Introduction
In the management of acute cerebrovascular syndrome (ACVS) [
1], the high prevalence of conditions that mimic stroke presents a challenge, particularly for first-line physicians. Such mimics include migraine, Todd’s paresis following seizure, delirium, compressive neuropathies, and many other entities [
2]. Unlike cardiology where an ECG and single blood test allows effective filtering, the first step with ACVS may be advanced imaging and/or specialist referral.
The development and validation of a reliable blood biomarker test capable of distinguishing ACVS from mimic has been challenging, despite numerous multi-center studies of varying size [
3‐
11]. Additionally, most stroke biomarker studies have used ELISA technology, with the exception of some studies using newer methods for protein quantification such as mass spectrometry (MS) [
12,
13]. ELISA uses immunoassay to measure protein expression but each protein requires a separate assay, even when a few are bundled together in a composite test. In contrast, MS allows simultaneous quantitation of large numbers of biomarkers, in a rapid, reproducible and sensitive assay at low cost per sample; the drawback being that low-abundance proteins remain a challenge to detect using MS. To date, no protein biomarkers have been successfully adopted into clinical practice, although commercial ELISA kits for stroke do exist, as performance has not been adequate [
14].
In this paper, we report on a small-scale exploratory case-control study to examine the natural abundance and variability of 147 candidate plasma proteins in ischemic stroke and stroke-mimic patients. These candidate proteins include markers of stroke and cardiovascular disease selected through a comprehensive literature review of stroke biomarker research published prior to 2014 [
15,
16]. We were required by our funders to examine published markers as discovery research is not eligible for this type of translational research funding. The objectives of this small-scale case-control study are (1) to confirm that the MS proteomic platforms yield useful information and (2) provide a preliminary vetting of many of the candidates for our eventual protein biomarker panel in development for transient ischemic attack (TIA) or mild stroke. This work is part of a larger SpecTRA study [
17,
18], a large-scale, multi-site, precision medicine project that uses MS to measure 141 proteins concurrently in a clinical research program involving 1860 patients, to verify and validate a clinically useful blood test for TIA and minor stroke. As well, we chose to study severe stroke in this small-scale study on the premise that this would provide a more robust target than TIA, in terms of greater differential abundance of up- or down-regulated proteins, as TIA represents a mid-way point on the ACVS continuum.
Discussion
The management of ACVS could be greatly improved by the availability of robust, accessible, and inexpensive biomarkers. Clinical decisions that could benefit from a proteomic blood test include differentiating TIA from its many mimics, and in the future, a validated and clinically useful blood test could provide guidance to select candidates for thrombolysis and thrombectomy. A pre-requisite to candidate selection would be that the assay development and validation would necessarily be contingent upon plasma collected during a tight time windom from symptom onset that mirrors that of the hyper-acute context in which those interventions are delivered. To date, no individual protein biomarker has emerged for such critical decision support, but there are many interesting candidates. The answer may lie in finding patterns of biomarkers rather than single entities. Given that the majority of publications report on a single protein or a small group of proteins, typically two to five, this puts weight on the ability to gather sufficient data to find such patterns. This study demonstrates how one technique, LC/MRM-MS, can be used to achieve larger data sets, laying a foundation for further research to demonstrate the utility of MS in translational settings at the bedside with benchtop MS machines in laboratories or as point of care devices, for example.
We examined 147 high interest proteins and found a subset of 30 proteins that, in conjunction with age, could distinguish stroke from stroke-controls with a high AUC, thus achieving our goal of demonstrating the potential value of biomarker panel for ACVS diagnosis. The 30 significant proteins are involved in blood coagulation, inflammation, neurovascular unit injury, cell adhesion, and atrial fibrillation. Using STRING analysis, we identified 20 of them to have high-confidence molecular interactions with one another. The significant proteins and pathways highlighted here are consistent with the known biology of how proteins are up- or down-regulated during ischemic stroke.
