Cerebral blood flow is associated with matrix metalloproteinase levels during the early symptomatic phase of concussion

Concussion is associated with disrupted cerebral blood flow (CBF), although there appears to be substantial inter-individual variability in CBF response. At present, the mechanisms of variable CBF response remain incompletely understood, but one potential contributor is matrix metalloproteinase (MMP) expression. In more severe forms of acquired brain injury, MMP up-regulation contributes to CBF impairments via increased blood-brain barrier permeability. A similar relationship is hypothesized for concussion, where recently concussed individuals with higher MMP levels have lower CBF. To test this hypothesis, 35 concussed athletes were assessed longitudinally at early symptomatic injury (median: 5 days post-injury) and at medical clearance (median: 24 days post-injury), along with 71 athletic controls. For all athletes, plasma MMPs were measured and arterial spin labelling was used to measure CBF. Consistent with our hypothesis, higher concentrations of MMP-2 and MMP-3 were correlated with lower global CBF. The correlations between MMPs and global CBF were also significantly diminished for concussed athletes at medical clearance and for athletic controls. These results indicate an inverse relationship between plasma MMP levels and CBF that is specific to the symptomatic phase of concussion. Analyses of regional CBF further showed that correlations with MMP levels exhibited some spatial specificity, with greatest effects in occipital, parietal and temporal lobes. These findings provide new insights into the mechanisms of post-concussion cerebrovascular dysfunction.


Introduction
Concussion is a form of mild traumatic brain injury (TBI) that is associated with complex disturbances in brain physiology. Cerebral blood flow (CBF) has been identified as being particularly sensitive to the effects of concussion [1]. As tight control of CBF is needed to meet timevarying physiologic and neurometabolic demands, its dysregulation can have major a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 consequences, including more severe post-concussive symptoms and prolonged recovery [2][3][4]. While more severe TBI is associated with major reductions in CBF at early injury [5], the effects of concussion are more variable, with studies variously reporting mean increases in CBF relative to controls [4], mean reductions in CBF [6] and non-significant differences [7] within the first week of injury. These findings suggest substantial inter-individual variability in CBF response after concussive injury. Despite its key role in concussion, the mechanisms underlying heterogeneous CBF disturbances remain incompletely understood.
One potential mechanism of altered CBF is through the up-regulation of matrix metalloproteinases (MMPs). MMPs are a family of intercellular calcium-dependent zinc-endopeptidases that play key roles in cell migration, tissue development, morphogenesis and repair [8,9]. Under pathological conditions, such as acquired brain injury, the elevated expression of these extracellular matrix-degrading proteinases may have negative physiological effects [10]. In rodent models, MMP expression tends to be increased acutely as part of the "traumatic cascade" and it is associated with increased permeability of the blood-brain barrier (BBB), due to the breakdown of basal lamina proteins and the tight junctions between vascular endothelial cells [11][12][13][14][15]. Secondary effects of increased BBB permeability include cerebral leukocyte infiltration, vasogenic edema [16,17] and an enhanced neuroinflammatory response [18,19]. Collectively, the processes initiated by MMP up-regulation after TBI have a detrimental effect on cerebrovascular function and may similarly contribute to CBF disturbances after a concussion. To date, however, the relationship between CBF and MMP expression has been largely unexamined in humans and for milder forms of TBI.
The present study examined associations between global CBF and peripheral MMP values for a cohort of athletes with sport-related concussion, evaluated in the early symptomatic phase of injury and at medical clearance to return to play (RTP), along with a cohort of athletic controls without recent injury. For both groups, a panel of plasma MMPs were measured, and magnetic resonance imaging (MRI) was used to assess resting CBF via arterial spin labelling (ASL). This imaging technique uses magnetically-labelled arterial blood water as an endogenous tracer that tracks the delivery of oxygenated blood to brain tissues. The primary study hypothesis was that recently concussed individuals with higher MMP values would have lower global CBF values (i.e., negative inter-subject correlation between these measures). It was further hypothesized that this relationship would be specific to symptomatic injury, with diminished effects among concussed athletes at RTP and among athletic controls without recent injury. If confirmed, these results would provide evidence that the variable up-regulation of MMPs is a potential mechanism contributing to heterogeneous post-concussion CBF response.

