Expert consensus has historically viewed sports-related concussion (SRC) as an injury resulting in clinical features that arise from transient disturbances in brain functioning occurring in the absence of structural brain injury (1). Clinical studies indicate that the majority of adolescents and adults who sustain a SRC achieve symptomatic and neurocognitive recovery within 1–4 weeks postinjury (1–3). However, approximately 10–30% of SRC patients will develop persistent postconcussion symptoms (4, 5) that are mediated by heterogeneous pathophysiological processes that remain poorly understood (6). While advanced neuroimaging studies have provided some insight into these pathophysiological processes, results to date have been largely limited to the detection of differences between groups of SRC patients and healthy control subjects (7, 8). As such, there remains an urgent unmet need for novel neuroimaging tools that can provide qualitative and quantitative assessment of these processes in individual SRC patients.

Among the processes implicated in the pathogenesis of acute concussion and traumatic brain injury (TBI) are disturbances in cerebrovascular physiology that result in a mismatch between the metabolic demands of the injured brain and the delivery of cerebral blood flow (CBF) (9–13). One of the means to assess the physiological capacity of the cerebral vasculature to respond appropriately during periods of metabolic stress or injury is cerebrovascular reactivity or responsiveness (CVR). CVR is defined as the change in CBF that occurs in response to a vasoactive stimulus (14, 15). Accurate measurement of CVR can be achieved by the administration of a precise, quantifiable, and reproducible vasodilatory stimulus while measuring CBF or its surrogate (16, 17). Using model-based prospective end-tidal targeting (MPET) of CO₂ during blood oxygenation level-dependent (BOLD) magnetic resonance imaging (MRI), we have previously demonstrated patient-specific qualitative and quantitative regional alterations in CVR among adolescent postconcussion syndrome patients while global mean resting CBF remained within normal limits (18). Although these results suggest that the cerebrovascular disturbances that characterize acute SRC can persist into the chronic phases of injury, available studies that have aimed to evaluate alterations in CVR using various techniques among acute and subacute SRC patients have yielded inconsistent results (19–21).

In this study, we examined group differences in global mean resting CBF and CVR in healthy control subjects and adolescent acute and subacute SRC patients. Using a leave-one-out receiver operating characteristic (ROC) analysis we examined individual differences in CVR among SRC patients compared to an institutional atlas of healthy adolescent control subjects assessed using the same neuroimaging assessment technique.

This study examined global mean resting CBF and global and regional CVR in adolescent SRC patients and healthy controls subjects. In a heterogeneous population of SRC patients who underwent neuroimaging assessment at various stages of acute and subacute injury, we observed significant group differences in CVR that occurred in the setting of normal global mean resting CBF. When compared to an atlas of healthy control subjects, we found that SRC patients demonstrated significant patient-specific alterations in CVR with a predominant pattern of increased CVR including those patients who were found to demonstrate exercise intolerance during graded aerobic treadmill testing and were diagnosed with autonomic/physiological PCD. Furthermore, a preliminary leave-one-out ROC analysis utilizing the quantitative biomarker of increased CVR response (voxels demonstrating an increased response to a controlled hypercapnic stimulus) generated from brain MRI CO₂ stress testing was found to reliably discriminate patients with and without SRC with an ROC AUC of 0.879 (p = 0.0001). At a cut-point of approximately 1,900 voxels for increased CVR derived from our second-level analysis at the 0.005 level, this biomarker resulted in a sensitivity of 87% and a specificity of 78% for correctly identifying patients with SRC. There was no discrimination seen for a decreased CVR response for these SRC patients (AUC = 0.468). Importantly, all neuroimaging studies were well tolerated with no complications.

At present, the natural history of CVR impairments following SRC remains poorly understood due in large part to the relative paucity of studies that have examined CVR during the acute, subacute, and chronic stages of injury. Among available studies, Len et al. (19) examined CVR in 21 healthy controls and 10 SRC patients within 7 days of injury using transcranial Doppler ultrasonography of the right middle cerebral artery (MCA) and a breath-holding and hyperventilation challenge. No significant differences in percentage change in MCA velocity were observed between the SRC and control groups during breath-holding; however, greater reductions in percent-grouped mean MCA velocity were observed among SRC patients during the hyperventilation challenge. Using the same technique, Len et al. (20) examined CVR in 20 SRC patients at 2, 4 and 8 days postinjury including 18 that underwent baseline preseason or postseason evaluations. During breath-holding mean MCA velocity was found to be increased on day 2 and decreased on days 4 and 8 with less robust findings observed during hyperventilation. Militana et al. (21) examined CVR in 11 healthy controls and six SRC patients 3–6 days postinjury by measuring alterations in BOLD MRI signal during an inhaled CO₂ challenge. In the SRC group, they observed increased CVR for all preselected regions of interests including significant increases within the median default mode network and the anterior cingulate. They also found that increased CVR in the hippocampus was positively correlated with increased functional connectivity between the hippocampus and precuneus in the SRC group but not in the healthy control group. The authors also found no group differences in resting mean CBF between groups similar to the results of our present study. Finally, in a previous study our group examined CVR in 17 healthy control subjects and 15 adolescent SRC patients who remained symptomatic during the chronic phase of injury (33–993 days postinjury) (18). Using MPET of CO₂ and BOLD MRI we observed significant group differences in CVR characterized by a predominant pattern of decreased CVR when group means were compared. More importantly when compared to the healthy controls on an individual basis, we found that each SRC patient demonstrated a unique signature of both increased and decreased regional CVR even in the setting of normal global mean resting CBF.

