Abstract
Objective. To identify proteomic biomarkers in cerebrospinal fluid (CSF) and develop a diagnostic proteomic model for neuropsychiatric systemic lupus erythematosus (NPSLE).
Methods. CSF proteomic spectra were generated by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS) combined with weak cation exchange (WCX) magnetic beads. The spectra were taken from 27 patients with NPSLE before and after treatment, and 27 controls including 17 patients with scoliosis and 10 patients with SLE but without neuropsychiatric manifestation. Discriminating peaks were processed by Biomarker Patterns Software to build a decision tree model for NPSLE classification. In addition, CSF samples of 12 patients with NPSLE, 12 patients with lumbar disc herniation, and 9 patients with other neurological conditions were used as a blind test group to verify the accuracy of the model.
Results. Twelve discriminating mass-to-charge (m/z) peaks were identified between NPSLE and controls: m/z peaks 7740, 11962, 8065, 7661, 6637, 5978, 11384, 11744, 8595, 10848, 7170, and 5806. The diagnostic decision tree model, built with a panel of m/z peaks 8595, 7170, 7661, 7740, and 5806, recognized NPSLE with both sensitivity and specificity of 92.6%, based on training group samples, and sensitivity and specificity of 91.7% and 85.7%, respectively, based on the blind test group. In addition, the root node m/z peak 8595 protein, which was downregulated in the CSF of patients with NPSLE after treatment, was identified and confirmed as ubiquitin by immunoprecipitation and ELISA.
Conclusion. Potential CSF biomarkers for NPSLE are identified by MALDI-TOF-MS combined with WCX magnetic beads. The novel diagnostic proteomic model with m/z peaks 8595, 7170, 7661, 7740, and 5806 is highly sensitive and relatively specific for NPSLE diagnosis. The level of ubiquitin in CSF is a promising biomarker for active NPSLE.
- NEUROPSYCHIATRIC LUPUS
- MATRIX-ASSISTED LASER DESORPTION/IONIZATION TIME-OF-FLIGHT MASS SPECTROMETRY
- WEAK CATION EXCHANGE MAGNETIC BEADS
- UBIQUITIN
Neuropsychiatric lupus (NPSLE)1,2 is a common yet severe manifestation of systemic lupus erythematosus (SLE), with prevalence of 15%–75% and mortality of 7%–13%3. Early diagnosis and prompt treatment could significantly improve its prognosis4. However, the diversity of its clinical presentations, the multiple potential etiologies, and the absence of sensitive and specific tests often make NPSLE diagnosis difficult5. In our study, we used matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS) combined with weak cation exchange (WCX) magnetic beads to screen and identify biomarkers from cerebrospinal fluid (CSF) in patients with NPSLE6. A decision tree model was built, and we found this novel diagnostic model was highly sensitive and specific for NPSLE.
MATERIALS AND METHODS
Adult patients aged > 18 years who were admitted to the Peking Union Medical College (PUMC) Hospital between November 2008 and May 2009 were enrolled. All patients with SLE fulfilled at least 4 of the 1997 American College of Rheumatology (ACR) revised criteria for SLE7. Each patient was defined as having NPSLE if he/she had significant clinical manifestation that fulfilled at least 1 of the 19 neuropsychiatric syndromes summarized by the ACR in 1992, identified by history, physical examination, and laboratory or radiographic tests, and further proved by clinical course and response to treatment. NPSLE was excluded if ≥ 1 of these criteria were present: hypertension (diastolic blood pressure > 120 mm Hg), hypoxia (PaO2 < 50 mm Hg), uremia (blood urea nitrogen > 35.5 mmol/l), serious electrolyte imbalance, or culture-proved central nervous system infection.
In all, 39 patients with NPSLE were enrolled, including 11 with seizure disorder, 8 with cognitive dysfunction, 5 with acute confusional state, 7 with psychosis, 3 with myelopathy, 4 with lupus headache, and 1 with cerebrovascular infarction. All enrolled patients with NPSLE were treated with methylprednisolone 1 g/day for 3 consecutive days, followed by prednisone 1 mg/kg/day plus cyclophosphamide or mycophenolate mofetil. They were also treated with a weekly intrathecal injection of 10 mg dexamethasone, plus 10 mg methotrexate, because this treatment regimen had been shown effective in treating NPSLE4,8. The NPSLE symptoms were improved by these treatments in 36 patients. Symptoms did not improve in 2 patients with acute confusional state and 1 patient with cerebrovascular infarction.
