Non-contrast assessment of blood-brain barrier permeability to water in mice: an arterial spin labeling study at cerebral veins

Abstract

Blood-brain barrier (BBB) plays a critical role in protecting the brain from toxins and pathogens. However, in vivo tools to assess BBB permeability are scarce and often require the use of exogenous contrast agents. In this study, we aimed to develop a non-contrast arterial-spin-labeling (ASL) based MRI technique to estimate BBB permeability to water in mice. By determining the relative fraction of labeled water spins that were exchanged into the brain tissue as opposed to those that remained in the cerebral veins, we estimated indices of global BBB permeability to water including water extraction fraction (E) and permeability surface-area product (PS). First, using multiple post-labeling delay ASL experiments, we estimated the bolus arrival time (BAT) of the labeled spins to reach the great vein of Galen (VG) to be 691.2±14.5 ms (N=5). Next, we investigated the dependence of the VG ASL signal on labeling duration and identified an optimal imaging protocol with a labeling duration of 1200 ms and a PLD of 100 ms. Quantitative E and PS values in wild-type mice were found to be 59.9±3.2% and 260.9±18.9 ml/100g/min, respectively. In contrast, mice with Huntington’s disease (HD) revealed a significantly higher E (69.7±2.4%, P=0.026) and PS (318.1±17.1 ml/100g/min, P=0.040), suggesting BBB breakdown in this mouse model. Reproducibility studies revealed a coefficient-of-variation (CoV) of 4.9±1.7% and 6.1±1.2% for E and PS, respectively. The proposed method may open new avenues for preclinical research on pathophysiological mechanisms of brain diseases and therapeutic trials in animal models.

1. Introduction

Blood-brain barrier (BBB) is formed by endothelial cells and pericytes of the capillary wall (Ballabh et al., 2004; Dickie et al., 2020). These structures in the microvasculature provide a barrier against the free diffusion of many molecules and only allow the passage of specific nutrients, ions, and macromolecules, e.g. amino acids that are critical for normal neural functioning (Lochhead et al., 2020). Therefore, BBB plays an important role in blocking toxins and pathogens from entering the parenchyma, regulating solute transportation, and waste clearance (Persidsky et al., 2006). BBB disruptions have been observed in normal aging (Banks et al., 2021; Dickie et al., 2021; Ohene et al., 2021) and a variety of diseases, including Alzheimer’s disease (Hussain et al., 2021; Lin et al., 2021b), Huntington’s disease(Drouin-Ouellet et al., 2015), cancer (Gerstner and Fine, 2007; Wilhelm et al., 2013), and multiple sclerosis (Correale and Villa, 2007; Cramer et al., 2014).

Dynamic contrast-enhanced (DCE) MRI is a commonly used method for evaluating BBB function by exploiting the T₁ relaxation property of gadolinium when leaked into the tissue (Ivanidze et al., 2019; Shao et al., 2020; Varatharaj et al., 2019). However, leakage of chelated gadolinium is a slow process unless there are severe BBB damages (e.g. tumor). As a result, the sensitivity of DCE MRI in studying diseases with subtle BBB damage at the early stage, e.g. Alzheimer’s disease, is limited. More recently, several approaches based on arterial-spin-labeling (ASL) MRI have been proposed to investigate the BBB’s permeability to water (Li et al., 2005a; Lin et al., 2018; Ohene et al., 2019; Shao et al., 2019; St Lawrence et al., 2012; Wang et al., 2007; Wengler et al., 2019). These methods utilized water spins in the blood as an endogenous tracer and probed the BBB property by assessing the water exchange across the capillary. Due to its small molecular weight, the water-based BBB measurements have been postulated to be more sensitive to the subtle BBB damages in brain diseases (Lin et al., 2021b). At present, the majority of such studies have been carried out on human subjects.

