Shared and divergent phase separation and aggregation properties of brain-expressed ubiquilins

The brain-expressed ubiquilins, UBQLNs 1, 2 and 4, are highly homologous proteins that participate in multiple aspects of protein homeostasis and are implicated in neurodegenerative diseases. Studies have established that UBQLN2 forms liquid-like condensates and accumulates in pathogenic aggregates, much like other proteins linked to neurodegenerative diseases. However, the relative condensate and aggregate formation of the three brain-expressed ubiquilins is unknown. Here we report that the three ubiquilins differ in aggregation propensity, revealed by in-vitro experiments, cellular models, and analysis of human brain tissue. UBQLN4 displays heightened aggregation propensity over the other ubiquilins and, like amyloids, UBQLN4 forms ThioflavinT-positive fibrils in vitro. Measuring fluorescence recovery after photobleaching (FRAP) of puncta in cells, we report that all three ubiquilins undergo liquid–liquid phase transition. UBQLN2 and 4 exhibit slower recovery than UBQLN1, suggesting the condensates formed by these brain-expressed ubiquilins have different compositions and undergo distinct internal rearrangements. We conclude that while all brain-expressed ubiquilins exhibit self-association behavior manifesting as condensates, they follow distinct courses of phase-separation and aggregation. We suggest that this variability among ubiquilins along the continuum from liquid-like to solid informs both the normal ubiquitin-linked functions of ubiquilins and their accumulation and potential contribution to toxicity in neurodegenerative diseases.


Results
Among brain-expressed ubiquilins, UBQLN4 is particularly prone to self-assemble into fibrillar aggregates in vitro. We previously identified the UBA domain as a key driver of UBQLN2 aggregation 13 .
The UBA domains of all three brain-expressed ubiquilins are highly conserved, sharing 93% sequence identity (Fig. 1a) 30 . This similarity led us to investigate whether, like UBQLN2, UBQLN1 and UBQLN4 can also aggregate. Using an amyloid-like structure predictor algorithm 31 , we calculated the aggregation propensity profile of the three constructs used in our in vitro studies (Fig. 1b). These constructs (UBQLN1 438-589 , UBQLN2 430-624 and UBQLN4 444-601 ) extend from the fourth STI1 motif to the end of the protein and contain a 6xHis tag at the amino-terminus (Fig. 1). Whereas the UBA domains have nearly identical aggregation profiles, the fibril formation prediction profiles differ for the three constructs in the region between the fourth STI1 motif and the UBA domain (Fig. 1b). This region (fourth STI1 to UBA) also has the largest sequence variability across the ubiquilin The UBA domains are highly conserved (93% identical) with the major sequence differences residing in the region between STI1-4 and UBA 30,51 . Identical residues are colored in red. (b) Zipper DB was used to predict regions that may drive aggregation of ubiquilin constructs 31 . In the histogram, each bar represents a six-residue-long segment. Segments with Rosetta energies smaller than − 23 kcal/mol (black line) are predicted to form β-sheet structures characteristic of amyloids, and are colored in red. In orange are the segments where Rosetta energies are between − 22 and − 23 kcal/mol, a unit larger than the threshold of − 23 kcal/mol. As with the sequence alignment, the aggregation profiles of these ubiquilin constructs differ mainly in the region between STI1-4 and the UBA domain. The UBQLN2 PXX repeat motif is not shown because none of its segments were predicted to form β-sheet structures. Asterisks designate the gaps in the multiple sequence alignment of the ubiquilins. www.nature.com/scientificreports/ constructs used for the in vitro studies and, in UBQLN2, is further distinguished by its unique proline-rich (PXX) domain. Because the UBA domain is known to be critical for UBQLN2 aggregation and phase separation 13,19 , we anticipated that recombinant UBQLN1 438-589 and UBQLN4 444-601 would similarly aggregate in vitro. To measure aggregation, we used ThioflavinT (ThT), a dye that produces characteristic fluorescence in the presence of β-sheet structures and has been used to study fibril formation of many amyloid proteins 32,33 .