Study participants
Thirty-five (35) athletes were recruited consecutively from university-level sports teams at a single institution through the sports medicine clinic, following a diagnosis of sport-related concussion. Diagnosis was determined by staff physician following sustained direct or indirect contact to the head, with assessment of clinical features as per Concussion in Sport Group guidelines [20] and neurologic assessment including examination of cranial nerves, gait, balance and gross motor function. Imaging was conducted within the first week of injury (median and interquartile range (IQR): 5 [2,7] days) and at RTP. There was some data loss, as imaging and blood could not always be acquired concurrently. The number of participants with both imaging and blood was 27/35 at symptomatic injury and 27/35 at RTP, with 19/35 participants having data in both sessions (54% overlap). Seventy-one (71) athletes without recent concussion were also recruited consecutively at the start of their competitive season.
All athletes participating in the varsity program had completed baseline clinical assessments with the Sport Concussion Assessment Tool (SCAT, version 3 or 5) [21,22] before the beginning of their season and all concussed athletes had follow-up SCAT assessments at early symptomatic injury and at RTP. None of the athletes had a history of neurological or psychiatric diseases or sensory/motor impairments. Recruitment and data collection were carried out between June 2015 and October 2017 and the study was conducted in accordance with the Canadian Tri-Council Policy Statement 2 and approval of University of Toronto and St. Michael's Hospital research ethics boards, with participants giving free and written informed consent. The datasets analyzed for this study can be found in the figshare repository at https://figshare.com/s/d879d6bfb929808835ad.

Matrix metalloproteinase data
For both groups, plasma MMPs were measured, including MMP-1 (collagenase), MMP-2 and MMP-9 (gelatinases), MMP-3 and MMP-10 (stromelysins). This panel includes MMPs primarily localized to the endothelium and identified as relevant to vascular injury and TBI pathogenesis [23,24]. Blood was drawn proximal to neuroimaging for concussed athletes (1 [0, 3] days from imaging) and controls (1 [1,7] days from imaging) by standard venipuncture into a 10-mL K 2 EDTA tube, equilibrated at room temperature for one hour, and centrifuged for two minutes using a PlasmaPrep 12 TM centrifuge (Separation Technology Inc., FL, USA). After centrifugation, plasma was aliquoted and stored at -80˚C until analysis. Blood draws were not performed on subjects who were knowingly symptomatic with a viral or bacterial infection, seasonal allergies, or on any medication other than birth control at the time of venipuncture. The MMPs were analyzed by immunoassay on the Meso Scale Discovery (MSD) sector imager 6000, using MSD 96-well MULTI-SPOT1 technology (MSD1, Gaithersburg, MD, USA). MMP-1, -3, and -9, were assayed using the Human MMP 3-Plex Ultra-Sensitive kit, while MMP-2 and -10 were assayed using the Human MMP 2-Plex Ultra-Sensitive kit. All assays were run according to manufacturer's instructions, with individual samples run in duplicate.
Samples were not used for analysis if they fell outside of the range of detection provided by the manufacturer, or if the coefficient of variability (CV) between duplicate samples was greater than 25%. Given these criteria, a single value was removed for MMP-9, with remaining assays providing 100% useable data. The average CV was below 5% for all markers: MMP-1 = 4.3%, MMP-2 = 3.9%, MMP-3 = 4.2%, MMP-9 = 3.6%, MMP-10 = 3.6%. To control for the presence of heavy distribution tails that vary by biomarker (skewness: MMP-1 = 1.
