Absolute Cerebral Blood Flow Quantification with Pulsed Arterial Spin Labeling During Hyperoxia Corrected with the Simultaneous Measurement of the Longitudinal Relaxation Time of Arterial Blood

Introduction

There is significant interest in the use of inhaled hyperoxic gas as a contrast agent in T₂*-weighted images (1,2) and T₁-weighted images (3), where the respective contrast mechanisms are the modulation of the local deoxyhemoglobin (dHb) concentration (the BOLD effect) and the paramagnetic effects of molecular oxygen itself (4). Hyperoxic contrast has a range of applications, including the calibration of quantitative BOLD fMRI data (5), the quantitative measurement of regional capillary-venous cerebral blood volume (CBV) (6), the partial pressure of oxygen in body fluids (7), and the characterization of ischemic lesions (8). As a contrast agent, oxygen inhalation is attractive because it is minimally invasive, widely available, has few contraindications, and exhibits fast washout times (9). However, hyperoxic gas inhalation, through increases in the arterial partial pressure of oxygen (PaO₂) and hemoglobin saturation (SaO₂), exerts significant effects on the vasculature and is not physiologically inert. Many studies have shown that oxygen has reduces cerebral blood flow (CBF) due to at least one independent vasoconstrictive mechanism, including the attenuation action of nitric oxide, an endogenous vasodilator (10-13). While the vasoconstrictive effect of oxygen is well known, the degree to which CBF changes occur during hyperoxia remains the subject of ongoing research. Moreover, if inhaled oxygen is to be used as a quantitative contrast agent, precise characterization of its changes to the underlying physiology is crucial.

Several studies have used arterial spin labeling (ASL) approaches to investigate the regional CBF changes during hyperoxia (9,11,12,14-16). Although it is well-known that inhaled oxygen creates a significant reduction in the T₁ of arterial blood (T_1a) (4,17), and that T_1a has a substantial effect on CBF measurements using ASL (18), only a small number of these studies have incorporated T_1a changes in their CBF calculations (9,15). For some of these studies, the T_1a changes during hyperoxia were assumed from the results of prior studies of arterial blood at 100% oxygen inhalation (15). Other studies modeled the expected change in T_1a during hyperoxia at intermediate hyperoxia levels, since there are no established values of T_1a in the literature for various oxygen concentration levels (9). These approaches may be problematic for several reasons. First, the methods used in these studies (4,17) to measure T_1a are prone to several artifacts and errors. These include the inflow of fresh, un-inverted spins from outside the transmit coil, partial-voluming, pulsatility, and motion. Second, even under equivalent conditions of hyperoxia, T_1a changes may vary significantly between individuals due to physiologic variations in the blood. Third, it is not clear how to effectively model the expected T_1a changes at arbitrary oxygen concentration levels, since T_1a may be influenced by various factors other than the concentration of molecular oxygen in the plasma (e.g., dHb concentration).

In this study, our aim was to simultaneously measure CBF and T_1a in vivo during various levels of hyperoxia. This will allow for an accurate correction of CBF for T_1a on a per-subject basis. In particular, we wanted to explore the finding reported in previous studies (9,15) that CBF change measured during hyperoxia is dominated by a reduction in T_1a. To achieve these measurements, we implemented pulsed ASL (PASL) approaches for both the measurement of T_1a (19) and CBF that were very similar in implementation. This allowed for a simultaneous measurement of these parameters in the brain without moving the subject. These measurements, however, are presented with technical challenges that complicate their implementation in human studies. These include the requirement that the transmit coil is larger than the subject to avoid the contamination of the measurement from inflowing spins (19), in addition to the long scan times inherent to inversion-recovery measurements of T₁. To solve these problems, adult rats were chosen as a model organism, since they experience similar physiological responses to humans in blood and brain tissue during hyperoxia.

