Induction of the Unfolded Protein Response in Familial Amyotrophic Lateral Sclerosis and Association of Protein-disulfide Isomerase with Superoxide Dismutase 1*

Mutations in Cu/Zn superoxide dismutase (SOD1) are linked to motor neuron death in familial amyotrophic lateral sclerosis (ALS) by an unclear mechanism, although misfolded SOD1 aggregates are commonly associated with disease. Proteomic analysis of the transgenic SOD1G93A ALS rat model revealed significant up-regulation of endoplasmic reticulum (ER)-resident protein-disulfide isomerase (PDI) family members in lumbar spinal cords. Expression of SOD1 mutants (mSOD1) led to an up-regulation of PDI in motor neuron-like NSC-34 cells but not other cell lines. Inhibition of PDI using bacitracin increased aggregate production, even in wild type SOD1 transfectants that do not readily form inclusions, suggesting PDI may protect SOD1 from aggregation. Moreover, PDI co-localized with intracellular aggregates of mSOD1 and bound to both wild type and mSOD1. SOD1 was also found in the microsomal fraction of cells despite being a predominantly cytosolic enzyme, confirming ER-Golgi-dependent secretion. In SOD1G93A mice, a significant up-regulation of unfolded protein response entities was also observed during disease, including caspase-12, -9, and -3 cleavage. Our findings therefore implicate unfolded protein response and ER stress-induced apoptosis in the patho-physiology of familial ALS. The possibility that PDI may be a therapeutic target to prevent SOD1 aggregation is also raised by this study.

gene are associated with 20% of familial amyotrophic lateral sclerosis (FALS) cases (1), and when these mutations are overexpressed in transgenic rodents (2,3), motor neuron degeneration reminiscent of ALS results. Although SOD1 is thermally very stable (4), abnormal mutant SOD1 (mSOD1) aggregates are present in spinal cords of FALS patients and transgenic mice (5). The mechanism of mSOD1-mediated toxicity is unclear but is non-cell autonomous and involves apoptotic signaling (reviewed in Ref. 6). The selective toxicity for motor neurons also remains unresolved.
The disulfide status of proteins is largely regulated by ER stress-inducible enzymes. ER stress is triggered when misfolded proteins accumulate within the lumen, inducing the unfolded protein response (UPR) (12). The 78-kDa chaperone immunoglobulin-binding protein (BiP) controls activity of the three major UPR sensors: the kinase and endonuclease IRE1, the basic leucine-zipper transcription factor ATF6, and the PERK kinase (13). The combined effect of the activation of these three molecules is the up-regulation of genes encoding ER-resident chaperones and down-regulation of protein synthesis. Proteindisulfide isomerase (PDI) and endoplasmic reticulum protein 57 (Erp57) are both ER chaperones of the PDI family that catalyze the formation and rearrangement of intra-and inter-molecular disulfide bonds (14,15).
Although UPR is usually a short term homeostatic mechanism and necessary for cell survival, prolonged UPR leads to activation of ER-resident caspase-12, triggering apoptosis (16). The transcription factor C/EBP homologous protein (CHOP) (17) also regulates the transition from pro-survival to pro-apoptotic signaling during ER stress (18). There is some evidence of ER stress in transgenic SOD1 mice (reviewed in Ref. 19). BiP was up-regulated in spinal motor neurons of SOD1 H46R and SOD1 L84V mice (20). However, contrary results came from studies of SOD1 G93A mice (21), but subsequently intraneuronal deposits of BiP were found in this strain implying up-regulation (22). Evidence that ER stress induces apoptosis in ALS models is even more contradictory. Caspase-12 induction and activation This article has been withdrawn by the authors. The authors have become aware of several errors in the way images were presented in this manuscript and withdraw the article in the interests of maintaining their publication standards and those of the journal. The authors state that these presentational errors do not impact the underlying scientific findings of the article, which have been confirmed in other laboratories. The authors stand by the original scientific results as described. The authors state the following: while all actin blots were performed as described in the article and confirmed similar loading in each case, a portion of the actin immunoblot image from rat lysates in Fig. 2A was inadvertently reused in the mouse lysate panel images in Fig. 2A, in Fig. 2D, and in Fig. 8A. Due to their strong similarity, lanes 2 and 9 of the PDI immunoblot and lanes 7 and 9 of the SOD immunoblot in Fig. 3B were accidentally duplicated. Similarly, in Fig. 3C, lanes 1 and 9 of the PDI immunoblot and lanes 3 and 9 of the SOD immunoblot were accidentally duplicated. However, all mutants described in the article were included in the study, and the findings obtained were as reported. In Fig. 3C, because the same set of COS cell lysates were used in both studies, lanes 3-6 of the SOD1 immunoblot was published previously (Turner, B. J., Atkin, J. D., Farg, M. A., Zang, D. W., Rembach, A., Lopes, E. C., Patch, J. D., Hill, A. F., and Cheema, S. S. (2005) Impaired extracellular secretion of mutant superoxide dismutase 1 associates with neurotoxicity in familial amyotrophic lateral sclerosis. J. Neurosci. 25, 108-117), and hence the expression of SOD1 in these lysates was equivalent for both publications. Therefore, while the presentation of some of the images in the paper is incorrect, the images fully represent the scientific findings of the study as reported. The original paper can be obtained by contacting the authors.
were reported in spinal cords of SOD1 G93A mice prior to disease onset (21), but Sathasivam et al. (23) did not find increased caspase-12 cleavage at any stage of disease. Similarly, expression of CHOP has been reported both as unchanged (21) or up-regulated at disease onset (24) in SOD1 G93A mice. To date, there has been no comprehensive analysis of the full spectrum of ER stress and related apoptotic markers in relation to ALS. Recent evidence also implicates adverse SOD1 and ER interactions in disease. Although lacking a leader peptide, secretion of SOD1 was demonstrated in neuronal and non-neuronal cell lines (25). We reported the first evidence linking mSOD1 secretion to neurotoxicity in cellular and transgenic rodent FALS models (26). This was subsequently confirmed when mSOD1 was shown to associate with chromogranins in neurosecretory vesicles (27). Inhibition of SOD1 secretion by brefeldin A, an inhibitor of ER to Golgi transport (28), dissipated export and induced cytoplasmic accumulation of SOD1 (26,29). Furthermore, fragmentation of the neuronal Golgi apparatus is a consistent finding in SOD1-linked familial and sporadic ALS (30,31) and in transgenic SOD1 G93A mice (32). On the basis of these findings, SOD1 is implicated in dysfunction of the neuronal ER-Golgi system, leading to the proposal that modulation of the secretory pathway has a wider participation in pathogenesis than previously recognized.
In this study we provide compelling evidence that up-regulation of UPR markers occurs in transgenic SOD1 G93A rodents, implicating ER stress in the pathogenesis of ALS. PDI was also found in association with SOD1 in cellular and animal models of FALS, implying a biochemical interaction. Furthermore, pharmacological inhibition of PDI triggered SOD1 inclusion formation, suggesting a possible neuroprotective role for PDI in reducing mSOD1-induced aggregation. Taken together, these findings suggest new avenues of investigation for ALS therapy based on modulation of PDI and proteins of the UPR.

