Galectin-3 is a candidate biomarker for ALS: Discovery by a proteomics approach

Tissue and cerebrospinal fluid samples

All animal procedures were approved by Emory University's IACUC committee. Human studies were approved by the IRB committees at Emory University and Massachusetts General Hospital. Transgenic C57BL/6J mice overexpressing human wild type or G93A SOD1 were maintained as hemizygotes. Transgenic animals were identified by PCR analysis on tail snip DNA. All mice were anesthetized with intraperitoneal injection of 4% chloral hydrate and perfused systemically with 0.9% saline (37°C) to remove blood from tissues. Tissue samples (ventral root and lumbar spinal cord) were dissected from the transgenic animals and non-transgenic littermates at different ages (approximately 45, 90, and 120 days). Human spinal cord tissues were collected at autopsy from normal controls and from patients dying with neurodegenerative diseases. Spinal fluid (CSF) samples were obtained from Dr. Merit Cudkowicz (Massachusetts General Hospital, Boston). Lumbar puncture (commonly called a spinal tap) is a routine and safe procedure performed in the clinic by the neurologist. CSF samples were centrifuged to remove any cellular elements and frozen within 2 hours after collection at -80°C until use.

Protein profiling by LC-MS/MS approach

Proteins in tissues were extracted by homogenization and sonication in an ice-cold buffer (1 V tissue in 10 V buffer, 50mM HEPES, pH 7.4, 150mM NaCl, 1mM MgCl₂, 0.5mM CaCl₂, and 1mM DTT) with addition of protease inhibitors (0.1mM PMSF, 1mg/ml leupeptin, and 1mg/ml pepstatin). Protein concentration was determined by Bradford protein assay (Bio Rad) using bovine serum albumin as standard, and further verified by Coomassie staining on SDS gel ⁷. Protein lysates of ventral roots (∼60μg protein/sample) were reduced with 10mM DTT at 95°C for 5 min, alkylated with 50mM iodoacetamide at 21°C for 30 min, resolved on a 9% SDS gel and stained with Coomassie G250. Each gel lane was cut into twenty bands followed by in-gel trypsin digestion ⁸. Extracted peptides were dried by Speed Vac. The spinal cord samples from control and ALS animal model at 120 days were analyzed similarly but with fewer fractions (four per sample).

Peptide samples were analyzed by nanoscale LC-MS/MS on an LTQ mass spectrometer (Thermo Finnigan). Peptides were dissolved in buffer A (0.4% acetic acid, 0.005% heptafluorobutyric acid and 5% acetonitrile), loaded onto a 75μm i.d. × 10cm C₁₈ column (5 μm magic C18AQ; pore size, 200 Å; Michrom Bioresources, Auburn, CA)., and then eluted during an 80-min gradient of 10-30% buffer B (0.4% acetic acid, 0.005% heptafluorobutyric acid, 95% acetonitrile; flow rate, ∼250 nl/min). Eluted peptides were analyzed by MS (400–1600 m/z, normal scan rate, 1 μscan, target value of 30,000 for automatic gain control), followed by eight data-dependent MS/MS scans (isolation width of 2 m/z, 35% normalized collision energy, 1 μscan, target value of 5,000 for automatic gain control, 60 sec dynamic exclusion).

Protein identification by tandem mass spectrometry

MS/MS spectra were searched against a mouse reference database from the National Center for Biotechnology Information using the SEQUEST Sorcerer algorithm (version 2.0, SAGE-N) ⁹ as previously described ⁷. Searching parameters included mass tolerance of precursor ions (±1.5 Da) and product ion (±0.5 Da), tryptic restriction, fixed mass shift for modification of carboxyamidomethylated Cys (+57.0215 Da), dynamic mass shifts for oxidized Met (+15.9949), three maximal modification sites and three maximal missed cleavages. Only b and y ions were considered during the database match. To evaluate false discovery rate during spectrum-peptide matching, all original protein sequences were reversed to generate a decoy database that was concatenated to the original database ^{10, 11}. The false discovery rate (FDR) was estimated by the number of decoy matches (n_d) and total number of assigned matches (n_t). FDR = 2*n_d/n_t, assuming mismatches in the original database were the same as in the decoy database. To remove false positive matches, assigned peptides were grouped by charge state and then filtered by matching scores (XCorr and deltaCn) to values to reduce protein FDR to approximately 0.5%. Furthermore, we removed all proteins identified by a single peptide, which eliminated all decoy matches. If peptides were shared by multiple members of a protein family, the matched members were clustered into a single group. Based on the principle of parsimony, the group was represented by the protein with highest number of assigned peptides, and by other proteins if they were matched by unique peptide(s).