Materials and Methods

Animals

All data were obtained on adult male Sprague-Dawley rats (n=15; 427-476 g; Charles River, MA, USA). All studies were conducted in compliance with our institutional animal care and use committee. Anesthesia was induced with 3% isoflurane in oxygen for approximately five minutes, and the animals were then maintained on spontaneous inhalation anesthesia consisting of 1.5% isoflurane in 30% oxygen delivered at 1.5 L/min through a close-fitting nose cone. If arterial blood gases were to be drawn, an arterial catheter was surgically inserted using a cut down of the ventral surface of tail. If the animal showed any signs of movement at any point in the experiment, the anesthesia was increased to 1.7-1.8% isoflurane for the remainder of the experiment. The fraction of inhaled oxygen (FiO₂) was kept at a minimum of 0.3 instead of 0.21 (oxygen concentration at 1 ATM) to ensure that physiologic stability was maintained under anesthesia and to prevent arterial desaturation. In this study, normoxia refers to an FiO₂ = 0.3, and hyperoxia refers to an FiO₂ = 1.0 (unless otherwise indicated). All gases were delivered to the animal with manually controlled oxygen and nitrogen flow meters, which were mixed and run through an inline isoflurane vaporizer. A temperature-controlled water blanket was placed under the rat to maintain body temperature at 37.0 ± 0.2 °C; temperatures were measured continuously with an optical probe inserted in the rectum. Head movement was restricted during imaging using a custom built MR-compatible animal cradle with attached ear bars.

Hyperoxia Challenge Paradigms

The response of CBF to oxygen was studied under two different hyperoxia paradigms: (1) a brief hyperoxic challenge (Group I) and (2) an increasing, graded hyperoxic challenge (Group II), as illustrated in Fig. 1. All data acquired during the first two minutes after switching gases were not analyzed to allow for physiological adjustment to the new gas condition. To investigate possible effects unrelated to variations in inspired oxygen (e.g. scanner drift, duration of anesthesia, etc.), data were collected for control animals (n=2) using the same measurement protocols of Group I and II described below without switching gas conditions (i.e., under continuous normoxia).

FIG. 1.

Hyperoxic challenge paradigms used for Group I and II animals. Group I animals underwent a baseline – stimulation – rest paradigm (5 – 10 – 5 min) to assess the effects of a brief hyperoxic inhalation challenge on CBF and T_1a. Group II animals experienced a graded hyperoxic challenge to test the effects of longer oxygen exposures on CBF and T_1a. Each gas inhalation period lasted approximately 40 minutes. To minimize temporal effects, two CBF measurements were placed between two T_1a measurements. In addition, in Group I animals, a graded hyperoxic inhalation was used to assess T_1a at different oxygen concentrations.

In Group I (n=7), each rat underwent a baseline – stimulation – rest paradigm. After a five minute baseline period, gases were switched from 30% O₂ to 100% O₂ for ten minutes of hyperoxia, and then were switched back to 30% for a five minute baseline. This paradigm was repeated three times during continuous measurement of T_1a and CBF (detailed below). In addition, a second experiment was performed in Group I animals to determine the degree of T_1a change with hyperoxia. Animals underwent graded hyperoxic inhalation epochs of 40%, 60%, 80%, and 100% (in increasing order) lasting approximately ten minutes each during which continuous measurement of T_1a were made.

In Group II (n=6), both T_1a and CBF were measured with graded hyperoxic inhalation epochs of 30%, 40%, 60%, 80%, and 100% (in increasing order). Each measurement of T_1a or CBF lasted approximately ten minutes each, with two measurements of CBF occurring between two measurements of T_1a to minimize temporal effects. After gases were switched, approximately ten minutes elapsed before the start of the measurements to allow for physiological equilibration to the new gas condition.

For Group II, arterial blood gases were drawn from a subset of animals (n=4) during the imaging experiment immediately after each inhalation epoch. For Group I, blood gases were also analyzed from a subset of animals (n=4), but samples were drawn on a subsequent day to eliminate possible effects of the blood draw on the imaging experiment (20). Arterial blood gas measurements were not considered to be a significant issue for Group II animals, as there was adequate recovery time after each blood withdrawal (8-9 min). All blood gas samples were measured using disposable cartridges and a hand-held blood gas analyzer (i-STAT System; Abbott Laboratories, Abbott Park, IL, USA).

MRI Hardware

All imaging experiments were performed using a whole-body clinical 3T MRI scanner (Siemens Trio; Siemens Healthcare, Erlangen, Germany). RF excitation pulses were transmitted with the scanner body coil, and MR signals were received with a custom built single loop (35 mm ID) receive-only head coil passively decoupled during transmit. The head coil was fixed firmly over the dorsal aspect of the head of the animal.

CBF Quantification with ASL

A pulsed ASL technique developed by Wong and Luh (21,22), quantitative imaging of perfusion using a single subtraction with thin-slice TI₁ periodic saturation (Q2TIPS) based on proximal inversion with control for off-resonance (PICORE) tagging, was used in this study for regional quantification of CBF. This method is insensitive to spatially varying transit delay (δt) and flow-through effects (23), which have been shown to be significant sources of systematic error in perfusion quantification in rats (24).

