The heat shock response plays an important role in TDP-43 clearance: evidence for dysfunction in amyotrophic lateral sclerosis

Insoluble TDP-43 inclusions are the pathological hallmark of ALS and tau-negative frontotemporal lobar degeneration. Chen et al. show that the heat shock response (HSR), which regulates chaperone expression, is compromised in an ALS mouse model and in patients. Activation of the HSR clears insoluble TDP-43 and increases cell survival.


Introduction
Amyotrophic lateral sclerosis (ALS) is the most common adult-onset motor neuron degenerative disease, characterized by progressive motor neuron loss from the spinal cord, brainstem and motor cortex, leading to muscle weakness and eventual respiratory failure. The average age of disease onset is 55 years and affected individuals typically die within 3 to 5 years of diagnosis. No effective treatments are available and consequently there is a high unmet need for novel therapeutic development (for review see Goyal and Mozaffar, 2014). Approximately 10% of ALS cases are familial, with the remaining 90% being sporadic (sporadic ALS). Regardless of the cause, $95% of patients share a common molecular pathology, which involves the accumulation of ubiquitinated, hyperphosphorylated and insoluble TDP-43 protein aggregates in the cytoplasm, and decreased levels of TDP-43 in the nucleus (Neumann et al., 2006;Mackenzie et al., 2007;Tan et al., 2007;Geser et al., 2008;Brettschneider et al., 2013).
TDP-43 is a DNA and RNA binding protein that plays a key role in regulating RNA transcription, editing, transport and translation. It is a predominantly nuclear protein whose levels are tightly controlled through a system of auto-regulation (Ayala et al., 2011;Polymenidou et al., 2011). Knockout of TDP-43 is lethal in embryonic mice, indicating that it plays a vital role in development and survival (Kraemer et al., 2010). Although TDP-43 is predominantly a nuclear protein, it is known to shuttle between the nucleus and cytoplasm and has been identified as a component of stress granules, which form in response to cellular stress (Belly et al., 2005;Rutherford et al., 2008;Colombrita et al., 2009;Dion et al., 2009;Nonaka et al., 2009). Multiple ALS-causing mutations have been identified on the TARDBP gene coding for TDP-43, which promotes TDP-43 translocation to the cytosol, stabilizes and promotes the aggregation of the mutant protein (Johnson et al., 2009;Barmada et al., 2010;Ling et al., 2010). TDP-43 is aggregation-prone and the accumulation of mutant protein in the cytoplasm correlates with toxicity (Barmada et al., 2010;Bilican et al., 2012). In addition, overexpression of wild-type TDP-43 in mammalian cells or transgenic animal models leads to TDP-43 accumulation, cytotoxicity and motor deficits (Ash et al., 2010;Barmada et al., 2010;Stallings et al., 2010;Wils et al., 2010). As TDP-43 cytoplasmic aggregates are observed in 95% of ALS and tau-negative frontotemporal lobar degeneration cases, dysregulation of TDP-43 protein levels are a common feature of disease pathogenesis. The targeted clearance of aggregated TDP-43 is therefore a key strategy for therapeutic intervention.
A balance of protein synthesis, folding and degradation is needed to maintain protein homeostasis (proteostasis) and this process is facilitated by molecular chaperones. When cells are acutely stressed by insults that affect proteostasis, such as heat shock, they induce a transcriptional programme orchestrated by the transcription factor heat shock factor 1 (HSF1) to upregulate selected chaperones, known as heat shock proteins (HSPs), to protect them from being overwhelmed by the accumulation of aberrant proteins. HSPs are categorized into several families according to their function and size (for review see Stetler et al., 2010). HSP90 and HSP70 are families of classic abundant chaperones that are expressed ubiquitously in most subcellular organelles. They execute their chaperone function by facilitating the folding of client proteins in an ATP-dependent manner. The human DnaJ (HSP40) family contains $50 proteins that act as cochaperones for HSP70 to promote the client-chaperone interaction by stimulating HSP70 ATP hydrolysis, producing a high affinity binding state for client protein refolding. Some members of the HSP40 family have HSP70-independent functions and can facilitate client proteins folding on their own or directly targeting them for degradation (Kampinga and Craig, 2010).
Aberrant protein aggregation is a pathological hallmark of almost all neurodegenerative diseases and the beneficial effect of increasing individual HSP expression has been demonstrated for several neurodegenerative disease models. For example, overexpression of the co-chaperones, DNAJB6b and DNAJB8, suppressed the aggregation and toxicity of poly-Q proteins (Hageman et al., 2010); overexpression of the small HSP, HSPB8, promoted autophagic removal of mutant SOD1 (Crippa et al., 2010); overexpression of HSP70 inhibited -synuclein fibril formation (Dedmon et al., 2005); overexpression of HSPB1 in mutant SOD1 transgenic mice improved neuronal survival in the early but not late stages of disease development (Sharp et al., 2008); and overexpression of DNAJB2a in SOD1 transgenic mice reduced SOD1 aggregation (Novoselov et al., 2013).
In this study, we explored the potential of HSPs to clear pathological TDP-43 aggregates and whether these chaperones might be deficient in animal models of TDP-43 proteinopathy and patients with ALS. We established a cellular model that recapitulated key features of the human TDP-43 proteinopathies and demonstrated that overexpression of dominant active HSF1, or the co-chaperone DNAJB2a, dramatically cleared insoluble TDP-43 aggregates and improved cell survival without altering levels of soluble TDP-43. We demonstrated that DNAJB2a mediates pathological TDP-43 clearance through a ubiquitin proteasome system (UPS)-independent, HSP70-dependent mechanism.
This suggests that the HSF1-DNAJB2a pathway refolds TDP-43 and returns it to its natural physiological state rather than facilitating its degradation. We also demonstrated that the levels of HSF1 and HSPs are markedly reduced in both a transgenic TDP-43 mouse model and in spinal cord tissues from sporadic ALS patients. These results highlight the potential of HSF1 and HSP activation as an important therapeutic strategy for the TDP-43 proteinopathies.

