TDP-43, a protein central to amyotrophic lateral sclerosis, is destabilized by tankyrase-1 and -2

ABSTRACT In >95% of cases of amyotrophic lateral sclerosis (ALS) and ∼45% of frontotemporal degeneration (FTD), the RNA/DNA-binding protein TDP-43 is cleared from the nucleus and abnormally accumulates in the cytoplasm of affected brain cells. Although the cellular triggers of disease pathology remain enigmatic, mounting evidence implicates the poly(ADP-ribose) polymerases (PARPs) in TDP-43 neurotoxicity. Here we show that inhibition of the PARP enzymes tankyrase 1 and tankyrase 2 (referred to as Tnks-1/2) protect primary rodent neurons from TDP-43-associated neurotoxicity. We demonstrate that Tnks-1/2 interacts with TDP-43 via a newly defined tankyrase-binding domain. Upon investigating the functional effect, we find that interaction with Tnks-1/2 inhibits the ubiquitination and proteasomal turnover of TDP-43, leading to its stabilization. We further show that proteasomal turnover of TDP-43 occurs preferentially in the nucleus; our data indicate that Tnks-1/2 stabilizes TDP-43 by promoting cytoplasmic accumulation, which sequesters the protein from nuclear proteasome degradation. Thus, Tnks-1/2 activity modulates TDP-43 and is a potential therapeutic target in diseases associated with TDP-43, such as ALS and FTD. This article has an associated First Person interview with the first author of the paper.

Recently, we discovered in Drosophila melanogaster that reduction of the Tnks-1/2 homologue mitigates the neurotoxicity of TDP-43, whereas upregulation exacerbates TDP-43-associated toxicity (McGurk et al., 2018a). Furthermore, we observed that downregulation of the Tnks-1/2 homologue led to an increase in nuclear TDP-43 and a decrease in cytoplasmic TDP-43 in Drosophila neurons (McGurk et al., 2018a). Given the role of Tnks-1/2 in protein degradation and the role of aberrant protein degradation in ALS/FTD, we sought to determine whether Tnks-1/2 promotes ubiquitination and degradation of TDP-43. We demonstrate that a highly selective inhibitor of Tnks-1/2 activity mitigates the neurotoxicity of TDP-43 to rodent neurons. We discovered that TDP-43 has a functional tankyrase-binding motif; however, our data show that TDP-43 is not degraded by Tnks-1/2dependent ubiquitination. By contrast, our results suggest that Tnks-1/2 stabilizes TDP-43 and that this may occur by inhibiting degradation of TDP-43 by the nuclear proteasome. These findings provide molecular and cellular insight into the interaction between Tnks-1/2 and TDP-43 and provide a foundation for developing novel therapeutic strategies for TDP-43-associated diseases.

RESULTS
Pharmacological inhibition of tankyrase-1/2 protects against TDP-43-associated toxicity in rodent primary neurons Our previous studies demonstrate that Tnks is a dose-sensitive modifier of TDP-43-associated toxicity in Drosophila: upregulation of Tnks enhances TDP-43-associated toxicity whereas downregulation of Tnks mitigates TDP-43-associated toxicity (McGurk et al., 2018a). To ascertain whether Tnks-1/2 inhibition is of therapeutic benefit in mammalian cells, we developed a TDP-43 neurotoxicity assay in rat primary cortical neurons ( Fig. 1A; Fig. S1A). Primary neurons were virally infected with 5 multiplicity of infection (moi) of an attenuated herpes simplex virus encoding either a control protein (LacZ) or human TDP-43. Cultures were maintained for 7 days post infection, then immunostained for the neuronal marker Tubulin β-III chain. Surviving neuronal cell bodies were quantified. TDP-43 expression resulted in a significant reduction in the number of cortical neurons compared with the LacZ control [44±25 versus 82±26 (s.d.), respectively] ( To determine whether Tnks-1/2 inhibition could protect neurons from TDP-43-associated toxicity, we examined the effect of the small-molecule inhibitor G007-LK, as it is highly selective for Tnks-1/2 and has no reported effect on PARP-1 activity (Voronkov et al., 2013). To determine whether treatment with G007-LK had any effect on neurons in the absence of TDP-43-associated toxicity, we treated cultured neurons infected with HSV-LacZ with either vehicle (DMSO) or G007-LK (1 or 10 µM) for 7 days and quantified the number of neuronal cell bodies immunolabeled with Tubulin β-III chain. At the concentrations tested, G007-LK treatment had no significant effect on neuronal number (Fig. 1B,C;  Fig. S1B,C, Fig. S2), indicating that G007-LK had little to no effect on general neuronal survival. By contrast, neuronal loss induced by HSV-TDP-43 was significantly reduced by treatment with 1 or 10 µM G007-LK (Fig. 1B,C; Fig. S1B,C, Fig. S2), indicating that G007-LK protects neurons from TDP-43-associated toxicity. These data demonstrate that treating rat primary cortical neurons with the Tnks-1/2 inhibitor protects against TDP-43-associated toxicity. Furthermore, they suggest that targeting Tnks-1/2 activity in TDP-43-associated disease is a potential therapeutic strategy.