A multi-shot fast spin echo sequence (FSE) was used for data acquisition. The inversion pulse for tagging consisted of a 40 ms adiabatic hyperbolic secant pulse with a 100 mm slab thickness (gradient turned off for the control image) positioned 10 mm proximal to the inferior imaging slice. To saturate the tag, a series of 90° three-lobe sinc pulses with 20 mm slab thickness, positioned 10 mm proximal to the inferior imaging slice were used for periodic saturation. The inferior saturation pulses were 20 mm thick and spaced 25 ms apart, creating a cutoff velocity of 80 cm/s (22). This cutoff velocity was considered sufficient to saturate inflowing spins because peak systolic velocity in normal rat carotid arteries measured with ultrasound are typically found to be less than 60 cm/s (25). Periodic saturation was continued until image acquisition to ensure full saturation of intravascular spins during imaging. The imaging parameters were: TI₁ = 900 ms, TI_1S = 1475 ms, TI₂ = 1500 ms, TE/TR = 6/4000 ms, slice thickness = 2 mm, slice gap = 2 mm, no. of slices = 3, FOV = 40 × 40 mm, matrix size = 64 × 64, no. of segments = 5, phase partial Fourier factor = 5/8, echo spacing = 6.03 ms, and time between slice acquisitions = 54 ms. The phase encode direction was set left-to-right to avoid pulsation artifacts from the carotid arteries beneath the brain. ASL tag/control pairs were obtained in 40 s (TR = 4 s × 5 segments × 2). The equilibrium magnetization scan was acquired with the exact same imaging parameters, except there were no saturation or inversion pulses, and TR was set to 15 s.

The ASL signal difference ΔM is independent of δt (21) and the signal can be expressed by:

$$\mathrm{\Delta }M=2\cdot \alpha \cdot M_{0a}\cdot f\cdot T{I}_1\cdot e^{-T{I}_2/T_{1a}}$$ [1]

$$\text{for}T{I}_1<\sigma \text{and}T{I}_2>T{I}_1+\delta t$$ [2]

where α is inversion efficiency (= 1 for perfect inversion), M_0a is the fully relaxed longitudinal magnetization of arterial blood, f is cerebral blood flow in ml blood/g tissue/min, T_1a is the longitudinal relaxation time of arterial blood, and σ is the time duration of tag. Since the inversion pulse covered the entire rat heart and lungs, the tag did not have an end; typical values of δt for vessels of the rat neck to the brain are less than 400 ms (24). Therefore, the parameters used here should meet the conditions of Eq. [2] for all imaging regions. Using an agarose phantom, the inversion efficiency was determined experimentally to be approximately 0.97. Although changing oxygen concentrations along the vascular tree and spin exchange with different tissue compartments make the use a single value T_1a a somewhat oversimplified approach (26), the model in Eq. [1] is a reasonable approximation since the tagged spins will spend the majority of TI in the arteries and arterioles. Furthermore, this model is important to consider since it represents the maximum degree of change in measured CBF could be expected from a change in T_1a (see Discussion).

The value of magnetization of arterial blood was estimated on a per voxel basis using the local transverse magnetization of the reference scan and scaling it to the blood-brain partition coefficient (18,27). This method of estimating local M_0a has the advantage of simultaneously correcting for inhomogeneities of the receive coil (28). Correcting for the proton density and the relaxation rate of the local transverse magnetization and arterial blood, the magnetization of arterial blood can be expressed as:

$$M_{0a}=R\cdot M_0\cdot e^{(1/T_{2LT}-1/T_{2a})\cdot TE}$$ [3]

where R is the signal ratio of local transverse magnetization to blood in a proton-density weighted image, M₀ is the voxel signal intensity, T_2LT is the T₂ of the local transverse magnetization, and T_2a is the T₂ of arterial blood. An approximate value of R=0.98 was taken from human studies (29). The T₂ of human gray matter at 3T (80 ms; (30)) was used to estimate a value of T_2LT, and an approximate value of T_2a (100 ms) was determined from the literature (31). Because of the minimal T₂-weighting of our sequence (TE = 6 ms), the expected oxygenation changes occurring the arteriole and capillary spaces during hyperoxia are expected to yield changes in T_2a that would change the measured signal by less than 1% (4,17,31). Therefore, we assumed a single estimate of T_2a for normoxic and hyperoxic states.

T₂-weighted anatomical images were acquired with a multi-shot FSE sequence with the same slice prescription as the ASL sequence, with the following parameters: TE/TR: 81/4000 ms, slice thickness = 2 mm, slice gap = 2 mm, no. of slices = 3, FOV = 40 × 40 mm, matrix size = 192 × 192, no. of segments = 24, and no. of averages=6.