A total of 114 CSF samples were obtained. CSF proteomic spectra were generated by MALDI-TOF-MS combined with WCX magnetic beads. A decision tree model for NPSLE classification was constructed with data from a training group, and validated with independent data from a blind test group. The training group consisted of (1) paired samples of 27 patients with NPSLE before and 2 weeks after treatment; (2) 27 samples from a control group, including 10 patients with SLE without neuropsychiatric manifestations (non-NPSLE); and (3) 17 patients with scoliosis, a noninflammatory disease. The blind test group consisted of 12 patients with NPSLE before treatment, 12 patients with lumbar disc herniation, and 9 patients with other autoimmune diseases that had neuropsychiatric involvements, including 2 patients each with primary Sjögren’s syndrome, multiple sclerosis, Wegener’s granulomatosis, and Behçet’s disease, and 1 patient with relapsing polychondritis (Table 1).
In addition, the root node mass-to-charge (m/z) peak of this decision tree model was searched against an online proteomic database, and the potential candidate protein was confirmed with immunoprecipitation experiment plus MALDI-TOF-MS combined with WCX magnetic bead analysis as well as Western blot, and further validated by protein-specific ELISA.
Our study was approved by the ethics committee of PUMC Hospital, and informed consent was obtained from each patient or family.
Reagents and equipment
The MALDI-TOF-MS (Protein Biological System IIc) and Au-chip were both provided by Vermillion, Fremont, CA, USA. The WCX magnetic beads were purchased from Beijing SED Science and Technology, Beijing, China. Antiubiquitin antibody, protein A/G Plus-Agarose, horseradish peroxidase (HRP)-conjugated goat polyclonal to rabbit IgG and normal rabbit IgG were purchased from Santa Cruz Biotechnology, Santa Cruz, CA, USA. Antiubiquitin ELISA kit was purchased from Abcam, Cambridge, MA, USA. Other reagents were purchased from Sigma-Aldrich, St. Louis, MO, USA.
Sample collection
CSF samples were collected in sterilized tubes by lumbar puncture, and were immediately centrifuged at 3000 g for 10 min at 4°C. The supernatant was divided into 1-ml aliquots and stored at −80°C for subsequent proteomics analysis. The sample collection time was within 1 week after the onset of neuropsychiatric symptoms in SLE and other autoimmune disease controls.
Magnetic bead-based sample preparation for MALDI-TOF-MS
WCX magnetic bead was activated according to the manufacturer’s operation manual9. A CSF sample of 40 μl was incubated with 40 μl of U9 solution (9 M urea, 32.5 mM CHAPS, 64.8 mM dithiothreitol, 50 mM Tris-HCl buffer, pH 9) for 30 min at 4°C, then diluted with 320 μl of reaction buffer (150 mM sodium acetate, pH 4). The diluted sample was incubated with the activated magnetic beads for 1 h at room temperature, and washed twice with 100 μl binding buffer (50 mM sodium acetate, pH 4). Then the proteins bound to the beads were eluted with 10 μl 1% (v/v) trifluoroacetic acid (TFA). Finally, 3 μl of elute supernatant and 1.5 μl saturated SPA (50% ACN, 0.5% TFA) was spotted onto Au-chip for subsequent MALDI-TOF-MS analysis.
MALDI-TOF-MS analysis
Prepared Au-chips were placed on the Protein Biological System IIc mass spectrometer reader (Vermillion), and time-of-flight spectra were generated by averaging 81 laser shots collected on each spot at laser intensity 140–145, detector sensitivity 8. The optimization range was from 2000 to 30,000 m/z ratio, with the highest m/z 50,0006. Data reproducibility was tested and validated10. All the samples were tested by MALDI-TOF-MS in 1 batch to optimize the stability and accuracy. In addition, mass accuracy was calibrated externally by standard procedures using the all-in-one peptide molecular mass standards (Vermillion), and the spectra were calibrated by ProteinChip software (Vermillion). A representative CSF spectrum of 1 patient tested at 2 different times is shown in Figure 1. The mass difference in peak 11752 was 2.3% (< 0.3%). The coefficient of variability (CV) of intensity in peak 11752 was 8.2%.
