ER–mitochondria associations are regulated by the VAPB–PTPIP51 interaction and are disrupted by ALS/FTD-associated TDP-43

Mitochondria and the endoplasmic reticulum (ER) form tight structural associations and these facilitate a number of cellular functions. However, the mechanisms by which regions of the ER become tethered to mitochondria are not properly known. Understanding these mechanisms is not just important for comprehending fundamental physiological processes but also for understanding pathogenic processes in some disease states. In particular, disruption to ER–mitochondria associations is linked to some neurodegenerative diseases. Here we show that the ER-resident protein VAPB interacts with the mitochondrial protein tyrosine phosphatase-interacting protein-51 (PTPIP51) to regulate ER–mitochondria associations. Moreover, we demonstrate that TDP-43, a protein pathologically linked to amyotrophic lateral sclerosis and fronto-temporal dementia perturbs ER–mitochondria interactions and that this is associated with disruption to the VAPB–PTPIP51 interaction and cellular Ca2+ homeostasis. Finally, we show that overexpression of TDP-43 leads to activation of glycogen synthase kinase-3β (GSK-3β) and that GSK-3β regulates the VAPB–PTPIP51 interaction. Our results describe a new pathogenic mechanism for TDP-43.

(PTPIP51) as a binding partner for the resident ER protein vesicle-associated membrane protein-associated protein-B (VAPB) 20 . Here, we show that the VAPB-PTPIP51 interaction regulates ER-mitochondria associations. Moreover, we demonstrate that expression of both wild-type and ALS/FTD-associated mutant TDP-43 perturbs ER-mitochondria associations and that this is accompanied by changes to the VAPB-PTPIP51 interaction. As such, our findings reveal a new mechanism for controlling ER-mitochondria interactions and highlight ERmitochondria associations as a target for disruption by TDP-43.

Results
VAPB and PTPIP51 interact in a range of biochemical assays. VAPB and PTPIP51 interact in a variety of biochemical assays, and the region of PTPIP51 that mediates binding involves its central coiled-coil domain (amino acids 84-174) 20 . To gain insight into the domain(s) of VAPB involved in binding PTPIP51, we generated glutathione S-transferase (GST) fusion proteins comprising the entire VAPB cytoplasmic domain (amino acids 1-220), its N-terminal major sperm protein domain (amino acids 1-124), its central coiled-coil domain (amino acids 142-207) and sequences encompassing the C-terminal part of VAPB (amino acids 89-207) and tested their abilities to bind to PTPIP51 in pull-down assays from PTPIP51-transfected cells.
Only the entire VAPB cytoplasmic domain bound PTPIP51 (Fig. 1a). Hence, the entire cytoplasmic domain of VAPB is required for binding to PTPIP51 or the interaction requires some topographical conformation that is lost in the smaller VAPB fragments. We also performed in vitro-binding studies with purified VAPB and PTPIP51 cytoplasmic domains that were generated in E. coli. PTPIP51 cytoplasmic domain (amino acids  was incubated with either GST or GST-VAPB and the GST moiety used to isolate bound PTPIP51. GST-VAPB cytoplasmic domain but not GST bound to PTPIP51 in these assays (Fig. 1b). Together with our previous studies 20 , these results demonstrate that VAPB interacts with PTPIP51 in a number of different assays including in vitro assays in the absence of other proteins.
