Changes in the Spinal Cord Proteome of an Amyotrophic Lateral Sclerosis Murine Model Determined by Differential In-gel Electrophoresis ^,

Abstract

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disorder characterized by loss of motor neurons resulting in progressive paralysis. To date, more than 140 different mutations in the gene encoding CuZn-superoxide dismutase (SOD1) have been associated with ALS. Several transgenic murine models exist in which various mutant SOD1s are expressed. We used DIGE to analyze the changes in the spinal cord proteome induced by expression of the unstable SOD1 truncation mutant G127insTGGG (G127X) in mice. Unlike mutants used in most other models, G127X lacks SOD activity and is present at low levels, thus reducing the risk of overexpression artifacts. The mice were analyzed at their peak body weights just before onset of symptoms. Variable importance plot analysis showed that 420 of 1,800 detected protein spots contributed significantly to the differences between the groups. By MALDI-TOF MS analysis, 54 differentially regulated proteins were identified. One spot was found to be a covalently linked mutant SOD1 dimer, apparently analogous to SOD1-immunoreactive bands migrating at double the molecular weight of SOD1 monomers previously detected in humans and mice carrying mutant SOD1s and in sporadic ALS cases. Analyses of affected functional pathways and the subcellular representation of alterations suggest that the toxicity exerted by mutant SODs induces oxidative stress and affects mitochondria, cellular assembly/organization, and protein degradation.

Amyotrophic lateral sclerosis (ALS)1 is a devastating neurodegenerative disease characterized by loss of motor neurons in the motor cortex, the brainstem, and the spinal cord. This results in progressive muscular atrophy, and the patients usually succumb to respiratory failure within a few years. About 10% of ALS cases are familial (1), and in some patients the disease is linked to mutations in the CuZn-superoxide dismutase (SOD1) gene (2). SOD1 is a ubiquitously expressed antioxidant enzyme. Overall about 6% of all cases with ALS show SOD1 mutations, and more than 140 such mutations have been identified (3). The mutations confer a cytotoxic gain of function of unknown character to the enzyme (4, 5). ALS caused by mutant SOD1s shows the same spectrum of disease phenotypes as is seen in sporadic cases lacking such mutations. This suggests that the pathogenesis of ALS induced by SOD1 mutations should show significant similarities with that in sporadic disease, e.g. similar pathogenic protein alterations.

ALS has been modeled in mice via transgenic overexpression of mutant SOD1s (5–8). To cause disease within the short lifespan of a mouse, the mutant SOD1s have to be expressed at rates around 25-fold higher than the rate of expression of the endogenous murine enzyme (9). Mostly structurally stable mutants have been used, resulting in up to 10-fold increases in SOD activity and 20-fold increases in SOD1 protein levels in the CNS that may cause overexpression artifacts. Overloading of mitochondria with mutant SOD1s and vacuolization have been observed in such models (10).

To explore ALS pathogenesis, we studied changes in the proteome of spinal cords of SOD1 transgenic mice using DIGE. To reduce the risk of overexpression artifacts, we used mice that express the unstable human SOD1 (hSOD1) truncation mutant G127insTGGG (G127X) (7). These mice develop an aggressive form of the disease, which is of short duration, despite the fact that the mutant lacks SOD activity and the fact that the level of G127X hSOD1 protein is less than half that of the endogenous murine SOD1. The mice were studied at their peak body weights just before development of paralytic symptoms. Here we present the identity of and discuss the possible significance of 53 proteins found to be differentially regulated in ALS transgenic mice.

EXPERIMENTAL PROCEDURES

Transgenic Mice—

The G127X hSOD1 mice (7) were developed in our laboratory and backcrossed with C57BL/6JBomTac mice for >25 generations. Mice expressing G85R mutant hSOD1 were obtained from Dr. Don Cleveland and were similarly backcrossed with C57BL/6JBomTac mice for >15 generations (6). The G127X mice attained their peak body weights at 195 ± 14 days of age, and onset of symptoms, as observed by changes in grip strength and stride pattern, occurred at day 208 ± 15. Mice were regarded as terminally ill (216 ± 15 days) when they were no longer able to reach the food. Five mice, 196 days old and without any symptoms, were chosen for the study. As controls, five non-transgenic mice of the C57BL/6JBomTac strain used for the backcrossing were used (Taconic Europe, Bomholt, Denmark). Mice were killed by intraperitoneal injections of pentobarbital. The thorax was cut open, and the animal was perfused through the left ventricle with 20 ml of 0.15 m NaCl. The spinal cord was dissected by cutting the vertebral column at the base of the scull and just above the hip bone. A syringe was inserted at the lower opening, and the spinal cord was flushed out with 0.15 m NaCl, washed, frozen in liquid nitrogen, and stored at −80 °C. For validation by Western immunoblotting, spinal cords from four G127X mice (age, 184 days), four C57BL/6JBomTac control mice (age, 199 days), and four G85R mice (at their peak body weights; age, 325 days) were collected.