Animals
Transgenic rats derived from the SD-TgN (SOD1G93A) L26H line (Taconic Farms, Germantown, NY) were bred on a Sprague-Dawley (SD) background. Transgenic mice carrying the G93A human SOD1 mutation (G93A) were obtained from the line TgN (SOD1-G93A)1Gur (The Jackson Laboratory, Bar Harbor, ME). Wild type (WT) controls were age-and gendermatched nontransgenic animals. Animals were killed by intraperitoneal lethal injection (200 mg/kg sodium pentobarbitone, Lethabarb, Virbac, Australia). All methods conformed to the Australian National Health and Medical Research Council published code of practice for animal research and were approved by the Howard Florey Institute Animal Ethics Committee.

Protein Extraction
Animal Tissue-Spinal cord samples from the lumbar enlargement (L4 and L5) region were dissected from animals at postnatal day (p) 90 or 120 and stored at Ϫ70°C. Frozen tissue was homogenized in a mortar and pestle prior to the addition of 10 l of RIPA lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% (w/v) SDS, 1% (w/v) sodium deoxycholate, and 1% (v/v) Triton X-100) with 1% (v/v) protease inhibitor mixture (Sigma) per 1 mg of tissue). After 20 min of incubation at 4°C followed by centrifugation for 5 min at 15,000 ϫ g, the supernatant containing SDS-soluble proteins was collected and frozen at Ϫ20°C until required.
Cells-Cells were lysed in either Tris-NaCl (TN) buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl) or TN buffer containing 0.1% (w/v) SDS, both with 1% (v/v) protease inhibitor mixture, for 10 min on ice. Lysates were clarified by centrifugation at 15,800 ϫ g for 10 min and were frozen at Ϫ20°C until required. The resulting pellets were resuspended in TN buffer with 1% (w/v) SDS and ultrasonicated for 15 s to give the insoluble cell fraction.

Two-dimensional Gel Electrophoresis Separation
Protein samples from p60 rat spinal cords were added to solubilization solution (5 M urea, 2 M thiourea, 2% (w/v) CHAPS, 40 mM Tris, 65 mM dithiothreitol, 2% (w/v) sulfobetaine 3-10, 1% (w/v) carrier ampholytes, 0.002% (w/v) bromphenol blue) and sonicated for 10 s. Lysates were then clarified by centrifugation at 20,000 ϫ g for 10 min, and 250 g was loaded for isoelectric focusing; pH 4 -7 gradient 17-cm strips (Bio-Rad) were loaded via in-gel rehydration and pH 6 -11 gradient 18-cm strips (Amersham Biosciences) were cup-loaded at the anode. First dimension isoelectric focusing was performed for 23 h (80.4 kV-h) at a maximum voltage of 5000 V. The separating gel gradient was 8 -18% T large format polyacrylamide slab gel, and two-dimensional electrophoresis was performed for 2 h at 3 mA/gel and then 16 h at 15 mA/gel. Gels were stained using SYPRO Ruby fluorescent stain (Molecular Probes TM , Eugene, OR). Triplicate gels of each sample were run, with a representative gel from each sample used to indicate differentially expressed spots. Z3 Image Analysis software (Compugen, Tel Aviv, Israel) was used for image analysis.

Peptide Mass Fingerprinting by MALDI-TOF-MS and Data Base Searching
Differentially expressed protein spots were excised from the two-dimensional gels and digested with trypsin for 16 h at 37°C, followed by cleaning and concentration with a ZipTip (Millipore, North Ryde, Australia). A 0.5-l aliquot was spotted onto a sample plate with 0.5 l of matrix (␣-cyano-4-hydroxycinnamic acid, 8 mg/ml in 70% (v/v) AcN, 1% (v/v) trifluoroacetic acid) and allowed to air dry. Matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF-MS) was performed with an Applied Biosystems 4700 Proteomics Analyzer (Foster City, CA) with TOF/TOF optics in MS/MS mode. A Nd:YAG laser (355 nm) was used to irradiate the sample. The spectra were acquired in reflectron mode in the mass range 750 to 3500 Da.
The data were exported in a format suitable for submission to the data base search program MASCOT (Matrix Science Ltd., London, UK), and searches were performed against the NCBI nr data base. High scores in the data base search indicated a likely match and were confirmed or qualified by operator inspection. Positive identification of a protein spot was assessed by the percentage of sequence coverage, how well the masses matched significant peaks in the MS spectra, the MS/MS ion score (if applicable), the number of missed cleavages (if missed cleavages were present their location in the sequence is critical), and how well the molecular weight and isoelectric point (pI) of the identified protein matched.

