Detection of TAR DNA-binding protein 43 (TDP-43) oligomers as initial intermediate species during aggregate formation

Aggregates of the RNA-binding protein TDP-43 (TAR DNA-binding protein) are a hallmark of the overlapping neurodegenerative disorders amyotrophic lateral sclerosis (ALS) and frontotemporal dementia. The process of TDP-43 aggregation remains poorly understood, and whether it includes formation of intermediate complexes is unknown. Here, we analyzed aggregates derived from purified TDP-43 under semidenaturing conditions, identifying distinct oligomeric complexes at the initial time points before the formation of large aggregates. We found that this early oligomerization stage is primarily driven by TDP-43's RNA-binding region. Specific binding to GU-rich RNA strongly inhibited both TDP-43 oligomerization and aggregation, suggesting that RNA interactions are critical for maintaining TDP-43 solubility. Moreover, we analyzed TDP-43 liquid–liquid phase separation and detected similar detergent-resistant oligomers upon maturation of liquid droplets into solid-like fibrils. These results strongly suggest that the oligomers form during the early steps of TDP-43 misfolding. Importantly, the ALS-linked TDP-43 mutations A315T and M337V significantly accelerate aggregation, rapidly decreasing the monomeric population and shortening the oligomeric phase. We also show that aggregates generated from purified TDP-43 seed intracellular aggregation detected by established TDP-43 pathology markers. Remarkably, cytoplasmic aggregate seeding was detected earlier for the A315T and M337V variants and was 50% more widespread than for WT TDP-43 aggregates. We provide evidence for an initial step of TDP-43 self-assembly into intermediate oligomeric complexes, whereby these complexes may provide a scaffold for aggregation. This process is altered by ALS-linked mutations, underscoring the role of perturbations in TDP-43 homeostasis in protein aggregation and ALS-FTD pathogenesis.

TDP-43 pathology is a hallmark of amyotrophic lateral sclerosis (ALS) 3 and frontotemporal dementia (FTD). TDP-43 aggregates accumulate in ϳ98% of ALS and 50% of FTD cases, also defined as ubiquitin-positive frontotemporal lobar degeneration (FTLD-U or FTLD-TDP) (1,2). In addition, TDP-43 pathology is found in ϳ50% of Alzheimer's disease (3,4). The direct role of TDP-43 in disease is underscored by greater than 40 ALS-associated dominant missense mutations in the TDP-43 gene (TARDBP) (5). These cause ϳ3-5 and 1% of familial and sporadic ALS, respectively. The clinicopathological characteristics of ALS and FTD associated with mutant and WT TARDBP are largely indistinguishable, and the mechanisms affected by the mutations linked to pathogenesis have not been clearly established. Whether disease results from gain of toxic properties through aggregation, from sequestration of functional TDP-43 into aggregates (1), or from a combination of both, it is increasingly evident that loss of TDP-43 homeostasis and aggregation play a critical role in pathogenesis.
TDP-43 is a highly conserved RNA-binding protein and, like other heterogeneous nuclear ribonucleoproteins (hnRNPs), is composed of modular domains that mediate single-stranded RNA/DNA binding and protein interactions (6 -8). Of the two canonical RNA recognition motifs (RRMs), RRM1 contributes to the high affinity for RNA/DNA and GU-rich RNA specificity (6,7). RRM2 is also highly evolutionarily conserved; however, its function remains unclear. An additional folded domain is at the N terminus, which mediates self-assembly as an isolated domain and presumably of the full-length protein (9 -11). The C-terminal domain (CTD) is intrinsically disordered and is a typical low sequence complexity domain, which is highly represented in RNA-binding proteins (12,13). This domain mediates self-assembly and interactions with hnRNP complexes important for RNA processing activity (8,14,15), but at the same time, the CTD drives protein aggregation and toxicity (16 -18). The CTD is characterized by an abundance of glutamine/asparagine residues, showing great similarity to prion domains in yeast proteins, such as that of the archetypal prion protein Sup35 (13,19). Significantly, almost all disease-associated TDP-43 mutations cluster in the CTD (5,20), strongly suggesting that these substitutions disrupt normal protein interactions and promote aggregate formation, driving the disease state.
The central mechanism in TDP-43 self-assembly and aggregation has been largely unexplored. TDP-43 aggregation assays using the full-length protein are encumbered by the extreme aggregation-prone characteristic of TDP-43, which makes production of pure soluble protein particularly challenging. Having recently established methods to generate soluble recombinant TDP-43 (21), we studied its aggregation to identify the factors that mediate and alter this process (e.g. ALS-associated mutations) and to gain insight into the structure of aggregates. We found that TDP-43 aggregates are formed through a biphasic process that initiates with oligomerization followed by aggregation into high-molecular-weight polymers. ALS-linked mutants potently affect aggregation by increasing the rate of assembly. In addition, we show that the aggregates derived from purified TDP-43 are capable of seeding intracellular aggregation following uptake. Our results support a model in which TDP-43 undergoes self-assembly into oligomeric complexes upon misfolding that act as templates for large aggregates. This process may be altered in disease conditions, such as in the presence of patient-linked mutations.

