The SH3 domain of Fyn kinase interacts with and induces liquid–liquid phase separation of the low-complexity domain of hnRNPA2

Liquid–liquid phase separation of proteins and nucleic acids into membraneless organelles (MLOs) spatially organizes cellular components and reactions. The RNA-binding protein heterogeneous nuclear ribonucleoprotein A2 (hnRNPA2) carries mRNA targets in MLOs called transport granules in neurons and oligodendrocytes. At sites of local translation, hnRNPA2 is phosphorylated by the tyrosine protein kinase Fyn, releasing the mRNA for translation. Fyn recognizes targets through its SH3 domain (Fyn-SH3). However, hnRNPA2 lacks canonical SH3-binding sequences, raising the question of how Fyn-SH3 binds hnRNPA2 in phase-separated transport granules. Here, we characterize the structural details of the interaction of the hnRNPA2 low-complexity domain (LC) with Fyn-SH3 and the effect of Fyn-SH3 on hnRNPA2 phase separation. We combined in vitro microscopy and solution NMR spectroscopy to evaluate assembly of hnRNPA2 and Fyn-SH3 into in vitro phase-separated granules and probe the structural details of their interaction. We observed that Fyn-SH3 induces hnRNPA2 LC phase separation and that Fyn-SH3 is incorporated into in vitro hnRNPA2 LC granules. Moreover, we identified hnRNPA2 LC interaction sites on the surface of Fyn-SH3. Our data offer a structural view of how hnRNPA2 LC may interact with Fyn. To our knowledge, our study provides the first example of a single globular domain inducing phase separation of a disordered MLO scaffold protein.

Liquid-liquid phase separation (LLPS) 5 is a proposed mechanism for the formation of membraneless organelles (MLOs) (1)(2)(3). LLPS of MLOs can serve to spatially organize cellular components and has been proposed to speed enzymatic reactions as a result of concentrating reactants (4) or by providing a unique chemical environment (5)(6)(7). LLPS is an equilibrium process where molecules assemble into two or more distinct phases that differ in concentrations and potentially in structures and contacts (6). Disparate MLOs from nucleoli and nuclear speckles to stress granules and processing bodies carry out different cellular functions and are composed of a variety of cellular components (8,9). RNA granules represent a class of MLOs that are enriched in RNA and RNA-binding proteins.
Here we focus on RNA transport granules containing the heterogeneous nuclear ribonucleoprotein hnRNPA2 (10).
hnRNPA2 is an RNA-binding protein involved in pre-mRNA splicing, translation control, and mRNA transport (10). The domain structure of hnRNPA2 includes two N-terminal RNA recognition motifs that together bind a consensus RNA sequence known as the A2 response element (11), followed by an intrinsically disordered domain of low amino acid sequence complexity (low-complexity domain (LC)) (12). This LC has a subregion that is classified as a prion-like domain with amino acid sequence composition resembling aggregation-prone yeast prions (i.e. enriched in glutamine, asparagine, serine, and tyrosine) (13). Many prion-like domain-containing proteins including hnRNPA2 have been implicated in the formation of cellular inclusions seen in neurodegenerative diseases (14,15). Importantly, the LC of hnRNPA2 is necessary and sufficient for LLPS and aggregation (12). In its physiological role in the nervous system, hnRNPA2 forms mRNA transport granules to bring mRNAs, including myelin basic protein mRNA and activity regulated cytoskeleton-associated protein (Arc) mRNA, to sites of local translation in dendrites and oligodendroglial processes (16). When the granule reaches sites of local translation, the Src-family tyrosine kinase Fyn phosphorylates hnRNPA2, which releases cargo mRNA molecules for translation (17).
This work was supported in part by NIGMS, National Institutes of Health, Grants R01GM118530 (to N. L. F.) and P20GM104937. The authors declare that they have no conflicts of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This article was selected as one of our Editors' Picks. This article contains Figs. S1-S3. The NMR spectroscopy data reported in this paper have been submitted to the Biological Magnetic Resonance Data Bank (BMRB) under BMRB accession number 27649. 1 Both authors contributed equally to this work. 2