Plasmids and antibodies
The green fluorescent protein (GFP)-TDP-43 constructs were generated and used in a previous study (Nishimura et al., 2010). HSF1( + ) and HSF1(À) (Fig. 1A) were generated by altering the HSF1 constructs kindly provided by Professor Richard Voellmy (Xia et al., 1999). They were cloned with a V5 tag fused to the N-terminus into the pcDNA5/FRT/TO expression vector (Life Technologies). V5-chaperone constructs were kindly provided by Professor Harm Kampinga (Hageman and Kampinga, 2009) and Myc-DNAJB2a constructs were generated and used in previous studies (Chapple and Cheetham, 2003;Westhoff et al., 2005). All plasmid sequences were verified by DNA sequencing.

Cell culture and DNA transfection
HEK293T and SH-SY5Y cells were cultured using Dulbecco's modified Eagle medium (DMEM) and DMEM/F12 (Life Technologies) supplemented with 10% foetal bovine serum (Life Technologies), and maintained at 37 C, 5% CO 2 . Cells were plated a day before transfection and media were refreshed before plasmid DNA transfection using Lipofectamine Õ 2000 (Life Technologies). Cells were left for 48 h after transfection to be harvested for analysis unless otherwise stated.

Animals
All experiments were performed under the terms of the UK Animals (Scientific Procedures) Act 1986, and were approved by the Kings College, London ethics review panel. Compound transgenic mice were generated by crossing TDP-43 WT and TDP-43 Q331K mice as previously described (Mitchell et al., 2015).

Mouse and human spinal cord lysate preparation
Whole spinal cords from 8-week-old TDP-43 WTxQ331K animals and their single and non-transgenic littermates (n = 3-4) or 50 mg of spinal cord post-mortem tissues from controls or patients with sporadic ALS were homogenized in the appropriate amount of RIPA buffer (50 mM Tris pH8.0, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulphate with protease and phosphatase inhibitor), sonicated and stored at À80 C. Protein concentration was determined with DC TM Protein Assay Kit (Bio-Rad). Western blotting was run with 10 mg of protein lysate.

Immunohistochemistry
Eight-week-old, end-stage TDP-43 WTxQ331K mice and agematched littermates were anaesthetized and transcardially perfused with phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (PFA) in phosphate buffer. Spinal cord was postfixed in 4% PFA in 15% sucrose for 5 h, cryoprotected in 30% sucrose for 24 h and cut into 30 mm sections on a cryostat. Sections were incubated in rabbit anti-HSF1 antibody (1:250; Cell Signaling), washed and incubated with a biotinylated anti-rabbit antibody (1:1000; Vector) and then incubated in an ABC kit (Vector). Sections were imaged using a Zeiss light microscope and Axiovision software.

Immunopurification
Cells were harvested in immunopurification buffer (50 mM Tris pH7.4, 150 mM NaCl, 1% Triton TM X-100 with protease and phosphatase inhibitor). After centrifuging (14 000 rpm for 30 s at 4 C), the supernatant was collected and incubated with immunopurification antibody and Dynabead protein A (Life Technologies) overnight at 4 C. The Dynabead Õ protein Aantibody-protein complex was purified using magnetic separation and washed with immunopurification buffer before eluted in loading buffer.

