Traumatic injury to the spinal cord can result in life-changing neurological deficits. Patients, family members, clinicians, and researchers alike face significant long-term challenges including (but not limited to) pain management, rehabilitation, and functional improvement. According to 2011 estimates, the incidence of spinal cord injury (SCI) is roughly 12,000 cases per year in the USA, 40% resulting from motor vehicle accidents and 27% result from falls, creating a bimodal age distribution in the patient population [1]. Long-term implications of these debilitating injuries are significant. Advances in patient care, specifically prevention and treatment of urinary tract complications and renal failure, have improved longevity such that causes of mortality are now similar to the general population (cardiovascular disease, cancer, and lower respiratory disease) [2, 3].

The molecular mechanisms underlying inflammation-induced damage and recovery of function at the acute, subacute and chronic stages after SCI are currently being pieced together. Although trauma to the spinal cord results in rapid loss of neurons and glia within gray and white matter and transection of white matter pathways, this acute insult also initiates a robust inflammatory response [4–6] that continues to damage tissue [7, 8], creating new deficits at subacute and chronic time points post-injury. Loss of blood-spinal cord barrier (BSCB) [9] and lymphocytic infiltrate into the injury site [10] potentiates inflammation, and reactive gliosis and formation of scar tissue prevent axon regeneration [5, 11, 12].

With clinical trials testing the effects of new therapies on neurologic outcomes post-SCI already underway (minocycline and erythropoetin to name a few) [13], and more still in the scientific pipeline, the role of cerebrospinal fluid (CSF) biomarkers as indicators of injury progression and repair is now more important than ever. Although functional improvement remains the gold standard and certainly the most important outcome for patients, Kwon et al. highlight the potential role for biomarkers in assessing initial injury severity and tracking changes in injury status over time. In addition, they emphasize the role of biomarkers in determining therapeutic response more objectively than current highly subjective functional outcome measures like the American Spinal Injury Association (ASIA) Score [14]. Other important potential applications include characterization of molecular signs of neurotoxic damage and endogenous neuro-repair, and discovery of new therapeutic targets. What is needed is a protein-based high-throughput assay that can detect biomarkers from a safe and clinically available source. Sensitivity is also critical because CSF (an important source of biomarkers given its proximity to neuronal tissue) contains relatively low protein concentrations even after BSCB disruption [15]. This ideal assay would be of significant value to researchers and clinicians alike, facilitating delivery of new therapies to SCI patients.

It is becoming increasingly apparent that inflammation plays key roles in both acute and secondary damage, and late-phase recovery after traumatic SCI, and that any novel therapy would almost certainly interact with these processes. In fact, much can be learned from the Immunology field, where researchers have adopted many of the latest technologies to characterize inflammatory processes previously too complex for traditional methods of investigation. Where PCR and standard ELISA were once popular for detecting changes in mRNA and protein respectively, newer multiplex arrays have emerged as the current assays of choice. Recently Lee et al. simultaneously profiled 10 different cytokine responses to H1N1 infection in hospitalized patients using a cytometric bead array system [16], and Mizutani et al. were able to track serial changes in 17 cytokines using a similar suspension array in patients with Crohn’s disease treated with Infliximab [17]. Oncology is another field that has made good on the promise of newer array technology. Tefferi et al. studied the prognostic value of 30 cytokines in primary myelofibrosis by multiplex sandwich immunoassay [18]: one of many truly comprehensive cytokine profiling studies. For a thorough review of multiplex sandwich ELISA technology read Nielsen and Geierstanger [19].

An effort has also been made to validate the reliability of these new assays, focusing mainly on the issues surrounding transition from a uniplex to multiplex format. Pang et al. studied two commercially available solution-phase microarray platforms against standard ELISA in detecting autoantibodies in human serum [20]. They note similar detection performance, and name high-throughput screening and reduced reagent consumption as major advantages over standard methods. This paper also emphasizes the importance of cutoff values (criteria for determining positive and negative signals in the dataset). In further considering data processing, the multiplex format necessitates different statistical treatment of the data: the need for multiple hypothesis testing. The concept of the false discovery rate (FDR) comes from earlier work in epidemiology and genome-wide association studies: the idea that as the number of statistical tests performed increases, so does the number of false positives. Hsueh et al. compare strategies like Benjamini-Hochburg [21] for estimating the FDR, and emphasize the importance of controlling the type I error rate in multiplicity testing [22]. Finally, important technical issues such as binding ‘inhomogeneity’ across immobilized spot-based arrays [23] are currently under investigation. To summarize, multiplex proteomic assays, although relatively new, offer many qualities of the ‘ideal assay’ described above, provided there is: correct statistical treatment of the data, careful consideration of technical issues, and verification of ‘hits’ with appropriate confirmatory methods.

