Introduction

TAR DNA-binding protein (TDP-43) is the major aggregating protein in amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) patients and also forms pathological aggregates in up to 50% of Alzheimer's disease patients (Neumann et al, 2006; Josephs et al, 2014). It is a ubiquitously expressed RNA-binding protein (RBP) with key functions in RNA processing, e.g., regulation of alternative splicing and polyadenylation, miRNA processing, mRNA stability and localization (Ratti & Buratti, 2016). In the affected brain regions of ALS and FTD patients, the physiological diffuse nuclear localization of TDP-43 is lost. Instead the protein forms cytoplasmic and occasionally nuclear inclusions in neurons and glial cells (Mackenzie et al, 2010). TDP-43 pathology closely correlates with neurodegeneration, and both loss-of-function mechanisms, e.g., misregulation of nuclear RNA targets, and gain-of-function mechanisms, e.g., aberrant interactions of the TDP-43 aggregates, are believed to contribute to neuronal dysfunction and eventually neurodegeneration (Ling et al, 2013; Tziortzouda et al, 2021).

Similar to other prion-like RBPs, TDP-43 is thought to aggregate through aberrant liquid–liquid phase separation (LLPS), i.e., the transition of liquid-like RBP condensates into a solid-like state (Nedelsky & Taylor, 2019). Aberrant phase transitions may occur in stress granules (SGs) or other membrane-less organelles (MLOs), where aggregation-prone RBPs are highly concentrated and exceed the critical concentration for LLPS (Alberti & Dormann, 2019; Alberti & Hyman, 2021). Subsequent liquid-to-solid phase transition, as demonstrated for various disease-linked RBPs in vitro (Molliex et al, 2015; Patel et al, 2015), may then cause formation of pathological RBP inclusions. LLPS is often driven by intrinsically disordered low complexity domains (LCDs), that tend to engage in weak multivalent interactions with other molecules (Alberti, 2017). TDP-43 harbors a long C-terminal LCD enriched in glycine, serine, asparagine and glutamine residues, which drives intermolecular TDP-43 interactions and assembly by phase separation (Conicella et al, 2016; Babinchak et al, 2019). The LCD is also the region that harbors numerous ALS-linked point mutations (Buratti, 2015), suggesting that small chemical changes to the TDP-43 LCD can cause neurodegeneration.

Liquid–liquid phase separation and MLO dynamics are often regulated by post-translational modifications (PTMs) in LCDs, as the introduction of small chemical groups or proteins changes the chemical nature of amino acids, e.g., their charge or hydrophobicity, which can alter their molecular interactions and LLPS behavior (Bah & Forman-Kay, 2016; Hofweber & Dormann, 2019). A highly disease-specific PTM on deposited TDP-43 inclusions is hyperphosphorylation on C-terminal serine residues in the LCD (Hasegawa et al, 2008; Inukai et al, 2008; Neumann et al, 2009; Kametani et al, 2016). Antibodies specific for C-terminal TDP-43 phosphorylation sites (e.g., S409/S410 and S403/S404) detect inclusion pathology in patients, without cross-reactivity with physiological nuclear TDP-43. Therefore, C-terminal TDP-43 hyperphosphorylation is considered a pathological hallmark and is generally believed to promote TDP-43 aggregation (Buratti, 2018). This view is largely based on the observations that C-terminal TDP-43 phosphorylation correlates with inclusion pathology and that overexpression of kinases that can phosphorylate TDP-43 enhance TDP-43 aggregation and neurotoxicity (Choksi et al, 2014; Liachko et al, 2014; Nonaka et al, 2016; Taylor et al, 2018). Based on these studies, inhibition of TDP-43 phosphorylation by specific kinase inhibitors has even been proposed as a potential therapeutic strategy for ALS (Liachko et al, 2013; Salado et al, 2014; Martinez-Gonzalez et al, 2020). However, the molecular consequences of this disease-linked PTM are still poorly understood, and its effects on TDP-43 LLPS and aggregation are still unknown.

Using in vitro, in silico and cellular experiments, we now demonstrate that disease-linked C-terminal hyperphosphorylation of TDP-43 suppresses TDP-43 condensation and insolubility. We show this through (i) in vitro phase separation and aggregation assays with recombinant, full-length TDP-43; (ii) coarse-grained and atomistic molecular dynamics (MD) simulations of condensates of TDP-43 LCDs, elucidating molecular driving forces; and (iii) experiments in HeLa cells, stable inducible U2OS cells and primary rat neurons, where C-terminal phosphomimetic mutations do not disturb nuclear import or RNA processing functions of TDP-43, but abrogate TDP-43 condensation into MLOs and enhance its solubility. Based on our findings, we speculate that C-terminal TDP-43 hyperphosphorylation may be a protective cellular response to counteract TDP-43 solidification, rather than being a driver of TDP-43 pathology, as has so far been assumed.