Cell Culture and Transfection
Wild type SOD1 (wSOD1) and mSOD1 constructs encoding EGFP-tagged human SOD1 at the C terminus were generated as described previously (26). Mouse motor neuron-like NSC-34 cells were a generous gift of Dr. Neil Cashman (University of Toronto, Canada), and simian fibroblast COS-7 cells were provided by George Christopoulous (Howard Florey Institute, Melbourne, Australia). The rat pheochromocytoma PC12 cell line was supplied by Dr. Ora Bernard (Walter and Eliza Hall Institute of Medical Research, Melbourne, Australia). Cell lines were maintained in Dulbecco's modified Eagle's medium with 10% (v/v) fetal calf serum and 1% (v/v) penicillin/streptomycin (Invitrogen). PC12 cells were maintained in Dulbecco's modified Eagle's medium, 5% (v/v) fetal calf serum, 10% (v/v) horse serum, 0.5% penicillin/streptomycin and were differentiated in 60 ng/ml nerve growth factor (Sigma) for 5 days prior to transfection. Cells were subcultured in 24-well plates at a density of 1 ϫ 10 5 cells per well and were transfected transiently with plasmids (1 mg of DNA per well) using a 1:1 ratio of lipofection reagent ("Transfast," Promega, Australia) to DNA. Cells were examined 72 h after transfection with an inverted fluorescent microscope (Olympus). Cells containing prominent aggregates were counted in at least 4 wells and expressed as a percentage of total EGFP-positive cell transfectants. For the PDI inhibition studies, the cells were treated with 5 mM bacitracin (Tocris, Avonmouth, UK) immediately after transfection.
For metabolic labeling studies, the medium was replaced with cysteine-free Dulbecco's modified Eagle's medium (Invitrogen) containing 100 mCi/ml L-labeled cysteine (PerkinElmer Life Sciences), 24 h after transfection. The cells were harvested at 72 h post-transfection. Autoradiography was performed using 1-5 days exposure to Super RX x-ray film (Fuji).

Immunoprecipitation
Mouse spinal cord extracts (20 g of total protein) or whole transfected cell lysates (250 l) were incubated with anti-SOD1 or -PDI (1:750) antibodies and 30 l of 50% (w/v) protein A-Sepharose CL-4B (Amersham Biosciences) in Tris buffer (50 mM Tris-HCl, pH 7.5, 0.02% (w/v) NaN 3 ) on a rotating wheel overnight at 4°C. After centrifugation at 15,800 ϫ g for 1 min, pellets were washed twice in Tris buffer for 10 min each. For Western blotting, immunoprecipitates were released by incubation in 2% (w/v) SDS sample loading buffer.

Immunocytochemistry
Transfected NSC-34 cells were fixed in 4% (w/v) paraformaldehyde in phosphate-buffered saline (PBS) for 5 min. Cells were then permeabilized in 0.1% (v/v) Triton X-100 in PBS for 2 min, blocked for 30 min in 5% (w/v) milk in PBS, and incubated with anti-PDI antibody (1:200) overnight. Following three PBS washes (5 min each), anti-rabbit Ig Alexa Fluor Red 1:100 (Molecular Probes) was added for 3 h. After applying a coverslip using Dako fluorescent mounting medium, the cells were imaged using confocal microscopy.

Subcellular Fractionation
Untransfected NSC-34 cells (4 ϫ 10 6 cells) grown in three confluent flasks were pelleted by centrifugation, resuspended in cold 0.25 M sucrose and 0.1 mM EGTA in 10 mM Tris-HCl, pH 7.4, buffer with 1% (v/v) protease inhibitors, and homogenized before differential centrifugation was performed. After sedimentation of the nuclear fraction at 600 ϫ g for 10 min at 4°C, mitochondria were sedimented from the supernatant by centrifugation at 6,800 ϫ g for 20 min. After discarding the fluffy layer, the pellet was resuspended in sucrose buffer and centrifuged at 12,100 ϫ g for 5 min. Microsomes were prepared by centrifugation of the post-mitochondrial supernatant fraction at 100,000 ϫ g for 50 min at 4°C. The pellet was washed three Analysis of pH 4 -7 protein spot intensities in wild type mice compared with SOD1 G93A mice (S1 and S2) at p60 The table includes the fold changes in relative expression; this is a measure of spot quantity based on all differential expression values related to the intensity of the spot between each sample, measured in parts/million of the total spot quantity of the image. The protein spots that were not present on one or more gels were given a value that represents background staining in the gels, thus allowing a fold increase to be estimated. times with sucrose buffer. For the last wash, sucrose buffer without EGTA was used. The supernatant was centrifuged at 300,000 ϫ g for 60 min, and the final supernatants were designated as the cytosolic fraction. All the pellets were sequentially collected and resuspended in PBS with 1% (v/v) protease inhibitor mixture.

Immunohistochemistry
Mice were perfused transcardially with PBS followed by 4% (w/v) paraformaldehyde in PBS. Fixed lumbar spinal cord tissue was dissected out and placed in 30% (w/v) sucrose in PBS for 48 h. The tissue was then frozen and cut into 50-m thick sections on gelatin-coated slides. The sections were stained with an anti-PDI antibody (1:1500) for 48 h, followed by anti-rabbit Ig biotin conjugate (1:500, Vector Laboratories, Burlingame, CA) for 2 h. Sections were then incubated for 1 h in avidin-peroxidase (1:5000, Sigma) and reacted with cobalt and nickel-intensified diaminobenzidine (Sigma) for 30 min. Hydrogen peroxide (0.01% v/v) was further added to the diaminobenzidine solution for 5 min. Rinses (three, per 10 min) in PBS were performed between each step. Sections were then dehydrated in ethanol solution, and cleared prior to the application of a coverslip with 1,3-diethyl-8-phenylxanthine (Merck).

Statistics
Data were tested using either one-way analysis of variance (ANOVA) with Tukey's post hoc test (with 0.05 significance) or two-way ANOVA using Bonferroni's post test, and expressed as means Ϯ S.E.

