Cancer mutations of the tumor suppressor SPOP disrupt the formation of active, phase-separated compartments

Summary

Mutations in the tumor suppressor SPOP (Speckle-type POZ protein) cause prostate, breast and other solid tumors. SPOP is a substrate adaptor of the cullin3-RING ubiquitin ligase and localizes to nuclear speckles. Although cancer-associated mutations in SPOP interfere with substrate recruitment to the ligase, mechanisms underlying assembly of SPOP with its substrates in liquid nuclear bodies, and effects of SPOP mutations on assembly are poorly understood. Here we show that substrates trigger phase separation of SPOP in vitro and co-localization in membraneless organelles in cells. Enzymatic activity correlates with cellular co-localization and in vitro mesoscale assembly formation. Diseaseassociated SPOP mutations that lead to the accumulation of proto-oncogenic proteins interfere with phase separation and co-localization in membraneless organelles, suggesting that substrate-directed phase separation of this E3 ligase underlies the regulation of ubiquitin-dependent proteostasis.

eTOC Blurb

Mutations in the tumor suppressor SPOP are known to cause solid tumors. Bouchard and Otero et al. show that SPOP phase separates with substrates in vitro; the same interactions mediate colocalization in membraneless organelles in cells. SPOP cancer mutations disrupt liquid-liquid phase separation, which correlates with loss of function.

Introduction

Cancer-driving mutations in enzymes typically reduce their activity or introduce an aberrant activity. However, mislocalization of mutant enzymes is another form of loss-of-function. Here, we explore the mechanism underlying SPOP loss-of-function cancer mutations, and show that mutants fail to colocalize with substrates, via disruption of phase separation.

SPOP is frequently mutated in solid tumors, particularly in prostate cancer (Cerami et al., 2012; Kim et al., 2011; Kim et al., 2013; Lawrence et al., 2014; Le Gallo et al., 2012). Its gene product SPOP (Speckle-type POZ protein) is a substrate adaptor of a cullin3-RING ubiquitin ligase (CRL3) that recruits substrates to the ligase for ubiquitination and subsequent proteasomal degradation (HernandezMunoz et al., 2005; Kent et al., 2006; Kwon et al., 2006; Li et al., 2014). Cancer-associated mutations in SPOP are typically found in the substrate binding MATH domain (Fig. 1A); accordingly, they interfere with substrate binding and ubiquitination, which increases substrate levels (Gan et al., 2015; Geng et al., 2017). SPOP substrates include the death-domain associated protein (DAXX), androgen receptor (AR), and other important signaling cascade effectors, epigenetic modifiers, and hormone signaling effectors (Gan et al., 2015; Gao et al., 2015; Geng et al., 2013; Geng et al., 2017; Janouskova et al., 2017; Li et al., 2014; Theurillat et al., 2014; Zhang et al., 2018; Zhang et al., 2009; Zhuang et al., 2009). Increased levels of proto-oncogenic substrates as a consequence of SPOP mutations can oncogenically transform sensitive cell types (An et al., 2014; Dai et al., 2017a; Dai et al., 2017b; Gan et al., 2015; Geng et al., 2013; Geng et al., 2017; Geng et al., 2014; Janouskova et al., 2017; Theurillat et al., 2014).

Figure 1. SPOP and DAXX co-localize in liquid organelles, and colocalization is disrupted by SPOP cancer mutations.

(A) Sequence and cartoon schematics for SPOP (left) and DAXX (right) constructs used in this study. Sequence cartoons at the top represent domain architecture. SPOP contains a substrate binding MATH domain, and two dimerization domains, BTB and BACK. DAXX contains a DAXX helical bundle (DHB) domain, a helical region, and a C-terminal disordered domain (Escobar-Cabrera et al., 2010). Predicted SPOP-binding motifs (based on the consensus sequence motif, nonpolar-polar-S-S/T-S/T (Zhuang et al., 2009)) are shown in orange, with stronger binding sites shaded darker. Lys residues available for ubiquitination are shown as K. (Below) Cartoon schematics represent self-association and substrate binding behavior, with mutated interfaces in SPOP shown curved instead of straight to indicate the inability to self-associate or bind substrate. WT SPOP forms higher-order oligomers of different sizes (Marzahn et al., 2016); we show hexamers as an example. Cancer mutations W131G and F133V in the MATH domain (green) reduce substrate binding. Mutation of the dimerization interface of the BACK domain (blue, mutation Y353E (van Geersdaele et al., 2013)) results in SPOP mutBACK dimers (Marzahn et al., 2016); the addition of mutations of the BTB interface (red, L186D, L190D, L193D, I217K, (Zhuang et al., 2009)) results in SPOP mutBTB/BACK monomers (Marzahn et al., 2016). SPOP constructs for expression in mammalian cells encode the full-length protein; those for expression in bacteria 28–359. DAXX mammalian expression constructs encode full-length protein unless labeled cDAXX which comprises residues 495–740, as the bacterial expression constructs. H-cDAXX harbors a His₆-tag. In cDAXX-0sb, the major SB motifs are mutated. (For details see Fig. 3 and Table S1.)

(B)SPOP and DAXX localize to SPOP/DAXX bodies. HeLa cells were transfected with the indicated GFP-DAXX and/or SPOP-mCherry and analyzed by confocal microscopy. mCherry (red) and GFP fluorescence (green) were observed for SPOP and DAXX, while SC-35 and PML (both magenta) were used as markers for nuclear speckles and PML bodies, respectively, and detected by IF. DAPI (blue) marks the nucleus. (See also Fig. S1A-D.)

(C) PML bodies behave like liquid droplets. HeLa cells were transfected with GFP-DAXX and GFP monitored in live cells. Snapshots at the indicated time points show a PML body fusion event in the area boxed on the left. (See also Video S1.)

(D)SPOP and DAXX form nuclear bodies with liquid properties. HeLa cells were transfected with GFP-DAXX and SPOP-mCherry and analyzed as in (C). (See also Video S2.)

(E)SPOP cancer mutants fail to localize to SPOP/DAXX bodies. HeLa cells were transfected with GFP-DAXX and either WT V5-SPOP or mutants F133V or W131G, and analyzed as in (B). (See also Fig. S1G.)

Understanding how SPOP targets its substrates is important for the development of therapeutic interventions for SPOP-driven cancers. However, unlike many prototypic cullin-RING ligases, which rely on post-translational modifications for substrate targeting (Petroski and Deshaies, 2005), mechanisms that could regulate SPOP/substrate interactions are largely elusive. SPOP typically localizes to nuclear speckles (Nagai et al., 1997) and has also been reported to localize to DNA damage loci, PML bodies and other substrate-containing bodies (Boysen et al., 2015; Gan et al., 2015; Kwon et al., 2006; Marzahn et al., 2016; Zhang et al., 2014). Substrate level dynamics in cells may therefore direct SPOP localization. However, how SPOP assembles with substrates and gets recruited to nuclear bodies, is not well understood.

The nuclear bodies with which SPOP associates are so-called membraneless organelles, i.e. mesoscale bodies that concentrate specific components within them, without being enclosed by a membrane. Evidence is mounting that membraneless organelles are formed through phase separation processes. Liquid droplet-like organelles result from liquid-liquid phase separation (LLPS) (Berry et al., 2015; Brangwynne et al., 2009; Nott et al., 2015), while solid assemblies may result from oligomerization/polymerization processes or maturation of liquid assemblies into solid ones (Boke et al., 2016; Cai et al., 2014; Kato et al., 2012). While mesoscale assemblies formed through different processes are on a spectrum between solid and dynamic assemblies and have different material properties (Halfmann, 2016), they all have the potential to co-localize enzymes with their substrates and therefore regulate activity in cells (Wu and Fuxreiter, 2016).

Here, we systematically examine the mechanism of SPOP/substrate co-localization. We have shown previously that SPOP self-associates into large higher-order oligomers through the synergistic function of two dimerization domains, the BTB and the BACK domains (Fig. 1A). This selfassociation is required for localization to membraneless organelles (Marzahn et al., 2016) but is insufficient to drive LLPS because it mediates the formation of assemblies whose size distribution continuously shifts to larger sizes with icreasing concentration, but never leads to the cooperative formation of mesoscale assemblies. Here, we show substrates act as the trigger for co-localization of SPOP and substrates in cells and phase separation with each other in vitro. CUL3 ubiquitin ligase activity is found in the resulting mesoscale assemblies. We find that cancer-associated mutations disrupt co-localization and liquid phase separation, which correlates with reduced protein ubiquitination. Our results suggest the possibility that substrate-mediated phase separation of this ubiquitin ligase is essential for concentrating the enzyme and its substrates in active liquid organelles in cells, and that such organelles may have functions essential to proteostasis.

Results

DAXX and SPOP co-localize in liquid nuclear organelles

Because many substrates contain multiple SPOP-binding (SB) motifs in intrinsically disordered regions (IDRs) (Pierce et al., 2016; Zhang et al., 2009; Zhuang et al., 2009) and can mediate multivalent interactions with SPOP, substrates are candidates for mediating phase separation with SPOP and regulating its subcellular localization and function. Increased substrate levels, or increased apparent levels via other signals, seems to mediate the redistribution of SPOP from nuclear speckles into different compartments. To examine roles of substrates in SPOP localization, we used transient expression of DAXX and SPOP in cultured HeLa cells. Increased DAXX levels caused by SPOP cancer mutations likely contribute to cancer pathogenesis by increasing angiogenic factors such as VEGFR2 (Sakaue et al., 2017). Furthermore, DAXX contains at least 8 predicted SB motifs in its IDRs (Fig. 1A), making it a likely candidate to regulate SPOP localization. SPOP-mCherry localized to nuclear bodies that stain for the typical nuclear speckle marker SC-35 (Fig. 1B, Fig. S1A), as previously reported (Marzahn et al., 2016; Nagai et al., 1997). Transiently expressed GFP-DAXX (and its intrinsically disordered C-terminal region, GFP-cDAXX) localized to small, spherical PML bodies (Fig. 1B, Fig. S1A, B) in agreement with previous reports (Kwon et al., 2006; Li et al., 2000; Weidtkamp-Peters et al., 2008). However, we found that transient co-expression of SPOP-mCherry and GFP-DAXX resulted in their colocalization in a different, largely spherical type of nuclear body distinct from nuclear speckles, PML bodies, nucleoli, and Cajal bodies (Fig. 1B, Fig. S1C, D); we will refer to them as SPOP/DAXX bodies from hereon. Different expression tags did not influence the colocalization of SPOP and DAXX (Fig. S1E). Importantly, co-expression of HA-SPOP and GFPDAXX in a PC-3 cancer cell line also resulted in colocalization (Fig. S1F).

DAXX and SPOP therefore both re-localized from their original subcellular location when expressed together. The SPOP/DAXX bodies possess liquid properties as evidenced by their ability to undergo fusion events within minutes (Fig. 1C, D, Supplemental Videos S1, S2). These properties place them into the category of liquid membraneless organelles.

