A single N‐terminal phosphomimic disrupts TDP‐43 polymerization, phase separation, and RNA splicing

TDP‐43 N‐terminal domain self‐assembly is disrupted by phosphomimetic variant S48E

Several groups have reported that TDP‐43 NTD mediates assembly into dimers or other higher‐order structures (Shiina et al, 2010; Chang et al, 2012; Wang et al, 2013), though the details of the assembly have only recently begun to be visualized with atomistic resolution (Afroz et al, 2017; Jiang et al, 2017; Mompean et al, 2017; Tsoi et al, 2017). Recombinant TDP‐43 NTD (in our case: residues 1–80, with only three additional N‐terminal residues, GHM, resulting from the cloning strategy and cleavage with TEV protease) is highly soluble when expressed and purified from bacteria. However, as described previously, increasing concentrations of NTD result in size‐exclusion chromatography (SEC) profiles smoothly shifting to smaller elution volume, which is consistent with higher apparent molecular weight (Chang et al, 2012). This continuous shift of a single peak that becomes more asymmetric at higher concentrations is consistent with weak (μM) interactions. Correspondingly, concentrations above 20 μM resulted in altered NMR chemical shifts and a decrease in NMR signals. These concurrent changes in chemical shifts, intensities, and the resulting low signal‐to‐noise ratio precluded previous direct characterization of the assembled state by NMR (Chang et al, 2012).

Inspired by the observation that pentamerization of nucleophosmin is regulated by phosphorylation (Mitrea et al, 2014, 2016), we investigated proteomic databases to determine whether weak self‐assembly of TDP‐43 might be regulated by post‐translational modification. TDP‐43 NTD contains one reported serine/threonine phosphorylation, pS48, which was found in multiple phosphoproteomic analyses of various cell lines (Rigbolt et al, 2011; Hornbeck et al, 2012, 2015). This site is highly conserved as a serine or threonine in all organisms with TDP‐43 orthologues examined except for Xenopus species (Appendix Fig S1A). In order to confirm that S48 is phosphorylated in cells, we created a custom polyclonal antibody serum against residues G40‐G53 of TDP‐43 that is specific to the peptide containing phosphorylated serine 48 (α‐TDP‐43 pSer48) (Fig 1A). HEK 293T cell lysate probed with α‐TDP‐43 pSer48 yields a band with a band at 43 kDa consistent with TDP‐43 (Fig 1B). Furthermore, phosphatase treatment of cell lysate or fractions immunoprecipitated with commercial α‐TDP‐43 antibody abrogates α‐TDP‐43 pSer48 immunoreactivity (Fig 1C), confirming that our custom serum is specific for phosphorylated TDP‐43 and suggesting that a small though measurable population of TDP‐43 is constitutively phosphorylated in these cells, consistent with the previous phosphoproteomic results. Because this position is in a region that showed large NMR chemical shift differences upon increasing concentration of protein, consistent with NTD oligomerization at this site (Fig 1G), we tested the hypothesis that phosphorylation could alter assembly using a phosphomimetic substitution, S48E. Surprisingly, SEC profiles of TDP‐43 NTD S48E are dramatically shifted from wild‐type, showing a longer retention time and symmetric elution profile consistent with a monomer (Fig 1D) and nearly complete lack of self‐assembly as indicated by a lack of concentration‐dependent chemical shift changes (Fig 1H). S48E TDP‐43 NTD appears well folded and nearly unchanged compared to the wild‐type by fingerprint (¹H ¹⁵N HSQC) NMR spectra (see below), suggesting that the lack of assembly is not due to unfolding or adoption of a different global structure. Therefore, these data are consistent with the region near S48 playing a role in assembly.

TDP‐43 NTD assembles into linear chains with a monomeric repeat unit

The disappearance of NMR signals and the continued elution peak shift in SEC as a function of concentration observed previously (Chang et al, 2012) suggests that the assembly formed by TDP‐43 NTD may be larger than a dimer. To determine the number of monomers making up the oligomeric form, samples were loaded onto SEC columns equipped with UV absorption and multi‐angle light scattering (SEC‐MALS) detectors at increasing concentrations. For TDP‐43 NTD at 150 mM NaCl, the elution volume decreases continuously with increasing injection concentration (Appendix Fig S1D) as observed previously for TDP‐43 NTD (Chang et al, 2012). The molar mass monitored over the eluting protein indicates an equilibrium between at least monomers and dimers. To test whether the NTD self‐associates into larger species than dimers, we performed in vitro cross‐linking experiments (Marzahn et al, 2016) of TDP‐43 NTD WT samples with the primary amine cross‐linker BS3. The cross‐linked products were separated on SDS–PAGE and appear as ladders of increasing molecular weight, indicating the degree of oligomerization. Cross‐linking was observed for all samples spanning concentrations of 10–200 μM. At low concentrations, monomers and dimers are observed while species up to heptamers are present at higher concentrations (Appendix Fig S1G). The NTD therefore self‐associates into higher‐order species and populates different oligomer sizes simultaneously.

To characterize the nature of this higher‐order self‐association of TDP‐43 NTD, we used the quantitative equilibrium methods composition gradient multi‐angle light scattering (CG‐MALS) and sedimentation equilibrium analytical ultracentrifugation (AUC). The results from both methods support a model of weak infinite linear self‐association, specifically isodesmic assembly resulting in unbranched linear chains. CG‐MALS data were collected for wild‐type TDP‐43 NTD in a wide range of concentrations up to more than 300 μM. The mass average molecular weight increases continuously as a function of the concentration of the wild‐type NTD (Fig 1E). Models for oligomerization into discrete dimers, trimers, or tetramers do not fit the data well. By contrast, an excellent fit can be achieved with an isodesmic self‐association model, in which each addition of a monomer to pre‐existing oligomers/polymers occurs with identical affinity, here with values of 95 μM for wild‐type NTD in the presence of 150 and 300 mM NaCl (Fig 1F). Higher‐order assembly is similar at the two different ionic strength conditions, suggesting that electrostatic interactions are not primarily responsible for stabilizing the polymeric NTD states. These results suggest self‐assembly of NTD monomers into linear polymers, with long polymers (up to 7–10mers at 300 μM protein concentration), short polymers, and monomers populated simultaneously in equilibrium (Appendix Fig S1B and C). S48E disrupts polymerization (Fig 1F, red), resulting in a K _D of approximately 2 mM assuming an isodesmic assembly process.

