Red cell exchange transfusions increase cerebral capillary transit times and may alter oxygen extraction in sickle cell disease

Abstract

Persons with sickle cell disease (SCD) suffer from chronic hemolytic anemia, reduced blood oxygen content, and lifelong risk of silent and overt stroke. Major conventional stroke risk factors are absent in most individuals with SCD, yet nearly 50% have evidence of brain infarcts by age 30 years, indicating alternative etiologies for ischemia. We investigated whether radiological evidence of accelerated blood water transit through capillaries, visible on arterial spin labeling (ASL) MRI, reduces following transfusion-induced increases in hemoglobin and relates to oxygen extraction fraction (OEF). Neurological evaluation along with anatomical and hemodynamic imaging with cerebral blood flow (CBF)-weighted pseudocontinuous ASL and OEF imaging with T₂-relaxation-under-spin-tagging (TRUST) were applied in sequence before and after blood transfusion therapy (n=32) and in a comparator cohort of non-transfused SCD participants on hydroxyurea therapy scanned at two time points to assess stability without interim intervention (n=13). OEF was calculated separately using models derived from human hemoglobin-F, hemoglobin-A, and hemoglobin-S. Gray matter CBF and dural sinus signal, indicative of rapid blood transit, were evaluated at each time point and compared to OEF using paired statistical tests (significance: two-sided p<0.05). No significant change in sinus signal was observed in non-transfused participants (p=0.650), but a reduction was observed in transfused participants (p=0.034), consistent with slower red cell transit following transfusion. The dural sinus signal intensity was inversely associated with OEF pre-transfusion (p=0.011), but not post-transfusion. Study findings suggest that transfusion-induced increases in total hemoglobin may lengthen blood transit times through cerebral capillaries and alter cerebral OEF in SCD.

Introduction

Sickle cell disease (SCD) is an inherited blood disorder resulting in the development of hemoglobin S (HbS) and corresponding chronic hemolytic anemia. SCD is also associated with a high risk of overt stroke and silent cerebral infarct (SCI) throughout the lifespan, with SCIs being present in approximately half of all adults by age 30 years ¹. In both children and adults with SCD, approximately monthly blood transfusions are indicated for stroke prevention in high-risk patients with elevated transcranial Doppler ultrasound velocities ^{1, 2} or progressive silent or overt infarcts ³. Blood transfusion treatments generally increase total hemoglobin by approximately 1–2 g/dl and reduce HbS fraction by 25–75% ⁴, thereby increasing blood oxygen content and presumed oxygen delivery to tissue.

As blood oxygen content reduces due to anemia, cerebral blood flow (CBF; ml blood/100g tissue/min) will increase to maintain an adequate amount of oxygen delivery to tissue ^{5, 6}. In the setting of SCD, cortical CBF is elevated from healthy values of 40–60 ml/100g/min to approximately 75–100 ml/100g/min but can increase even higher when total hemoglobin is more substantially reduced ^7–9. While such hyperemia is expected to improve oxygen delivery, the oxygen delivery to tissue additionally depends on the capillary transit time, which is commonly estimated to be approximately 0.9–1.5 seconds in healthy parenchyma ¹⁰ but may be reduced in the presence of high flow scenarios such as in SCD. Here, red cells and plasma will traverse the capillary bed faster, and as such there will be less time for oxygen delivery to tissue within the capillary exchange site. Therefore, oxygen delivery to tissue may not scale directly with increases in CBF but may also depend on the time the red cell spends in the capillary bed itself. Capillary transit times should influence tissue oxygen delivery, even in the presence of normal or elevated CBF ¹¹. In anemic individuals where increased CBF and blood velocity are required to compensate for reduced blood oxygen content, shorter vascular transit times can lead to reduced time for oxygen offloading; capillary flow disturbances have been suggested to alter oxygen delivery in individuals with anemia ¹², stroke ¹¹, traumatic brain injury ¹³, aging-related white matter disease ¹⁴, Alzheimer’s disease ¹⁵, and migraine ¹⁶.

In support of this premise, relatively small or negligible CBF changes in response to transfusion-induced increases in total hemoglobin and decreases in HbS fraction have been observed in persons with SCD, and the implications of such changes on oxygen delivery have been discussed in separate work ^{6, 12, 17}. These small changes are observed despite well-documented improvements in symptomatology and reduced stroke rates, suggesting that transfusion-induced increases in hemoglobin in individuals with SCD have an impact on oxygen delivery beyond simply modifying total CBF and hemoglobin. One possibility is that the increase in total hemoglobin reduces flow velocities and increases associated capillary transit times, which allows for increased oxygen extraction from the blood compartment (Figure 1). In support of this possibility, oxygen extraction has recently been shown to be reduced in some adults with SCD, specifically those with evidence of aberrant blood-tissue water exchange on perfusion-weighted imaging ¹², and in a single case report, that blood transfusion can reduce arterio-venous shunting and increase oxygen extraction ¹⁸.

Figure 1.

Potential relevance of capillary transit time on arterial, venous, and tissue oxygen delivery in sickle cell disease. Warm colors depict high oxygen saturation and cool colors show low oxygen saturation. (A) In healthy brain parenchyma, capillary transit times of 0.9–1.5s lead to an oxygen extraction fraction of 30–40%. (B) In the proposed situation of compensated anemia, smooth muscle surrounding arterioles will relax to facilitate vasodilation and increase cerebral blood flow; when these changes are sufficient to maintain cerebral oxygen delivery despite anemia, normal amounts of oxygen are delivered to tissue. (C) In non-compensated anemia, the cerebral hyperemic response is typically larger and leads to a rapid transit of red cells and plasma through capillaries (white arrows). This rapid transit may result in insufficient time for oxygen to be extracted by tissue, leading to reduced oxygen delivery to brain tissue and increased oxygen concentrations in venules. In this model, the presence of arterio-venous shunting is generally associated with higher cerebral blood flow, which is demonstrated in vivo in Figure 2A. SaO₂=arterial oxygen saturation. SvO₂=venous oxygen saturation.