SPOP cancer mutants disrupt co-localization

The typical prostate cancer mutations in SPOP, W131G and F133V in the substrate-binding MATH domain, disrupted SPOP and DAXX co-localization upon co-expression; V5-SPOP W131G and F133V localized to nuclear speckles, and GFP-DAXX localized to PML bodies, as they do when expressed alone (Fig. 1E, Fig. S1G). The SPOP cancer mutations thus disrupt re-localization of both proteins.

It is clear from these observations that SPOP and DAXX not only bind to each other, but binding shifts them to a different liquid organelle. We therefore hypothesized that SPOP and DAXX undergo LLPS with each other, a process in which a macromolecule (or a set of macromolecules) demixes from the solution and forms a separate, condensed liquid phase, often visible as liquid droplets

DAXX and SPOP undergo phase separation in vitro

To test this hypothesis, we purified SPOP²⁸⁻³⁵⁹ (the extreme termini are missing in in vitro purified protein to improve protein behavior, (Marzahn et al., 2016)) and the intrinsically disordered C-terminal region of DAXX, DAXX^495−740 (cDAXX from here on) (Fig. 1A) and studied their interaction in vitro. We have previously reported that SPOP self-associates into linear higher-order oligomers (Marzahn et al., 2016). In the presence of molecular crowders such as Ficoll-70, these oligomers are large enough to be observed by light microscopy (Fig. 2A, top, and Fig. S2A-C). At concentrations of above ~100 μM, H-cDAXX forms condensed droplets (Fig. S2A). However, the tendency of H-cDAXX to undergo phase separation is strongly enhanced in the presence of SPOP, and this was the case in the presence of both polymer and protein crowders (Fig. S2B). This observation suggests that weak multivalent interactions between SPOP and DAXX result in a sol-gel transition coupled to phase separation, as defined by Harmon et al (Harmon et al., 2017).

Figure 2. SPOP and DAXX undergo phase separation in vitro and in cells, which depends on SPOP oligomerization.

(A)SPOP and DAXX undergo phase separation in vitro. Fluorescence microscopy images of increasing concentrations of WT SPOP (green) alone, H-cDAXX (red) alone, and SPOP + H-cDAXX at a constant molar ratio of 1 SPOP : 5 DAXX. Camera settings are the same across rows. The panel boxed red is shown as separate green, red and DIC channels below. All samples in (A) and (C) contain 10% w/v ficoll 70, 500 nM ORG-SPOP and/or Rhodamine-H-cDAXX. Samples were in 25 mM Tris pH 7.6, 150 mM NaCl, and 1 mM T-CEP. (See also Fig. S2A-C.)

(B) SPOP multivalency is required for SPOP/DAXX phase separation in vitro. Fluorescence microscopy images of SPOP variants (green) with reduced self-association ability, in the presence or absence of H-cDAXX (red). Camera settings are the same down columns.

(C)SPOP multivalency is required for SPOP/DAXX co-localization in cells. HeLa cells were transfected with GFP-DAXX and V5-WT SPOP or mutants mutBACK or mutBTB/BACK and analyzed by confocal microscopy. GFP fluorescence was observed for DAXX, while V5-SPOP (red), and PML bodies (magenta) were detected by IF. (See also Fig. S2D.)

(D) SPOP mutants are defective in DAXX ubiquitination in cells. Western Blots showing GFP-cDAXX ubiquitination in HEK293 cells that were transfected with His₆-Ubiquitin, Myc-Cul3, HARbx1 and one of the SPOP variants each. The asterisk * indicates the IgG heavy chain. (See Fig. S2E for in cell ubiquitination assay with pull-down on His₆-Ubiquitin.)

SPOP oligomerization promotes phase separation and co-localization of DAXX and SPOP

We hypothesized that phase separation of DAXX and SPOP was mediated by the interaction of multiple SB motifs in DAXX with multiple MATH domains in oligomeric SPOP, as described previously for multivalent systems (Li et al., 2012). If this was in fact the case, reducing the multivalency of SPOP and DAXX should decrease their propensity to undergo phase separation. We therefore tested this hypothesis using previously established SPOP mutants of the BTB and BACK interfaces that lack the ability to form higher-order oligomers and are instead constitutively dimeric (SPOP^mutBACK, in which Y353 in the BACK interface is mutated (van Geersdaele et al., 2013)), or monomeric (SPOP^mutBTB/BACK, in which the substitutions L186D, L190D, L193D and I217K are made in the BTB interface (Zhuang et al., 2009)) (Fig. 1A, (Marzahn et al., 2016). Indeed, these SPOP mutants do not or only slightly enhance H-cDAXX phase separation (Fig. 2B).

Similarly to our in vitro observations, SPOP^mutBACK and SPOP^mutBTB/BACK were unable to relocalize DAXX to SPOP/DAXX bodies in HeLa cells. Instead, SPOP^mutBACK and SPOP^mutBTB/BACK were diffuse in the nucleus, in agreement with our earlier observation that SPOP self-association was required for recruitment to liquid organelles (Marzahn et al., 2016). Consequently, DAXX was found in PML bodies in the presence of SPOP^mutBACK and SPOP^mutBTB/BACK (Fig. 2C, and Fig. S2D). Therefore, SPOP-self-association into higher-order oligomers via its BTB and BACK domains promotes phase separation with DAXX in vitro and co-localization in cells.

We tested whether the lack of co-localization of DAXX and SPOP mutants resulted in decreased substrate ubiquitination. While DAXX and cDAXX were readily ubiquitinated by WT SPOP, ubiquitination was substantially reduced with both SPOP mutants, SPOP^mutBACK and SPOP^mutBTB/BACK (Fig. 2D, and Fig. S2E), even though their levels in cells were higher. This defect of oligomerization-deficient SPOP mutants is consistent with our previous reports on the same mutants in an in vitro ubiquitination assay (Marzahn et al., 2016; Pierce et al., 2016). Our data suggest that the underlying mechanism of reduced ubiquitination in the cell is mislocalization due to disrupted phase separation.

Multiple weak SPOP-binding motifs in DAXX mediate phase separation with SPOP

Since SPOP multivalency is critical for phase separation, DAXX multivalency might also be critical. We identified SB motifs in cDAXX, and scrambled the strongest sites in the construct cDAXX-0sb (see methods section for details), to test the ability of SPOP to enhance phase separation of cDAXX vs cDAXX-0sb.

To identify SB motifs with sequence homology to the SB consensus motif (Ф-П-S-S/T-S/T, where Ф is a nonpolar and П is a polar residue (Zhuang et al., 2009)) in cDAXX, we performed a bioinformatic search allowing for one mismatched position per motif. We found 5 potential binding sites with different agreement with the consensus sequence and thus varying predicted strength (Fig. 1A and Table S1). To characterize binding of these SB motifs by two complementary biophysical methods, we mapped regions in cDAXX that bound to SPOP^MATH by NMR spectroscopy, and quantified the binding of each predicted site, and the full cDAXX constructs, by Fluorescence Anisotropy (FA).

We assigned the NMR resonances of ¹⁵N,¹³C cDAXX (Fig. S3A, B), which have the limited chemical shift dispersion and sharp linewidths of a typical intrinsically disordered protein region (IDR) (Fig. 3A (Eliezer, 2009; Mittag and Forman-Kay, 2007)). These spectral properties indicate that many cDAXX conformations are in fast exchange and that the IDR exhibits little structure. Any SB motifs present in cDAXX should thus be accessible for SPOP binding. A titration of unlabeled SPOP^MATH into ¹⁵N,¹³C cDAXX and monitored by CON spectra, resulted in a dose-dependent loss of signal intensity along several cDAXX regions (Fig. 3A, B). The regions of broadening coincided with most of the predicted motifs and revealed one other weak site (Fig. 3C, solid and dashed orange lines, respectively). Additional broad regions experienced intensity loss, indicating there may be additional cryptic binding motifs.

Figure 3. Multiple weak SPOP-binding motifs in DAXX mediate phase separation with SPOP.

(A) cDAXX is intrinsically disordered and binds SPOP via several SB motifs. ¹⁵N,¹³C CON NMR spectrum of cDAXX at 600 MHz and 25 °C, without SPOP^MATH (black) and in the presence of 2 molar equivalents of SPOP^MATH (red). (For spectra annotated with all assignments, see Fig. S3A, B.)

(B) Titration of SPOP^MATH into cDAXX leads to identification of SB motifs. Intensity ratios of CON correlations for cDAXX upon titration with SPOP^MATH (I/I₀) are plotted as a function of residue number. Broadening of CON resonances of cDAXX in the presence of SPOP^MATH reveals multiple SB motifs, the 5 predicted (solid orange lines), one unpredicted (dashed orange lines), and other broadened regions.

(C) Sequence schematic for cDAXX constructs updated based on binding data in (B), (D) and (E). Stronger SB motifs are shown in darker shades of orange. In cDAXX-0sb, the nonpolar residue in each SB motif was replaced with a polar residue, the second residue replaced with a proline, and the rest of the motif sequence scrambled. (See Table S1 and S2 for sequences.)

(D - G) cDAXX binds SPOP in an SB motif-dependent manner. Representative fluorescence anisotropy competition binding isotherms for peptides containing cDAXX binding sites (D) or mutated binding sites (E) into SPOP^MATH and fluorescein-Puc⁹¹⁻¹⁰⁶, and direct binding isotherms for SPOP^MATH (F) and WT SPOP (G) into full-length Rhodamine-cDAXX constructs. Symbols are experimental data points; continuous or dashed lines are non-linear least-squares fits (Roehrl et al., 2004). All measurements were conducted in triplicate. (Average K_Ds are shown in Tables S2 and S3, respectively.)

(H) DAXX-0sb does not phase separate with SPOP in vitro. Fluorescence microscopy images of SPOP with cDAXX or cDAXX-0sb. All samples contain 10% w/v ficoll 70, 500 nM ORG-SPOP and/or Rhodamine-cDAXX.

(I) DAXX-0sb does not localize predominantly to SPOP/DAXX bodies in cells. HeLa cells were transfected with GFP-cDAXX or GFP-cDAXX-0sb and SPOP-mCherry and analyzed by confocal microscopy. cDAXX-0sb in the absence of endogenous SPOP localizes to PML bodies (Fig. S3C).

(J)Quantification of partition coefficient of GFP-cDAXX and GFP-cDAXX-0sb into SPOP/cDAXX bodies in (I). Each point in the whisker plot signifies an individual cell, the mean is shown as a line. Error bars indicate the SD.

(K)The cDAXX-0sb mutant is defective for ubiquitination in cells. Western Blots showing GFP-cDAXX and GFP-cDAXX-0sb ubiquitination in HEK293 cells that were transfected and analyzed as in Fig. 2D.