We observe evidence of the same self‐association behavior for TDP‐43 NTD in AUC experiments. In sedimentation velocity (SV) experiments, at least three species are observed when the data are modeled to determine the sedimentation coefficient distribution of the present species, and their fractional populations shift to larger species in a concentration‐dependent manner (Appendix Fig S1E). However, interconversion of multimeric species with different and unknown shape asymmetries precludes straightforward quantitative interpretation of SV‐AUC fits. Sedimentation equilibrium (SE) AUC profiles, which do not suffer from assumptions of shape, for TDP‐43 NTD WT protein in 150 mM NaCl fit well to an isodesmic self‐association model with a K _D ~ 40 μM (Appendix Fig S1F). The discrepancy between the values calculated by CG‐MALS and SE‐AUC is likely due to the difficulty of obtaining the precise equilibrium constant for a weak association.

NTD oligomerization contributes to TDP‐43 full‐length phase separation in vitro

We next sought to understand the role of TDP‐43 NTD oligomerization in the context of the full‐length protein. As described above, TDP‐43 NTD is highly soluble even up to 20 mg/ml in our hands. Although it self‐assembles into chains (Fig 1 and Appendix Fig S1), these interactions are relatively weak and NTD does not aggregate or undergo LLPS at any conditions tested. Conversely, TDP‐43 C‐terminal domain alone is sufficient to undergo LLPS at > 5 μM concentrations (Conicella et al, 2016). Therefore, we tested the hypothesis that TDP‐43 NTD polymerization enhances phase separation of full‐length TDP‐43 by providing additional interaction sites creating cooperativity between NTD self‐interactions and CTD self‐interactions. Supporting this notion is the recent finding that a TDP‐43 variant (where the RNA‐binding RRM domains are replaced by GFP, leaving intact NTD and CTD) readily undergoes LLPS in cells (Schmidt & Rohatgi, 2016). However, difficulties in purifying soluble, monomeric full‐length TDP‐43 have so far prevented detailed biochemical characterization of TDP‐43 full‐length phase separation in vitro (Molliex et al, 2015). Our own attempts using the N‐terminal maltose‐binding protein (MBP) solubilizing fusion tag that worked for the related ALS‐associated aggregation‐prone protein FUS (Burke et al, 2015) were not successful. Therefore, we instead created a bacterial expression construct with a TEV cleavable C‐terminal MBP fusion, placing the solubilizing MBP protein directly adjacent to the aggregation‐prone C‐terminal domain. We combined this refined strategy with the purification of the fusion protein under high salt conditions to reduce nucleic acid binding and further promote solubility. Purification of this construct via column nickel affinity chromatography followed by SEC results in a significant fraction of the fusion protein eluting at a position consistent with monomer or low‐order oligomer (Appendix Fig S2A). Immediately of interest, the low molecular weight fraction of full‐length, wild‐type TDP‐43 elutes with a skewed peak consistent with assembly, while the same region of the full‐length S48E fusion elutes with a symmetric peak at a longer retention time, consistent with monomeric NTD (Appendix Fig S2A). This observation directly mimics what we observed for TDP‐43 NTD alone (Fig 1D), suggesting that S48E disrupts the dynamic equilibrium for full‐length TDP‐43 fusion protein assembly. Importantly, these fractions of both full‐length wild‐type and S48E are essentially devoid of nucleic acid as measured by ratio of UV absorbance at 260–280 nm, suggesting they will be useful for evaluating the behavior of full‐length TDP‐43 assembly and binding.

TDP‐43 is a highly abundant protein, present at up to 3 μM in human cell lines assuming a uniform distribution across an entire cell volume (Ling et al, 2010; Beck et al, 2011), and known to be even more concentrated in the nucleus where it is primarily localized (Neumann et al, 2006). We therefore tested the ability of full‐length purified low molecular weight/monomeric TDP‐43 to undergo LLPS at low micromolar range concentrations. Addition of TEV protease to cleave the solubilizing C‐terminal MBP fused to 2.5 μM full‐length TDP‐43 wild‐type results in the formation of turbidity at physiological salt concentration (150 mM NaCl) (Fig 2A). Turbidity is absent before TEV addition (Appendix Fig S2B, time zero), and the increase in turbidity is due to robust phase separation, as observed by the formation of round liquid micron‐sized phases seen by differential interference contrast (DIC) microscopy (Fig 2B). In contrast to wild‐type full‐length TDP‐43, the single‐point variant S48E fails to phase separate at these conditions (Fig 2A, top B), suggesting that TDP‐43 NTD polymerization contributes to full‐length TDP‐43 phase separation. Phase separation of full‐length TDP‐43 S48E (and wild‐type) can be observed at higher protein concentrations (Fig 2B, bottom), implying that the S48E variant increases the critical protein concentration required for phase separation rather than disrupt it entirely. These data support a model where the C‐terminal domain, which can readily phase separate on its own, is primarily responsible for mediating a network of multivalent contacts and that additional interactions between NTDs enhance phase separation. Interestingly, as observed for the C‐terminal domain alone (Conicella et al, 2016), after 2‐h incubation these assemblies appear irregularly shaped, no longer flowing and fusing, consistent with aggregation within the LLPS state (Fig 2B).

NTD oligomerization contributes to TDP‐43 in cell phase separation and RNA splicing