Measurement of Arterial Blood T₁ (T_1a)

A pulsed ASL method based on the work of Thomas, et al. (19) was used to measure T_1a in each animal. A schematic of the pulse sequence is shown in Fig. 2. The same PICORE inversion described above was used to label arterial blood. The inversion was preceded by an in-plane saturation pulse to limit the interaction of the inversion pulse with the static tissue spins in the imaging region. The approach used by Thomas, et al. (19) employed a global saturation pulse to prepare the magnetization along with a flow-sensitive alternating inversion recovery (FAIR) ASL sequence. We used a global inversion pulse instead of a saturation pulse to prepare the magnetization, which has the advantage of increased dynamic range. Furthermore, the use of the PICORE ensures that there is no contamination with venous blood in the measurement, as is possible when using FAIR. It is important to note that because the transmitter coil is much larger than the rat, there is no potential for error in the measurement due to the inflow of uninverted blood into the imaging region. Image acquisition was performed with a single-slice, single-shot FSE sequence. Imaging parameters were: TE/TR: 25/8000 ms, slice thickness = 8 mm, FOV = 64 × 64 mm, matrix size = 64 × 64, and phase encode direction = left-to-right.

FIG. 2.

Pulse sequence diagram of the PICORE sequence with a global inversion preparation. An in-plane presaturation pulse was followed by a global hyperbolic secant inversion pulse. After a variable inversion preparation time, τ, a second in-plane saturation pulse was followed by either a slab-selective hyperbolic secant pulse inferior to the imaging slice (tag) or the same pulse without the slab-selective gradient (control). Control and tag pulses were interleaved for each acquisition. After an inversion time TI, a single-shot FSE acquisition was performed. By keeping TI fixed (1.5 s) and using a series of values of τ, a value of the T₁ of arterial blood (T_1a) was obtained.

If TI is kept at a constant value while the inversion preparation time τ is varied, the PICORE signal difference can be expressed as:

$$\mathrm{\Delta }M=A\cdot (1-2\cdot \alpha \cdot e^{\tau /T_{1a}})$$ [4]

where A is a voxel-specific constant independent of τ. To allow for a robust, overdetermined fit to Eq. [4] in each animal, ten values of τ (65, 400, 800, 1200, 1600, 2000, 2400, 3200, 4000, and 6000 ms) were used with a constant TI of 1500 ms. A randomized acquisition order was used, and tag and control images were interleaved for each value of τ. Ten repetitions of each value of τ were obtained with a total imaging time of 27 min. To increase the number of T_1a measurements during hyperoxic challenges, the same approach was used, except only four values of τ (65, 900, 2500, and 6000 ms) were obtained. To increase the robustness of the fit, these data were only fit for A and T_1a, with the value α taken from the fit from the previous experiment.

Phantom Study of Oxygen Longitudinal Relaxivity

To characterize T₁ as a function of dissolved oxygen concentration, a phantom study was performed by bubbling in a mixture of nitrogen and oxygen into phosphate-buffered saline (PBS) in a centrifuge tube held at 37° C in a water bath. After allowing ten minutes to allow the gas to fully dissolve in the solution, an air-tight lid was secured, and the tube was immediately transferred to a foam-insulated holder positioned inside the magnet. Since the measurement of T₁ lasted only approximately three minutes, it was not necessary to heat the phantom while it was in the magnet. An inversion-prepared single-shot FSE sequence was used with ten inversion times (90, 200, 400, 800, 1200, 1600, 2400, 4000, 6000, 1000 ms) with a TR = 20 s. Temperature measurements were performed before and after, and all temperature decreases were less than 0.5° C. This process was repeated for 0 to 100% oxygen (0 to 760 mm Hg) in steps of 10%. To investigate the possible influence of blood proteins on the relaxivity of molecular oxygen, this experiment was repeated with 5% bovine serum albumin in PBS. All measurements were repeated five times using separate samples.

Data Analysis

All images were analyzed using routines and scripts written in MATLAB (MathWorks, Inc.). For analysis of the T_1a data, whole brain ROIs were drawn on the PICORE signal difference images (see Fig. 3) and the mean values were determined. The acquisition using ten values of τ was fit to Eq. [4] for α, A, and T_1a using a non-linear least squares fit (MATLAB function lsqcurvefit). The value of α determined from this fit was used in the fit of data from the acquisition using four values of τ, so that only A and T_1a were fit to Eq. [4]; also, the values from the prior fit were input as starting values for the subsequent fit. Before this analysis, the T_1a data for each animal and gas condition in Group I and Group II were compiled. For Group I T_1a data, the difference in T_1a between normoxic and hyperoxic states across all animals was compared with a paired Student's t-test. A one-way analysis of variance (ANOVA) was used to compare the difference in T_1a values between different gas conditions pooled from Groups I and II.