TDP-43 oligomers assemble at the initial aggregation stage followed by high molecular weight aggregates
We have successfully developed methods to generate fulllength bacterial recombinant TDP-43 (rTDP-43) to characterize TDP-43 interactions (21) (Fig. S1A). We expressed and purified homogeneous and soluble TDP-43 N-terminally fused to SUMO (21), which is cleaved off by the SUMO deconjugating protease Ulp1 (Fig. S1B). Using this protein, we analyzed TDP-43 self-assembly and aggregation. Prior to carrying out the aggregation assays, we ensured that purified protein preparations did not include preformed aggregates using highspeed ultracentrifugation and by routinely measuring protein activity in RNA-binding assays (Fig. S1C). The binding affinity for the TDP-43-A(GU) 6 RNA interaction was determined with our previously established fluorescence-based method (21). This accurately measures the apparent dissociation constant for A(GU) 6 (typically K d, app ϭ 2.3 Ϯ 0.7 nM) and also provides an estimate of the active protein concentration. By measuring these parameters and assuming that active protein concentration is equal to the concentration of soluble protein, we determined the purity of our protein preparations. Our aggregation assays consisted of analyzing rTDP-43 complexes after brief shaking followed by incubation at room temperature (ϳ22°C). To detect the assembled complexes, we adapted a method widely used for the characterization of yeast prion aggregates (22)(23)(24), based on the resolution of soluble and large aggregate complexes by semidenaturing detergent-agarose electrophoresis (SDD-AGE). SDD-AGE generates large pores allowing the detection of large protein complexes that are resistant to SDS detergent, typically ranging from 0.1 to 2% (24). This method has also been used to detect neurodegeneration-associated aggregates, such as Tau (25), and in a limited number of studies to analyze TDP-43 pathology from patient tissue and cultured cells (26,27). We analyzed rTDP-43 aggregates formed over time and obtained the resolution of rTDP-43 monomers and large polymers, which increased in size (Fig. 1A). In addition to the expected high-molecular-weight aggregates, we observed an oligomeric pattern of assembly at initial time points. These initial complexes (Figs. 1A, days 1-4, arrows) were replaced by larger, heterogeneous species at longer incubation times (Fig.  1A, days 8 -15). The later-day large complexes resembled aggregates formed by prion proteins and other protein aggregates analyzed with this method (22,25,28). Addition of polyethylene glycol, used as a crowding agent, accelerated the aggregation rate, increasing the formation of the larger complexes (Fig. S2A). Fig. 1 shows complexes formed by SUMOtagged TDP-43, and removal of SUMO had no significant effect in the size of the aggregates or in the formation of oligomers (Fig. S2B). In the absence of SUMO, the assays were initiated immediately after removal of the tag because the cleaved Figure 1. Detection of TDP-43 oligomers during the initial steps of aggregation. SDD-AGE and immunoblotting analysis to detect TDP-43 aggregates using purified protein, cells treated with proteotoxic stress, and FTD patient samples. A, aggregation of full-length purified TDP-43 (rTDP-43) was triggered by brief shaking (2 M in reaction buffer), and samples were analyzed as a function of time. Day 0 was collected immediately after triggering aggregation by shaking. Arrows point to the initial oligomeric species. B, lysates from HEK293 cells treated with sodium arsenite, the ubiquitin proteasome system inhibitor MG132 (10 M, 4 h), heat shock (43°C for 30 min), and nontreated control. Either 300 or 100 g of total protein was loaded. C, the sarkosylinsoluble fraction from FTD and control (lane C) brain samples analyzed by SDD-AGE. These were compared with day 0 and day 10 rTDP-43 aggregates prepared in the same buffer conditions as the brain samples. All blots were probed with TDP-43 antibody and are representative of more than three individual experiments.

Oligomeric species are intermediates in TDP-43 aggregation
TDP-43 could not be stored without significant protein loss. Therefore, we used SUMO-tagged protein in our experiments, unless noted.
We compared aggregates from purified TDP-43 to cell-derived aggregates and insoluble TDP-43 fractions from FTD brain tissue samples to validate our methods for aggregate characterization and to assess the physiological relevance of the rTDP-43complexes detected using SDD-AGE. Lysates from human embryonic kidney (HEK293) cells under conditions known to promote TDP-43 aggregation, oxidative stress treatment with sodium arsenite (29), showed high-molecularweight complexes similar to the later-day rTDP-43 aggregates (Fig. 1B). They were absent in nontreated control cell lysates. Milder proteotoxic conditions, such as inhibition of the ubiquitin proteasome system with 10 M MG132 or 30 min of heat shock (21,30,31), increased the presence of a distinct highmolecular-weight complex migrating above monomeric TDP-43. The physiological oligomerization of TDP-43 into higher order complexes in cells, ranging in size from monomers to much larger macromolecular assemblies has been previously reported (9,27). In our analysis, TDP-43 from control-treated lysates migrated as monomers, indicating that these complexes were reversible and destroyed in the presence of detergent. On the other hand, distinct higher order complexes detected by SDD-AGE upon treatment with MG132 and heat shock suggested that misfolding under these conditions promoted strong interactions resistant to semidenaturing agents. Based on these findings, we propose that the early-day oligomers during the initial rTDP-43 aggregation stage are the counterparts of misfolded complexes in cells. These species may be detected under specific conditions, resulting in loss of proteostasis and perhaps even in physiological complexes, such as the recently described TDP-43 myo-granules found in skeletal muscle (32). Next, we analyzed postmortem human brain tissue from control and FTD cases following sequential extraction of gray matter dissected from frontal cortex samples (Fig. 1C). We used the sarkosyl insoluble fraction for our analyses as described under "Experimental procedures" (33). Our results showed that the levels of TDP-43 aggregates in FTD samples increased compared with control, consistent with a previous report (26). In addition, the migration pattern of TDP-43 in control tissue was different from the disease cases. We included 0-and 10-day rTDP-43 aggregates in the same solution conditions as brain tissue for comparison (Fig. 1C). This showed that migration of later-day rTDP-43 aggregates was similar to TDP-43 complexes from FTD but not from control tissue. Collectively, these results suggest that our methods may be used to detect and analyze aggregate species generated from the purified protein, which show conformational and biochemical similarities to cell and patient-derived TDP-43 inclusions.
We investigated whether TDP-43 complexes could be detected without triggering aggregation by shaking and found that 10 min of incubation of soluble protein at increasing temperatures enhanced the formation of oligomers as seen by SDD-AGE ( Fig. 2A). The left panel of Fig. 2A shows aggregates formed at 0, 3, 5, and 10 days after shaking, for comparison. The TDP-43 complexes, which increase at higher temperatures, are similar to the intermediate species in the aggregation assay. To estimate the oligomeric state of the early TDP-43 complexes, we performed cross-linking experiments under reducing conditions (Fig. 2B). These complexes were analyzed by standard denaturing SDS-PAGE after the addition of ␤-ME (100 mM). As control, non-cross-linked samples were analyzed in the presence and absence of ␤-ME. Immunoblots showed cross-linked complexes corresponding to tetramers at day 3, according to the estimated molecular weight. This complex was not detected in non-cross-linked samples independent of ␤-ME treatment. Thus, we propose that initial rTDP-43 self-assembly favors tetrameric complexes formed through interactions not mediated by disulfide bonds. . B, days 0 and 3 TDP-43 complexes were chemically cross-linked (X-link) in the presence and absence of 250 mM ␤-ME). Samples were analyzed by SDS-PAGE and immunoblot. The estimated number of TDP-43 molecules cross-linked, and the molecular mass markers (kDa) are shown. C, SDD-AGE analysis of rTDP-43 aggregates under reducing conditions, TCEP was added throughout the purification and supplemented at day 5 (reducing). Arrows point to the time of addition of extra TCEP. In nonreducing conditions, TDP-43 was purified in the presence of the short-lived reducing agent ␤-ME, and no additional reducing agents were added during the assay. D, day 3, 7, and 14 aggregates prepared in the presence of TCEP and treated with high concentrations of reducing agents (250 mM ␤-ME, DTT, and TCEP). A longer gel was used to increase oligomer separation. All blots were probed with TDP-43 antibody and are representative of more than three individual experiments.