Among the three ubiquilins, UBQLN4 444-601 aggregated most rapidly (Fig. 2). At a 10 µM protein concentration, the UBQLN4 444-601 aggregation reaction was the fastest to transition to a steady-state, characterized by a constant fluorescence over time. Indeed, the time to half transition to steady-state for UBQLN4 444-601 was similar to that of the isolated UBA domain from UBQLN2, which drives aggregation of the full protein 13,19 . In contrast, UBQLN1 438-589 and UBQLN2 430-624 aggregated more slowly and displayed decreased ThT signal (Fig. 2b).
We note that for this study we used a UBA construct (UBA 581-624 ) that is seven residues shorter than UBA 574-624 , which we previously described 13 . UBA 581-624 lacks the thrombin cleavage site and a linker sequence present in the longer UBA 574-624 construct 13 . These differences likely explain the shortened lag phase of UBA 581-624 compared to UBA 574-624 13 .
A previous study revealed that similarly truncated UBQLN2 constructs formed liquid-like condensates in solutions with higher ionic strength than the physiologically relevant PBS buffers 19 . This observation prompted us to compare the liquid-liquid phase separation potential of UBQLN1 438-589 and UBQLN4 444-601 to that of UBQLN2 430-624 . In our liquid-liquid phase separation experiments (Fig. 2e), we opted to use 10 µM as the starting protein concentration, which matched the protein concentrations of our aggregation assays, and screened buffers with increasing salt concentrations. The liquid-liquid phase separation assays, reproduced twice, were carried out with protein samples unlabeled with fluorescent tags. These data show that all ubiquilins liquid-liquid phase separate, but the images of UBQLN2 430-624 samples differed from the ones of UBQLN1 438-589 and UBQLN4 444-601 . We also found that the minimum salt concentration for liquid-liquid phase separation differed among the constructs (≥ 3 M NaCl for UBQLN1 438-589 ; ≥ 1.5 M NaCl for UBQLN2 430-624 , and ≥ 2.5 M NaCl for UBQLN4 444-601 ).
Among tested ubiquilins, UBQLN4 is most prone to partition into the PBS-insoluble fraction of cell lysates. In our in vitro ThT and liquid-liquid phase separation assays we used truncated ubiquilin constructs that included the region of the protein known to drive phase separation and aggregation 13,19 . To determine whether the divergent aggregation and liquid-like behavior of ubiquilins observed in vitro extended to cells, we chose to analyze full-length proteins. We transiently expressed full-length FLAG-tagged UBQLNs 1, 2 and 4 (FLAG-UBQLN1, FLAG-UBQLN2, and FLAG-UBQLN4) in HEK-293 cells and separated cell lysates into PBS-soluble (supernatant) and PBS-insoluble (pellet) fractions. Because our mechanical shearing lysis protocol efficiently releases soluble proteins that normally reside in the nucleus (Fig. 3a), the vast majority of the proteins in the PBS-insoluble fraction are likely to be in an aggregated state as opposed to trapped in a subcellular structure. All three proteins partitioned into both PBS-soluble and insoluble fractions. Compared to FLAG-UBQLN1 and FLAG-UBQLN2, a significantly higher percentage of FLAG-UBQLN4 partitioned into the PBS-insoluble fraction (Fig. 3a,b). Moreover, on denaturing Western blots only FLAG-UBQLN4 was noted to electrophorese as a higher molecular weight, possibly dimeric species, and a high molecular weight (HMW) smear (Fig. 3c). This pattern of relative solubility/insolubility and formation of SDS-resistant HMW species by full-length ubiquilins mirrors the relative aggregation behavior of the C-terminal ubiquilin constructs in vitro.