To determine the affinity of the five predicted SB motifs to the SPOP^MATH-groove, we generated 13-residue long peptides encompassing the motifs (Table S2), and performed FA competition binding experiments. Indeed, all candidate SB motifs interacted with SPOP weakly (Fig. 3D) with K_D values from 40 μM to 1 mM (Fig. 3D and Table S2), in agreement with the role of weak multivalent interactions in LLPS.

Next, we scrambled the motifs, and they indeed lost the ability to interact with SPOP^MATH; only one mutant motif still interacted with a K_D value of 2 mM (Fig. 3E and Table S2). Consequently, cDAXX-0sb, the cDAXX version with scrambled motifs, bound to SPOP^MATH with a K_D in the range of hundreds of micromolar while WT cDAXX had a K_D of 40 μM (Fig. 3F and Table S3). The interaction of cDAXX-0sb with multivalent SPOP²⁸⁻³⁵⁹ was even more dramatically decreased compared to WT cDAXX, with K_D values of hundreds of micromolar vs 1.7 μM, respectively (Fig. 3G, Table S3). Residual weak SB motifs must still be present in cDAXX-0sb (Fig. 3C), and they result in weak binding to SPOP and SPOP^MATH. But, as expected, the reduced multivalency of cDAXX-0sb prevented SPOP-enhanced phase separation (Fig. 3H). Our results therefore confirm that cDAXX contains multiple SB motifs that are critical for its ability to form condensed droplets with SPOP.

We wondered whether the multivalency of DAXX is similarly important for co-localization in cells as it is for phase separation in vitro. First, the localization of GFP-cDAXX in cells showed the same dependence on SPOP expression as that of full-length DAXX; GFP-cDAXX was diffuse with some fraction of the protein in PML bodies (Fig. S1B), and it co-localized with SPOP when transiently expressed together (Fig. 3I). Second, cDAXX-0sb was less enriched in SPOP/DAXX bodies than WT cDAXX (Fig. 3I-J, S3C). Residual co-localization is likely caused by the same weak multivalent interactions that we detected by NMR and FA (Fig. 3B, E, C). Therefore, the localization in cells and the ability to phase separate in vitro depend similarly on SPOP oligomerization and DAXX multivalency. Furthermore, ubiquitination of cDAXX-0sb is significantly reduced compared to cDAXX, as is the case for SPOP oligomerization-deficient mutants compared to SPOP (Fig. 3K, 2D, S2E). Based on the dependence on multivalency for DAXX and SPOP that we have shown, it is reasonable to conclude that SPOP and DAXX also undergo phase separation in the cell.

Material properties of SPOP/DAXX mesoscale assemblies

When we explored formation of mesoscale SPOP/H-cDAXX assemblies in more detail, we observed that droplets formed at H-cDAXX/SPOP molar ratios above ~3 (Fig. 4A and magnification at top right). At molar ratios below ~2, the assemblies have a filamentous morphology, as especially noticeable in DIC images (Fig. 4A and magnification on bottom right; the DIC image is incorporated in the composite image to highlight the texture of the filamentous assemblies). We wondered whether these two different types of mesoscale assemblies were formed through two different assembly processes, or whether they had different material properties because of their different composition, and whether the filamentous assemblies represented non-native, irreversible aggregates from maturation of liquid droplets.

Figure 4. Material properties of SPOP/DAXX mesoscale assemblies.

(A) SPOP and H-cDAXX form filamentous assemblies as well as liquid droplets. Fluorescence microscopy images of SPOP/H-cDAXX as a function of protein concentration. All samples contain 10% w/v ficoll 70, 500 nM ORG-SPOP and/or Rhodamine-H-cDAXX. Images in red boxes are shown zoomed in with DIC overlaid at the right.

(B) Quantification of protein concentration in mesoscale assemblies in the first row of (A, blue box). Error bars represent the SD from three replicate images. (For standard curves and additional conditions see Fig. S4A-C.)

(C) Filamentous assemblies are not irreversible aggregates. (top) Time course of fluorescence microscopy/DIC images of a 15 μM SPOP : 50 μM H-cDAXX sample, which develops its typical droplet appearance over time. (middle) Addition of extra H-cDAXX to a filamentous sample incubated for 2 hr. The assemblies change from the filamentous to the droplet-like morphology. (bottom) Fusion events between SPOP/H-cDAXX droplets (red boxes).

(D) Schematic of the proposed nature of assemblies at different SPOP/H-cDAXX molar ratios. SPOP alone forms oligomers (left). Oligomers are stabilized in the presence of low molar ratios of H-cDAXX, leading to large filamentous assemblies (middle left). At higher molar ratios of H-cDAXX, intermolecular interactions are favored, SPOP oligomers are smaller, and H-cDAXX contributes to liquid behavior (middle right). H-cDAXX alone forms droplets (right). (See also Fig. S4 D-G.)

Some dense liquid assemblies can mature into gel-like states or nucleate fibrillization over time (Lin et al., 2015; Mackenzie et al., 2017; Molliex et al., 2015; Monahan et al., 2017; Murakami et al., 2015; Patel et al., 2015; Zhang et al., 2015). Importantly, this maturation of dense liquid states has been proposed as the precipitating event in protein aggregation diseases (Lin et al., 2015; Mackenzie et al., 2017; Molliex et al., 2015; Monahan et al., 2017; Murakami et al., 2015; Patel et al., 2015). Interestingly, we indeed observed morphological changes over time before assemblies reached their final state; i.e., samples that result in assemblies with droplet character after incubation can have filamentous character when first mixed (Fig. 4C top). Notably, it was also possible to manipulate the morphology of samples by adding one of the components after incubation; when we added additional H-cDAXX to a sample with filamentous assemblies (condition as in the inset in Fig. 4A, bottom), the sample evolved to a droplet state (Fig. 4C; as in Fig. 4A, starred condition; please note camera settings are different, and thus assemblies appear to have different colors). This evolution suggests that the filamentous assemblies do not represent irreversible, non-native aggregates.

We attribute these morphological changes to slow off-rates of SPOP building blocks from SPOP oligomers in the presence of crowders and multivalent substrate, as we saw previously with the multivalent substrate Gli3¹⁻⁹⁰ (Pierce et al., 2016). All samples were therefore imaged after 4–6 hours of incubation. The assemblies with droplet appearance undergo fusion events (Fig. 4C bottom), which demonstrate liquid-like properties.

We wondered whether the filaments and droplets were both formed through LLPS and merely represented different material states, or whether they formed through different assembly processes. We generated samples consisting of 15 μM SPOP and increasing concentrations of H-cDAXX, equivalent to the top row in the phase diagram in Fig. 4A (blue box). After incubation, the protein concentrations within the mesoscale assemblies were determined via their fluorescence intensity in confocal micrographs (Fig. 4B, Fig. S4A-C). The concentrations followed a biphasic behavior. They first increased with the H-cDAXX concentration in the sample, consistent with increasing levels of protein incorporated into assemblies in a typical oligomerization/polymerization process. The H-cDAXX concentration in assemblies then plateaued above a H-cDAXX input concentration of 30 μM (Fig. 4B), while the SPOP concentration decreased continuously. This is in agreement with a mechanism, in which H-cDAXX is able to mediate phase separation independently and SPOP promotes H-cDAXX phase separation further (Fig. S2A-B, Fig. 4D).

These results support a scenario, in which SPOP forms linear, filamentous higher-order oligomers as previously described (Marzahn et al., 2016) (Fig. 4D, left). The addition of multivalent H-cDAXX (at levels below the saturation concentration for phase separation) leads to the stabilization of the oligomeric state, observed as larger species in cross-linking assays (Fig. S4D, E), resulting in solid filaments through an oligomerization/polymerization process (Fig. 4D left middle, S4E left and middle boxes). With increasing H-cDAXX concentrations, H-cDAXX molecules compete for SPOP, and SPOP oligomers are less stabilized; this is observed as smaller species in cross-linking experiments (Fig. S4E right box). Eventually, more strongly networked complexes form, resulting in the formation of dense SPOP/DAXX-containing droplets via LLPS (Fig. 4D, middle right). High concentrations of H-cDAXX can undergo LLPS alone (Fig. 4D, right).

In support of this interpretation, the H-cDAXX mobility as measured by fluorescence recovery after photobleaching (FRAP) is larger than the SPOP mobility, notwithstanding the type of assembly DAXX is incorporated in (Fig. S4F, Table S4). Interestingly, the DAXX mobility is nearly identical in filaments and liquids, suggesting a steady flux of DAXX in and out of assemblies. The SPOP mobility is similar in the absence of DAXX and in the filamentous assemblies with DAXX. This points to SPOP as the scaffold, formed from large, adhesive oligomers, with DAXX binding to it from the outside. DAXX forms the liquid-promoting contacts. Only in SPOP/DAXX droplets does SPOP have an increased mobility, suggesting a change in the underlying structure from filaments to more crosslinked complexes. FRAP analysis of GFP-DAXX in cells shows a higher mobility of DAXX in PML bodies than in SPOP/DAXX bodies, reflecting the different nature of the interactions driving DAXX localization into the different nuclear bodies. SPOP has a low mobile fraction in nuclear speckles as well as in SPOP/DAXX bodies, indicating viscoelastic properties of SPOP-containing bodies (Fig. S4G).

Together, our results suggest that the concentration-dependent multivalency of SPOP, coupled to the phase separation-propensity encoded in its substrate, result in a rich phase diagram with several different types of mesoscale assemblies (Fig. S4H).

SPOP cancer mutants disrupt phase separation and DAXX ubiquitination

Given the role of phase separation for SPOP/substrate colocalization, we hypothesized that SPOP cancer mutations would disrupt this behavior, explaining the cellular mislocalization shown in Fig. 1E. Since SPOP cancer mutants disrupt substrate binding (Gan et al., 2015; Geng et al., 2017), we first tested the binding of SPOP W131G to a peptide with a single SB motif (fPuc). Indeed, SPOP W131G did not bind fPuc (Fig. S5A and Table S5), and we therefore expected that the phase separation propensity of SPOP cancer mutants with substrates would be reduced. Indeed, in vitro, SPOP W131G did not form droplet assemblies up to a concentration of ~75 μM H-cDAXX, while only 30 μM was required for the formation of droplets with SPOP WT (at a constant SPOP concentration of 15 μM; Fig. 5A). SPOP F133V failed to form droplet assemblies with H-cDAXX over the full concentration range tested and maintained filamentous character throughout. The propensity of SPOP to form condensed liquid droplets with DAXX was therefore markedly decreased by cancer mutations. These results demonstrate that SPOP/substrate interactions are substantially weakened by cancer mutations, that multivalency of both binding partners can still result in a physical association and the formation of large, solid assemblies, but LLPS requires increased protein concentrations.