To determine whether there is a region within TDP-43 that may directly interact with Tnks-1/2, we computationally aligned the tankyrase-binding motif (RxxΦDG) to the human TDP-43 protein sequence. This highlighted an evolutionarily conserved region (amino acids 165-170) with 84% identity to the tankyrase-binding motif ( Fig. 2C-E; Fig. S3A), which we call the tankyrase-binding domain (TBD) (Fig. 2C). To establish whether the predicted tankyrase-binding motif in TDP-43 could mediate interaction with Tnks-1/2, we deleted the TBD from TDP-43 (TDP-43-ΔTBD-YFP) and tested the ability of the mutant protein to co-immunoprecipitate with Tnks-1/2. The results revealed that deletion of the TBD abolished the capacity of TDP-43 to co-immunoprecipitate with Tnks-1/2 in mammalian cells (Fig. 2F). Importantly, deletion of the TBD did not affect all interactions, as it had no effect on the capacity of TDP-43-ΔTBD to co-immunoprecipitate with endogenous TDP-43 from cellular lysates (Fig. S3D). Collectively, these data suggest that the TBD is essential for the interaction between TDP-43 and Tnks-1/2.
To define further the Tnks-1/2 interaction domain in TDP-43, we mutated each amino acid in the TBD (RxxΦDG) individually to alanine (Fig. 2G) and tested the ability of the mutated protein variants to co-immunoprecipitate with Tnks-1/2. This demonstrated that mutation of either H166 or I168 to alanine was sufficient to abolish the interaction between TDP-43-YFP and Tnks-1/2, whereas mutation of R165, D169 or G170 to alanine had little to no effect (Fig. 2H). We note that the ALS-associated mutation of TDP-43 D169G (Kabashi et al., 2008) resides within the TBD of TDP-43; however, similar to the alanine mutation in D169, the mutation to glycine had no effect on the co-immunoprecipitation of TDP-43 with Tnks-1/2 (Fig. S3E).
Analysis of the previously solved NMR structure of RNA recognition motifs (RRM1 and RRM2) of TDP-43 (Lukavsky et al., 2013) revealed that the TBD and the RNA-binding regions are on opposite sides of RRM1 (Fig. 2I). Recent studies have also demonstrated that mutation in the TBD region (D169G) has no effect on RNA-binding (Chen et al., 2019). The TBD spans a loop, a β-strand and a second loop ( Fig. 2J) and, intriguingly, the amino acids essential for the interaction with Tnks-1/2 (H166 and I168) are positioned on the internal side of the β-strand (Fig. 2J). The nonessential amino acids of the TBD are located on the unstructured loops (R165, D169 and G170) or on the external surface of the β-strand (M167) (Fig. 2J). These combined data indicate that TDP-43 and Tnks-1/2 interact and that this interaction is dependent upon H166 and I168, which are positioned in the β-strand in the TBD of TDP-43.

Proteasomal turnover of TDP-43 occurs in the nucleus
To gain an understanding of how Tnks-1/2 could lead to stabilization of TDP-43, we examined TDP-43 localization by immunofluorescence. Under normal conditions, both TDP-43-WT-YFP and TDP-43-ΔTBD-YFP localized diffusely to the nucleus. In response to MG132 treatment, both proteins formed nuclear foci that co-labeled with ubiquitin ( Fig. 5A,B). However, ubiquitin colabeling occurred significantly earlier for nuclear TDP-43-ΔTBD foci than for TDP-43-WT foci (2 h versus 4 h of treatment) (Fig. 5A,B). These data are consistent with our finding that TDP-43-ΔTBD is more rapidly ubiquitinated than TDP-43-WT (see Fig. 4) and indicates that proteasome inhibition causes both TDP-43 proteins (WT and ΔTBD) to accumulate in ubiquitin-positive foci in the nucleus.