The differences in CVR findings observed across the available studies are undoubtedly impacted by the techniques used to measure CVR. To achieve accurate within- and between-subject measures of CVR requires a precise, quantifiable, and reproducible vasodilatory stimulus to be administered to all study participants during assessment of CBF. Although transcranial Doppler and BOLD MRI are commonly used techniques to assess cerebral blood velocity and CBF, respectively, the vasoactive stimulus chosen to provide the stress for the CVR study each have their own strengths and limitations. The most common techniques used to provide a vasoactive stimulus in CVR studies include breath-holding, inhaled CO₂, and MPET CO₂ targeting (13, 16). Breath-holding is one of the most widely used methods to induce hypercapnia due to ease of use and lack of requirement for specialized personnel or equipment. Unfortunately, breath-holding fails to result in a linear rise in the arterial partial pressure of carbon dioxide (PaCO₂) over time (27); the P_ETCO₂ detected at the end of breath-holding (the measured stimulus) is not always equal to the PaCO₂ (the stimulus acting on the cerebral vasculature); and P_ETCO₂ and P_ETO₂ levels cannot be sampled throughout the breath-holding periods (16). Together, these limitations can lead to significant differences in the vasodilatory stimuli administered across groups and subjects thereby impacting CVR measurement and contributing to experimental noise. This finding is illustrated in the study by Len et al. (19) discussed above, where mean P_ETCO₂ levels of 41.2 and 37.9 mm Hg among healthy controls and SRC patients, respectively, were obtained during breath-holding. An alternative method for providing a hypercapnic stimulus is inhaled CO₂. Although this technique relies on the administration of a gas containing a fixed fractional concentration of CO₂, the resultant effect on PaCO₂ has also been found to vary considerably between subjects and within the same subject over serial administrations (28, 29). Importantly, both breath-holding and inhaled CO₂ techniques can also lead to changes in PaO₂ that can result in independent and unaccounted for changes in CBF and CVR, especially with BOLD MRI approaches. In the study by Militana et al. (21) utilizing inhaled CO₂, resting P_ETCO₂ levels at baseline and during the CO₂ challenge were not reported. In contrast, the MPET CO₂ used here and in our previous studies (18, 22) has emerged as the only technique that is capable of providing a precise, quantifiable, and reproducible CO₂ stimulus under controlled isoxic conditions (16). This technique employs end-inspiratory rebreathing that ensures that the P_ETCO₂ is equivalent to the PaCO₂ (25) as well as continuous breath-by-breath sampling (30). By combining this vasodilatory stimulus with BOLD MRI and standardized data analysis techniques, brain MRI CO₂ stress testing results in highly reliable within (15, 31) and between subject measures of CVR that can be quantified on an individual subject basis (13, 17, 32). In this study, we observed no significant group differences in baseline P_ETCO₂ or P_ETCO₂ and P_ETO₂ during MPET targeting suggesting the results of brain MRI CO₂ stress testing obtained here are indeed a true reflection of abnormal CVR in these patients and are not due to variations in the vasoactive stimuli applied to these subjects.

Taken together, the available preliminary literature suggests that SRC is associated with patient-specific impairments in CVR that are detectable by transcranial Doppler and MRI-based techniques during acute injury but, to date, only by MRI-based techniques during the subacute and chronic phases of injury. Results from MRI-based studies suggest that SRC may be characterized by a predominant pattern of increased CVR among acute and subacute patients as well as those that develop autonomic/physiological PCD and a predominate pattern of decreased CVR among patients that develop more chronic symptoms. Interestingly, the results of the available MRI-based CVR studies do not provide supportive evidence of a correlation between the severity of postconcussion symptoms and the magnitude of CVR alterations detected among SRC patients at different stages of recovery. Most importantly, this study confirms that brain MRI CO₂ stress testing is well tolerated in patients with acute and subacute SRC and may be capable of quantifying the extent of CVR impairments in individual SRC patients.