Solubility fractionation
The fractionation for protein solubility was performed using a protocol described by Winton et al. (2008) with some minor modifications. Cells were harvested in RIPA buffer, sonicated and centrifuged at 12 000g for 20 min at 4 C. After centrifugation, the supernatant was collected as the RIPA solubility fraction. The pellet, after being washed once with RIPA buffer, was then suspended in 20% of the original lysis volume with urea buffer (7 M Urea, 2 M Thiourea, 4% CHAPS and 30 mM Tris pH8.5) and collected as the insoluble, detergent-resistant fraction. . Two-way ANOVA analysis indicates the difference in TDP-43 solubility is not contributed by TDP-43 genotypes or interaction between TDP-43 and HSF1, but by HSF1 alone (P 5 0.001). Both HSF1( + ) and HSF1(À) cause significant changes in the level of insoluble TDP-43 when co-transfected with GFP-TDP-43 when compared to the vector only control (P 5 0.001, Bonferroni post-test).

Western blotting and densitometry analysis
Protein quantification and western blotting were performed as described before (Nishimura et al., 2010). Five micrograms of cell lysate from the RIPA fraction and the equivalent liquid volume from the urea fraction were loaded. Western blot quantification was performed using the image analysis software, ImageJ (http://imagej.nih.gov/ij/). Integrated band intensities were normalized to that of loading control or the RIPA fraction.

Immunofluorescence
Cells for immunofluorescent analyses were fixed in 4% PFA (VWR) for 20 min and washed with PBS three times for 5 min. Cells were permeablized by incubation in PBS containing 0.5% Triton TM X-100 (Sigma) for 15 min at room temperature, followed by blocking in PBS containing 1% donkey serum for 1 h at room temperature. Cells were incubated with primary antibody diluted in blocking solution overnight at 4 C. After washing in PBS, cells were subsequently incubated with fluorescent secondary antibodies diluted in blocking solution for 1 h at room temperature. DAPI (Sigma) was then used to stain for nuclei before being mounted on coverslips using FluroSave (Calbiochem).

Cell survival assay
For cell survival assay, SH-SY5Y cells were trypsinized and stained with Calcein Violet 450 AM Viability Dye (500 nM, eBioscience) for 30 min at room temperature. After washing twice with PBS, cells were resuspended in PBS and analysed with the BD FACSCanto II (BD Biosciences).

Co-expression of HSF1( + ) reduces levels of insoluble TDP-43 protein
This study used dominant active and negative forms of HSF1 to investigate the effect of the HSR and HSP upregulation on TDP-43 proteinopathy in cellular models. HSF1 is a transcription factor that upregulates HSPs during the stress response. In the unstressed condition, HSF1 is bound to HSP90 via its regulatory domain and is kept in an inactive form. During cellular stress, HSF1 dissociates from HSP90, trimerizes, translocates into the nucleus, and binds to the heat shock element (HSE) through its DNA binding domain (DBD), where it promotes the transcription of downstream genes via its transactive domain (TAD) (Voellmy, 2005;Velichko et al., 2013). To activate the HSR we used a dominant positive HSF1, HSF1( + ), in which the regulatory domain is deleted (Fig. 1A). This modified HSF1 is therefore unable to be inactivated by HSP90 binding. Conversely, for comparison, we used a dominant negative construct, HSF1(À), with the TAD domain deleted (Fig. 1A). Although this HSF1(À) can still form trimers with endogenous wild-type HSF1 during stress, it is incapable of inducing the transcription of its targeted genes. The expression of HSF1( + ) increased the levels of endogenous HSP40 (DNAJB1) and HSP70 (HSPA1A), demonstrating HSF1( + ) activated the HSR as expected (Fig. 1B). On the other hand, levels of HSP40 and HSP70 were unchanged, or slightly decreased in HSF1(À) expressing cells, indicating HSF1(À) overexpression did not induce the HSR, and furthermore, may even inhibit basal HSR activation by dominant negative inhibition of endogenous HSF1.
The expression level of endogenous TDP-43 in our cells was not high enough to cause any proteinopathy phenotype similar to that seen in disease tissues, and was not found in the insoluble fraction ( Supplementary Fig. 1A and B). Overexpression of wild-type or mutant TDP-43 in HEK293 cells resulted in high levels of TDP-43 accumulating in the detergent-resistant fraction ( Fig. 1B and C). This result is consistent with previous reports showing that high cellular levels of TDP-43 caused protein aggregation resulting in a decrease in protein solubility. However, co-expression of TDP-43 with HSF1( + ) led to a significant reduction in the level of insoluble TDP-43 to $25% ( Fig. 1B and C, P 5 0.001). Conversely, HSF1(À) significantly increased the level of insoluble TDP-43 by $250% (Fig. 1B and C, P 5 0.001). The mutation status of TDP-43 when co-transfected with HSF1 did not significantly change TDP-43 protein solubility, nor did the presence of a mutation in TDP-43 when transfected alone (Fig. 1C). HSF1( + ) showed no obvious effect on the level of soluble TDP-43 ( Fig. 1B and  D). However, soluble GFP-TDP-43 was found to slightly decrease when co-expressed with HSF1(À) especially for the cytosolic TDP-43, GFP-dNLS TDP-43 ( Fig. 1B and D, P 5 0.01 for dNLS-TDP-43), suggesting the expression of HSF1(À) results in an increased tendency for overexpressed cytosolic TDP-43 to shift from soluble monomer to insoluble aggregates.
In summary, these results show that increased activation of the HSR by HSF1( + ) substantially enhances the clearance of insoluble TDP-43, whereas inhibition of the HSR via HSF1(À) expression results in an increase in insoluble TDP-43, likely resulting from an inhibition of the endogenous, basal HSR.