Results In vitro phosphorylation with Casein kinase 1δ reduces condensation of TDP-43

To examine how phosphorylation affects TDP-43 phase transitions, we expressed and purified unphosphorylated full-length TDP-43 with a solubilizing MBP tag and a His6-tag in Escherichia coli (Wang et al, 2018) (Appendix Fig S1A–E). We then in vitro phosphorylated the purified protein with casein kinase 1 delta (CK1δ), a kinase previously reported to phosphorylate TDP-43 at disease-associated sites (Kametani et al, 2009), and confirmed phosphorylation of C-terminal serines (S403/S404; S409/S410) with phospho-specific antibodies (Fig EV1A). Mass spectrometric analysis detected phosphorylation on several additional serine/threonine sites (Fig EV1B), and the running behavior in SDS–PAGE suggests hyperphosphorylation on multiple sites (Figs 1B and EV1A). We then induced phase separation of the unphosphorylated vs in vitro phosphorylated TDP-43 by cleaving off the MBP tag with TEV protease (Wang et al, 2018) and used centrifugation to separate the condensates (C) from the cleared supernatant (S; Fig 1A). Cleaved TDP-43 was mostly in the condensate fraction (S/[S + C] ratio ~0.25), whereas in vitro phosphorylated TDP-43 was predominantly in the supernatant (S/[S + C] ratio > 0.6; Fig 1B and C). Reduced sedimentation of TDP-43 was not seen upon addition of adenosine triphosphate (ATP) or CK1δ alone, suggesting that it is indeed caused by the addition of phospho-groups to TDP-43.

Click here to expand this figure.

Figure EV1. Identification of TDP-43-MBP-His6 phospho-sites after in vitro phosphorylation with CK1δ

Identification of TDP-43 phospho-sites on in vitro phosphorylated TDP-43 (+CK1δ, +ATP) in comparison to controls (−CK1δ −ATP; CK1δ only; ATP only) by Western blot. Samples were analyzed by SDS–PAGE and Western blot using a rabbit anti-TDP-43 N-term antibody (Proteintech) to detect total TDP-43, rat anti-TDP-43-phospho Ser409/410 (clone 1D3, Helmholtz Center Munich) and mouse anti-TDP-43-phospho Ser403/404 (Proteintech, Cat. No.: 66079-1-Ig) antibodies. Schematic diagrams showing sequence coverage in mass spectrometry after trypsin digest (underlined) and phosphorylated serine/threonine residues (orange) of in vitro phosphorylated TDP-43-MBP-His6 with CK1δ + ATP (one out of two representative experiments is shown).

Source data are available online for this figure.

Figure 1. TDP-43 phosphorylation by CK1δ and C-terminal phosphomimetic substitutions reduce TDP-43 condensation in vitro

A. Scheme of sedimentation assay (created in BioRender.com): phase separation of TDP-43 was induced by TEV protease cleavage of TDP-43-MBP-His6, and condensates were pelleted by centrifugation. B. Sedimentation assay to quantify condensation of unmodified TDP-43 versus in vitro phosphorylated TDP-43 (+CK1δ, +ATP) and controls (CK1δ or ATP only); TDP-43 detected by Western blot (rabbit anti-TDP-43 N-term). Due to incomplete TEV cleavage, some TDP-43-MBP-His6 remains present and co-fractionates with cleaved TDP-43, due to TDP-43 self–self interaction. C. Quantification of band intensities of cleaved TDP-43 shown as mean of Supernatant/(Supernatant + Condensate) (S/[S + C]) ratio of three independent experimental replicates (n = 3) ± SD. ***P < 0.0002 by one-way ANOVA with Dunnett's multiple comparison test to Wt. D. Schematic diagram of TDP-43 and sequence of C-terminal region (aa. 370–414) for Wt, phosphomimetic (S-to-D) variants and control (S-to-A) variants. NTD, N-terminal domain; RRM, RNA recognition motif; LCD, low complexity domain with α-helical structure. E. Turbidity measurements (optical density [OD] at 600 nm) to quantify phase separation of the indicated TDP-43 variants at three different concentrations (in Hepes buffer). Values represent mean of three independent experimental replicates (n = 3) ± SD. *P < 0.0332, **P < 0.0021 and ***P < 0.0002 by one-way ANOVA with Dunnett's multiple comparison test to Wt, comparing the respective concentration condition (5, 10 and 20 µM). F–I. Representative bright field microscopic images of TDP-43 condensates (in Hepes buffer), Bar, 25 µm (F) and quantification of condensate number (G), size (H) and roundness (I). Box plots show the comparison of median and inter-quartile range (upper and lower quartiles) of all fields of view (FOV) from Min to Max (whiskers) of two replicates (≥ 22 FOV per condition). *P < 0.0332, **P < 0.0021 and ***P < 0.0002 by one-way ANOVA with Dunnett's multiple comparison test to Wt, comparing the respective concentration condition (5, 10 and 20 µM).