VAPB and PTPIP51 mediate ER-mitochondria associations. The binding of VAPB to PTPIP51 suggests that these proteins act as tethers to link ER with mitochondria. We therefore monitored how modulating VAPB and PTPIP51 expression affects ER-mitochondria associations via EM in mouse NSC34 motor neuron cells. We quantified ER-mitochondria associations by determining the proportion of the mitochondrial surface that was closely apposed (o30 nm) to ER. Such an approach has been used by others 21,22 . We first reduced VAPB and PTPIP51 expression using short interfering RNAs (siRNAs) and identified one potent mouse VAPB siRNA and two PTPIP51 siRNAs. Consistent with previous studies in HEK293 cells 20 , siRNA knockdown of VAPB did not influence expression of PTPIP51 and knockdown of PTPIP51 did not influence expression of VAPB in NSC34 cells (Fig. 2a). Approximately 12% of the mitochondrial surface was closely associated with ER in untreated and control siRNA cells. However, siRNA knockdown of either VAPB or PTPIP51 significantly reduced these values (Fig. 2b). We also monitored how loss of VAPB and PTPIP51 affected ER-mitochondria associations in human HEK293 cells using previously characterized siRNAs that involved different sequences 20  We next elevated the expression of VAPB and PTPIP51 via transfection. Cells were transfected with either enhanced cyan fluorescent protein (ECFP) control vector, ECFP-VAPB or PTPIP51 that was expressed using a bicistronic internal ribosome entry site vector with AcGFP. The ECFP/AcGFP tags were then utilized to isolate transfected cells using a cell sorter. The proportion of the mitochondrial surface closely associated with ER in control cells expressing ECFP alone was not significantly different to that in untreated cells or cells treated with control siRNA (Fig. 3a). However, expression of VAPB or PTPIP51 markedly increased ER-mitochondria associations (Fig. 3a). We also monitored the effect of co-transfecting both VAPB and PTPIP51 and obtained a remarkable phenotype where large proportions of the mitochondrial surface were closely associated with ER ( Fig. 3a,b). Analyses of these images revealed the presence of structures that appeared to tether ER with mitochondria ( Fig. 3c); the average spacing between ER and mitochondria in the VAPB þ PTPIP51 co-transfected cells was 14.7 nm.
Confocal microscopy has also been utilized to study ER-mitochondria interactions (for example, refs 1,10). Since overexpressing VAPB and PTPIP51 produced particularly striking EM images, we reasoned that such phenotypes would be clearly discernible by light microscopy. We therefore monitored the effect of transfecting VAPB and/or PTPIP51 on ER mitochondria co-localization by confocal microscopy. For these investigations, we utilized CV1 cells since these have a flattened and spread morphology, which facilitates such morphological analyses. Cells were transfected with ECFP control, ECFP-VAPB, haemagglutin-tagged PTPIP51 (PTPIP51-HA) or ECFP-VAPB þ PTPIP51-HA, and ER and mitochondria co-localization monitored by immunostaining. ER was detected by labelling for protein disulphide isomerase (PDI) and mitochondria by labelling for translocase of the outer mitochondrial membrane protein-20 (TOM20). Transfection of VAPB and/or PTPIP51 all noticeably increased co-localization of ER and mitochondria and in agreement with the EM studies, this phenotype was particularly striking in VAPB þ PTPIP51 co-transfected cells (Fig. 4). To obtain unbiased quantitative data on the effects of VAPB and PTPIP51 overexpression on ER-mitochondria co-localization, we used intensity correlation analysis (ICA) to analyse the co-localization of PDI and TOM20 signals in the different samples 23 . ICA is one of the most rigorous methods for studying co-localization of proteins by confocal microscopy since it compares the scatter plots of two stains against the product of the difference of the pixel intensities of each of the two stains from their respective means. Thus, unlike other methods, which simply compare co-localization of pixels, ICA also determines whether the pixel intensities from the two signals vary in synchrony. ICA yields an intensity correlation quotient, which is a statistically testable single-value assessment of the relationship between two stained protein pairs 23 . Transfection of VAPB, PTPIP51 and VAPB þ PTPIP51 all significantly increased co-localization of ER and mitochondria in these assays (Fig. 4). Thus, modulating VAPB and PTPIP51 expression significantly influences ER-mitochondria interactions in both EM and confocal light microscopy assays.
