In this proof-of-concept study, we demonstrate the feasibility of temporal proteomic profiling of cell culture media for HBEC response to oxidative stress, followed by exploratory bedside plasma testing. This novel approach may provide a quick and reproducible screening tool for candidate markers and/or targets of stroke therapy, in which it otherwise would be difficult to find low-abundance biomarkers using discovery proteomic technology. Our data showed that there were clear differences between healthy HBEC vs oxidatively stressed HBEC. Healthy cells developed networks of cross-talk involving secreted factors, membrane receptors, and matrix components,13
whereas stressed HBEC released potential biomarkers, demonstrating different patterns over time.
In general, the endothelial response to oxidative stress was highly dynamic and cellular machinery and inflammatory markers peaked gradually, secreted factors tended to oscillate over time, and endothelial injury and repair were manifested in dynamic changes in factors related to matrix integrity. Reactive/inflammatory markers specific to ischemia increased over time, as represented by TSP-1, chemokine ligand-1, heat shock protein-1, serum amyloid-A1, annexin-5, and serum amyloid-A1. Neurotrophic factors and neuropeptides such as prosaposin, nucleobindin-1, annexin-2, calsyntenin-3, calsyntenin-1, protease inhibitors (eg, tissue inhibitor of matrix metalloproteinase-1), and tachykinin precursors (tachykinin-1, which converts to various neuropeoptides such as substance p and neurokinin) tended to peak mid-treatment (12 hours) and then rapidly decreased by 24 hours (). In general, these responses are consistent with the idea that cellular stress leads to an overall downregulation of beneficial trophic mediators and an increase in potentially deleterious inflammatory signals.14
Loss of vascular trophic coupling may be an important part of stroke pathophysiology.6,15
In addition to trophic and inflammatory markers, our data also suggested that oxidative stress may trigger deterioration and active remodeling of endothelial extracellular matrix. Over time, basement membrane components decreased (fibronectin-1, desmoglein-1, profiling-1), whereas other cytoskeleton components (nidogen-1, actin, vimentin, and filamin B) appeared to increase over time after oxidative stress (). In the context of brain endothelium, these responses are consistent with blood–brain barrier alterations. Our findings are supported by other proteomic studies demonstrating that cytoskeletal proteins contribute to the dynamic blood–brain barrier responses when bovine brain endothelial cells are co-cultured with astrocytes.16
In comparison to the study by Haqqani et al in which both rat brain endothelial cellular and secreted proteins were studied by 2-D gel (n=38 proteins) vs isotope-based (n=138 proteins) proteomics, our study only measured protein secreted by human brain endothelial cells and found a higher number (n=277) of lower-abundance proteins, correlating to human plasma at a low picogram level, a concentration previously difficult to achieve by direct human plasma screening.17,18
Taken together, our findings suggest that brain endothelium should be a rich source of brain injury-specific biomarkers comprising trophic, inflammatory, and barrier properties that then become accessible in the systemic circulation. There is an overlap between proteins found in our endothelial cell culture and the published human plasma proteome.19
In this initial study, a large number of potential markers were profiled. One of the high-ranking candidates was TSP-1, a pleiotropic antiangiogenic factor involved in coagulation and atherosclerosis.12
In our brain endothelial cells, TSP-1 was produced and then actively degraded after 12 hours of oxidative stress. Our data are consistent with those of previous studies of mouse models of cerebral ischemia in which TSP-1 increases within 1 hour after occlusion.20
Because, in stroke, focal ischemia leading to oxidative injury begins with a specific clinical event with an onset that can be timed, we explored our finding further by measuring TSP-1 levels in a small cohort of acute ischemic stroke patients whose exact time of stroke onset was known. Consistent with our in vitro findings, TSP-1 levels were higher in acute stroke, within 8 hours of initial symptom onset, compared to age-matched controls with similar clinical risk factors. Two other rising markers (chemokine ligand-1 and nidogen-1) also appeared to be higher in stroke patients, although these results did not reach statistical significance. Importantly, our cell culture markers showed a good dynamic range of pg to ng/mL concentration in human plasma, as confirmed by our enzyme-linked immunosorbent assay measurements, highlighting the potential for detecting low-abundance candidates otherwise not feasible in direct human plasma proteomic screening.18
However, it must be acknowledged that there are many caveats. We did not assess potential markers from the oscillating category because it will be difficult to match the timing of our cell cultures to variable stroke onset and sample collection times in patients. The same is true for proteins from the falling category. For example, although fibronectin-1 is listed in this category, its temporal profile peaks at 6 hours and then falls over the next 24 hours (). So, depending on the time of sample collection, fibronectin-1 levels may be high or low in patients. However, our cell-based finding of an early fibronectin-1 peak is at least consistent with a previous study showing elevated plasma fibronectin levels within the first 6 to 8 hours after stroke onset.21
Similarly, elevations in markers such as IL-6 and tissue inhibitor of matrix metalloproteinase-2 also have been reported in stroke patients previously.22,23
Ultimately, our initial attempt at validation is obviously limited by our sample size. However, our findings are consistent with the idea that specific markers derived from our cell culture model were measurable in actual human stroke samples.
Proteomic profiling of mouse and rat endothelium after ischemia has been actively investigated.17,24
However, to our knowledge, temporal proteomic profiles of human brain endothelial responses have not been reported. Differences between human and rodent cells are not fully understood. For example, our findings of increased vimentin and decreased fibronectin at 24 hours are similar to those of the Haqqani et al17
study that looked at rat endothelial cells. However, our highest-ranking and clinically validated marker TSP-1 was not detected in the rat cells at all.17
The dynamic changes of secreted factors may have important roles in cell signaling after ischemia. Furthermore, analysis of fragment sizes may help identify neurovascular proteases. Although our data are driven by oxidative stress because this is the central trigger after stroke,25
this relatively simple methodology can be generalizable to other insults. Because it is challenging to characterize unknown low-abundance secreted factors by discovery proteomics directly at the bedside, an initial screening in a “cleaner” cell culture system may yield candidates that later can be tested clinically. Ultimately, larger collaborative efforts are required to validate more candidates of interest, as advocated by major proteomic and research organizations such as the Human Proteome Organization and the National Institutes of Health.
The present study provides proof of concept. A full in vivo validation of all potential candidate biomarkers is outside the scope of our initial study. This type of label-free gel-based intact protein discovery proteomics, even when optimized in a serum-free cell-culture system, still may not be able to discover exceedingly low levels of secreted factors or give the most accurate quantification. Isotope-labeled techniques offer more accurate quantitation26
than the semiquantitative label-free ratio we report. Because endothelial cells do not work in isolation, contribution by neurons, astrocytes, and hematologic agents may alter their secretory function. In addition, oxygen glucose deprivation also might be a better “stroke mimic.” In our model in which SNP is used, nitrosylated proteins would be of interest for future study. And dose-response is important because U-shape curves also may be present, depending on the protein involved. Further studies using RNA, quantitative protein microarrays for target validation, co-cultures with glial cells, validation in other model systems with dose-response curves, and larger clinical cohorts with timed samples to validate other candidates are required to confirm our findings.