Curiously, we observed that MG132 treatment did not lead to an increase in the percentage of cells with cytoplasmic foci of TDP-43-WT or TDP-43-ΔTBD, or with cytoplasmic foci of the protein co-labeled with ubiquitin ( Fig. 5A,C). This appears to be in contrast to some studies, but consistent with others, and we suggest that this difference is due to differing time periods of MG132 treatment (see Discussion). This result, however, raised the possibility that proteasome inhibition may preferentially promote accumulation of TDP-43 selectively in the nucleus. To explore proteasomal-turnover of TDP-43 in the context of the cellular milieu, we examined the effect of MG132 on a form of TDP-43 that cannot be imported into the nucleus and instead localizes to the cytoplasm. TDP-43 nuclear localization is dependent upon a bipartite nuclear localization sequence (NLS) that also acts as a PAR-binding motif (PBM) in the N-terminal portion of the protein (Fig. S7A); mutation of this region (TDP-43-ΔNLS/PBM) prevents nuclear import and binding to PAR (McGurk et al., 2018a;Winton et al., 2008). Under conditions in which MG132 treatment increased the percentage of cells with nuclear TDP-43-WT-GFP foci [from 9.4±1.3% to 45±3.6% (s.e.m.); Fig. 5C], TDP-43-ΔNLS/PBM-GFP remained diffusely cytoplasmic (Fig. 6B,D; Fig. S7B-D). It is important to note that under the same conditions, the cytoplasmic protein G3BP1 formed cytoplasmic foci in response to MG132 treatment (Fig. S8A,B), which is consistent with previous reports (Mazroui et al., 2007). Thus, our data suggest that, under the conditions tested, TDP-43 localized to the cytoplasm (TDP-43-ΔNLS/PBM) does not respond to MG132.
To determine whether TDP-43-ΔNLS/PBM remained diffuse upon MG132 treatment because of its cytoplasmic localization or, alternatively, because the NLS/PBM mutation impaired the ability of the protein to respond to proteasome inhibition, we generated TDP-43-ΔNLS/PBM-GFP with an exogenous NLS sequence. We compared the bipartite NLS/PBM from the TDP-43 protein (TDP-43-ΔNLS/PBM TDP-43 ) to the proline-tyrosine NLS (PY-NLS) from heterogeneous nuclear ribonucleoprotein A1 (hnRNPA1; TDP-43-ΔNLS/PBM A1 ) (Fig. S7A). The NLS from TDP-43 differs from the PY-NLS not only in amino acid sequence but also in the transport system used to direct proteins to the nucleus. Importin α/β (also known as karyopherin α/β1) directs TDP-43 to the nucleus whereas transportin (also known as karyopherin β2) directs hnRNPA1 to the nucleus (Lee et al., 2006;Winton et al., 2008). Under normal conditions TDP-43-ΔNLS/PBM TDP-43 and TDP-43-ΔNLS/PBM A1 localized to the nucleus ( Fig. 6A) and, upon treatment with MG132, both TDP-43-ΔNLS/PBM TDP-43 and TDP-43-ΔNLS/PBM A1 formed ubiquitin-labeled nuclear foci ( Fig. 6B-D). These data suggest that TDP-43 must be in the nucleus to form MG132-induced nuclear foci. These data also suggest that the NLS/PBM sequence of TDP-43 is not required for the response to MG132 treatment, because the addition of a different NLS sequence (TDP-43-ΔNLS/PBM A1 ) also rescued MG132-induced accumulation of the protein in the nucleus.
To gain a molecular understanding of TDP-43 turnover in the context of the cell, we assessed the ubiquitination levels of immunoprecipitated TDP-43-GFP localized to the nucleus (-WT, -NLS/PBM TDP-43 ) or cytoplasm (-ΔNLS/PBM). Upon MG132 treatment, both nuclear forms of TDP-43 (-WT and -ΔNLS/ PBM TDP-43 ) were ubiquitinated (Fig. 6E,F). Although MG132 treatment led to an increase in ubiquitination of cytoplasmic TDP-43 (-ΔNLS/PBM) compared with baseline, the levels were significantly lower than for TDP-43-WT (Fig. 6E,F). This finding indicates that upon MG132 treatment, ubiquitinated TDP-43-GFP localized to the nucleus accumulates more rapidly than ubiquitinated TDP-43-GFP in the cytoplasm. These data further suggest that, in this assay, TDP-43 is preferentially degraded by the nuclear proteasome.