FIG. 3.

A representative set of PICORE signal difference ΔM (control minus tag) images in arbitrary units (a.u.) from the pulse sequence shown in Fig. 1. Note that the ΔM signal recovered globally through zero from an initial negative value as the inversion preparation time τ increased. The mean value from a whole brain ROI (outlined in white in the control image) was used to analyze the signal behavior.

To analyze the difference in T_1a in the blood versus PaO₂, the effects of molecular oxygen and deoxyhemoglobin (dHb) were considered. R₁ (1 / T₁) is known to increase linearly with increasing dHb concentration (32). Therefore, the R₁ of blood versus PaO₂ (longitudinal relaxivity of blood as a function of PaO₂, or r₁^PaO2) was modeled as the addition of the linear longitudinal relaxivities of blood as a function of molecular oxygen concentration ([O₂]; determined by blood gas PaO₂) and dHb concentration (1-Y, where Y is the fraction of oxyhemoglobin to total hemoglobin; determined by P₅₀ and blood gas PaO₂), which can be expressed simply as:

$$r_1^{\mathit{\text{Pa}}{O}_2}=r_1^{[O_2]}+r_1^{[\mathit{\text{dHb}}]}$$ [5]

The experimental R_1a values were fit to this model using a non-linear least squares fit in MATLAB.

CBF maps were generated for each gas condition for all animals in Group I and II using the T_1a measured during normoxia (T_1a uncorrected), and another CBF map was generated for the hyperoxic condition using the value of T_1a measured in that animal for the given level of hyperoxia (T_1a corrected). CBF data were analyzed regionally for normoxia and hyperoxia (FiO₂ = 0.3 and 1.0) and whole brain averages for all gas conditions in both experimental groups. Regional analysis was performed because several studies have suggested that there are significant regional variations in the reduction of CBF during hyperoxia (9,10,15). Using the anatomical images, regions-of-interest (ROIs) were manually drawn using a standard rat brain atlas (33) and were then transferred to the perfusion maps. The differences across all ROIs between the two gas conditions were measured using a two-way ANOVA with replication. If the ANOVA showed significant differences in the means between the two gas conditions across all ROIs, analysis of the source of the differences using paired Student's t-test between individual ROIs was considered to be warranted. In all cases, a value of P < 0.05 was considered a statistically significant difference.

Results

The arterial blood gas parameters from Group I and Group II are shown in Tables 1 and 2, respectively. Data in Group I show a statistically significant increase in PaO₂ (402.3 mm Hg), producing a 7% increase in SaO₂. Data from Group II were in close agreement with Group I, showing a very similar increase in PaO₂ (401.9 mm Hg), with increase in PaO₂ in the intermediate steps closely approximating the relative change in FiO₂. A mild but significant increase in the arterial partial pressure of carbon dioxide (PaCO₂) was measured from FiO₂ = 0.3 to 1.0 in Group I (2.3 mm Hg) and Group II (3.2 mm Hg). Mean arterial blood pressure (MABP) (measured in Group I only) and pH were not found to change significantly.

Table 1. Arterial Blood Gas Parameters for Normoxia and Hyperoxia in 1.5% Isoflurane Anesthesia (Group I).

FiO2	pH	PaCO2 (mm Hg)	PaO2 (mm Hg)	SaO2 (%)	MABP (mm Hg)
0.3	7.44 ± 0.03	45.3 ± 1.8	97.0 ± 8.3	92.9 ± 1.3	73.5 ± 4.9
1.0	7.43 ± 0.04	47.6 ± 1.6*	499.3 ± 32.2*	99.9 ± 0.1*	73.9 ± 5.3

^*P < 0.05 from normoxia.

Table 2. Arterial Blood Gas Parameters for Graded Levels of Hyperoxia in 1.5% Isoflurane Anesthesia (Group II).