To further evaluate assembly formation and aggregation of each ubiquilin, we quantified the formation of intracellular puncta by each protein in fixed cells. Assessed by immunofluorescence, FLAG-UBQLN1 primarily distributed diffusely in cells whereas a much greater percentage of FLAG-UBQLN2 and FLAG-UBQLN4 localized to cytoplasmic puncta (Fig. 3d,e). Characterization of puncta characteristics revealed that UBQLN1 puncta were smaller and more heterogeneous in number per cell than the larger puncta formed by both UBQLN2 and UBQLN4 (Fig. 3f,g). Analysis of the circularity of puncta revealed that UBQLN2 puncta were significantly less circular than either UBQLN1 or UBQLN4 puncta, which may be a function of their larger size and therefore physical constraint in the cellular space. To evaluate whether interactions between the three ubiquilins are necessary for the formation of UBQLN puncta in living cells, we assessed the formation of intracellular puncta by each protein in UBQLN1/2/4 triple knockout cells and found that, similar to control cells, all three UBQLNs form puncta in the absence of the other two ubiquilins (Supplemental Fig. S2). Divergent ubiquilin solubility in human brain. To gain insight into the physiological relevance of this divergence in ubiquilin aggregation, we analyzed levels of each brain-expressed ubiquilin in the PBS-soluble versus PBS-insoluble fractions of human brain lysates, assessing both age-matched controls and various neurodegenerative diseases.
Results in human brain samples supported our results in vitro and in cellular models: the percentage of UBQLN4 present in the PBS-insoluble fraction was elevated over that of UBQLN1. The relative levels of UBQLN2 insolubility varied, but was intermediate between UBQLN1 and UBQLN4 (Fig. 4). We assessed cingulate cortex from two synucleinopathies, Parkinson's disease (PD) and dementia with Lewy bodies (DLB), and frontal cortex from the tauopathy, progressive supranuclear palsy (PSP), as well as cortex from age-matched control brains www.nature.com/scientificreports/ (Supplemental Table S1). Comparisons of the soluble and insoluble fractions from both PD and DLB demonstrated that UBQLN2 and UBQLN4 behavior is similar, partitioning into the insoluble fraction to a greater extent than UBQLN1. Similarly, in PSP brain tissue, UBQLN4 insolubility was significantly elevated over UBQLN1 and  www.nature.com/scientificreports/ UBQLN2. The inherent heterogeneity of human disease tissue samples likely explains the sample-to-sample variability in degree of solubility for the three ubiquilins. Notably, UBQLN4 displayed the least variation, consistently showing a high degree of insolubility in all three diseases, as well as in age-matched controls.
No disease-dependent differences in UBQLN4 insolubility were detected in PD (Fig. 4a), DLB (Fig. 4b) or PSP (Fig. 4c) when compared to age-matched control samples. Thus, both in control and disease brain tissue, UBQLN4, and to a lesser extent UBQLN2, preferentially exist in an insoluble form.
Divergence in ubiquilin mobility within condensates assessed by recovery after photobleaching. Using fluorescence recovery after photobleaching (FRAP), we previously showed that UBQLN2 puncta in HEK-293T cells resemble liquid-like condensates and established a link between UBQLN2 condensate forma- Figure 3. In transfected cells, UBQLN4 and UBQLN2 display enhanced aggregation and puncta formation compared to UBQLN1. (a) Of the three tested ubiquilins expressed in HEK-293 cells, FLAG-UBQLN4 is most prone to deposit in the PBS-insoluble (P) fraction versus the PBS-soluble (S) fraction (n = 9). Both cytoplasmic (GAPDH) and nuclear protein (Histone H3) is isolated in the soluble fraction. (b) All three ubiquilins were expressed equally as measured by total FLAG band density (normalized to Ponceau). For measurements of insolubility, values are expressed as percentage of protein that is PBS-insoluble, determined from the calculated total pool of PBS-soluble