Oligomeric species are intermediates in TDP-43 aggregation
Thioflavin reactivity with TDP-43 aggregates, including samples derived from patient brain, remains controversial (34 -37). The fluorescence of this compound increases upon binding to amyloid structures that accumulate cross-␤ sheets. We tested binding of rTDP-43 and cellular aggregates to thioflavin T (ThioT) and S (ThioS). ThioT was incubated with rTDP43 monomer for 2 h at room temperature to allow formation of early TDP43 aggregates. We did not observe any significant binding of ThioT to early rTDP43 aggregates under low-salt aggregate-forming conditions. In contrast, the same concentration of ␣-synuclein fibrils showed a concentration-dependent increase in ThioT fluorescence signal, consistent with previous findings (38) (Fig. S3A). Similar results were obtained upon ThioT incubation with 2-and 10-day rTDP-43 aggregates (data not shown). In addition, cellular aggregates, positive for the pSer-409/410 marker of TDP-43 pathology, were not detected by ThioT or ThioS (Fig. S3B). These results are in agreement with previous data showing the lack of ThioT binding to rTDP-43, especially when compared with established amyloid aggregates, such as A-␤ and Sup35 (16,37). Recently, Vogler et al. (32) reported ThioT reactivity with TDP-43 complexes in myogranules from the muscle of a Vcp mutant mouse model of multisystem proteinopathy and inclusion body myopathy, which is characterized by TDP-43 aggregation. The discrepancy in the reports from various studies may be caused by differences in TDP-43 complex/aggregate structures formed under the various conditions. Some isoforms may adopt cross-␤ sheet structure, which may be absent or buried in others. Consistent with this idea, isolated C-terminal peptides show ThioT/S binding upon fibrillization (27, 39 -42).

Early TDP-43 complexes are not mediated by disulfide bonds
To further understand whether the TDP-43 complexes detected by SDD-AGE were mediated by covalent interactions, we analyzed their sensitivity to reducing agents. The reducing agent Tris(2-carboxyethyl)phosphine (TCEP) was used throughout rTDP-43 purification and during the aggregation assays. In Fig. 2C we added supplementary TCEP on day 5 to ensure that the samples remained under reducing conditions. This was compared with a "nonreducing" assay, whereby aggregates were generated using rTDP-43 purified in the presence of ␤-ME, a less stable reducing agent compared with TCEP. No additional reducing agent was used in the aggregation assay (nonreducing; Fig. 2C). Nonreducing conditions generated large heterogeneous aggregates immediately, beginning at the initial time point (day 0). From these experiments we observed that the initial and intermediate aggregates were detected in the presence of TCEP; however, the larger aggregates decreased significantly under these conditions (i.e. day 10 in reducing and nonreducing samples). To further confirm our results, we treated aggregates formed at days 3, 7, and 14 with high concentrations of reducing agents (TCEP, ␤-ME, and DTT, 250 mM) (Fig. 2D). This did not significantly affect the initial complexes but greatly decreased formation of the largest aggregates at later time points. Based on these collective observations we propose that initial complexes that are detected as ordered molecular assemblies by SDD-AGE are partially detergent-resistant oligomers and are not mediated by covalent bonds. On the other hand, the higher-molecular-weight heterogeneous aggregates form through disulfide interactions. This is consistent with the effect of oxidative stress in promoting TDP-43 aggregation in cells and with the previous identification of intra and intermolecular TDP-43 Cys interactions in patient tissue (43). Our results and this previous work highlight the relevance of disulfide bond formation on TDP-43 pathology.

Role of liquid-liquid phase transitions in oligomer formation
RNA-binding proteins associated with ALS/FTD, including TDP-43, FUS, TIA-1, Matrin 3, and hnRNP A1 undergo a process of condensation through liquid-liquid phase separation (LLPS) according to in vitro and in cell-based evidence (44). This activity mediates the formation of membraneless RNAprotein rich organelles in the nuclear and cytoplasmic cellular compartments (e.g. nucleoli and stress granules) (12,(45)(46)(47). Liquid droplets assembled through LLPS "mature" over time into solid-like complexes such as fibrils. Importantly, diseaselinked conditions are proposed to disrupt LLPS and RNA granule assembly, thereby promoting aggregation and neurotoxicity in ALS and FTD (48 -50). Therefore, self-assembly of these ALS and FTD-associated proteins is central to their physiological function but also plays a major role in aggregate formation. The mechanisms connecting these processes of LLPS and aggregation remain poorly characterized.
We asked whether the oligomeric complexes seen at the initial time points in the TDP-43 aggregation assay represent assemblies formed by LLPS. Microscope visualization of fluorescently labeled rTDP-43 in low salt (25-50 mM NaCl) showed immediate phase separation and formation of droplets similar to those recently reported by Maharana et al. (54) (Fig. 3A). Droplet formation was salt-dependent because the number of

Oligomeric species are intermediates in TDP-43 aggregation
droplets was dramatically reduced at higher salt concentrations (250 mM), consistent with previous studies (Fig. 3B) (11,47). Furthermore, phase separation was reversed by increasing the salt concentration to 250 mM, indicating that the observed complexes were not irreversible aggregates (data not shown). Interestingly, extending the incubation time of the sample beyond 30 min resulted in the growth of fibrillar structures as seen by fluorescence microscopy (Fig. 3A). We analyzed samples corresponding to those used for droplet formation at low and high salt and at longer incubation times by SDD-AGE (Fig.  3C). For control, we loaded samples from the aggregation assay days 0 and 10 (Figs. 1A and 2A). We found no macromolecular complexes by SDD-AGE in droplet conditions (25-50 mM NaCl) or at high salt (250 mM). However, incubation of the sample for 30 min to 2 h, parallel to the conditions that resulted in liquid droplet maturation, showed an increase in the oligomeric species similar to the pattern detected at the initial time points of the aggregation assay (Fig. 3B). These observations suggest that maturation of droplets into assemblies with solidlike properties may be detected by SDD-AGE and provide further evidence that the detergent-resistant oligomers form as the initial steps of misfolding and aggregation.