TDP-43 disrupts ER-mitochondria associations. We next enquired whether expression of ALS/FTD-associated TDP-43 influenced ER-mitochondria associations. To do so, we transfected NSC34 cells with EGFP control vector, EGFP-TDP-43, or familial ALS mutant EGFP-TDP-43M337V, TDP-43Q331K, TDP-43A382T or TDP-43G348C. Transfected cells were again isolated using a cell sorter via the EGFP tags and ERmitochondria associations were quantified by EM as described above. We detected no change in the numbers of mitochondria or ER profiles in the presence of either wild-type or mutant TDP-43. However, expression of both wild-type and mutant TDP-43 all led to significant reductions in ER-mitochondria associations (Fig. 5a). The finding that both wild-type TDP-43 and mutant TDP-43 disrupted ER-mitochondria interactions is consistent with the phenotypes seen in transgenic mice, where expression of both wild-type and mutant TDP-43 all induce disease [24][25][26][27][28] . We also enquired whether siRNA-mediated loss of TDP-43 influenced ER-mitochondria associations or the VAPB-PTPIP51 interaction. However, loss of TDP-43 had no effect on either of these ( Fig. 6). We next monitored whether transfection of wild-type or mutant TDP-43 influenced expression of VAPB, PTPIP51 or mitofusin-2, a further protein that has been linked to ER-mitochondria associations 10 . To do so, cell-sorted control EGFP and EGFP-TDP-43-transfected cell proteins were probed on immunoblots for VAPB, PTPIP51 and mitofusin-2. However, no differences in expression of any of these proteins were detected in the different samples (Fig. 5b). Thus, the reduction in ER-mitochondria interactions seen in cells expressing wild-type TDP-43 or mutant TDP-43 is not due to changes in expression of VAPB, PTPIP51 or mitofusin-2.
To complement the above studies, we also quantified ER-mitochondria associations in spinal cord motor neurons of 1-month-old homozygous TDP-43 transgenic and their nontransgenic littermates 26 . At this age, these TDP-43 transgenic mice develop motor defects and symptoms characteristic of ALS 26 . Similar to control NSC34 cells, B12% of the mitochondrial surface was closely apposed to ER in the non- transgenic animals. However, in TDP-43 transgenic mice, ER-mitochondria interactions were significantly reduced (Fig. 5c).
TDP-43 disrupts VAPB-PTPIP51 interactions. Since VAPB and PTPIP51 mediate ER-mitochondria associations, and since expression of TDP-43 disrupted ER-mitochondria associations, we enquired whether the expression of TDP-43 influenced binding of VAPB to PTIP51. To do so, we transfected NSC34 cells with PTPIP51-HA and either EGFP control vector or wild-type or ALS mutants of EGFP-TDP-43, and monitored the amounts of endogenous VAPB bound to PTPIP51-HA in immunoprecipitation assays. Consistent with the EM studies, expression of both wild-type and ALS mutant TDP-43 all reduced the amount of VAPB that was bound to PTPIP51 (Fig. 7a). We obtained similar results in HEK293 cells (Fig. 7b).
We also monitored the effect of TDP-43 on the VAPB-PTPIP51 interaction in TDP-43 transgenic mice using immunoprecipitation assays. In spinal cords, endogenous VAPB bound to endogenous PTPIP51 in these assays (Fig. 7c), which is consistent with previous results on endogenous VAPB-PTPIP51 binding in other tissues 20 . In line with the data from the TDP-43-transfected NSC34 cells, no differences in the levels of VAPB, PTPIP51 or mitofusin-2 were detected in spinal cords of TDP-43 and non-transgenic mice (Fig. 7d). However, the amounts of PTPIP51 bound to VAPB were significantly reduced in the TDP-43 transgenic mice (Fig. 7d). Thus, expression of ALS/FTDassociated TDP-43 disrupts ER-mitochondria interactions and this is associated with a decrease in the binding of VAPB to PTPIP51.