and PBS-insoluble. Data were analyzed by ANOVA. (c) UBQLN1, 2 and 4-specific antibodies (cropped to present all three antibodies in parallel) show that UBQLN4 partially electrophoreses as detergent-resistant high molecular weight species whereas UBQLNs 1 and 2 do not (uncropped Western blots are shown in Supplemental Figure S1). (d,e) Immunofluorescence of FLAG-UBQLNs 1, 2 and 4 shows that FLAG-UBQLN1 in HEK-293 cells is primarily diffusely expressed in the cytoplasm, whereas a higher percentage of FLAG-UBQLN2 and 4 is sequestered into large, heterogeneous puncta when normalized to corrected total cell fluorescence (CTCF) (n = 20). Characterization of puncta (n = 35-45/group) revealed that UBQLN1 puncta tend to be significantly smaller (f) and more numerous per cell (g) than UBQLN2 or UBQLN4 puncta, and UBQLN2 puncta are significantly less circular than either UBQLN1 or UBQLN4 puncta (h). Data were analyzed by Kruskal-Wallis test. Means and SEMs are displayed. Scale bar = 50 μm. www.nature.com/scientificreports/ tion and aggregation 13 . The differential aggregation propensity of the three ubiquilins prompted us to compare the ubiquilins' condensate mobility in cells. To assess mobility, we used wild-type human ubiquilins fused to eGFP and measured puncta FRAP in transfected cells. To reduce the chance that puncta exchanged contents with neighboring puncta, we examined their surroundings for other puncta using a Z-height scan. Data were recorded only from puncta that did not have immediately neighboring puncta and when the bleached area was 80% or smaller than the total punctum area. We bleached only the center of each punctum and used fixed ROI . UBQLN2 and UBQLN4 are more insoluble than UBQLN1 in human brain tissue. Representative Western blots (uncropped Western blots shown in Supplemental Figure S3) of cingulate cortex from human controls (n = 9), PD (a; n = 6) and DLB (b; n = 15), and mid-frontal gyrus from human controls (n = 4) and PSP (c; n = 12). Results display increased partitioning of UBQLN4 into the insoluble fraction compared to UBQLN1, analyzed by the Kruskal-Wallis test grouping disease and controls together for each analysis. UBQLN2 insolubility is greater than UBQLN1 insolubility in both DLB and PD and respective controls, but significantly lower than UBQLN4 insolubility in PSP and respective controls. Results are expressed as a percentage of PBSinsoluble ubiquilin, determined from the calculated total PBS-insoluble (P) and PBS-soluble (S). No significant differences were seen between disease and controls for PD, DLB or PSP. Both UBQLN1 and UBQLN2 were detected and quantified as a doublet, as previously reported 13,52,53 . Data were analyzed by the Kruskal-Wallis test using the Dunn's post-hoc test. Box-and-whisker plot (median, first and third percentiles, range of quantified bands) is displayed with the scatter plot of the raw data. www.nature.com/scientificreports/ across all puncta, so that fluorescence recovery that did occur would likely take place from nonbleached regions of the same punctum. Consistent with the above results, eGFP-UBQLN1 fluorescence recovery was faster than that of UBQLN2 or UBQLN4 (Fig. 5c). To minimize any effect of puncta drift from the photobleached region (region of interest (ROI)), we chose shorter recording times of 300 s for UBQLN1 puncta, which were the fastest to recover fluorescence signal. Although the mobile fractions for all ubiquilins were similar, the mobile fractions for UBQLN2 and 4 trended lower than for UBQLN1 (Fig. 5d). Recovery times of UBQLN2 and 4 puncta are similar to one other and longer than those of UBQLN1 puncta (Fig. 5c). In addition to FRAP experiments, we recorded examples of puncta fusion by UBQLNs 1, 2, and 4 ( Supplementary Movies 1-7). Supplementary Movie 8 shows an example of puncta fission by UBQLN4. Taken together, these results suggest that UBQLN2 and UBQLN4 display similar self-association in liquid-like condensates, differing from UBQLN1.