Brain maps of absolute resting CBF in grey matter were obtained for each participant using a 2D pulsed ASL imaging sequence (PICORE QUIPSS II; TE/TR = 12/2500 ms, TI1/TI1s/ TI2 = 700/1600/1800 ms, θ = 90 o , 14 oblique-axial slices with FOV = 256x256 mm, 64x64 matrix, 8.0 mm slice thickness with 2.0 mm gap, 4.0x4.0 mm in-plane, BW = 2368 Hz/px). For this sequence, 45 tag-control image pairs were obtained, where taking the difference in signal intensity between image pairs generates a perfusion-weighted image. A single calibration image M 0 was also acquired, in order to rescale signal intensities into absolute units of flow. To control for noise and artifact that may confound CBF estimates, the data were processed after acquisition, using Analysis of Functional Neuroimages (AFNI; afni.nimh.nih.gov) software and customized in-house algorithms. Rigid-body correction of between-scan head movements was performed using AFNI 3dvolreg to align tag and control scans to the calibration image M 0 . To control for signal spikes, e.g., due to head motion, physiology and scanner noise, filtering of outlier tag-control pairs was performed using the protocol of Tan et al. [25], followed by smoothing the scans with AFNI 3dmerge to reduce spatial noise, using a 3D Gaussian kernel with 6 mm isotropic full width at half-maximum. Voxel-wise estimates of CBF were then obtained by taking the mean of the signal difference between all tag-control pairs, rescaled into flow units of mL/100g/min based on M 0 values and established kinetic modelling parameters [7]. To compare regional CBF values between study participants, the brain maps were afterwards transformed into a common anatomical space, using the Montreal Neurological Institute MNI152 template as a reference. This was achieved using the FMRIB Software Library (FSL; https://fsl.fmrib.ox.ac.uk) and in-house scripts. For each participant, FSL flirt first computed the rigid-body alignment of their mean ASL volume to their T1-weighted image, along with the affine alignment of their T1 image to the MNI152 template. The net transform of ASL data into MNI space was then calculated with xfm_convert and applied to the CBF maps, resampled at 3 mm isotropic resolution. The study focused on cortical grey matter, to ensure a high signal-to-noise ratio with minimal partial volume effects when analyzing associations with MMPs. This was achieved using the Brainnetome Atlas (BNA V1.0; https://atlas. brainnetome.org) by constructing a binary mask that included all voxels with parcel labels 1 through 210. The CBF values were subsequently analyzed for all voxels in this mask.

Analysis of clinical and demographic data
Participant demographics are listed in Table 1, including age, sex and concussion history. Clinical scores are also reported for SCAT symptoms. A symptom severity score was obtained by summing across a 22-item symptom scale, with each item receiving a 7-point Likert scale rating. A total symptoms score was also obtained by counting all symptoms with non-zero ratings. For symptom scores, non-parametric Wilcoxon tests (1-tail) determined whether postconcussion values were significantly higher than controls. Paired-measures Wilcoxon tests (1-tail) also evaluated whether post-injury values were significantly higher than pre-injury. For all sets of statistical tests, significance was determined after adjusting for multiple comparisons at a False Discovery Rate (FDR) of 0.05.

Analysis of MMPs and CBF
To characterize the athlete cohorts, the mean MMP and global CBF values (i.e., averaged over all grey matter voxels) were calculated for controls and for concussed athletes at symptomatic injury and RTP, along with the bootstrapped 95% confidence intervals of the mean (95%CIs), based on 1,000 bootstrap samples. The mean differences between groups were also reported at each imaging session, along with 95%CIs and p-values, obtained from two-sample bootstrap analyses that compared the mean values of concussed and control groups. To test the main study hypothesis, Spearman correlations between MMP values and global CBF were calculated for controls and concussed athletes at symptomatic injury and at RTP, with bootstrapped 95% CIs and p-values, and significant associations were identified at an FDR of 0.05. For significant MMPs and sessions, the difference in Spearman correlation between concussed and control groups Δ ρ = ρ concussedρ control was also reported, with 95%CIs and p-values obtained from two-sample bootstrap analyses. Supplemental analyses examined whether the MMP-CBF correlations were confounded by demographic factors, including age, sex, prior concussion history and time post-injury. This was done by computing the partial Spearman correlation, adjusted for each covariate in turn, and testing whether there were significant changes in correlation strength, based on bootstrapped p-values. For all of the above analyses, missing data were handled using a pairwise deletion approach.
Additional analyses measured the effects of MMPs on regional CBF values. For the set of MMPs showing significant associations with global CBF, a single composite MMP score was generated by z-scoring and summing these variables. Spearman correlations were calculated between the composite MMP score and CBF for each grey matter voxel, with bootstrapped pvalues. Significant brain regions were identified by thresholding at a voxel-wise p = 0.005, followed by cluster-size thresholding at an adjusted p = 0.05, using AFNI 3dFWHMx to estimate spatial smoothness of the CBF maps and running 3dClustSim to obtain the minimum cluster size threshold. This procedure obtained a "consensus map" of the brain regions where MMPs affected regional CBF. The relationship was further quantified by calculating mean CBF values within significant voxels, and regressing mean CBF onto significant MMPs, with reporting of the regression coefficient b, coefficient of determination R 2 and bootstrapped 95%CIs. Regression diagnostic checks were also conducted, including plotting residuals against fitted values to assess linearity and homoscedasticity, quantile-quantile plots to assess normality of residuals and Cook's distance to test for high-leverage outliers, none of which showed evidence of substantial deviations from modelling assumptions. Supplemental analyses examined whether taking the mean CBF over all significant clusters provided a good representative summary of cluster-specific relationships between CBF and MMP levels. For each cluster of contiguous voxels k, the mean CBF values were taken and separately regressed against MMP values, providing cluster-specific coefficient b k . The difference score Δ k = b k −b was then obtained, measuring deviation of the cluster-specific model from the all-clusters model. Bootstrap resampling estimated 95%CIs and p-values of these differences, with significantly different clusters identified at an FDR of 0.05. For cluster-specific models deviating from the all-clusters model, the difference statistics were then reported.