Ribonucleoprotein granules and membraneless organelles assembled via LLPS of their components have been proposed to act as functional centers for RNA processing (Brangwynne et al, 2009, 2011; Hyman et al, 2014). Our data above suggest that polymerization of the TDP‐43 NTD may contribute to this function/process as well. Indeed, it has been pointed out recently that the cooperation and interactions between the various domains of RNA‐binding proteins have only begun to be systematically probed (Wei et al, 2016). We thus sought to test the effect of NTD polymerization on TDP‐43 assembly and function in cells. Using a reporter assay for TDP‐43 intranuclear phase separation (Schmidt & Rohatgi, 2016), we determined the contribution of NTD interactions on TDP‐43 phase formation by comparing wild‐type and S48E forms of TDP‐43 where the RNA‐binding domains are replaced by GFP, termed here TDP‐43_RRM‐GFP. As described previously, when expressed in human (293T) cells, TDP‐43_RRM‐GFP localizes to the nucleus and concentrates into a single large liquid–liquid phase separated assembly per cell over time (Schmidt & Rohatgi, 2016; Fig 2C and D). We hypothesized that the phosphomimetic S48E variant would alter the assembly behavior and phase properties of these model TDP‐43 phases. When expressed under identical conditions (Fig 2E), S48E TDP‐43_RRM‐GFP also forms nuclear assemblies; however, these phases differ from wild‐type TDP‐43_RRM‐GFP in several critical parameters. Most strikingly, the S48E TDP‐43_RRM‐GFP phase is much less viscous than those made by the wild‐type TDP‐43_RRM‐GFP (Fig 2E and G). This observation is consistent with reduced interprotein contacts for S48E compared to wild‐type TDP‐43_RRM‐GFP. We also find that this effect of S48E on droplet viscosity appears to be “dominant‐negative”—when S48E and wild‐type are co‐expressed (Fig 2F and H), S48E/wild‐type TDP‐43 mixed phases show rapid kinetics like S48E‐alone phases, suggesting that they are well mixed at the molecular level and co‐constitute the phase. The dominant fluidizing effect of S48E suggests that wild‐type NTD normally engages in chain‐forming interactions that are disrupted in the presence of co‐expressed S48E variant. Additionally, although qualitatively the phases formed by S48E and wild‐type alone appear similar [and expression levels are indistinguishable (Fig 2D and Appendix Fig S2C and D)], quantitatively S48E TDP‐43_RRM‐GFP partitions less readily into these phases and forms a slightly less dense concentration of TDP‐43 inside the phase, as seen by the relative quantification of TDP‐43_RRM‐GFP fluorescence intensity in the surrounding nucleoplasm and inside the phase, respectively (Appendix Fig S2E). Given that the wild‐type NTD appears to stabilize TDP‐43_RRM‐GFP, we also tested the hypothesis that phase separation would occur earlier in wild‐type TDP‐43_RRM‐GFP transfected cells compared to S48E. Although transfection heterogeneity precluded quantification of kinetics, wild‐type TDP‐43_RRM‐GFP does form liquid phases earlier than S48E in a time course after transfection (see Appendix Fig S2F for representative images). Taken together, these data suggest that NTD polymerization contributes to TDP‐43 phase separation in cells.

Given that the phase separation of TDP‐43 in cells is altered by the S48E variant that controls TDP‐43 NTD polymerization, we tested whether TDP‐43 function is also altered by S48E. Previously, a correctly folded NTD has been shown to be important for TDP‐43 splicing function (Zhang et al, 2013). To determine the effect of the Ser48 phosphomimetic variant (S48E) on TDP‐43 function, we analyzed TDP‐43 splicing regulatory activity in HeLa cells using the well‐established cystic fibrosis transmembrane conductance regulator (CFTR) exon 9 minigene reporter (Buratti et al, 2001; Ayala et al, 2006). The levels of splicing are quantified as the percent of exon inclusion (Appendix Fig S2G), and the splicing activity is measured as the ratio of percent exon inclusion relative to WT (Fig 2I). As expected, siRNA‐mediated knock‐down of TDP‐43 greatly decreased the splicing activity (Fig 2I and Appendix Fig S2G). The RNA‐binding‐deficient mutant F147/149L (Buratti & Baralle, 2001) used as control failed to rescue TDP‐43 activity. Compared to wild‐type, S48E showed approximately 40% reduction in regulatory function (Fig 2I), similar to other variants disrupting TDP‐43 NTD assembly, as recently demonstrated (Afroz et al, 2017; Jiang et al, 2017; Mompean et al, 2017). The corresponding Ala substitution (S48A) also decreased TDP‐43 splicing regulatory activity, by approximately 25%, consistent with decreased self‐association in vitro (see below). These findings suggest that the details of the NTD structure and interactions in the region surrounding S48, not just phosphorylation at S48, are critical for mechanisms that mediate TDP‐43 control of splicing. To further determine the requirement for NTD oligomerization in TDP‐43 splicing regulatory function, we studied NTD Y4R and E17R, two other self‐assembly deficient variants (see below), in our cellular assays. We observed a dramatic decrease in TDP‐43 activity upon introduction of both mutations (Fig 2I) and in particular E17R, which shows approximately 80% loss in activity. These findings, combined with our structural data showing that S48A and S48E disrupt polymerization, strongly suggest that TDP‐43 acts as a polymer or oligomer during splicing regulation and that this activity may be modulated by modification of S48 or other interfacial residues.

Structural details of TDP‐43 N‐terminal domain head‐to‐tail polymerization

Because the polymerization of TDP‐43 NTD regulates in cell phase separation and splicing function, we next sought to characterize the structural details of NTD polymerization and the structural mechanism by which interface modifications control assembly and function. Because the formation of a dynamic, high molecular weight NTD polymer makes direct NMR structural determination difficult (Chang et al, 2012), we began by first distinguishing the repeat unit for NTD polymerization in solution. The CG‐MALS and NMR data we collected (Fig 1E and H) are consistent with a repeat unit of a monomer with two distinct interfaces. Therefore, one possibility is that TDP‐43 NTD self‐assembles into a linear chain with a head‐to‐tail assembly. Unlike a symmetric dimer where the same interface mediates self‐interaction on both monomers, a head‐to‐tail assembly will have two interfaces—one interface, the “head”, that forms contacts with the other monomer via a distinct region or second interface, “the tail”. Above we demonstrated that S48E disrupts polymerization and lies in one interface identified by NMR chemical shift deviations of wild‐type NTD (Fig 1H and I). Therefore, we expected that TDP‐43 NTD S48E would only be able to bind the TDP‐43 NTD wild‐type interface that includes S48 and not the other interface. Hence, NMR spectra observing (¹⁵N‐labeled) TDP‐43 NTD wild‐type titrated with increasing concentrations of NMR‐invisible (natural isotopic abundance) S48E TDP‐43 NTD would show chemical shift deviations only near S48. Indeed, chemical shift deviations are only observed in the C‐terminal half (residues 40–80) of wild‐type (Fig 3A, top). The complementary experiment where increasing quantities of NMR‐invisible wild‐type combined with NMR visible (¹⁵N‐labeled) TDP‐43 NTD S48E shows chemical shift deviations at residues in the N‐terminal half (residues 1–35) of the S48E protein (Fig 3A, bottom). Therefore, this experiment identifies the residues in the N‐terminal region as participating together in the second interface. Only residues 75–80 show significant chemical shifts in both titrations, suggesting these residues are near or connected to both interfaces. This paradox may be due to these residues forming part of the final β‐strand that spans the width of the protein and contacts both interfaces. Hence, an allosteric connection may link residues 75–80 to both interfaces. Furthermore, the structure of NTD residues 1–77 solved by Mompean and coworkers (Mompean et al, 2016) (PBD ID: 2N4P) depicts V75 forming a hydrogen bond with residue of E9, the latter of which is in the N‐terminal interface. Therefore, the CSD observed in the titration of either ¹⁵N S48E with wild‐type (or ¹⁵N S48E with Y4R, see below) may be the consequence of the conformational change caused by the interaction occurring in the N‐terminal interface. These residues are the only outliers from the single interface behavior observed for other residues in the domain. Collectively, these data immediately point to a head‐to‐tail assembly (Fig 3B), where S48E reduces the affinity at the interface.