Here, we extend these preliminary findings to evaluate evidence of cerebral arterio-venous shunting on magnetic resonance imaging (MRI) of the brain in a larger cohort of adult and pediatric SCD participants scanned before and after blood transfusion. Measurements of dural venous sinus hyperintensity from arterial spin labeling MRI are used as a proxy for capillary transit time and paired with measurements of tissue structure, CBF, and oxygen extraction fraction (OEF). We postulated that evidence of arterio-venous shunting is associated with hyperemia and reduced OEF within the setting of SCD, and evidence of arterio-venous shunting reduces following transfusion-induced increases in total hemoglobin.

Methods

Demographics

Adult and pediatric participants with SCD provided informed, written consent in accordance with the local institutional review board for this prospective study. The study goal was to understand how cerebral hemodynamics and dural venous ASL hyperintensity, consistent with a change in capillary flow or permeability, responded to blood transfusion. Patients scheduled for clinically-indicated red cell exchange transfusions were enrolled. To additionally understand the stability of the study measures in the absence of an intervention, a second cohort of participants stable on hydroxyurea without blood transfusion intervention were included. Participants with SCD were recruited from a local sickle cell clinic continuously between 2015 and 2020. All enrolled participants had either hemoglobin-SS (HbSS) or hemoglobin-Sβ0-thalassemia (HbSβ0) phenotype and no 3T MRI contraindications (dental braces, implantable devices that were 3T MRI prohibitive, intracranial clips, or metallic foreign body in the eyes).

Blood transfusion and hematological measures

Hematocrit and hemoglobin levels, including hemoglobin-S percentage (HbS%), and arterial oxygen saturation fraction were measured in all participants on the day of the scan. MRI was performed within seven days prior to an exchange transfusion and repeated within 12 days after the transfusion. Time before and after transfusion were recorded. The decision of whether to perform a simple or exchange transfusion was based on clinical criteria and was not a research procedure.

Magnetic resonance imaging and angiography

To characterize prior infarct and vasculopathy extent, a standard non-contrast anatomical head and neck MRI and angiography (MRA) protocol was performed including: T₁-weighted imaging (magnetization-prepared rapid gradient echo; spatial resolution = 1.0 × 1.0 × 1.0 mm³; 3D turbo-field-echo; repetition time/echo time = 8.2/3.7 ms); T₂-weighted imaging (spatial resolution = 0.6 × 0.6 × 4.0 mm³; turbo spin echo; repetition time/echo time = 3000/80 ms); 2D T₂-weighted fluid attenuated inversion recovery (FLAIR) acquired separately in two orthogonal planes (axial and coronal) (spatial resolution = 0.9 × 1.1 × 3.0 mm³; turbo inversion recovery; repetition time/inversion time/echo time = 11,000/2800/120 ms); and intracranial time-of-flight magnetic resonance angiography (spatial resolution = 0.6 × 0.6 × 1.4 mm³; 3D gradient echo; repetition time/echo time = 23/3.5 ms).

2D pseudocontinuous arterial spin labeling (pCASL) was used to evaluate gray matter CBF as well as dural venous signal intensity. In each participant, an identical scan was performed before and after transfusion, or in intervention-control participants at two time points. Imaging parameters varied slightly between adults and children. Both scans had a common spatial resolution of 3 × 3 × 7 mm³ and utilized dual-pulse background suppression, however, adults utilized an ASL scan with post-labeling delay = 1900 ms, averages = 20, scan duration = 168s, whereas children utilized an ASL scan with labeling parameters of post-labeling delay = 1650 ms, averages = 20, scan duration = 168s. The difference in label parameters was due to known differences in blood arrival between adults and children and a desire to titrate each sequence to the population. Importantly, an identical pCASL scan was performed in each participant before and after transfusion, or in the non-transfused cohort, at each time point. Other scan parameters: TR=4200 ms, TE=13 ms, spatial pre-saturation, bandwidth=2665 Hz, slice time = 23 ms, field-of-view=240×240 mm, slice thickness=7 mm, slices=17, SENSE-factor=2.3, echo planar imaging (EPI) factor = 35, k-space trajectory=cartesian. The M₀ scan was acquired with identical geometry but with the pCASL preparation removed, TR=20s, and scanner gain, shimming, and scaling unchanged from the pCASL acquisition.

T₂-relaxation-under-spin-tagging (TRUST) ¹⁹ data were acquired twice per session (field-of-view=230×230 mm, spatial resolution=3.4×3.4×5 mm³, τ_CPMG=10 ms, effective echo time=0, 40, 80, and 160 ms, TR/TE=1978/3.6 ms, averages=3, slice thickness=5 mm, single-slice 2D EPI, bandwidth=3201 Hz, SENSE-factor=3) ^{19, 20} for OEF determination. Control and venous-labeled (transfer insensitive labeling technique; TILT) TRUST images were acquired from a slice planned orthogonal to the sagittal sinus and 20 mm superior to the torcula.