EDITORS' PICK
Fyn kinase is composed of several different Src homology domains in addition to the catalytic domain (16). The SH3 domain mediates protein-protein interactions with target proteins and putatively binds peptides containing a PXXP motif (18), whereas the SH2 domain binds primarily to phosphorylated tyrosines. Initially, Fyn is in a closed confirmation, but when it is phosphorylated and activated, the protein adopts an open conformation, making the SH2 and SH3 available to downstream targets (19). Fyn activation triggers phosphorylation of hnRNPA2 and other granule components, including hnRNPF, resulting in the dissociation of the granule and removal of translational repression (17). Whereas the functional interaction between hnRNPA2 and Fyn is well-understood, the physical interaction between Fyn kinase and hnRNPA2 is still unclear. Indeed, hnRNPA2 lacks a PXXP motif, the canonical SH3-binding motif. Therefore, a clearer picture of the structural details and interaction between hnRNPA2 LC and the SH3 domain of Fyn kinase is necessary. We hypothesized that the SH3 domain of Fyn interacts with disperse and phase-separated forms of hnRNPA2 LC. Here, we used NMR spectroscopy and microscopy to probe hnRNPA2-Fyn-SH3 proteinprotein interactions and the effect of hnRNPA2 diseaseassociated mutations on their interaction and hnRNPA2 assembly.

Fyn-SH3 interacts with hnRNPA2 LC
We performed titration experiments using NMR to determine which regions of hnRNPA2 LC mediate interactions with Fyn-SH3. Initially, we attempted to quantify the effect of the addition of unlabeled (NMR-silent) hnRNPA2 LC WT on the two-dimensional "fingerprint" 1 H-15 N NMR correlation (HSQC) spectrum of Fyn-SH3 incorporating 15 Fig. 1A). Further increase in hnRNPA2 LC concentration decreased the NMR intensity significantly (Fig. 1B), and NMR signal was lost entirely with 1:2.5 15 N-labeled Fyn-SH3 to hnRNPA2 LC. Correspondingly, when we titrated 15 N-labeled hnRNPA2 LC WT with natural abundance Fyn-SH3, we saw a decrease in intensity at 1:0.5 hnRNPA2 LC-Fyn-SH3 (Fig. 1C), and signal was lost entirely in at 1:1. Importantly, hnRNPA2 LC alone does not phase-separate at these solution conditions (in the absence of salt) (12); therefore, these data suggest that Fyn-SH3 induces phase separation of hnRNPA2 LC. In an attempt to reduce phase separation of hnRNPA2 LC/Fyn-SH3 to reveal

Fyn-SH3 induces hnRNPA2 LC phase separation
the specific binding sites on Fyn-SH3, we used a highly soluble form of hnRNPA2 LC with an N-terminal maltose-binding protein solubility tag, MBP-hnRNPA2 LC (12). This tagged form of hnRNPA2 LC does not phase-separate in buffers with conductivities in the physiological range (here, we use 20 mM NaP i , 50 mM NaCl, pH 6.1). Surprisingly, we still observed a loss of signal intensity upon mixing the peptides at a ratio of 1:0.5 [ 15 N]Fyn-SH3 to MBP-hnRNPA2 LC, indicative of the formation of a large, NMR-invisible species, and the solution turned cloudy (Fig. 1D). Quantification at ratios beyond 1:0.5 was impossible as no Fyn-SH3 signal remained (consistent with Fyn-SH3 partitioning into the protein-dense phase), and chemical shifts were small at 1:0.5 (Fig. 1, E and F). Significantly, these experiments with tagged hnRNPA2 LC were done in 50 mM NaCl, indicating that the interaction between monomeric Fyn-SH3 and monomeric hnRNPA2 LC is not dependent on weak electrostatic interactions artificially increased in the 0 mM salt conditions used for untagged hnRNPA2 LC. Therefore, Fyn-SH3 robustly induces phase separation of hnRNPA2 LC. As such, we decided to investigate the effect of Fyn-SH3 on hnRNPA2 LC phase separation further.