Data analysis
The data were analyzed with ProteinChip software6,9. Step 1: Peak detection including (1) baseline subtraction, (2) mass accuracy calibration, and (3) automatic peaks detection. Using Biomarker Wizard Version 3.1.0 (Vermillion), biomarkers were generated that represent consistent protein peak sets across multiple spectra9. Baseline subtraction was performed on all spectra. The settings for autodetect peaks to cluster were first 5, mini peak 0%, cluster mass 0.3%, second pass 2. Step 2: Selection of differently expressed peaks that may represent potential biomarkers of NPSLE with SPSS software, p values < 0.05, were considered statistically significant (after normality analysis, data in normal distribution were analyzed with independent-samples T test or paired-samples T test; data in nonnormal distribution were analyzed with nonparametric Kruskal-Wallis test or Wilcoxon signed-rank test). Step 3: Construction of a decision tree model for NPSLE classification with Biomarker Patterns Software (BPS) 5.0 (Vermillion) with data from the training group, and validated with independent data from the blind test group. BPS is a pattern analytical tool developed by Breiman, et al11. BPS builds a binary decision tree algorithm with the peak information of the training set. The algorithm assigns each sample into 1 of the 2 nodes by rules based on the intensity of certain peaks12,13.
Protein identification
For immunoprecipitation, a volume of 180 μl CSF was incubated with 2 μg antiubiquitin polyclonal antibodies for 2 h. As negative controls, aliquots of the same CSF sample were incubated with normal rabbit polyclonal antibodies. Then 20 μl protein A/G agarose was added and incubated overnight. The samples were then spun down and the supernatants were analyzed by MALDI-TOF-MS combined with WCX beads.
For Western blotting, samples included ubiquitin standard protein, NPSLE CSF, NPSLE CSF supernatant after immunoprecipitation with antiubiquitin antibody, non-NPSLE CSF, and scoliosis CSF. They were subjected to sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transblotted to immobilon P membrane. The antiubiquitin antibody was incubated with the membrane at 1:200 dilution for 2 h. Then goat polyclonal IgG to rabbit (HRP-conjugated) was added for 1 h and detected with an enhanced luminol-based chemiluminescent kit.
For ubiquitin ELISA, CSF samples from 16 cases of NPSLE and 7 cases of non-NPSLE were tested by antiubiquitin ELISA kit according to the manufacturer’s instructions. OD 450 nm was read with a 2-wavelength microplate photometer.
RESULTS
Within the m/z peak range of 2000–20,000, there were 12 discriminating m/z peaks between NPSLE and the control group (p < 0.05), including 10 peaks that were upregulated in NPSLE and 2 peaks downregulated (Table 2).
Construction of decision tree model for NPSLE classification
The discriminating m/z peaks were analyzed by Biomarker Patterns Software 5.0 (BPS) to establish the optimal classification tree (Figure 2), with the following measurements9: method, 0; advanced, 10; testing, 10; costs, 1:1. The m/z peaks 8595, 7170, 7661, 7740, and 5806 were selected in the classification tree, and m/z peak 8595 was the root node. The peaks of m/z 7740, 7661, and 8595 were upregulated in patients with NPSLE compared with the control group, while the peaks of m/z 5806 and 7170 were downregulated (Figure 3). All 54 training group CSF samples were allocated to 6 terminal nodes. Samples allocated to terminal nodes 2 and 6 were classified as NPSLE; those allocated to terminal node 1, 3, 4, and 5 were classified as control. For example, if an unknown sample had peaks of m/z 8595 (intensity > 1.86) and m/z 7740 (intensity > 3.13), then the sample was placed in terminal node 6 and classified as NPSLE. If the sample was placed in terminal node 1, it was ruled out from NPSLE diagnosis. The model showed both sensitivity and specificity of 92.6% in classifying the training group CSF samples, and a sensitivity of 91.7% and a specificity of 85.7% in the blind test group (Table 3). The corresponding receiver-operating characteristic (ROC) curve of the optimal decision tree was supplied by the BPS. The ROC curve integral was 0.963 (Figure 4).
When CSF proteomic profiles of patients with NPSLE before and after treatment were compared, 4 discriminating peaks were identified. Peaks of m/z 4963 and 8595 were downregulated, and peaks of m/z 6637 and 6896 were upregulated after treatment (Figure 5).