Tankyrase-1/2 promotes cytoplasmic accumulation of TDP-43 Previously, we found that Tnks-1/2 regulates the cytoplasmic accumulation of TDP-43 in aging neurons in Drosophila and in the cytoplasm of mammalian cells exposed to the chemical stressor Black asterisks indicate the evolutionarily conserved amino acids. The aspartic acid at position 169 is mutated to glycine (D169G) in a case of sporadic ALS (Kabashi et al., 2008). (E) Alignment of TDP-43 amino acids 165-170 to tankyrase-binding domains in Tax1-binding protein 1 (TXBP151), telomererepeat binding factor 1 (TRF-1), ubiquitin-specific peptidase 25 (USP25), nuclear mitotic apparatus protein (NuMA), 182 kDa tankyrase-binding protein (TAB182), homeobox B2 (Hox-B2), L-type calcium channel (L-type), protein phosphatase 1 (PP1) and insulin-responsive aminopeptidase (IRAP) (Sbodio and Chi, 2002). Asterisks indicate conserved amino acids in this set of proteins. (F) Deletion of the predicted TBD in TDP-43 prevents endogenous Tnks-1/2 co-immunoprecipitating with human TDP-43-WT-YFP in COS-7 cells. Experiment was repeated six independent times. (G) A series of TDP-43-YFP constructs were generated with the amino acids in the TBD mutated to alanine (highlighted in red). (H) Mutation of either H166 or I168 to alanine (red asterisks) was sufficient to abolish the interaction between endogenous Tnks-1/2 and TDP-43-YFP in COS-7 cells, whereas mutation of R165, M167, D169 or G170 to alanine had little to no effect. Experiment was repeated three independent times. (I) The previously reported NMR structure of RRMs of TDP-43 binding to UG-rich RNA (4BS2) (Lukavsky et al., 2013). The TBD was identified using FirstGlance in Jmol. The yellow halos mark the structure of each amino acid in the TBD. (J) The amino acids (H166 and I168) in the TBD that are crucial for the interaction between Tnks-1/2 and TDP-43 are marked with a red asterisk and those that have no to little effect on the interaction with Tnks-1/2 when mutated to alanine are marked with a white asterisk.
The tankyrase-binding motif is a loosely conserved motif of RxxΦDG (Guettler et al., 2011;Sbodio and Chi, 2002). The first and last amino acids (R1 and G6) in peptides or fragments that span the tankyrase-binding motif of 3BP2, axin1, RNF146 and IRAP abolish the interaction with Tnks-1/2 and are considered to be invariant amino acids (DaRosa et al., 2018;Guettler et al., 2011;Morrone et al., 2012;Sbodio and Chi, 2002). Mutation analysis of a peptide spanning the TBD of 3BP2 demonstrated that at position 4, glycine, proline, alanine and cysteine are preferred whereas isoleucine, leucine and valine prevent Tnks-1/2 binding to the 3BP2 peptide (Guettler et al., 2011). This appears in contrast with our detailed mutational analysis of the tankyrase binding domain in TDP-43, which shows that mutation of R1 or G6 has little to no effect on the interaction between TDP-43 and Tnks-1/2. Furthermore, not only is position 4 of the TBD an isoleucine, but mutation of this amino acid in the full-length form of TDP-43 abolishes the interaction with Tnks-1/2. Thus, we suggest that TDP-43 harbors a tankyrase-binding motif with non-canonical features. It is important to note that non-canonical and extended tankyrase-binding motifs have been identified and include an extra two amino acids after position 6 in 3BP2 and motifs with additional amino acids between positions 1 and 2 in APC2, axin1 and RNF146 (Croy et al., 2016;DaRosa et al., 2018;Guettler et al., 2011;Morrone et al., 2012). Furthermore, the second tankyrase-binding motif in axin is not dependent on R1 (Morrone et al., 2012) and tankyrase-binding motifs in several other proteins harbor amino acids that are not tolerated at the fourth position in the 3BP2 peptide including a valine (motif 1 of axin 1, motif 2 of RNF146, motif 1 of APC2 and motif 3 of PEX14) and a leucine (motif 2 of MDC1) (Croy et al., 2016;DaRosa et al., 2018;Li et al., 2017b;Morrone et al., 2012;Nagy et al., 2016). Thus, in the context of different proteins or perhaps context-dependent functions of Tnks-1/2, the constraints on the consensus of the tankyrase-binding motif might differ.