Fyn-SH3 increases the turbidity of hnRNPA2 LC WT
Our observation of reduced NMR signal intensity and enhanced optical turbidity for 15 N-labeled hnRNPA2 LC to Fyn-SH3 titrations upon mixing intrigued us. To quantify phase separation of hnRNPA2 LC as a function of Fyn-SH3 concentration, we measured sample turbidity ( Fig. 2A). As the concentration of Fyn-SH3 increases, the turbidity of hnRNPA2 LC WT increases up to equimolar concentrations of Fyn-SH3. In addition, we measured the turbidity over the course of 6 h (Fig. 2B). The turbidity increased for each titration over a short time, consistent with sample mixing, and then decreased until the turbidity reached baseline levels, consistent with droplet coarsening to larger sizes than the wavelength of light and droplet sedimentation and adherence to walls of the well. At 1:1 and 1:2, similar time courses and maximum turbidity values are observed. At 1:10 only, turbidity does not return to baseline by the 6-h time point (Fig. 2B).
To analyze the effects of the disease mutations of hnRNPA2 on the interaction with Fyn-SH3, we performed turbidity experiments with hnRNPA2 LC disease mutants (D290V or

Fyn-SH3 induces hnRNPA2 LC phase separation
P298L) with increasing concentrations of Fyn-SH3. As the concentration of Fyn-SH3 increases, the turbidity of both D290V and P298L samples increase (Fig. S1, A and B). These results were similar to the results obtained in the hnRNPA2 LC WT experiment ( Fig. 2A).

Fyn-SH3 incorporates into hnRNPA2 LC droplets
As loss of NMR signal and turbidity could arise from aggregation or LLPS, we used differential interference contrast (DIC) microscopy to show that the turbidity observed for Fyn-SH3 and hnRNPA2 LC samples arises from LLPS. Furthermore, we did this experiment along with fluorescence microscopy to better understand the role that Fyn-SH3 plays in LLPS. To test the hypothesis that Fyn-SH3 co-phase separates with hnRNPA2 LC, we placed an Alexa Fluor label on Fyn-SH3 away from the proline binding site that was identified previously (20) (Fig. 2C) by mutating a surface-exposed serine residue (Ser-25) to cysteine. In 0 mM NaCl buffer, DIC microscopy images show that the addition of Fyn-SH3 induces the formation of hnRNPA2 LC WT liquid droplets at a concentration and solution conditions where hnRNPA2 LC alone does not phase-separate (Fig. 2D, second column, bottom panel) (12). Additionally, fluorescence microscopy data demonstrate that Fyn-SH3 incorporates into the droplets (Fig. 2D, second column, top panel). Furthermore, Fyn-SH3 induces droplet formation of the hnRNPA2 LC D290V mutant (Fig. 2E).
Now that we could conclude that Fyn-SH3 induces hnRNPA2 LC phase separation at 0 mM NaCl, we performed experiments at increasing salt concentrations to determine whether Fyn-SH3 is incorporated into LLPS droplets of hnRNPA2 LC at physiological salt conditions. Regardless of the concentration of NaCl, incorporation of Fyn-SH3 in the hnRNPA2 LC WT droplets was still observed, and the morphology of the droplets did not change (Fig. 2D). These results demonstrate that the interaction between Fyn-SH3 and hnRNPA2 LC is not an artifact of the low-salt conditions used for NMR experiments.