Modeling with normal (Gaussian) distributions revealed that the size of puncta formed by all three ubiquilins followed similar area distributions, consisting mainly of two populations (Fig. 5g-i). The first population, comprising 37-43% of puncta, was relatively small in size (mean areas: 0.8 ± 0.6 µm 2 for eGFP-UBQLN1, 0.4 ± 0.3 µm 2 for eGFP-UBQLN2, and 0.9 ± 0.8 µm 2 for eGFP-UBQLN4), and the second population was much larger and distributed over a wider size range (mean areas: 11.2 ± 10.1 µm 2 for eGFP-UBQLN1, 8.4 ± 8.1 µm 2 for eGFP-UBQLN2, and 14.8 ± 10.7 µm 2 for eGFP-UBQLN4). Linear modeling of punctum area in live cells overexpressing ubiquilin proteins showed that larger puncta tend to be less circular (Supplemental Table S2, Fig. 5g-i). Overall, punctum areas of eGFP-UBQLN1 and eGFP-UBQLN4 were larger for both the small and large populations compared to eGFP-UBQLN2. We note that while overall patterns of differences between UBQLN2 and 4 versus UBQLN1 are consistent between multiple systems we have used to analyze the three ubiquilins, the circularity and puncta area of eGFP-tagged proteins differs from that of FLAG-tagged proteins visualized in fixed cells. Differences between the size of the protein tags and/or the dynamic species measured in live-imaging versus in fixed cells may underlie differences. Puncta selected for FRAP all fall within the larger population of condensates ( Fig. 5g-i).

Discussion
All three brain-expressed ubiquilins are linked to neurodegenerative diseases characterized by protein dyshomeostasis [1][2][3]5,29 . Recent studies highlight UBQLN2 propensity to form amyloid-like fibrils and liquidlike condensates 13,19,20 , but the relative aggregation behavior of these three highly homologous proteins has not been investigated. Here we report that the three ubiquilins share the property of liquid-liquid phase separation (Figs. 2e, Fig. 5), yet differ in condensate properties (Fig. 5), propensity to form fibrillar aggregates (Fig. 2), and accumulation in insoluble species in cells and brain tissue (Figs. 3, 4).
Although all recombinantly-expressed ubiquilins phase separate, expanding upon previous study of recombinant UBQLN2 phase separation 19,20 , we observed that the minimum salt concentration for liquid-liquid phase separation differed among the constructs (≥ 3 M NaCl for UBQLN1 438-589 ; ≥ 1.5 M NaCl for UBQLN2 430-624 , and ≥ 2.5 M NaCl for UBQLN4 444-601 ). Because we are using unlabeled proteins in our in vitro assays, we cannot formally conclude that any specific ubiquilin protein has partitioned inside or outside of the liquid droplets. Our results suggest that similar to our studies with mammalian cells, all recombinantly expressed ubiquilins have the capacity to phase separate under a specific set of conditions.
With respect to the shared property of liquid-liquid phase transition, UBQLN1 is most liquid-like and UBQLN4 most aggregation-prone among the three tested ubiquilins in mammalian cellular systems overexpressing full-length proteins (Fig. 5). Previous studies have suggested that liquid droplets formed by aggregation-prone proteins proceed to a less liquid state (from hydrogel to amyloid) at a faster rate 34 . Consistent with our earlier findings with UBQLN2 and studies by others of additional aggregation-prone proteins 13,15,17,35 , the increased recovery times after photobleaching for UBQLN4 puncta (Fig. 5) correlate with its tendency to form high molecular weight aggregates and partition into the insoluble fraction (Figs. 3, 4). Some studies have shown that amyloid fibrils can induce liquid-liquid phase transition, implying that the relationship could be bi-directional 36,37 . The correlation between increased recovery times of UBQLN4 liquid-like droplets and heightened fibril formation suggests that regulation of liquid-liquid phase transition by ubiquilins is linked to aggregation.
The ability of all ubiquilins to form liquid-like droplets suggests it is a fundamental shared characteristic of this class of proteins that relates, in a still undefined manner, to their function in quality control pathways. While the relevance of liquid-like droplet formation to ubiquilin function is not yet understood, studies of other protein quality control factors may provide insight into potential mechanisms. For example, proteasome components form liquid-like droplets that depend on the interaction of polyubiquitin chains with the UBA-UBL ubiquitin-proteasome shuttle factor, Rad23B 38 . Thus, ubiquilins may form concentrated molecular assemblies via their UBL and UBA domains to regulate multiple cellular processes.