Source data are available online for this figure.

C-terminal phosphomimetic substitutions mimicking disease-linked phosphorylation suppress TDP-43 phase separation

To study defined disease-linked phosphorylation sites, we generated phosphomimetic proteins harboring different numbers of phosphomimetic serine-to-aspartate (S-to-D) mutations or corresponding serine-to-alanine (S-to-A) mutations as control. Phosphomimetic substitutions rely on the replacement of a phosphorylated serine or threonine with a negatively charged amino acid (D or E), thus mimicking the negative charge of the phospho group. Even though they under-appreciate the charge change (net charge of aspartate = −1 instead of −2 for a phospho-group) and do not always accurately mimic the chemistry of a phospho group, phosphomimetics have been successfully used to probe the biological function of phosphorylated residues (Martin et al, 2014). Phosphorylation on S409/S410 is a highly specific and consistent feature of aggregated TDP-43 in all ALS/FTD subtypes (Inukai et al, 2008; Neumann et al, 2009), and five phosphorylation sites (S379, S403, S404, S409 and S410) were detected with phosphorylation site-specific antibodies in human post-mortem tissue (Hasegawa et al, 2008). Therefore, we mutated these serines to create "2D" and "5D" variants as well as the corresponding “2A” and “5A” controls (Fig 1D). Based on a mass spectrometric study that found phosphorylation on 12 out of 14 serines in the C-terminal LCD of TDP-43 in ALS spinal cord (Kametani et al, 2016), we also mutated these 12 sites (S373, S375, S379, S387, S389, S393, S395, S403, S404, S407, S409 and S410) to create “12D” or “12A” variants (Fig 1D). Interestingly, the PLAAC web tool (http://plaac.wi.mit.edu/) that allows prediction of probable prion subsequences using a hidden-Markov model (HMM) algorithm (Lancaster et al, 2014), predicted a reduced prion-like character of the C-terminal region in the phosphomimetic 12D variant compared with the wild-type (Wt) and 12A protein (Appendix Fig S2).

To study phase separation experimentally, all variants were expressed and purified as TDP-43-MBP-His6 fusion proteins (Appendix Fig S1A–E), and phase separation induced by TEV protease-mediated cleavage of the MBP tag was examined by turbidity, sedimentation or microscopic condensate assays. Turbidity measurements revealed a concentration-dependent increase in phase separation for TDP-43 Wt, as expected, whereas the increase was less pronounced for the 2D and 5D variants and no concentration-dependent increase was seen for the 12D mutant (Fig 1E). The gradual decrease in turbidity caused by the phosphomimetic mutations (Wt > 2D > 5D > 12D) was not seen to the same extent for the corresponding S-to-A control mutations (Fig 1E), hence suppression of phase separation is not due to the loss of serines at these positions, but can be attributed to the additional negative charges introduced by the D substitutions. Turbidity assays in phosphate buffer instead of Hepes buffer gave similar results (Fig EV2A), and sedimentation assays confirmed that TDP-43 condensation is gradually suppressed by increasing numbers of phosphomimetic mutations (Fig EV2B and C).

Click here to expand this figure.