Fyn-SH3 may reduce the aggregation of hnRNPA2 diseaseassociated mutants
We previously showed that hnRNPA2 mutants D290V and P298L lead to conversion of liquid, phase-separated hnRNPA2 LC droplets to protein aggregates (12). To determine whether Fyn-SH3 binding can disrupt or encourage conversion of hnRNPA2 from a liquid form to an aggregated form associated with disease, we incubated hnRNPA2 LC and its disease-associated variants D290V and P298L with and without Fyn-SH3. We mixed Fyn-SH3 with hnRNPA2 LC WT, D290V, or P298L and then visualized the morphology of the phase-separated assemblies using DIC microscopy at two time points (0 and 3 h). The addition of Fyn-SH3 to hnRNPA2 LC WT did not change the morphology of the droplets as the droplets predominantly maintained a round shape (Fig. 3A, top). In comparison, after 3 h, the hnRNPA2 LC disease mutant proteins in the absence of Fyn-SH3 appear irregularly shaped, consistent with the formation of irreversible aggregation (12,21) (Fig. 3A, middle and  bottom). Aggregation was most prominent in the P298L disease mutant (Fig. 3A, bottom). When Fyn-SH3 was added to the hnRNPA2 LC P298L mutant, after 3 h, aggregation into irregular structures was not apparent, and round droplets were present instead (Fig. 3A, bottom). For the D290V mutant, both aggregation and droplets are observed, suggesting that the effect is not as prominent as for the P298L mutant (Fig. 3A,  middle). These results suggest that Fyn-SH3 delays hnRNPA2 LC P298L aggregation. We repeated this same experiment at the same conditions but with different protein stocks and observed the same change in P298L aggregation (Fig. S2A). Furthermore, Fyn-SH3 still incorporates into hnRNPA2 LC P298L droplets despite the propensity to aggregate over time (Fig.  S2B). Unfortunately, we were not able to quantify the effect by other approaches; turbidity is enhanced by both LLPS and aggregation, and amyloid-labeling dyes also fluoresce in liquid droplets (22). To test the effects of salt on the aggregation of hnRNPA2 disease mutants, we ran similar experiments to the DIC microscopy experiments in Fig. 2A but used buffer with physiological 150 mM NaCl in place of 0 mM NaCl. Unlike at low salt, hnRNPA2 LC aggregation still occurs at 150 mM NaCl even in the presence of Fyn-SH3 (Fig. 3B). After 3 h, the WT, D290V, and P298L samples were still aggregated, with or without Fyn-  SH3. Therefore, Fyn-SH3 is not able to prevent hnRNPA2 LC aggregation in all conditions tested.

NMR experiments mixing hnRNPA2 LC subpeptides with Fyn-SH3 identify potential binding sites on Fyn-SH3
The purpose of our earlier two-dimensional NMR experiments was to characterize the structural details of the interaction between hnRNPA2 LC and Fyn-SH3. This proved difficult as hnRNPA2 LC phase-separated upon the addition of Fyn-SH3, sending the interacting proteins into viscous phase-separated droplets that are difficult to detect by solution NMR (12). To use NMR to identify the site of interac-tion, we hypothesized that smaller pieces of hnRNPA2 LC will discourage LLPS but retain the ability to bind Fyn-SH3. We started with a subpeptide of hnRNPA2 containing residues 266 -341, which contains the prion-like domain (residues 266 -303; Fig. 4A). The prion-like domain in hnRNPA2 has been indicated in past studies to induce aggregation in neurodegenerative disease (15), and this domain contains the residues that are mutated in hnRNPA2-related degenerative diseases (Pro-298 and Asp-290). Furthermore, this subpeptide contains four of the nine total proline residues in the LC, which may act as binding sites for SH3 domains. As we had hoped, mixtures of Fyn-SH3 and hnRNPA2 266 -341 in