Specific binding to RNA rich in GU repeats decreases TDP-43 oligomer formation and aggregation
RNA is one of the principal physiological partners of TDP-43 mediating protein binding to hundreds of RNA transcripts. Genome-wide studies have shown a strong preference of TDP-43 for GU-rich sequences, an interaction of tight binding affinity (21,51,52). We tested whether RNA modulates the aggregation of rTDP-43 using our assay and focused on the effect of GU repeats. For these experiments we used 2Ј O-methyl, phosphorothioate-modified RNA to reduce its degradation during the assay, at 2-fold molar excess of RNA to protein. We found that addition of (GU) 6 repeats strongly inhibited both the initial oligomerization and large aggregate formation of rTDP-43, compared with control (Fig. 4A). Significantly, (GU) 6 maintained higher levels of the monomer throughout the experiment. This was not observed when the protein was incubated with RNA composed of (CA) 6 repeats.
To further test the role of specific RNA-binding interactions, we carried out similar aggregation assays with the RNA binding-deficient mutant F147L/F149L. Phe-147 and Phe-149 in the RNP1 motif of RRM1 make specific contacts with U and G bases, respectively (53), and double substitution of these amino acid residues (F147L/F149L) drastically decreases RNA binding (6,21). We found that F147L/F149L oligomerization and aggregation resembles WT rTDP-43 but is not significantly affected by (GU) 6 RNA (Fig. 4B). Together, these experiments provide strong evidence that specific RNA binding stabilizes TDP-43 and prevents aggregation. This suggests that TDP-43 binding to target RNA transcripts, which are predominantly GU-rich sequences plays a major role in maintaining cellular TDP-43 homeostasis. Our results are in agreement with recent studies showing that RNA modulates LLPS and aggregation of RNA-binding proteins, proposing that RNA acts as a chaperone for this class of proteins (54).

Contribution of TDP-43 domains on protein assembly
To determine how each of the modular domains contributes to the two stages of assembly, we compared oligomer formation and aggregation of rTDP-43 deletion mutants (Fig. 5A). SDD-AGE analysis of the complexes showed that removal of the N-terminal domain (⌬N) decreased the formation of the initial oligomers and increased the formation of the large aggregates relative to the full-length protein (Fig. 5B). Conversely, deletion of the C-terminal low complexity region (⌬C) maintained the pattern of oligomerization and reduced the levels of large aggregates, consistent with its prion-like function. Removal of both N-and C-terminal domains (RRM1-2) triggered the formation of multiple oligomeric complexes starting at day 0 with little accumulation of high-molecular-weight species even after prolonged growth time. These results were supported by solubility assays in which protein samples (days 0 and 5 of the aggregation assay) were fractionated into RIPA-soluble and -insoluble fractions (Fig. 5C). The ratio of urea-soluble to RIPAsoluble protein was similar for all constructs at the initial time point (day 0). As expected, the later time point showed a significant increase of insoluble protein for full-length and ⌬N (day 5) (Fig. 5, C and D). Deletion of the C-terminal low complexity domain reduced insoluble aggregates 6-fold compared with full-length TDP-43. The RRM1-2 fragment showed the highest solubility with little increase in the insoluble fraction over time (Fig. 5, C and D). These results suggest a strong contribution of RRM1-2 in TDP-43 oligomerization and that adding only the C-terminal domain in the absence of the entire N-terminal region accelerates large aggregate accumulation. This is consistent with the strong aggregation-prone characteristic of the C-tail (16 -18).
Next, we investigated the structural changes associated with the different stages of TDP-43 aggregation seen by SDD-AGE using atomic force microscopy (AFM) and immuno-EM. Using these techniques, we studied both full-length TDP-43 and truncation mutants to compare the structure of large aggregates versus oligomeric patterns. AFM experiments showed that RRM1-2 does not form visible particles at day 0, whereas the full-length protein revealed small round oligomeric structures without incubation (Fig. 6A). After 1 day, the full-length formed  6 repeats, or control was added to rTDP-43 prior to triggering aggregation. B, aggregation of the RNA binding-deficient mutant F147L/F149L over time in the presence and absence of (GU) 6 . All four blots were probed with TDP-43 antibody and are representative of more than individual experiments.

Oligomeric species are intermediates in TDP-43 aggregation
larger density oligomers compared with RRM1-2. Nascent fibrillar structures of full-length rTDP-43 were observed after 2 days of incubation but were absent in RRM1-2 (Fig. 6A). Similarly, EM analysis showed fibrils of full-length TDP-43 at day 2, which grew in size after 7 and 10 days of incubation (Fig. 6B). This is consistent with previously obtained complexes formed by full-length TDP-43 and C-terminal domain protein fragments (27,55). In contrast, aggregates of RRM1-2 and ⌬C fragments appeared as smaller structures. Unlike full-length and ⌬N aggregates, RRM1-2 and ⌬C fragments failed to grow into large fibril-like assemblies by days 7 and 10. This is consistent with the assembly pattern of full-length and mutant rTDP-43 seen by SDD-AGE (Fig. 5B). Binding of anti-TDP-43 immunogold particles confirmed that the assemblies were formed by the respective proteins. Collectively, these results strongly suggest that the initial assembly during TDP-43 aggregation, which we detected as an ordered oligomeric pattern by SDD-AGE, are structurally different from later-day aggregates. The larger aggregates form fibril-like structures similar to those previously observed for full-length TDP-43 using similar microscopy methods (16,55).
To further dissect the contribution of TDP-43 domains during the different stages of aggregation, we studied two additional mutations (Fig. 7A). We generated a mutant harboring double residue substitutions, Y4R/E17R, to address the role of N-terminally mediated assembly without deleting the entire N-terminal region. These mutations do not affect folding but disrupt N-terminal domain self-assembly in vitro and splicing regulatory function in cells (11). To confirm that Y4R/E17R disrupts self-assembly through the N-terminal domain, we studied its LLPS behavior. Compared with WT rTDP-43, Y4R/ E17R showed a markedly reduced ability to form liquid droplets (Fig. 7B), in agreement with previous studies showing that TDP-43 LLPS decreases upon disruption of N-terminal domain-mediated association (11). The aggregation assays showed that Y4R/E17R did not affect oligomerization (Fig. 7C), suggesting that N-terminally mediated assembly does not play a

