD condensates varied with their subcellular localization. Spherical condensates formed within the nucleus and irregular, yet dynamic, assemblies were observed in the cytoplasm (Fig. S3c, d). Such differences could arise from the distinct intracellular microenvironment of the cytoplasm and the nucleus, such as the viscoelasticity of chromatin fibers, altered post-translational modifications, and high abundance of RNAs in the nucleus, which can markedly influence the coarsening behavior and biophysical properties of condensates^{55,56,57,58,59}.

Tyrosine residues play a dominant role in mSWI/SNF subunit PLD phase separation

The phase separation capacity of PLDs from RNA-binding proteins has been attributed to multivalent interactions predominantly mediated by the distributed aromatic and arginine residues^18,28. While SS18^PLD has a lower fraction of aromatic and arginine residues (0.11) than FUS^PLD (0.14; Tables S1−3), it possesses a greater number of aromatic and arginine residues (39) than FUS^PLD (24). To test if increasing the number of aromatic residues can improve the phase separation driving force of FUS^PLD without changing the overall sequence composition, we created a dimer of FUS^PLD, termed FUS^2XPLD (Fig. 2a), which possesses a total of 48 aromatic residues at a fixed fraction of 0.14. In contrast to the FUS^PLD, which remained diffused at all expression levels, we observed that FUS^2XPLD formed phase-separated condensates in the cell nucleus at a relatively low expression level (Fig. 2b) similar to the three mSWI/SNF subunit PLDs (Fig. 1f). The estimated intracellular saturation concentration of FUS^2XPLD was observed to be similar to that of ARID1A^PLD, ARID1B^PLD, and SS18^PLD (Figs. 1g, 2b; Figs. S3, S4a). Analogous to mSWI/SNF subunit PLD condensates, FRAP experiments revealed that FUS^2XPLD condensates have a high degree of dynamic behavior (Fig. S4b). To test whether the stronger driving force for phase separation of FUS^2XPLD primarily stems from the greater number of tyrosine residues and not simply from its increased length, we further created a variant of FUS^2XPLD, termed FUS^{2XPLD halfYtoS}, where we replaced the tyrosine residues to serine in the second half of the FUS^2XPLD (Table S4). This sequence variation led to a complete loss of phase separation of FUS^2XPLD in living cells even at 10-fold higher intracellular concentrations (Fig. 2a, b; Fig. S3), implying that the number of tyrosine residues is a key determinant of phase separation in this PLD. Next, to test if tyrosine residues are also important for the phase separation of mSWI/SNF subunit PLDs, we first created an ARID1A^PLD variant where we mutated all 29 tyrosine residues to serine (29Y-to-S), termed ARID1A^{PLD YtoS}. We observed that 29Y-to-S substitution abolished the ARID1A^PLD phase separation in the cell at all expression levels (Fig. 2c; Fig. S3). Apart from tyrosine residues, ARID1A^PLD primary sequence shows enrichment of glutamines with multiple polyQ tracts with stretches of three to four Gln residues in the C-terminal region (Tables S1, S2, S4). Previous studies have suggested that polyQ regions can promote LCD self-association^60,61. However, when we mutated these Gln residues to Gly and created an ARID1A^PLD variant, termed ARID1A^{PLD 30QtoG}, we observed only a modest (~2 fold) increase in intracellular C_sat (Fig. 2c; Fig. S3). These data suggest that Tyr residues play a greater role in driving homotypic ARID1A^PLD phase separation, similar to the FUS^PLD, whereas polar residues such as Gln play a moderate role.

Based on the results obtained from the sequence perturbations of FUS^2XPLD and ARID1A^PLD, it appears that intracellular phase separation of PLDs can be tuned by the number of Tyr residues. Since BRG1^PLD, which only contains seven aromatic residues but a large number of proline residues, does not phase separate in cells, we attempted to improve its condensation driving force by increasing the tyrosine content. To this end, we created two variants where we mutated 17 proline and 41 proline residues to tyrosine residues, termed BRG1^{PLD Aro+} and BRG1^{PLD Aro++}, respectively. We observed that both BRG1^PLD variants can form intracellular condensates with comparable C_sat to other mSWI/SNF PLDs (Fig. 2d; Fig. S3). However, the BRG1^{PLD Aro++} was observed to form large irregular aggregates in the cytoplasm and was predominantly excluded from the nucleus (Fig. 2d; Fig. S3), which is likely due to strong homotypic interactions mediated by a large number of Tyr residues in this synthetic BRG1^PLD variant.

mSWI/SNF PLD condensates recruit low-complexity domains of transcriptional machinery and RNA polymerase II via heterotypic interactions

An emerging feature underlying transcriptional regulation by prion-like low complexity domains in transcription factors is their ability to directly engage with chromatin remodeler SWI/SNF complexes^34,40,45 and RNA polymerase II (RNA pol II)²⁵, the carboxy-terminal domain (CTD) of which also has a prion-like sequence (Fig. S5)….