FiO2	pH	PaCO2 (mm Hg)	PaO2 (mm Hg)	SaO2 (%)
0.3	7.45 ± 0.03	44.7 ± 1.2	99.3 ± 11.6	93.0 ± 2.1
0.4	7.44 ± 0.03	45.3 ± 1.0	154.3 ± 13.6	97.7 ± 0.6
0.6	7.44 ± 0.02	46.1 ± 0.8	266.8 ± 16.8	99.4 ± 0.1
0.8	7.44 ± 0.01	46.7 ± 0.6	388.8 ± 13.4	99.8 ± 0.0
1.0	7.43 ± 0.02	47.9 ± 0.8	501.2 ± 23.0	99.9 ± 0.0

A representative set of ΔM images during normoxia with ten values of τ is shown in Fig. 3. As predicted from the model, the signal increased exponentially from an initial negative value of ΔM and changed globally to the final expected positive value of ΔM as τ increased. The values of ΔM around the periphery of brain are reduced due the partial voluming from the large voxel size. However, these effects remain the same across all values of τ and do not affect the calculation of T_1a.

Figure 4a shows R₁ versus pO₂ in the PBS calibration phantom, and Fig. 4b shows R₁ versus PaO₂ from the pooled data from Groups I and II. Using a linear regression, the longitudinal relaxivity of molecular oxygen in PBS was found to be 1.61 ± 0.02 × 10^-4 s^-1 mm Hg^-1, with R² = 0.99. No statistically significant difference was found in the relaxivity of the phantom containing bovine serum albumin. Using the fit to the model shown in Eq. [5] above, the r₁^[O2] was determined to be 1.59 ± 0.21 × 10^-4 s^-1 mm Hg^-1, while the r₁^[dHb] was found to be 0.246 ± 0.051 s^-1 (1-Y)^-1. There was found to be a clear influence of dHb on T_1a values measured between FiO₂ = 0.3 and 0.6, with the T_1a at FiO₂ = 0.4 actually showing a slight increase. At higher FiO₂ levels, the longitudinal relaxivity with PaO₂ appeared to be linear. The values of T_1a were found to be 1632 ± 37 ms, 1641 ± 35 ms, 1606 ± 33 ms, 1549 ± 24 ms, and 1514 ± 27 ms for FiO₂ = 0.3, 0.4, 0.6, 0.8, and 1.0, respectively. The values of T_1a measured for animals in Group I and II for each gas condition did not show statistically significant differences. There was no statistically significant change in the T_1a measured during FiO₂ = 1.0 between sections (a) and (b) for the Group I paradigm (see Fig. 1).

FIG. 4.

Effect of oxygen concentration on R₁ in a phosphate-buffered saline (PBS) phantom (a) and effect of arterial oxygen tension on R₁ of arterial blood (b). The marker indicates the experimental data, and the lines indicate the fits to the data. The effect of oxygen on R₁ in (a) was shown to be well-described by a linear fit to the data. Data shown in (b) are pooled from Groups I and II (n=13). The data were well described by a model incorporating the linear relaxivity effect of both molecular oxygen and deoxyhemoglobin. Values shown are mean ± SD.

An anatomic dataset from a representative animal from Group I is shown in Fig. 5, along with outlines of the ROIs used in the analysis of the perfusion maps. Quantitative CBF maps during normoxia and hyperoxia are shown below. A reduction in calculated CBF values in several regions across the brain can be clearly visualized. Correcting for T_1a measured during hyperoxia clearly reduced the degree of the differences between normoxia and hyperoxia, but reduced CBF values can still be visualized. To demonstrate the degree of the CBF reduction due to hyperoxia in our experimental setup, Fig. 6 shows the time course of the average ΔM across all the ROIs during the hyperoxic challenges for a representative animal. The signal intensity of the control image was modulated by the BOLD signal; although the signal change was small due to the short TE, it clearly shows the effects of inhaled oxygen across the experiment. Overlays of the measured reduction (in percent decrease) on anatomical images of the CBF maps corrected for T_1a from Group II are shown in Fig. 7a. The observed CBF reduction seems to be somewhat regional in nature, with larger effect in the cortex compared to other regions. The effect of FiO₂ on whole brain CBF measured from Group II animals is shown in Fig. 7b. CBF shows a consistently decreasing trend with time and increasing FiO₂.

FIG. 5.

T₂-weighted structural images with ROIs (outlined in white) for analysis of regional CBF values (corresponding columns below), from a representative animal. ROI values from both hemispheres were averaged. In the left column, the inferior slice (bregma -6.5 mm) contains (1) visual/auditory cortex and (2) hippocampus/subiculum. In the middle column, the middle slice (bregma -2.5 mm) contains (3) sensory/auditory cortex, (4) hippocampus, and (5) thalamus. In the right column, the superior slice (bregma 1.5 mm) contains (6) motor/sensory cortex and (7) caudate putamen. CBF values (in ml/100 g/min) in the second row were calculated during normoxia (FiO₂ = 0.3) from Eq. [1] using T_1a value measured during normoxia, while CBF values in the in third and fourth row were calculated during hyperoxia (FiO₂ = 1.0) using the T_1a measured during normoxia (uncorrected) and hyperoxia (corrected).