Both ER and mitochondria are stores for intracellular Ca 2 þ and the ER-mitochondria interface, and MAMs regulate Ca 2 þ exchange between the two organelles 1,2,10,20 . Indeed, we have shown that siRNA loss of VAPB and PTPIP51 perturbs cellular Ca 2 þ homeostasis following its release from ER stores 20 . The TDP-43-induced disruption to the VAPB-PTPIP51 interaction and ER-mitochondria associations thus predict that TDP-43 may influence Ca 2 þ homeostasis. We therefore monitored the effect of overexpressing TDP-43 on cytosolic and mitochondrial Ca 2 þ levels after induction of inositol 1,4,5-trisphosphate receptor (IP3R)-mediated Ca 2 þ release from ER stores. For these experiments, we used HEK293 cells co-transfected with the M3

TDP-43 activates GSK-3b to inhibit VAPB-PTPIP51 binding.
To gain insight into the mechanism by which TDP-43 might influence the VAPB-PTPIP51 interaction, we first tested whether TDP-43 bound to either VAPB or PTPIP51. To do so, we performed immunoprecipitation assays from TDP-43 þ VAPB and TDP-43 þ PTPIP51-transfected cells. However, we did not detect binding of TDP-43 to either protein in these assays (Fig. 9a) or in yeast 2-hybrid library screens with VAPB or PTPIP51 baits 20 (and our unpublished data). Others have also reported that TDP-43 does not bind VAPB 29 . Phosphorylation is a common mechanism for controlling protein-protein interactions, and TDP-43 has recently been shown to activate glycogen synthase kinase-3b (GSK-3b) 30 . A major route for regulating GSK-3b is by inhibitory phosphorylation of serine-9, and so we monitored GSK-3b serine-9 phosphorylation in control and TDP-43-transfected cells by immunoblotting. In agreement with previous studies 30 , we found that overexpression of wild-type and four different TDP-43 mutants all activated GSK-3b as evidenced by reduced serine-9 phosphorylation (Fig. 9b). To determine whether GSK-3b might regulate the VAPB-PTPIP51 interaction, we modulated GSK-3b activity in transfected cells and monitored the effects on binding of VAPB to PTPI51 again using immunoprecipitation assays. Two structurally different GSK-3b inhibitors both increased, whereas transfection of GSK-3b to elevate its activity 31 reduced binding of VAPB to PTPIP51 (Fig. 9c,d). Thus, TDP-43 activates GSK-3b and GSK-3b regulates the VAPB-PTPIP51 interaction.

Discussion
Despite their fundamental importance to a variety of physiological processes, the mechanisms that mediate and regulate ERmitochondria associations are not properly understood. Recently, we demonstrated an interaction between the resident ER protein VAPB and the mitochondrial protein PTPIP51 (ref. 20). VAPB is anchored in the ER via a C-terminal membrane-spanning domain such that its N terminus projects into the cytoplasm, and PTPIP51 is an outer mitochondrial membrane protein whose C terminus projects into the cytoplasm 20 . VAPB is also a MAM protein 20 . VAPB and PTPIP51 interact in yeast 2-hybrid, immunoprecipitation (including transfected and endogenous proteins), GST pull-down assays and proximity ligation assays, and recombinant VAPB and PTPIP51 cytoplasmic domains interact in vitro in the absence of other proteins demonstrating direct binding 20 (and results presented here). Thus, VAPB and PTPIP51 represent plausible candidates for mediating ER-mitochondria interactions.