Homogenization of Tissue—

For 2-D gels, the spinal cords were supplied with 10 volumes of lysis buffer (1 m Tris-HCl, pH 8.5, 2 m thiourea, 7 m urea, 4% CHAPS) and homogenized using an Ultraturrax (IKA, Stufen, Germany) followed by sonication using a Sonifier Cell Disruptor (Branson, Danbury, CT) for 1 min. Tissues for Western blot were supplied with 25 volumes of PBS (10 mm potassium phosphate, pH 7.0, 0.15 m NaCl containing EDTA-free Complete antiproteolytic mixture (Roche Diagnostics)) and homogenized as above.

Labeling Proteins for DIGE—

Flourescent DIGE labeling of proteins was performed as described in the manufacturer's manual (GE Healthcare). For further information, see the supplemental information.

Analytical Two-dimensional Electrophoresis—

Preparation of 2-D gels for DIGE was carried out as described in the manufacturer's manual (GE Healthcare). Samples were run on 24-cm pI 3–11 NL and pI 4–7 strips (GE Healthcare). For further information, see the supplemental information.

Gel Analysis—

CyDye-labeled DIGE gels were scanned without removing the glass plates in a Typhoon 9410 imager (GE Healthcare) using excitation and emission filters specified by GE Healthcare (Cy2, Cy3, and Cy5, excitation 480, 540, and 620 nm and emission 530, 590, and 680 nm, respectively). Individual gels were processed further using DeCyder 6.5 software, and spot detection was done using the differential in-gel analysis module. Data from this application were then loaded into the biological variation analysis software for spot matching between gels and variation analysis. Statistical significance was calculated using the Student's t test, and spots that were present in all gels, with a difference between the groups of at least 25% and with a p value of <0.05, were considered for further analysis.

Protein Identification—

Quantitative 24-cm 2-D gels were run using 0.5–1 mg of total protein. For Coomassie staining, gels were fixed in 10% methanol, 7% acetic acid for 1 h. Staining was performed overnight with G-250 staining solution (0.8% phosphoric acid, 8% ammonium sulfate, 20% methanol, 0.08% Coomassie Brilliant Blue G-250), briefly washed with 25% methanol, and stored in distilled water. Selected spots were picked manually and put in a tube with distilled water. Coomassie-stained spots were dehydrated twice with a solution of 35% (v/v) acetonitrile in 20 mm ammonium hydrogen carbonate and dried in a SpeedVac (Savant; GMI, Inc., Ramsey, MN) and finally incubated with trypsin (V511a; dilution, 10 μg/ml; Promega, Madison, WI) overnight at 37 °C. One microliter of protein digest was applied to a MALDI-TOF target plate (Applied Biosystems) and dried. Then 0.5 μl of a 1:1 mixture of α-cyana-4-hydroxycinnamic acid solution (Agilent Technologies, Santa Clara, CA) and 1% trifluoroacetic acid, 50% acetonitrile was added on top. One microliter of calibrant (a 1:1 mixture of 50% acetonitrile, 1% trifluoroacetic acid and α-cyana-4-hydroxycinnamic acid solution mixed with Sequazyme calibration mixtures 1 and 2 (P2-3143-00, Applied Biosystems)) was spotted onto the plate next to every protein digest and used for external calibration. When completely dry, the plate was put in a Voyager DE-STR mass spectrometer (Applied Biosystems), and the peptide spectra were recorded. Peptide peak lists were generated using Data Explorer software version 4 (Applied Biosystems) with external calibration, advanced base-line correction, noise filter set to 0.7, and deisotoping. A cutoff was generally set to 1–5% of base peak intensity. Advanced peak detection was used for the peptide spectra of some weak protein spots, with few peptides and high background noise, using a higher cutoff in the low molecular weight area for efficient noise reduction. Proteins were identified by searching the Swiss-Prot (version 50.9) and International Protein Index (version mouse_V3_21_20060915) databases using an in-house Mascot server that was licensed to Umeå University by Matrix Science (Boston MA). Following glycine 127, the G127X protein contains five novel amino acids before the carboxyl-terminal truncation. To get a Mowse score for the G127X dimer, this sequence was added to the Swiss-Prot database searched. Search parameters were set to allow mass deviations of ±50 ppm and one or two missed cleavage sites as well as fixed modifications such as carbamidomethylation of cysteine and oxidation of methionine. When human keratin contaminants were present, the “Re-search unmatched” function was used. Identifications that were statistically significant (with a Mowse score in the range of at least 60–66 and a p value <0.05) and that showed a gel position close to the theoretical mass and isoelectric point were regarded as positive.

Solubilization and Reduction of the 33-kDa SOD1 Species—

Formic acid efficiently solubilizes aggregated proteins (11). Homogenates were prepared as for Western blots, and an aliquot was dried in a SpeedVac and then incubated in 100% formic acid for 15 min. The acid was evaporated under a stream of argon, and the dry material was dissolved in sample buffer (100 mm Tris-HCl, pH 6.8, 20% glycerol, 4% sodium dodecyl sulfate, reductant) for immunoblotting. Because 2-mercaptoethanol has been shown to occasionally induce aggregation (12), both 10% 2-mercaptoethanol and 100 mm dithiothreitol were tested as reductants prior to immunoblotting (13).