Label-free protein quantification by spectral counts and statistical analysis

To identify differences between the ALS mice and control animals we quantified the proteins in multiple samples based on spectral counts (SC). The spectral counts were first normalized to ensure that the average SC per protein was the same in all datasets ¹². G-test was used to judge statistical significance of protein abundance difference ¹³. Briefly, the G-value of each protein was calculated as shown in equation (1).

where S₁ and S₂ are the detected spectral counts of a given protein in any of two samples for comparison, respectively, and “ln” is the natural logarithm. Although theoretical distribution of the G values is complex, these values approximately fit to the χ² distribution (one degree of freedom), allowing the calculation of related p-values ¹³.

Immunoblotting and Immunocytochemistry

Standard protocols for SDS-PAGE immunoblotting and for immunocytochemistry on paraffin-embedded mouse and human tissues were described previously ^{14, 15}. Antibodies included APOE (Calbiochem # 178468), Gal3 (Santa Cruz, sc-32790 mouse monoclonal Ab for immunocytochemistry, sc-20157 rabbit polyclonal Ab for immunoblotting), SOD1 (Calbiochem), β-tubulin (Developmental Studies Hybridoma Bank), GAPDH (Abcam ab9485-100). Appropriate secondary antibodies were used, and visualization of signals was by ECL for immunoblots (Amersham), and by DAB for immunocytochemistry. Quantitation of immunoblot bands was done on digitized images using Image-J software (http://rsbweb.nih.gov/ij/).

Galectin-3 ELISA

Measurement of Gal3 in spinal fluid samples was performed based on manufacturer's protocol (R and D Systems). Briefly, a standard curve was generated using 0.15 to 10 ng/ml of pure Gal3 provided in the kit. 50μl of CSF from each patient was added to 50μl of biotin conjugate containing a proprietary ELISA buffer. After 2 hours of incubation at room temperature, a colorometric signal was generated by addition of streptavadin-HRP followed by TMB. Relative intensity of the TMB signal was measured at 450nm on a microplate reader, and the amount of Gal3 in the sample was calculated from fitting to the standard curve. After subtracting for background we found the sensitivity in CSF to be about 1 ng/ml. ELISA signals that were below background were calculated as “zero”, which translated to < 1 ng/ml. Group data were compared by non-parametric ANOVA, comparing all motor neuron diseases, other neurological diseases, and healthy controls. Post-test correction for multiple comparisons was done using Mann-Whitney analysis.

Identification of novel proteins altered in ALS mouse model

The experiments were performed in two phases (Fig 1A): (i) discovery phase for large-scale mass spectrometry (MS) analysis of ALS mice and statistical inference, and (ii) validation phase for confirming potential biomarker proteins by traditional approaches in ALS mice and human patients. As ventral roots are composed primarily of axons emanating from motor neurons, we dissected ventral roots from two diseased animals (SOD1^G93A), two non-transgenic littermates and another control mouse overexpressing wild type SOD1. The extracted proteins were first compared by SDS PAGE and silver staining. The overall pattern of major bands looked highly similar (Fig 1B), indicating the vast majority of proteins are not altered in ALS mice, whereas the difference in SOD1 level was affirmed by immunoblotting (Fig 1C). To identify differentially expressed proteins, the protein samples resolved on the SDS gel were subjected to in-gel digestion and the resulting peptides were analyzed by extensive LC-MS/MS. Tens of thousands of MS/MS spectra were acquired and analyzed by database searching and stringent data filtering. Finally we accepted 1,299 proteins that were identified by at least two peptides (supplemental Table S1), and these proteins were matched by a total of 9,197 peptides (supplemental Table S2).

To compare protein levels in ventral roots of ALS and control mice, we used a label-free quantitative method based on spectral counts ¹⁶. The spectral count of a given protein is the summed number of its matching MS/MS spectra. Since the peptide ions are chosen for MS/MS analysis according to the rank of ion intensity, the spectral count of an identified protein is a reasonable index that reflects its abundance in the samples. The label-free method requires reproducibility of the LC-MS/MS runs, and we thus tested run-to-run variation in a series of five runs (Fig 2A). Base peak profiles for the five comparative runs were almost identical, indicating of high consistency of our LC-MS/MS system.