To directly test this possibility, we used EM to monitor how modulating VAPB and PTPIP51 expression affects ERmitochondria associations. siRNA knockdown of VAPB or PTPIP51 reduced, whereas overexpression of VAPB and/or PTPIP51 markedly increased ER-mitochondria associations. Indeed, overexpression of both VAPB and PTPIP51 produced a total GSK-3b and GSK-3b phosphorylated on serine-9. Phosphorylation of GSK-3b serine-9 is the principal mechanism for regulating its activity; serine-9 phosphorylation inhibits GSK-3b activity. Bar chart shows relative levels of GSK-3b serine-9 phosphorylation following quantification of signals from immunoblots and normalization to total GSK-3b signals. Data were analysed by one-way analysis of variance (ANOVA) and Tukey's post hoc test. N ¼ 5; *Po0.05, **Po0.01,***Po0.001; error bars are s.e.m.. (c) Inhibition of GSK-3b increases the amount of VAPB bound to PTPIP51. Cells were transfected with control empty vector (EV) or HA-tagged PTPIP51 and treated with either vehicle, GSK-3b inhibitor AR-A014418 (1 mM) or GSK-3b inhibitor CT99021 (100 nM) for 16 h. PTPIP51 was immunoprecipitated using the HA tag and the amounts of endogenous bound VAPB detected by immunoblotting. Both inputs and immunoprecipitations (IP) are shown. Bar chart shows relative levels of VAPB bound to PTPIP51 in the immunoprecipitations following quantification of signals from immunoblots. VAPB signals were normalized to immunoprecipitated PTPIP51-HA signals. Data were analysed by one-way ANOVA and Tukey's post hoc test; N ¼ 3, error bars are s.e.m., *Po0.05. (d) Transfection of GSK-3b decreases the amount of VAPB bound to PTPIP51. Cells were transfected with empty vector (EV), HA-PTPIP51 or HA-PTPIP51 þ GSK-3b. PTPIP51 was immunoprecipitated using the HA tag and the amounts of endogenous bound VAPB detected by immunoblotting. Both inputs and immunoprecipitations (IP) are shown. Bar chart shows relative levels of VAPB bound to PTPIP51 in the immunoprecipitations following quantification of signals from immunoblots. VAPB signals were normalized to immunoprecipitated PTPIP51 signals. Data were analysed by the unpaired t-test; N ¼ 3, error bars are s.e.m., *Po0.05. remarkable phenotype detectable by both EM and confocal microscopy, in which ER and mitochondria redistribute to form close associations. Alterations to ER-mitochondria interactions influence Ca 2 þ exchange between these organelles, and we have demonstrated previously that modulating VAPB and PTPIP51 expression induces such changes to Ca 2 þ homeostasis 20 . Together, these findings support the conclusion that VAPB and PTPIP51 form at least one of the scaffolds that mediate ER-mitochondria associations.
In support of this notion, we show here that overexpression of wild-type and mutant TDP-43 reduce ER-mitochondria interactions, and that these reductions are associated with decreased binding of VAPB to PTPIP51. We also demonstrate that TDP-43 increases cytosolic Ca 2 þ and decreases mitochondrial Ca 2 þ levels following IP3R-mediated Ca 2 þ release from ER stores, which is consistent with a decrease in ER-mitochondria associations 20 . Our findings that overexpression of wild-type as well as mutant TDP-43 affects ER-mitochondria associations, the VAPB-PTPIP51 interaction and associated Ca 2 þ homeostasis are consistent with many studies which show that wild-type as well as mutant TDP-43 all induce similar, aggressive forms of disease in transgenic mice and in other models (for example, refs 24-28,41). Thus, ER-mitochondria and VAPB-PTPIP51 interactions may be targets for damage by TDP-43. The precise mechanisms underlying this damage are not clear. However, in agreement with previous studies we show that overexpression of TDP-43 activates GSK-3b a kinase strongly implicated in both ALS and dementia 30,42,43 . Moreover, we demonstrate that modulating GSK-3b activity affects the VAPB-PTPIP51 interaction; inhibiting GSK-3b increased, whereas overexpression of GSK-3b decreased binding of VAPB to PTPIP51. Thus, the effect of TDP-43 on the VAPB-PTPIP51 interaction may involve activation of GSK-3b, which then leads to decreased VAPB-PTPIP51 binding. This may be via direct phosphorylation of VAPB or PTPIP51 by GSK-3b to inhibit their binding, or signalling via GSK-3b to downstream effectors that somehow influences VAPB/PTPIP51 phosphorylation and/or binding. Future studies to tackle this will involve mass spectrometric sequencing of VAPB and PTPIP51 to identify any GSK-3b phosphorylation sites, and the analyses of their function using mutagenic approaches.