HSF( + ) rescues TDP-43 hyperphosphorylation
Aside from reduced solubility, TDP-43-linked pathology is also characterized by the presence of hyperphosphorylated TDP-43, which can be detected by a phospho-TDP-43 specific antibody. Phospho-TDP-43 was only detectable in the insoluble urea fraction, confirming that the pathological TDP-43 in our model is both insoluble and hyperphosphorylated ( Fig. 2A). The accumulation of insoluble, phosphorylated TDP-43 was completely inhibited in HSF1( + ) co-expressing cells ( Fig. 2A and B, P 5 0.01) and enhanced by HSF1(À) ( Fig. 2A and B, P 5 0.001). No phospho-TDP-43 staining was identified in GFP-vector transfected cells alone (Fig. 2C), indicating that TDP-43 phosphorylation was not caused by the stress of transient transfection but by the accumulation of aggregated TDP-43, which contained endogenous TDP-43 ( Supplementary Fig. 1C). Staining for phospho-TDP-43 in cells confirmed the western blot findings, with evidence of TDP-43 phosphorylation in $10% of GFP-TDP-43 expressing cells. HSF1( + ) substantially reduced levels of TDP-43 phosphorylation in both wild-type and mutant TDP-43 expressing cells without altering the TDP-43 transfection efficiency (Fig. 2D-H). On the other hand, HSF1(À) increased TDP-43 phosphorylation, which coincided with a marginal reduction in the frequency of GFP-dNLS TDP-43 expressing cells (Fig. 2H, P = 0.012). The genotype of the overexpressed TDP-43 had no significant impact on TDP-43 phosphorylation per se ( Fig. 2B and G). We also observed a slight change in TDP-43 protein subcellular localization in this model, as cells that co-expressed GFP-TDP-43 and HSF1( + ) tended to have a higher proportion of nuclear TDP-43, while cells that co-expressed GFP-TDP-43 and HSF1(À) displayed a higher proportion of cytosolic TDP-43 ( Supplementary  Fig. 2). In the aggravated condition where TDP-43 was co-expressed with HSF1(À), not only was phosphorylated TDP-43 identified in both the nucleus and cytoplasm, but $30% of the cytosolic phosphorylated TDP-43 inclusions were also ubiquitin-positive (Fig. 2I).
To further validate these findings in a neuronal model, we overexpressed GFP-TDP-43 with HSF1 in rat primary cortical neurons (Fig. 3). Overexpression of GFP vector did not induce TDP-43 phosphorylation in transfected neurons (Fig. 3B), which again demonstrates that exogenous protein expression is not the cause of TDP-43 hyperphosphorylation. When primary neurons were transfected with GFP-TDP-43 alone, expression was largely confined to nuclei (65.9 AE 1.48%) but cytoplasmic localization of phosphorylated TDP-43 was also observed in a significant minority of cells (31.2 AE 0.54%) (Fig. 3C, top). In HSF1( + )/GFP-TDP-43 expressing cells, all TDP-43 displayed a nuclear localization with no evidence of phosphorylation (Fig. 3C, middle). However, in cells co-expressing HSF1(À) and GFP-TDP-43, there was a dramatic reduction of nuclear TDP-43 and accumulation of hyperphosphorylated TDP-43 in the cytoplasm (Fig. 3C, bottom). The clearance of nuclear TDP-43 in cells containing large cytoplasmic aggregates is in agreement with observations of pathological TDP-43 in ALS patient tissues (Neumann et al., 2006). Of note, GFP-positive cells co-expressing HSF1(À) were very rare, and many had lost their neuron-like morphology were transfected with GFP-WT-TDP-43 (green) and V5-HSF1 (red). bIII-tubulin (magenta) is used as a neuronal marker. (B and C) GFP vector (B) or GFP-wild-type TDP-43 with V5-HSF1 (C) are transfected into the rat primary cortical neurons. Cells were fixed 6 days after transfection and stained for V5 (magenta), phospho-TDP-43 (red) and nuclei (DAPI, blue). Scale bar = 10 mm.
( Fig. 3A and C), suggesting that TDP-43 cytotoxicity is enhanced when the HSR is suppressed.
Our cellular model of TDP-43 proteinopathy replicates several features of human disease with the accumulation of insoluble phosphorylated and ubiquitinated TDP-43 in the cytoplasm. We have also shown that the activation of the HSR rescues TDP-43 pathology whereas blocking this response exacerbates TDP-43 accumulation.