Despite being highly similar, the brain-expressed ubiquilins differ in condensate and aggregation behavior. These differences suggest that relatively minor sequence changes can alter the protein's behavior. In this light, one can envision that missense mutations that cause UBQLN2-mediated disease 1 , or possibly UBQLN4-mediated disease 5 , would change these proteins' behavior along the liquid-like droplet-hydrogel-aggregate spectrum. Published evidence from our group and others supports this view 13,19,20 . While the ubiquilins' UBL and UBA domains are nearly identical, the three brain-expressed UBQLNs are distinguished by the middle region located between the UBL and UBA domains 30 . In evaluating the C-terminal regions of UBQLNs 1, 2 and 4 for fibril formation, we demonstrated that UBQLN4 444-601 was the most aggregation-prone, while UBQLN1 438-589 and UBQLN2 430-624 displayed slower aggregation kinetics. These results suggest that C-terminal differences among brain-expressed ubiquilins lead to altered aggregation behavior that corresponds to the differences observed for the full-length proteins. www.nature.com/scientificreports/ www.nature.com/scientificreports/ Although UBQLN4 exhibited the greatest propensity to aggregate in in vitro assays, cellular models, and human brain tissue, UBQLN2 also displayed heightened aggregation propensity in cellular models and human brain when compared to UBQLN1. Since UBQLN2 is known to accumulate in aggregate-like structures in various neurodegenerative diseases 1, 3,29,39 , it will be important now to determine whether UBQLN4 similarly accumulates in one or more neurodegenerative diseases.
Liquid-like condensates are complex structures with antagonistic behaviors. While many important cellular functions are carried out in liquid-like condensates, phase separation has been implicated in abnormal protein aggregation and neurodegeneration [22][23][24][25] . Interestingly, it has been observed that UBQLN2 can both protect and facilitate abnormal protein aggregation in the context of liquid-liquid phase separation. Indeed, wild-type UBQLN2 protects against pathogenic protein aggregation in FUS-related ALS/FTD by decreasing stress granule formation 26 . In contrast, liquid-like condensates of UBQLN2 may facilitate TDP43 aggregation in ALS/FTD 21 . Although UBQLN4's involvement in protein aggregation is yet to be determined, it is possible that UBQLN4 liquid-like condensates, like UBQLN2, are sites that initiate and facilitate abnormal protein association.
Our results demonstrating that UBQLN4 insolubility is elevated over UBQLN1-and in some cases over UBQLN2-in human brain suggest that heightened UBQLN4 aggregation behavior may be an intrinsic feature of the protein that informs its function. Furthermore, the percent UBQLN4 in the insoluble fraction showed little variability between samples despite the inherent wide variation present when using human tissue. While no differences in UBQLN4 solubility were detected in PD, DLB or PSP samples compared to age-matched controls, our limited analysis of available human samples does not allow us to conclude that there are no disease-dependent changes. Further studies are needed to elucidate whether UBQLN4 aggregation status relates to specific diseases or more generally to brain aging.
Although the relationship between dynamic molecular assembly formation and ubiquilin function is still uncertain, we anticipate that the phase separation and aggregation behavior of all three brain-expressed ubiquilins will prove to be regulated by ubiquitin and will inform their function as proteostasis regulatory factors 19 . Our results reveal two populations of condensates formed by all three ubiquilins, one being uniformly small and the second being larger and more variable in size. Small condensates conceivably represent functional assemblies, whereas larger condensates do not. Studies have also shown that the three ubiquilins interact with one another, indicating that their cellular functions and liquid-liquid phase transition properties may be linked 40 , though we show that all three ubiquilins form puncta in cells absent the presence of inter-ubiquilin interactions. Further research will be needed to better understand how the shared and divergent condensate and aggregation properties of ubiquilins influence their cellular functions and roles in disease. Protein expression and purification. All plasmids for bacterial expression contained a 6xHis tag and a tryptophan (Trp) at the N-terminus. UBQLN1 438-589 , UBQLN2 430-624 and UBQLN4 444-601 constructs were transformed in Rosetta (DE3) Escherichia coli bacteria. All LB agar plates and LB media were supplemented with 100 μg/mL carbenicillin and 34 μg/mL chloramphenicol. Transformed cells were grown overnight at 37 °C on LB agar media plates. The following day, cells were transferred to 100 mL LB starter cultures and allowed to grow for 60 min, then upscaled into 1 L LB media. At OD601 ≈ 0.6-0.8, cells were induced with 0.5 mM isopropyl β-d-1thio-galacto-pyranoside (IPTG) and collected after 4 additional hours of incubation. Bacteria were collected by centrifugation for 10 min at 10,322×g. Bacterial pellets were stored at − 80 °C.