In order to further define the interface of the self assembled NTD, we also measured intermolecular ¹H_N paramagnetic NMR relaxation enhancements (PREs) that are high for positions in close proximity to a stabilized nitroxide spin label, even in transiently populated states. Assuming a two‐state monomer–dimer equilibrium, the relative orientation of two NTD monomers in the assembled dimer state can be discerned given a sufficient number of samples with conjugated nitroxide radical spin labels at single positions spatially separated across the monomer surface. Therefore, MTSL spin labels were conjugated to the NTD at engineered cysteine variants at positions 2 or 29 or at native cysteine positions 39 or 50 and mixed with ¹⁵N isotopically enriched NTD C39S/C50S in order to measure ¹H_N transverse paramagnetic relaxation enhancements, Γ₂, arising only from intermolecular contacts (Fig 3C). Addition of MTSL spin labels did not result in perturbations to the native monomeric structure as confirmed by ¹H‐¹⁵N HSQC spectra (Appendix Fig S3A). Comparing the CSDs of each PRE variant to the diamagnetic spin label analog (AcMTSL)‐labeled species suggested that cysteine variants and addition of spin labels did not disrupt the self‐assembly for S2C and C50 and decreased affinity but maintained a similar pattern of chemical shifts compared to the reference for the variants of S29C and C39 (Appendix Fig S3B and C). The PRE measurements then provide complementary contact details, with C39 and C50 reporting on the C‐terminal interface, and S2C and S29C reporting on the N‐terminal interface. Interestingly, the cysteine‐free C39S/C50S variant shows similar chemical shift patterns but approximately 3× weaker assembly compared to wild‐type in reducing conditions (Appendix Fig S3B), suggesting that cysteine side chains have some contribution to assembly, though no evidence of disulfide dimers in the wild‐type is observed, as expected for reducing (1 mM DTT) conditions.

For each MTSL spin label, we observed a specific and distinct pattern of PREs that could each be mapped to a distinct cluster of residues on the NTD surface (Fig 3C). Data from C39 and S29C labels together suggest that the α‐helix in N‐terminal interface contacts with the β‐strands in the C‐terminal side, which forms the upper part of the interface. The PREs from C50 and S2C reveal that the lower part of the interface mainly involves the loop of N45‐G53 from the C‐terminal side and the antiparallel beta‐strands on the N‐terminal side. Using rigid‐body docking (Clore & Schwieters, 2003; Zwolak et al, 2010) of the published solution NMR structure of residues 1–77 of the NTD monomer (PDB 2N4P) based on ambiguous distance restraints generated from CSP and PRE data, we generated a low‐resolution model of the NTD dimer interface. The model that best fits the experimental data depicts a head‐to‐tail interface with approximately a translation of one monomer width (Fig 3B). The head interface is comprised of N‐terminal residues including those in β‐strands 1 and 2 as well as the α‐helix. The tail interface, on the other hand, is comprised of C‐terminal residues, including the β‐strands and the S48‐V54 loop.

Structure of a stable TDP‐43 NTD dimer by mixing two distinct interface‐disrupting variants

Based on our model of the NTD dimer interface, we hypothesized that mutations introduced on the head interface (residues 1–30) would inhibit self‐assembly as potently as S48E. In order to test this hypothesis, we generated NTD variants E17R and Y4R and examined their effects on self‐assembly. When the NMR chemical shift deviations for increasing concentrations of each mutant alone were compared to wild‐type, we found that E17R and Y4R NTD variants do not disrupt folding (Appendix Fig S4A) but do disrupt self‐assembly as measured by SEC elution volume (Appendix Fig S4B), lack of NMR chemical shift deviations upon increasing concentration (Appendix Fig S4C), and disruption of splicing function (Fig 2I), consistent with E17 and Y4 positions being in the interface (Fig 3A and B). Given that Y4R and S48E each block distinct interfaces, we hypothesized that mixing these two variants together would result in saturation of the dimer state without the formation of higher‐order oligomers. Furthermore, this dimer has an expected molecular weight of 20 kDa and therefore should be readily visible by standard solution NMR techniques. Titration of increasing concentrations of unlabeled Y4R into 100 μM NMR visible (¹⁵N labeled) S48E resulted first in disappearance of peaks due to the kinetics of exchange between monomer and complex (Fig 4A), as observed in wild‐type (Fig 1G). At millimolar concentrations of the binding partner, the resonances reappear with very large chemical shift differences from the unbound positions, consistent with saturation of a complex with micromolar affinity [as observed for EIN:HPr complex (Fawzi et al, 2010)]. Significant NMR chemical shift deviations occur on the N‐terminal side of ¹⁵N S48E due to addition of Y4R and on the C‐terminal side of ¹⁵N Y4R due to addition of S48E (Appendix Fig S4D). The residues with large CSD in the titration of ¹⁵N S48E with Y4R, such as S2 and T32 (Fig 4B left), did not experience significant shifts in the titration of ¹⁵N Y4R with S48E (Fig 4B right), which is additional evidence that there are two distinct interfaces involved in the formation of the asymmetric dimer. Together, these data demonstrate that S48E and Y4R form an asymmetric dimer complex, rather than polymers, that preserves the wild‐type head‐to‐tail interface.