PDI and Erp57 Are Significantly Up-regulated in Spinal
Cords of Both Transgenic SOD1 G93A Rats and Mice-A proteomic approach was employed to identify differentially expressed proteins associated with the onset and progression of motor neuron disease. Spinal cord homogenates from WT and SOD1 G93A rats at postnatal age 60 days (p60) were separated by two-dimensional gel electrophoresis. Thirty protein spots showed differential expression between the genotypes using either pH 4 -7 or pH 6 -11 gradients. Proteins that showed consistent expression patterns in different SOD1 G93A animals were selected for further analysis by MALDI-TOF-MS; 5 upregulated and 10 down-regulated proteins were chosen. Twelve of these proteins were confidently identified (defined as a MAS-COT data base score of 100 or greater); 3 of these from the pH 4 -7 gradient are detailed in Tables 1 and 2. Triplicate gels of each sample were run, and representative gels are shown in Fig.  1. Spot 16 was identified as the human SOD1 transgene product, which was increased up to 20-fold in the SOD1 G93A rats, with 50% sequence coverage by MALDI-TOF-MS analysis. Spot 13 was increased up to 7-fold and identified as PDI, with 36% coverage by MALDI-TOF-MS. Spot 14 was characterized as Erp57, a second member of the PDI family, increased up to 6-fold, with 39% coverage by MALDI-TOF-MS analysis. A previous proteomic study did not identify the differential regulation of PDI family members (33). However, pooled tissue extracts were used that may mask identification of some proteins, and SOD1 G93A mice were used instead of rats.
The proteomic results were supported by quantitative Western blot analysis of spinal cord extracts from SOD1 G93A rats and mice, performed at two postnatal ages. These findings confirmed a significant up-regulation (ϳ8-fold) of PDI at disease end stage (p120) in both the SOD1 G93A rats ( p Ͻ 0.01) and SOD1 G93A mice (ϳ30-fold, p Ͻ 0.001; Fig. 2, A-C). Similarly, FIGURE 1. PDI and Erp57 protein spots are up-regulated in transgenic SOD1 G93A rats compared with wild type (WT) controls. Lumbar spinal cord extracts from one WT and two SOD1 G93A animals (S1 and S2) at p60 were separated by isoelectric focusing using 17-cm, pH 4 -7, IPG strips. Second dimension SDS-PAGE was performed on 8 -18% gradient polyacrylamide gels, and the proteins were visualized by SYPRO Ruby fluorescent staining. The panel shows the specific regions of the gel containing the spots of interest. MALDI-TOF-MS analysis identified spot 13 as rat PDI, spot 14 as rat Erp57, and spot 16 as human SOD1 (Tables 1 and 2). Western blot analysis of Erp57 in the SOD1 G93A mouse revealed a significant increase compared with WT at p120 (ϳ2.5-fold, p Ͻ 0.01; Fig. 2, D and E). Expression of hSOD1-EGFP Induces PDI in the Motor Neuron-like Cell Line NSC-34 but Not in Fibroblast (COS-7) or Nonmotor Neuronal (PC12) Cells-We showed previously that mutant, but not wild type SOD1 (wSOD1), forms intracellular fluorescent SOD1 aggregates that correlate with cell death in transfected motor neuron-like NSC-34 cells (26). Six mutants were studied, with either similar enzymic activity to wSOD1 (A4V, G37R, D90A, G93A, G93C) or with negligible activity (H46R). All mutants except D90A form SOD1 inclusions. Hence, we examined the expression of PDI in NSC-34 cells transiently expressing wSOD1 or mSOD1 tagged with EGFP. NSC-34 transfectants revealed increased amounts of PDI (measured by quantitative Western blotting) compared with FIGURE 2. PDI and Erp57 are up-regulated in SOD1 G93A rats and mice at disease end stage. Soluble extracts of lumbar spinal cords (20 g per lane) from SOD1 G93A rats or mice with wild type controls were separated by SDS-PAGE and immunoblotted using anti-PDI antibody (A) or anti-Erp57 antibody (D). Two postnatal ages were examined as follows: p90, the onset of symptoms, and disease end point (p120). Immunoblots were stripped and re-probed for ␤-actin as a loading control. Quantification of immunoblots for rat PDI (B), mouse PDI (C ), and mouse Erp57 (E ) are shown. The total amount of protein per lane was quantified for each SOD1 G93A animal and plotted as a ratio relative to the corresponding wild type amount at each postnatal age. Five different animals for each age and genotype were analyzed. Data are presented as mean Ϯ S.E.; **, p Ͻ 0.01; ***, p Ͻ 0.001 versus wild type by one-way ANOVA with Tukey's post hoc test.
untransfected and vector alone expressing cells (Fig. 3A, left  panel). In addition, A4V and G93C mutants induced the greatest expression of PDI in NSC-34 cells (Fig. 3D). Interestingly, these mutants also formed significant numbers of inclusions (p Ͻ 0.05 and p Ͻ 0.01, respectively; see Fig. 3E). SOD1 immunoblotting confirmed similar transfection efficiency in each group of cells (Fig. 3A, right panel). The induction of PDI observed in NSC-34 cells was not as great as that seen in vivo, probably because only a subpopulation of cells was expressing wSOD1 or mSOD1 because of the transient transfection procedure.
The induction of PDI by SOD1 mutants was specific to motor neuron-like cells because transient expression of hSOD1-EGFP plasmids in nerve growth factor-differentiated PC-12 cells and non-neuronal COS-7 cells failed to alter PDI expression (Fig. 3,  B and C). In addition, far fewer mSOD1 inclusions were prevalent in these two cell lines compared with transfected NSC-34 cells (26) (data not shown).
PDI Binds to Endogenous Mouse SOD1, wSOD1, Mutant hSOD1-EGFP in NSC-34 Cells-We then used immunoprecipitation to determine whether there was an association between PDI and SOD1. Control experiments were performed to determine the specificity of the precipitating antibodies. Immunoprecipitations with either anti-SOD1 or anti-PDI antibodies were followed by Western blotting with anti-ac-