Disruptions to a number of physiological processes are seen in both ALS and FTD. These include damage to mitochondria and mitochondrial ATP production, disruption to Ca 2 þ homeostasis, perturbation to axonal transport of mitochondria, damage to autophagy and finally damage to the ER including activation of the unfolded protein response (UPR) (see review (ref. 44)). One conundrum is how so many apparently disparate pathological features might link together in a common disease pathway. However, all of these features are regulated by ER-mitochondria associations. Thus ER-mitochondrial Ca 2 þ exchange following release from ER stores affects both cellular Ca 2 þ homeostasis and mitochondrial ATP production since several mitochondrial dehyrogenase enzymes involved in ATP synthesis are Ca 2 þ regulated 4,45 . Decreased mitochondrial ATP production is a feature of at least some forms of ALS and FTD 46 . Likewise, ER-mitochondria associations are linked to axonal transport. Mitochondria are anterogradely transported through axons on kinesin-1 motors and attach to kinesin-1 via the outer mitochondrial membrane protein Miro, which acts as a Ca 2 þ sensor; elevated Ca 2 þ levels disrupt transport 47 . Recently, Miro has been shown to localize to ER-mitochondria contact sites and a proportion of ER has been shown to be co-transported with mitochondria 48,49 . Also, defects in autophagy are features of ALS/ FTD involving TDP-43; autophagosomes are now known to form at ER-mitochondria contacts 7,50,51 . Finally, ER-mitochondria associations impact on ER stress and the UPR since a number of ER chaperones involved in protein folding localize to MAM and structural uncoupling of ER from mitochondria induces ER stress and ER-UPR 3,52 . Indeed, VAPB itself is linked to the UPR (see review (ref. 53)). Thus, the TDP-43 loosening of ER-mitochondria associations that we described here may contribute to some of the pathological features of ALS/FTD. Several recent studies have implicated defective ER-mitochondria signalling in some related neurodegenerative diseases including Alzheimer's and Parkinson's diseases [11][12][13][14][15]54 . Damage to the ER-mitochondria axis may thus be a common feature of neurodegenerative diseases. The results described here, which show that VAPB and PTPIP51 function to mediate ER-mitochondria associations and which demonstrate that the VAPB-PTPIP51 interaction is disrupted by TDP-43, will facilitate future studies on the role of ER-mitochondria associations in neurodegenerative diseases. Antibodies and inhibitors. Rat and rabbit antibodies to VAPB and PTPIP51 were generated by immunization with GST-VAPB(1-220) and GST-PTPIP51(36-470), respectively 20 . Rabbit anti-PTPIP51 was from Atlas, rabbit anti-haemagglutin (HA), mouse anti-a-tubulin (DM1A) and rabbit anti-Mitofusin-2 were from Sigma. Rabbit anti-TOM20 was from Santa Cruz Biotechnology. Rabbit anti-GFP and mouse anti-bactin were from Abcam. Mouse PDI (RL77) was from Affinity Bioreagents. Rabbit anti-TDP-43 (12892-1-AP) was from Proteintech. Antibodies to total and ser9-phosphorylated (inactive) GSK-3b were from BD Transduction Labs (mouse 610201) and Cell Signalling (rabbit 9336), respectively. The concentrations used for the different antibodies are shown in Supplementary  Table 1. Secondary antibodies were horseradish peroxidase-coupled goat antimouse, anti-rabbit and anti-rat Igs (GE Healthcare), and Alexa fluorophore (350, 488 and 568)-coupled goat anti-mouse and anti-rabbit IgGs (Invitrogen). AR-A014418 and CT99021 were from Abcam and Cayman, respectively.