The mechanism by which Tnks-1/2 stabilizes TDP-43 may include cellular localization of the protein. Using a combination of biochemistry and immunofluorescence, we observed that upon proteasome inhibition, TDP-43 primarily accumulates in the nucleus and is more rapidly ubiquitinated than cytoplasmic TDP-43. Our data, presented here and previously (McGurk et al., 2018a), suggest that Tnks-1/2 promotes cytoplasmic accumulation of TDP-43. We hypothesize that, in doing so, Tnks-1/2 may either sequester TDP-43 from the nuclear proteasome or inhibit nuclear import of the protein, which ultimately leads to stabilization (Fig. 8). Tnks-1/2 activity is known to regulate protein localization, for example by promoting axin localization to the Wnt receptor, or alternatively by directing its degradation (De Rycker and Price, 2004; Mariotti et al., 2016;Wang et al., 2016;Yang et al., 2016). Whether the regulation of TDP-43 localization by Tnks-1/2 is direct or indirect and whether this involves modification of TDP-43 (e.g. by PARylation) remain to be established. Impairment of the proteasome has been hypothesized to be a disease-causing mechanism in ALS/FTD (Scotter et al., 2015;Taylor et al., 2016); thus, many studies have examined TDP-43 cellular localization upon proteasome inhibition with MG132. Treatment with MG132 for extended time periods (ranging from 12 to 72 h) leads to diffuse accumulation of TDP-43 in the cytoplasm (Klim et al., 2019;van Eersel et al., 2011;Walker et al., 2013) or cytoplasmic aggregates of TDP-43 (Huang et al., 2014;Li et al., 2017a;Scotter et al., 2014). Our data demonstrate that treatment with MG132 for short periods of time (3-5 h) leads to nuclear foci formation of TDP-43-WT, whereas TDP-43 localized to the cytoplasm (TDP43-ΔNLS/PBM) remains diffuse. Our findings are consistent with previous studies showing that TDP-43-WT forms nuclear foci in cells treated with MG132 for 8 h (Wang et al., 2010) and that TDP-43 with mutation the NLS/PBM remains diffuse upon treatment with MG132 for 6 h (Nonaka et al., 2009). Thus, we propose that, upon proteasomal inhibition, TDP-43 accumulation in the nucleus occurs earlier than in the cytoplasm. Our data further suggest that turnover of TDP-43 by the nuclear proteasome is important for regulating TDP-43 degradation.
An impact on global protein levels is thought to be involved in ALS/FTD. Many disease-associated mutations occur in proteins that function in protein turnover such as the UPS and autophagy, including C9orf72, charged multivesicular body protein 2b, optineurin, sequestrome 1, serine/threonine protein kinase TBK1, ubiquilin 2 and valosin-containing protein (VCP), suggesting broad impairment of protein turnover in neurodegenerative disease (Balendra and Isaacs, 2018;Gao et al., 2017;Taylor et al., 2016). Furthermore, altered TDP-43 protein levels have also been implicated in disease. For example, mRNA levels for TDP-43 are upregulated in post-mortem tissue from patients with ALS/FTD as well as in a knock-in mouse model for the Q331K disease-causing mutation in TDP-43 (Gitcho et al., 2009;Mishra et al., 2007;White et al., 2018). In post-mortem tissue, TDP-43 protein abnormally accumulates in ubiquitin inclusions in the cytoplasm of affected neurons and glia, suggesting that the affected cells cannot remove and degrade TDP-43 in the cytoplasm (Arai et al., 2006;Mackenzie et al., 2007;Neumann et al., 2007). In patient fibroblasts (sporadic and those harboring a G 4 C 2 -hexanucleotide expansion in C9orf72 or mutation in TDP-43), the levels of cytoplasmic TDP-43 and of total TDP-43 protein are significantly higher than in control cells (Lee et al., 2019;Sabatelli et al., 2015). Additionally, upon proteasome inhibition, TDP-43 protein levels remain unaltered in these ALS/FTD patient fibroblasts (Lee et al., 2019), suggesting that UPS turnover of TDP-43 is impaired. It is intriguing to postulate that UPS-mediated turnover of TDP-43 is impaired in ALS/FTD patient fibroblasts because TDP-43 accumulates in the cytoplasm and thus is sequestered from the nuclear proteasome.