Fyn-SH3 induces hnRNPA2 LC phase separation
NMR buffer showed no phase separation at the concentrations tested.
We obtained NMR spectra at 1:1, 1:1.5, and 1:2 titrations of 15 N-labeled Fyn-SH3 to hnRNPA2 266 -341 (20 M Fyn-SH3) ( Fig. 4B and Fig. S3A). As the concentration of the hnRNPA2 subpeptide was increased, we observed small but increasing chemical shift perturbations of Fyn-SH3 resonances ( Fig. 4B and Fig. S3A), consistent with our microscopy experiments that suggest an interaction between Fyn-SH3 and hnRNPA2 LC. 1 H and 15 N chemical shift perturbations increased as a function of increasing concentration of the hnRNPA2 subpeptide (Fig. 4, C  and D). These shifts are small, consistent with weak interactions often present in phase-separating protein partners (23). When the residues of Fyn-SH3 that experienced shifts are mapped on the structure of Fyn-SH3, they occur at surprising locations. We initially hypothesized that hnRNPA2 would interact with Fyn-SH3 through the n-Src loop, as shown for a proline-rich segment solved by NMR (20). However, the chemical shifts mainly occur in the RT loop, the ␤4 strand, and the distal loop of Fyn-SH3 (Figs. 4E and 5). This SH3 surface overlaps with the binding surface region identified in multiple studies to bind to proline-rich motifs (24) and has been referred to as the "specificity zone." The specificity zone is composed of the ␤3 and ␤4 strand, flanked by the far end of the n-Src loop and part of the RT loop (24) (Fig. 5). The most significant chemical shift perturbations occur in a part of this region, suggesting that Fyn-SH3 interacts with hnRNPA2 through this specificity zone. Specifically, Thr-53 and Gly-54 on the ␤4 strand and Leu-24 in the RT loop show the highest shifts (Fig. 4E). Unlike NMR experiments on Fyn-SH3 with the entire hnRNPA2 LC, these mixtures are stable and not prone to LLPS, as evidenced by intensity ratios around 1 (Fig. 4F). However, despite no visual evidence for phase separation, at 1:2, the intensity ratio decreases to ϳ0.7 (Fig. 4F), which may occur due to formation of higher-order assemblies.
We also performed the inverse experiment with 15 N hnRNPA2 266 -341 and natural abundance Fyn-SH3 to attempt to identify Fyn-SH3-binding sites on hnRNPA2. We observed small chemical shift perturbations (Fig. 4 (G-I) and Fig. S3B) in hnRNPA2 266 -341 across the whole subpeptide. In the region of 310 -320, we observed a cluster of larger 15 N shifts, suggesting increased Fyn-SH3 binding with this region (Fig. 4I). Interestingly, when Fyn-SH3 was added at 5 molar eq to hnRNPA2 266 -341, we observed slight turbidity in the NMR sample and corresponding signal loss, leaving about 60% of the 1:1 signal (Fig. 4J), suggesting that Fyn-SH3 can still induce phase separation of this small truncation of hnRNPA2.
In an attempt to probe the interaction of Fyn-SH3 with the remainder of hnRNPA2 LC, we generated a subpeptide encompassing hnRNPA2 190 -303. However, in our standard conditions, this peptide immediately precipitated. Therefore, we performed NMR experiments at a lower pH (4.5), where hnRNPA2 LC is less assembly-prone. Here, the hnRNPA2 190 -303 phaseseparated upon the addition of equimolar Fyn-SH3, so we were unable to observe any significant chemical shift perturbations (Fig. S3, C-E). Hence, it appears that protein-protein interaction occurs, but LLPS precludes straightforward evaluation of the site-specific details of the interaction by NMR. This subpeptide does have more proline residues (7 total) than the 266 -341 subpeptide; hence, this region may be a major contributor to Fyn-SH3 interaction with hnRNPA2.

Discussion
These NMR experiments identified hnRNPA2 LC interactions with the ␤4 strand, RT loop, and distal loop of Fyn-SH3. Other groups have characterized similar interactions with the SH3 domain and other proline-rich proteins and refer to this surface region as the specificity zone (24). We saw significant chemical shift perturbations of the residues in the ␤4 strand and the RT loop that are conserved across many SH3 domains (25), suggesting that these residues are important for SH3 function. Residues Asp-23 and Leu-24 in the RT loop of Fyn-SH3 showed large chemical shift perturbations in the presence of hnRNPA2 266 -341. Chemical shift perturbations were observed in these same conserved residues of Noxa1-SH3 upon binding of Noxo1 proline-rich region (25). Another study claims that the binding of Fyn-SH3 to proline-rich proteins is mediated by weak interactions between the hydroxyl groups of the tyrosines in the n-Src loop region of SH3 and the backbone of the proline residues (26). Here, we see chemical shift perturbation at Asp-41 of the n-Src loop and the spatially proximate Tyr-14, suggesting that there may be a second interface for hnRNPA2 LC binding. Additionally, the negatively charged binding surfaces of Fyn-SH3 have been implicated in other