Rodent models (mice and rats) represent important avenues of research in understanding different aspects of human physiology and disease. They provide critical insights into the mechanisms of brain diseases and are used extensively in the preclinical study of efficacy and safety of potential therapeutic interventions (Ericsson et al., 2013). In vivo assessment of BBB permeability to water has not been a routine practice in mice and has only been performed in a limited number of studies (Dickie et al., 2021; Dickie et al., 2019; Ohene et al., 2021; Wells et al., 2013). In this work, we aim to develop a non-contrast MRI technique to estimate BBB permeability to water in mice. This technique utilizes a principle similar to water-extraction-with-phase-contrast-arterial-spin-tagging (WEPCAST) MRI (Lin et al., 2018) that has been applied to humans and determines the relative fraction of labeled spins that were exchanged into the brain tissue as opposed to those that remained in the vein. Technical developments were carried out to optimize the labeling duration and post-labeling duration (PLD) of the sequence. Reproducibility of the measurement was characterized. Finally, an initial application of the proposed technique to a mouse model of Huntington’s disease (HD) was performed to demonstrate the utility of the proposed method in pathological conditions.

2. Materials and Methods

2.1. Theory

According to the Renkin-Crone Model (Crone, 1963; Renkin, 1959), BBB permeability can be defined as the permeability surface-area product (PS) per unit mass of tissue, specifically,

$$PS=-\text{ln}\left(1-E\right)\times \mathit{\text{CBF}},$$ (1)

where E denotes the extraction fraction of water and CBF denotes cerebral blood flow. For the purpose of the present study to quantify global PS, CBF is determined by phase contrast (PC) MRI, which measures bulk blood flows across the major feeding arteries of the brain (Wei et al., 2019).

The main task lies in the measurement of E. If we consider arterially labeled spins that reach the capillary-tissue interface, a portion of the spins will be exchanged into the tissue and the remaining spins will stay in the vessel and be drained to large veins such as the great vein of Galen (VG). The signal in the vein when using a pseudo-continuous arterial spin labeling (pCASL) can be written as:

$$\Delta M_{\mathit{\text{vein}}}\left(t\right)=2\alpha \left(1-E\right)M_{0,\mathit{\text{vein}}}{e}^{-\frac{BA{T}_{\mathit{\text{vein}}}}{T_{1b}}}c\left(t\right),$$ (2)

where α denotes the inversion efficiency of pCASL, M_0,vein denotes the equilibrium magnetization of venous blood, BAT_vein is the bolus arrival time of the vein, T_1b denotes the blood T₁, and c(t) = H(t – BAT_vein + LD) – H(t – BAT_vein) denotes the arterial input function where H(t) is the Heaviside step function, LD is the labeling duration, and bolus dispersion is ignored. If the bolus dispersion is considered, the equation becomes

$$c\left(t\right)=\left[H\left(t-BA{T}_{\mathit{\text{vein}}}\right)-H\left(t-BA{T}_{\mathit{\text{vein}}}-LD\right)\right]\otimes G\left(t\right),$$ (3)

where $G\left(t\right)=\frac{1}{\sqrt{2\pi {\sigma }^2}}{e}^{-t^2/\left(2{\sigma }^2 \right)}$ denotes the Gaussian kernel with variance σ² and ⊗ denotes the convolution operation. Here, we focus on VG because it is a large vein easily visible in the mouse brain, it has few variations across animals (thus can be identified in every mouse), and its flow trajectory is relatively straight. Unlike in human cases, we found that the size of VG in mice is greater than that of the sagittal sinus.

In Eq. (2), blood T₁ was assumed to be 2813 ms at the field strength of 11.7 T (Lin et al., 2012). The labeling efficiency of pCASL was experimentally determined by using PC CBF as a reference, similar to those conducted in a human study (Aslan et al., 2010). BAT_vein can either be determined experimentally for each animal or use an assumed value based on group-averaged data (see “3. Results” for more details). The equilibrium blood magnetization, M_0,long-TR, can be obtained with a scan using a long TR (e.g., 20 s). However, the long-TR signal of a voxel may contain partial volume (PV) effects from tissue spins. To further account for this effect, we performed the following additional scans to determine the PV factor of blood (β), denoting the fraction of blood in the VG ROI.