FIG. 6.

Representative signal time course of ΔM (control minus tag) and control image signal intensity during normoxia (FiO₂ = 0.3; white regions) and hyperoxia (FiO₂ = 1.0; green regions). Signal values were calculated at each time point from the mean of all ROIs (see Fig. 4). Units were calculated as the percent difference from starting baseline. Control image signal intensity is increased during hyperoxia due to BOLD response; percent signal change is low (∼0.5%) due to the degree of T₂-weighting (TE = 6 ms).

FIG. 7.

Overlays of the measured reduction (% decrease) on anatomical images of the CBF maps from a representative animal (a) and the effect of FiO₂ on whole-brain relative CBF. Both sets of data come from Group II animals. The data shown (a) have been corrected for the per subject change in T_1a; the observed CBF reduction appear to be regional in nature, with the largest effects observed in the cortex. Relative CBF in (b) shows a consistently decreasing trend with time and increasing FiO₂. The squares and circles represent the data uncorrected and corrected for the change in T_1a, respectively. Values are mean ± SD.

ROI analysis of mean calculated CBF across animals for normoxia and hyperoxia (FiO₂ = 1.0) in Groups I and II is shown in Figs. 8a and b, respectively. CBF was significantly reduced across all ROIs before and after correction for T_1a in both Groups I and II, as determined by ANOVA. CBF decreases were larger in the Group II animals, with all regions still showing statistically significant reductions in CBF after correction for T_1a. There appeared to be some regional nature to the CBF reduction, particularly in Group II animals where, after correction for T_1a, the sensory/auditory cortex region showed a 23.5% reduction, while the caudate putamen region only showed a 10.9% reduction. After T_1a correction, whole brain values of CBF exhibited a normoxia to hyperoxia reduction of 109.2 ± 12.9 ml/100 g/min to 103.9 ± 10.4 ml/100 g/min in Group I and 108.3 ± 11.0 ml/100 g/min to 90.2 ± 12.5 ml/100 g/min in Group II. The correction for the change in T_1a dominated the Group I correction, accounting for 63% of the increase in the observed reduction in CBF. In Group II, the T_1a change contributed significantly less, accounting for 28% of the increase in the observed reduction in CBF.

FIG. 8.

Mean calculated CBF values (ml/100 g/min) in seven ROIs (both hemispheres; see Fig. 4) during normoxia (FiO₂ = 0.3; dark gray bars) and hyperoxia (FiO₂ = 1.0) uncorrected (light gray bars) and corrected (white bars) for T_1a during hyperoxia. Data are were taken from Group I (a) and Group II (b). Values are mean ± SD. * P < 0.05 from normoxia.

Discussion

In this study, T_1a and quantitative CBF were measured in vivo at 3T during normoxia and hyperoxia using PASL approaches. The measured T_1a values were used to correct the calculated CBF in each experimental animal under a brief hyperoxic inhalation and graded hyperoxic inhalation paradigms. The average value of T_1a measured during normoxia shows close agreement to values from previous studies in vivo (1623 ms; (34)) and in vitro (1664 ms; (35)). The measured value of T_1a in response to oxygen was shown to closely agree with phantom data after accounting for the effect of changing dHb concentration on T_1a. The measured relaxivity of dHb was found to be in agreement with previous results (32,35). The reduction in T_1a due to oxygen was observed to be rapid and consistent across large differences in the duration of oxygen exposure. The changes in CBF, however, were significantly different depending on the oxygen exposure duration. The longer exposures of the slow, graded hyperoxic inhalation paradigm of Group II produced significantly more reduction in CBF. This is in agreement with established effects of hyperoxia on cerebral blood flow, as the attenuation of effects of nitric oxide effects have been shown to depend on the duration of hyperoxic exposure (13). Overall, the results of this study suggest that the observed reduction in CBF measured by ASL are dominated by T_1a reduction during short term inhalation epochs, but physiologic effects become the primary source of the reduction over longer hyperoxic exposures.