Figure 5. SPOP cancer mutants disrupt phase separation and DAXX degradation.

(A) SPOP cancer mutants are defective at phase separation in vitro. Fluorescence microscopy/DIC images of WT SPOP or cancer mutants as a function of H-cDAXX concentration. All samples contain 10% w/v ficoll 70, 500 nM ORG-SPOP construct and/or Rhodamine-cDAXX. Camera settings were optimized in samples containing ~1:1 molar ratios for each row.

(B) SPOP cancer mutants are defective at co-localization with DAXX in HeLa cells. SC-35 (magenta) was used as marker for nuclear speckles. Cells with SPOP-DAXX co-localization or lack thereof are indicated. (For co-staining with PML, see Fig. S5B.)

(C) SPOP cancer mutants co-localize with DAXX when expressed at high levels. Whisker plot showing the signal intensity of V5-SPOP or V5-F133V (red points) and GFP-DAXX (green points) from (B) in which the V5-SPOP construct and GFP-DAXX co-localize or fail to co-localize. Each point represents a single cell. Horizontal lines indicate the mean; error bars indicate SD.

The observation that SPOP W131G was able to form droplet assemblies with H-cDAXX at high concentrations in vitro prompted us to test whether the mislocalization defect of SPOP cancer mutants with DAXX could be rescued at high protein concentrations in cells as well, as would be expected if their normal co-localization is dependent on phase separation. Indeed, while V5-SPOP W131G and V5-SPOP F133V typically remained in nuclear speckles when co-expressed with GFP-DAXX, a fraction of cells with significantly higher SPOP and DAXX levels showed colocalization in SPOP/DAXX bodies (Fig. 5B,C, S5B). These results also suggest that the filamentous assemblies do not mediate the formation of liquid-like bodies in the cell, and that these instead require a phase transition process.

We next tested whether the lack of co-localization of DAXX and SPOP cancer mutants resulted in decreased substrate ubiquitination. While GFP-cDAXX was readily ubiquitinated by WT SPOP, ubiquitination was much reduced with the SPOP cancer mutants (Fig. 2D). We also observed reduced ubiquitination of full-length FLAG-tagged DAXX in the presence of SPOP mutants (Fig. S2E), in agreement with previous reports showing reduced substrate ubiquitination by SPOP cancer mutants (An et al., 2014; Dai et al., 2017a; Dai et al., 2017b; Gan et al., 2015; Geng et al., 2013; Geng et al., 2017; Geng et al., 2014; Janouskova et al., 2017; Theurillat et al., 2014). Together, our data suggest that the underlying mechanism of reduced ubiquitination is the disruption of phase separation, which results in a failure of the SPOP cancer mutants to co-localize with DAXX.

SPOP/DAXX bodies are active ubiquitination compartments

Given that SPOP and substrates co-localize in liquid compartments, and that this colocalization is disrupted by functionally deficient, cancer-associated mutants, we tested whether the liquid compartments are indeed active for SPOP-mediated ubiquitination.

We first tested whether other subunits of the CRL3^SPOP ubiquitin ligase are present in the SPOP/DAXX bodies. Transient expression of Myc-Cul3, the scaffold that bridges the substrate and the E2 ubiquitin-conjugating enzyme in Cullin/RING ubiquitin ligases, resulted in its diffuse localization in the cytoplasm in the absence of SPOP. In contrast, Myc-Cul3 localized in nuclear speckles in the presence of SPOP, and in SPOP/DAXX bodies in the presence of both SPOP and DAXX (Fig. 6A). Therefore, Cul3 is recruited to liquid organelles by SPOP. HA-Rbx1, the RING protein associating with Cul3, also localizes to SPOP/DAXX bodies (Fig. 6B). In vitro, the active form of the scaffold, i.e. neddylated Cul3/Rbx1 (N8~Cul3/Rbx1), likewise partitioned into the mesoscale assemblies (Fig. 6C). Therefore, it is plausible that CRL3^SPOP could carry out its function in cellular SPOP-containing liquid assemblies.

Figure 6. SPOP/DAXX bodies are active ubiquitination compartments.

(A) SPOP recruits Cul3 to SPOP/DAXX bodies. HeLa cells were transfected with the indicated constructs. Cul3-Myc (magenta) was detected by IF.

(B) SPOP recruits Cul3 and Rbx1 to SPOP/DAXX bodies. Cul3-Myc (blue) and Rbx1-HA (magenta) were detected by IF.

(C) Cul3 partitions into SPOP/DAXX assemblies in vitro. Fluorescence microscopy images of N8~Cul3/Rbx1 (blue channel), SPOP (green) and H-cDAXX (red). The blue-only channel images were pseudo-colored to black/white for clarity. All samples contain 500 nM of each Alexa647N8~Cul3/Rbx1, ORG-SPOP, and Rhodamine-H-cDAXX. Samples were in 25 mM HEPES pH 7.5, and 150 mM NaCl (top row) and 25 mM Tris pH 7.6, 150 mM NaCl, and 1 mM T-CEP (bottom row).

(D) Conjugated ubiquitin in SPOP/DAXX bodies depends on SPOP-Cul3 interaction. HeLa cells were transfected with the indicated constructs. Cul3-Myc (blue) and conjugated ubiquitin (magenta, with FK2 antibody) were detected by IF.

(E) Disruption of SPOP/Cul3 interaction results in increased GFP-cDAXX levels and decreased conjugated ubiquitin levels. Quantification of signals from GFP (green bars), conjugated ubiquitin (magenta bars), and conjugated ubiquitin normalized by GFP (open bars) for n=20 cells per condition in (D).

(F) Schematic representation of in vitro ubiquitination assay. Transfer of ubiquitin is monitored by SDS-PAGE and the incorporation of fluorescent *UB into assemblies microscopically.

(G) Ubiquitinated H-cDAXX accumulates in SPOP/H-cDAXX assemblies in vitro. Fluorescence microscopy/DIC images showing the time-course of in vitro ubiquitination assays described in (F) at the indicated SPOP (green) / H-cDAXX (red) molar ratios plus 1.25 μM N8~Cul3/Rbx1, 20 nM ARIH1 as indicated on the left, and 1.5 μM UbcH7~*UB (blue). All reactions contain either ficoll 70 or sucrose as indicated, and 500 nM ORG-SPOP and Rhodamine-H-cDAXX; *UB denotes stoichiometrically labeled Alexa647-Ubiquitin. (See Fig. S6C for images of control reaction conditions.)

(H) Ubiquitination can occur in the presence or absence of SPOP/DAXX assemblies. Representative fluorescent scan of non-denaturing gels showing time course of in vitro ubiquitination reactions described in (F). Blue UBCH7~*UB band diminishes and blue H-cDAXX~*UB band appears over the course of reactions containing WT ARIH1 + ficoll70 or sucrose, but not in reactions containing no ARIH1 or the catalytically inactive mutant, ARIH1^C375S.

(I) Quantification of in vitro ubiquitination assay from blue fluorescence intensity of assemblies in fluorescence microscopy images in (G and Fig. S6C)—left; and gel band intensity of product HcDaxx~*UB in (H and Fig. S6C)—right. Data points represent average of triplicate experiments. Error bars indicate SD.

We next tested whether ubiquitination activity resides in the SPOP-containing bodies. Indeed, SPOP/cDAXX/Cul3-containing bodies in cells were positive for conjugated ubiquitin as determined by positive staining with the FK2 ubiquitin antibody, which does not stain free ubiquitin (Fig. 6D). To determine if the conjugated ubiquitin signal is dependent on CRL3^SPOP, we quantified ubiquitin conjugation in cells expressing Cul3 or SPOP mutants. For Cul3, we used Cul3^H2H5, a previously described mutant that interferes with SPOP binding (Furukawa and Xiong, 2005). For SPOP, we mutated residues that we predicted would affect the SPOP/Cul3 interaction (Fig. S6A) and called the resulting mutant SPOP^CBM or Cul3-binding mutant. Indeed, both Cul3^H2H5 and SPOP^CBM led to a reduced ubiquitin signal in the SPOP/cDAXX bodies (Fig. 6D-E). The combination of the Cul3 and the SPOP mutant further reduced the level of conjugated ubiquitin in the bodies, demonstrating that a significant fraction of conjugated ubiquitin in the SPOP/DAXX bodies stems from CRL3^SPOP-mediated ubiquitination. Furthermore, cDAXX levels in cells rose with decreasing ubiquitination in the bodies (Fig. 6E), supporting our hypothesis that DAXX is ubiquitinated within the bodies and subsequently degraded. Immunoprecipitation of cDAXX and blotting for ubiquitin confirmed that cDAXX was ubiquitinated in a Cul3 and SPOP dependent manner in these cells (Fig. S6B).

To exclude the possibility that other ubiquitin ligases in the SPOP/DAXX bodies are responsible for ubiquitination, we moved to an in vitro ubiquitination assay with purified, recombinant proteins. We tested an array of E2 conjugating enzymes for their activity towards cDAXX in the presence of neddylated CRL3^SPOP, and observed the strongest activity with ARIH1/UBCH7 (Fig. S6D) (Scott et al., 2016). We then designed an in vitro ubiquitination assay to test whether transfer of ubiquitin onto H-cDAXX correlated with the appearance of ubiquitin in mesoscale assemblies. We charged UBCH7 with an equimolar amount of fluorescently-labeled ubiquitin (N-terminally labeled Alexa647-UB, from now on *UB), quenched the charging reaction with EDTA, and added *UB~UBCH7 to pre-formed SPOP/H-cDAXX/N8~Cul3/Rbx1 assemblies, in the presence and absence of ARIH1 or a catalytically inactive mutant (ARIH1^C357S, (Scott et al., 2016)) (Fig. 6F). This reaction mixture is competent for a single turnover, i.e. discharge of *UB and potentially transfer onto an acceptor lysine of a substrate. With live fluorescence imaging, we observed the appearance of *UB in the assemblies over time in the presence of enzymatically active ARIH1, resulting in co-localization of H-cDAXX, SPOP and *UB (Fig. 6G, I left). Visualization of the reaction products by SDS-PAGE showed ubiquitinated H-cDAXX in the presence of active ARIH1, but not with an enzymatically inactive mutant or in its absence (Fig. 6H, I right, Fig S6C). Both filamentous and droplet-like assemblies with WT SPOP were able to mediate activity. However, we also observed effective transfer of *UB onto DAXX in the presence of sucrose instead of ficoll (Fig. 6I, right, Fig. S6C). Under these conditions, SPOP and DAXX did not form mesoscale assemblies, but the reaction went to completion with similar kinetics. The SPOP cancer mutants W131G and F133V were hardly able to mediate DAXX ubiquitination. (Fig 6G, H, I)

We come to four conclusions from these results: (1) Since most of the SPOP/DAXX is concentrated in the assemblies and little protein is diffuse under phase separation conditions, the reaction must occur largely within the assemblies. (2) Filamentous and droplet-like WT SPOP/DAXX assemblies have similar activities, in agreement with the similar protein mobilities we observed within them. (3) The assemblies do not enhance enzymatic turnover compared to diffuse reactions. This may be explained by competition between enhancing and decelerating factors such as high local concentrations and increased viscosity in the dense assemblies, respectively. FRAP indeed shows a considerable immobile fraction of DAXX and in particular of SPOP in all assemblies (Fig. S4G). (4) SPOP cancer mutants show reduced activity, in agreement with their reduced ability to assemble and their expected higher off-rates.