Image analysis

MRI and MRA findings were independently evaluated by two board-certified neuroradiologists. Disagreement was resolved by consensus. Number of participants with intracranial vasculopathy and cerebral infarcts were calculated based on brain MRI and MRA findings. Vasculopathy was defined as at least 50% stenosis of the first segment of anterior, middle, or posterior cerebral artery and/or intracranial segment of internal carotid artery (ICA) or basilar artery. Silent cerebral infarcts were determined as hyperintense lesions on FLAIR ≥ 3mm, normal neurological examination, and no history of stroke-like symptoms. Overt strokes were diagnosed via a history consistent with clinical stroke and corresponding lesion hyperintense on FLAIR and hypointense on T₁-weighted MRI approaching CSF signal.

pCASL data were corrected for motion and baseline drift, normalized by the equilibrium magnetization, M₀, and the solution to the flow-modified Bloch equation, using the measured hematocrit for blood T₁ determination and SCD labeling efficiency (α=0.72), was applied ⁷. Here, the recently-recommended white paper simplified kinetic model ²¹ was not used and rather we used the final stage of a previously-published three-stage kinetic model which has been described in detail ^{22, 23} and evaluated previously in SCD ⁷,

$$\mathrm{\Delta }{M}_{tiss}=2\alpha M_{0,b}f{T}_{1,app}{e}^{-\frac{BAT}{T_{1,b}}}{e}^{-\frac{\left(t-BAT-\tau \right)}{T_{1,app}}}\left(1-e^{\frac{-\tau }{T_{1,app}}}\right)$$ [1],

where ΔM_tiss is the difference magnetization (i.e., control – label), α=0.72 is the pCASL labeling efficiency in participants with SCD ⁷, M_0,b is the equilibrium magnetization of blood water calculated in each participant, f is the perfusion, T_1,app is the apparent relaxation time of tissue defined as 1/T_1,app = 1/T_1,tiss+f/λ, λ=0.9 ml/g is the blood partition coefficient, T_1,tiss=1.209s ^{24, 25}, ΒΑΤ=1.02s is the approximate bolus arrival time (measured in SCD previously ⁷), T_1,b is the T₁ of blood water as calculated below, and τ is the labeling duration. We calculated the participant-specific blood T₁ using the measured hematocrit and previously published relationship between arterial (oxygenation=92±7%) blood water T₁ and hematocrit: 1/T₁ = 0.52 x Hct + 0.38 ²⁶. An alternative method has been proposed whereby individual T₁ measurements are made in venous blood water in vivo ²⁷. We chose not to measure venous T₁ as the flow-modified Bloch equation requires arterial blood water T₁, and the venous and arterial T₁ differ by approximately 200 ms, depending on oxygenation status. However, it should be noted that both approaches are commonly used in the literature. T₁-weighted images were segmented into gray matter, white matter, and CSF using FSL-FAST ²⁸. Perfusion maps were transformed to the native T₁-weighted space using the M₀ image for co-registration and gray matter perfusion recorded.

Next, dural sinus signal was evaluated in the pCASL data. Dural sinus pCASL signal was quantified in two ways: (i) using a previously reported categorical scoring system ²⁹ and (ii) using a continuous, quantitative measure of venous signal at a standardized location in the superior sagittal sinus. First, categorical shunting scores of 0, 1 or 2 were calculated for each MRI using the previously reported categorical scoring system ²⁹ and were assigned to each image according to the intensity and continuity of ASL signals extending ventrally from the superior sagittal to straight and finally transverse sinus. A shunting score of 0 corresponded to a weak venous signal with little-to-no enhancement in any segment of the sagittal sinus. A shunting score of 1 corresponded to an increased venous signal through the superior sagittal or straight sinuses, which did not extend to the torcula. A shunting score of 2 corresponded to a continuous, conspicuous venous hyperintense signal that extended through all segments of the sagittal sinus to the confluence of the sagittal and transverse sinuses. As such, the categorical scoring system is ordinal with increasing score consistent with increased venous hyperintensity and reduced blood-tissue water exchange. Additionally, a continuous measure of venous signal was calculated. A circular region of interest (ROI) of approximately 40 mm² was identified at the approximate location of the parieto-occipital sulcus. While the signals in the dural sinuses are calibrated in an identical way as in the brain tissue (normalization by M₀ and scaled identically as defined above), we utilize the nomenclature of flow for signal in the sinuses, and perfusion in the tissue, however both measures are presented in normalized units of ml/100g/min. This calibrated mean venous flow signal within this region of the superior sagittal sinus was extracted from the quantified CBF map, which was achieved to reduce variability from differences in scanner gain between participants that would otherwise be present without signal normalization.

TRUST data were pair-wise subtracted, and venous blood water T₂ was quantified in the superior sagittal sinus; venous blood water T₂ values were then converted to venous oxygen saturation (Y_v) using previously characterized calibration curves. Since the effect of hemoglobin phenotype (e.g., Hemoglobin-S vs. Hemoglobin-A) on blood water relaxation at 3 Tesla has been shown to be negligibly ³⁰, moderately ³¹, or highly ³² important, we present findings from all models that have a hematocrit dependence and are calibrated over an anemic range: human hemoglobin-AA ³³, human hemoglobin-F ³⁴, and a model that incorporates data from SCD participants from two different research groups ³⁵. It should be noted that the data merged from the two different groups ^{31, 32} exhibit different trends for calibration parameters, which is considered in the Discussion. For transparency, results from all models are presented and we discuss the strengths and limitations of the different models in the Discussion. Y_v was then utilized along with arterial oxygenation saturation (Y_a), which was acquired from two-wavelength pulse oximetry, to calculate OEF=(Y_a-Y_v)/Y_a. The two repeated OEF values from both TRUST measurements in the same session were averaged to obtain the OEF value for each participant. The quantified variable from this analysis was a global measure of the OEF.

Statistical analysis

Descriptive statistics were first calculated, including means and standard deviations for continuous parameters. Assessment of outliers (2.5 standard deviations from the group mean) and assumptions for statistical analysis (e.g., normality) were made.

The primary hypothesis was to analyze whether non-invasive surrogate measures of arterio-venous shunting (e.g., shunting score and dural sinus flow signal) changed significantly following red cell exchange transfusions. Paired statistics were calculated between pre-transfusion and post-transfusion and between time 1 and time 2 using paired t-tests for parametric data.