HSF1( + ) rescues TDP-43-induced cell death
As reported in other cellular and transgenic animal studies, overexpression of TDP-43 can be cytotoxic to some cells, especially neurons (Ash et al., 2010;Barmada et al., 2010;Stallings et al., 2010;Wils et al., 2010). When either wildtype or ALS-associated mutant TDP-43 were overexpressed in the SH-SY5Y neuronal cell line, a significant reduction in cell survival was observed from those GFP-TDP-43 expressing cells (Fig. 4A and B and Supplementary Fig. 3), indicating that the overexpression of TDP-43 is indeed toxic to neuronal cells. This cytotoxicity can, however, be partially, but significantly rescued, by the presence of HSF1( + ) (Fig.  4C and D). Similar findings were obtained in cells overexpressing ALS-linked mutant TDP-43. Conversely, although overexpression of GFP-dNLS TDP-43 resulted in similar levels of cell death as the other GFP-TDP-43 constructs, expression of HSF1( + ) was only able to induce a . T-test was used to determine whether HSF1( + ) significantly rescues TDP-43 induced cell death. Ã P 5 0.05; ÃÃ P 5 0.01; ÃÃÃ P 5 0.001. non-significant (P = 0.246) reduction in cell death. Overexpression of TDP-43 in HEK293 cells did not show cytotoxicity (data not shown), supporting the observation in ALS studies that neuronal cells are particularly sensitive to TDP-43 cytotoxicity.
In summary, we have demonstrated that overexpressing TDP-43 in cells recapitulates pathological features of the TDP-43 proteinopathies and that the presence of these features results in an increase in cell death. Manipulating the heat shock response (HSR) by expression of HSF1( + ) reduces this pathological TDP-43, and improves cell survival. In contrast, when cells are deprived of their basal heat shock response activity by the expression of HSF1(À), the degree of TDP-43 proteinopathy is intensified.
DNAJB2a as a potential mediator of HSF1-induced TDP-43 aggregation suppression HSF1 is a transcription factor that is responsible for the induction of a wide range of HSPs and other stress-responsive proteins upon its activation (Velichko et al., 2013;Vihervaara et al., 2013). As overexpression of HSF1( + ) increased endogenous levels of HSP40 and HSP70, this indicated that HSF1( + ) was triggering the heat shock response as expected (Fig. 1B). To determine which chaperone proteins may participate in TDP-43 clearance, we screened a selection of HSPs, including HSP90, HSP70, Hsc70, HSP27 and most of the HSP40s from the DNAJA and DNAJB families that were previously screened for their ability to suppress polyQ aggregation (Hageman et al., 2010). Most of the HSPs tested displayed little or no effect on TDP-43 solubility, however, two HSP40s, DNAJB2a and DNAJB8a, resulted in a reduction of insoluble TDP-43 similar to that observed with HSF1( + ) (Fig. 5). As DNAJB8a has a very restricted expression pattern and can only be found in testis tissue, we did not test this further. DNAJB2a displays a neuronal enriched expression profile (Cheetham et al., 1992;Chapple and Cheetham, 2003;Hageman and Kampinga, 2009), and was chosen for further investigation.