Our results therefore support a model in which the UB-charged E2 diffuses into the SPOP/DAXX assemblies and CRL3^SPOP-mediated ubiquitination occurs within the assemblies. We therefore propose that SPOP-mediated ubiquitination occurs largely within membraneless organelles in the cell, e.g. within SPOP/DAXX bodies.

SPOP phase separates with androgen receptor

To determine whether the synergistic recruitment of SPOP and DAXX to liquid organelles is a general feature of other SPOP substrates, we predicted SB motifs in a number of known SPOP substrates from the literature. Most of SPOP’s substrates likely harbor weak SB motifs in their intrinsically disordered regions, in addition to the experimentally confirmed motifs in the literature (Table S6). We thus hypothesized that SPOP recruits at least a subset of its substrates via phase separation. To test this hypothesis, we generated a large N-terminal fragment of the androgen receptor that contained the majority of the predicted SB motifs (nAR, Fig. 7A). AR is a well-known substrate of SPOP; increased AR levels caused by SPOP cancer mutations likely contribute to prostate cancer pathogenesis (An et al., 2014). nAR had the ability to form liquid-like droplets at high concentrations in vitro, and their formation was strongly enhanced by SPOP (Fig. 7B). The nAR/SPOP assemblies had similar properties to SPOP/DAXX assemblies, ranging from filamentous to liquid droplet-like depending on the nAR/SPOPmolar ratio (Fig. 7B). Similarly, transiently expressed full-length GFPAR and SPOP-mCherry co-localized in punctate membraneless bodies in HeLa cells (Fig. 7C). nAR binding to SPOP²⁸⁻³⁵⁹ is enhanced compared to binding to the MATH domain (Fig. S7A), supporting the existence of multiple SB motifs in nAR, which can mediate phase separation with multivalent SPOP, analogously to DAXX. We thus propose that SPOP has the ability to undergo phase separation with multivalent substrates as a general mechanism for targeting substrates.

Figure 7. SPOP phase separates with androgen receptor and may phase separate with other substrates in an evolutionarily conserved fashion.

(A) Sequence schematic for AR and the N-terminal fragment used for in vitro experiments (nAR). AR contains an N-terminal disordered domain, DNA binding domain (DBD), and a ligand binding domain (LBD) (Centenera et al., 2008).

(B) SPOP phase separates in vitro with nAR. Fluorescence microscopy images of SPOP (green) and nAR (red). All samples contain 10% w/v ficoll 70, and 500 nM ORG-SPOP and/or Rhodamine-nAR. Samples were in 20 mM NaPO₄ pH 7.4, 50 mM NaCl, 2 mM T-CEP, and 1 mM EDTA.

(C) SPOP co-localizes with the AR in cells. HeLa cells were transfected with the indicated constructs and analyzed by confocal microscopy.

(D) Covariation analysis of SPOP shows evolutionary coupling across the BTB and BACK interfaces. Co-evolutionary couplings in SPOP from covariation of ~2600 SPOP homologues (Table S7) sharing all three structural domains. The co-evolutionary couplings (top 600) are reported in the upper triangle of the matrix as black dots with size proportional to the relative coupling strength, overlapping both intra- and intermolecular contacts. The couplings are compared to contacts between pairs of residues with a distance of up to 5 Å between sidechain heavy atoms, based on a structural model of SPOP²⁸⁻³⁵⁹ (built using two available crystal structures (PDB ID 3HQI (Zhuang et al., 2009) and 4HS2 (van Geersdaele et al., 2013)), and no further assumptions); intra- and intermolecular contacts are shown in blue and red, respectively.

(E) BTB and BACK interface residues coevolve with residues across the interface, not with domain core residues. Evolutionary domains, obtained by the analysis of the patterns of couplings (Granata et al., 2017) are reported on the SPOP monomer structure model for the subdivision in 2 groups of coevolving residues (Q=2) (top), and on the oligomer structure model for 5 groups of coevolving residues (Q=5) (bottom). Other meaningful subdivisions are reported in Fig. S7B-D.

(F) Schematic of proposed mechanism. SPOP phase separates with multivalent substrates, and is able to target and ubiquitinate substrates localized to membrane-less organelles. SPOP cancer mutants are defective at phase separation and therefore co-localization and ubiquitination.

SPOP oligomeric interfaces are evolutionarily conserved

Since SPOP multivalency was required for phase separation with substrates, we wondered whether the ability to form higher-order oligomers was under evolutionary pressure. We employed a co-evolutionary coupling analysis (Ekeberg et al., 2013; Marks et al., 2011), which assesses statistical coupling of correlated mutations between all residue positions, from the covariation in sequence alignments of ~2600 SPOP orthologs (Table S7 for their taxonomy). In particular, we assessed whether there was any covariation between residues in SPOP that could not be explained by intramolecular contacts within a SPOP monomer, and that instead coincide with the known intermolecular interfaces.

The contact map of the SPOP oligomer (Marzahn et al., 2016; van Geersdaele et al., 2013) highlights pairs of residues that are in proximity across the oligomerization interfaces, but not within SPOP monomers (Fig. 7D). Comparison with the 600 strongest non-local (|i-j|>3) couplings reveal close agreement between coevolving residue pairs and contacts in the protein structure, confirming that co-evolution reports on spatial proximity. Notably, we also observe numerous couplings across the BTB and the BACK interfaces (Fig. 7D).

We further analyzed the patterns of the evolutionary couplings to find natural groups of coevolving residues, also called evolutionary domains (Granata et al., 2017). A subdivision into two evolutionary domains results in a good fit to the data (Fig. S7B), and points to the BTB and BACK domains as a single coevolving unit, while the MATH domain forms a separate unit (Fig. 7E, top, Fig. S7C). Progressively finer subdivision next results in a good fit for 5 evolutionary domains (Fig. S7B, C). With this subdivision, it becomes clear that the residues of the BACK domain interfacial helices coevolve across the interface, while the rest of the BACK domain coevolves with the BTB domain (Fig. 7E, bottom). Even finer subdivision shows a similar co-evolution of BTB interfacial residues across the interface (Fig. S7D). Based on this analysis, the BTB and BACK domains in SPOP appear to have coevolved their ability to dimerize synergistically into higher-order oligomers, and our results suggest that this property is under evolutionary pressure. This finding supports the possibility that SPOP has evolved multivalent properties to target substrates through phase separation.

Discussion

SPOP localizes to several membraneless organelles in the nucleus, but the mechanism underlying its redistribution between these organelles, and how this is related to substrate targeting, has so far been unclear. Here, we show that substrates trigger SPOP co-localization and CRL activity by mediating phase separation, and that SPOP cancer mutants disrupt both processes (Fig. 7F). We propose the disruption of phase separation as the mechanism underlying SPOP mutant-driven oncogenesis. The ability of SPOP to form higher-order oligomers, where the SPOP concentration determines the size of the oligomers and therefore the valency of SPOP for substrates, gives rise to a rich phase diagram; SPOP can undergo oligomerization/polymerization processes and LLPS with multivalent substrates, depending on concentrations and molar ratios. LLPS resulting in droplet-like assemblies seems to be the basis for colocalization and activity in cells.

Phase separation may allow targeting substrates in membraneless organelles

Phase separation is a potential mechanism for concentrating enzymes and substrates and enhancing turnover (Banani et al., 2017; Li et al., 2012). In vitro, condensed BugZ droplets enhance tubulin polymerization (Jiang et al., 2015), and Nephrin/Grb2 phase separation enhances actin polymerization (Li et al., 2012); whether the respective liquid compartments in the cell also enhance polymerization is unclear. While the nucleolus is a phase-separated, enzymatically active compartment (Berry et al., 2015; Mitrea et al., 2016), it is less clear whether phase separation is needed for activity. Smaller oligomeric assemblies may be able to sustain the activities found in typical membraneless organelles (Smith et al., 2016; Wallace et al., 2015). Many membraneless organelles may be sequestering proteins, RNA or DNA, rather than forming active compartments. Here, we show strong evidence that SPOP- and substrate-containing membraneless organelles contain enzymatic activity.

But given our observation that SPOP ubiquitinates the substrate DAXX equally efficiently in the absence of large assemblies, why has SPOP evolved the ability to phase separate? In cells, when not incorporated in SPOP/DAXX bodies, SPOP and DAXX are not largely diffuse but incorporated into two different nuclear bodies, i.e. nuclear speckles and PML bodies. Presumably only a small fraction of diffuse protein has the possibility to interact. In contrast, in our in vitro ubiquitination assay, the fraction of unassembled protein is not sequestered in separate bodies but is available for interaction and turnover. Importantly, there is evidence for several SPOP substrates that they localize to nuclear bodies (Kwon et al., 2006; Li et al., 2000; Weidtkamp-Peters et al., 2008) (Klokk et al., 2007; Tomura et al., 2001; Tyagi et al., 2000) (Zhang et al., 2014). ERG and BET proteins may also participate in phase separation processes proposed for transcription factories and chromatin (Hnisz et al., 2017; Larson et al., 2017; Strom et al., 2017). Indeed, SPOP substrates are predicted to have a high propensity to phase separate, which is well above those of proteins with PDB structures (Fig. S7E) (Vernon et al., 2018). SPOP may have evolved an ability to phase separate in order to target substrates localized to membraneless organelles. Phase separation may be an efficient mechanism to achieve specific colocalization with substrates in various membraneless organelles. While diffuse samples without mesoscale assemblies are active in vitro, the full activity in cells requires co-localization of SPOP and substrates via phase separation.

Redistribution between membraneless organelles as a function of available substrates

Our results that SPOP cancer mutants with decreased substrate binding maintained localization to nuclear speckles, support a view in which nuclear speckles act as storage sites for SPOP in the absence of high substrate levels in the cell, either triggered by the presence of pseudo-substrates in nuclear speckles or by molecules retaining oligomeric SPOP via other interactions. We propose that the rise of substrate levels, potentially accompanied by additional signals, leads to substrate phase separation with SPOP and recruitment of SPOP to the respective organelle. Since the signals driving SPOP redistribution towards substrates are not well understood, we here used transient expression of SPOP and substrates to trigger assembly. The resulting dense phase may form a separate organelle, may be miscible with a pre-existing organelle, or may localize to the original organelle of the substrate. Several cases have recently been described in which phase separation of proteins or protein complexes supports their recruitment to preexisting membraneless organelles, including the recruitment of the miRISC complex to P bodies (Sheu-Gruttadauria and MacRae, 2018) and of UBQLN2 to stress granules (Dao et al., 2018).