Oligomeric species are intermediates in TDP-43 aggregation
role in mediating the initial phase of TDP-43 aggregation. To address the role of RRM1-2 in the context of the full-length protein, we replaced the RNA-binding domains with monomeric EGFP (N-GFP-C) (Fig. 7A). We analyzed N-GFP-C aggregation focusing on the early days of the assay and found almost-complete inhibition of oligomerization with the chimera, compared with WT (Fig. 7D). Even at later time points, whereas WT rTDP-43 accumulated larger aggregates, N-GFP-C remained mostly monomeric. This provides strong evidence for the central role of RRM1 and 2 in the initial oligomerization during TDP-43 misfolding and confirms the results obtained with the RRM1-2 fragment (Fig. 5B). Moreover, the data showed that disruption of oligomer formation greatly delays aggregation in our assays. It is possible that inhibition of large aggregates in N-GFP-C were partially due to the loss of disulfide bonds and would be consistent with our results in Fig. 2C. The RRM1-2 region contains four of the six Cys residues in full-length TDP-43. This would suggest that oligomerization at the initial phase of TDP-43 aggregation creates a structural scaffold for high-molecular-weight aggregates and that this also depends on covalent disulfide bond formation. This model may be tested by analyzing the aggregation of mutants substituting RRM1-2 Cys residues to examine how these affect oligomer and large aggregate assembly. It is also possible that during misfolding, TDP-43 oligomerization and aggregation represent two alternative pathways that follow different assembly kinetics. This may be tested with additional TDP-43 variants that specifically stabilize or block oligomeric complexes.

Disease-linked mutations increase the assembly and aggregation of TDP-43 in vitro
The pathogenesis of ALS-causative mutations in TDP-43, for the most part, remains to be elucidated. Understanding how these mutations affect TDP-43 homeostasis and function is critical in uncovering the disease mechanisms of ALS and FTD. The clinicopathological features or TDP-43 lesions are similar in patients with or without TDP-43 mutations. This suggests that alternative methods are needed to understand the differences between WT and mutant TDP-43 at the molecular level. To test whether our assay may be used to determine how these mutations affect the aggregation process, we studied two single-site substitutions: A315T and M337V (Fig. 8A). Both mutations are highly associated with disease and previously reported to increase aggregation in vitro and promote neurotoxicity in animal models (5,27,56,57). We performed aggregation assays with A315T and M337V and analyzed the complexes formed over time by SDD-AGE (Fig. 8B). We observed that both mutations potently impacted the rate of aggregation as seen by significantly accelerated oligomer formation at the earliest time point (day 0), which was mostly absent in WT. The oligomer pattern typical of days 3 and 5 in WT TDP-43 was substituted by the larger aggregates in the mutants, particularly in the case of M337V. We also analyzed the initial soluble samples by SDD-

Oligomeric species are intermediates in TDP-43 aggregation
AGE to examine whether or not oligomers were already present before triggering aggregation. Fig. 8C shows that oligomers or aggregates are absent from the starting WT and mutant samples. Quantifying aggregation rates from the loss of monomer proteinatdifferenttimepointsshowedthatalthoughWTmonomer decreased linearly as a function of time, A315T and M337V monomers decreased at a significantly faster exponential rate (Fig. 8D). This, particularly the effect of M337V, is in strong agreement with previous observations that these substitutions increase amyloid formation of the isolated disordered C-tail (58,59) and full-length protein (16) and disrupt phase separation in cell models (60 -62). Our results on the macromolecular assembly of A315T and M337V may reflect distinct structures formed in the presence of the substitutions that have faster rates of complex association. Alternatively, these results may suggest that disease mutations decrease the dissociation of TDP-43 molecules making more stable complexes that result in stable fibrils or aggregates.

rTDP-43 aggregates seed and propagate intracellularly and this is enhanced by ALS mutations
Propagation of aggregates through prion-like transcellular transmission has been documented for a number of proteins associated with neurodegeneration, including TDP-43 (63,64). Previous studies showed TDP-43 seeding in cells; however, these assays used TDP-43 protein fragments, and the resulting intracellular aggregates were not detected by established markers of TDP-43 pathology, i.e. phosphorylation at Ser-409/410 (65)(66)(67). To investigate the aggregate seeding function of our rTDP-43 aggregates in cells, we generated a stable cell line to detect cytoplasmic aggregates in a reproducible and sensitive assay. A single copy of mCherry-tagged TDP-43 was stably integrated in HEK293 cells for tetracycline-induced expression (HEK-TDP-43 NLS ) (Fig. S4). This facilitated visualization of TDP-43 and prevented high and nonhomogeneous levels of transgene overexpression that could disrupt protein homeostasis. This system offers an advantage to studying TDP-43 aggregation relative to transient transfection, which often causes overexpression of the protein. We expressed full-length TDP-43 NLS harboring alanine substitutions disrupting the nuclear localization signal (NLS) and increasing cytoplasmic TDP-43, as previously observed (68,69) (Fig. S4, A and B). Expression of a similar TDP-43 NLS-mutant causes neurotoxicity accompanied by pathological inclusions in brain and spinal cord in mice (70). Our HEK-TDP-43 NLS cells expressed soluble TDP-43 NLS under control conditions but could be induced to form cytoplasmic aggregates readily detected by microscopy after addition of proteotoxic agents (e.g. MG132 and sodium arsenite) (Fig. S4, B and E). We confirmed that the cytoplasmic inclusions triggered by proteotoxic stress were recognized by an antibody recognizing phosphorylated Ser-409/410 (Fig. S4, C  and D).
We used lipofection-mediated transduction of rTDP-43 aggregates in HEK-TDP-43 NLS and observed protein internalization with fluorescently labeled rTDP-43. The protein was labeled with Oregon Green before initiating the aggregation assays. rTDP-43 samples collected shortly after triggering aggregation (day 0) (Fig. 1A) were delivered into cultured HEK-TDP-43 NLS . The cells were trypsinized and replated 24 h posttransfection to remove rTDP-43 from the medium and from association with the extracellular membrane. We observed internalization of labeled rTDP-43 48 h post-transfection (Fig.  9A). Aggregate formation of mCherry-TDP-43 NLS was monitored for several days. We observed a significant increase in cellular TDP-43 aggregation after 6 days of treatment, as seen by formation of large mCherry cytoplasmic foci (Fig. 9A). The mCherry-positive inclusions colocalized with the Oregon Green signal, suggesting the nucleation of de novo cellular aggregation by the rTDP-43 internalized aggregates. Cells treated with rTDP-43 aggregates showed a significant 2-fold increase in cells positive for mCherry aggregates compared with those treated with soluble rTDP-43 or no protein control (Fig. 9B). During our initial experiments, we tested seeding activity using rTDP-43 aggregates collected at various time points (0, 3 and 10 day aggregates). We did not find significant differences in the levels of rTDP-43 internalization or aggregate seeding between rTDP-43 aggregates collected at different time (Ͼ200 cells/sample counted in n Ͼ 3, S.D., GraphPad Prism two-tailed t test analysis; black asterisk and red asterisk, p Յ 0.04). C, percentage of HEK-TDP-43 NLS cells with mCherry-positive cytoplasmic aggregates quantified 6 days after transfection with full-length rTDP-43 and truncation mutant ⌬N, ⌬C, and RRM1-2 aggregates. Statistical significance calculated with Ͼ200 cells/sample, n Ͼ 3, S.D., using GraphPad Prism two-tailed t test analysis (two black asterisks, p Յ 0.004).