Methods
For protein purification, pellets of bacteria grown in 1 L of LB media were resuspended in 25 mL of pre-chilled lysis buffer containing 2% glycerol, 1 mM EDTA, 25 mM Na phosphate, pH 7.4, EDTA-free cOmplete protease inhibitor cocktail tablet (Roche; 1 tablet per 10 mL lysis buffer) and 6 μL/mL of saturated phenylmethylsulfonyl fluoride (PMSF). Bacteria were then lysed using EmulsiFlex B-15 high pressure homogenizer (Avestin). The lysates were centrifuged at 31,000×g for 20 min and the supernatant was added to Ni-NTA agarose slurry, After incubation for 15 min at 4 °C, the slurry was washed with 25 mM Na phosphate pH 7.4 containing 0.1 M NaCl, 6 μL/mL from saturated PMSF 2% glycerol, and 20 mM imidazole. Proteins were eluted with 25 mM Na phosphate pH 7.4, which was supplemented with 200 mM imidazole 0.1 M NaCl, 6 μL/mL from saturated PMSF, and 2% glycerol. In the case of the UBQLN2 430-624 construct, the eluted protein was diluted with two volumes of 25 mM Na phosphate pH 7.4 containing 2% glycerol. Then it was filtered through 50 kDa concentrator (Amicon Ultra-15 Centrifugal Filter Unit, Millipore Sigma) and the flow through was collected for dialysis. Proteins were dialyzed against 5 mM Na phosphate pH 7.4 with two buffer exchanges overnight. Immediately after dialysis, proteins were frozen in liquid nitrogen and stored at − 80 °C. Protein concentration was determined by Pierce BCA protein assay Kit (cat. # 23225, ThermoFisher Scientific). Transmission electron microscopy (TEM). Negatively stained specimens for TEM were prepared by applying 5 μL of protein (UBQLN1 438-589 , UBQLN2 430-624 and UBQLN4 444-601 ) sample to hydrophilic 400 mesh carbon-coated Formvar support films mounted on copper grids (Ted Pella, 01702-F). Samples were allowed to adhere for 4 min, rinsed twice with distilled water, and stained for 60-90 s with 5 μL of 1% uranyl acetate (Ted Pella, Inc.). All samples were imaged at an accelerating voltage of 80 kV in a JEOL JSM 1400 Plus (JOEL). Human disease brain tissue. Frozen brain tissue from the cingulate cortex was obtained from subjects with PD, DLB, and age-matched control subjects as well as mid-frontal cortex from PSP and age-matched control subjects (Supplemental Table S1) from the Michigan Brain Bank (University of Michigan, Ann Arbor, MI, USA). Brain tissue was collected with the informed consent of the patients. Protocols were approved by the Institutional Review Board of the University of Michigan and abide by the Declaration of Helsinki principles. Samples were examined at autopsy by neuropathologists for diagnosis.
To quantify the percent insoluble protein for each sample, normalized insoluble ubiquilin band densities were corrected for ubiquilin concentration based on lysate volume, and then divided by total (soluble + insoluble) normalized ubiquilin protein levels and multiplied by 100. Two-way ANOVA revealed UBQLN-dependent differences, but no differences based on disease state. Therefore, control and disease samples were pooled for oneway ANOVA analyses of differences in UBQLN insolubility. Analyses were completed in triplicate and analyzed by the Kruskal-Wallis test with the Dunn's post-hoc test. Data were analyzed using Statview (SAS Institute).