Additional regulatory processes, such as post-translational modification, could serve to change the apparent substrate concentration sensed by SPOP. Phosphorylation of SB motifs negatively impacts SPOP binding (Zhuang et al., 2009) and could increase the tolerated substrate concentration before it phase separation with SPOP. Furthermore, phosphorylation of SPOP was recently shown to modulate its function (Zhang et al., 2018). Non-equilibrium regulatory mechanisms may influence phase separation, and therefore fine-tune SPOP substrate levels. Future studies will provide further insights into how SPOP moves from one organelle to another.

SPOP evolution and conservation highlights the functional importance of SPOP oligomerization

The multivalency of SPOP towards substrates, generated through linear SPOP oligomerization, is evolutionarily encoded in the sequence. We have previously reported that self-association deficient SPOP mutants can disrupt normal SPOP function through dominant-negative effects in a fly model (Marzahn et al., 2016). Furthermore, prostate cancer patients with SPOP mutations typically have one normal allele, i.e. no loss of heterozygosity. Intriguingly, we find that SPOP is extremely highly conserved within the human population. For example, we find no common missense variants of SPOP present at an allele frequency greater than 10^–4 (Table S8) in a database of more than 100,000 sequenced exomes (Lek et al., 2016). In contrast, we find multiple missense mutations in the very similar SPOPL, which is not able to form higher order oligomers (Errington et al., 2012). We therefore hypothesize that dominant-negative phenotypes resulting from SPOP mutations causes this protein to be extremely conserved. SPOP self-association is evolutionary conserved, functionally critical, and the resulting multivalency is required for SPOP-mediated phase separation.

Conclusion

We propose that phase separation is an evolutionary adaptation of the SPOP CRL substrate adaptor to target substrates in membraneless organelles. SPOP mutants not only disrupt substrate binding, but also phase separation, resulting in the failure to co-localize with and turn over the substrate. There is precedence that mislocalization of SPOP can result in oncogenesis; incorrect localization of SPOP into the cytoplasm under hypoxic conditions unleashes CRL activity on the tumor suppressor PTEN, which is not usually a SPOP substrate (Li et al., 2014). Other ubiquitin ligases also form higher-order oligomers (Yin et al., 2009), and multivalent interactions are prevalent in the ubiquitin proteasome pathway (Liu and Walters, 2010). Triggering activation of ubiquitin ligases by their substrates via phase separation may therefore be a common principle for attaining proteostasis.

STAR Methods text

Contact for Reagent and Resource Sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Tanja Mittag (Tanja.Mittag@stjude.org).

Experimental Model and Subject Details

Cell Lines

HeLa and HEK293T cells were grown under sterile conditions on Dulbecco’s Modified Eagle’s Medium (DMEM). PC-3 cells were grown under sterile conditions on RPMI 1640 Medium. Culture media were supplemented with 10% fetal bovine serum (FBS), 2mM L-Glutamine and antibiotics. Cells were grown at 37 °C with 5% CO₂.

Method Details

Prediction of SB motifs

The position of SB motifs was predicted using a bioinformatics search with the online tool PattinProt (https://npsa-prabi.ibcp.fr/NPSA/npsa_pattinprot.html), using the query sequence [GAVLIMFWP]-[STCYNQ]-S-[ST]-[ST], and allowing for 1 mismatch to the consensus SB motif sequence (Ф-П-S-S/T-S/T, where Ф is a nonpolar and П is a polar residue (Zhuang et al., 2009)). In order to create the cDAXX-0sb sequence, the SB motifs were scrambled in the following way to preserve the length and charge distribution of the cDAXX sequence: 1) The first position was mutated from a hydrophobic to polar residue. 2) The second position was mutated from a polar residue to a proline. 3) The third position was mutated to a charged residue (if one was in the original sequence). 4) The fourth or fifth position was mutated from S/T to a hydrophobic residue. The resulting mutated sequences slightly increased the Pro content from 11 to 12.6% and reduced the Ser content from 18.3 to 16.7% in cDAXX-0sb relative to cDAXX.

Plasmids

We purchased full-length DAXX cDNA in vector pDONR221 from DNASU (clone: HsCD00079589) and introduced it into pcDNA6.2-EmGFP by Gateway technology (Life Technologies) for expression in mammalian cells. We also excised the cDAXX (495–740) sequence by nested PCR and introduced it into pDEST17 (Thermo Fisher) by Gateway technology for bacterial expression of His-tagged protein and protein purification for NMR experiments. For all other in vitro experiments, we purchased synthetic, codon-optimized genes for cDAXX, and cDAXX^0SB, which included Cys to Ser mutations at positions 629, 664, 699, and 720 (Thermo Fisher),and introduced them into pCDNA6.2-EmGFP by Gateway technology for expression in mammalian cells and into pDEST17 for bacterial expression and protein purification. Site directed mutagenesis was used to add a TEV protease site to remove the His₆ tag from bacterially expressed cDAXX. Plasmids for V5-tagged SPOP-WT, mutBACK, and mutBTB-BACK were generated by switching the HA tag to V5 in pcDNA3-HA-SPOP (Marzahn et al., 2016) by site directed mutagenesis. We generated V5-SPOP F133V, and W131G by site directed mutagenesis of V5-SPOP-WT. The SPOP-mCherry plasmid was constructed by using our previously described HA-SPOP plasmid (Marzahn et al., 2016) as a template for PCR and SPOP was introduced into vector pmCherry-N1 by restriction digest followed by ligation. To generate the SPOP^CBM-mCherry mutant, we mutated M233 to E, D278 to A, and K279 to A by site directed mutagenesis using the restriction free cloning method (van den Ent and Lowe, 2006). Plasmids pcDNA3-myc-CUL3, pcDNA3-HA2-ROC1 (Rbx1) (Ohta et al., 1999), pcDNA3-Myc3-Cul3 H2M/H5M (Furukawa and Xiong, 2005), and pEGFP-C1-AR (Stenoien et al., 1999) were obtained from Addgene. The plasmid for His₆-Ubiquitin was a kind gift from Wenyi Wei (Harvard). We purchased the nAR (AR 1–559 Uniprot P10275) gene for bacterial expression (Thermo Fisher) and introduced it into pDEST17 (Thermo Fisher) by Gateway technology (Life Technologies).

The following plasmids were previously published: pFastbac GST-UB E1, pGEX4T1 GSTThrombin-UBCH7 (Huang et al., 2008), pGEX4T1 GST-Thrombin-UBC12, pGEX4T1 APPBP1UBA3, GST pGEX4T1 GST-ThrombinNEDD8 (Duda et al., 2008; Walden et al., 2003), pGEX2TK GST-Thrombin-UB (C>S) (Scott et al., 2014), pGEX4T1 GST-TEV-ARIH1 (and C357S mutant) (Scott et al., 2016), pET-DUET-1-His-Cul3/Rbx1 (Small et al., 2010)

Transfections

Cells were transfected with Lipofectamine 3000 (Thermo Fisher) or with Effectene (Qiagen) according to the manufacturer conditions.

Immunofluorescence

Cells were transfected in 8-well glass chambers (Millipore) and fixed with 4% paraformaldehyde 24 hours later. Cells were permeabilized with 0.1% Triton-X100 and blocked with 10% donkey serum. GFP and mCherry fluoresce was detected directly. The primary antibodies used were: SC-35 (1:300; Abcam, ab11826), PML (1:50; Santa Cruz, sc-966), coilin (1:500; Santa Cruz Biotechnology, sc-32860), B23 (1:200, Santa Cruz Biotechnology, sc-56622) Myc-tag (1:500 Cell Signaling Technologies, 71D10), HA-tag (1:250; Clone 3F10, Roche, 11867423001), FK2 (1:50; Enzo Life Sciences, BML-PW8810), V5-tag (1:300, Novus Biological, NB600–379). The secondary antibodies used were Alexa 555, 647 (1:5,000; Thermo-Fisher), and CF405S (1:1000; Biotium). Samples were mounted on ProLong Gold antifade with or without DAPI and cured before imaging on a Zeiss LSM 780 NLO microscope. Images were prepared with the Fiji software.

Live Cell Imaging

Cells were transfected in 4-well borosilicate chambers (LAB-TEK). Twenty-four hours after transfection, media was changed to phenol red-less DMEM and imaged in a Marianas spinning disk confocal microscope at 37 °C in the presence of CO₂ for the duration of the experiment. Analysis, image and video preparation was done with the Slidebook software.

Immunoprecipitation and Western Blots

HEK293T cells were transfected with plasmids expressing the indicated proteins. 24 h posttransfection, cells were incubated with MG132 at a final concentration of 20 ȝM for 4 h. For immunoprecipitation of the substrates, cells were lysed 24 h after transfection in Nonidet P-40 lysis buffer (50 mM Tris/HCl, pH 7.5, 150 mM NaCl, 0.5% Nonidet P-40, 50mM NaF) supplemented with 10 mM N-Ethylmaleimide (NEM) and protease inhibitors (Roche Applied Science). Immunoprecipitation was performed on clarified cell lysates with GFP antibody (Santa Cruz sc-9996) overnight at 4 °C and the resulting proteins were analyzed by immunoblotting with anti-GFP and antiHis₆ antibody. Input materials were also checked by immunoblotting using anti-GFP, anti-V5, antiMyc, anti-HA, and anti-GRP170 antibodies (loading control). Immune complexes were isolated with protein A-agarose beads, washed with NP-40 buffer in 2× reducing Laemmli buffer. Whole cell lysates were mixed with 4× reducing Laemmli buffer and analyzed by SDS-PAGE and followed by immunoblotting with the indicated antibodies. Cells were lysed, GFP-cDAXX was pulled down

For pull-down of ubiquitinated proteins, cells were incubated with MG132 or DMSO at 20 μM for 4 hours cells were lysed 24h after transfection in buffer A [6 M guanidine-HCl, 0.1 M Na₂HPO₄/NaH₂PO₄ (pH 8.0) and 10 mM imidazole]. The lysates were sonicated, cleared, and incubated with Ni-NTA sepharose (Qiagen) for 3 hr at room temperature. The beads were washed twice with buffer A, twice with A/T composed of one volume of buffer A and three volumes of buffer T [25 mM Tris (pH 8.0) and 20 mM imidazole], and twice with buffer T. The beads were finally boiled in SDS-PAGE loading buffer containing 100 mM imidazole.