Figure EV2. C-terminal phosphomimetic substitutions reduce TDP-43 condensation in vitro

Turbidity measurements (optical density [OD] at 600 nm) to quantify phase separation of different S-to-D and S-to-A mutants in comparison to TDP-43 Wt using phosphate buffer. Values represent mean of three independent experimental replicates (n = 3) ± SD. *P < 0.0332, **P < 0.0021 and ***P < 0.0002 by one-way ANOVA with Dunnett's multiple comparison test to Wt, comparing the respective concentration condition (5, 10, 20 µM). Sedimentation assay to quantify condensation of different S-to-D mutants in comparison to TDP-43 Wt (in Hepes buffer). TDP-43 was detected by TDP-43 Western blot (rabbit anti-TDP-43 N-term). Quantification of band intensities of cleaved TDP-43 corresponding to supernatant (S) and condensates (C) fractions is shown as mean of S/(S + C) ratio of three independent experimental replicates (n = 3) ± SD. **P < 0.0021 and ***P < 0.0002 by one-way ANOVA with Dunnett's multiple comparison test to Wt. Representative bright field microscopic images of TDP-43 condensates formed from TDP-43 Wt vs different S-to-D or S-to-A variants in phosphate buffer (Bar, 25 µm).

Source data are available online for this figure.

Phosphomimetic S-to-D substitutions lead to rounder TDP-43 condensates, whereas S-to-A mutations cause an amorphous, aggregate-like morphology

Interestingly, bright field microscopy revealed that TDP-43 Wt formed relatively small, amorphous condensates, suggestive of solid-like material properties (Fig 1F). In contrast, the phosphomimetic S-to-D proteins formed fewer, but much larger and rounder condensates (Fig 1F, see G–I for quantification), suggesting a more liquid-like behavior and therefore fusion of condensates into larger droplets. Again, the observed changes were correlated with the number of phosphomimetic mutations, i.e., they were most pronounced for the 12D mutant, which formed very few, but large and perfectly circular protein droplets. (Note that these few large condensates most likely escape detection in the turbidity assay due to rapid sedimentation during the assay.) In contrast, the S-to-A control variants formed numerous small, amorphous condensates and had a more irregular, aggregate-like appearance than TDP-43 Wt (Fig 1F). This phenotype suggests that the OH groups in the respective serines influence the material properties of TDP-43 and contribute to preventing its aggregation. Similar results were obtained when the assay was carried out in phosphate buffer instead of Hepes buffer, except that 12D formed only very few, small condensates in phosphate buffer (Fig EV2D), possibly because the ions in phosphate buffer may screen certain attractive interactions between TDP-43 molecules and disfavor phase separation. Together, these results demonstrate that phosphomimetic substitutions mimicking disease-linked C-terminal TDP-43 phosphorylation reduce the tendency of TDP-43 to phase separate into amorphous condensates and suggest a more dynamic, liquid-like behavior of C-terminally phosphorylated TDP-43.

C-terminal phosphomimetic substitutions yield more liquid-like, dynamic TDP-43 condensates

To test whether the phosphomimetic mutations indeed render TDP-43 more liquid-like, we performed live imaging of Alexa488-labeled Wt, 5D and 12D condensates by spinning disc confocal microcopy. For TDP-43 Wt, no fusion events were observed over a time course of several minutes. Instead the small condensates stuck to each other in a chain-like arrangement (Movie EV1, Fig 2A). In contrast, 5D condensates occasionally and slowly fused with each other, and 12D condensates readily fused upon contact and relaxed into perfectly round spheres, indicating a liquid droplet-like nature (Movies EV2 and EV3, Fig 2A). To assess the mobility of TDP-43 molecules in condensates, we performed half-bleaches of condensates and analyzed fluorescent recovery after photobleaching (FRAP) in the bleached half. In TDP-43 Wt condensates, fluorescence recovered very slowly, indicating a low mobility of TDP-43 molecules, whereas recovery was faster in 5D and even faster in 12D condensates (Fig 2B and C), in line with an increased mobility of phosphomimetic TDP-43 compared with “unmodified” TDP-43. Taken together, phosphomimetic S-to-D substitutions in the C-terminal region enhance the liquidity of TDP-43 condensates, suggesting that phosphorylation in this region might counteract TDP-43's tendency to form solid, irreversible aggregates.