We applied PC MRI to the VG to isolate pure blood signals without tissue contributions. Since we do not know the exact velocity of VG and it could vary from one voxel to another, we employed a series of encoding velocity (VENC) values, specifically, 2, 4, 6, 8, 10, and 12 cm/s. The complex-difference signal intensity of PC MRI, S_CD, can be given as:

$$S_{CD}=2S_{\mathit{\text{vein}}}\left|\text{sin}\left(\frac{\pi \nu }{2\mathit{\text{VENC}}}\right)\right|,$$ (4)

where S_vein denotes blood signal at vein, v denotes the flow velocity, and VENC is the encoding velocity. We fit the multiple-VENC S_CD signals to Eq. (4) to estimate S_vein. We also performed a gradient-echo (GRE) MRI with identical parameters to the PC MRI but without velocity-encoding to estimate S_total. The ratio between S_vein and S_total provides an estimation of β. M_0,vein can then be estimated as β·M_0,long-TR.

2.2. MRI and immunoassay experiments

All MRI experiments were performed on an 11.7T Bruker Biospec system (Bruker, Ettlingen, Germany) with a horizontal bore as well as an actively shielded pulse field gradient (maximum intensity of 0.74 T/m). The local institutional animal care and use committee approved the experimental protocol. All mice had free access to food/water and were housed in a quiet environment with a 12-h day/night cycle. Anesthesia of mice was induced using 1.5% isoflurane in medical air (78% N₂ + 21% O₂) for 15 min and maintained throughout the experiment using 1.25% isoflurane. At approximately 10^th minute after anesthetic induction, mice were relocated to a water-heated animal bed with temperature control and positioned with a bite bar and ear pins. Images were acquired using a 72-mm quadrature volume resonator as a transmitter, and a four-element (2×2) phased-array coil as a receiver. The homogeneity of B₀ field over the mouse brain was improved with a global shimming (up to 2^nd order) based on a subject-specific pre-acquired field map.

A total of 15 C57BL/6 mice (age: 14–44 weeks; 8 females and 7 males; body weight: 21–40 g) and 28 zQ175 model mice (age: 20–24 weeks; wildtype [WT]: 5 females and 9 males; HD: 6 females and 8 males) were studied in this work, divided into four sub-studies.

2.2.1. Study 1: Estimation of BAT_vein in mice

To estimate E in Eq. [2], it is important to know BAT_vein. There have not been any reports on BAT_vein in mice in the literature (there were only reports on BAT to tissue). Therefore, Study 1 aimed to quantify BAT_vein in mice. Five mice (3 females and 2 males) were scanned. We performed a pCASL MRI (Hirschler et al., 2018; Wei et al., 2021) with a short labeling duration and multiple post-labeling delay (PLD) values. Specifically, the imaging parameters were: TR/TE = 3000/11.8 ms, labeling duration = 500 ms, FOV = 15×15 mm², matrix size = 96×96, slice thickness = 1 mm, labeling-pulse width = 0.4 ms, inter-labeling-pulse delay = 0.8 ms, flip angle of labeling pulse = 40°, two-segment spin-echo echo-planar-imaging acquisition, partial Fourier acquisition factor = 0.7, number of average =10, and scan duration = 2.0 min per PLD value. PLD values were 25, 100, 200, 300, 400, 500, 600, 700, 800, 900 and 1000 ms. The imaging slice was positioned to cover the midline of the mouse brain in the sagittal orientation.

2.2.2. Study 2: Optimization of labeling duration

Eq. [2] predicts that, if the labeling duration is greater than BAT_vein, the signal ΔM_vein should have reached a plateau, i.e., c(t)=1 for 0<PLD<BAT_vein. However, due to the dispersion of the labeled bolus along the vascular tree, the signal plateau will appear at a later time. Study 2 therefore aimed to examine the dependence of the ΔM_vein signal on labeling duration. Five mice (2 females and 3 males) were scanned. A pCASL sequence was performed, in which a fixed PLD of 25 ms was used and a total of 11 labeling durations, i.e. 50, 100, 200, 300, 500, 700, 900, 1100, 1300, 1600, and 2000 ms, were employed. Other parameters of the sequence were identical to Study 1.