Role of PDI and UPR in ALS
tin antibodies. Actin was not present in immunoprecipitated untransfected NSC-34 cell lysates but was present in the total cell lysates (Fig. 4A, right panel). The same immunoprecipitates were immunoblotted using an anti-p38MAPK antibody, and p38 MAPK was not detected (data not shown). A lack of reactivity between the anti-SOD1 antibody and PDI and the anti-PDI antibody and SOD1 was shown by performing anti-PDI and anti-SOD1 enzyme-linked immunosorbent assays (data not shown).
Cell lysates from NSC-34 cells transfected with wSOD1 or mSOD1-EGFP were subjected to immunoprecipitation using the anti-SOD1 antibody. Western blot analysis of immunoprecipitated proteins revealed that PDI was co-precipitated by the anti-SOD1 antibody (Fig. 4B, top panel, ϳ50-kDa band), supporting a physical interaction between SOD1 and PDI in culture. PDI was also co-immunoprecipitated from untransfected cells, revealing that endogenous mouse SOD1 also binds to PDI. When this blot was stripped and reprobed with a SOD1 antibody, both human EGFP-tagged SOD1 and endogenous mouse SOD1 were detected, indicating that both proteins were successfully immunoprecipitated (Fig. 4B, bottom panel).
The reverse experiment was performed to confirm the observed binding between SOD1 and PDI. Cell lysates from NSC-34 transfectants were immunoprecipitated using an antimouse PDI antibody. Western blot analysis confirmed the presence of both the human EGFP-tagged SOD1 (Fig. 4C, top panel, ϳ50-kDa band) and endogenous mouse SOD1 (ϳ16-kDa band). PDI immunoprecipitation was shown by re-staining with the anti-mouse PDI antibody (Fig. 4C, bottom panel).
In Response to SOD1 Expression in NSC-34 Cells, Newly Synthesized PDI Binds to SOD1-The PDI detected in SOD1 NSC-34 transfectants may represent an endogenous pool already present in cells, or may represent protein induced specifically in response to SOD1 transfection. To assess which was the case, metabolic labeling using L-[ 35 S]cysteine was performed (SOD1 does not contain methionine residues), beginning at 24 h post-transfection and continuing until 72 h. Using an anti-PDI antibody, Western blotting of soluble cell lysates previously immunoprecipitated with SOD1 was performed, followed by autoradiography when the chemiluminescence was completely extinguished (Fig. 4D). A major band of ϳ60 kDa was observed on the autoradiograph (Fig. 4D), which when superimposed on the immunoblot co-localized with the PDI signal. Although it is possible that several SOD1-binding proteins of the same size may contribute to this band, it is likely that significant amounts of PDI were interacting with SOD1. In contrast, untransfected cells and cells transfected with EGFP alone did not induce a significant amount of PDI. This suggests that SOD1 expression by transfection had directly induced the production of PDI, confirming the findings of Fig. 3. Conversely, an immunoprecipitation with the PDI antibody revealed a large number of bands whose identity could not be resolved, con-