Oligomeric species are intermediates in TDP-43 aggregation
points (data not shown). The cellular aggregates were also positive for pSer-409/410 (Fig. S5, A and B), suggesting that the inclusions share structural and mechanistic similarities with those formed in disease. The relevance of our cellular reporter to study TDP-43 aggregate seeding is further supported by recent work using a similar HEK293 GFP-tagged TDP-43 NLS stable cell line (71). FTD-derived extracts positive for TDP-43 pathology showed cytoplasmic aggregate seeding in these cells and were also capable of propagating TDP-43 pathology in an in vivo model upon intracerebral injection in mice.
Next, we analyzed whether the ALS/FTD-linked mutations altered aggregate formation in cells as seen in our in vitro assays. Transduction of HEK-TDP-43 NLS with A315T and M337V rTDP-43 day 0 aggregates showed a significant 2-fold increase in mCherry-positive inclusions already at 48 h (Fig.  9B), whereas WT rTDP-43 aggregates did not show a significant difference compared with control at this time point. The effect of the disease-associated mutations was even greater after 6 days, showing a 3-fold increase in mCherry-TDP-43 NLS aggregates compared with control (Fig. 9B). A315T and M337V also caused significantly greater intracellular aggregation compared with WT, 50 and 67% increase, respectively, 6 days posttransfection. Of note, cellular internalization of rTDP-43 aggregates was similar for WT and mutant, and ϳ40% of cells showed cytoplasmic labeled rTDP-43. These results strongly suggest that the mutants do not alter cellular uptake but are more efficient at seeding and propagating relative to WT. This is in agreement with the increased rate of large aggregate assembly seen by SDD-AGE in vitro (Fig. 8B).
Similar experiments were performed with aggregates derived from rTDP-43 truncated mutants, ⌬N, ⌬C, and RRM1-2 to examine the contribution of oligomers versus large aggregates in cellular aggregate seeding. Deletion of both N-and C-terminal domains showed significantly reduced seeding function compared with WT. The proportion of cytoplasmic inclusions at 6 days post-transduction was similar (14%) in cells treated with RRM1-2 aggregates, soluble rTDP-43, or no protein control (Fig. 9, B and C). The seeding function of ⌬N and ⌬C were not statistically significantly different from full-length aggregates but showed increased and reduced mCherry-positive aggregates, respectively. Collectively, these results suggest that the oligomeric intermediates are not sufficient to nucleate cytoplasmic TDP-43 aggregation in the absence of the N terminus and particularly, upon deletion of the C-terminal region. Additional studies are needed to determine whether preventing oligomerization in the context of full-length TDP-43 disrupts aggregate seeding function.

Discussion
Our studies shed new light into the molecular steps that lead to TDP-43 aggregation and highlight methods to study the factors that control TDP-43 proteostasis and/or trigger protein misfolding. We also report a previously unappreciated role for disease-associated mutations on the aggregation mechanisms, providing important clues on how these mutations lead to pathology.
The use of SDD-AGE to analyze rTDP-43 aggregates provided the first evidence of a defined oligomeric pattern that precedes formation of TDP-43 high-molecular-weight aggregates (Figs. 1A and 2A). These intermediates may not be observed by other methods used to monitor aggregation, such as turbidity and other spectroscopic techniques. We confirmed the differences in the conformation and assembly state of earlyand late-stage aggregates by AFM/immuno-EM analysis (Fig.  6). Interestingly, our findings suggest that the oligomers detected by SDD-AGE are key intermediates during the transition from liquid droplets to irreversible aggregates (Fig. 3). This is consistent with the model that TDP-43 liquid-liquid phase separation may lead to aggregation over time, as suggested by studies of similar RNA-binding proteins linked to ALS/FTD (48,50). This mechanism resembles the aggregation process of other neurotoxic proteins, such as Tau, ␣-synuclein, and ␤-amyloid, in which these intermediate products have been proposed to be the toxic seeds that transmit cell damage and pathology (72)(73)(74)(75)(76)(77)(78). Future experiments will examine whether the TDP-43 oligomers at the initial aggregation time points are also more toxic to cells and/or propagate more efficiently than large aggregates in animal models.
The mechanisms and structural determinants of TDP-43 oligomerization have not been widely reported, although strong evidence supports TDP-43 physiological oligomerization in cells and in brain tissue (9,27,37). The initial complexes we observe in our aggregation assays represent at least partially misfolded species that may act as precursors of protein aggregation. These are likely to be different from the soluble assemblies formed in cells, which would be expected to disassemble under the semidenaturing conditions used for SDD-AGE analysis. Our SDD-AGE results showing that the size and levels of oligomeric complexes increase at higher temperatures further suggest that these form as the protein undergoes misfolding (Fig. 2A). Based on our findings, TDP-43 shows key similarities with the canonical yeast prion protein Sup35, which was recently found to assemble into functional, soluble complexes through liquidliquid phase separation in vitro and in yeast models (79). However, Sup35 also forms partially SDS-resistant oligomers that promote prion propagation (80).
Our observation that aggregates generated from the purified protein seed cellular cytoplasmic inclusions ( Fig. 9) are consistent with previous reports and our own observations (data not shown) that TDP-43 aggregates from disease tissue and isolated TDP-43 C-terminal fragments promote cellular TDP-43 aggregation (41,64,81) and more recently in mouse brain (71). Moreover, these results show that rTDP-43 aggregates are relevant to cell and patient-derived inclusions and suggest that, similar to Tau and ␣-synuclein, TDP-43 aggregates act as templates in cellular TDP-43 aggregation (63). We also provide strong evidence that enhanced TDP-43 aggregation rates translate into greater cellular seeding function, as seen with the ALS-linked mutations A315T and M337V. By studying the aggregation and seeding of A315T and M337V, we propose that pathogenesis is strongly influenced by factors that increase the assembly and/or alter the dissociation rates. Changes in the aggregation rate may be due to structurally distinct complexes associated with the mutations. However, our observations that A315T and M337V aggregates nucleate cellular TDP-43 suggest that mutant complex structure and assembly pathways are sufficiently similar to