Figure 2. C-terminal phosphomimetic substitutions enhance liquidity of TDP-43 condensates and reduce TDP-43 aggregation in vitro

Representative still images of Alexa488-labeled TDP-43 condensates by spinning disc timelapse confocal microscopy. Wt condensates do not fuse, 5D condensates fuse slowly and 12D condensates readily fuse upon contact and relax into spherical droplets. Bar, 5 µm. Representative images of FRAP experiments at indicated time-points. Boxes indicate bleached area (half-bleach of condensate). Bar, 5 µm. FRAP curves after half-bleach of freshly formed Alexa488-labeled TDP-43 condensates. Values represent mean ± SD of three independent experimental replicates (n = 3) of ≥ 9 droplets analyzed per condition. ***P < 0.0002 by one-way ANOVA with Tukey's multiple comparison test for area under the curve (AUC) of individual droplets. Confocal images of Alexa488-labeled TDP-43 aggregates formed in an in vitro aggregation assay (with TEV protease cleavage). Bar, 100 µm. Zoom shows magnified view of aggregates at the 24 h time point. Bar, 20 µm. SDD-AGE followed by TDP-43 Western blot to visualize SDS-resistant oligomers/high-molecular-weight species of TDP-43-MBP-His6 in an in vitro aggregation assay (without TEV protease cleavage). Asterisks represent monomeric (*), oligomeric (**) and polymeric (***) species. Input of TDP-43-MBP-His6 variants used in the SDD-AGE assay, detected by Western blot (anti-TDP-43 N-term).

Source data are available online for this figure.

C-terminal phosphomimetic substitutions reduce TDP-43 aggregation

To address whether phosphorylation indeed counteracts TDP-43 aggregation, we performed in vitro aggregation assays modified from published protocols (Halfmann & Lindquist, 2008; French et al, 2019). Under the assay conditions, TEV cleavage of fluorescently labeled TDP-43-MBP-His6 yields amorphous TDP-43 aggregates that can be visualized by confocal microscopy. In contrast to Wt or 12A, the phosphomimetic 5D or 12D proteins formed much smaller and fewer aggregates, respectively (Fig 2D), suggesting that C-terminal TDP-43 phosphorylation can efficiently suppress TDP-43 aggregation. For biochemical characterization of the formed aggregates, we performed semi-denaturing detergent-agarose gel electrophoresis (SDD-AGE) under the same assay conditions, just in the absence of TEV, as MBP-tagged TDP-43 aggregates slower than TDP-43 and distinct oligomeric/polymeric species resistant to 0.5% SDS can be visualized under these conditions (Appendix Fig S3). In comparison to TDP-43 Wt and 5D, 12D showed reduced and delayed oligomerization and formation of high-molecular-weight species (Fig 2E, equal protein input shown in Fig 2F). In contrast, 12A formed SDS-resistant oligomers/high-molecular-weight species at a higher rate, which together with our microscopic images of TDP-43 condensates (Fig 1F), suggests that C-terminal alanine substitutions make TDP-43 more aggregation-prone. Taken together, C-terminal phosphomimetic substitutions that mimic the phosphorylation pattern in ALS patients reduce the formation of SDS-resistant high-molecular-weight oligomers and TDP-43 aggregates in vitro.

Multi-scale simulations of the TDP-43 LCD reveal reduced protein-protein interactions through enhanced solvation of phosphomimetic residues

To understand the effect of C-terminal TDP-43 phosphorylation on phase separation at the molecular level, we used coarse-grained and atomistic MD simulations of the disordered TDP-43 LCD (aa. 261–414) with and without phosphomimetic substitutions. In coarse-grained simulations, we can access the relevant long time and large length scales to characterize phase behavior, while in atomistic simulations we can resolve the interactions of condensates with high resolution and high accuracy (Dignon et al, 2018; Pietrek et al, 2020; Benayad et al, 2021). We found that phosphomimetic substitutions locally reduce protein–protein interactions (Fig EV3) and increase protein–solvent interactions (Fig 3). In the coarse-grained simulations, the LCD of both TDP-43 Wt and 12D phase separated spontaneously to form condensates (shown for Wt in Fig 3A and Movie EV4). Yet, phosphomimicking residues are less prone to interact with protein in the phase-separated condensates and are somewhat more solvated than the corresponding serine residues (Figs 3B and EV3A and B). The aspartate side chains in 12D LCDs engage in partially compensatory interactions with arginines, showing that introduction of charged side chains gives rise to both stabilizing and destabilizing interactions in condensates. Importantly, our simulations are in line with previous studies that have highlighted the importance of aromatic sticker–sticker interactions in driving phase separation of prion-like domains and the TDP-43 LCD (Li et al, 2018; Schmidt et al, 2019; Martin et al, 2020).

Click here to expand this figure.