Although the measured longitudinal relaxivity of T_1a as function of PaO₂, and therefore the CBF correction, determined in this study were found to be significantly smaller than those used in previous studies (9,15), we have grounds to be confident in the findings of our study. The observed T_1a changes of our in vivo data very closely approximated the more robust phantom data. Perhaps more significantly, the correction made in this study most likely represents the maximum degree to which the reduction in T_1a would decrease the measured CBF reduction. We have assumed that spins measured in the ASL experiment have spent the vast majority of TI (inversion time) in a compartment that approximates arterial blood. However, toward the end of TI, a significant portion of tagged spins will enter the capillary and tissue spaces. In these regions, the change in pO₂ during hyperoxia will be substantially lower (36), yielding a smaller change in T₁. In addition, in the capillary space, the mild dilution of deoxyhemoglobin is likely to yield a slight increase in T₁. In order to accurately account for these effects, it will be necessary to develop a model of the average change in pO₂ experienced by the tagged spins over the duration of TI. It will also necessary to know how the relative amount of spins in each compartment as a function of time. For these reasons, the ASL signal model used in this study is not complete, and the effective reduction in the T₁ experience by the tagged spins is likely to be significantly less than the measured reduction in T₁ of arterial blood. However, the use of this model is important, since it represents the upper bound to which the reduction in CBF measured can be attributed to a reduction in T₁ of the spins measured by the ASL experiment.

Baseline mean values of CBF across the rat brain during normoxia were in close agreement to literature values (11,16). In agreement with previous studies in humans and mechanically ventilated animals (9,11,14,15), a reduction in regional CBF with hyperoxia was observed. The response to brief inhalation epoch was more uniformly global and appeared to demonstrate less regional variance compared to the longer hyperoxic exposure of the graded inhalation. As mentioned above, this may have physiologic foundation, since T_1a changes by definition are a global effect, but it remains unclear whether a local physiologic effects, like nitric oxide attenuation, exhibit variations that are more regional in nature. Contrary to the results of previous study of hyperoxia in free-breathing rats that showed a mild increase in CBF (12), we found that CBF consistently decreased with hyperoxia. The presumed source of the increase in CBF in the previous study was mild hypercapnia caused to hyperoxia-induced hypoventilation, and we also observed a slight increase in pCO₂ during hyperoxia. However, very short hyperoxic epochs (∼2 min) were used in the previous study (12), which are likely too short to observe the vasoconstrictive effects of hyperoxia (9,11,14). Furthermore, the levels of isoflurane used in the present study were significantly higher (1.15-1.25% versus 1.5-1.8%), and isoflurane is known to substantially suppress the effects of CO₂ on the vasculature (16). We believe these factors explain why the effects of elevated pO₂ on CBF were dominant in the present study. However, the elevation in pCO₂ is a likely confounding factor in this experiment, and if normocapnia was maintained, it is likely that a larger reduction in CBF would have been observed.

Although not a trivial operation, the measurements of T_1a in rat blood made in this study can be translated to human studies. Given the similarity in the constitution of blood between the two species, the longitudinal relaxivity as a function molecular oxygen is likely to be very similar. The most important difference between species is effect of deoxyhemoglobin, which will exist in different concentrations at the same PaO₂ (due to difference in P₅₀) and will likely exhibit different longitudinal relaxivity. However, if these effects can be characterized and accounted for, it should be possible to accurate estimate the change in T_1a during any oxygen challenge, given knowledge of PaO₂ during normoxia and hyperoxia. To this end, it is possible to estimate PaO₂ from end-tidal oxygen concentration, given normal lung physiology (5).

Given the challenges of making in vivo T_1a measurements in humans, the use of small animal model may be the most practical method for T_1a calibration for human experiments. However, a direct measurement of T_1a on a per subject and per condition basis is still a desirable goal for accurate quantification of CBF using ASL, given the relatively large inter- and intrasubject variability in T_1a in baseline normoxia as well as in response to oxygen challenge. Several changes from the present approach are likely to be necessary to obtain this goal, including minimizing the number of acquired inversion preparation times, and the application of the constraint on the duration of the inversion preparation times to values short enough to avoid contamination of spins flowing in from outside the transmit coil.

Conclusion

Simultaneous in vivo measurements of T_1a and CBF were performed using PASL approaches during normoxia and hyperoxia in rat brains at 3T under a brief hyperoxic inhalation and graded hyperoxic inhalation paradigms. Baseline values of T_1a and CBF measured during normoxia were in close agreement with previous studies. Significant reductions in T_1a and CBF were measured during hyperoxia. The results of this study suggest that the measured reduction in CBF using ASL is dominated by T_1a reduction during short term inhalation epochs, but physiologic effects become the primary source of the ASL signal reduction for longer hyperoxic exposures. Furthermore, the inter- an intrasubject variability during normoxia and in response to hyperoxia observed in this study underscores the importance of the accurate measurement of T_1a in quantitative ASL studies.