The challenge of characterizing the inflammatory response to acute SCI has been met using a variety of methods from flow cytometry, to PCR, to ELISA. Streit et al. showed transient increases in IL-1β, TNF-α, IL-6, and M-CSF mRNA by PCR from cord tissue in a rat model of SCI (C8 contusion) [24]. Wang et al. also showed elevated IL-1β mRNA in contused cord, but extended their analysis by showing elevated protein levels in cord by ELISA. Of note, CSF samples also contained low levels of IL-1β [25]. These studies, and others, established the presence of inflammatory molecules in the setting of acute SCI, and yet they were clearly limited by the number of molecules that could be studied. More recently, the emergence of high-throughput multiplex techniques has allowed for more comprehensive analyses. Kwon et al. studied 25 cytokines (as well as a number of growth factors) simultaneously in human CSF samples from acute SCI patients [26]. Lubieniecka et al. identified 42 possible biomarkers (cytokines, growth factors, etc.) in CSF from a rat model 24 hours post-injury by mass spectrometry of which 10 correlated with injury severity (ASIA Score) including Ywhaz, Itih4, and Gpx3 [27].

In an effort to characterize inflammatory markers present at subacute time points after traumatic SCI, thought to represent a clinically relevant ‘window of opportunity’ for therapeutic intervention, we have conducted a proteomics analysis of CSF from rats at 12 days after cervical spinal cord contusion for changes in 34 cytokines using a multiplex sandwich ELISA microarray. Initial statistical analysis yielded seven proteins (IL-1α, Leptin, B7-2/CD86, GM-CSF, IL-1β, matrix metalloproteinase-8 (MMP-8), and tissue inhibitor of metalloproteinase-1 (TIMP-1)), the levels of which were all increased in the CSF from injured animals compared with normal uninjured controls. The FDR was controlled (limit type I errors) by adjusting the P-values using both Benjamini-Hochburg and Bonferroni methods (although we chose to focus our analysis on Benjamini-Hochburg adjustments because this method generally allows more potential hits to reach significance). To account for technical replicates, as well as variation in regional signal intensity within the array, a linear mixed model was applied, which showed only MMP-8 was significantly elevated in injured CSF, a finding subsequently confirmed by western blot.

The primary objective of this study was to characterize the inflammatory profile present in CSF at a subacute time point in a clinically relevant rodent model of traumatic SCI. Our other aim was to test a microarray proteomics platform specifically for this application. Studying inflammation at 12 days post-SCI is highly relevant in that it falls within a clinical ‘window of opportunity’ where patients have been stabilized and are generally more receptive to therapy. We also agree with the SCI field in general that CSF is an important source of potential biomarkers given that it is safe to obtain, readily available, and in close proximity to neuronal tissue of interest. Using small volumes of CSF we were able to study 34 cytokines/growth factors simultaneously, detecting significant changes in seven inflammatory markers. After applying statistical models designed for multiplex testing, we also found robust increases in MMP-8 within CSF, which to our knowledge represents a novel subacute phase biomarker for SCI.

As with any new technology, comparison of results to previous data is crucial in assessing the reliability of the new system. It is generally accepted that a multiphasic inflammatory response occurs after SCI: an early phase peaking at 24 hours and extending from 9 to 10 days, and a later phase between 14 and 180 days post-injury [31]. Studies of inflammation in the setting of SCI have primarily focused their analysis on this early phase between 0 and 72 hours post-injury when neuroprotective strategies are most effective, making comparisons to this study somewhat difficult. Changes in mRNA from rat cord lysate observed by Streit et al. (IL-1β, TNF-α, IL-6, and M-CSF) were transient, returning to normal levels by 24 hours post-injury, and remaining at these levels for the remainder of their analysis (10 days) [24]. In this study, TGF-β levels started to rise at 24 hours, peaked at 7days, and remained elevated at 10 days; TGF-β was not included in our analysis. The study by Wang et al. showed elevated IL-1β protein in CSF at one hour [25]. Their result differed from Streit et al. in that protein levels remained elevated at 7 days, but fits nicely with our IL-1β simple model result. Although changes in IL-1β did not hold up to mixed model analysis, it is quite possible that this result is real, and that we are observing a lull in protein concentration between early- and late-phase inflammation. Further analysis is needed to confirm this result. Yang et al. showed elevated levels of IL-1β, IL-6, and TNF-α by western blot from cord lysates, a result that peaked at 6 hours, returned to normal by 24 hours, and was dependent on injury severity [6]. IL-6 and TNF-α levels in our study, although elevated in injured animals, did not reach statistical significance. Finally, the human CSF study by Kwon and colleagues did not show any changes in IL-1β or GM-CSF protein over a 5-day period [26] (both elevated in our simple model at 12 days), a difference that may be explained by the observation window, and/or cross-species differences in the inflammatory response.