Figure EV3. Analysis of contacts in biomolecular condensates formed by the TDP-43 LCD in coarse-grained simulations

A, B. Contact maps for Wt (A) and 12D (B) TDP-43 LCD from simulations with the explicit solvent Martini coarse-grained model. Residue i and residue j are defined to be in contact if any of the coarse-grained beads are within 4.5 Å. The relative contact probability is calculated by averaging over all 118 protein chains and the last 5 of 20 μs simulations each. Intra-chain contacts with the two preceding and following residues are excluded from the analysis. Aromatic residues form prominent contacts and are highlighted by black arrows. For example, looking at the column for F276 and following it upwards one can see that F276 interacts with F276 in other chains and irrespective of the chain, with F283, F289, F313, F316, W334, F367, Y347, W385, F401, and W412. The sites of the phosphomicking S-to-D mutations are highlighted by purple arrows. At these sites differences between Wt and 12D LCD can be seen, with Wt forming more contacts close in protein sequence and 12D instead interacting with R268, R272, R275, R293, and R361 further away in the sequence. C. Differences in contact probability Pi,j = Pi,j(Wt) − Pi,j(12D) from simulations with the explicit-solvent Martini coarse-grained model. Differences highlight that wild-type S residues, unlike phosphomicking D residues, favor interactions with residues close in sequence, while demonstrating that most contacts are not affected by the phosphomicking S-to-D mutations. Black and purple arrows correspond to aromatic residues and phosphomicking S-to-D mutations, respectively.

Figure 3. Atomistic and coarse-grained simulations of TDP-43 LCD: phosphomimicking residues form fewer protein–protein interactions and more protein–solvent interactions

A. TDP-43 LCD phase separates in coarse-grained simulations with explicit solvent. Condensate of TDP-43 Wt LCD is shown, protein colored according to chain identity. Water omitted for clarity. Ions shown in cyan. B. Normalized probability of protein-protein contacts by phosphomimicking aspartates in 12D and serines in Wt resolved by amino acid type from coarse-grained simulations. Error bars smaller than symbols. Inset: Distributions of the number of water molecules within 5 Å of side chains of phosphomimicking aspartates of 12D and corresponding serines in Wt from 15 µs of coarse-grained molecular dynamics simulations. C. Atomistic simulation setup of 32 TDP-43 LCDs. Different LCD chains shown in different colors in space-filling representation. For one chain (lower left), a transparent surface reveals its atomic structure as sticks. D. Normalized probability of protein–protein contacts by phosphomimicking aspartates in 12D and serines in Wt resolved by amino acid type from atomistic simulations. Two 1 µs simulations are distinguished by color intensity. Inset: distributions of the number of water molecules within 5 Å of the side chains of phosphomimicking aspartates of 12D and the corresponding serines in Wt from atomistic simulations. E. Representative snapshots of atomistic simulations showing water within 3 Å of (left) Wt S407, S409 and S410 with nearby LCDs in surface representation and (right) 12D D407, D409 and D410. Protein surfaces are colored according to chain identity. F. Density profiles in TDP-43 LCD condensates (peak at center) coexisting with dilute solutions for Wt, 12D, 5pS, 12pS and 12A from coarse-grained simulations with the implicit solvent coarse-grained HPS model. G, H. Snapshots of 12D condensate (G) and fragmented 12pS clusters (H) in simulations with the coarse-grained HPS model. Side view on elongated boxes (blue lines).

To characterize the interactions of TDP-43 LCDs further, we performed atomistic MD simulations of dense protein condensates (Fig 3C, Movie EV5) assembled with hierarchical chain growth (HCG; Pietrek et al, 2020) to enhance the sampling of polymeric degrees of freedom. In microsecond dynamics with explicit solvent and a highly accurate atomistic description of molecular interactions (Robustelli et al, 2018), we again found serine residues in the Wt protein to be more prone to interact with other protein residues than interacting with solvent (Fig 3D). By contrast, phosphomimicking aspartate side chains bind comparably more water molecules and show an overall reduced tendency for protein-protein interactions (Fig 3D and E, Appendix Fig S4A). Enhanced side chain solvation is consistent across the 12 phosphomimetic substitution sites (Appendix Fig S4B). The atomistic simulations are consistent with an increase in charge favoring solvated states and thus weakening TDP-43 condensates.