The results of the present study must be considered in light of several important limitations. First, the sample size is relatively small and included adolescent patients who were recruited from a multidisciplinary pediatric concussion program. Patients evaluated at tertiary concussion programs are likely to have sustained more severe injuries that take longer to recover and therefore may be more likely to exhibit a greater impairment in CVR compared to a more generalized population of adolescent SRC patients. Second, the patient population examined in this study included those with a significant history of previous concussions. Although approximately 50% of patients that present to our multidisciplinary pediatric concussion program have a history of previous concussion (33), the lifetime burden of concussions among the SRC patients evaluated in this study may have had an impact on the magnitude of CVR alterations detected in this cohort. Indeed, preliminary longitudinal studies suggest that the cerebrovascular disturbances that accompany SRC may persist beyond symptomatic recovery (22, 34, 35). Third, the SRC patients in this study were imaged at non-uniform time points following injury, thus limiting direct comparisons between individual subjects. As such, future prospective longitudinal studies incorporating larger samples of adolescents and adults without a history of previous concussion who are imaged at uniform time points are needed to evaluate the impact of individual concussions on CVR alterations during the acute and clinically recovered stages of injury. Fourth, the incorporation of individualized masks during imaging postprocessing removed from analysis some of the prefrontal cortex and periventricular white matter, thus potentially resulting in an under-detection of abnormal voxels within these regions. Fifth, one of the most important limitations of this neuroimaging assessment technique is the effect of subject motion on accurate CBF and CVR measurements. It is well documented that both BOLD EPI and arterial spin-labeling techniques are highly susceptible to head motion (36, 37), which can have a confounding influence on BOLD and pCASL MRI CVR measurements. Although ongoing modifications to our brain MRI CO₂ stress testing technique, including a reduction in the magnitude of the CO₂ stimulus, have helped reduce the work of breathing and head movement during imaging acquisition, this limitation did result in the exclusion of a number of participants from the present study. Future studies should consider the use of head restraints and preemptive motion correction techniques (38, 39) as well as ramp CO₂ breathing sequences (40, 41) that may be more tolerable than the abrupt changes associated with square-wave sequences as used here. Progressive CO₂ breathing sequences will also allow help facilitate the measurement of dynamic cerebrovascular responses to hypercapnic stimuli (42). Sixth, although the leave-one-out ROC analysis in this study suggests that quantitative increases in CVR may be a promising biomarker to reliably distinguish between SRC patients and normal control subjects, quantitative voxel cut points will likely vary across future studies as a consequence of institutional and study differences in SRC patients, normal control subjects, and the vasoactive stimuli and neuroimaging platforms utilized. These methodological issues should be carefully considered when comparing results across previous and future CVR studies in this patient population. Finally, the neuroimaging assessment techniques utilized in this study only allowed for accurate and reliable measurement of global mean resting CBF and CVR and did not assess for other mechanisms that may contribute to the pathophysiology of TBI and patient symptoms including alterations in white matter integrity and resting state functional brain networks (7, 8) which may have accounted for the small false-positive and false-negative rates observed in this study and may have contributed to the lack of correlation observed between CVR alterations and symptom severity scores in the SRC patients. This later finding is not surprising; however, given the heterogeneity of pathophysiological processes (i.e., coexisting vestibulo-ocular and cervical spine dysfunction) (6) and non-injury patient factors (somatization, resilience) (43–45) that have been found to contribute to symptom severity following concussion and mTBI. Consequently, future studies combining brain MRI CO₂ stress testing with other advanced neuroimaging techniques such as quantitative susceptibility-weighted imaging, DTI, and resting state fMRI are needed to provide a more comprehensive assessment of the complex pathophysiological mechanisms mediating SRC and mTBI and examine the potential diagnostic and clinical utility of these multimodal neuroimaging tools in individual patient diagnosis and management.

In conclusion, this study provides evidence that acute and subacute SRC is associated with patient-specific alterations in CVR that occur in the presence of normal global mean resting CBF. This study suggests that with meticulous attention paid to end-tidal gas control and head motion, brain MRI CO₂ stress testing is capable of detecting and quantifying abnormal CVR among individual SRC patients over a wide range of injury acuities and postinjury time points. The results from our preliminary leave-one-out ROC analysis suggests that alterations in CVR detected by brain MRI CO₂ stress testing may be a useful qualitative and qualitative biomarker to help reliably identify individual subjects with SRC. Larger prospective studies are needed to evaluate the clinical utility of brain MRI CO₂ stress testing in the diagnosis, prognosis, and management of individual concussion and mild TBI patients.

Institutional research ethics approval was obtained for this study. Informed patient, and where applicable, parental consent was obtained for all participants prior to participating in the study. The approving body was the Biomedical Research Ethics Board of the University of Manitoba, Max Rady College of Medicine.

WM and ME: conception of study, conduct of study, analysis of data, interpretation of data, writing, and final approval. LR: conduct of study, analysis of data, interpretation of data, writing, and final approval. PM: conception of study, interpretation of data, writing, and final approval. MM and PP: conduct of study, writing, and final approval. ME: conduct of study, interpretation of data, writing, and final approval. DM, JD, and JF: interpretation of data, writing, and final approval.

JF, JD, and DM are senior scientists at Thornhill Research Inc. (TRI), a company affiliated with the University Health Network that developed the RespirAct™, a non-commercial research tool assembled by TRI to enable cerebrovascular reactivity studies. The other authors have no financial disclosures or conflicts of interest to declare.