. Wild type and mutant SOD1 co-precipitate with PDI and induce PDI expression in NSC-34 cells.
A, the anti-SOD1 and anti-PDI antibodies do not co-precipitate actin or p38MAPK (data not shown). Abbreviations are as follows: TCL, total cell lysate; IP, immunoprecipitate; VEC, vector. B, top panel, coprecipitation of NSC34 transfectants with an anti-SOD1 antibody, followed by immunoblot analysis using an anti-PDI antibody. Bottom panel, SOD1 immunoprecipitation was confirmed by re-probing the membrane with the anti-SOD1 antibody. C, top panel, the reverse experiment was performed; co-precipitation of NSC34 transfectants with an anti-PDI antibody, followed by immunoblot analysis using an anti-SOD1 antibody, confirming the interaction between SOD1 and PDI. Both hSOD1-EGFP and endogenous mSOD1 are precipitated. Bottom panel, PDI immunoprecipitation was confirmed by re-probing the membrane with the anti-PDI antibody. D, NSC34 hSOD1-EGFP transfectants were metabolically labeled with L-[ 35 S]cysteine for 48 h, and the cell lysates were coprecipitated using an anti-SOD1 antibody. Immunoblot analysis revealed the PDI band (data not shown). Figure shows autoradiograph of this blot to reveal the proteins synthesized 24 -72 h post-transfection. The 60-kDa major band co-localized with the PDI band on the immunoblot, suggesting that PDI is a major protein interacting with SOD1. E, left panel, soluble extracts of lumbar spinal cords (20 g per lane) from SOD1 G93A mice plus wild type littermate controls (two duplicate lanes of each) were immunoprecipitated using an anti-SOD1 antibody, followed by immunoblot analysis using an anti-PDI antibody. Right panel, co-precipitation of NSC34 transfectants with an anti-PDI antibody followed by immunoblot analysis using an anti-SOD1 antibody confirms the interaction between SOD1 and PDI in these extracts. At least three separate immunoprecipitations were performed in each case (A-E ).
firming the ability of PDI to bind to a large number of proteins that require folding within the cell (data not shown).
More PDI was found in total cell lysates from the most aggregating mSOD1 transfectants (Fig. 3). However these mutants did not appear to bind any more PDI in the immunoprecipitation studies. This may relate to a greater ability of PDI to bind to endogenous SOD1 or to the possibility that despite the induction of PDI with mSOD1, mSOD1 could not bind to PDI as efficiently as wSOD1. Alternatively, as a much greater proportion of mSOD1 forms intracellular inclusions than wSOD1, it is possible that PDI may interact with insoluble mSOD1, given its role as a chaperone, which would not be detected in the soluble cell lysates used in this study. Indeed, preliminary analyses of the insoluble fraction of cell lysates of mSOD1 and wSOD1 indicated that a much greater proportion of PDI was associated with A4V and G93A mSOD1 than wSOD1 (data not shown). However, further studies are required to determine the relative binding abilities of mSOD1 and wSOD1 to PDI.
PDI and SOD1 Are Closely Associated in Mouse Spinal Cord Tissue-Spinal cord homogenates from transgenic SOD1 G93A and WT mice at p120 were then subjected to immunoprecipitation using the anti-SOD1 antibody. Western blot analysis revealed that PDI was co-precipitated by the SOD1 antibody (Fig. 4E, left panel), indicating a physical interaction between SOD1 and PDI in the lumbar spinal cord extracts. A much fainter PDI band was observed in nontransgenic spinal cord, probably reflecting the much greater quantities of SOD1 present in SOD1 G93A mice because of expression of the mutant transgene product and induction of PDI at p120 in these animals. The reverse experiment was also carried out; spinal cord homogenates from SOD1 G93A and WT mice at p120 were immunoprecipitated with the anti-PDI antibody, followed by Western blotting with anti-SOD1 antibody. In both cases, a band of ϳ20 kDa was observed (Fig. 4E,  right), suggestive of a PDI-SOD1 association in mouse spinal cord tissue.
In support of this we observed that spinal motor neurons express high levels of PDI. At p120 there was intense immunoreactivity to PDI in the cell bodies and axons of spinal motor neurons of both wild type and SOD1 G93A animals (Fig. 5, A  and B, left). Furthermore, PDI was distributed to large, intracellular inclusions within neurons of SOD1 G93A mice (Fig. 5B, right) that were not present in control animals.
Inhibition of PDI Augments the Burden of SOD1 Inclusions in Transfected NSC-34 Cells-The metallo-antibiotic bacitracin inhibits PDI activity (34 -36) and was used to investigate the effect of depleting functional PDI on the formation of inclusions in NSC-34 cells. Cells were transiently transfected with wSOD1 and mSOD1-EGFP constructs in the presence and absence of 5 mM bacitracin; the proportion of cells containing inclusions was quantified. In the presence of bacitracin, most hSOD1-EGFP mutants formed inclusions with increased frequency (Fig. 6). Interestingly, the two proteins least prone to aggregation, wSOD1 and D90A, were most sensitive to PDI inhibition as assessed by increased inclusion frequency. wSOD1 rarely produces inclusions in the absence of bacitracin, suggesting that PDI normally prevents wSOD1 aggregation. In contrast, the most aggregating mutants were least sensitive to PDI inhibition implying that PDI could not prevent these mSOD1 proteins from aggregation to the same degree.
PDI and SOD1 Co-localize with Aggregates in NSC-34 Cells-Next we examined the distribution of PDI in cells that formed large mSOD1 aggregates. NSC-34 cells expressing mutant A4V were chosen because these cells form large, multiple SOD1 inclusions that are easily visualized (Fig. 7A, i). When these cells were immunostained with an anti-PDI antibody (Fig. 7A, ii), PDI co-localized with the green fluorescent SOD1 inclusions FIGURE 5. Spinal cord motor neurons express high levels of PDI that recruit to abnormal inclusions in SOD1 G93A mice. A, lumbar spinal cord sections of WT mice at end stage (p120) were immunostained with an anti-PDI antibody. Intense immunoreactivity was observed in the motor neurons (solid arrows). B, lumbar spinal cord sections of transgenic SOD1 G93A mice at end stage (p120) were immunostained with an anti-PDI antibody. Intense immunoreactivity was observed in the motor neurons (solid arrows), and recruitment to spinal motor neuron inclusions was visualized (dotted arrows). Staining both in the cytoplasm and axons of motor neurons was observed. Magnification: ϫ20; inset ϫ100; scale bar, 50 m. (Fig. 7A, iii) consistent with the notion that PDI and mSOD1 are both components of these aggregates.

SOD1 Is Present and Interacts with PDI in the Microsomal Fraction of NSC-34 Cells-
The association of mSOD1 with PDI, primarily an ER-resident enzyme, implies that SOD1 may localize to the ER. Hence, the microsomal fraction of untransfected NSC-34 cells was examined for the presence of SOD1. Cytoplasmic, nuclear, mitochondrial, and microsomal fractions of NSC-34 cells were prepared by subcellular fractionation. Western blots of each fraction demonstrated PDI and SOD1 in all cellular compartments (Fig. 7B), although PDI was enriched in microsomes as expected which was hence used as a microsomal marker. Although conventionally thought of as an ER molecule, PDI is also found in the nucleus, cytosol, extracellular space (15), and mitochondrial outer membrane (37). Cross-contamination from the other fractions was excluded by Western blotting for marker proteins native to each subcellular fraction. Importantly, endogenous mouse SOD1 was present in the microsomal fraction of NSC-34 cells despite its usual cytosolic location, supporting a physiological interaction with ERresident proteins such as PDI. To determine whether the interaction of SOD1 and PDI also occurs in the secretory pathway, the microsomal fraction of lysates from untransfected NSC-34 cells was immunoprecipitated using either the anti-SOD1 or anti-PDI antibodies. Western blot analysis of immunoprecipitated proteins revealed that PDI was co-precipitated by the anti-SOD1 antibody (Fig. 7C, left panel), and SOD1 was coprecipitated by the anti-PDI antibody (Fig. 7C, right panel), supporting a physical interaction between SOD1 and PDI in the ER. Taken together, the results from the mSOD1 NSC-34 transfectant studies suggest that PDI protects against aggregate formation, co-localizes with intracellular aggregates formed by mSOD1, and binds to both wild type and mSOD1.
UPR Sensors and Caspases Are Up-regulated in Symptomatic SOD1 G93A Mice-Based on the induction of ER chaperones from the proteomic analysis, we hypothesized that elements of the UPR may be activated in transgenic SOD1 G93A rodents. We therefore examined the SOD1 G93A mouse model for evidence of further ER stress and SOD1-PDI interaction at symptom onset (p90) and at disease end stage (p120).
Quantitative Western blotting revealed up-regulation of UPR markers in SOD1 G93A mice at both postnatal ages (Fig. 8). PERK, ATF6, CHOP, and IRE1 were all significantly up-regulated ( p Ͻ 0.05) at p90 and at disease end stage (p120). By p120 the ER chaperones BiP, Erp57, and PDI were also up-regulated ( p Ͻ 0.05), particularly PDI (up to 30-fold) (Fig. 2C). Both PERK (5-fold) and IRE1 (10-fold) had also increased between p90 and p120. This evidence indicates that up-regulation of the chaperone molecules was foreshadowed by an ER stress response. The blots were stripped and re-probed for ␤-actin, demonstrating that loading was consistent in all lanes. The induction of all three major UPR components, PERK, IRE1 and ATF-6, implies ER stress signaling in mutant SOD1-mediated neurodegeneration (Fig. 8B).
Activated caspase-12, indicated by the appearance of the cleaved form of the enzyme, was found at p90 and p120 in SOD1 G93A mice (Fig. 8A). Together with up-regulation of CHOP, this suggested that the ER stress response in SOD1 G93A mice had engaged apoptotic pathways. Quantitative Western blotting revealed significant activation of downstream caspases-9 ( p Ͻ 0.001) and -3 ( p Ͻ 0.05) in SOD1 G93A mice at disease end stage (Fig. 8C), confirming apoptotic signaling had commenced by this time point.

