In this study we performed ALS biomarker discovery and verification on CSF from 241 subjects using the SELDI-TOF-MS platform. Protein mass peaks corresponding to C-reactive protein (CRP), cystatin C and transthyretin provided the best individual diagnostic predictive value for ALS, though the overall accuracy for any one mass peak remained less than 70%. A biomarker panel identified using the RL algorithm generated 82% accuracy in a 10-fold cross-validation study using all CSF samples. Three of the biomarker peaks identified by RL are identical to those we previously reported 8
. We also performed direct group comparisons and obtained mass peaks that could distinguish between the subject groups. However further studies using additional ALS disease mimics (such as pure upper motor neuron disease, pure lower motor neuron disease, multifocal motor neuropathy) are required to access the ability of this biomarker panel to differentiate between motor neuron diseases. The current study included 4 LMN subjects, and 3 of the 4 were classified as ALS by the biomarkers. However 2 of these 3 LMN cases later exhibited UMN symptoms and were subsequently re-classified as either definite or probable/lab supported ALS. Therefore the protein biomarkers identified LMN subjects that later developed clinical symptoms for the classification of ALS. Our study did not detect any normal age-related changes in biomarker protein levels, though it was not powered to examine age-associated changes in protein levels. This issue must be explored in future studies.
The 23,030 Da mass spectral peak exhibited the single best sensitivity and specificity for ALS and has been reported to be C-reactive protein (CRP), an acute-phase inflammatory protein 17,19
. CRP was not detected as a potential biomarker in our prior study due to the mass range cut-off of 20 kDa used for that study. We observed a statistically significant increase of CRP in ALS subjects when compared to HC or AD subjects but no significant difference to MS or other disease control subjects, indicative of inflammatory processes in these other neurologic disorders. We also failed to detect significant differences for CRP between ALS and the upper and lower motor neuron disease mimics. These results were confirmed by ELISA using a subset of the total subjects used for mass spectrometric analysis (). A recent study found increased blood levels of CRP in ALS patients that correlated with ALS-FRS measurements 20
. We also identified alterations in another acute-phase protein, transthyretin, during ALS.
Little is known about the role of CRP in the central nervous system, but it is expressed by neurons 21
and has been observed in neurofibrillary tangles during AD 22
. The mean levels of CRP in control CSF have been reported to be 3.2 to 8 μg/L 23,24
, values similar to those in our study. A 2-fold elevated CSF level of CRP in ALS suggests low level activation of the immune system, consistent with prior studies that indicate increased inflammation in ALS and the potential for a CSF inflammatory profile to be indicative of the disease 25,26
. By comparison, this CRP level is much lower than that observed in acute bacterial or viral infections, such as meningitis 27
. It would be interesting to determine if CRP levels increase during disease progression within individual patients, suggestive of increased inflammation during ALS. We found no correlation between the CRP level and the time from symptom onset to lumbar spinal tap used in this study.
Our SELDI-TOF-MS analysis demonstrated reduced levels of native transthyretin and increased levels of CysGly-transthyretin, a modified form generated by oxidative damage 28,29
. Oxidative modifications to transthyretin were increased in ALS patients, but they also occurred in other neurodegenerative conditions (). Oxidized transthyretin retains its tetramer structure and is resistant to formation of amyloid fibrils that occur in familial amyloidotic polyneuropathy patients 30
. Thus oxidation of transthyretin may stabilize transthyretin function during metabolic stress conditions associated with ALS. Statistically significant reductions in cystatin C were also noted in ALS patients. These results confirm prior studies from our lab and others 7,8,12
. In addition, we identified mass spectral peaks corresponding to a number of proteins that were previously shown to exhibit altered protein levels or contribute to pathogenic mechanisms of ALS, including chromogranin B, neuroendocrine protein 7B2, β2-microglobulin, and β-amyloid peptide 8,26,31-33
We failed to identify a specific protein peak that correlated with disease duration across ALS patients (). A recent study using a small number of subjects also failed to identify a correlation between cystatin C levels and disease duration 12
. However data in both studies represents only a single time point within the disease for each patient. Further studies using longitudinal samples collected from individual ALS patients are required to identify candidate prognostic biomarkers of disease progression. Interestingly, cystatin C levels correlated with patient survival in limb-onset ALS. This suggests that cystatin C levels in the CSF may be useful as a marker of survival in specific sub-types of ALS. However bulbar-onset ALS patients in our study exhibited a significantly shorter disease course when compared with limb-onset patients. Therefore levels of cystatin C may not correlate with survival of rapid disease progressors. Further studies are required to explore this possibility. A functional role for cystatin C in ALS remains uncertain but worthy of pursuit. While cystatin C is a known component of Bunina bodies 34
, reduced cystatin C in CSF indicates a reduction in extracellular cystatin C levels. Extracellular cystatin C is a regulator of cysteine proteases such as cathepsins and calpains. Reduced cystatin C in ALS may enhance cysteine protease-mediated degradation of extracellular matrix proteins and retard regenerative and/or repair events 35
. Extracellular cysteine proteases also modulate degradation of cell surface proteins and may facilitate localized neuronal damage and death. Reduction of cystatin C in transgenic mice results in enhanced neuronal cell death following focal ischemia 36
. Cystatin C has also been shown to modulate cerebral amyloidosis, and reduced cystatin C may enhance cerebral vessel damage due to amyloidosis 37
. Recently, reduced levels of cystatin C were observed in spinal motor neurons and astrocytes in ALS patients and were correlated with formation of TDP-43 inclusions 38
. Therefore cystatin C exhibits multiple extracellular and intracellular functions that may contribute to the pathogenesis of ALS.
The rules generated by the RL algorithm predicted ALS with a high level of specificity (94%) and 82% overall accuracy. The large cohort size in this study may explain the increased number of mass peaks necessary to distinguish ALS from healthy and other disease controls, when compared to our prior study. The ability to rule out ALS with a high degree of confidence (94% specificity) while predicting ALS with only 63% sensitivity is suggestive of ALS disease heterogeneity. A high degree of specificity would provide diagnostic value to the clinician to rule out ALS, but the lower sensitivity would initially miss many ALS patients. Alternatively, the biomarker peaks identified by SELDI-TOF-MS may not provide the best panel or platform to predict ALS with high sensitivity. To enhance our ability to identify biomarkers with high sensitivity for ALS, future studies will require a larger ALS patient cohort containing sufficient numbers of motor neuron disease subtypes based on clinical parameters, and the use of multiple experimental platforms. A consortium of clinical sites, each following standard procedures of sample collection and processing, will be required to complete such a study. While we replicated findings for many mass peaks from our prior study, issues of SELDI-TOF-MS reproducibility across laboratories may limit its utility as a diagnostic platform 10,11,39
. ELISA or more quantitative mass spectrometry techniques will likely be required to obtain the high level of sensitivity and reproducibility across laboratories necessary to generate clinical utility for a diagnostic assay. A recent review article described the biomarker development pathway for ALS 40
. Our current verification study represents a next-step in this biomarker development pathway, but further studies are required to first validate the biomarkers and then ultimately qualify them for ALS. Future prospective studies using CSF collected from subjects at the initial neurology clinic visit are required to first validate any candidate diagnostic biomarkers and demonstrate clinical utility by comparing the biomarker assay predictions to the subsequent clinical diagnosis. Such a prospective study will reduce the time from symptom onset to lumbar tap observed in our current study () and determine the ability of the biomarkers to identify ALS soon after clinical symptom onset. Finally, large prospective studies would be required to evaluate the utility of any ALS specific biomarker for use in the general population.