The antibodies used for Western blots were: GFP (1:10,000; Cell Signaling Technologies, 4B10), V5 (1:5,000; Thermo-Fisher, R960), Myc-tag (1:100 Cell Signaling Technologies, 71D10), FLAG-tag (1:1,000; Sigma, M2), GAPDH (1:5,000; Abcam ab9485). HA-tag (1:500), and GRP170 (1:1,000) were a gift from Linda Hendershot. Anti-mouse and anti-rabbit HRP conjugated antibodies (Jackson immunoresearch) were used at 1:20,000

Protein Purification

All His-SUMO-SPOP^28–359 constructs were expressed in BL21 RIPL cells in ZYM-5052 autoinduction media (Studier, 2005), and purified as previously described by Ni column, TEV cleavage during dialysis, Ni-pass-back, and SEC (Marzahn et al., 2016). GST-SPOP^MATH was expressed in BL21 RIPL cells in LB medium, and purified as previously described by Glutathione-sepharose, TEV cleavage during dialysis, ion-exchange by SP column, and SEC (Pierce et al., 2016). Small fractions of each construct were labeled with 4X Oregon Green 488 Carboxylic Acid, Succinimidyl Ester, 5 isomer (Cat #O6147 Thermo Fisher) at 4 ºC overnight. Un-reacted label was removed from the proteins by PD-10 column (Thermo Fisher 45–000-67).

All cDAXX constructs were expressed in BL21 RIPL cells in LB media at 18°C for ~20 hr. Protein expressed for NMR experiments were isotopically labeled with ¹⁵N, ¹³C by growing cells in M9 minimal media supplemented with ¹³C glucose and ¹⁵N ammonium chloride (Cambridge Isotopes). at 37°C until an OD₆₀₀ = 0.8. Expression was then induced with 0.6 mM IPTG for 18 hours at 20°C. Cells were lysed in 50 mM Tris pH 8.0, 500 mM NaCl, 30 mM Imidazole, 2 mM ȕǦME, and Complete protease inhibitor cocktail (Roche) with a microfluidizer at 20,000 psi. The cleared lysate was loaded onto a gravity NiǦNTA column, washed with lysis buffer, and eluted in 50 mM Tris pH 8.0, 50 mM NaCl, 300 mM Imidazole, and 2 mM β-ME. When indicated, the tag was cleaved with 1mg TEV protease/100 mg protein at 4°C overnight. Both cleaved and uncleaved solutions were diluted 2–4 fold, passed over a Q column, and eluted by salt gradient. Peak fractions were then concentrated (Amicon 10 MWCO) and subjected to SEC on a Superdex 200 16/60 column (GE Healthcare) equilibrated with 25 mM Tris pH 7.5, 150 mM NaCl, and 5 mM DTT (or 1 mM T-CEP). Small fractions of each cDAXX construct were labeled with 10X Rhodamine Red C2 maleimide (cat #R6029 from Thermo Fisher) at 4°C overnight. Un-reacted label was removed from the proteins by PD-10 column.

UBA1 was expressed in insect cells as a GST-fusion protein, and purified by glutathione affinity, thrombin cleavage, and ion-exchange, as previously described (Huang et al., 2008; Scott et al., 2016). UBCH7 was expressed in E. coli BL21 Gold (DE3) cells as a GST-fusion protein, and purified by glutathione affinity, thrombin cleavage, ion-exchange, and SEC chromatography, as previously described (Huang et al., 2008; Scott et al., 2016). NEDD8 and UB were expressed in E. coli BL21 Gold (DE3) cells as GST-fusion proteins, and purified by glutathione affinity, thrombin cleavage, glutathione pass-back, and SEC chromatography, as previously described (Scott et al., 2014; Walden et al., 2003). UB was fluorescently labeled with 4X Alexa 647 maleimide (cat # from A20347 Thermo Fisher) for 2 hr at RT, quenched with 10 mM DTT, and passed over a PD-10 column to remove excess unreacted label. Cul3/Rbx1 were co-expressed in E. coli BL21 Gold (DE3) and purified by Ni, and SEC chromatography, as previously described (Small et al., 2010). Neddylation of 12 μM Cul3/Rbx1 was accomplished by incubating with 1 μM UBC12, , 0.1 μM APPBP1-UBA3, and 20 μM NEDD8 with ATP and MgCl₂ as previously described (Duda et al., 2008). A small fraction of N8~Cul3/Rbx1 was fluorescently labeled with 4X Alexa 647 Acid, NHS (Succinimidyl) Ester (cat #A2006 from Thermo Fisher) at RT for ~1.5 hr, and excess unreacted label was removed by PD-10 column.ARIH1 and ARIH1^C375S were expressed in E. coli BL21 Gold (DE3) cells as GST-fusion proteins, and purified by glutathione affinity, TEV cleavage, ion-exchange, and SEC chromatography as previously described (Scott et al., 2016).

His-nAR was expressed in E. coli Rosetta (DE3) cells and purified from inclusion bodies. Cells were lysed by sonication. The insoluble fraction was washed with PBS buffer pH=7.4 containing 1% Triton, and dissolved in 8 M Urea, 20 mM Tris pH 7.8, 500 mM NaCl, and 14 mM β-ME. The solution was then passed over a Ni-HP or gravity column and eluted with the same buffer with 500 mM Imidazole. The tag was cleaved with 1mg TEV protease/100 mg protein at 4°C overnight under dialysis, and removed by passing back over Ni resin in 8 M Urea, 50 mM Tris pH 7.8, 100 mM NaCl, and 1 mM β-ME. The flow-through was then concentrated (Amicon 3 MWCO) and subjected to SEC on a Superdex 200 16/60 column (GE Healthcare) equilibrated with 20 mM Phosphate buffer pH 7.4, 2 mM T-CEP, and 1 mM EDTA. Due to protein stability, protein was immediately used, and microscope samples were imaged within 1 hour of set-up. A small fraction of the protein stock labeled with 10X Rhodamine Red-X Succinimidyl Ester, 5 isomer (cat # R6160 from Thermo Fisher) for 1 hr at 4°C. Un-reacted label was removed from the proteins by PD-10 column.

Microscopy analysis for in vitro LLPS

Samples were prepared by mixing the determined amount of protein, buffer, and ficoll PM 70 (Sigma). Sealed sample chambers containing protein solutions comprised coverslips sandwiching two layers of 3M 300 LSE high-temperature double-sided tape (0.34 mm). For each given cDAXX and/or SPOP concentration, the sample was equilibrated at room temperature and incubated for 4–6 hours. Samples were imaged on a Nikon C2 laser scanning confocal microscope with a 20X (0.8NA) Plan Apo objective. Images and movies were processed with the Nikon NIS Elements software. All images within figures were taken with the same camera settings, unless otherwise noted.

In vitro crosslinking reactions

Samples were prepared at 15 μM SPOP and the indicated concentrations of H-cDAXX in buffer containing 25 mM HEPES pH 7.5, 150 mM NaCl, 5 mM DTT. The amine crosslinker BS³ (bis(sulfosuccinimidyl)suberate, cat #21580 from Fisher Scientific) was added for a final concentration of 0.15 mM in samples from Fig. S2C and 0.3 mM in samples from Fig. S4E. Reactions were incubated at room temperature for 30 min. The reactions were quenched by the addition of 100 mM Tris pH 7.6 and were incubated at room temperature for at least 15 min prior to loading samples onto SDS-PAGE gel.

NMR data collection and analysis

NMR data were acquired on Bruker Avance 600 and 800 MHz spectrometers equipped with TCI triple-resonance cryogenic probes and pulsed-field gradient units at 5 °C. All samples were prepared in an NMR buffer consisting of phosphate-buffered saline (137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄, 1.8 mM KH₂PO₄), 10 mM DTT pH 6.9 and 10% D₂O.

Assignment of cDAXX backbone resonances were done in two steps. Initially, approximately 0.75 mM¹⁵N,¹³C cDAXX samples were prepared and standard 3D assignment experiments based on sensitivity enhanced ¹H,¹⁵N HSQC (8 scans, 2048 × 320 complex data points) were collected. These included a HNCACB and CBCA(CO)NH (16 scans, 1024 (¹H) × 32 (¹⁵N) × 128 (¹³C) complex data points, with 11 ppm, 24 ppm, and 70 ppm as ¹H, ¹⁵N and ¹³C sweep width, respectively), a HN(CA)CO (16 scans, 1024 (¹H) × 32 (¹⁵N) × 75 (¹³C) complex data points, with 11 ppm, 24 ppm, and 18 ppm as ¹H, ¹⁵N and ¹³C sweep widths), a HNCO (16 scans, 1024 (¹H) × 32 (¹⁵N) × 75 (¹³C) complex data points, with 11 ppm, 32 ppm, and 22 ppm as ¹H, ¹⁵N and ¹³C sweep widths), and a HNCA (16 scans, 1024 (¹H) × 32 (¹⁵N) × 95 (¹³C) complex data points, with 16 ppm, 25 ppm, and 30 ppm as ¹H, ¹⁵N and ¹³C sweep widths). Additionally, HN(CA)NNH (16 scans, 1024 (¹H) × 32 (¹⁵N F1) × 60 (¹⁵N F2) complex data points, with 11 ppm, 24 ppm, and 24 ppm as ¹H, ¹⁵N F1 and ¹⁵N F2 sweep widths) provided connectivity between N, N+1 and N-1 facilitating a “backbone walk”(Weisemann et al., 1993). Using these methods, approximately 70% of the sequence was confidently assigned.

Gaps in assignments were filled using carbon-detect experiments. Initial HSQC based resonance assignments were transferred to a ¹³C,¹⁵N 2D carbon-detect CON spectrum using IPAP decoupling (32 scans, 1024 (¹³C) × 256 (¹⁵N) complex data points with 18 ppm and 35 ppm as ¹³C and ¹⁵N sweep widths). Carbon-detect assignments were performed using 3D CANCO (16 scans, 1024 (¹³C) × 32 (¹⁵N) × 48 (¹³C) complex data points, with 20 ppm, 35 ppm, and 35 ppm as ¹³C F3, ¹⁵N and ¹³C F1 sweep widths), COCON (8 scans, 1024 (¹³C) × 40 (¹⁵N) × 32 (¹³C) complex data points, with 18 ppm, 35 ppm, and 12 ppm as ¹³C F3, ¹⁵N and ¹³C F1 sweep widths) and CCCON (8 scans, 1024 (¹³C) × 42 (¹⁵N) × 40 (¹³C) complex data points, with 12 ppm, 35 ppm, and 72 ppm as ¹³C F3, ¹⁵N and ¹³C F1 sweep widths) spectra (Bermel et al., 2005; Bermel et al., 2006a; Bermel et al., 2006b). Overlapping peaks were resolved by using a combination of nitrogen-detect 2D NCO experiments and carbon-detect amino acid specific 2D experiments (Sahu et al., 2014; Takeuchi et al., 2010).