Immunoblotting—

One-dimensional immunoblotting and quantitation of proteins were performed as described previously (10). All immunoblots were run at least in triplicate. Two-dimensional immunoblotting was performed with 11-cm pI 3–11 NL strips (GE Healthcare) run on regular SDS-PAGE (Criterion IPG+1 gels (Bio-Rad)). For further information, see the supplemental information.

Immunohistochemistry—

Brain and spinal cords from 200-day-old G127X and G85R transgenic mice and C57BL/6JBomTac control mice were fixed with 4% paraformaldehyde in phosphate-buffered solution (10 mm potassium phosphate, pH 7.4, 0.15 m NaCl) and embedded in paraffin. Four-micrometer-thick transverse paraffin sections were prepared for immunohistochemistry, which was carried out using the same set of antibodies as for immunoblotting. For enhancement, microwave heating for 20 min in citrate buffer (pH 6.0) was performed, and the sections were blocked for 30 min in goat serum prior to staining procedures. Single immunohistochemistry was performed on the sections using the Ventana 3-amino-9-ethylcarbazole (Ventana Medical Systems, Tucson, AZ) and alkaline phosphatase red immunohistochemistry system according to the protocol of the manufacturer.

Stereology—

Stereological quantification of ventral horns in murine spinal cord was performed as described previously (8). For further information, see the supplemental information.

Network Analysis—

Network analysis was performed using the Ingenuity Pathways Knowledge Base. Protein identifiers of the altered proteins were uploaded into the Ingenuity Pathways Knowledge Base to reveal networks of interacting proteins. Function and canonical pathways analysis revealed the possible impact of the proteins identified on cellular functions and pathways. See Ref. 14 and the supplemental information for further information.

Statistical Analysis—

Orthogonal projections to latent structures-discriminant analysis (OPLS-DA) (15) was used for multivariate statistical evaluation in the SIMCA-P+ software version 11.5 (Umetrics AB, Umeå, Sweden). OPLS-DA is a supervised multivariate classification method used to extract the systematic differences between known classes of samples in predictive variation (systematic variation correlated to the class separation) and orthogonal variation (systematic variation unrelated to the class separation). Cy3 and Cy5 values were corrected by dividing by the corresponding Cy2 values. The data were mean-centered and Pareto scaled before being subjected to OPLS-DA. The number of significant OPLS-DA components was decided by full cross-validation (16). Protein spots that differed significantly between wild-type and G127X SOD1 mice were detected through a simultaneous inspection of the OPLS-DA covariance and correlation loadings (w*p and p(corr)p, respectively). In addition, an iterative threshold approach, a modified version of the approach presented by Karp et al. (17), was applied to identify the protein spots that contributed most to the calculated OPLS-DA model (VIP analysis). This was done by successively removing the 20 most significant protein spots, based on p(corr)p, and recalculating the model until the cross-validated score vector (t_cv) revealed the first overlap between the two classes. For this separation, the cross-validated predictive ability (VIP-Q2) and the p value were calculated.

RESULTS

The DIGE analysis of the gels covering pI 3–11 found about 1,800 potential protein spots that could be matched between the different gels and that could be subjected to statistical analysis. When analyzed by OPLS-DA, it was found by VIP-Q2 scoring that 420 of these spots significantly contributed to the separation between the groups (Fig. 1). The criteria we used regarding choice for identification (differences greater than 25% with p < 0.05 by Student's t test) yielded 178 apparent protein spots. Because of streaking, multiple protein isoforms (trains of spots), differences in staining between DIGE and Coomassie, and other uncertainties, 111 spots could be reliably picked. The omissions occurred mainly at molecular masses above 80 kDa. Finally the MALDI-TOF analysis resulted in 65 identified proteins, most located in the 15–80-kDa range. Most failures in identification occurred in low abundance proteins below 15 kDa because of the presence of few significant peptides and difficulties in achieving high coverage.

Fig. 1.

VIP analysis. The cross-validated predictive ability (Q2) versus the number of excluded variables (protein spots) for successive OPLS-DA models removing 20 spots at the time based on p(corr)p is shown. The first overlap between the two sample classes in the cross-validated OPLS-DA scores was detected after exclusion of 420 variables. For this separation, Q2 = 0.555 with a p value (p(t_cv)) of 0.0009.

A problem in analysis of DIGE data sets is that the number of variables/protein spots is much greater than the number of observations and that the variables often are correlated and contain noise. Multivariate methods are well suited for extraction of systematic information from such materials, and they can robustly and reliably detect differences between groups. Combined with full cross-validation, the number of false positives can be minimized. In our final analysis of alterations caused by expression of the G127X mutant SOD1, we only considered proteins that showed good correlations (p(corr)p > 0.70) by the multivariate OPLS-DA and that differed from controls with significances better than p < 0.01 using the parametric t test. This gave us a final list of 53 proteins (Fig. 2 and Table I). Eleven proteins with borderline significance levels of between 0.01 and 0.05 and one spot that was identified as a mixture of two proteins are listed in supplemental Table 1.

Fig. 2.