A secondary hypothesis was to understand whether any such changes were associated with changes in OEF. The mean of the two OEF values for the sequential TRUST measurements was calculated and used for analysis. To understand repeatability, the intraclass correlation coefficient (ICC) and Pearson’s R between repeated measures were both calculated. To test the secondary hypothesis, a Wilcoxon rank sum test was applied to evaluate hypothesized differences in data between the SCD transfused and SCD non-transfused participants. Here, OEF (using each model), tissue CBF, and flow signal values were compared between groups. Additionally, a Spearman rank sum test was applied to evaluate potential relationships between the flow signal and the OEF, separately before and after transfusion. In exploratory analyses and to understand whether the pre-transfusion arterio-venous shunting effect may predict transfusion response, we evaluated whether the pre-transfusion venous hyperintensity was related to the CBF reduction induced from increasing blood oxygen content by comparing the change in tissue CBF and pre-transfusion venous hyperintensity using a Spearman rank-sum test.

Finally, to understand whether the findings were similar for participants with pre-transfusion and post-transfusion scans on the same day, we repeated the above analyses for only the subgroup of participants with pre-transfusion and post-transfusion data obtained on the same day as the transfusion.

In all cases, two-sided p < 0.05 was required for significance.

Results

Table 1 summarizes participant demographics. We enrolled 13 non-transfused SCD participants (age = 21.2 ± 8.0 years, 38% male) and 32 transfused SCD participants (age = 18.2 ± 9.1 years, 53% male). All participants were African American with either hemoglobin-SS (HbSS) or hemoglobin-Sβ0-thalassemia (HbSβ0) phenotype. Six of 13 non-transfused participants had silent cerebral infarcts; one had a history of prior overt stroke and vasculopathy. At time 1, mean hemoglobin was 8.7 ± 1.2 g/dL while at time 2, mean hemoglobin was 8.8 ± 1.1 g/dL. Twenty of 32 transfused participants had silent cerebral infarcts, five had a history of prior overt stroke, and ten had a history of vasculopathy. All participants with an overt stroke also had a history of vasculopathy. Mean Hb pre-transfusion was 8.9 ± 1.7 g/dL compared to 9.9 ± 1.5 g/dL post-transfusion. Typically, transfusions were monthly, thus pre-transfusion scans occurred at least three weeks after the most recent blood transfusion when hematocrit was near nadir. Post-transfusion scans were performed 2.0 ± 3.5 days (range = 0 – 12 days) following blood transfusion.

Table 1.

Participant demographics and imaging measures.

Demographics	SCD Non-transfused			SCD Transfused
N	13			32
Age (years)	21.2 ± 8.0			18.2 ± 9.1
Sex (percent male)	38			52
Race (percent African American)	100			100
SCI (percent present)	46			63
Overt stroke (percent present)	7.7			16
Vasculopathy (percent present)	7.7			31
Relationship Between CBF and OEF Response	SCD Non-transfused			SCD Transfused
	Time 1	Time 2	P-value	Pre-transfusion	Post-transfusion	P-value
Hb (g/dL)	8.7 ± 1.2	8.9 ± 1.7	0.391	8.8 ± 1.1	9.9 ± 1.5	*<.0001
HbS (percent)	66.5 ± 18.8	70.0 ± 21.3	0.569	39.9 ± 15.3	28.9 ± 14.2	*<.0001
CBF in tissue (ml blood/100 g tissue/min)	92.2 ± 13.5	96.1 ± 24.6	0.735	83.7 ± 24.4	74.5 ± 18.3	*0.031
OEF (HbAA#)	34.5 ± 3.9	33.7 ± 7.1	0.719	33.7 ± 6.5	34.5 ± 6.5	0.666
OEF (HbF#)	39.8 ± 4.6	39.8 ± 6.3	0.992	38.7 ± 7.2	37.3 ± 6.0	0.386
OEF (Merged Mixture#)	21.9 ± 5.6	19.8 ± 9.6	0.435	26.0 ± 6.9	28.9 ± 6.9	0.147
Categorical shunting score (range=0–2)	1.53 ± 0.51	1.47 ± 0.52	0.673	1.38 ± 0.61	1.34 ± 0.55	0.711
Flow signal in superior sagittal sinus (ml blood/100g tissue/min)	117.4 ± 30.1	117.1 ± 32.1	0.650	113.6 ± 37.1	98.9 ± 33.9	*0.034
Arterial oxygen saturation (percent)	96.3 ± 2.6	93.8 ± 4.9	0.071	95.7 ± 2.4	96.4 ± 2.4	0.193

Abbreviations: SCD, sickle cell disease; Hb, hemoglobin; SCI, silent cerebral infarct; CBF, cerebral blood flow; OEF, oxygen extraction fraction.

^#Denotes the model used to calculate the OEF.

Figure 2 shows three SCD participants, each with different categorical shunting scores according to previously published criteria ²⁹. Representative images from SCD participants are shown below. To standardize the location the region for quantifying sagittal sinus signal intensity, the flow maps were registered to a standard atlas as described above; Figure 3 shows how the conspicuity of the dural sinus hyperintensities compares before versus after this registration process.

Figure 2.

Quantification of dural sinus signal in pseudocontinuous arterial spin labeling (pCASL) data using both a (A) continuous measure of sinus signal and (B-D) categorical scoring system. (A) Location of the standardized region of interest (ROI) selected with a cross-sectional circular area of approximately 35–45 mm². Orthogonal depictions of cerebral blood flow (CBF)-weighted pCASL maps are shown as references for shunting scores of (B) 0, (C) 1 and (D) 2, assigned according to strength and continuity of venous signal. Note that the conspicuity of the dural sinus hyperintensity generally scales with the extent of cortical hyperemia. (E) Participant with sickle cell disease (SCD) and silent cerebral infarcts on regular blood transfusion therapy with no history of vasculopathy or overt stroke. CBF-weighted pCASL maps are shown prior to and following a blood transfusion. Baseline hemoglobin (Hb) was 9.1 g/dl, which elevated to 10.2 g/dl following transfusion. White arrows indicate dural sinus hyperintensity in the superior sagittal sinus, which reduces in conspicuity post-transfusion. (F) Participant with silent cerebral infarcts and SCA stable on hydroxyurea with no blood transfusion intervention and no history of vasculopathy or overt stroke. CBF-weighted pCASL maps are shown at time 1 and time 2. This participant’s baseline hemoglobin was 11.6 g/dl, which increased slightly to 12.2 g/dl at time 2. Dural sinus hyperintensities are less conspicuous compared to the transfusion participant and are largely stable across time points. Group results are summarized quantitatively in Figure 4.