DNAJB2a maintains the solubility of TDP-43 by interacting with HSP70
DNAJB2a, encoded by DNAJB2 and also known as HSJ1a, is a target of HSF1 (Vihervaara et al., 2013) and functions as a co-chaperone for HSP70 interacting through its J domain ( Fig. 6A; Cheetham et al., 1994). In addition to working with HSP70, DNAJB2a also binds client proteins independently and can target them to the proteasome for degradation through its ubiquitin interacting motif (UIM) (Fig. 6A; Westhoff et al., 2005). To investigate the nature of this HSF1-DNAJB2a-mediated TDP-43 aggregation clearance, we expressed GFP-wild-type TDP-43 with either wild-type or selected mutants of DNAJB2a. The H31Q mutation in DNAJB2a disrupts the interaction between DNAJB2a and HSP70, thereby impeding the cochaperone activity of DNAJB2a. The DNAJB2a-dUIM bears four point mutations in the UIM, which prevent DNAJB2a from delivering client proteins to the proteasome for degradation (Westhoff et al., 2005). Interestingly, the DNAJB2a-mediated reduction in insoluble/phosphorylated TDP-43 was not altered by the presence of the dUIM mutations, whereas the presence of the H31Q mutation completely disrupted this activity (Fig. 6B). These findings suggest that DNAJB2a reduces the level of insoluble TDP-43 by co-operating with HSP70 to enhance refolding and maintain solubility, rather than by causing the degradation of the insoluble protein. In support of this hypothesis, treatment of GFP-WT-TDP-43 expressing cells with the proteasome inhibitor, MG-132, which led to the build-up of insoluble TDP-43 (Fig. 6C), had minimal impact on HSF1( + )/DNAJB2a-mediated TDP-43 aggregation suppression (Fig. 6C). Both the DNAJB2a functional mutations study and the study of ubiquitin proteasome system inhibition by MG-132 suggest that protein degradation does not play a major role in HSF1-DNAJB2a-mediated TDP-43 aggregation clearance. This is further supported by our observation that autophagy inhibition by bafilomycin treatment showed minimal effect on HSF1( + )/DNAJB2amediated TDP-43 aggregation clearance ( Supplementary  Fig. 4).
An immunopurification study provided further evidence to support the existence of a DNAJB2a-HSP70-TDP-43 refolding complex, since TDP-43 was shown to interact with both DNAJB2a and HSP70 ( Fig. 6D and E). Although the co-chaperone's binding to the client protein is in general transient and dynamic, we were able to detect TDP-43 binding to all three forms of DNAJB2a. The binding to H31Q-DNAJB2a was stronger than the others, probably due to its inability to transfer the client protein to HSP70, therefore prolonging binding (Fig. 6D).
These results show that the HSF1-induced co-chaperone, DNAJB2a, reduced levels of insoluble TDP-43 not through degrading these aggregated, phosphorylated proteins, but by directing them to HSP70 to be refolded ( Fig. 6 and Supplementary Fig. 5). As TDP-43 is tightly controlled to maintain its cellular level, and is known to be aggregationprone especially under stress, our finding suggests a promising therapeutic strategy which uses the heat shock response mechanism to regain the neutral TDP-43 protein homeostasis.

HSF1/heat shock proteins are reduced in ALS mouse model and affected patient tissues
The HSR is a powerful mechanism to cope with cellular stress, which we show in this study can efficiently refold insoluble TDP-43 and dissolve features of TDP-43 proteinopathy; however, it has been shown to be compromised in some neurodegenerative conditions (Hay et al., 2004;Maheshwari et al., 2014). To characterize further the involvement of HSF1/HSPs in TDP-43 proteinopathy in ALS, levels of HSF1 and HSPs proteins were investigated in ALS mouse model and patient tissues.
To study the role of TDP-43 in the development and progression of ALS, lines of transgenic mice expressing either wild-type-TDP-43, Q331K-TDP-43 or both were established (Mitchell et al., 2015). A correlation between the level of TDP-43 protein and phenotypic severity was noted, with the strain that expressed the highest level of TDP-43 (TDP-43 WTxQ331K double transgene) suffering from the most severe motor deficits and TDP-43 pathology, and a drastically reduced life span compared to both single and non-transgenic mice (Mitchell et al., 2015).
Interestingly, HSF1 was decreased in the TDP-43 Q331K single transgenic mice that displayed a mild, non-lethal motor defect at later stage ( Fig. 7A and Mitchell et al., 2015), but was most significantly lost in diseased TDP-43 WTxQ331K mice (Fig. 7A). As a result of the loss of HSF1, levels of HSP70 and HSP40 were also decreased substantially in TDP-43 WTxQ331K mice, whereas HSP90 remained unaffected. Immunohistochemical staining also showed reduced levels of HSF1 in the remaining motor neurons in the end-stage diseased mice spinal cord sections, (Fig. 7B) indicating that the reduction in HSF1 protein was not purely a result of motor neuron loss.
Examination of HSF1 and the HSPs in human tissue identified a significant decrease in HSP70 (Fig. 7D) and HSP40 (Fig. 7E) despite no significant change in HSF1 levels between sporadic ALS and control tissues (Fig. 7C). It is known that total level of HSF1 is not always correlated with its level of activity, instead, certain post-translational modifications are required to activate the protein (Westerheide et al., 2009;Calderwood et al., 2010;Yih et al., 2012). This regulation in HSF1 activity is likely to be responsible for the difference we see in affected tissues in ALS, where HSP40 and 70 are reduced, despite no apparent change in the total levels of HSF1. We therefore propose that a reduction in HSF1 activity results in a reduction in HSPs, which in turn contributes to the enhanced accumulation and aggregation of TDP-43 in ALS patients.   Taken together, our results suggest that the HSR is compromised in both the TDP-43 transgenic ALS mouse model and in patient tissues, which could be initiated either directly or indirectly by long term TDP-43 protein overexpression. Additionally, these results demonstrate that a reduction in the HSR leads to decreased TDP-43 refolding efficiency, promoting the accumulation of phosphorylated insoluble TDP-43.