DIGE gel. Proteins were picked from 24-cm pI 3–11 NL gels and identified with MALDI-TOF MS. Spots with significant p values (<0.01) from DeCyder software as well as OPLS-DA (p(corr)p > 0.7) and identified with significant MASCOT scores are presented here. Proteins are indicated by their Swiss-Prot accession numbers and numbered as displayed in Table I. Spot 19 was not adequately resolved in pI 3–11 NL gels but found to be differentially regulated on the pI 4–7 gels.

Table I.

Proteins significantly down- or up-regulated

Gel IDs are numbers from Fig. 1, and accession numbers are from Swiss-Prot. Abbreviated gene names are as indicated from the Ingenuity Pathways Knowledge Base (14). -Fold change refers to ratios as identified by DeCyder software. Further data on identifications and statistics are presented in supplemental Table 2.

Gel ID	Abbreviation/gene	Protein accession no.	Name of protein	-Fold change
Down-regulated proteins
1	MBP	P04370	Myelin basic protein	−3.10
2	ETFA	Q3THD7	Electron transfer flavoprotein subunit α	−2.12
3	MTAP	Q9CQ65	S-Methyl-5-thioadenosine phosphorylase	−2.03
4	DSTN	Q9R0P5	Destrin (actin-depolymerizing factor)	−1.97
5	INSL3	Q5RL10	Insulin-like 3	−1.92
6	ABAT	P61922	4-Aminobutyrate aminotransferase	−1.76
7	FABP5	Q05816	Fatty acid-binding protein, epidermal	−1.65
8	ALDH5A1	Q8BWF0	Succinate-semialdehyde dehydrogenase	−1.56
9	VDAC1	Q60932	Voltage-dependent anion-selective channel protein 1	−1.50
10	FIS1	Q9CQ92	Mitochondrial fission 1 protein	−1.50
11	ESD	Q9R0P3	S-Formylglutathione hydrolase	−1.48
12	ARPC4	P59999	Actin-related protein 2/3 complex subunit 4	−1.48
13	HINT3	Q9CPS6	Histidine triad protein 4	−1.47
14	FABP5	Q05816	Fatty acid-binding protein, epidermal	−1.46
15	DPYSL2	O08553	Dihydropyrimidinase-related protein 2	−1.44
16	HAGH	Q99KB8	Hydroxyacylglutathione hydrolase	−1.42
17	DBI	P31786	Acyl-CoA-binding protein	−1.39
18	CNP	P16330	2′,3′-Cyclic-nucleotide 3′-phosphodiesterase	−1.39
19	CPLX1	P63040	Complexin	−1.38
20	TST	P52196	Thiosulfate sulfurtransferase	−1.32
21	NUDT2	P56380	Nucleoside diphosphate-linked moiety X type motif 2	−1.31
22	ANXA5	P48036	Annexin A5	−1.30
23	RPS27A	P62991	Ubiquitin	−1.30
24	APOA1BP	Q8K4Z3	ApoA-I-binding protein	−1.29
25	GRB2	Q60631	Growth factor receptor-bound protein 2	−1.25
Up-regulated proteins
26	IDE	Q9JHR7	Insulin-degrading enzyme	1.31
27	DNM1	Q05193	Dynamin	1.31
28	PGAM1	Q9DBJ1	Phosphoglycerate mutase	1.32
29	ACTG1	Q3TSB7	Actin-γ	1.34
30	SUCLA2	Q9Z2I9	Succinyl-CoA ligase	1.34
31	STIP1	Q60864	STI (stress-induced) phosphoprotein 1	1.35
32	CYC1	Q9D0M3	Cytochrome c1, heme protein, mitochondrial precursor	1.35
33	SOD2	P09671	Mn-SOD	1.36
34	C3ORF10	Q91VR8	Brick 1	1.37
35	CAPNS1	O88456	Calpain smaller subunit	1.40
36	MLF2	Q99KX1	Myeloid leukemia factor 2	1.40
37	AKR1A1	Q9JII6	Alcohol dehydrogenase (NADP+)	1.42
38	MDH1	P14152	Malate dehydrogenase	1.43
39	COPS4	O88544	COP9 (signalosome complex) subunit 4	1.43
40	CKB	Q04447	Creatine kinase B-type	1.44
41	YWHAG	P61982	14-3-3 protein γ	1.44
42	BLVRA	Q9CY64	Biliverdin reductase A precursor	1.47
43	COPS8	Q8VBV7	COP9 (signalosome complex) subunit 8	1.47
44	TPRG1L	Q9DBS2	Tumor protein p63-regulated gene 1-like protein	1.49
45	MDH1	P14152	Malate dehydrogenase	1.52
46	IDH1	O88844	Isocitrate dehydrogenase (NADP) cytoplasmic	1.52
47	ERP29	P57759	Endoplasmic reticulum protein 29	1.53
48	ACTG1	Q3TSB7	Actin-γ	1.53
49	ACTG1	Q3TSB7	Actin-γ	1.56
50	CKB	Q04447	Creatine kinase B-type	1.62
51	COPS8	Q8VBV7	COP9 (signalosome complex) subunit 8	1.71
52	UCHL1	Q9R0P9	Ubiquitin carboxyl-terminal hydrolase isozyme L1	1.75
53	UBE2L3	P68037	Ubiquitin-conjugating enzyme E2 L3	1.81

As a partial technical replicate and to achieve a higher resolution, the transgene/control set was also subjected to separation in pI 4–7 gels and DIGE analysis. In the optimally resolved area, 25 of the proteins in the pI 3–11 gels that fulfilled all criteria for inclusion were also found to be differentially regulated (p < 0.01) in the pI 4–7 gels. One such protein was not replicated in the pI 4–7 gels and was omitted from Table I. Another protein, complexin, which was poorly resolved in the pI 3–11 gels, fulfilled the significance criterion (p < 0.01) and was added to Table I.