Immunofluorescence. Fixed cells were washed in PBS, permeabilized with 0.5% Triton-X 100 and blocked in 5% goat serum for 1 h. Cells were incubated in anti-FLAG, clone M2 (Sigma cat. # F3165, 1:100) overnight at 4 ο C. The following day, cells were washed in PBS three times for 10 min each and incubated with goat antimouse IgG Alexa-568 (Invitrogen cat. # A-11004; 1:500) for 1 h. Sections were then washed in PBS three times for 10 min each and incubated with DAPI (Sigma) to label nuclei for 5 min at room temperature, washed three times for 5 min each, and were mounted with Prolong Gold Antifade Reagent (Invitrogen). Slides were imaged using an IX71 Olympus inverted microscope. Images were analyzed with the Analyze Particles tool in Image-J (National Institute of Health) 43 to determine the corrected total cellular fluorescence (CTCF; used as a measure of overall expression of each ubiquilin protein per cell for normalization), the corrected total fluorescence for each punctum (CTFP), circularity and area of each punctum and total number of puncta per cell. To calculate the percent fluorescence of puncta, the sum of CTFP of all puncta in the cell was divided by CTCF of the whole cell. Immunofluorescence experiments were repeated in triplicate and statistical analyses utilized means from the three replicates. Analyses were completed by the Kruskal-Wallis test and Dunn's post-hoc test. Data were analyzed using Statview (SAS Institute).

Fluorescence recovery after photobleaching (FRAP). HEK293-T cells (Batch # 70008735, ATCC)
were plated on Nunc Lab-Tek II Chambered Coverglass (cat. # 155409, Fisher) in DMEM, supplemented with 10% FBS, 10 mM Glutamine and 100 U/mL penicillin/streptomycin. The next day cells were transfected with eGFP-UBQLN1, eGFP-UBQLN2, and eGFP-UBQLN4 plasmid DNA. Using Lipofectamine 2000 Transfection Reagent (cat. # 11668027; Fisher) according to the manufacturer's instructions. Cells were imaged 24 h after transfection with a Nikon A-1 confocal microscope (40 × WI Lens) using Nikon Elements software with perfect focus engaged. For FRAP data collection, an area of 512 pixels was scanned using speed 518 frames/sec. FRAP imaging consisted of three phases: pre-bleach, bleaching, and post-bleach imaging. During pre-bleach imaging, each punctum was imaged every 2 s. In the bleach phase, a 5-6 μm 2 ROI was drawn in the middle of the punctum covering no more than 2/3. This ROI defined the stimulation area for a 488 nm laser with 20% power. The post-bleach phase consisted of two periods. For the first minute, images were acquired every 5 s, while for the subsequent 6 min, images were acquired every 10 s. This was done to provide greater temporal resolution during the rising phase of fluorescence recovery. FRAP analysis was performed in Fiji 44 . To fix granules in place, stack registration (Rigid Body) was performed. Following thresholding, one ROI was generated that corresponded to the pre-bleach signal and another to the post-bleach signal. A mask was created from the pre-bleach ROI, www.nature.com/scientificreports/ inverted, and the post-bleach ROI was subtracted from this image. This created a region of bleached signal deemed the FRAP ROI. At each time point the integrated density of the FRAP ROI was divided by the integrated density of the pre-bleach ROI. These values were then normalized such that the mean of the five pre-bleach values was set to 1, and the first postbleach value was set to 0. To calculate the mobile fraction and half time, the normalized fluorescent signal measured from each granule was fitted to the equation A * (1 − exp(− t/τ)), where A is the mobile fraction, t is the collection time and τ is the half time 45 . Puncta area, circularity, roundedness, aspect ratio. and solidity were calculated using Fiji 44 . The area distribution of UBQLN1, 2, and 4 puncta was modeled using Normal (μ, σ2) distribution. The process of distribution model-fitting involved maximum likelihood estimation of the model distribution parameters 46 . We employed the R packages mixtools 47 and ggplot2 48 to obtain and plot the desired models. The model parameters and the mixture distribution weights are used as proxy measures to discriminate between different puncta areas. For statistical analysis of FRAP, Kruskal-Wallis (KW) test was used, which is an alternative when parametric assumptions are not guaranteed 49,50 . FRAP data are shown as means ± SEM.