Statistical analyses of the datasets were carried out in multiple steps. (i) The effect of spectral counts on the precision of quantitative analysis was tested by plotting the distribution of spectral counts in two identical samples. For example, the two ALS animals have no theoretical difference in protein composition, and their comparison represents a null experiment and should lead to no change in any protein (Fig 2B). When spectral counts were decreased from a few thousands to the minimal number of 1, the variations of measurement were gradually increased. To account for this effect, we used G-test to derive p values for each protein comparison (see Methods), which indicates the probability of protein change between ALS and control mice. (ii) We defined a p value cutoff to select proteins with significant changes (Table 1). Ideally, the cutoff should accept a very small number of proteins (i.e. false discoveries) in null experiments (e.g. comparison of control 1 versus control 2; or comparison of ALS 1 versus ALS 2). When the same cutoff was applied to ALS-to-control comparison, the list is deemed acceptable with a low false discovery rate. To determine the appropriate cutoff for excluding false discoveries, we dynamically adjusted the p value cutoff from 0.2 to 0.001, and identified that p value of 0.05 was a reasonable threshold with almost zero false discovery rate. (iii) Based on this cutoff, we identified a list of 38 proteins after comparing two ALS mice with two non-transgenic littermates. After further comparison of ALS mice with the third control overexpressing wild type SOD1, the list was reduced to 14 proteins (Table 2). We were able to classify the proteins that were elevated in ALS mice into several functional groups, including ER stress, secretory pathways, cellular structure, metabolism, and proteolysis, suggesting that cellular events involved in ALS pathogenesis are complex and diverse.

Validation of protein changes in ALS mice

To verify the findings from the proteomic analysis we analyzed mouse tissues by immunoblotting of selected candidates. Comparison of ventral roots from ALS mice vs. littermate and SOD1^WT controls confirmed the relative specificity of ApoE and Gal3 in SOD1^G93A (Fig 3A). Further confirmation of specificity was sought in spinal cord homogenates from these animals (Fig 3B). Interestingly, the protein that remained specific for the SOD1^G93A was Gal3, and the Gal3 level was only increased in symptomatic stage (118 days) but not in presymptomatic stage (47 days) (Fig 3c). At 118 days SOD1^WT mice showed very little expression of Gal3 in the lumbar spinal cord, suggesting that the increase in Gal3 in the SOD1^G93A was not just a phenomenon of ageing. Strikingly, immunocytochemistry using the Gal3 antibody corroborated the findings on immunoblot, showing intense staining in SOD1^G93 lumbar spinal cord, and little to no staining in controls (Fig 3D).

Analysis of Gal3 in human tissues

Spinal cord and brain tissues from patients who died with ALS (n = 9), other neurodegenerative diseases including Alzheimer's disease (n = 13) and Lewy body dementia (n = 2) and from “control” tissues without neuropathological abnormalities (n = 2) were subjected to immunoblotting and immunocytochemistry. Spinal cord homogenates from ALS patients showed significantly higher amounts (∼4 fold) of Gal3 as compared to both disease and normal controls (Fig 4). This same differential staining was seen by immunocytochemistry (Fig 5). Spinal cord and brainstem sections from ALS patients showed a relative increase in immunoreactivity with Gal3 antibodies, most consistently within the regions of the corticospinal tracts (Fig 5A-D). In sections containing both ventral and dorsal spinal roots, the ventral roots displayed higher intensity of staining as compared to the dorsal roots (Fig 5E-F). We did not see staining of neuronal cell bodies – staining was generally confined to cells with the appearance of glia. In an effort to identify the cells expressing Gal3 we stained serial sections with antibodies to Gal3, GFAP (for astrocytes), and Iba1 (for microglia/macrophages). As shown in Fig 5G-N, this analysis identified the majority of cells expressing Gal3 as microglia/macrophages.

As Gal3 is a secretory protein, we further examined Gal3 level in human spinal fluids by ELISA. The samples were collected from patients of motor neuron diseases (MND, n = 37), other neurological diseases (OND, n = 24), and normal controls (n = 21) (Fig 6). The MND group included patients characterized at the time of sampling as sporadic ALS (n = 30), lower motor neuron disease (n = 4) and upper motor neuron disease (n = 3). The OND group included patients with peripheral neuropathy (n = 9, including one patient with multifocal motor neuropathy), hydrocephalus (n = 5), stroke or hemiparesis (n = 3), dementia (n = 2), and 1 patient each with Lyme disease, Bell's palsy, multiple sclerosis, Parkinsonism, and HSV meningitis. Gal3 was below the limit of detection of 1 ng/ml in 10 samples from patients with MND (27%), 10 from OND (41%), and 7 from normal controls (33%). In those samples where Gal3 was detected, the mean level was significantly higher in MND (2.17 ± 1.5 ng/ml) as compared to both OND (1.6 ± 0.48 ng/ml) and normal controls (1.59 ± 0.34 ng/ml), suggesting that Gal3 in the CSF may be a marker for motor neuron diseases.