In order to determine the atomic resolution structure of the S48E/Y4R dimer complex, we performed ¹³C/¹⁵N‐edited and ¹³C/¹⁵N‐filtered/edited NOESY measurements using the ¹³C/¹⁵N‐labeled S48E (or Y4R) complexed with natural abundance Y4R (or S48E) (Appendix Fig S5A). The filtered NOESY allowed us to specifically probe the intermolecular contacts involved in the dimer interface. Most of intermolecular NOEs arise from the interfacial residues, S29, T30, A33, and Q34 on the α‐helix and I16‐P19 on the β‐strand 2 in the N‐terminal half and C39‐L41, C50‐G53 on β‐strand 5, V54‐ L56 on β‐strand 6 and terminus residues N76‐D80 in the C‐terminal half, consistent with the interface determined by CSDs and PREs (Fig 3B and Appendix Fig S5B). With the intra‐ and intermolecular NOEs used as distance restraints, we calculated the dimer structure constructed by Y4R and S48E, shown in Fig 4C. The α‐helix (on the N‐terminal half of S48E) lies against the β‐strands on the C‐terminal of Y4R. The first half of the α‐helix, including residues S29 and T30, mainly contacts V54, R55, and L56 on β‐strand 6 (Fig 4D). A33, on the second half of α‐helix, contacts C39, G40L41 (Fig 4D). The residue at the end of α‐helix, Q34, as well as adjacent F35 and P36 interacts with the C‐terminal residues N76, P78, and K79 (Fig 4E). In the lower part of interface, the β‐strand 2 of I16‐P19 (N‐terminal) forms a parallel sheet with β‐strand 5, residues C50‐G53 (C‐terminal) (Fig 4F).

The structure of each subunit is largely consistent with the previously published NTD monomer structure of 2N4P (residues 1–77 Cα RMS deviation = 1.48 Å; Appendix Fig S5E; Mompean et al, 2016), except for deviations at β‐strands 4 and 5 and the extreme C‐terminal residues. Additionally, our NTD dimer structure is even more similar to a recent 2.1 Å resolution X‐ray crystal structure of a TDP‐43 NTD (5MDI) (Afroz et al, 2017) where the asymmetric unit comprises two nearly identical monomers (Cα RMS deviation of 0.6 Å) with a similar head‐to‐tail orientation which forms a polymer by crystallographic symmetry. The Cα RMS deviation of the individual Y4R and S48E subunits in our structure and the corresponding monomer subunits in the crystal structure is low, 0.899 and 1.357 Å, respectively, indicating agreement of the NTD monomer component structures. However, the overall Cα RMS deviation for our dimer and the asymmetric unit that comprises the dimer in the crystal structure is 2.033 Å. The larger deviation is due mainly to differences in the orientation of the interface (Appendix Fig S5D)—primarily a rotation of −17° about an axis approximately perpendicular to the surface and translation of ~2 Å (Appendix Fig S5D).

Cloning, expression, purification of recombinant proteins

The DNA sequence encoding TDP‐43 with an N‐terminal Thioredoxin leader sequence, 6xHis (Thio‐His) tag, and TEV cleavage site (Peti & Page, 2007) was codon optimized and synthesized by DNA2.0 in the pJ411 bacterial expression vector (designated pJ411/TDP‐43). In order to generate the expression construct for the TDP‐43 NTD, the codon for residue N81 was mutagenized by Quikchange (Agilent Genomics) to a stop codon.

The expression vector containing a cleavable C‐terminal maltose‐binding protein (MBP) fusion tag (designated pJ4M) was constructed using overlap extension PCR (Horton et al, 1989) of MBP and pJ411 DNA fragments. The DNA sequence encoding for MBP was amplified from an N‐terminal MBP fusion construct optimized for bacterial expression (Peti & Page, 2007) using primers 5‐TTCCAGGGTAGCGGTACCAAAATCGAAG‐3 and 5‐GGGTTATGCTAGGGGGTCAATGATGGTGATGGTGATGGGTACCGCCTTCAGC‐3. The pJ411 DNA sequence was amplified from the pJ411/TDP‐43 construct using primers 5‐GCTGAAGGCGGTACCCATCACCATCACCATCATTGACCCCCTAGCATAACCC‐3 and 5‐GTACAGGTTCTCCTCGAGTCATCACATATGGGTATATCTCCTTCAAAAGTTAAAC‐3.

The TDP‐43‐MBP expression construct (designated pJ4M/TDP‐43) was created by subcloning the DNA sequence for TDP‐43, which was amplified from pJ411/TDP‐43 by PCR using primers 5‐TAATACGACTCACTATAGGG‐3 and 5‐TCATCACTCGAGCATGCCCC‐3, into pJ4M via restriction sites NdeI and XhoI.

Point mutations were introduced either by Quikchange mutagenesis or by overlap extension PCR with mutagenic primers. Proteins were overexpressed in Escherichia coli BL21 Star DE3 (Life Technologies). Cells were grown in LB media for natural abundance isotope labeling. For expression of [U‐99% ¹³C, 99% ¹⁵N] isotope labeling, cells were grown in M9 minimal media supplemented with ¹³C‐glucose and/or ¹⁵NH₄Cl (Cambridge Isotope Laboratories). In order to purify the TDP‐43 NTD, cell cultures were grown at 37°C to an OD600 of 0.6–0.9. Protein expression was induced by addition of isopropyl β‐D‐1‐thiogalactopyranoside (IPTG) (GoldBio) to 1 mM, followed by incubation at 37°C for 4 h. Cells were harvested by centrifugation, resuspended in NTD binding buffer (20 mM Tris–Cl pH 8.0, 500 mM NaCl, 10 mM imidazole, 1 mM DTT) supplemented with complete EDTA‐free protease inhibitor cocktail (Roche), and lysed using an EmulsiFlex C3 (Avestin). Cell lysates were clarified by centrifugation and applied to a 5 ml Histrap HP column (GE Healthcare). The column was washed with five volumes of binding buffer, and TDP‐43 NTD was eluted with a linear gradient of NTD elution buffer (20 mM Tris–Cl pH 8.0, 500 mM NaCl, 500 mM imidazole, 1 mM DTT). Cleavage and removal of the N‐terminal Thio‐His tag was achieved by incubation with 0.03 mg/ml TEV protease concomitant with dialysis into NTD dialysis buffer (20 mM Tris–Cl pH 8.0, 200 mM NaCl, 1 mM DTT), followed by a subsequent purification with the 5 ml Histrap HP column. TDP‐43 NTD was further purified by SEC over a Superdex 75 26/60 column (GE Healthcare) with NTD SEC buffer (20 mM Tris–Cl pH 8.0, 50 mM NaCl, 1 mM DTT). Purified TDP‐43 NTD was concentrated to ~500 μM, flash frozen, and stored at −80°C.

