Magnetic resonance imaging metrics of oxygen extraction fraction: Contradictions or insight into pathophysiological mechanisms?

In this issue of the American Journal of Hematology, Fields et al. report cerebral oxygen extraction fraction (OEF) measurements using an Asymmetric Spin Echo Magnetic Resonance Imaging (ASE-MRI) technique for a large cohort ($N=120$) of children including healthy controls (34), anemic controls without sickle cell disease (27), and those with sickle cell anemia (SCA) (59).¹ They found that average OEF in anemic controls was significantly higher than in healthy controls, and that OEF is higher still in SCA. In addition, they found that the volume of brain with elevated OEF was larger for SCA than for anemic controls. In both cohorts, the largest OEF increases were in the white matter and included regions where infarcts were identified by fluid-attenuated inversion recovery (FLAIR) MRI.

The observations of increased OEF by Fields et al. in SCA contradict the decreased OEF recently reported by Vu et al. Vu et al. used T2-relaxation under spin tagging MRI (TRUST-MRI) to measure OEF in a large cohort ($N=118$) of young adults including healthy controls (44), anemic controls (27), and individuals with SCA (47).² This study also found that OEF in anemic controls differed from healthy controls and SCA had even larger OEF differences, however, the differences were in the opposite direction with SCA having the lowest OEF. This discrepancy could lead readers to dismiss both methods, not knowing which is correct, but once the details of each measurement technique are considered a reconciliation between these opposite observations is possible.

All MRI-based techniques for quantifying OEF rely on the different magnetic properties of oxy- and deoxy-hemoglobin. There are two main families of techniques that quantify cerebral OEF, those that provide estimates of OEF on a voxelwise basis from signals that arise from the local tissue, and those that quantify whole-brain OEF by observing large draining veins most often the superior sagittal sinus (SSS).³ The techniques from the contradictory papers come from these two different families.

ASE-MRI is a tissue-based method used to detect the extravascular effects of deoxyhemoglobin contained in microvessels within the brain, including cortex, white matter, and deep gray structures. Microvessels containing deoxygenated blood create local field variations that affect MR signals from the nearby extravascular tissue water protons. Simplifying assumptions about microvasculature structure yield a signal model that can be used to estimate voxelwise OEF.⁴ Key model assumptions are that signals originate only from the vicinity of small deoxygenated microvessels (vessels with radii smaller than the imaging voxel dimension), not from larger vessels where intravascular signal components predominate, and that these microvessels are randomly oriented with respect to the main magnetic field. It must be emphasized that the estimated OEF from a single voxel originates from the magnetic field variations in the volume of tissue around deoxygenated microvessels. This means that OEF from this technique, even when talking about a single voxel, is a spatial average. Whole-brain OEF can be calculated by taking the average of the voxelwise estimates.

TRUST-MRI is a large-vessel method used to estimate whole-brain OEF from a measurement of intravascular blood transverse relaxation time (T2) in the SSS. Spin tagging is used to isolate a pure-blood signal in the SSS for the determination of T2, and then T2 is converted to SO₂ based on empirical and biophysical models. These models have been gradually refined to account for individual HbS and HbA concentrations, which has improved the accuracy of TRUST for subjects with SCA.⁵ It must be emphasized that TRUST-MRI estimates of OEF are flow averaged, meaning the blood mixture in the SSS is determined by the various volume outflows from each vascular territory drained.² It is also important to note that although often described as a “whole brain” estimate, the SSS primarily drains cortex, with white matter draining primarily to the deep venous system. Thus, this is an estimate of cortical OEF where vascular territories contributing to this measure are primarily the superior frontal, parietal, and occipital cortex.

Observations by Fields et al. using the ASE-MRI method found increased OEF for SCA patients. In contrast, the most recent and technically refined observations using TRUST-MRI found decreased OEF.² If we set aside differences between the observed cohorts, there remain two ways to reconcile these observations. The first is that ASE-MRI and TRUST-MRI have different sensitivities to anatomical brain regions, with ASE-MRI measuring a true whole-brain average including the white matter, and TRUST-MRI biased to the cortex. The second is that ASE-MRI produces a spatial average of OEF and TRUST-MRI a flow average. The first possibility is partially true. Fields et al. observed that OEF increased predominantly in white matter. However, an increase was also observed in gray matter. We believe the second possibility may allow the observations to be reconciled.