Identification of m/z peak 8595 protein
Because the m/z peak 8595 was the root node of this decision tree model, and it was downregulated in the CSF of patients with NPSLE after treatment, we set out to search this discriminating peak against an online proteomic database (Geneva: Swiss Institute of Bioinformatics; http://expasy.org), and ubiquitin was identified as the potential protein. We then conducted an immunoprecipitation experiment to validate this candidate protein. Antiubiquitin antibody was used to deplete the CSF sample of ubiquitin. When the depleted sample was analyzed by MALDI-TOF-MS combined with WCX magnetic beads, we found that the m/z peak 8595 was dramatically reduced in the MALDI-TOF-MS spectra as compared with the sample treated with control antibody, while other protein peaks remained unchanged (Figure 6A–6C). This was also confirmed by Western blot after antiubiquitin antibody immunoprecipitation (Figure 6D). In addition, with a ubiquitin ELISA, we also found that the optic density of the NPSLE group (0.200 ± 0.080) was significantly higher than that of the non-NPSLE group (0.090 ± 0.021; p < 0.001; Figure 7).
DISCUSSION
MALDI-TOF-MS is a reliable, high-throughput technique for identifying proteins and/or peptides. WCX magnetic beads separate the proteins and/or peptides of different isoelectric points from complex biological fluids with specific anionic ligands. The techniques of MALDI-TOF-MS combined with WCX magnetic beads incorporate both of their advantages9, allowing the identification of comprehensive “fingerprints” of protein profiles within biological fluids, and were used to identify biomarkers of various diseases14. In the field of SLE, Mosley, et al15, Suzuki, et al16, and Rovin, et al17 used this technique to study the urine proteomics of lupus nephritis, and profiled the urine proteomic signature to distinguish lupus nephritis from lupus without nephritis, and/or active lupus nephritis from inactive lupus nephritis, independently. Huang and colleagues profiled the serum proteome of patients with SLE, established a decision tree for SLE classification, and provided a novel approach for the diagnosis of SLE18.
We compared the CSF proteomic profiles of patients with NPSLE and patients without NPSLE, and established a decision tree model for NPSLE classification, with both a sensitivity and a specificity of 92.6% in learning data classification, and a sensitivity of 91.7% and specificity of 85.7% in a blind test group, indicating that this model had high sensitivity and specificity for the diagnosis of NPSLE.
The discriminating m/z peaks included in the nodes of the decision tree model were those with the most significant difference, especially the root node peak at m/z peak 8595. We found the peak intensity at m/z peak 8595 was significantly higher in the CSF of patients with NPSLE than in the control group (Figure 3), and was downregulated after effective treatment (Figure 5), indicating that protein of m/z peak 8595 strongly correlates with NPSLE activity. In our study, the protein at m/z peak 8595 was identified and validated as ubiquitin by immunoprecipitation experiment plus MALDI-TOF-MS combined with WCX magnetic bead analysis as well as Western blot. In addition, ELISA also found that the level of ubiquitin in the CSF of patients with NPSLE was significantly higher than that of patients without NPSLE, suggesting ubiquitin could be a promising biomarker indicating active NPSLE.
Ubiquitin is a small protein that is ubiquitously expressed in eukaryotes19. It consists of 76 amino acids and has a molecular weight of about 8.5 kDa. The most important function of ubiquitin is labeling various target proteins for proteasome degradation. The ubiquitination system functions in various cellular processes, including apoptosis, immune response, and inflammation, as well as neural and muscular degeneration. There were also reports showing that ubiquitin was elevated in the CSF of patients with some other neurological conditions, including Alzheimer’s disease20,21 and Creutzfeldt-Jakob disease22, suggesting ubiquitin might participate in neural apoptosis and degeneration. We showed that in patients with SLE, an elevated level of ubiquitin in CSF indicates neuropsychiatric involvement and it correlates with NPSLE disease activity.
We identified several discriminating protein peaks for NPSLE by MALDI-TOF-MS combined with WCX magnetic beads. The decision tree model with m/z peaks 8595, 7170, 7661, 7740, and 5806 was highly sensitive and relatively specific for NPSLE classification. In addition, the level of ubiquitin in CSF was identified as a promising biomarker for active NPSLE. This study provides useful biomarker data in NPSLE, and more studies are needed before these biomarkers can be introduced in clinical practice.
Acknowledgment
We thank Dr. Haiteng Deng from the Proteomics Resource Center, Rockefeller University, New York, NY, USA, for his helpful comments.
Footnotes
-
Supported by New Century Excellent Talents Project, Ministry of Education of China (NCET04-0191), National Natural Science Foundation of China (30400410, 30972731), National Natural Science Foundation of Beijing (7052052), and the National High Technology Research and Development Project of China (973 Project) (2007CB512405).
- Accepted for publication October 28, 2010.