Figure 3.

Arterial spin labeling images after (A) and before (B) co-registration to the 2 mm standard atlas demonstrate that dural sinus hyperintensities are conspicuous at both resolutions. It is possible to disambiguate the sinus hyperintensities from extra-axial noise and motion by ensuring that the signal traces through slice, which can be appreciated in the orthogonal planes shown. Yellow arrows depict the sagittal sinus and magenta arrows depict the straight sinus, when visible.

The bottom half of Table 1 summarizes results of the shunting analysis across all participants. Quantification of categorical and continuous shunting measures pertained to the primary hypothesis of the study that the dural venous hyperintensity decreases post-transfusion, as total hemoglobin increases. Using the categorical scoring system, mean shunting score in non-transfused participants was 1.53 ± 0.51 at time 1 and 1.47 ± 0.52 at time 2 (p=0.673). Mean categorical shunting score in transfused participants was 1.38 ± 0.61 pre-transfusion and 1.34 ± 0.55 post-transfusion (p=0.711). Therefore, there was no significant change observed in the categorical shunting score for either the non-transfused or transfused SCA participants. Evaluation of the continuous pCASL mean venous signal provided additional information. We observed no significant difference in mean flow signal between time 1 (117.4 ± 30.1 ml/100g/min) and time 2 (117.1 ± 32.1 ml/100g/min) of non-transfused participants (p=0.650), yet in transfused participants a significant decrease in mean flow signal was observed between pre-transfusion (113.6 ± 37.1 ml/100g/min) and post-transfusion (98.9 ± 33.9 ml/100g/min) (p=0.034). As noted in the Methods, the flow signal is written as ml/100g/min, simply to allow for direct, normalized comparison to the perfusion signal. These findings provided evidence in favor of the primary hypothesis of the study and specifically that dural venous ASL signal hyperintensity reduces following transfusion-induced changes in total hemoglobin.

Figure 4 summarizes group level imaging findings. A decrease in gray matter CBF was observed following blood transfusion for the transfused participants (p=0.031), but no change in gray matter CBF was observed between time points for non-transfused participants (p=0.735). Consistent with the primary study hypothesis, transfused participants had significantly decreased dural sinus flow signal after transfusion (p=0.034). Alternatively, there was no change between time points in the non-transfused participants (p=0.650).

Figure 4.

(A) Gray matter cerebral blood flow and (B) flow signal in the sagittal sinus at two time points without interval intervention or before versus after transfusion. In sickle cell disease participants not receiving interval transfusion, gray matter cerebral blood flow and sagittal sinus flow signal are not significantly different between two time points. In participants receiving interval transfusion, gray matter cerebral blood flow (p=0.031) and flow signal in the sagittal sinus (p=0.034) both reduce. Note that the flow signal is shown in calibrated units of ml/100g/min to allow for direct comparison with cerebral blood flow, however, this measurement depicts the intravascular signal and not the rate of blood delivery to tissue. * p < 0.05.

For the two TRUST measurements performed sequentially in each scan, the quantified venous relaxation rates (R₂) had an ICC=0.92 and Pearson’s correlation of 0.96. Figure 5 compares mean flow signal in the dural sinus and OEF (HbAA model) before and after blood transfusions. We observed that with increasing mean dural sinus flow signal, consistent with increased capillary shunting, participants before transfusion had a reduced OEF (Wilcoxon test, p=0.011), consistent with prior reports ¹². The post-transfusion relationship between OEF and flow signal became non-significant for all models.

Figure 5.

Before transfusion, there is an inverse relationship (p=0.011) between flow signal in the sagittal sinus and oxygen extraction fraction (OEF), whereas after transfusion (gray) the relationship becomes non-significant. Findings are shown for the human HbAA calibration model (see Methods and Discussion). Note that the flow signal is shown in calibrated units of ml/100g/min to allow for direct comparison with cerebral blood flow, however, this measurement depicts the intravascular signal and not the rate of blood delivery to tissue.

As an exploratory analysis we evaluated whether the pre-transfusion shunting effect may portend transfusion response. Here, it was observed that pre-transfusion ASL flow signal in the sagittal sinus was inversely related to the CBF change in the tissue after transfusion (Spearman’s ρ=0.410; p=0.020). Specifically, those participants with the highest pre-transfusion flow signal had the largest reduction in tissue CBF after transfusion.

Finally, as a sub-analysis for completeness, we evaluated whether trends were similar or different if only data from participants that had MRI measurements on the same day of transfusion (n=19) were considered. Here, the trends were the same as described above for the full treatment cohort (n=32). The hemoglobin increased from 8.9±0.7 g/dl to 10.3±1.2 g/dl (p<0.001). This paralleled a trend for a reduction in gray matter CBF as expected (pre: 82.0±22.2 ml/100g/min; post: 74.4±14.8 ml/100g/min; p=0.065) and a trend for a reduction in ASL superior sagittal sinus flow signal (pre: 110.5±21.3 ml/100g/min; post: 95.2±34.5 ml/100g/min; p=0.085). Before transfusion, the venous hyperintensity signal was inversely related to the OEF (HbAA model) (p=0.022) and after transfusion the hypertense signal trended to be positively related to OEF, but as in the full cohort, this was not significant (p=0.386). As such, the direction of the findings is similar and consistent regardless of whether the full cohort is considered or only the subgroup of participants with pre-transfusion and post-transfusion scans on the same day (n=19). The level of significance does vary, however, as the same-day cohort is approximately 30% smaller than the full cohort.