Discussion
In this study, we generated a cellular model of the human TDP-43 proteinopathies that developed detergent-resistant, hyperphosphorylated cytoplasmic insoluble TDP-43 aggregates that are cytotoxic. Co-expression of a dominant positive version of the master regulator of the heat shock response HSF1 induced a dramatic clearance of insoluble TDP-43 aggregates without affecting the levels of soluble TDP-43. Conversely the dominant negative version of HSF1 lead to a massive accumulation of insoluble and hyperphosphorylated TDP-43 aggregates confirming a vital role of the HSF1-HSP axis in physiological TDP-43 proteostasis. We have shown that HSF1-induced TDP-43 clearance is partly mediated by HSP70 and its co-chaperone DNAJB2a in a proteasome-and autophagy-independent manner. We propose that DNAJB2a recognizes and binds to insoluble TDP-43 taking it to the chaperone protein HSP70, which then refolds the protein. Once it has been returned to its soluble, properly folded status, TDP-43 is then released to perform its biological functions (Fig. 8). Because DNAJB2a mediates TDP-43 refolding without involving the proteasome or autophagy, it is able to reduce TDP-43 toxicity without changing the levels of soluble TDP-43.
We have also shown that levels of several HSR proteins are reduced in transgenic mice and sporadic ALS tissues implying that the HSF1-HSP axis may be compromised due to the disease process. Q331K TDP-43 transgenic mice develop a slowly progressive motor phenotype and have moderately reduced levels of HSF1 and HSP-40, whereas the severely-affected double transgenic mice have a dramatic loss of HSF1, HSP40 and HSP70. In human spinal cord tissues the levels of HSP70 and HSP40 were significantly reduced but not HSF1.
The data from our cellular, animal and human tissue studies indicate that the HSF1/HSP70/DNAJB2a pathway plays a vital role in the physiological maintenance of TDP-43 proteostasis. In the disease state, however, the HSR appears to be compromised and may contribute to the accumulation of insoluble TDP-43 protein aggregates and exacerbate defective proteostasis (Fig. 8).