MMPs and their endogenous inhibitors (TIMPs) have been studied extensively in the setting of SCI. de Castro et al. showed increases in gelatinases MMP-9 and MMP-2 at 12 to14 hours, and 5 days respectively in injured rat spinal cord by zymography. They suggest (using neutralizing antibody studies) that these MMPs are released by infiltrating neutrophils responding to inflammation [32]. Wells et al. performed a comprehensive MMP/TIMP mRNA profile in an acute murine model of SCI. They showed an increase in MMP-3, 7, 10, 11, 19, and 20 at 24 hours post-SCI, and elevated levels of MMP-2, 12, and 13, but not MMP-8 at 5 days, colocalizing MMP-12 mRNA to Iba1 positive macrophages in the lesion epicenter [33].Using microarray proteomics, we present a new finding, an increase in MMP-8 protein, also known as neutrophil collagenase or collagenase-2, in CSF at 12 days after SCI in rats. This result is consistent with findings of Veeravalli et al. who showed an increase in MMP-8 mRNA following SCI between 3 and 21 days post-injury in mouse cord lysate samples [34]. Increases in TGF- β mRNA 10 days post-SCI reported by Streit et al.[24] may in part explain the elevated levels of MMP-8 in our study, as TGF- β is a known inducer of other collagenases like MMP-13, through the Smad3 pathway [35], as well as MMP-2, and MMP-9 though ERK-1/2, and p38 MAPK [36]. In the Streit study [24], TGF-β levels started to rise at 24 hours, peaked at 7 days, and remained elevated at 10 days; TGF-β however was not included in our analysis. Our finding also correlates with data from the stroke field, where Cuadrado and colleagues reported upregulation of MMP-8 protein (and other MMPs) in ischemic brain tissue [37]. Finally, the fact that increases in CSF MMP-8 protein levels were not reported by Lubieniecka et al.[27], despite the use of sensitive mass spectrometry analysis at time points ranging from 12 to 72 hours after injury, indicates that our observed increase in MMP-8 protein in CSF represents a novel subacute-phase SCI biomarker. Increases in a variety of MMPs within the spinal cord following SCI are thought to significantly contribute to further tissue damage particularly at acute time points after SCI. At present, however, relatively little is known about CSF levels of specific MMP mRNAs and proteins or their impact on tissue damage at subacute time points after SCI. Elevated levels of MMP-12 mRNA in spinal cord injured tissue observed by Wells and colleagues [33] have been shown to correlate with disruption of the BSCB, however this study did not extend beyond 5 days post-SCI. Whether the significant increase in MMP-8 protein we have observed in the CSF contributes to BSCB disruption and other mechanisms of tissue damage post-SCI requires further investigation beyond the scope of the present study.

Comparing MMP-8 and TIMP-1 signals in both microarray statistical models illustrates the importance of data interpretation in multiplex testing. MMP-8, by far the greatest protein change detected, was significant across both models, suggesting the ability of the array and mixed statistical model combination to correctly identify true changes in the dataset. We would like to reiterate the importance of methods like western blot in verifying protein size when using an ELISA-based system because other immunoreactive proteins may disrupt the analysis.

TIMP-1, by contrast, only reached statistical significance in our simple model. Lack of significance in the linear mixed model indicates that this result is more likely to be a type 1 error than a true positive. Western blot was performed to test this hypothesis. Although it is possible that our inability to detect TIMP by western blot reflects the ability of the mixed model to control the FDR, it is also possible that TIMP-1 is truly elevated following SCI and, as discussed above, the microarray was just more sensitive at identifying biologically significant changes in this molecule.

Clearly, uncertainty exists when interpreting the results of multiplex testing. Different statistical treatment of the data gives us clues, but follow-up testing is often required by conventional methods. This is especially true when using the Benjamini-Hochburg correction, accepting a certain number of false positives per assay. However, there are ways, within the confines of the multiplex format, to reduce uncertainty. One obvious option would be to increase the sample size of each dataset. If we were to run two chips instead of one would we not increase our power to detect differences in the data? This brings up the concept of batch variability: the observation of significant variation between batches (chips in our experiment). This variability commonly offsets any statistical ‘gain’ from the corresponding increase in sample size. A better alternative would be to run sequential arrays, one for screening, as presented here, and another ‘confirmation array’, where only hits from the screening array are spotted, in greater numbers. This sequential approach, although more expensive and time consuming, avoids the problem of batch variation, and prevents controls for the FDR from making costly type II errors.

To our knowledge our study is the first to demonstrate significant increases in MMP-8 protein in CSF after traumatic injury to the spinal cord. We propose that MMP-8 is a novel subacute-phase CSF biomarker for SCI. In addition, our study supports the use of sandwich microarray proteomics as a valid approach for studying CSF in the setting of SCI in rats, and potentially human SCI patients. This assay reliably detects low levels of CSF protein, and has several advantages over conventional methods including quick and easy protocol, high-throughput format, and use of relatively small amounts of precious sample. Disadvantages include relative cost compared with traditional methods, and the need to confirm hits (either with repeat microarray analysis or other assays). Moreover our study stresses the importance of controlling the FDR given the multiplex format. In summary, this assay has the potential to detect new target molecules, characterize complex time-dependent pathophysiologic responses to SCI (like inflammation) and, as more therapies reach the clinic, identify biomarkers with which to assess injury severity and grade response to treatment.