2.2.3. Study 3: Assessment of BBB permeability in wild-type mice.

This study integrated the optimized parameters in Study 1 and 2, and conducted a complete assessment of BBB permeability. Reproducibility was also evaluated. Five mice (3 females and 2 males) were utilized in Study 3. A complete BBB MRI protocol consisted of: 1) a multi-PLD pCASL with short labeling duration to measure BAT_vein, 2) a single-PLD pCASL with optimized labeling duration to measure ΔM_vein, 3) a long-TR M₀ scan, and 4) a 4-artery PC scan to measure CBF. Imaging procedures for the PC CBF scan have been optimized and described previously (Wei et al., 2019), the parameters of which were TR/TE = 15/3.2 ms, FOV = 15×15 mm², matrix size = 300×300, slice thickness = 0.5 mm, encoding velocity =20/10 cm/s (ICA/VA), number of average = 4, dummy scan = 8, receiver bandwidth = 100 kHz, flip angle = 25°, partial Fourier acquisition factor = 0.7, and scan duration per artery = 0.6 min (Wei et al., 2020a; Wei et al., 2021). The multi-PLD pCASL scan followed the parameters in Study 1 except that fewer PLD values (150, 300, 450, 600, 750, 900, and 1050 ms) were used to shorten the scan time. The single-PLD pCASL scan followed the parameters in Study 2. The M₀ scan was performed by using the same acquisition module as the pCASL but employed a repetition time (TR) of 20,000 ms. This complete BBB protocol was repeated three times to test the reproducibility of the technique.

As described in the Theory section, the partial volume effect in the M₀ signal was accounted for by performing a series of PC MRI sequences covering VG using multiple VENC values (2, 4, 6, 8, 10, and 12 cm/s). Parameters for these multi-VENC PC MRI were: TR/TE = 15/3.2 ms, FOV = 15×15 mm², matrix size = 96×96, slice thickness = 1 mm, number of average = 4, dummy scan = 8, receiver bandwidth = 100 kHz, flip angle = 25°, partial Fourier acquisition factor = 0.7, and scan duration = 0.2 min for each VENC.

In addition, a T₂-weighted fast-spin-echo MRI protocol (Wei et al., 2020a) (TR/TE = 4000/26 ms, FOV = 15×15 mm², matrix size = 128×128, slice thickness = 0.5 mm, echo spacing = 5 ms, 35 axial slices, and scan duration = 1.1 min) was performed to estimate the brain volume.

2.2.4. Study 4: BBB permeability in a mouse model of HD

A total number of 17 mice (20–24 weeks; HD mice: 4 females and 5 males; WT mice: 3 females and 5 males) were included in the MRI scan of Study 4. The HD model was the heterozygous zQ175 model and original breeders were purchased from The Jackson Laboratory (Liu et al., 2021). This sub-study was designed to test the sensitivity of our proposed method in detecting BBB permeability alteration induced by disease pathology. Parameters of PC MRI and pCASL MRI were identical to Study 3.

A separate cohort of 11 mice (20–24 weeks; HD mice: 2 female and 3 males; WT mice: 2 female and 4 males) were used to determine levels of tight junction proteins, including immunohistochemistry and Western blotting. A pair of WT and HD mice were euthanized following reported procedures (Liu et al., 2021) for immunofluorescence analyses. Brain sections were stained with Claudin-5 antibody (1:200, Invitrogen) (red channel) and DAPI (blue channel) for qualitative examination of alterations in tight-junction proteins. Fluorescence images were acquired with a Zeiss LSM 700 confocal microscope (objective 63×). Moreover, for quantitative evaluation of the tight-junction proteins, 9 mice (4 HD and 5 WT mice) were conducted with Western blotting. Three tight-junction proteins, i.e., ZO-1, Occludin, and Claudin-5, were examined. Specifically, brain samples were homogenized in a buffer containing 50mM Tris-HCl (pH 8.0), 150mM NaCl, 0.1% (w/v) SDS, 1.0% NP-40, 0.5% sodium deoxycholate, and 1% (v/v) protease inhibitors. Proteins (10 μg) were separated in a 4–20% gradient gel and transferred to a PVDF membrane which was activated by the methanol. The membrane was blotted with the following primary antibodies: ZO-1 polyclonal antibody (1:1000, Invitrogen), Anti-Occludin monoclonal antibody (1:1000, Abcam), Claudin-5 monoclonal antibody (1:1000, Invitrogen), and anti-β-actin monoclonal antibody (1:5000, Sigma-Aldrich). After incubation with HRP-conjugated secondary antibodies, bound antibodies were visualized with ECL Plus reagents (Thermo) and imaged on the ChemiDoc Imaging System (BioRad).