Oligomeric species are intermediates in TDP-43 aggregation
WT to act as seeds for aggregation. These findings open the avenue for future studies aimed at addressing whether the mechanisms observed with A315T and M337V apply to other TDP-43 disease-associated mutations. This will increase our understanding of the molecular basis of ALS-FTD pathogenesis. Importantly, studies on the mechanisms by which these mutations alter structural and kinetic parameters of TDP-43 self-assembly will shed light on the regulation of the equilibrium between soluble complexes and irreversible aggregates.
Until recently, most efforts to understand TDP-43 aggregation centered on the low complexity C-terminal region. Our results provide strong evidence supporting a central role of the RNA-binding domains in TDP-43 aggregation, particularly during the initial phase of misfolding. We showed that replacing both RRMs with monomeric GFP drastically disrupts oligomerization and reduces large aggregate accumulation over time (Fig. 7D). First, this suggests that in the absence of the RRM-dependent oligomers, the N terminus and the C-tail are not sufficient to form detergent-resistant aggregates, which share common features with FTD and cell-derived inclusions (Fig. 1, B and C). Second, it supports a model in which oligomerization provides a scaffold for TDP-43 aggregation. Previous studies showed that isolated RRM2 forms aggregates (82) and that RRM2 polypeptides that are normally buried under native conditions assemble into amyloid, leucine zipper structures (83)(84)(85)(86). Based on these previous findings and our results, we predict that RRM2 forms a hydrophobic core, intermediate structure upon TDP-43 unfolding that mediates the oligomers detected by SDD-AGE. Interestingly, the strong effect of RNA binding in preventing TDP-43 oligomerization and aggregation (Fig. 4) provides further support that stabilizing a specific RRM conformation maintains protein solubility.

Experimental procedures
All reagents are from Sigma-Aldrich unless otherwise specified.

Plasmid construction
Construction of the TDP-43 vector for bacterial expression (SUMO-TDP43) was previously described (21). Oligonucleotides used for mutagenesis and cloning are described in Table  S1. Deletion constructs for bacterial expression were made by standard PCR and cloning between BamHI and SacI restriction sites in pET28b/His-SUMO (87). Single-site substitutions, including modification of the TDP-43 nuclear localization sequence, were made by site-directed mutagenesis using the QuikChange site-directed mutagenesis kit (Agilent Technologies) as described (15). The template used for mutagenesis of the mammalian expression vector, HA-mCherry-TDP-43, was made by cloning hemaggluttinin peptide-mCherry cDNA into pCDNA5/FRT upstream of TDP-43 cDNA (NM_007375.3). TDP-43 was cloned between BamHI and NotI sites. N-GFP-C was generating by PCR-mediated fusion of TDP-43 residues 2-101, mEGFP, and TDP-43 residues 261-414. This chimera was cloned between BamHI and SacI restriction sites in pET28b/His-SUMO.

Sequential extraction of TDP-43 from postmortem human brain tissue
Samples of frozen postmortem human brain tissue from control and FTD deidentified cases were obtained from the Cognitive Neurology and Alzheimer's Disease Center at Northwestern University. Gray matter was dissected from frontal cortex samples using a scalpel while maintaining the tissue in a frozen state. Extraction methods were adapted from (33). Samples of dissected tissue (ϳ100 mg) were homogenized in 5 ml/g Trissucrose buffer (25 mM Tris-HCl, pH 7.4, 5 mM EDTA, 1 mM DTT, 10% sucrose, protease inhibitor (PI) mixture) with Kontes Dounce tissue grinders (Kimble KT885300-0002). The homogenate was spun at 18,000 ϫ g (average relative centrifugal force) for 30 min at 4°C. The resulting pellet was next extracted in 5 ml/g Triton X-100 buffer (Tris sucrose buffer, 1% Triton X-100, 0.5 M NaCl) and centrifuged at 135,000 ϫ g (RCF average) for 30 min at 4°C. To float and remove myelin, the pellet was further extracted in TX-30% sucrose buffer (25 mM Tris-HCl, pH 7.4, 5 mM EDTA, 1 mM DTT, 10% sucrose, 1% Triton X-100, 0.5 M NaCl, 30% sucrose, PI mixture) and centrifuged at 135,000 ϫ g (RCF average) for 30 min at 4°C. Finally, the pellet was homogenized in 5 ml/g sarkosyl buffer (25 mM Tris, pH 7.4, Oligomeric species are intermediates in TDP-43 aggregation N-lauroylsarcosine sodium salt, 0.5 M NaCl) and incubated at room temperature for 1 h with shaking at 750 rpm on Advanced Vortex Mixer (VWR 89399-884). The sarkosyl extracted homogenate was centrifuged at 135,000 ϫ g (RCF average) for 30 min at 4°C. The resulting pellet was resuspended in 100 l of RIPA buffer (50 mM Tris, pH 8.0, 150 mM NaCl, 1% Nonidet P-40, 0.5% SDC, 0.1% SDS, 5 mM EDTA, PI mixture, PhosStop (Roche)), sonicated with a nanoruptor (Diagenode) for 10 cycles of 30 s on, 30 s off. SDD-AGE analysis was carried out with 25 l of this sample, representing 12.5% of the initial sample.