Fyn-SH3 induces hnRNPA2 LC phase separation
interactions between proline-rich proteins and Fyn-SH3 (27). hnRNPA2 is positively charged, with a pI of 9.30, so electrostatic interactions may contribute to the interaction and phase separation of the mixture. However, the interaction is not due to weak electrostatics contact artifacts from low-conductivity buffers as we still observed the formation of a large, NMRinvisible species when hnRNPA2 LC with a solubility tag was mixed with Fyn-SH3 in the presence of salt. With a truncated hnRNPA2 LC(266 -341), we were able to observe some chemical shift perturbations upon Fyn-SH3 binding. The largest shifts were localized to residues 314 -325, a region that contains a proline-tyrosine nuclear localization signal (PY-NLS) (residues 318 -324) that binds the importin transportin-1 (karyopherin ␤-2). Intriguingly, transportin-1 binding to the PY-NLS of FUS, a related hnRNP with a prion-like domain, disrupts FUS phase separation (28 -30). Therefore, PY-NLSs may serve as prominent binding sites regulating phase separation. However, our data do not definitively identify the PY-NLS as the highest affinity Fyn-SH3-binding site in hnRNPA2 LC because the addition of Fyn-SH3 to both the intact LC and an N-terminal fragment (residues 190 -303) resulted in phase separation at lower concentrations than with the 266 -341 truncation, possibly due to higher affinity binding and precluding determination of chemical shift perturbations.
Understanding how Fyn-SH3 binds to target proteins and identifying the binding interfaces between Fyn-SH3 and those targets, like hnRNPA2, is important for the potential development of therapeutics for degenerative diseases. Past studies have shown that the Nck-SH3 domains induce phase separation via multivalent interactions (31)(32)(33). Phase separation of Nck and N-WASP depends on the number of SH3 domains; one SH3 domain is not sufficient to induce phase separation, and Nck normally contains three SH3 domains (33). In contrast, our study shows that the single Fyn-SH3 domain is sufficient to induce phase separation of hnRNPA2 LC. The Fyn-SH3 domain itself may be a multivalent binding platform for hnRNPA2 LC; our NMR data indicate multiple sites on Fyn-SH3 that can bind to hnRNPA2 LC. Interestingly, the presence of multiple interaction sites implies a potential mechanism for how this single, monomeric SH3 domain induces phase separation of hnRNPA2 LC. This model has multiple hnRNPA2 LC molecules binding to different sites on Fyn-SH3 simultaneously, enhancing hnRNPA2 LC intermolecular interactions and initiating network-spanning contacts. Alternatively, multiple hnRNPA2 LC molecules could also bind to a single site/ region on Fyn-SH3 (e.g. a site that can accommodate several proline residues simultaneously).
The induction of LLPS of hnRNPA2 LC is not likely to be the physiological role of Fyn-SH3, as hnRNPA2 would already be phase-separated within an RNA granule when it then interacts with membrane-tethered Fyn-kinase present at the oligodendroglial site of local translation. We suggest that multivalent Fyn-SH3-hnRNPA2 contacts aid Fyn incorporation into hnRNPA2 granules to spatially control the interaction of the kinase domain of Fyn and hnRNPA2, leading to hnRNPA2 phosphorylation and granule disassembly at sites where local translation is needed. Partitioning of Fyn into hnRNPA2 granules may also speed the phosphorylation reaction due to increased local concentrations of hnRNPA2, Fyn-kinase, and ATP (6). Tight spatial and temporal control of the dissociation of mRNA transport granules is critical to ensure that the mRNAs are translated only when and where the protein is needed. As phosphorylation also reduces phase separation of another granule protein, FUS (34), it is possible that other kinases co-phaseseparate with their target granule proteins and dissociate the phase-separated granule from inside the phase-separated state. More work is needed to understand the molecular details underlying granule dissociation and whether incorporation of kinases into the granule is a common mechanism that can be targeted to prevent aggregation of mutant protein from the phase-separated state.