Effects of phosphomimicking mutations and phosphorylation on TDP-43 LCD phase behavior

To characterize possible differences between phosphomicking mutations and phosphorylation, we employed the highly efficient hydrophobicity scale (HPS) coarse-grained model (Dignon et al, 2018). The HPS implicit solvent model enabled us to quantify differences in the phase behavior of TDP-43 LCD variants. In line with experiments on full-length TDP-43 (Fig 1), 12D LCD phase-separated, but more protein remained in the dilute phase compared with Wt (Fig 3F–H). Indeed, computing the excess free energy of transfer ΔGtrans from the density profile (Appendix Fig S5D), which reports how favorable it is to move one chain from dilute solution at the saturation density to the dense phase of the condensate, showed that 12D LCDs are less prone to interact with each other in a condensate than Wt LCDs (Appendix Table S1). Loss of local contacts in the C-terminal region due to phosphomimetic substitutions was only partially compensated by new protein–protein interactions with arginines (Appendix Fig S5A–C), in accordance with coarse-grained simulations with explicit solvent (Fig EV3C). The 12A substitutions stabilized the TDP-43 LCD condensates, as expected based on our experiments, with little protein remaining in the dilute phase (Fig 3F). Phosphorylation modulates the stability of LCD condensates in a dose-dependent way. Attaching five phospho groups (5pS) led to a somewhat less-dense LCD condensate, but overall the excess free energy of transfer is on par with Wt (Appendix Table S1, Appendix Fig S5D). By contrast, fully phosphorylating all twelve sites (12pS) dissolved the LCD condensate in our simulations, with no clear peak in the density profile (Fig 3F). Overall, the simulations with the HPS model rank the saturation density to form condensates as 12A ≫ Wt ~5pS > 12D ≫ 12pS. The calculations thus predict that (i) phosphorylation may indeed dissolve condensates, and (ii) that phosphorylation may have an even stronger effect than phosphomicking substitutions, due to the larger negative charge of phospho-serine compared with aspartate.

C-terminal phosphomimetic substitutions do not impair nuclear import and RNA regulatory functions of TDP-43

Next, we turned to cellular experiments to investigate how C-terminal TDP-43 phosphorylation affects the behavior and function of TDP-43 in cells. As TDP-43 hyperphosphorylation is found in the disease state, it seems possible that this PTM has detrimental effects on the protein and contributes to mislocalization and/or malfunction of TDP-43, thus driving neurodegeneration. To address this possibility, we transiently expressed different myc-tagged TDP-43 variants (Wt, 12D, 12A) in HeLa cells and analyzed their intracellular localization, nuclear import and RNA processing functions. All three TDP-43 variants showed a predominantly nuclear steady-state localization (Fig EV4A). We also compared their nuclear import rates in a hormone-inducible reporter assay by live cell imaging (Hutten et al, 2020). In this assay, a protein-of-interest harboring a nuclear localization signal (NLS) is fused to a tandem EGFP and two hormone binding domains of the glucocorticoid receptor (GCR), which retains the reporter protein in the cytoplasm. Upon addition of a steroid hormone (dexamethasone), the reporter protein is released from the cytoplasm and imported into the nucleus, by virtue of the NLS in the protein-of-interest. We examined reporters containing the different TDP-43 variants (Wt, 12D, 12A) and found that their import rates were indistinguishable (Fig 4A and B).

Click here to expand this figure.

Figure EV4. Phosphomimetic substitutions do not alter nuclear localization, UG-rich RNA binding and autoregulation of TDP-43

Immunostainings showing nuclear localization of myc-TDP-43 Wt, 12D and 12A in HeLa cells. Endogenous TDP-43 expression was silenced by siRNAs, followed by transient transfection of the indicated siRNA-resistant myc-TDP-43 constructs. After 24 h, localization of TDP-43 Wt, 12D and 12A variants was visualized by TDP-43 immunostaining (mouse anti-TDP-43 antibody, Proteintech). G3BP1 (rabbit anti-G3BP1 antibody, Proteintech) and DAPI signal is shown to visualize the cytoplasm and nuclei, respectively. In the merge (right column), DAPI is show in turquoise, TDP-43 in green, and G3BP1 in magenta. Bar, 30 µm. Electrophoretic mobility shift assay (EMSA) of TDP-43-MBP-His6 variants (Wt, 12D and 12A) in a complex with (UG)12 RNA. Representative confocal images of U2OS cells stably expressing the indicated myc-TDP-43 variants (Wt, 12D and 12A) after siRNA KD of endogenous TDP-43 and induction of myc-TDP-43 expression with doxycycline. Cells were stained with mouse monoclonal anti-myc 9E10 antibody (IMB protein production facility) and DAPI. For clarity, signals were converted to grey values in the individual channels (upper two rows). In the merge (lower row), DAPI is shown in turquoise) and myc-TDP-43 is shown in green. Bar, 20 µm. Western Blot showing the expression levels of myc-TDP-43 variants in stable inducible Flp-In T-Rex U2OS cell lines before and after addition of doxycycline (dox). Samples were analyzed by SDS–PAGE and Western blot using a rabbit anti-TDP-43 N-term antibody (Proteintech, upper blot), mouse anti-myc 9E10 antibody (IMB protein production core facility), and rabbit anti-Histone H3 antibody (Abcam) to detect the loading control Histone H3. Quantification of TDP-43 autoregulation after dox-induced expression of myc-TDP-43 variants in U2OS cell lines. Values represent the mean ± SD of four independent experimental replicates (n = 4) of endogenous TDP-43 expression levels normalized to Wt (−Dox) condition. *P < 0.0332 and ***P < 0.0002 by one-way ANOVA with Šídák's multiple comparisons test of TDP-43 endogenous expression levels, comparing the respective non-induced (−Dox) and induced (+Dox) lines.