Results
Participant demographics are reported in Table 1 for control and concussed groups. Both cohorts are comparable in age range, proportions of male and female athletes and prior history of concussion. Concussed athletes had a median of 3-4 weeks from injury to medical clearance, which included completion of a graded exercise protocol and cognitive testing, although there was substantial variability in individual recovery times. Concussed athletes had significantly elevated total symptoms and symptom severity at early symptomatic injury, relative to their own baseline and the control cohort (p<0.001, all tests) at an FDR of 0.05. Symptom scores at RTP were no longer significantly elevated (p�0.991, all tests), but tended to be slightly lower than baseline. Table 2 summarizes the distributions of MMP and global CBF values for the different groups. For all parameters, the control and concussed groups had similar group means and heavily overlapped 95%CIs. Moreover, the mean differences between groups were small relative to the group means themselves, with 95%CIs that overlapped zero. None of the The mean MMP and CBF values are reported for athletic controls and for concussed athletes at early symptomatic injury (SYM) and return to play (RTP), along with bootstrapped 95%CIs of the mean. The mean differences between concussed and control groups and 95%CIs are also reported for each post-concussion imaging session. None of the comparisons between concussed and control groups attained a value of p<0.05, uncorrected.  Table 3. Significant effects were seen for extensive clusters within occipital, parietal and temporal Thus, associations between MMPs and CBF exhibit some spatial specificity, with greatest effects confined to posterior and dorsal brain regions. Supplemental analyses of cluster-specific CBF found no significant differences in regression coefficient values, compared to the all-cluster results. One cluster (cluster #8 in Table 2, right inferior frontal area) had a slightly higher coefficient value when regressing CBF against MMP-2 (p = 0.016) with a coefficient difference Δ k = 3.65x10 -4 (95%CI: 0.96x10 -4 to 5.75x10 -4 ) but it was non-significant at an FDR of 0.05. All other clusters showed minimal  Table 3. Cluster report for correlations between matrix metalloproteinase (MMP) concentrations and regional cerebral blood flow (CBF). Cluster centers of mass are given in MNI coordinates and the brain regions are identified based on the nearest labelled grey matter region in the automated anatomical labelling (AAL) atlas.

Discussion
CBF disturbances are a key component of concussion pathophysiology, however, the underlying mechanisms are incompletely understood in humans. The present study examined whether there was an inverse relationship between peripheral MMPs and global CBF among individuals with sport-related concussion. Consistent with the main study hypothesis, MMPs were negatively correlated with global CBF for recently concussed athletes. The correlations were no longer significant at RTP or for athletic controls, further indicating that the relationship is limited to the early symptomatic phase of injury. The correlations were seen in the absence of group-level differences in mean MMP and global CBF values between concussed athletes and controls, which is consistent with our understanding of concussion as a diffuse, heterogeneous form of brain injury [7,26]. Previous studies of sport-related concussion have often found spatially limited and study-dependent CBF effects, including increases [4], decreases [6] and non-significant differences [7] relative to controls. The variable CBF response may be due to multiple factors, including differences in time post-injury [7] and in the extent of microvascular injury [19], with the latter interpretation supported by the present MMP-related findings. Significant negative correlations between global CBF and MMP levels are consistent with the main study hypothesis that MMP expression contributes to reduced CBF post-injury. Interestingly, the effects were not present for all measured MMPs, nor were they confined to a single subtype. MMP-2 (gelatinase) and MMP-3 (stromelysin) showed significant associations with CBF, whereas MMP-1 (collagenase), MMP-9 (gelatinase) and MMP-10 (stromelysin) did not. The identification of MMP-2 is consistent with studies of ischemic injury showing its involvement at acute injury [12,27,28]. Conversely, although MMP-9 has been linked with traumatic injury in rodents and is associated with greater lesion volumes and motor deficits [15,29,30], it did not show significant associations in this study. The mechanisms by which increased MMP expression may affect CBF are multifactorial, as MMPs are associated with injury to cerebral vasculature [15] and with BBB permeability, where the subsequent release of autocoids and free radicals can impair autoregulation [31]. In terms of secondary mechanisms, injury is followed by metabolic perturbations including ischemia, hypoxia, and vasospasm, all of which may perpetuate BBB dysfunction and edema [32,33]. In addition, the infiltration of peripheral immune cells across the dysfunctional BBB may contribute to CBF impairments, via ongoing cellular damage and neuroinflammatory response [34,35].