Involvement of heat shock proteins in neurodegenerative disorders
Molecular chaperones are essential for maintaining proteostasis (Smith et al., 2015). Neurodegenerative disorders are characterized by the presence of aggregated aberrant proteins in affected neural tissues, so the role of HSR proteins and the opportunities for therapy is of great interest. Reduced levels of HSR components are found in affected neuronal tissues from patients with Huntington's disease (Maheshwari et al., 2014) and ALS (Anagnostou et al., 2010) and a failed stress response has been implicated in ageing and neurodegenerative disorders (Brehme et al., 2014). Overexpression of individual HSPs in cellular and transgenic animal models of disease have shown variable degrees of success. In some cases, overexpression of a single HSP was sufficient to reduce protein aggregation and improve survival, such as: HSP70 for -synuclein (Auluck et al., 2002;Dedmon et al., 2005), DNAJB6b and DNAJB8 for poly-Q (Hageman et al., 2010), HSP27 for tau and SOD1 (Shimura et al., 2004;Sharp et al., 2008) and HSPB8 for SOD1 (Crippa et al., 2010). In other cases, both HSP70 and HSP40 were required to clear aberrant protein aggregates (Krobitsch and Lindquist, 2000;Lotz et al., 2010). The differences are likely to be due to substrate specificity and the particular model used in the study. The HEK293T cells used in our study have high levels of HSP70 in basal conditions but relatively low levels of HSP40 (Fig. 1B). As the level of HSP40 was a limiting factor in our cellular model, it may explain why we observed no rescue of TDP-43 pathology in response to HSP70 overexpression alone (Fig. 5A), despite evidence that the interaction between HSP70 and DNAJB2a is essential for insoluble TDP-43 clearance (Fig. 6B). Unlike HSP70, DNAJB2a expression is specifically enriched in neuronal tissues (Chapple and Cheetham, 2003;Hageman and Kampinga, 2009) and mutations in DNAJB2 are linked to a rare recessive distal hereditary motor neuropathy and Charcot-Marie-Tooth disease type 2 (Blumen et al., 2012;Gess et al., 2014) implying a critical need for this co-chaperone. Overexpression of DNAJB2a was protective in G93A-SOD1 mouse model of ALS (Novoselov et al., 2013) and mutant huntingtin in R6/2 transgenic mice (Labbadia et al., 2012). These results, in combination with our own data on TDP-43, provide convincing evidence that DNAJB2a plays an important role in neuronal survival and the capacity of neurons to cope with aberrant Figure 8 Proposed model of HSR involvement in TDP-43 protein refolding and proteinopathy. In the healthy condition with a functional HSR, HSF1 activates the expression of HSPs such as HSP70 and DNAJB2a. DNAJB2a recognizes misfolded TDP-43 in the cytosol and brings it to the chaperone protein, HSP70. HSP70 catalyzes the refolding of TDP-43, returns it to its native soluble state, which enables TDP-43 to shuttle back to the nucleus. In the diseased condition, the HSR is impaired by either reduced protein levels or the decreased activity of HSF1. In either condition, this leads to a reduction of HSPs, which subsequently fail to efficiently engage misfolded TDP-43. As a result, the insoluble and phosphorylated TDP-43 proteins accumulate and form cytosolic aggregates in disease affected cells.
protein aggregation. Interestingly, both the refolding (J) and proteasome (UIM) domains of DNAJB2a were required for mutant SOD1 and huntingtin protein clearance, whereas we demonstrated that TDP-43 clearance was achieved by HSP70/DNAJB2a-mediated refolding and not proteasome clearance. The marked reduction in TDP-43 C-terminal phosphorylation indicates that the protein is in a near native and probably functional state. A similar result was seen where DNAJB2a suppressed the aggregation of mutant parkin and enhanced mitophagy in a J domain-dependent and UIM-independent manner (Rose et al., 2011). This result demonstrates that the diverse function of DNAJB2a to mediate protein clearance either through refolding or degradation depends on the nature and the structure of the protein client.
Interactions between HSPs and TDP-43 have been documented previously. Freibaum et al. (2010) conducted a proteomic screening which identified a group of HSPs interacting with TDP-43, including HSP70 and HSP40. These findings were later validated in cellular model Udan-Johns et al., 2014). Both studies suggested that the HSP-TDP-43 interaction plays a role in stress-induced TDP-43 aggregation. However, as the HSP expression pattern is cell type specific, the exact role of the HSP-TDP-43 interaction in the context of neurodegeneration is still yet fully explored.
Cellular stress and TDP-43 aggregation: is heat shock response activation enough to help? TDP-43 is an aggregation-prone protein that is found to form detergent-resistant aggregates or stress granules during conditions of cellular stress such as heat shock (Supplementary Fig. 6; Udan-Johns et al., 2014), osmotic, and oxidative stress (Dewey et al., 2011;McDonald et al., 2011;. Studies in cellular models show that despite TDP-43 aggregates forming quickly in response to stress, those aggregates can be readily disassembled when the stress is removed (Supplementary Fig. 6; Scotter et al., 2014). Several cellular mechanisms have evolved to participate in resolving potentially detrimental protein aggregates. The ubiquitin proteasome system has been found to be responsible for the majority of TDP-43 protein degradation, whereas the autophagy pathway is required to clear larger aggregates (Scotter et al., 2014). Indeed, we observed an increase in TDP-43 aggregation when the proteasome was inhibited by MG132 (Fig. 6B); however, neither proteasome nor autophagy inhibition reduced the effectiveness of HSF1 or DNAJB2a overexpression, further confirming that they do not appear to be targeting TDP-43 for degradation. Upstream of these pathways, the chaperone system facilitates protein folding and directs the degradation of misfolded proteins. However, due to their highly specified structures, neurons have higher demand for these chaperone and protein degradation systems to maintain protein homeostasis, which appear to fail in ageing and disease (Brehme et al., 2014). In addition, motor neurons have a higher threshold for the activation of the HSR than most other cells (Batulan et al., 2003). Thus there may be a delay in the activation of the chaperone system in response to stress, creating a cascade where the protein quality and degradation systems begin to fail and aberrant protein aggregates form. If and when, the HSR eventually does come into play, it may be too little too late. Indeed, there is no evidence to date to suggest that the HSR is activated in diseased tissues and may actually be inhibited (Fig. 7) (Anagnostou et al., 2010;Maheshwari et al., 2014). Our data suggest that the enforced activation of the HSR, in particular the neuronal specific co-chaperone, DNAJB2a, could effectively rescue TDP-43 proteinopathy by restoring aggregated TDP-43 to its native state.
HSP90 binds to and inactivates HSF1 by preventing it from entering the nucleus and inducing the HSR. HSP90 inhibitors such as 17-AAG have therefore been tested in cellular and animal models of neurodegeneration with some success (Waza et al., 2005;Fujikake et al., 2008;Rusmini et al., 2011;Gregory et al., 2012). However, as HSP90 is involved in cell cycle and cell survival pathways (Taipale et al., 2010), the risk of toxic side effects of HSP90 inhibitors limit their attractiveness as a potential therapy. We have shown that activation of HSF1/HSP70/ DNAJB2a pathway results in a dramatic clearance of TDP-43 aggregates and that directly targeting these HSR components is an important therapeutic opportunity for ALS.