The difference between spatial and flow averaging means that ASE-MRI and TRUST-MRI measure two different aspects of oxygen extraction in the brain. ASE-MRI offers a specific window into the extraction occurring in the microvasculature. Naive models of the cerebral microvasculature where SO₂ monotonically declines across arterioles to capillaries to venules, are not accurate. Observations in the mouse cortex using optical methods instead show that there is a tremendous heterogeneity in oxygen saturation and flow among the microvessels. This heterogeneity gives rise to the unintuitive observation that venular SO₂ increases with cortical vessel diameter.⁶ The most likely explanation of this observation is that among the heterogeneous flow paths between arterioles and venules those with higher flow tend to have higher SO₂.⁶ Venular SO₂ is therefore a flow-weighted averaging of the SO₂ of the microvascular paths. Thus, although the voxel space is predominately made up of low SO₂ capillaries there are flow paths with disproportionately high flow and higher SO₂ that cause the flow-averaged SO₂ in draining veins to be higher. In other words, OEF is higher and SO₂ lower in the capillaries than in the larger veins into which they drain (black arrow, Figure 1).

FIGURE 1.

(A) The naive picture where vessel size defines vessel type and each type is distinct and homogeneous with arterioles in red, capillaries in green and venules in blue. (B) An actual observation using two-photon microscopy of the pO₂ in the vessels with pO₂ heterogeneity in the capillary bed and counterintuitive higher pO₂ in the venule (arrow). Scale bars, 200 μm. From Reference [6] [Color figure can be viewed at wileyonlinelibrary.com]

How this differential sensitivity between spatial and flow averaging plays out in SCA is not entirely known, but we offer our current thinking. In cases of SCA, anemia and compensatory hyperemia developing alongside the sludging and occlusion caused by HbS polymer, likely leads to increased heterogeneity of flow and SO₂ in the microvasculature. Tissue starved for oxygen under these conditions means spatially averaged OEF observed in the capillaries via ASE-MRI is greater than in healthy controls. However, some microvascular paths have even higher flow than normal with an even lower OEF than normal leading to the diminished OEF observed in the SSS. ASE-MRI OEF is simply not comparable to TRUST-MRI OEF, but each is important. Higher tissue OEF indicates ischemic injury risk, while lower whole brain OEF is related to compensatory hyperemia.

Holding open the possibility that both observations are correct leads to the conclusion that OEF measured with ASE-MRI may be a tissue-specific biomarker of brain health in SCA, while OEF measured with TRUST-MRI provides quantitative insight into whole-brain oxygen physiology. To fully appreciate this difference, we must consider OEF in its larger physiological context. The Fick principle states that oxygen that flows into the brain but not out must be consumed. Formally, CMRO₂ = 0.547 · OEF · SaO₂ · [Hb] · CBF, where CMRO₂ is the cerebral metabolic rate of oxygen consumption (μmol/100 g tissue/min), $\text{OEF=Sa}{\text{O}}_{\text{2}}-\text{Sv}{\text{O}}_{\text{2}}\text{/Sa}{\text{O}}_{\text{2}}$, SaO₂, and SvO₂ are the arterial and venous blood oxygen saturations, respectively [Hb] is the blood hemoglobin concentration (g/dl), and CBF is cerebral blood flow (ml/min/100 g tissue). To maintain CMRO₂ in the face of diminished blood oxygen-carrying capacity in anemia (lower [Hb]), CBF and OEF must adapt. Thus, OEF is at once a sensitive, but incomplete metric of brain oxygen consumption. It is only half of the equation.