Cell culture and transfection. NSC34 cells were provided by Professor Dame Pamela Shaw, (University of Sheffield, UK). CV1 and HEK293 cells were obtained from the ATCC. Cells were grown in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum supplemented with 2 mM glutamine. NSC34 and HEK293 cells were transfected using Lipofectamine 2000 (Invitrogen); CV1 cells were transfected using Exgen (Fermentas) according to the manufacturer's instructions. Transfected cells were analysed 36 h post transfection; siRNA-treated cells were analysed 4 days post treatment. For fluorescence-activated flow cytometry cell sorting, NSC34 were trypsinized, resuspended in phosphate-buffered saline (PBS) and then sorted for ECFP and EGFP expression using a BD FACSAria cell sorter, re-plated and cultured for a further 16 h before processing.
Electron and light microscopy. For EM, NSC34 cells were fixed with 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.2) for 3 h and then harvested by scraping gently with a plastic scraper. The cells were pelleted by centrifugation at 800g(av) for 10 min, washed in buffer and post-fixed for 1 h in 1% osmium tetroxide in 0.1 M sodium cacodylate buffer. The cells were then stained for 1 h with 1% uranyl acetate in water before dehydration and embedding in Taab resin. Spinal cords from 4% paraformaldehyde-perfused mice were immersed in 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer, post-fixed in 1% osmium tetroxide, dehydrated in alcohol and propylene oxide and finally infiltrated and embedded in Epon 812 (ref. 26). Samples were viewed on a Tecnai 12 electron microscope at 6,800 or 23,000 magnification. Digital images were acquired and the circumference of each mitochondria and the proportions of the mitochondrial surface (circumference) closely associated (o30 nm) with ER were calculated. Cells were randomly selected for analyses without prior knowledge of transfected plasmid, siRNA or mouse genotype. All clearly identified mitochondria in the samples were scored. Image analyses were performed using Adobe Illustrator and ImageJ.
For immunostaining of CV1 cells, cells were fixed in 4% paraformaldehyde/ 0.1% glutaraldehyde in PBS, quenched in 0.05 M ammonium chloride in PBS and permeabilized with 0.25% Triton X-100 in PBS. Following blocking in 3% bovine serum albumin in PBS, samples were incubated with primary antibodies diluted in blocking solution, washed with PBS and incubated with secondary antibodies. Following washing, the samples were mounted in Mowiol-DABCO mounting medium containing 10% (w/v) Mowiol 4-88 (Calbiochem), 25% (w/v) glycerol and 2.5% (w/v) DABCO (1,4-diazobicyclo[2.2.2]octane) in 100 mM Tris-HCl pH 8.5. Cells were imaged using a Zeiss LSM510Meta confocal microscope equipped with a Â 63 Plan-Apochromat 1.4 NA objective and analysed using ImageJ with the ICA plug-in 57 . Statistical analyses were performed using GraphPad Prism.