DISCUSSION
The major findings of this study are that UPR-induced chaperones, stress sensor kinases, and apoptotic effectors are up-regulated in transgenic SOD1 G93A mouse spinal cords, implying that ER stress plays a role in SOD1-mediated neurodegeneration. Furthermore, the ER chaperone PDI associates with SOD1 and co-localizes with mSOD1 inclusions. The bacitracin inhibition results suggest that PDI prevents the formation of SOD1 aggregates in NSC-34 motor neuronal cells. Taken together, these findings point to the potential importance of SOD1-PDI interactions in mSOD1 aggregation and hence disease.
The Role of ER Stress in ALS-This study provides strong evidence that UPR induction leads to caspase-12 cleavage and hence ER stress promoted apoptosis in SOD1 G93A mice. After exposure to ER stress, the first pathway activated is translational repression mediated by PERK (38), which prevents further influx of nascent proteins into the ER lumen. The activated form of ATF6 is a transcription factor for UPR-responsive genes, including CHOP (39). There was already a large up-regulation of PERK, ATF6, and IRE1 by p90 (Fig. 8), indicating that UPR pathways were already underway at disease onset. BiP, PDI, and Erp57 are up-regulated by the ATF6 pathway, and this study shows that their induction occurs later in the UPR process than CHOP.
The activation of CHOP and the cleavage of caspase-12 by p90 suggests that pro-apoptotic pathways were already triggered at the onset of symptoms. The activation of downstream caspase-9 and the executioner caspase-3 by disease end stage is consistent with previous reports (21) and implies that apoptosis was already progressing. However, other studies have not detected caspase-12 cleavage in transgenic SOD1 G93A mice (23). Further studies are required to correlate the timing of up-regulation of these executioner molecules with cell loss and symptoms. Caspase-12 cleavage caused by misfolded protein has been implicated in cell death of cultured neurons (16,40).
However, caspase-12 cleavage was not detected in NSC-34 cells expressing mSOD1 G93A, although sensitivity to ER stress-induced apoptosis may be lost in generation of the stable cell transfectants used (23). Furthermore, aggregated SOD1 must be expressed at high levels before ER stress is evident in cell culture (20).
The demonstration of ER stress in ALS suggests an ER localization for at least a proportion of the pools of SOD1 in cells. Wild type SOD1 is normally cytoplasmic, with smaller amounts in the nucleus (41)(42)(43), mitochondrial outer membrane, inter-membrane space, and matrix (44); and SOD1 is not traditionally associated with the ER. However, morphological ER abnormalities have been reported in human ALS patients and transgenic mice (45,46). In mSOD1-linked FALS culture models, SOD1 aggregates co-localize with ER markers, and the ER volume is expanded (20,47). mSOD1 also co-localized with ER-resident BiP in transfected fibroblasts (20) and BiP-immunoreactive Lewy body-like inclusions in spinal motor neurons of transgenic mice (22), implying SOD1 aggregates at least partially in the ER. A recent study showed that mutant SOD1 was associated with chromogranins in secretory vesicles, in the trans-Golgi network, and in microsome preparations from SOD1 G93A mice (27). In hepatitis and cirrhosis, SOD1 is found in the ER, vesicles, and Golgi apparatus of hepatocytes (48). Export of SOD1 is linked to a brefeldin A-sensitive pathway via microvesicles, implying an ER-Golgi route, also suggesting ER localization of SOD1 (29,49). Therefore, this study combined with previous findings confirms an ER presence for SOD1.
The reason for the existence of SOD1 in the ER is unclear, although one possibility is to form native SOD1 disulfide bonds before transit to other locations. In neurons and glia, the cytosol is strongly reducing (GSH/GSSG ratio of 1) (50). SOD1 is highly unusual among intracellular proteins in forming stable disulfide bonds under these conditions. PDI is also found and is potentially active in the nucleus, cytosol, and extracellular space (15). In contrast, the ER is oxidizing (GSH/GSSG ratio of 0.5) (51), hence facilitating disulfide bridge formation. The wSOD1 disulfide bond is remarkably stable, even under iii, the yellow staining shows areas of both SOD1 and PDI immunoreactivity. Scale bar, 30 m. At least five groups of cells were examined. B, subcellular fractionation of untransfected NSC-34 cells lysates. Cytoplasmic (Cyt), nuclear (Nuc), mitochondrial (Mito), and microsome (Mic) fractions were prepared, and Western blotting for SOD1, PDI, and marker proteins native to each subcellular fraction was performed. Cytochrome oxidase (COX) was used as a mitochondrial marker; Sp1 was used as a nuclear marker. Three independent experiments were performed. C, left panel, co-precipitation of the microsomal fraction from untransfected NSC34 cells with an anti-SOD1 antibody, followed by immunoblot analysis using an anti-PDI antibody. Right panel, co-precipitation of the microsomal fraction from untransfected NSC34 cells with an anti-PDI antibody, followed by immunoblot analysis using an anti-SOD1 antibody. strongly reducing conditions (10). Hence, if formed correctly in the ER, the bond should remain intact in the cytosol, despite unfavorable conditions. SOD1 is highly abundant, making up ϳ1% of total cytosolic protein. Thus, even a fraction of total mSOD1 misfolded in the ER may be sufficient to induce ER stress. It is also noteworthy that ER targeting of wSOD1 or mSOD1 induced strong oligomerization in culture, revealing that the ER environment is favorable to SOD1 aggregation (47).