Compartments, Networks, and Functions—

The proteins were divided into groups defined by their main tissue compartments (Fig. 3) based on information from European Molecular Biology Laboratory-European Bioinformatics Institute or Swiss-Prot. Altered proteins were mainly from the cytoplasm or mitochondria. In a recent comprehensive study, Pollak et al. (18) tried to identify as many proteins as possible in another part of the murine central nervous system (the hippocampus of the brain) by the same techniques as those used here, i.e. 2-D gels and MALDI-TOF. The compartment distributions of the 469 different proteins they identified were compared with the present findings (Fig. 3). The most distinct difference is the overrepresentation of altered mitochondrial proteins in the ALS model mice.

Fig. 3.

Compartments. The altered proteins were categorized according to subcellular compartment. Information was collected from the European Molecular Biology Laboratory-European Bioinformatics Institute and Swiss-Prot databases. Relative numbers (%) are presented, and these are compared with the results of a previous study (data within parentheses) of the proteome of murine CNS (18).

The proteins identified were further evaluated by Ingenuity Pathways Knowledge Base analysis (14). This tool builds protein networks based on known direct or indirect interactions described in the literature and defines common functional and canonical pathways. Of the 53 proteins, 44 could be used for network analysis, and 42 could be used for functional/canonical pathway analysis. The discrepancy was due to protein triplicates or duplicates (e.g. actin-γ) and to a few proteins having very limited information in the literature. Four protein networks were identified with high significance and contained 41 of the 44 proteins included (Fig. 4). The networks generated were composed of the proteins identified and their interaction partners known from the literature. Network A contained 13 identified proteins and their interaction partners and involved functions such as cellular assembly/organization (filament dynamics) and the cell cycle. Networks B and C contained 12 and 11 identified proteins, respectively, and involved functions in nervous system development/function and cellular assembly/organization. Network D contained five identified proteins, and the main functions involved the cell cycle.

Fig. 4.

Networks of interacting proteins. Networks of interacting proteins were generated through Ingenuity Pathways Analysis. Proteins identified (displayed as abbreviated gene names) were connected with interaction partners known from the literature and information stored in the Ingenuity Pathways Knowledge Base. Red means up-regulated, and green means down-regulated. Solid lines represent direct interaction, and dashed lines represent indirect interaction. A, network A contained 13 identified proteins and involved top functions in cellular assembly/organization, the cell cycle, and cell death. B, network B contained 12 identified proteins and top functions in nervous system development/function and inflammatory disease. C, network C contained 11 identified proteins and involved top functions in cellular assembly/organization, cellular compromise, and cell morphology. D, network D contained five proteins and top functions in the cell cycle, cancer, and cellular development.

The material was then subjected to functional analysis (Table II). In this analysis, the p value is calculated by comparing the number of identified proteins that participate in a given function or pathway relative to the total number of occurrences of these proteins in all functional pathway annotations stored in the Ingenuity Pathways Knowledge Base. The four top functional pathways involved fundamental mechanisms of cell survival but also pointed at proteins known earlier to take part in neurological disease (nucleic acid metabolism, neurological disease, cellular assembly/organization, and cellular function/maintenance). Ten canonical pathways reached significance levels of <0.01 (Table III). Pyruvate metabolism was the top pathway (p < 0.001) represented by four identified proteins followed by the citrate cycle (three proteins, all up-regulated), and the NRF-2-mediated oxidative stress response (five proteins, all up-regulated).

Table II.

Functional analysis

Proteins identified were uploaded to the Ingenuity Pathways Knowledge Base and subjected to functional analysis. This identifies functions and diseases in which the proteins have previously been found to be altered. Abbreviations are as in Table I. Up-regulated proteins are shown in bold.

Global function	p value	Proteins
Nuclei acid metabolism	0.0003	DPYSL2, MTAP, NUDT2, VDAC1, IDH1
Neurological disease	0.0005	MBP, CNP, DPYSL2, VDAC1, SOD2, IDE, UCHL1, MDH1
Cellular assembly and organization	0.0006	CNP, DPYSL2, DSTN, MBP, ANXA5, TST, CPLX1, FIS1, GRB2, CAPNS1, SOD2, UCHL1, DNM1
Cellular function and maintenance	0.001	VDAC1, SOD2

Table III.

Canonical pathways

Proteins identified were uploaded to the Ingenuity Pathways Knowledge Base and subjected to canonical pathway analysis. This identifies canonical pathways that are overrepresented among the identified proteins in a way that is not likely to be by chance. Abbreviations are as in Table I. Up-regulated proteins are shown in bold. GABA, γ-aminobutyric acid.