For purification of full‐length TDP‐43‐MBP fusion protein, cell cultures were grown at 37°C to an OD600 of 0.6–0.9. Cell cultures were then cooled to 16°C, after which protein expression was induced with addition of IPTG to 1 mM, followed by incubation at 16°C overnight. Cells were harvested by centrifugation, resuspended in TDP‐43 binding buffer (20 mM Tris–Cl pH 8.0, 1 M NaCl, 10 mM imidazole, 10% (v/v) glycerol, 1 mM DTT) supplemented with complete EDTA‐free protease inhibitor cocktail, and lysed with the EmulsiFlex C3. Cell lysates were clarified by centrifugation, applied to the 5 ml Histrap HP column, and washed with five volumes of TDP‐43 binding buffer. TDP‐43‐MBP was eluted with a linear gradient of TDP‐43 elution buffer (20 mM Tris–Cl pH 8.0, 1 M NaCl, 500 mM imidazole, 10% (v/v) glycerol, 1 mM DTT).

Eluant was further purified by SEC over a Superdex 200 26/60 column (GE Healthcare) with TDP‐43 SEC buffer (20 mM Tris–Cl pH 8.0, 300 mM NaCl, 1 mM DTT). Purified TDP‐43‐MBP was concentrated to 250 μM, flash frozen, and stored at −80°C.

NMR spectra processing

NMR spectra were processed with NMRPipe (Delaglio et al, 1995) and analyzed with Sparky (T. D. Goddard and D. G. Kneller, SPARKY 3, University of California, San Francisco).

Backbone amide resonance assignments for the C39S/C50S were determined using standard Bruker three‐dimensional triple‐resonance experiments: HNCA, HNCO, HN(CA)CO, CBCA(CO)NH, and HNCACB.

NMR chemical shift deviations

¹⁵N and ¹H chemical shifts were measured from the crosspeaks on the HSQCs of 5, 20, 100, 200, 350 and 500 μM WT and 5, 100 and 200 μM variants in the condition of 20 mM HEPES pH 6.8, 1 mM DTT, and 10% D₂O. Chemical shift deviations for ¹H and ¹⁵N were quantified by subtracting the chemical shift of the 5 μM sample from the respective value of each other samples. The average chemical shift deviations (CSDs) were calculated by the equation:

For the titrations of S48E and Y4R, the 100 μM of either ¹⁵N S48E was titrated by 50 μM, 100 μM, 400 μM, 1 mM, 1.6 mM, and 2 mM of natural isotopic abundance Y4R. ¹⁵N Y4R was titrated by 50 μM, 100 μM, 400 μM, 800 μM, 1 mM, and 2 mM of natural isotopic abundance S48E. The CSDs for ¹⁵N and ¹H were quantified by subtracting the chemical shifts of 100 μM ¹⁵N S48E (or ¹⁵N Y4R) from the respective value of each mixture samples and then calculated by the equation above.

¹H‐¹⁵N HSQCs were acquired at 500 MHz ¹H Larmor frequency with 32 transients per data point with 64* and 1,536* complex pairs in the indirect ¹⁵N and direct ¹H dimensions, with corresponding acquisition times of 35 and 219 ms, and sweep widths of 36.0 and 14.0 ppm centered at 116.5 and 4.7 ppm, respectively.

NMR paramagnetic relaxation enhancement

Since TDP‐43 1–80 contains two intrinsic cysteines, C39 and C50, either cysteine was substituted by serine to generate the variants C39S and C50S, respectively. Other PRE variants including single cysteine substitution at S2 and S29 were engineered from the construct of C39S C50S TDP‐43 1–80. One mM MTSL (Toronto Research Chemicals, O875000) or 1 mM AcMTSL (diamagnetic analog of MTSL, Toronto Research Chemicals, A167900) was added into 50 μM of each variant protein under the condition of 20 mM Tris–Cl pH 8.0, 150 mM NaCl. After incubating 1 h at room temperature, the excess label was removed by HiPrep desalting column and the protein was exchanged to the buffer of 20 mM HEPES pH 6.8. Finally, the paramagnetic probe MTSL (or the diamagnetic acetylated MTSL analog, AcMTSL) was conjugated at C2, C29, C39, and C50, respectively. The samples were concentrated and flash frozen. For the intermolecular PRE measurement, 100 μM MTSL (or AcMTSL)‐conjugated PRE variant was mixed with 100 μM ¹⁵N‐labeled C39S/C50S in 20 mM HEPES pH 6.8, 10% D₂O.

Backbone amide proton transverse relaxation rate constant, ¹H_N R _2, were measured at 500 MHz ¹H frequency for paramagnetic and diamagnetic samples, with 192* and 1,536* complex pairs in the ¹⁵N indirect and ¹H direct dimensions, corresponding acquisition times of 105 and 21.9 ms, and sweep widths of 36 and 14 ppm centered at 116.5 and 4.7 ppm, respectively. Each ¹H_N R ₂ experiment comprised three interleaved ¹H_N R ₂ relaxation delays: 0.1, 2.1, 5.1 ms. PRE values (Γ₂ = ¹H_N R ₂ ^para − ¹H_N R ₂ ^dia) were obtained from the data collected with 0.1‐ and 2.1‐ms delays using the two time point method as described previously (Iwahara et al, 2007).

NMR structure calculation

NMR experiments were carried out at 25°C on Bruker Avance 500 or 850 MHz spectrometer on two separate samples of TDP‐43 NTD dimer described below. Our first sample was made up of 0.7 mM ¹³C, ¹⁵N‐labeled S48E 1–80 NTD mixed with 2 mM non‐labeled Y4R, while our second sample was reciprocally prepared with labeled 0.7 mM ¹³C, ¹⁵N‐labeled Y4R mutant mixed with 2 mM unlabeled S48E mutant. Both samples were prepared in identical buffer comprising of 20 mM HEPES pH 6.8 and 1 mM DTT. Each of these samples contains roughly 92% dimer species and facilitated NMR observation of single subunit of the dimer. Practically, complete NMR resonance assignments were attained using triple‐resonance experiment performed using standard pulse programs available in Topspin 3.2 software (HNCA, HNCO, CBCACONH, HCCH‐COSY, HCCH‐TOCSY, HBHACONH, hbCBcgcdceHE, and hbCBcgcdHD). Distance restraints were derived from sets of 3D‐ NOESY experiments acquired with mixing time of 120 ms. Isotope‐filtered NOESY experiments (¹³C, ¹⁵N‐filtered/edited 3D NOESY‐¹H‐¹⁵N/¹H‐¹³C HSQC) were performed to identify intermolecular NOE, which were further classified in strong, medium and weak restrains corresponding to 3.5, 4.5, and 6.0 Å upper limits. Dihedral angle restraints were calculated by using software TALOS+ (Shen et al, 2009), while hydrogen bond restrains were identified from hydrogen–deuterium exchange and interpretation of NOESY patterns. Semi‐automated NOESY assignment and initial structure calculation was performed using software CYANA (version 3.9) (Guntert, 2004), and final water refinement was carried using software Xplor‐NIH (version 2.34) (Schwieters et al, 2003). The NMR assignments and structure coordinates for TDP‐43 NTD dimer between S48E and Y4R mutants are deposited under BMRB entry number 30345, and PDB ID number 6B1G.