Thioflavin binding
To test binding of initial rTDP-43 aggregates to the amyloiddetection dye, ThioT, rTDP-43 monomer was first spun down at 100,000 ϫ g for 30 min at 4°C to remove any aggregates formed during the freeze-thaw process. In a Corning Black 96-well plate (Fisher, catalog no. 07-200-762), 50 l of 0.5 M rTDP-43 monomer was combined with increasing concentrations (0 -1000 nM) of ThioT in a final volume of 100 l. Nonspecific signal was assessed by incubating ThioT alone with buffer. The reaction mixture was incubated for 2 h at room temperature to allow formation of rTDP43 aggregates. At the end of incubation, ThioT fluorescence was measured in a BioTek plate reader using a 440/30-nm excitation filter, a 485/ 20-nm emission filter, and top 50% optical setting. In parallel, we incubated 0.5 M of recombinant ␣-synuclein fibrils with ThioT under the same conditions used for rTDP43. Methods to produce recombinant ␣-synuclein monomer and fibrils were as described previously (38). ThioT and ThioS binding to TDP-43 cellular aggregates was performed using HEK-TDP-43 NLS cells treated with sodium arsenite (0.5 mM, 30 min). Binding was analyzed by immunofluorescence of paraformaldehyde-fixed cells using 1 mM dye.

Liquid-liquid phase separation assays
Oregon Green 488 -labeled rTDP-43 was mixed with unlabeled protein at a 1:10 ratio at 0.5 M in 18 mM MES, pH 7.0, 5 mM Tris, 50 mM NaCl, 1% glycerol, 1% sucrose, 30 mM imidazole. The samples were visualized by fluorescence microscopy immediately or incubated at 22°C for 30 -120 min prior to analysis or SDD-AGE. Reversal of phase separation was achieved by raising the NaCl concentration to 250 mM immediately after droplet formation.

Cross-linking and labeling rTDP-43
For cross-linking analysis, rTDP-43 (0.5 or 1 M) was incubated with 0.25 mM disuccinimidyl suberate (Thermo Scientific) for 1 h at 22°C. The reaction was quenched by addition of Tris, pH 8.0, to 50 mM final concentration and incubated for 15 min at 22°C. rTDP-43 the was labeled with Oregon Green 488 maleimide (ThermoFisher Scientific) according to manufacturer's protocol. Labeled protein was purified with ZebaSpin columns (ThermoFisher Scientific) and used to label aggregates. Aggregation or LLPS assays included a 1:10 ratio mixture of labeled to nonlabeled rTDP-43.

Atomic force and immuno-EM
Aliquots of rTDP-43 aggregation time points (10 l) were placed on a clean, freshly cleaved grade V-1 mica (catalog no. 01792-AB, Structure Probe, Inc.). After 10 min, the solvent was wicked off by filter paper, and the mica was washed six times with 20 l of water to remove salts and buffer from the sample. Samples were dried overnight, and AFM images were acquired in tapping mode on a Veeco Dimension 3100 machine (Bruker) with Bruker FESP tips. AFM images were analyzed using the Gwyddion SPM data visualization tool. Carbon films on 200mesh copper grids (Ted Pella) were incubated with 5 l of sample in the dark side of the grid. After 10 min, the sample was wicked off from the grid and was incubated with 10 l, 1% BSA in PBS to block any nonspecific binding. The grid was incubated with primary rabbit TDP-43 antibody 1:100 diluted in PBS solution containing 0.1% BSA for 45 min. The grid was then washed by seven drops of PBS buffer. The grid was incubated with secondary anti-rabbit IgG antibody conjugated to 5-nm-diameter gold nanoparticles (Sigma-Aldrich) diluted 1:20 in PBS buffer containing 0.1% BSA. 45 min later, the grid was washed by seven drops of PBS followed by seven drops of water. Finally, the grid was stained by 4% uranyl acetate solution for 10 min, washed with seven drops of water, and air-dried before collecting the images on a JEM-1400 Plus transmission electron microscope.

Cell culture
The cells were grown in growth medium-Dulbecco's modified Eagle's medium, 4,500 mg/liter glucose, L-glutamine, and sodium bicarbonate and supplemented with filtered fetal bovine serum at 10%. The cells were incubated at 37°C, 5% CO 2 . Stable human embryonic kidney cells expressing TDP-43 NLS (HEK-TDP-43 NLS ) upon induction with tetracycline were generated from Flp-In TM T-REX TM 293 cells (Thermo-Fisher) according to the manufacturer's instructions. The cells were grown in the presence of hygromycin (50 g/ml), and transgene expression was induced with 1 g/ml tetracycline.

TDP-43 aggregate cellular seeding
HEK-TDP-43 NLS were plated and induced with 1 g/ml tetracycline for 16 h to reach 80 -90% confluency. Recombinant TDP-43, day 0 aggregates were transduced with Pierce protein transfection reagent according to the manufacturer's instructions at a final protein concentration of 0.1 M in the culture media. As controls, soluble rTDP-43 or transfection reagent

Oligomeric species are intermediates in TDP-43 aggregation
only were used. The medium was changed 6 h post-transfection. After 24 h, the cells were trypsinized, washed, and replated. For immunofluorescence, the cells were plated on poly-D-lysinecoated coverslips 48 h prior to fixation in 4% paraformaldehyde for 20 min. For fluorescence microscopy, slides were permeabilized for 5 min at 4°C with 0.3% Triton X-100 and DAPI-stained. Coverslips were mounted, and the cells were observed on a Leica DMI3000B inverted microscope and Leica AF6000E software (Leica Microsystems Inc.). A 63ϫ/ 1.4 oil immersion objective was used for confocal microscopy studies on a TCS SP5 microscope (Leica) using the LAS AF software. Images were taken with the DAPI filter to eliminate bias selection of aggregates. Cells positive for cytoplasmic aggregation were quantified using ImageJ.