SOD1 Interacts with PDI-One of the major functions of ERresident PDI is the formation, reduction, and isomerization of disulfide bonds, producing the correctly bonded native structures. The PDI family is characterized by an active site containing a pair of cysteine residues (WCGHCK) that shuttle between the oxidized disulfide and the reduced dithiol form. During disulfide bond formation, an intra-chain bond between these cysteine residues and the substrate is formed (52). The importance of the Cys 57 -Cys 146 disulfide bond in stabilizing dimeric SOD1 and preventing aggregation is increasingly recognized (10). mSOD1 proteins are more prone than wSOD1 to dissociate into monomers prior to aggregation (7)(8)(9)(10)(11)53) suggesting reduction of the SOD1 disulfide bond is important in FALS pathogenesis (49). Indeed, two recent studies have shown that SOD1 aggregates contain abnormally disulfide-bonded multimers in the spinal cord of affected SOD1 mutant mice, suggesting that a pathological hallmark of this disease is the incorrect disulfide crosslinking of the immature, misfolded mutant proteins leading to insoluble aggregates via oxidation of SOD1 cysteine residues (54,55). It is possible that PDI may play an important role in forming the correct disulfide bonds in SOD1, and the observed up-regulation and interaction with SOD1 in this study is the reason for this.
A second role of PDI is that of a molecular chaperone, assisting in polypeptide folding (52). Co-expression of other chaperones such as hsp27, hsp70, and ␣␤-crystallin has been shown to shield the cell from mSOD1-induced toxicity (56). Chaperones associate directly with mutant but not wSOD1 in tissue lysates (57). Abnormal SOD1 aggregates are present in spinal cords of human FALS patients and transgenic mSOD1 mice, especially at disease end stage (5,58). We found PDI and Erp57 to be greatly up-regulated in the lumbar spinal cords of SOD1 G93A rodents in this study, correlating with the formation of SOD1 aggregates in vivo. Erp57 is a close homologue of PDI (59) whose up-regulation may relate to SOD1 folding (60) or participation in the ER stress response (61).
Bacitracin-mediated inhibition of PDI in cell transfectants increased the frequency of mSOD1 inclusions, suggesting that PDI activity plays a role in the prevention of insoluble SOD1 aggregate formation. Although wSOD1 and D90A do not readily form inclusions in vitro (26), bacitracin treatment provoked the greatest increase in SOD1 inclusion formation in these transfectants. Conversely, bacitracin influenced the most aggregating mutants, G93A and G93C, the least. This finding implies that PDI may normally be protective toward wSOD1 and mSOD1 associated with milder ALS phenotypes (62) by modifying SOD1 aggregation, implying a neuroprotective role for PDI in ALS. PDI is additionally up-regulated in Parkinson disease (63,64), Creutzfeldt-Jakob disease (65), and Alzheimer disease (66). Moreover, PDI family members can protect neu- from SOD1 G93A mice plus wild type littermate controls were separated by SDS-PAGE and immunoblotted using anti-BiP, -PERK, -IRE1, -ATF6, -CHOP, -caspase (Casp) 12, cleaved caspase-9, and cleaved caspase-3 antibodies. Two postnatal ages were examined as follows: pre-symptomatic (p90) and disease end point (p120) animals. Immunoblots were stripped and re-probed for ␤-actin as a loading control. B and C, the total amount of each protein per lane was quantified for SOD1 G93A animals and plotted as a ratio to the corresponding wild type control at each postnatal age as follows: p90 (B) and p120 (C ). Data are presented as mean Ϯ S.E.; n ϭ 5; **, p Ͻ 0.01; ***, p Ͻ 0.001 versus wild type by one-way ANOVA with Tukey's post hoc test. rons in cerebral ischemia and prion disease (67,68), attesting to a potential neuroprotective role.
The immunoprecipitation study revealed that endogenous, wSOD1, and mSOD1 formed complexes with PDI in NSC-34 cells. In addition, PDI bound to SOD1 in spinal extracts of transgenic ALS and wild type animals. The relative ability of endogenous wSOD1 and transgenic mSOD1 or wSOD1-EGFP and mSOD1-EGFP to bind to PDI requires future study involving a detailed analysis of the PDI binding properties of each protein. The greater effect of bacitracin on wSOD1 and D90A hints at a reduced ability of mSOD1 to bind to PDI. One may envisage that PDI may normally bind to wild type enzyme in the ER to form a disulfide-linked dimer that is transported to the cytosol in the correct conformation. Misfolded aggregates, if formed, may be refolded into native conformations or eliminated. However, if mSOD1 cannot bind to PDI as efficiently, it is possible that disulfide-reduced monomers are formed that aggregate easily. This may lead to induction of the UPR and eventually neurotoxicity via apoptosis if unresolved. However, these propositions must be confirmed by further studies.
However, the association between PDI and mSOD1 suggests chaperone-like activity preventing SOD1 misfolding. Thus, upregulation of PDI may represent a cellular defense against aggregate formation. Indeed, thioredoxin, a protein that shares two active sites closely resembling PDI, is also up-regulated in human ALS spinal cords (69). This study points to a relationship between SOD1 aggregation, ER stress, and apoptosis that requires future investigations and introduces new opportunities for FALS therapy in the future.