Turbidity measurements

Turbidity was quantified by absorbance of 600 nm wavelength light for wild‐type or S48E TDP‐43‐MBP diluted to 2.5 μM in 20 mM HEPES, 1 mM DTT, pH 7.0 with 150 mM, 300 mM, or 500 mM NaCl. Phase separation was initiated by addition of TEV protease to 0.003 mg/ml, and turbidity was monitored over a 12‐h time period at 5‐min intervals using a Spectra Max M5 microplate reader. All measurements were performed in triplicate.

Microscopy of in vitro TDP‐43 phase separation

Full‐length TDP‐43‐MBP was diluted to 2.5 μM or 20 μM in 20 mM HEPES, 1 mM DTT, pH 7.0 supplemented with 150–300 mM NaCl. Phase separation was initiated by the addition of TEV protease to 0.006–0.030 mg/ml, and phase separation was monitored by DIC micrographs acquired at 30‐min intervals over a 2‐h time period. All images were acquired with a Zeiss Axiovert 200M with a 40× objective with a numerical aperture of 0.75. Images were processed with ImageJ.

Modeling of the dimer interface

A molecular model for the TDP‐43 NTD dimerization interface was generated with XPLOR‐NIH (Schwieters et al, 2003). The published solution structure of the TDP‐43 NTD (Mompean et al, 2016) (PBD ID: 2N4P), with the first 12 residues encompassing the His tag removed, was duplicated and randomly rotated a maximum of 360° and offset by 10–100 Å using the XPLOR‐NIH python interface. This process was repeated 100 times in order to generate 100 randomized initial structures used for the IVM rigid body, torsion angle docking procedure (Schwieters & Clore, 2001). The dimerization interface was represented by highly ambiguous distance restraints derived from ¹H‐¹⁵N chemical shift perturbations as described previously (Clore & Schwieters, 2003). Cutoffs of 0.02 and 0.01 ppm were used to define contact residues on the WT and S48E binding surfaces, respectively. Residues without solvent exposed side chains were excluded, as were C‐terminal residues of S48E due to their disorder and relatively large distance from the interface. Relative orientation of the monomer subunits was restricted using ambiguous distance restraints derived from ¹H_N PRE data (Zwolak et al, 2010), with paramagnetic spin labels positions at S2, S29, C39, or C50, using a cutoff of > 10 s⁻¹. A hard square well‐potential term with r⁻⁶ sum averaging was used for both CSP‐derived and PRE‐derived distance restraints, with lower and upper cutoffs of 1.2–5 and 1.2–20 Å, respectively. A radius of gyration potential term corresponding to 2.2N^0.38–1, where N is the number of residues in the dimer (Kuszewski et al, 1999), was geometrically scaled over the course of the simulated annealing protocol to ensure compaction of the monomeric subunits in the dimer model. Each subunit was treated as a rigid body, with rotational and translational degrees of freedom. The lowest energy structure was chosen to represent the TDP‐43 NTD dimer model.

CG‐MALS data collection and analysis

Light‐scattering measurements for TDP‐43 NTD were performed in a buffer containing 20 mM HEPES pH 6.8, 150 mM NaCl, and 1 mM DTT, for NTD WT also at 300 mM NaCl. Measurements were collected using a Calypso system (Wyatt Technology Corporation), which uses a software‐controlled multiple syringe pump to create a concentration gradient, and a DAWN‐HELEOS multi‐angle light‐scattering photometer to collect data from the incoming sample stream. Static light‐scattering data were collected at 14 scattering angles as a function of protein concentration. For each injection, the solution was allowed to come to equilibrium for 60 s. Equilibrium MALS data as a function of concentration were analyzed as described by Attri et al (2010), and data obtained at all angles in a single experiment were combined for subsequent model generation. Simple monomer–dimer, monomer–trimer, etc., interactions and isodesmic self‐association models were used to model the data. In a solution, in which the different scattering species X_i correspond to different association states of a single protein, the theory of Rayleigh scattering from multicomponent solutions yields the concentration‐dependent Rayleigh ratio R

in which M _x is the molar mass of protein X, and [X _i] is the concentration of the species X _i. R is normalized to an optical constant K* defined as

where n ₀ denotes the refractive index of the solvent, λ₀ the vacuum wavelength of incident light (690 nm), N _A, Avogadro's number, and dn/dc is the refractive index increment of the sample (a standard value for proteins of 0.185 was used). The concentrations of each species are related to the equilibrium constants and total protein concentration, resulting in the following equations for typical monomer–dimer, monomer–trimer, monomer–i‐mer equilibria:

Above, i = 1 represents the free monomer, the total molar concentration [X]_total is known at each gradient injection, and R(0)/K* is measured. Non‐linear least square optimization is used to obtain a single K _A value that fits the data across the entire concentration range of interest.

To describe isodesmic self‐association, we used equations previously described (Attri et al, 2010):

Fitting parameters were K _A (i.e., 1/K _D) and the molecular weight of the building block, which was in agreement with a TDP‐43 NTD monomer within error.

SEC‐MALS data collection and analysis

The SEC multi‐angle light‐scattering (SEC‐MALS) experiments were carried out using a WTC‐0150S5 (MW range 500–150,000 Da) size‐exclusion column (Wyatt Technologies, Santa Barbara, CA, USA) with three detectors connected in series: an Agilent 1200 ultraviolet (UV) detector (Agilent Technologies, Santa Clara, CA, USA), a Wyatt DAWN‐HELEOS‐II multi‐angle light‐scattering (MALS) detector, and a Wyatt Optilab T‐rEX differential refractive index (RI) detector (Wyatt Technologies, Santa Barbara, CA, USA). The column was equilibrated with 20 mM HEPES pH 6.8, 1 mM DTT, and either 0 or 150 mM NaCl. All data were collected at 25°C. A 100 μl sample was injected into the column using an auto‐sample injection method with a flow rate of 0.5 ml/min. Protein in the eluent was detected via UV absorbance at 280 nm, light‐scattering, and refractive index detectors. The data were recorded and analyzed with the Wyatt Astra software (version 7.0).