Bacterial culture and isotope labeling
Proteins were expressed in either Luria broth or M9 minimal medium containing 15 N ammonium chloride as the sole nitrogen source for 15 N labeling. BL21 Star DE3 E. coli cells were grown at 37°C with 50 mg/liter kanamycin or 100 mg/liter ampicillin as appropriate until the A 600 reached 0.6 -1, and then the cultures were induced with 1 mM isopropyl-␤-D-thiogalactoside at 37°C for 4 h. Cells were collected by centrifugation. The cell pellet was resuspended in 20 mM sodium phosphate (NaP i ), pH 7.4, 300 mM NaCl, 10 mM imidazole for hnRNPA2 proteins and 50 mM Tris, pH 8.0, 500 mM NaCl, 10 mM imidazole for Fyn-SH3 protein. Once cleared by resuspension, cells were lysed using an Emulsiflex C3 homogenizer (Avestin). The cell lysates were cleared by centrifugation at 20,000 ϫ g for 60 min at 4°C.

Protein purification
hnRNPA2-hnRNPA2 constructs were purified as described (12). Briefly, after expression and cell lysis, inclusion bodies were resuspended in 8 M urea, 20 mM NaP i , pH 7.4, 300 mM NaCl, 10 mM imidazole. After centrifugation at 20,000 ϫ g for 1 h, the supernatant was collected, filtered, and loaded onto a HisTrap HP 5-ml column. Protein was eluted using a gradient from 10 to 300 mM imidazole over 5 column volumes. Pooled fractions containing the desired protein were buffer-exchanged into 8 M urea, pH 5.5, MES and concentrated to 1 ml. Urea was diluted out by adding 20 ml of 20 mM MES, pH 5.5 (pH-adjusted with BisTris). The protein was incubated with TEV protease Fyn-SH3 induces hnRNPA2 LC phase separation overnight at room temperature. To resolubilize precipitated protein, 8 M urea, 20 mM NaP i , 300 mM NaCl, 10 mM imidazole was added to the protein. Protein was filtered and loaded onto a HisTrap HP 5-ml column to remove the TEV protease, cleaved His-tag, and uncleaved protein. Flow-through fractions containing the purified product were concentrated and buffer-exchanged into 8 M urea, pH 5.5, MES and flash-frozen.
Fyn-SH3-As Fyn-SH3 is soluble, the cleared lysate was filtered using a 0.2-m filter and loaded onto a HisTrap HP 5-ml column. Protein was eluted using a gradient from 10 to 300 mM imidazole over 5 column volumes. The fractions containing Fyn-SH3 were pooled and incubated with TEV protease overnight at room temperature. To remove the His-tag, the protein was loaded onto a Superdex 75 column equilibrated in 50 mM Tris, pH 8.0, 150 mM NaCl, 2 mM EDTA. The fractions containing the purified Fyn-SH3 protein were concentrated and bufferexchanged into pH 5.5 MES and flash-frozen for storage. For labeling experiments, Fyn-SH3 S25C was expressed and purified as described above, with the addition of 1 mM DTT to all buffers.

Alexa Fluor labeling
Fyn-SH3 S25C was labeled using Alexa Fluor TM 488 C5 maleimide (Life Technologies), which reacts with free thiol groups. After Fyn-SH3 S25C was purified, 2 ml Zeba TM spin desalting columns (Thermo Fisher Scientific) were used to remove the DTT and buffer-exchange the protein into pH 7.5, 50 mM Tris buffer. After desalting, the protein was incubated with excess Alexa Fluor C5 maleimide at room temperature for 2 h (covered in aluminum foil), according to the manufacturer's procedures. This reaction mixture was then placed into an equilibrated desalting column to remove excess label. The protein was buffer-exchanged into 20 mM MES, pH 5.5 (pH-adjusted with BisTris) and flash-frozen. The concentration of the protein was quantified by Nanodrop 2000c using the Alexa-488 setting, which is programmed to account for Alexa Fluor TM 488 labeling.

Solution NMR experiments
NMR titration experiments were acquired using a Bruker Avance III HD NMR spectrometer operating at 850 MHz, 1 H Larmor frequency equipped with HCN TCI z-gradient cryoprobe. The temperature was set at 25°C. Experimental sweep widths, acquisition times, and the number of transients were optimized for the necessary resolution, experiment time, and signal/noise ratio.