Titration of the SPOP^MATH domain into cDAXX was performed using samples containing 0.4 mM ¹⁵N, ¹³C cDAXX and 0.2 and 0.4 mM SPOP^MATH for 1:0.5 and 1:1 molar ratios, respectively, and 0.2 mM cDAXX and 0.4 mM SPOP^MATH for the 1:2 molar ratio. Carbon detect CON-IPAP spectra were recorded with identical parameters to assignment experiments however the number of scans were increased to 80 and 128 for 0.4 mM and 0.2 mM cDAXX samples respectively.

Data were processed using BRUKER Topspin version 3.4, NMRPipe (v.7.9) (Delaglio et al., 1995) and analyzed using CARA (v.1.8.4) (Keller, 2004). All spectra were referenced directly using DSS for the ¹H dimension;¹³C and ¹⁵N frequencies were referenced indirectly.

Peptide Synthesis and Preparation

Peptides encompassing the SB motifs of cDAXX and the mutated motifs of cDAXX-0sb were synthesized at the Hartwell Center at St. Jude Children’s Research Hospital. Each peptide contained the 5 residue SB/mut motif and 4 residues on either side, with N-terminal acetyl and C-terminal carboxy modifications. Each peptide was solubilized in water, pH-corrected, lyophilized, and resolublized in minimal amounts of water. Stock concentrations were calculated from A₂₀₅ readings using the extinction coefficient predicted by the online “Protein Calculator” (Anthis and Clore, 2013). The fPuc peptide, comprising residues 91–107 from the protein Puckered [amino acid sequence Ac-ENLACDEVTSTTSSSSTǦNH2 (Pierce et al., 2016; Zhuang et al., 2009)] and N-terminally labeled with fluorescein was purchased from GenScript, and solubilized in buffer containing 20 mM Tris pH 7.6, 150 mM NaCl, and 5 mM DTT.

Fluorescence Anisotropy

All DAXX assays were performed in 20 mM Tris pH 7.6, 150 mM NaCl, 5 mM DTT, 0.01% Triton X-100, and 10 mg/mL BSA. nAR FP assays were performed in the same buffer minus BSA. For direct binding measurements, serial dilutions of each SPOP construct, ranging from from 0.006 to 100 ȝM, were prepared on a 384-well plate (Greiner BioOne). Then fluorescently tagged- cDAXX construct, nAR, or fPuc peptide was added for a final concentration of 40 nM into each well. For competition binding measurements, serial dilutions of each peptide were prepared in 384-well plates ranging from ~10 mM to ~ 0.5 μM. MATH domain and fPuc were added to final concentrations of 6 μM and 40 nM, respectively. Anisotropy was measured using a CLARIOstar plate reader (BMG LABTECH).

FRAP measurements

FRAP experiments were performed using a Marianas spinning disk confocal (SDC) imaging system on a Zeiss Axio Observer inverted microscope platform using a Zeiss Plan-Apochromat 63X (1.4 NA) oil objective and Evolve 512 EMCCD camera (Photometrics). For in vitro samples, timelapse images were collected with 200 ms exposure time for 4 to 6 min, every 300 ms for the first min, then every 5 s for the remainder of the time-lapse. Photobleach settings were: 1 ms, 1 repetition, 2050% of 488 channel intensity or 35–80% of 561 channel intensity. For in cell FRAP, time-lapse images were collected every 500 ms for 3.5 to 4 min. Photobleach settings were: 1 ms, 1 repetition, 60% of 488 or 561 channels. Images were analyzed with SlideBook 6 software (3i).

In vitro ubiquitination assays

Singe-turnover assays were conducted to monitor the paths of UB transfer. E2~*UB was prepared by mixing 15 μM UBCH7, 0.6 μM UB E1, and 20 μM *UB in 25 mM HEPES pH 7.5, 100 mM NaCl, 1 mM ATP, 10 mM MgCl₂ at room temperature for 15 minutes. The reaction was quenched by the addition of EDTA to a final concentration of 50 mM. The single-turnover reactions consisted of mixing the E2~*UB thioester conjugate (1.5 μM final concentration) with pre-incubated complexes (30 min – 2 hr at RT) of the indicated combinations of 20 nM ARIH1 or indicated mutants, 1.25 μM N8~Cul3/RBX1, 5 or 15 μM SPOP and 20 or 50 μM H-cDAXX at room temperature in 25 mM HEPES pH 7.5, 150 mM NaCl, 1 mM DTT, in the presence of 4 or 10% w/v ficoll-70 or sucrose. Samples were visualized as for LLPS analysis or loaded on SDS-PAGE gels and analyzed in a Typhon FLA scanner (GE Healthcare).

Co-evolutionary analysis

We obtained a multiple sequence alignment containing SPOP homologues by first building a hidden Markov model of the protein family, based on 4 iterations of jackhmmer (Finn et al., 2015), and extracting the sequences from the Uniprot Uniref100 database (Suzek et al., 2015). We refined the alignment by requiring that all sequences contain all three structural domains (MATH, BTB and BACK) and excluding all positions that contain more than 50% of gaps. We then used the asymmetric plmDCA algorithm (Ekeberg et al., 2013), using default input parameters (including a 90% cutoff in sequence similarity resulting in 2603 sequences), to find pairs of residues with direct correlated mutations along evolution. We used the derived couplings to divide the SPOP sequence into Evolutionary Domains (Granata et al., 2017), i.e. to find groups of residues that evolved together and almost-independently from each other. In this analysis with used the webserver at spectrus.sissa.it/spectrus-evo_webserver with default parameters.

Quantification and Statistical Analysis

Partition Coefficient for cDAXX and cDAXX^0SB

Signal intensities were obtained with the Fiji software. The cellular average intensity for GFPc-DAXX and GFP-cDAXX^0SB in nuclear condensates and in the diffuse nuclear fractions were calculated by determining the signal intensities in 3 regions of interest (ROIs) per cell for condensates and for the diffuse signal. Background was subtracted by using an ROI of the same size in the area outside of the cell. Statistical significance was determined with the paired student t-test. Signal intensities were plotted using the GraphPad Prism software.

GFP-DAXX and V5-SPOP Signal Intensity

Signal intensities for V5 (SPOP) and GFP-DAXX in cells expressing WT or F133V V5-SPOP were obtained with the Fiji software. The nuclear signal intensity for GFP-DAXX and V5-SPOP was measured by using an ROI the size of the whole nucleus to determine the signal intensity for each cell. Background was subtracted by using an ROI of the same size in the area outside of the cell. Statistical significance was determined with the paired student t-test. Signal intensities were plotted using the GraphPad Prism software.

K_D Determination from Anisotropy Measurements

K_D values were obtained by fitting to the following equations, adapted from Roehrl et al,

$$A_{obs}=Q{F}_b{A}_b+\frac{A_f\left(1-F_b\right)}{1-F_b(1-Q)}$$

where A_obs is the observed anisotropy, A_b is the anisotropy of the bound state, A_f is the anisotropy of the free state, Q is the ratio of Intensity_free/Intensity_bound, and F_b is the fraction bound of the fluorescent species, which is given by the following equation for direct binding assays,

$$F_b=\frac{\left(K_D+SPOP+substrate\right)-\sqrt{{\left(K_D+SPOP+substrate\right)}^2-4(SPOP)(substrate)}}{2(substrate)},$$

where substrate is the total concentration of fluorescently labeled cDAXX construct or fPuc peptide, SPOP is the total concentration of SPOP (Roehrl et al., 2004).

For competitive binding assays, F_b is given by the following equations adapted from Roehrl et al,

$$F_b=\frac{2\sqrt{\left(a^2-3b\right)}\cos (\theta /3)-a}{3K_{D,Puc}+2\sqrt{\left(a^2-3b\right)}\cos (\theta /3)-a}$$

where

• a = K_{D, Puc} + K_D + Puc + peptide-MATH,

• b=(Puc-MATH)KD, Puc +(peptide - MATH)K_D + K_D,PucK_D,

• c=-K_{D, Puc} K_DMATH,

• $\theta ={\cos }^{-1}\left[\frac{2a^3+9ab-27c}{2\sqrt{{\left(a^2-3b\right)}^3}}\right],$

and K_D,Puc is the dissociation constant determined from a direct binding assay of fPuc and MATH, Puc is the total concentration of fluorescently labeled fPuc peptide, peptide is the variable concentration of each SB/mut peptide, and MATH is the total concentration of MATH domain. For each FA assay, three independent experiments were performed and fit; the average K_D and standard deviation are reported in Tables S2, S3 and S5.

Quantification of in vitro assemblies

Signal intensities of proteins in assemblies were measured in NIS Elements from images without signal saturation. For quantification of protein concentrations, overlapping intensity-based threshold layers in each channel were applied to select the assemblies and measure the mean intensities within them. A standard curve of the intensity vs fluorescent protein was then used with the ratio of labeled to unlabeled protein in each sample to calculate the concentration of each protein construct in the assemblies.

FRAP Analysis and Fitting

For FRAP analysis, mean fluorescence intensities from ROIs were background-corrected, and corrected for photobleaching due to imaging. Fluorescence intensity versus time graphs were expressed in fractional form normalized by the pre-photobleach intensity (Axelrod et al., 1976) and fitted to equations for single- or double-exponential recovery. See also Table S4.

Quantification of in vitro ubiquitination

Quantification of *UB appearance in assemblies from microscope images was conducted in NIS Elements, overlapping intensity-based threshold layers in the green and red channels were applied to select assemblies. The average intensity in each channel within the assemblies was then measured at each time point, as well as the average intensity of the background (from a location in the image with no assemblies). The intensity within assemblies was determined by plotting the backgroundcorrected blue intensity divided by the background-corrected average of green and red intensities, for each time point. Due to the variability in signal intensity, reactions with increases in the blue channel within assemblies over time were normalized to the intensity of the last point in the WT SPOP + ficoll conditions.

Supplementary Material

3

Video S1. Time lapse video of GFP-DAXX in bodies undergoing fusion, Related to Figure 1. HeLa cells were transfected with GFP-DAXX and their fluorescence monitored in live cells at 37 °C in the presence of CO2. Cells were imaged every minute for 30 minutes. The video was compiled by creating maximum intensity projections of Z-stacks at each time point.

4

Video S2. Time lapse video of GFP-DAXX and SPOP-mCherry in SPOP/DAXX bodies undergoing fusion, Related to Figure 1. HeLa cells were transfected with GFP-DAXX and SPOPmCherry and their fluorescence monitored in live cells at 37 °C in the presence of CO2. Cells were imaged every 6 minutes for 2 hours, and every ten minutes for 1 hour. The video was compiled by creating maximum intensity projections of Z-stacks at each time point.

Highlights.

• Substrates drive phase separation of the tumor suppressor SPOP.

• Phase separation and co-localization of SPOP and substrate depend on multivalency

• Mesoscale SPOP/substrate assemblies mediate enzymatic activity.

• SPOP cancer mutations disrupt phase separation co-localization, and activity.