There are no effective treatments for ALS/FTD and related disorders such as Parkinson's disease. In ALS, nuclear PAR is elevated in motor neurons of post-mortem spinal cord tissue and total PAR is elevated in the cerebrospinal fluid of patients with Parkinson's disease (Kam et al., 2018;McGurk et al., 2018c). These findings suggest that inhibiting PARP activity may have therapeutic potential. Similar to inhibition of Tnks-1/2, chemical inhibition of nuclear PARP-1/2 activity reduces accumulation of TDP-43 in stress foci in the cytoplasm (McGurk et al., 2018a). Thus, nuclear PARP activity may promote nuclear export of TDP-43, which may also lead to reduced turnover of the protein. Finding agents that regulate the levels of TDP-43 is important for modulating TDP-43 protein homeostasis, as well as removing misfolded and possibly toxic forms of the protein that accumulate in affected brain regions. We propose that inhibition of Tnks-1/2, which regulates TDP-43 stability and neurotoxic properties, is a potential therapeutic target for ALS/FTD and related disorders.

Rat cortical neuron culture and neurotoxicity assay
Rat cortical neurons were from embryos isolated from female Sprague Dawley wild rats that were 16-18 days pregnant (Neuron R Us, neuron service center, University of Pennsylvania). About 100,000 neurons were plated out on poly-D-lysine coated coverslips (12 mm diameter and thickness #1; Neuvitro, Vancouver, WA, Canada) in neurobasal medium supplemented with serum-free B27, penicillin streptomycin and Glutamax (all from ThermoFisher Scientific Waltham, MA, USA). Neurons were cultured at 37°C with 5% CO 2 . Three times per week, half of the medium was removed and replaced with fresh prewarmed medium. The primary neurons were infected at 14-17 days in vitro with HSV-TDP-43 or HSV-LacZ with either the Tnks-1/2 inhibitor G007-LK (SelleckChem, Houston, TX, USA) or DMSO (Sigma Aldrich, St Louis, MI, USA). Every 2 days, half of the medium was removed and replaced with fresh medium containing G007-LK or DMSO. The drug-containing medium was made up at 2× concentration so that when it was diluted twofold in the well it was of the appropriate concentration. At 7 days post infection, the neurons were fixed in 4% paraformaldehyde, blocked in 10% normal donkey serum (Sigma Aldrich) in Tris-buffered saline containing 0.05% Tween 20 (TBST) (ThermoFisher Scientific) for 1 h at room temperature, and immunostained overnight at 4°C with antibodies directed to the neuronal marker tubulin β-3 chain (1:500; Abcam, Cambridge, UK). After three sets of 5 min washes in TBST, neurons were incubated with mouse AlexaFluor 488 (1:500; ThermoFisher Scientific) in the dark for 1 h at room temperature. Neurons were washed three times (5 min each) with TBST, counterstained for 15 min with 1 µg/ml Hoechst 33342 (ThermoFisher Scientific), washed in deionized H 2 O and mounted in ProLong Diamond (ThermoFisher Scientific). Five images (10× magnification) were captured from each coverslip and the remaining neuronal cell bodies in each image counted. Each condition was repeated three times on three independent cultures, each from a different pregnant rat.

Mammalian cells and culture details
COS-7 cells originally purchased by ATCC were a gift from Virginia M. Lee (University of Pennsylvania). Prior to purchase, the COS-7 cells were authenticated by ATCC. HEK293T cells were kindly provided by Aaron Gitler (Stanford University). COS-7 cells were routinely grown in Dulbecco's modified Eagle's medium (DMEM) containing high glucose and L-glutamine (ThermoFisher Scientific), 10% filter-sterile FBS (Sigma Aldrich) and penicillin-streptomycin (ThermoFisher Scientific). HEK-293T cells were grown in DMEM with high glucose, L-glutamine and sodium pyruvate (ThermoFisher Scientific), 10% filter-sterile FBS (Sigma Aldrich) and penicillin-streptomycin (ThermoFisher Scientific). Cells were grown at 37°C with 5% CO 2 ; a water bath was used for humidification. Cells were washed with Dulbecco's PBS without calcium or magnesium (ThermoFisher Scientific) and trypsinized in trypsin with 0.25% EDTA (ThermoFisher Scientific). No commonly misidentified cell lines were used.

Identification of the tankyrase-binding domain
We computationally aligned the tankyrase-binding motif (RxxΦDG) to the human TDP-43 protein sequence using the PATTINPROT search engine (Combet et al., 2000). To map the TBD to the reported NMR structures of RRM1 and RRM2 of TDP-43 (4BS2) (Lukavsky et al., 2013), we used the open source Java viewer 'FirstGlance in Jmol'.