Two-dimensional NMR titration experiments
hnRNPA2 LC was diluted to the appropriate concentration for the experiment from 8 M urea, 20 mM MES, pH 5.5 (pHadjusted with BisTris), ensuring that the urea concentration was consistently 150 mM. Fyn-SH3 was diluted to the appropriate concentration from 20 mM MES, pH 5.5. Each sample contained 10% D 2 O. 1:0 samples (regardless of 15 N-labeled species) contained 20 M protein in the absence of unlabeled titrant protein; protein concentration ratios for the titrant protein or peptide are calculated from that concentration (e.g. 1:2 is 20 M 15 N-labeled protein plus 40 M unlabeled titrant protein). This applies to all data except for Fig. 1A, as labeled.
Chemical shift perturbations were calculated by subtracting the chemical shifts of the reference peaks (1:0 titrations) from the chemical shifts acquired in the titration experiments. Intensity ratios were calculated from the height of each peak, and the uncertainty was estimated from the signal/noise ratio of each spectrum. Data were plotted using Prism (GraphPad Software, Inc.

Fluorescence and DIC microscopy
DIC images and fluorescence microscopy of Fyn-SH3, Fyn-SH3 Alexa Fluor, and hnRNPA2 LC WT, P298L, and D290V proteins were collected at pH 5.5, 20 mM MES buffer conditions. NaCl was omitted unless specified. Experimental samples were prepared by diluting proteins into experimental conditions with a final volume of 50 l. Differential interference contrast on an Axiovert 200M microscope (Zeiss) was used to observe hnRNPA2 LC droplets and aggregates. Fluorescence microscopy on an Axiovert 200M microscope (Zeiss) was used to observe Alexa Fluor-labeled Fyn-SH3 incorporation in hnRNPA2 LC droplets. Samples were spotted on a coverslip for imaging. Images were processed with ImageJ (National Institutes of Health).

A600 turbidity experiments
Turbidity experiments were performed in 96-well clear plates (Costar) with 65-l samples sealed with clear optical adhesive film (MicroAmp, Thermo Fisher Scientific). Cytation 3 Cell Imaging Multi-Mode Reader (BioTek) was used to record the turbidity by measuring the optical density at 600 nm at 5-min time intervals for up to 12 h. The samples were premixed for 5 s before the start of the recordings, and readings were made from the bottom of the plate. Each sample was placed in three wells to acquire average and S.D. Sample turbidity values were subtracted from the buffer control value.
Turbidity experiments of hnRNPA2 LC and Fyn-SH3 were conducted by diluting hnRNPA2 LC WT to 20 M (maintaining a final concentration of 150 mM urea). Both proteins were diluted into 20 mM MES, pH 5.5. The amount of Fyn-SH3 con-Fyn-SH3 induces hnRNPA2 LC phase separation centrations was added in appropriate ratios to hnRNPA2 LC concentrations. Statistical analysis was performed with Microsoft Excel 360, and data were graphed using Prism (GraphPad Software).

Aggregation by microscopy assay
To study the effects of Fyn-SH3 on the progression of hnRNPA2 LC WT aggregate formation, Fyn-SH3 was mixed with either hnRNPA2 LC WT, D290V, or P298L. Each protein was diluted into 50-l samples at 20 mM MES, pH 5.5. NaCl was omitted unless otherwise indicated. hnRNPA2 LC variants were diluted to 20 M, which diluted the 8 M urea from the storage solution to a concentration of 150 mM residual urea for the final experimental conditions. Fyn-SH3 was diluted to 20 M concentrations as well and placed into each tube as appropriate. Each sample contained either a 1:0 ratio of the hnRNPA2 LC variant to Fyn-SH3 or a 1:1 ratio of the hnRNPA2 LC variant to Fyn-SH3. 20 l of each sample was pipetted to a coverslip and observed using differential interference contrast on an Axiovert 200M microscope (Zeiss). This was done immediately after the sample was prepared. Two replicates were prepared, one for observation at 0 h and the other for observation at 3 h. Images were processed and analyzed using ImageJ (National Institutes of Health).