Source data are available online for this figure.

Figure 4. Phosphomimetic substitutions do not alter the rate of TDP-43 nuclear import and do not impair TDP-43 autoregulation, RNA-binding or alternative splicing function

Hormone-inducible nuclear import assay, representative still images of GCR2-EGFP2-TDP-43 Wt, 12D and 12A before and during import triggered by addition of dexamethasone. Images were live recorded by spinning disc confocal microscopy. Bar, 20 µm. Quantification of the hormone-inducible nuclear import measured during a total time course of 50 min. Values represent the mean fluorescence intensity of GCR2-EGFP2-TDP-43 in the cytoplasm for three independent replicates ± SEM (≥ 42 cells per condition). Phosphomimetic 12D TDP-43 is competent in autoregulating TDP-43 expression. SDS–PAGE followed by TDP-43 Western blot showing downregulation of endogenous TDP-43 through autoregulation (60) after 48 h expression of Wt, 12D and 12A variants in HeLa cells. TDP-43 was detected using rabbit anti-TDP-43 C-term antibody (Proteintech), Histone H3 (rabbit anti-Histone H3 antibody, Abcam) was visualized as a loading control. * denotes an unspecific band. Electrophoretic mobility shift assays (EMSA) of TDP-43-MBP-His6 variants (Wt, 12D and 12A) in a complex with TDP-43 autoregulatory RNA binding site (60). All TDP-43 variants form TDP-43-RNA complexes equally well. Splicing analysis by RT–PCR of known TDP-43 splice targets (SKAR exon 3 and Bim exon 3) in HeLa cells. Silencing of endogenous TDP-43 by siRNA leads to altered splice isoforms of SKAR and Bim (second vs first lane). These splicing alterations can be rescued by re-expression of TDP-43 Wt, but also 12D or 12A variants, demonstrating that phosphomimetic TDP-43 is fully competent in regulation splicing of these TDP-43 splice targets.

Source data are available online for this figure.

To assess whether hyperphosphorylated TDP-43 shows functional impairments in RNA processing, we first assessed its ability to autoregulate its own levels when transiently overexpressed in HeLa cells (Ayala et al, 2011; Avendano-Vazquez et al, 2012). However, endogenous TDP-43 was downregulated to the same degree by all three myc-TDP-43 variants (Fig 4C), indicating that hyperphosphorylated TDP-43 can normally bind to its own 3′UTR and autoregulate its own levels. In line with these findings, recombinant TDP-43 Wt, 12D and 12A showed comparable RNA binding in electrophoretic mobility shift assays (EMSAs) with in vitro transcribed RNA comprised of the autoregulatory TDP-43 binding site (Fig 4D) or synthetic (UG)12 RNA (Fig EV4B). Second, we examined splicing of two known TDP-43 splice targets that get mis-spliced upon loss of TDP-43 (Tollervey et al, 2011; Fiesel et al, 2012). After siRNA-mediated silencing of endogenous TDP-43 expression and re-expression of siRNA-resistant myc-TDP-43 Wt, 12D or 12A (Appendix Fig S6A), splicing of SKAR and Bim exon 3 were fully restored by all three TDP-43 variants (Fig 4E), indicating normal function of phosphomimetic TDP-43 in splicing regulation. Normal nuclear localization and autoregulation of TDP-43 were also replicated in a cellular system that avoids high overexpression and has homogenous expression levels, namely stable inducible Flp-In U2OS cell lines that express the different myc-TDP-43 variants (Wt, 12D and 12A) after overnight doxycycline addition (Fig