In vitro cross‐linking reactions

Protein samples of TDP‐43 NTD WT and S48E were prepared at 400 μM in the following buffer: 20 mM HEPES pH 7.5, 150 mM NaCl, and 1 mM DTT. The amine cross‐linker BS3 (bis(sulfosuccinimidyl)suberate, Fisher Scientific) was added to create a molar ratio of 50. Reactions were incubated at room temperature for 30 min. The reactions were quenched by the addition of 100 mM Tris pH 7.5 and were incubated at room temperature for at least 15 min prior to taking samples for analysis on a gel.

Analytical ultracentrifugation experiments

Sedimentation velocity experiments were conducted in a ProteomeLab XL‐I analytical ultracentrifuge (Beckman Coulter, Indianapolis, IN, USA) following standard protocols unless mentioned otherwise (Zhao et al, 2013). The samples of TDP‐43 NTD at 32.78, 98.83, and 295.60 μM concentration in a buffer containing 20 mM HEPES pH 6.8, 150 mM NaCl, and 1 mM DTT were loaded into a cell assembly comprised of a double‐sector charcoal‐filled centerpiece with a 12‐mm path length and sapphire windows. Buffer density and viscosity were determined in a DMA 5,000 M density meter and an AMVn automated micro‐viscometer (both Anton Paar, Graz, Austria), respectively. The partial specific volumes and the molecular masses of the proteins were calculated based on their amino acid compositions (Cohn & Edsall, 1943) in SEDFIT (https://sedfitsedphat.nibib.nih.gov/software/default.aspx).The cell assembly, containing identical sample and reference buffer volumes of 360 μl, was placed in a rotor and temperature equilibrated at rest at 20°C for 2 h before it was accelerated from 0 to 50,000 rpm. Rayleigh interference optical data were collected continuously for 12 h separately. The velocity data were modeled with diffusion‐deconvoluted sedimentation coefficient distributions c(s) in SEDFIT, using algebraic noise decomposition and with signal‐average frictional ratio and meniscus position refined with non‐linear regression (Schuck, 2000, 2016). The s‐values were corrected for time, and finite acceleration of the rotor was accounted for in the evaluation of Lamm equation solutions (Zhao et al, 2015). Maximum entropy regularization was applied at a confidence level of P‐0.68.

Sedimentation equilibrium was attained at a rotor temperature of 20°C at increasing speeds of 12,000 (for 36 h), 25,000 (for 24 h), and 43,000 rpm (for 24 h) (Zhao et al, 2013). Protein at concentration of between 28.28 and 89.23 μM (130 μl) was loaded into double‐sector centrepieces and absorbance distributions recorded at 280 in 0.001‐cm‐radial intervals with 20 replicates for each point. Global least squares modeling was performed at multiple rotor speeds with the software SEDPHAT (https://sedfitsedphat.nibib.nih.gov/software/default.aspx) using the reversible one‐step self‐association model. The association scheme used in this analysis was [A] ↔ [A₂] ↔ [A₃] ↔ [A₄] ↔ …. [isodesmic] with A being the TDP‐43 NTD monomer, and K _D ^iso the single dissociation constant for the association steps were obtained from the analysis. Errors of the fits represent the 68.3% confidence interval (CI) using an automated surface projection method (Zhao et al, 2015). All plots were generated with the program GUSSI (kindly provided by Dr. Chad Brautigam).

Custom antibodies, HEK293T cell culture, and immunoblotting

Oligopeptides and phospho‐specific rabbit antibodies against TDP‐43 amino acids 40–53 (GLRYRNPVSQCMRG) were custom produced by GenScript (Piscataway, NJ USA). Confirmation of antibody specificity was performed by serially diluting TDP‐43 oligopeptides in 8 M urea and spotting to nitrocellulose membranes, which were then blocked in 6% milk in Tris‐buffered saline (TBS). Membranes were then probed with polyclonal antibodies specific to the phospho‐peptide (α‐TDP‐43 pSer48) or non‐specific to the phosphorylated state (PAN antibody). Primary antibodies were detected with IRDye conjugated secondary antibodies (LI‐COR 926‐32211 or 925‐68020, Lincoln, NE, USA), and their fluorescence was measured and imaged using the Odyssey LI‐COR Imaging System.

HEK293T cells were obtained from ATCC (CRL‐11268) and maintained under standard conditions at 37°C in DMEM‐based media (Dulbecco's modified Eagle's medium) supplemented with 10% fetal bovine serum and 1% P/S/G (penicillin–streptomycin–glutamine). Lysates were prepared by suspension and incubation of cells in nondenaturing lysis buffer (200 mM NaCl, 100 mM Tris–HCl, 0.5% sodium deoxycholate, 1% Triton X‐100, 0.2% sodium dodecyl sulfate, 660 μM phenylmethylsulfonyl fluoride, 100 μl protease inhibitor cocktail (Sigma‐Aldrich P8215, St. Louis, MO, USA) 100 μl phosphatase inhibitor cocktail (Sigma‐Aldrich P0044), 1,250 units Benzonase nuclease (Sigma‐Aldrich E1014).

Western blotting was performed using standard laboratory procedures with Mini‐PROTEAN^® Precast AnyKD™ acrylamide gels (Bio‐Rad 456‐9034) and nitrocellulose membranes (Bio‐Rad 162‐0146). Immunoblotting with custom and commercial α‐TDP‐43 antibodies (CST 3449) and detection with the LI‐COR imaging system were performed as described above. Immunoprecipitation of HEK293T cell lysates was performed according to the manufacturer's protocol with Protein G Dynabeads (Invitrogen 10007D) and mouse monoclonal α‐TDP‐43 antibody (Proteintech 60019‐2).

Phosphatase treatment of nitrocellulose membranes was performed using calf intestinal phosphatase (CIP; NEB M0290S). Following detection of the 43 kDa species by Western blotting, membranes were stripped of primary and secondary antibodies using stripping reagent (Thermo Fisher 46430). Membrane sections containing putative TDP‐43 species were cut and blocked with 5% bovine serum albumin and 0.1% Triton X‐100 in TBS. Identical membrane pieces were treated with and without CIP in reaction buffer (100 mM NaCl, 50 mM Tris–HCl, pH 8, 10 mM MgCl₂, 1 mM DTT) at 37°C. The membranes were then probed as described above.

In cell phase separation assays

TDP‐43 in cell splicing assays