Discussion

We provide evidence that dural sinus hyperintensity on arterial spin labeling MRI, a hypothesized surrogate of cerebral arterio-venous shunting, reduces following transfusion-induced increases in hemoglobin in adults and children with sickle cell disease (SCD). While previously shown in a single case study ¹⁸, we extend these preliminary findings to demonstrate this effect, and its potential relation to oxygen extraction, in a larger cohort of SCD participants before and after transfusion. Prior to transfusion, an inverse correspondence between dural sinus hyperintense signal and global OEF was observed; after transfusion, this relationship changed and was either null or positive, depending on calibration model used. We also reinforce the dural sinus hyperintensity findings by demonstrating comparative stability of dural sinus hyperintensity signal in a comparator non-transfused SCD cohort scanned at two time points on hydroxyurea therapy. We observed that while mean flow signal in the superior sagittal sinus reduced post-transfusion in the transfused cohort, no significant change on average was observed in the non-transfused SCD cohort. The above findings were similar regardless of whether only data on the same day of the transfusion were considered, or whether data from all participants extending out to 12 days post-transfusion were considered.

Continuous versus ordinal assessment of arterio-venous shunting

To quantify dural venous sinus signal in pCASL data, we employed both categorical and new continuous measures of venous signal, with the latter presumed to be more sensitive to small adjustments in capillary transit. While pairwise comparisons of continuous flow signal measures in the sagittal sinus were found to be significant between pre- and post-transfusion, comparisons of categorical shunting score measures were not significant. It is anticipated that the ordinal, categorical measures are simply less sensitive to subtle differences in dural signal intensity as only three scores (0, 1, 2) are utilized to encompass a presumed wide range of differences in flow velocities. As such, the continuous measure of mean venous signal may be a more sensitive indicator of changes in capillary transit time and therefore critical tissue-level impairment for patients with SCD. Future work that extends this line of work to incorporate multi-delay arterial spin labeling data is ongoing and will be necessary to better quantify the arterio-venous transit time and better understand how rapid capillary transit contributes to oxygen extraction in persons with SCD. Such measurements require a larger range of post-labeling delays of approximately 200 – 3500 ms, spanning arterio-venous circulation times in persons with varying intracranial flow velocities and capillary kinetics.

Oxygen extraction fraction

While the primary focus of this work was to characterize dural sinus hyperintensities on ASL imaging before and after transfusion, we also calculated OEF to understand whether any possible dural sinus changes may be associated with oxygen delivery to tissue. Prior work has suggested that blood transfusions can slightly reduce otherwise elevated OEF in patients with SCD ¹², which has been hypothesized to occur due to an increase in blood oxygen content associated with the transfusion-related increase in hemoglobin, which in turn also increases total oxygen delivery to tissue. This finding remains controversial, however. One reason is methodological, as the dephasing of water will depend on the magnetic susceptibility of the deoxygenated red blood cells ³⁶, which may vary for red cells with HbA ³³, HbS ^{30, 35}, and HbF ³⁴. Therefore, the calibration model that calculates the blood oxygenation from the measured blood water T₂ and hemoglobin can yield different results if different hemoglobin phenotypes affect this relationship differently ^30–35. This becomes additionally complex given that patients will have varying levels of HbA, HbS, and HbF and these fractions will change following treatment.

In this study, we utilized three separate OEF calibration models utilizing data from three different groups for completeness. One model was calibrated in healthy HbAA human blood over a wide and anemic hemoglobin range ³³, another from an independent group that was calibrated in neonate blood with HbF over a moderately anemic range (hematocrit=25–50%) ³⁴, and finally a model that merged data acquired in SCD participants from two separate research groups (e.g., merged mixture) albeit with the caveat that the two data sets combined yielded different trends for calibration parameters ³⁵. We did not include one HbSS model simply because it was measured over a narrow hematocrit range and did not include a hematocrit dependence in the simplified terms ³²; since hematocrit changes pre- and post-transfusion, we determined that utilizing a model that did not have a hematocrit dependence was not appropriate for this study. Importantly, results from each considered model are significantly correlated when considering HbAA vs. HbF (p<0.001), merged mixture vs. HbAA (p<0.001), and merged vs. HbF (p<0.001), however the magnitude of the values are highest for the HbF model, slightly reduced for the HbAA model, and reduced below values from participants without interval interventions for the merged mixture model (e.g., Table 1).

Similarly for all models, before transfusion we observed an inverse relationship between the OEF and the dural sinus signal, which was significant for the HbAA and merged mixture models. After transfusion, there was a significant positive relationship between OEF and remaining dural sinus signal when using the HbAA and HbF models, but no relationship when using the merged mixture model. As such, the finding of how OEF changes after transfusion remains somewhat speculative, however, all models suggest that the relationship may differ pre- versus post-transfusion.