Canonical pathways	p value	Proteins
Pyruvate metabolism	0.0001	HAGH, ALDH5A1, MDH1, AKR1A1
Citrate cycle	0.0002	MDH1, SUCLA2, IDH1
NRF-2-mediated oxidative stress response	0.0004	AKR1A1, SOD2, ERP29, STIP1, ACTG1
GABA receptor signaling	0.0006	ABAT, ALDH5A1, DNM1
β-Alanine metabolism	0.0009	ABAT, DPYSL2, ALDH5A1
Propanoate metabolism	0.001	ABAT, ALDH5A1, SUCLA2
Glycolysis/gluconeogenesis	0.005	ALDH5A1, AKR1A1, PGAM1
Integrin signaling	0.005	GRB2, ARPC4, CAPNS1, ACTG1
Actin cytoskeleton signaling	0.006	GRB2, ARPC4, C3ORF10, ACTG1
Glutamate metabolism	0.008	ABAT, ALDH5A1

G127X SOD1 Dimer—

A spot with a molecular mass of 17 kDa, representing G127X SOD1 monomer, was found by the DIGE analysis and was not considered further (Fig. 2). In the pI 4–7 2-D gels, another spot was found at ∼32 kDa, pI ∼5.5 that was only present in the transgenic samples (Fig. 5A). When analyzed by MALDI-TOF, all the major peaks were found to represent peptides from G127X hSOD1 with 90% sequence coverage (Fig. 5B). If these were removed from the peptide spectrum, the remaining peaks yielded only an insignificant hit for human keratin. With one-dimensional and 2-D Western blotting for G127X hSOD1, in addition to the monomers, a major band/spot was found at about 32 kDa, i.e. close to the expected molecular mass of a dimer (Fig. 5C). A similar band is commonly seen in spinal cord extracts from humans (7) and transgenic mice expressing various mutant hSOD1s (9, 19), and it has recently been observed also in sporadic ALS patients (20). The present analysis shows conclusively that the 32-kDa band is an hSOD1 dimer at least in the case of G127X mice. To further study the nature of this dimerization, the protein samples were subjected to solubilization with SDS or formic acid to disrupt any non-covalent binding (Fig. 5D). This treatment had no effect on the ratio of monomer to dimer. To determine whether the dimer is attached by way of inefficiently reduced disulfide bonds, homogenates were treated with two different reducing agents (Fig. 5E), both of which failed to change the quantity of dimer, showing that the SOD1 dimer is linked covalently but by means other than disulfide bonding.

Fig. 5.

Identification of a G127X SOD1 dimer. DIGE analysis on high resolution 24-cm pI 4–7 gels revealed a spot that was only present in the G127X mice at an approximate molecular mass of 32 kDa. This spot was identified on MALDI-TOF as G127X protein with 90% sequence coverage. No other significant proteins could be detected. A, DIGE gels from control and G127X mice. The arrow indicates the specific spot. B, MALDI-TOF spectra with G127X SOD1 peptides indicated by their masses. * denotes a porcine trypsin autodigest product. When G127X peptides were subtracted from the spectra, human keratin was the top hit (not significant) represented by the background low abundance peptides. C, one-dimensional and two-dimensional human SOD1 immunoblots of G127X murine homogenates. The arrow represents the isoform identified with MALDI-TOF. D, to investigate a possible non-covalent interaction between SOD1 monomers, homogenates were solubilized with SDS or formic acid (FA). E, 2-mercaptoethanol (ME) or DTT was used to explore the risk of inefficient reduction of disulfide bonds. This did not eliminate the dimer in G127X or G85R mice.

Transglutaminases cross-link proteins via isopeptide linkages, and such cross-linking has been implicated in the pathogenesis of several neurodegenerative diseases (21). 2-D immunoblots from G127X and control mice were carried out using an anti-isopeptide antibody. More spots were indeed found in the blots from the ALS model mice, but none of them coincided with the position of the mutant SOD1 dimer (supplemental Fig. 1).

When quantified from Western immunoblots, the dimer appears to represent about 40–50% of the total G127X hSOD both in transgenic mice (Fig. 5C) and in human cases (7). However, on CyDye- and Coomassie-stained 2-D gels the dimer appears to represent about 5–10% of the total amount of G127X hSOD1. This suggests that the dimer has a very high immunoreactivity and that it becomes overestimated relative to the monomer in immunoblots. Difficulties in finding the dimer in protein-stained 2-D gels may explain why its composition has not been determined previously.

Validations—

To examine the validity of the differences found by the DIGE procedure and the biological variation analysis software, the G127X and control mice were randomized into new groups (e.g. 2 + 3 from each group) that were still mixed regarding Cy3 and Cy5 (which is a confounder if not mixed). Generally fewer than 10 proteins were found to be significantly different between the groups using the initial criteria for selection (p < 0.05 and ±25% regulation). If p was set to <0.01 in practice no significant spots remained in the randomized setup.