Immunoprecipitation and immunoblotting COS-7 cells were seeded at a density of 2.7×10 5 cells in a six-well plate overnight in DMEM with high glucose and L-glutamine (ThermoFisher Scientific), 10% FBS (Sigma Aldrich) and penicillin-streptomycin (ThermoFisher Scientific). Each well was transfected with 2.5 µg plasmid DNA, 2.5 µl PLUS reagent and 7.9 µl LTX in 400 µl OPTIMEM I (all from ThermoFisher Scientific) in DMEM with high glucose (ThermoFisher Scientific) and 10% FBS (Sigma Aldrich). Control IgG (2.5 µg; Santa Cruz Biotechnology) and anti-GFP-3E6 (2.5 µg; ThermoFisher Scientific) were coupled to 50 µl Protein G dynabeads (ThermoFisher Scientific). For MG132-induced ubiquitination of TDP-43, cells were treated with 10 µM MG132 (Sigma Aldrich) or an equivalent volume of DMSO (Sigma Aldrich) at 23 h post-transfection for the indicated times. Treatment with the DMSO control always matched the longest MG132 treatment. For MG132induced ubiquitination of TDP-43 in the presence or absence of Tnks-1/2 inhibitor, cells were treated with 10 µM MG132 (Sigma Aldrich) combined with either DMSO (Sigma Aldrich) or XAV939, G007-LK or IWR1-endo (all SelleckChem, Houston, TX, USA) at the indicated times and concentrations. Cells were lysed by adding 300 µl ice-cold RIPA buffer (Cell Signaling, London, UK) containing Halt Protease Inhibitor Single-Use Cocktail (ThermoFisher Scientific) and 5 mM N-ethylmaleimide (NEM) (ThermoFisher Scientific) to each well and incubating the plate on a platform shaker at medium speed for 10 min at 4°C. Lysates were passed three times through a 20G1½ 1 ml syringe (BD Biosciences, San Jose, CA, USA), transferred to centrifuge tubes and rotated at 15 rpm for 10 min at 4°C. Lysates were centrifuged for 10 min at 16,873 g and 4°C. Then, 25 µl of lysate was removed for input and the remaining lysate was made up to 1 ml and incubated with the antibody-coupled beads for 18 h at 4°C with rotation.
Beads were washed three times in 500 µl lysis buffer for 10 min at 4°C with 15 rpm rotation. Elution was performed at 95°C for 5 min in 40 µl 1× LDS Sample Buffer (ThermoFisher Scientific) containing 5% βmercaptoethanol (Sigma Aldrich). Input samples (10 µl) were denatured Fig. 8. Model for effects of Tnks-1/2 to modulate the subcellular localization of TDP-43. The data show that Tnks-1/2 promotes cytoplasmic accumulation of TDP-43, thereby inhibiting access of TDP-43 to the nuclear proteasome. In this way, Tnks-1/2 stabilizes TDP-43 in the cytoplasm. In human disease, accumulation of TDP-43 in the cytoplasm is observed in affected brain cells of >95% of ALS cases and ∼45% of FTD cases. Thus, therapeutic inhibition of Tnks-1/2 in ALS/FTD may maintain TDP-43 in the nucleus where misfolded or mutated forms of the protein can be degraded by the nuclear proteasome.

Cycloheximide pulse chase
To examine the stability of TDP-43-YFP, COS-7 cells were seeded at a density of 2.7×10 5 cells in a six-well plate overnight in DMEM with high glucose (ThermoFisher Scientific), 10% FBS (Sigma Aldrich) and penicillinstreptomycin (ThermoFisher Scientific). The following day, each well was transfected with 2.5 µg plasmid DNA, 2.5 µl PLUS reagent and 7.9 µl LTX in 400 µl OPTIMEM I in DMEM with high glucose and L-glutamine (all ThermoFisher Scientific) and 10% FBS (Sigma Aldrich). Cells were treated with 100 µg/ml cycloheximide (Sigma Aldrich) in DMEM with high glucose and L-glutamine (all ThermoFisher Scientific) containing 10% FBS (Sigma Aldrich) and penicillin-streptomycin (ThermoFisher Scientific) starting at 24 h after transfection. The media with DMSO (ThermoFisher Scientific) or 100 µg/ml cycloheximide (Sigma Aldrich) were replaced every 24 h during this period. Cells were lysed at 0, 24 and 48 h (see paragraph below for cell lysis). To examine the stability of endogenous TDP-43, a six-well plate was inoculated with COS-7 cells at a seeding density of 5.6×10 5 cells. Cells were grown in DMEM with high glucose and L-glutamine (ThermoFisher Scientific) containing 10% FBS (Sigma Aldrich) and penicillinstreptomycin (ThermoFisher Scientific) in six-well plates overnight. Cells were treated with 100 µg/ml cycloheximide (Sigma Aldrich) in DMEM containing high glucose and L-glutamine (ThermoFisher Scientific) and 10% FBS (Sigma Aldrich) with penicillin-streptomycin (ThermoFisher Scientific) and lysed (see paragraph below) 1 h after drug treatment.