It is interesting to note diminished correlation strength between MMPs and CBF at medical clearance to RTP, which was a median of 3-4 weeks post-injury. This suggests that the underlying effects of MMPs on CBF are transient in nature, unlike some mechanisms that show persistent effects on cerebral perfusion, such as inflammatory cytokines [36,37]. These findings are also consistent with previous studies showing dissipation of concussion-related disturbance in CBF following RTP [4,6]. However, there are also potentially chronic effects of concussion that emerge at longer time intervals post-injury. For example, previous studies of concussion in this cohort found delayed changes in frontal CBF emerging over a year post-RTP that are potentially linked to grey matter volume loss [4,38]. Moreover, the stroke literature suggests that elevated MMPs may confer beneficial effects chronically [39]. It may therefore be important to further examine the evolution of MMPs and CBF over longer postconcussion time intervals.
Analyses correlating the MMPs with regional CBF show broadly affected territories in the brain, as expected. The most consistently affected regions are predominantly occipital, parietal and temporal. These are brain areas in which resting CBF also tends to be high, suggesting that they may be sensitive targets for investigating associations with MMP expression. However, it is also possible that this spatial distribution reflects brain regions experiencing the greatest MMP-mediated dysregulation of CBF. These regions are implicated in diverse processes including visuo-spatial processing and memory, which frequently show symptom impairments following concussion [22,40]. Future research should therefore consider the potential links between MMP expression and specific post-concussion impairments.
This study fills a knowledge gap, but there are some limitations that should be acknowledged. One consideration is the post-injury time interval, as initial imaging and blood draws were conducted a median of 5 days post-injury. This enables comparisons with other imaging studies of concussion that are often on a similar timeline, but it is beyond the 24 hour window in which MMP expression is most pronounced [41,42]. Moderately strong MMP-CBF correlations were identified, but given the rapidly evolving post-concussion neurometabolic cascade [43], these relationships may differ at acute injury. This must be verified before the present findings can be applied to emergency care settings. Another important consideration is demographic variability. Although the study found no significant effect of adjusting for age, sex or concussion history, there may be other factors that contribute to modelling error. For example, athletes with higher training loads and more frequent subconcussive blows may show greater CBF disruptions due to increased BBB permeability [44,45], the consequences of which should be further investigated. Lastly, the detailed mechanisms linking MMP expression to CBF are not yet fully elucidated. At present, further investigation is required to directly assess the effects of proteolytic activity on cerebrovascular integrity, edema and neuroinflammatory response. The assessment of MMPs in peripheral blood also only indirectly reflects MMP levels in the CNS. Previous literature has established that cytosolic and membrane-bound proteins are released into the CSF and interstitial fluid after injury. From there, they can pass across the disrupted BBB or into the paravenous space, ultimately reaching the peripheral circulation [11,[46][47][48]. This provides a plausible mechanism relating peripheral MMP levels to brain injury. Nevertheless, the indirect nature of MMP measurements is a potential confound, as systemic vasculature is also a source of MMPs [49,50], and MMP levels may be increased by, for example, peripheral nerve injury [51]. It is critical to validate the study findings, using animal models of injury or, alternatively, human studies that employ ventriculostomies to directly sample the CSF [52].
In sum, this study demonstrates that greater expression of peripheral MMP-2 and MMP-3 are correlated with reduced global CBF during the early symptomatic phase of concussive injury. These findings provide evidence for MMP expression as a potential contributor to the variable CBF response among concussed individuals. Although further research is needed to validate these findings, it is a critical first step that establishes a need for further research into the specific pathways by which MMPs and its natural inhibitors impact CBF regulation. Such research may lead to new insights into the mechanisms of CBF dysregulation and subsequent recovery, which are needed to improve clinical management and to develop targeted interventions, particularly for patients experiencing more severe or prolonged post-concussion impairments.