Performing a response permutation test in the SIMCA-P+ software (VALIDAT) showed that the Q2 values for OPLS-DA models calculated against permuted response vectors (class identity) were significantly decreasing with declining correlation to the original response vector. Twenty permutations gave an intercept for the extrapolation of Q2 of −0.296, implying that a permuted response with zero correlation to the original response would not yield a significant model. Of the 20 permutations none gave a Q2 higher or even close to the Q2 for the original response.

As to the differences in amounts of protein, seven of the proteins found to be altered were analyzed by Western immunoblots (Fig. 6A and supplemental Fig. 2). Spinal cords from independent sets of four G127X and C57BL/6JBomTac controls were used as were tissues from transgenic mice representing another ALS model, G85R hSOD1, at peak weight. Of the four up-regulated proteins, isocitrate dehydrogenase and HSC70 (with borderline significance in DIGE analysis; see supplemental Table 1) were found to be increased in both ALS models. From this additional analysis, it can be concluded that HSC70 is significantly up-regulated in transgenic ALS models. The increases found on Western blot for Mn-SOD and COP9s8 were not statistically significant. Of the three down-regulated proteins, FABP was significantly reduced in the G127X mice but not in those expressing G85R hSOD1. FIS1 was significantly reduced in the G127X mice but could not be compared in the G85R ALS model because of a different immunoblot pattern with double bands (supplemental Fig. 2). No changes could be found for CNP1 in the validation sets (Fig. 6B and supplemental Fig. 2). We investigated this result further by immunoblotting 2-D gels for CNP1 (Fig. 6B). CNP1 was found to exist in multiple isoforms, most of them with different isoelectric points. The species detected by the DIGE analysis in controls appears to be in much lower amounts in the G127X sample, which in turn contains a species with a lower isoelectric point (arrow). The latter was not detected by the DIGE analysis possibly because of shielding by other proteins stained by the CyDyes. The other differences between DIGE analysis and the one-dimensional Western immunoblots may have similar explanations, i.e. differences in isoforms with distinct isoelectric points that are only resolved in the 2-D analysis.

Fig. 6.

Validation. A, to validate quantitative data from the DIGE experiment, Western immunoblotting was performed in four new spinal cords each from control, G127X, and G85R mice. A series of proteins was tested, and gels were run at least in triplicate. Down-regulation of FABP and FIS1 was statistically significant in G127X, and up-regulation of isocitrate dehydrogenase (ICD) was statistically significant in G85R mice. HSC70 was significantly up-regulated in both strains. Western blots to detect COP9s8 and Mn-SOD did not reveal any statistically significant changes, but the trend was in the same direction as in the DIGE experiment. DIGE ratios are indicated by a dashed line (in the case of FABP and COP9s8 as means of the two spots) (see Table I). The error bars indicate standard deviations. B, Western blots for detection of CNP1 did not reveal any sign of down-regulation of protein levels. We therefore performed two-dimensional immunoblotting on 11-cm 3–11 NL strips for isoform resolution. As compared with material from control mice, we could see a change in isoform pattern with one spot shifted toward a lower isoelectric point (arrows). C, immunohistochemical staining of mouse ventral horn sections from control and G127X SOD1 transgenic mice with antibodies to HSC70, FIS1, COP9s8, and Mn-SOD.

To investigate the cellular specificity of these proteins, sections from brain and cervical, thoracic, and lumbar spinal cord were examined in G127X, G85R, and control mice using the above mentioned antibodies. Two animals were studied for each group. The FABP and CNP1 antibodies failed to give a reliable staining. Staining with the Mn-SOD antibody showed small dotlike specific staining in the cytoplasm of spinal motor neurons in all mice investigated (Fig. 6C). Mn-SOD was also seen in neurons of the dorsal horn, a few reactive astrocytes, and the neuropil. Staining with the Cop9s8, FIS1, and HSC70 antibodies showed diffuse immunoreactivity in the cytosol of ventral horn motor neurons as well as in dorsal horn neurons and in a few reactive astrocytes (Fig. 6C). No staining was seen in the majority of astrocytes or the neuropil, indicating that the staining principally was neuron-specific. If this pattern can be extrapolated to the rest of the proteins, it may indicate that the majority of protein alterations in ALS take place in neurons.

DISCUSSION

In this study we analyzed ALS model mice at their peak body weight, a time point at which there is significant pathology in the spinal cord but still limited deterioration of the global status of the animals. In the murine transgenic ALS models, the motor neurons in the ventral horns of the spinal cords are primarily affected. Because murine spinal cords are small (weighing about 70 mg), whole spinal cords had to be used. By stereology, we found that the ventral horns account for 20% of the volume of the spinal cords. Motor neuron somas and neurites have been estimated to occupy 20% of the volume of the ventral horns (22), and they thus account for about 4% of the whole spinal cord. Changes in a specific cell type (e.g. motor neurons) in proteins that are expressed in multiple cell types therefore run the risk of being diluted in the present study, whereas changes in more motor neuron-specific proteins will be less affected. There is evidence, however, that the demise of motor neurons is caused by toxic effects of mutant SOD1s in multiple cell types (for a review, see Boillée et al. (23)), which is why many protein changes can be expected to be widespread among the cells in the spinal cord.