Immunofluorescence
Immunofluorescence was carried out in a 24-well format. Cells (COS-7 and HEK293T) were seeded at a density of 60,000 cells per well onto glass coverslips (Neuvitro) in DMEM with high glucose and L-glutamine (ThermoFisher Scientific) and 10% FBS (Sigma Aldrich) with penicillinstreptomycin (ThermoFisher Scientific) and incubated overnight at 37°C with 5% CO 2 ; a water bath was used for humidification. The following day (∼18 h later), each well was transfected with 500 ng of plasmid DNA, 1.75 µl lipofectamine LTX and 0.5 µl PLUS reagent in 100 µl of OPTIMEM I (all ThermoFisher Scientific). At 21 h post transfection, cells were treated with 10 µM MG132 (Sigma Aldrich) for the indicated amount of time. Treatment with the DMSO control always matched the longest MG132 treatment. For localization studies in the presence of Tnks-1/2 inhibitors, cells were transfected in the presence of the indicated amount of G007-LK or IWR1endo (both SelleckChem) and fixed for 21 h post-transfection. Cells were fixed in 4% paraformaldehyde (Electron Microscopy Sciences, Hatfield, PA, USA) in 50 mM HEPES pH 7.4, 150 mM NaCl, 1 mM MgCl 2 and 1 mM EGTA (all Sigma Aldrich). Fixed cells were permeabilized by three treatments of 3 min in 100 mM PIPES, 1 mM MgCl 2 , 10 mM EGTA pH 6.8 and 0.1% Triton-X 100 (PEM-T buffer) (all Sigma Aldrich). Cells were blocked in 10% normal donkey serum (Sigma Aldrich) in TBST (ThermoFisher Scientific). Primary antibodies in TBST (ThermoFisher Scientific) were applied overnight at 4°C in a humidified chamber. Cells were washed three times in PEM-T buffer (3 min each); then, secondary antibody in TBST (ThermoFisher Scientific) was applied for 45 min at room temperature and in the dark. Cells were washed three times in TBST (3 min each), stained with 1 µg/ml Hoechst 33342 (ThermoFisher Scientific) for 15 min, washed in deionized H 2 O and mounted in ProLong Diamond (ThermoFisher Scientific). All experiments were performed at least three independent times.

Image acquisition and quantification
All imaging was performed on fixed cells at 18-23°C on a Leica DMI6000, widefield epifluorescent microscope (Leica Microsystems, Buffalo Grove, IL, USA). To quantify MG132-induced foci, four or five independent images at 20× were captured. The number of transfected cells was quantified in each image; the number of transfected cells with either nuclear or cytoplasmic foci was counted and the number of cells with nuclear or cytoplasmic TDP-43 foci that co-labeled with ubiquitin was counted and the percentage calculated. To compare the effect of MG132 on TDP-43-WT-YFP versus TDP-43-ΔTBD-YFP, 139-298 transfected cells were quantified per condition. To compare the effect of MG132 on TDP-43-WT-GFP, TDP-43-ΔNLS/PBM-GFP, TDP-43-ΔNLS/PBM TDP-43 -GFP and TDP-43-ΔNLS/ PBM A1 -GFP, 129-514 transfected cells were quantified per condition. To quantify the effect of Tnks-1/2 inhibition on TDP-43-GFP localization, 4-5 independent images at 20× were captured at the same exposure time. Cells with diffuse cytoplasmic GFP and/or GFP foci in the cytoplasm were scored as cells with cytoplasmic TDP-43-GFP. Up to 540 transfected cells were quantified per condition. All experiments were repeated at least three independent times and the mean±s.e.m. calculated.

Statistical analysis
Each graph gives the mean (±s.e.m. or s.d.). n is the number of biological repeats and is indicated in each figure legend. Student's t-tests, one-way ANOVA, two-way ANOVA and multiple comparison tests were performed, as indicated in each figure legend. Significance was set at P<0.05; values for asterisks are presented in each legend. All statistical analyses were carried out using Graphpad prism6 software (GraphPad software, San Diego, CA, USA).