It is worthwhile considering the assumptions in the calibration models when interpreting these findings, as several factors beyond simply the blood type differ between the models and this is not always noted. An appropriate model must take into account the fundamental determinants of blood water relaxation, including hematocrit and hemoglobin type, and ideally should include the fraction of deoxyhemoglobin for each hemoglobin type (an assay that is difficult to perform). The PET literature can be used to guide our expectations, as ¹⁵O-PET OEF mapping does not rely on a relaxometry calibration model and has reported normal-to-mildly elevated OEF in SCD participants (OEF=44+/−7%) ³⁷. Recent quantitative MRI measures of susceptibility provide evidence for no significant change in susceptibility of water near red cells from individuals with HbSS vs. HbAA ³⁰. When utilizing this assumption, multiple groups have observed similar OEF values as those measured from PET in comparable SCD participants without clinical indicators of advanced disease severity such as vasculopathy or infarcts ^{8, 12}. Prior to this more recent work being available, earlier MRI studies ³⁸ used a model of bovine blood for calibration under the assumptions that (i) blood from persons with SCD had similar MRI properties as bovine blood, and (ii) the model could be extrapolated accurately to low hematocrit levels. The former assumption was based on the long-standing expectation that bovine and human HbAA blood behave similarly in terms of MRI relaxation times ³⁹, and also evidence provided in the supplementary material of that manuscript ³⁸ whereby it was observed that HbAA and HbSS blood, matched for total hemoglobin, exhibited similar susceptibility effects on water relaxation rate, a finding that has now been independently quantified by another group in a larger study ³⁰. The latter assumption has recently been supported to some degree as well, as oxygenation values in piglets, calculated using this bovine model, over a low hematocrit range of 23.6 ± 6.5%, were closely related to gold standard blood gas measurements obtained from catheterization of draining veins ⁴⁰.

Nonetheless, it is reasonable that these models may be less accurate in SCD, and as such we present data here from a merged mixture model calculated from blood that contains both HbAA and HbSS ³⁵. It is important to understand the pre-requisite studies included in the model: those of Bush et al. ³² and Li et al ³¹. Bush et al ³² reported a separate, simplified calibration model from persons with SCD and using this reported that OEF was approximately 24 ± 4% in SCD participants, compared to 44 ± 7% using gold-standard PET ³⁷ and 38 ± 8% using individualized MRI calibration curves from a separate study performed by Li et al ³¹. These measurements were made in different patients. The merged mixture model includes data from two different groups and a larger sample, albeit with the caveat that the data from the two groups yielded a different directional dependence between the calibration coefficients reported and hematocrit. The model also yielded lower OEF values from those reported using historical gold-standard ¹⁵O-PET ³⁷. While historical PET studies were performed prior to hydroxyurea treatments become standardized and OEF may be higher in untreated patients, it is unlikely that the OEF discrepancy (24 ± 4% measured using MRI by Bush et al. ³² vs. 44 ± 7% measured using ¹⁵O-PET by Herold et al. ³⁷) could be explained by the effects of hydroxyurea alone. For completeness, we have analyzed our data here using these three models and ongoing MRI and PET work in this field will likely be important to clarify which models are most appropriate.

Despite some potential inconsistencies between models and modalities, in all calibration models we observed an inverse trend between dural flow signal and OEF pre-transfusion, which provided evidence for elevated shunting being associated with reduce oxygen extraction within participants with SCD, as has been previously reported ¹². As the dural sinus signal generally scales with the cortical CBF, and CBF and OEF are generally inversely related, this finding is not surprising. This finding was significant for the HbAA and merged mixture models but not for the HbF model, which is not surprising given that participants on chronic blood transfusion have low HbF fractions and most blood donors are HbAA. After transfusion, HbS% is targeted to less than 30% ⁴¹. It is common that after transfusion, the HbS% is 15–20% and generally increases to only 30% by the time of the next monthly transfusion. Therefore, the majority hemoglobin type is HbAA, with a smaller approximately 10% fraction of HbF as most patients on transfusions are not also on hydroxyurea. This provides support that an HbAA model may be reasonable for many transfusion participants.

After transfusion, we observed that the inverse relationship between dural sinus hyperintensity and OEF was not significant, although trended positive when using the HbAA and HbF models. It is possible that persons with SCD with ongoing shunting effects, even after transfusion, are in a state of higher impairment where oxygen delivery remains insufficient to meet demand, thereby leading to an increase in OEF. Alternatively, the shift in the hemoglobin dissociation curve following a reduction in HbSS relative to HbAA may in itself lead to higher oxygen binding affinity and reduced OEF, which may partly offset the effect of capillary residency time. As has been proposed by Jespersen and Ostergaard ¹⁰, oxygen delivery may not be simply related to CBF and blood oxygen carrying capacity. Specifically, in other more common ischemic conditions such as arterial steno-occlusive disease, it has been suggested that reductions in capillary transit time heterogeneity (CTTH) may improve tissue oxygenation; however, when capillary flow is altered, a malignant CTTH condition may ensue whereby higher CBF leads to reduced oxygen availability ¹⁰. The observed relationships in our study cannot be fully explained by simple changes in CBF and hemoglobin alone, but may have origins in CTTH and related capillary residency kinetics. It is not possible to disambiguate these findings from the data available in this study, however, these data suggest that the relationship between the OEF and venous hyperintensity differs, on average, before versus after transfusion. Furthermore, SCD is a complex disorder, and patients have varying degrees of vasculopathy, as well as silent and overt stroke. Thus, it is not expected that all individuals with SCD operate under identical hemodynamic and metabolic mechanisms and a personalized approach to risk assessment and care is likely required. For instance, the data in this study are being used as part of five-year, prospective, longitudinal trial of biomarkers of silent cerebral infarct progression which is targeted to include more than 200 patients with SCD and infarct development status.

Capillary transit as a potential biomarker

We observed that the extent of the pre-transfusion shunting effect scaled inversely with the CBF change in the tissue after transfusion. In other words, the participants with the highest evidence of arterial-to-venous shunting effect, on average, were most likely to have a greater reduction in CBF following transfusion. This finding provides preliminary data that the dural sinus signal intensities may have prognostic potential for treatment response, however, additional data in more participants is required to rigorously confirm this possibility.

