The Brain Thermal Response as a Potential Neuroimaging Biomarker of Cerebrovascular Impairment

Discussion

The results support the presence of a BTR as a potential neuroimaging biomarker reflecting a hypothesized, convolved interaction of hemodynamic and metabolic tissue properties. The measurement of a BTR is feasible in routine clinical MR imaging implementations, with significant relationships identified between both BTR and baseline brain temperature relative to CVR, supporting further exploration of BTR as a fully noninvasive and dynamic biomarker. The use of voxelwise measurements facilitated a nuanced description of local hemodynamics and temperature, allowing contemporaneous, multiparametric classification of the diseased hemisphere, all potentially achievable in a single imaging session. MR chemical shift thermometry, exploiting the temperature-dependent difference between water and the methyl proton resonance of NAA, was chosen to approximate temperature using our previously reported protocol for phantom as well as human and nonhuman primate in vivo thermometry.^19,24,25,33 The present study takes advantage of the unique pathophysiologic attributes of chronic cerebrovascular ischemia, in a cohort allowing multiphasic interrogation of a flow-temperature relationship, which may not be pragmatic in the clinical acute stroke setting. Ishigaki et al³⁶ previously reported the use of PET combined with localized, single-voxel MR spectroscopy–based thermometry to explore the relationship of brain temperature in steno-occlusive disease, observing positive correlations between the brain temperature difference (ipsilateral-contralateral hemisphere) with both CBV and OEF. A limited number of additional studies using single-voxel MR spectroscopy thermometry have observed higher temperatures in ischemic brain tissues, further supporting the investigation of brain temperature in the development of a pathophysiologic framework defining cerebral energetics.^{16,36⇓–38}

Our analysis of the relationship between BTR and baseline brain temperature with CVR was performed voxelwise to examine whole-brain effects in patients with steno-occlusive disease. CVR and flow augmentation are generally hypothesized to reduce brain temperature and might, therefore, have been expected to produce a negative BTR. Our results were, however, more complex, and the observed flow-temperature relationship may not be immediately intuitive but, on deeper examination, can be reconciled with prevailing theories regarding cerebral blood flow, temperature, oxygen metabolism, and BOLD signal evolution. Existing models of cerebral temperature homeostasis propose a dynamic modulation of heat-producing mechanisms (oxygen cleavage from hemoglobin, glycolysis, and generation of metabolic by-products) and heat dissipation through the inflow of cooler systemic blood with lesser contributions from conduction to the brain surface. This is predicated on the notion of heat sink/radiator effects of inflowing blood, cooling tissues through the removal of metabolic waste.

We would also anticipate relative heating among those tissues with insufficient blood flow to meet metabolic demands. Conceptually, a linear or at least monotonically changing thermal response as a function of flow might be expected. However, several factors could interfere with this conceptual framework: 1) the temperature-flow relationship may vary radially within the brain, whereby superficial structures closer to the surface CSF are cooler than arterial inflow, thus warming with increasing flow; and 2) oxygen extraction, a primary contributor to the production of thermal waste, may be variably upregulated depending on the depth of misery perfusion, ostensibly at its highest among tissues with the most impaired flow. However, as demonstrated in several past studies, the capacity for flow augmentation may remain even among tissues with OEF elevation. The downregulation of OEF among such tissues during hemodynamic augmentation may occur on a time scale shorter than the duration of the BOLD scan and therefore shorter than the separation in time between CBF or thermometry imaging. This may, in part, drive a component of time shifts observed in BOLD signal augmentation if, for instance, a very high OEF must be overcome before BOLD augmentation can be detectable (ie, for deoxyhemoglobin dilution to occur sufficiently).³⁹ Accordingly, CVR augmentation following ACZ was initially anticipated to produce negative BTR values as a result of the increased inflow of cooler arterial blood, acting as a heat sink for warmer cerebral temperatures and removing metabolic by-products, including thermal waste.

Such interactions could underlie the observed quadratic relationship between BTR with CVR_BOLD, wherein a negative BTR-CVR relationship was present at a low CVR but a positive trend was observed for CVR_BOLD augmentation upward of approximately 4%. The far left of the curve follows the expected relationship, with decreasing (ie, more negative) BTR accompanied by increasing CVR augmentation and, indirectly, CBF. The source of the observed positive correlation between BTR and CVR above 4% remains inconclusive within the limitations of this exploratory analysis, in the absence of concurrent OEF and CMRO₂ interrogation (see below).

An opposing relationship (positive-versus-negative quadratic curve) was observed between baseline brain temperature and CVR_BOLD, suggesting that resting brain temperature alone may not fully explain the dynamics of the brain temperature response to a hemodynamic maneuver. This difference may, in part, reflect that baseline brain temperature is inherently a known indicator of diseased tissue energetics, with higher temperature associated positively with hypoperfusion and ischemia.^{16,19,36⇓–38} The benefit of BTR is the ability to quantify a maneuver response, thus producing a form of physiologic contrast with the potential for characterizing not only static tissue impairment, as with baseline brain temperatures, but also the dynamics of the hemodynamic response in a manner similar to that of CVR. The overall quadratic relationships suggest the presence of multiple regimes of temperature regulation or dysregulation across the brain, determined by additional hemodynamic parameters such as OEF and CMRO₂.

In addition to the relationships observed among BTR, baseline temperature, and CVR_BOLD for all voxels, significant associations were also observed for the CVR_BOLD values thresholded to 70% of the contralateral hemispheres, demonstrating a robust interaction between BTR and CVR_BOLD. The use of thresholding to explore voxelwise relationships enabled discrimination between impaired and healthy voxels, given the absence of a healthy control cohort undergoing the MR imaging ACZ protocol for comparison in this study.