In the end, CMRO₂ will either be satisfied or not, achieved by some amount of oxygen delivery and oxygen extraction. Although tissue- and large-vessel-based OEF and flows can differ, whole-brain CMRO₂ should be the same. Large-vessel estimates of CMRO₂ are decreased in SCA compared to controls suggesting insufficient oxygen delivery.² ASE-MRI measures of tissue OEF in SCA are increased suggesting compensation for decreased oxygen delivery. Also, the increased tissue OEF is compatible with decreased large vessel OEF if there are flow paths with disproportionately high flow and high SO₂ in vessels that participate less in oxygen exchange. To fully reconcile ASE-MRI with TRUST-MRI measures, whole-brain CMRO₂ needs to be calculated from tissue estimates and compared to the large-vessel estimates. However, this requires an MRI-based measure of capillary CBF. Current Arterial Spin Labeling (ASL) methods, such as pseudocontinuous ASL (PCASL) and pulsed ASL (PASL) are often contaminated by large vessel flow making them inaccurate tissue measures.⁷ Velocity selective ASL mitigates the transit time uncertainty that affects PCASL and PASL in hyperemic states, but cut-off velocities must be refined. Given the ischemic strokes observed in the SCA subjects by Fields et al. it seems reasonable to assume that in white matter oxygen needs are not being met and CMRO₂ is likely reduced, consistent with the diminished whole-brain large-vessel CMRO₂.

Separate from this quantitative approach to oxygen physiology, Fields et al. have provided persuasive evidence that they have defined an imaging biomarker that may permit improved monitoring of brain health in SCA. Thus, a key priority for the field should be to investigate the microvascular changes caused by SCA as part of a theory of physiological mechanisms that would formally unify the seemingly contradictory OEF observations we have discussed. Understanding the pathophysiological mechanisms and their effects on imaging biomarkers requires collaboration between imaging and basic scientists. As Fields et al. note, their findings suggest that there are aspects of SCA pathophysiology aside from decreased arterial oxygen content that drive changes in OEF. Unlike other types of anemia, in SCA, pathologic polymerization of hemoglobin reduces overall blood oxygen affinity at a given oxygen tension,^8,9 enhancing oxygen extraction and possibly OEF. Combining imaging-based measurement of tissue oxygenation with measurement of intra-erythrocyte hemoglobin polymerization for individual patients may provide a more complete picture of oxygen transport in SCA and how it contributes to cerebral pathology. In our group, we are attempting to couple the insights from MRI and near-infrared spectroscopy (NIRS) with those from in vitro single-cell assays of oxygen-dependent hemoglobin polymerization for individual patients.¹⁰ Eventually, we intend to confirm whether low-oxygen tension that causes polymerization and sickling leads to capillary transit abnormalities detectable by MRI or NIRS.

This examination of physiological mechanisms can also be accelerated by relying on multiple imaging modalities to corroborate findings. Our group and others have been developing advanced optical technology (frequency-domain near-infrared combined with diffuse correlation spectroscopies, FDNIRS-DCS) for quantitative, bedside monitoring of regional OEF, CBF, and CMRO₂ in pediatric populations which are now being applied to SCA.^11,12 Increased OEF in cortical tissue observed with FDNIRS corroborates the findings presented by Fields et al., but cannot probe the deeper white matter where they found the largest effects.¹² Furthermore, while MRI-based measurements could provide a more comprehensive understanding of the cerebrovascular physiology at both tissue and whole-brain level at discrete times, FDNIRS-DCS is more feasible for frequent screening or monitoring of the cortex and can be utilized in under-resourced areas. Our current effort has focused on exploring the clinical utility of this combined approach (MRI-NIRS) under a variety of clinical conditions (i.e., chronic transfusions and gene therapy) to offer a complementary view of SCA hemodynamic and CMRO₂ response to treatment (NCT04166526).

In summary, observations of decreased large-vessel OEF and increased tissue OEF in SCA can both be correct, and both techniques can be leveraged to probe aspects of the pathophysiology of oxygen delivery and consumption in SCA while simultaneously developing clinically useful MRI biomarkers of brain health. However, to fully integrate these methods to more deeply understand cerebral pathology in SCA, tissue-based measures of CMRO₂ are also needed.