Both rapid red cell transit and capillary flow heterogeneity ¹³ have been hypothesized to contribute to a range of neurological conditions, including neurodegeneration ¹⁰, cerebrovascular disease ¹¹, and headache ¹⁶. Importantly, these studies have elucidated the potential role of capillary transit on oxygen delivery and consumption, however, also utilized dynamic susceptibility contrast (DSC) MRI and exogenous contrast, whereas the approach here does not require exogenous agents. The ASL MRI method used here also does not yet allow for quantitative estimates of capillary transit, however, multi-delay ASL performed over arterial and venous structures is being developed and should enable this possibility. Finally, additional data may also relate the venous hyperintense signal to capillary permeability, which can be modeled using the Renkin-Crone model, and which also depends on accurate measures of the capillary radius and flow velocity ⁴². It is not possible to reliably model these parameters using single-delay ASL data, however, future work will benefit from additional measurements which may render this analysis possible with sufficient rigor.

Alternative explanations and limitations

Alternative physiological origins for the dural sinus hyperintensity should also be considered. Specifically, dural sinus hyperintensity on ASL most fundamentally indicates incomplete labeled blood water and unlabeled tissue water exchange at the capillary exchange site. While prior work in SCD has shown that this contrast is associated with higher flow velocities and CBF, and as such rapid red cell transit is a likely explanation, this could also be interpreted as altered water exchange through blood-brain barrier dysfunction in the absence of arterio-venous shunting. This alternative explanation has been suggested in pediatric patients with SCD ⁴³. In this study, this explanation is unlikely as there is no evidence to suggest that a blood transfusion would alter blood-brain barrier permeability, and we observe significant reductions in dural sinus signal intensity post-transfusion. Additionally, blood water relaxation times will vary with hematocrit and oxygenation level ⁴⁴. ASL is minimally T₂*-weighted owing to the short echo time and normalized difference magnetization calculation implemented here. However, changes in T₁ owing to differences in hematocrit pre- versus post-transfusion and between participants may contribute. Given that the transfusion-induced hemoglobin change observed was only 1–2 g/dL, only a small variation in blood water T₁ would be expected ²⁶, which would be unlikely to explain the findings here. Differences due to arterial blood arrival time may be worthwhile of consideration; however, the elevated flow velocities present in participants with SCD and lack of cervical steno-occlusion in most participants with SCD ⁴⁵ suggests that delayed blood arrival is unlikely in most patients. A related issue is that the labeling efficiency in pCASL will depend on the flow velocity and geometry of the labeled vessels relative to the labeling plane. In our acquisitions, we plan the labeling plane orthogonal to the feeding cervical vessels. We also include labeling efficiencies that have been measured in patients with SCD based on prior work. It is possible that some variation in flow velocity between patients may lead to a difference in the amount of label bolus generated, which would affect the CBF measurement by pCASL as well as potentially venous hyperintensity signal. This potential confound, however, also occurs in healthy adults and other patient populations (e.g., cervical vasculopathy) and in these cohorts dural sinus hyperintensities are not observed as in SCD. Therefore, we do believe that the presented findings are more specific to the microvasculature and to sickle cell physiology. Nonetheless, the present findings should be considered in the context of these potential additional contributors.

The findings of this study should also be considered in the context of several limitations. First, the sample size of 32 SCD transfused participants was modest, which did not allow for a combined analysis using all time points and both groups to be evaluated with sufficient statistical power. However, participants were well-characterized by neuroimaging, and pre- and post-intervention scans were performed at similar times across participants. The sample size does preclude evaluating multiple covariates on more sophistical analyses, which may be the focus of future multi-site studies. Second, our distribution of participants’ sex was 38% male in the non-transfused group and 53% male in the transfused group. In view of the equal prevalence of SCD across sexes, sex was slightly imbalanced between groups. We performed pair-wise comparisons (rather than group-wise comparisons) to reduce this confound, however, our study was not sufficiently powered to investigate whether responses differed by sex. Third, we chose to incorporate both adults and children with SCD in our analysis. Further studies that separate adults and children into larger, distinct cohorts will permit evaluation of differences in effects of blood transfusions on CBF, flow signal and efficiency of tissue oxygenation with respect to age. Finally, as discussed above the quantification of OEF from TRUST requires knowledge between MRI relaxation times, hematocrit, and blood oxygenation. While these calibrations have been performed ex vivo by multiple groups, many assumptions still remain and different groups have reported negligible or large effects of HbS, relative to HbA, on MRI relaxation times. Additional measurements from ¹⁵O-PET or from MR susceptometry-based blood oximetry ⁴⁶, which have fewer or different assumptions, may help to corroborate these findings in the future.

Conclusion

We performed arterial spin labeling and dural sinus signal analysis in adults and children with SCD before and after red cell exchange transfusions, together with independent measures of oxygen extraction and tissue anatomy. These data provide evidence that transfusion-induced increases in hemoglobin yield imaging contrast consisted with a lengthening of capillary transit times, which may affect oxygen extraction.

Abbreviations

2D

Two-dimensional

3D

Three-dimensional

ASL

Arterial spin labeling

CBF

Cerebral blood flow

CMRO₂

Cerebral metabolic rate of oxygen

CSF

Cerebrospinal fluid

DSC

Dynamic susceptibility contrast

EPI

Echo planar imaging

FLAIR

Fluid attenuated inversion recovery

FSL-FAST

FMRIB Software Library-FMRIB’s Automated Segmentation Tool

GM

Gray matter

HbAA

Hemoglobin AA

HbF

Hemoglobin F

HbSS

Hemoglobin SS

HbSβ0

Hemoglobin-Sβ0-thalassemia

Hct

Hematocrit

ICA

Internal carotid artery

MRA

Magnetic resonance angiography

MRI

Magnetic resonance imaging

OEF

Oxygen extraction fraction

pCASL

Pseudocontinuous arterial spin labeling

PET

Positron emission tomography

ROI

Region of interest

SCA

Sickle cell anemia

SCD

Sickle cell disease

SCI

Silent cerebral infarct

TE

Echo time

TILT

Transfer insensitive labeling technique

TR

Repetition time

TRUST

T₂-relaxation-under-spin-tagging

Y_a

Arterial oxygen saturation

Y_v

Venous oxygen saturation