Using CVR_ASL to study the same relationships resulted in a single significant correlation for BTR with CVR_ASL of ≥30%. Inherent challenges to ASL perfusion in highly delayed or low-flow environments, such as expected even in compensated territories downstream from steno-occlusive disease, may impact its use in this context, even when using multi-TI approaches to the extent governed by tag decay. We suspect that several, potentially divergent errors in ASL quantitation in such patients may have confounded the BTR relationship with CVR_ASL, among them: 1) exaggerated hypoperfusion if tagged spins are undelivered to the vascular network at the selected TI; 2) exaggerated hyperperfusion if tagged spins are delivered but linger in the arterial network overlying the tissue of interest (particularly when crusher gradients are not used, or when partial volume effects preclude reliable segmentation of the cortical tissues from such overlying vessels, both of which exist in our protocol); or 3) exaggerated hypoperfusion due to excessive tag decay, which may affect our measurements, particularly at the longest TI of 3000 ms. An additional circumstance of negative CVR_ASL, which we have observed to occur despite preserved BOLD and dynamic susceptibility contrast augmentation, exists in tissues with large delays. Arterial transit artifacts at baseline that spuriously elevate perfusion are somewhat cleared following ACZ but appear not to increase the apparent tissue perfusion to the same extent that the arterial transit artifacts previously had, therefore leading to apparently negative CVR. While exploratory, we suspect that temperature dynamics, together with flow, blood volume, and oxygen use, may be better subsumed by CVR_BOLD, albeit in an entangled manner requiring deeper, multiparametric study. Taken together, the observed trends imply the presence of at least 2 distinct regimes for BTR as a function of CVR augmentation: 1) decreasing BTR in response to increasing CVR (negative relationship) as predicted from existing models of brain temperature regulation, which we hypothesize may be dominated by heat-dissipating CBF augmentation over metabolic heat production, with decreasing temperatures in response to ACZ^13,14; and 2) a positive BTR-CVR regime in which increasing CVR augmentation is accompanied by increasing, or more positive, BTR values.

In line with previous work examining hemispheric differences in the study of hemodynamics and brain temperature,³⁶ we compared BTR and baseline brain temperature across hemispheres. Our analysis, however, did not reveal differences in baseline brain temperature and BTR between the ipsilateral and contralateral hemispheres such as those previously observed following ischemic injury.^16,36 However, the lack of hemispheric temperature differences is not surprising because even in cases of unilateral disease, consequential hemodynamic impairment may not be limited to the affected hemisphere and compromised function and tissue damage may exist bilaterally and heterogeneously.³ Consequently, temperature differences resulting from impaired cerebral perfusion and impaired autoregulation may be averaged out in coarse hemispheric comparisons. Nevertheless, the presence of significant random-effects differences in BTR and baseline brain temperature between hemispheres could suggest that in a larger sample size, a difference may exist.

We acknowledge several study limitations. While it was designed to test our hypothesis regarding the association between brain temperature changes and hemodynamic augmentation, the lengthy, multiparametric protocol necessitated a 2-day study, which limited the sample size. The findings nevertheless emphasize the feasibility of dynamic brain thermometry in this setting and compel more rigorous study of BTR as a prognostic biomarker. A challenge in the analysis was the observation of negative CVR values, suggesting the possibility of vascular steal and limitations in the measurement of CBF.^10,29 As discussed above, several possible mechanisms may contribute to negative CVR values, particularly arterial transit artifacts, collateral rather than anterograde flow after ACZ, and the time scale of ASL-based measures of CVR. These may also contribute to spuriously high CVR values. As a result, continuous BOLD measurements may be a more reliable surrogate for the hemodynamic response to ACZ. The temporal dynamics of the ACZ response are worth considering. To mitigate potential errors related to a rapidly changing baseline, the initial BOLD signal used for CVR_BOLD in our study was derived from a pre-ACZ phase of the examination spanning the initial 5 minutes of BOLD acquisition. A uniform protocol was prescribed in which a slow infusion of ACZ was then administered for 3–5 minutes without patient repositioning or scan interruption, followed by ∼10 more minutes of uninterrupted BOLD signal acquisition. It was recently reported that CVR_BOLD increases continuously following ACZ administration, reaching a plateau at approximately 8.5 minutes, which continues through the remainder of the acquisition.⁴⁰ While CVR delays beyond 20 minutes are difficult to exclude, we expect that the window of maximal augmentation is likely captured within this study design.

While we cannot conclusively parcellate the competing hemodynamic effects from the technical challenges, further studies with MR-based methods to characterize functional activity, CBF, and metabolism in cerebrovascular disease would complement our results. While direct estimation of OEF and CMRO₂ with MR-based techniques remains challenging, ongoing efforts to optimize MR oximetry measurements will enable further studies into the relationship of BTR with these metabolic parameters.⁸ A more thorough description could be facilitated through study of a healthy control population; however, administration of intravenous vasodilatory stimuli is not common practice in our experience. Other approaches such as induced hypercapnia or breath-hold may be valuable in this regard, though the mechanism, depth, and magnitude of flow augmentation may differ. Last, the relatively small sample size and limited longitudinal follow-up information in this cohort of patients limit our characterization of BTR as a prognostic tool to predict future stroke risk in such patients. Further study in a larger and, ideally, prospectively collected cohort is warranted to better assess the prognostic power of BTR in a clinical population.