SDS-PAGE, immunoblotting, immunoprecipitation and GST assays. Cells were harvested for SDS-PAGE and immunoblotting by scraping into SDS-PAGE sample buffer containing 2% SDS, 100 mM dithiothreitol, 10% glycerol, 0.1% bromophenol blue and protease inhibitors (Complete Roche) in 50 mM Tris-HCl pH 6.8 and heating to 100°C for 5 min 57 . For immunoprecipitation assays, transfected HEK293 cells were lysed in ice-cold immunoprecipitation buffer comprising 50 mM Tris-citrate pH 7.4, 150 mM NaCl, 1% Triton X-100, 5 mM ethylene glycol tetraacetic acid, 5 mM EDTA and protease inhibitors (Complete, Roche) for 30 min. Transgenic mouse samples were homogenized and diluted into the same buffer. NSC34 cells were lysed in RIPA buffer containing 25 mM Tris-HCl pH 7.5, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS, 150 mM NaCl and protease inhibitors (Complete Roche). The samples were then centrifuged at 13,000g for 20 min and the supernatants were transferred to fresh tubes and precleared by incubation with Protein-G-sepharose beads (prepared as a 50% slurry in PBS containing 0.1% Triton X-100). Following centrifugation at 2,000 g for 30 s to settle the beads, supernatants were transferred to fresh tubes and protein concentrations determined using a Bio-Rad protein assay kit. The protein concentrations were adjusted to 1 mg ml À 1 , and 500 mg protein incubated with appropriate primary antibodies on a rotary shaker for 16 h at 4°C. Antibodies were then captured by addition of 30 ml Protein-G-sepharose beads (50% slurry in PBS/0.1% Triton X-100). Following washing in PBS/0.1% Triton X-100, bound proteins were prepared for SDS-PAGE by addition of 50 ml SDS-PAGE sample buffer and heating to 100°C. Signals on immunoblots were quantified using ImageJ after scanning with an Epson Precision V700 Photo scanner essentially as described by us in previous studies 57 . To ensure the signals obtained were within the linear range, the mean background-corrected optical density (OD) of each signal was interpolated for an OD calibration curve created using a calibrated OD step tablet (Kodak). Only film exposures that gave OD signals within the linear range of the OD calibration curve were used for the statistical analyses.
GST, GST-VAPB and GST-PTPIP51 fusion proteins were prepared on glutathione-sepharose beads from E. coli BL21(DE3) according to the manufacturer's instructions (GE Healthcare). For cellular pull-down assays, GST and GST fusion proteins were incubated with transfected cell lysates prepared in immunoprecipitation buffer as described above 58 . For in vitro pull-down assays of recombinant VAPB and PTPIP51, the cytoplasmic domain of PTPIP51 (aa36-470), was prepared as a GST fusion protein and PTPIP51(36-470) released from the GST moiety by incubating GST-PTPIP51(36-470) bound to glutathionesepharose beads with 1 mg factor XA protease in 50 ml buffer containing 20 mM Tris-HCl pH 8.0, 100 mM NaCl, 2 mM CaCl 2 for 16 h at 4°C according to the manufacturer's recommendations (New England Biolabs). 50 mg of purified recombinant PTPIP51(36-470) was then incubated for 16 h at 4°C with either GST or GST-VAPB(1-220) fusion proteins on glutathione-sepharose beads in PBS containing 1% Triton X-100. GST-fusion protein complexes were pelleted and washed three times with incubation buffer by centrifugation at 2,000 g for 30 s, and analysed by SDS-PAGE and immunoblotting as described above.
Full blots are shown in Supplementary Fig. 1.
Ca 2 þ measurements. Cytosolic and mitochondrial Ca 2 þ levels were measured following IP3R-mediated release from ER stores in HEK293 cells 20 . To do so, cells transfected with M3R and either empty vector control plasmid or TDP-43 plasmids were loaded with either 2 mM Fluo4-AM or Rhod2-AM dye (Invitrogen) in external solution (145 mM NaCl, 2 mM KCl, 5 mM NaHCO 3 , 1 mM MgCl 2 , 2.5 mM CaCl 2 , 10 mM glucose and 10 mM Na-HEPES pH 7.25) containing 0.02% Pluronic-F27 (Invitrogen) for 15 min at 37°C, followed by washing in external solution for 15 min. Fluo4 and Rhod2 fluorescence were timelapse recorded (1-s intervals) with MetaMorph (Molecular Dynamics) on an Axiovert S100 microscope (Zeiss) equipped with appropriate filtersets (Chroma Technology), a Â 40/1.3NA Plan-Neofluar objective (Zeiss) and a Photometrics Cascade-II 512B EMCCD. The cells were kept under constant perfusion with external solution (0.5 ml min À 1 ). IP3Rmediated Ca 2 þ release from ER stores was triggered by application of 100 mM Oxotremorine-M (Tocris) for 2 min. Ca 2 þ levels were calculated as relative Fluo4 or Rhod2 fluorescence compared with baseline fluorescence at the start of the measurement.