The results of this study are limited by sample size. We acknowledge that 40 patients do not justify representation of the tremendous heterogeneity inherent in stroke and mimic populations, yet this was designed specifically as a pilot study to assess (i) feasibility of plasma collection, processing, consent, and logistics for the larger SpecTRA project (for TIA biomarkers) with enrollment that will surpass 1600 ED patients and (ii) detection of a proteomics signal in stroke to serve as a baseline against which we will compare our larger TIA biomarker study and its control group.
Further, our results are also limited by an imbalance in baseline patient characteristics between strokes and controls as we were not able to match patients in both groups for health and sociodemographic variables, such as age. We acknowledge that the generalizability of the results is affected by limiting the model to age. It is likely that measures of stroke severity and related comorbidities would be associated with the diagnosis of stroke. Patient age, though, is a clinical variable easily collected in patient care settings. Other clinical measures, such as time from symptom onset, are often imprecisely measured or reported by patients. However, the aim of our study was to assess whether MS platforms are capable of yielding useful information for the diagnosis of stroke. The results of our study suggest that after accounting for patient age, such is the case. The generalizability of these results may be attenuated by other competing clinical measures not included in our model (e.g., stroke size) that may share information with the proteins we assessed. In the larger TIA biomarker study, we demonstrate the discriminative capacity of these proteomic markers over and above readily available clinical variables that are often collected in the prehospital and acute care settings during stroke assessments, for example, vascular comorbidities, presence of atrial fibrillation, and motor deficit.
The generalizability of our results, though, we suggest, is a secondary consideration to our main aim of demonstrating that the MS platform in conjunction with a readily available clinical variable (i.e., age) could yield clinically useful information. Yet as we state above, detecting patterns of biomarkers may ultimately lead to greater clinical value than identifying single biomarker candidates. As such, our study is focused on demonstrating the diagnostic potential of the combination of protein biomarkers, as assessed by MS, rather than on the generalizability of any specific protein biomarker. Related to that, we note that with a MS approach there are challenges in measuring low-abundance proteins. An illustrative example is MMP-9 which figures highly in many previous studies but lies below the measurable threshold of regular LC/MRM-MS. Other limitations include the absence of time since onset in the analysis; this shortcoming has been addressed in our larger TIA study in which patients are enrolled based on symptom onset within 24 h. Finally, we did not include intracerebral hemorrhage at this stage of the project.
Despite this pilot study’s limitations, discriminative power was high, making the pursuit of further plasma biomarker validation studies using MS worthwhile. Further, the results add weight to the argument for multi-protein panels and the use of how one technique, LC/MRM-MS, can be used to achieve larger data sets, laying a foundation for further research to demonstrate the utility of MS in translational settings at the bedside with benchtop MS machines in laboratories or as point of care devices, for example. However, despite its capacity for highly multiplexed and very specific determination of protein abundance, it is clear that biomarker assays will necessarily require translation from the LC/MRM-MS research platform onto much more rapid and clinic-ready platforms (e.g., iMALDI, ELISA). Sample turnaround times measured in hours and days from blood draw to test result are not appropriate in the acute care setting; turnaround times measured in minutes would be required. Assays with these characteristics are the focus of several ongoing follow-up projects in this area.
Where we measured proteins by multiple techniques, only 1 of 27 was found to be differentially expressed at the 5% significance level by more than one method. It is not clear which technique gives the more clinically relevant results. This highlights the question of analytical validity that needs to be considered when we apply these various proteomic techniques in clinical medicine. Where ELISA may suffer from lack of specificity being confounded by the presence of different isomers or other proteins sharing epitopes, it can be very sensitive; on the other hand, MS is highly specific, but may lose sensitivity due to protein modification during processing or the presence of a variant protein in an individual. It is recognized that some plasma protein levels fluctuate significantly in the general population due to heritable factors, individual and common environmental factors, and as yet unknown factors [
31]. Many potential stroke protein markers are low-abundance proteins (i.e., not easily or readily quantifiable) [
7], and their variability in the general population is not well known.
These limitations notwithstanding, we are confident that the findings from this preliminary study are a useful contribution to the published literature, providing more contexts for those of us trying to navigate in the world of protein markers in stroke. Our aim was to explore variability of protein markers within these two small patient groups, recognizing diversity within disease and across healthy populations and not to discover stroke markers.