Multiple comparisons of the present magnitude are likely to give many “false positives.” This led us to use a combination of parametric (p < 0.01) and multivariate analysis (OPLS-DA p(corr)p > 0.7) criteria to reduce this risk. A subset of 53 (Table I) of all the proteins contributing to the differences between ALS/model mice and controls (Fig. 2) was finally selected for evaluation. Most proteins accepted for network analysis (44 of 53) could be divided into four networks of interacting proteins (Fig. 4). This indicates that the proteins identified were not altered in a random way but rather showed a high degree of interaction, which points to functional relationships. Functional analysis revealed changes in pathways for basal cellular functions such as nucleic acid metabolism and cellular organization. One can hypothesize that the mechanisms behind most of these reflect compensatory actions taken from cells that are affected by the disease. The alterations in eight proteins were classified as common with other neurological diseases.

The NRF-2-mediated oxidant stress response was found to be one of the top canonical pathways differentially regulated. All identified proteins were up-regulated. The presence of the covalently linked SOD1 dimers also supports the occurrence of oxidant stress because protein-protein cross-linkages are often generated by oxidants (24). The findings are in accordance with evidence of increased oxidative stress reported previously in the ALS context (25, 26). Furthermore in mice expressing mutant SOD1 and a reporter for NRF-2 (27), activation was found in several cell types in the spinal cord during the course of the disease. However, it should be noted that in cultured cells as well as in motor neurons isolated from rats expressing the stable high level G93A mutant SOD1 (28, 29) NRF-2 and its downstream targets have been reported to be down-regulated.

Proper degradation of protein is essential for cells to function, particularly neurons as they lack the ability to regenerate. The ubiquitination system is one of the major degradation pathways, and here it was found to be differentially regulated. Ubiquitin is reduced, and ubiquitin-conjugating enzyme E2 L3 and ubiquitin carboxyl-terminal hydrolase isozyme L1 (UCH-L1) are up-regulated in G127X mice (Table I, up-regulated proteins). Oxidation and altered expression of UCH-L1 have been found previously in brain tissue in Alzheimer and Parkinson disease (30). Furthermore intense immunostaining for UCH-L1 was seen in motor neuron somata from both a patient and transgenic mice carrying the G127X SOD1 mutation (7). The COP9 signalosome complex, which is preferentially seen in neurons in the CNS (31), is multifunctional and involved in protein degradation and apoptosis for example (for a review, see Wei et al. (32)). An important function is to inhibit ubiquitination through deneddylation of the large cullin-RING family of ubiquitin E3 ligases. The up-regulation of subunits 4 and 8 suggests that inhibition of ubiquitination of many substrates and hence interference with proper protein degradation might occur. Recently knock-out of components involved in the autophagic pathway for protein degradation has been shown to cause neurodegeneration in mice (33). This points to the neuronal dependence on proper protein degradation and the potentially deleterious consequences of such disturbances.

A few proteomics studies of spinal cord from the highly expressing G93A ALS transgenic mice have been published (34, 35). Lukas et al. (34) performed a study including non-transgenic mice and transgenic mice expressing wild-type human SOD1 (wthSOD1) mice as controls. Using fractionation of spinal cord homogenates and LC-MS/MS, they compared the proteome from different fractions on the basis of functional categories. Many of these categories overlap with ours (e.g. mitochondria and oxidative stress) despite the fact that very different techniques and mouse models were used. In a study comparing G93A mice with wthSOD1-expressing mice using Coomassie-stained 2-D gels and MALDI-TOF, Massignan et al. (35) reported a mutant SOD1-specific change in 15 proteins in presymptomatic mice. As in our study, alterations in mitochondrial proteins were found, but none of the specific protein changes are common. This may be due in part to the use of wthSOD1 mice as controls as they have been shown to show pathology, although delayed, similar to that of the ALS-inducing mutants (8, 36).

Several previous studies of ALS have implicated oxidative stress and mitochondrial impairment in mutant SOD1-induced injury. Such changes have also been found in ALS cases lacking SOD1 mutations. In these studies, mostly wild type-like mutants such as G93A have been used. Such SOD1s have been found to carry out increased catalysis of strong oxidant formation (37, 38) and to artificially overload into mitochondria (10). The truncation mutant G127X is present in minute quantities, lacks SOD activity, and is unlikely to bind potentially oxidant-producing copper ions in vivo (7, 9). Furthermore there is no evidence of association between G127X protein and mitochondria in mice (10). The fact that we found signs of both oxidant stress and mitochondrial disturbances in this study suggests that such effects are not necessarily directly caused by mutant SOD1s but rather could be secondary to attacks on other primary targets. Recent studies in cells lacking Apaf1 are compatible with the occurrence of mutant SOD1-induced feedback loops, which originate in the cytosol and attack mitochondria (39). In conclusion, our findings suggest that interventions with antioxidants, agents that support mitochondria, and agents that preserve or enhance degradation of misfolded proteins might be interesting to test in transgenic ALS models even if such interventions do not affect the primary cytotoxic mechanisms of mutant SOD1s.