2.3. Data processing

All processing was conducted using custom-written MATLAB (MathWorks, Natick, MA) scripts. PC MRI data were processed according to a previous report (Wei et al., 2019). Specifically, the artery of interest was first manually delineated on the complex-difference image, which showed an excellent contrast between vessels and surrounding tissue. The mask was then applied to the velocity map and blood flow (ml/min) through that artery was calculated using the integration of arterial voxels. To estimate brain volume, the T₂-weighted images were analyzed manually by delineating the brain boundary on a slice-by-slice basis while referencing a mouse brain atlas (https://atlas.brain-map.org/) (Wei et al., 2020a; Wei et al., 2020b). Voxels inside the masks were summed to yield the total brain volume in cm³, after which brain weight was calculated with an assumed density (Leithner et al., 2010) (1.04 g/cm³).

For the pCASL data, pair-wise subtraction between control and labeled images (M_ctr − M_lbl) were first applied to yield a difference image (i.e., ΔM). An ROI was manually drawn on the ΔM/M₀ image to encompass the VG. Ten voxels with the highest signal intensities in the ROI were automatically located with MATLAB functions and then averaged to yield a ΔM_vein/M_0. Experimental measures of ΔM_vein/M₀, β, and BAT_vein were used to estimate E based on Eq. [2]. E and CBF were in turn used to estimate PS based on Eq. [1]. Additionally, to test the necessity of measuring BAT_vein individually, we also estimated E and PS using an assumed BAT_vein value based on the group average.

The images of Western blot were analyzed with the Image J software.

2.4. Statistical analyses

Parametric values were reported in mean ± standard error. Linear mixed-effect model was used to examine the time-dependence of physiological parameters across repetitions. Pearson’s correlation was used to investigate the relationship between PS values estimated with individually measured BAT_vein and with assumed BAT_vein based on group average. Student’s t-test was employed to examine the statistical difference between HD and WT mice. In all analyses, a P value less than 0.05 was considered statistically significant.

3. RESULTS

3.1. Study 1

Figure 1A presents a TOF image allowing visualization of the vessels in the mid-sagittal plane of the brain, including the great vein of Galen (Dorr et al., 2007). It can be observed that the VG is the largest vein in this location (e.g., in comparison with the superior sagittal sinus), thus was the focus of this study. Figures 1B and 1C show the control and labeled images of all PLD values. Figure 1D shows the difference images obtained with pair-wise subtraction between Figures 1B and 1C. It can be observed that the signal intensities in VG show dependence on PLD values (red arrows). Figure 1E shows the quantified ΔM/M_control intensities in VG. By fitting the VG signal intensities into Equation (3), it was estimated that BAT_vein=691.2±14.5 ms (N=5). For comparison, signal intensities in brain tissue are shown in Supplemental Figure S1. BAT of tissue was found to be 370.0±21.4 ms, which was significantly shorter than BAT_vein (P<0.001).

Figure 1.

Estimation of BAT_vein in mice. (A) shows the time-of-flight (TOF) image at the mid-sagittal plane of a mouse. (B-D) present the control, labeled, and difference images at different PLD values. (E) shows the dependence of ΔM/M_control signal in VG on PLD values after averaging across mice (N=5). Error bar stands for standard error across mice.

3.2. Study 2

Figures 2A–2C show the control, labeled, and difference images under different labeling durations ranging from 50 to 2000 ms. Figure 2D shows ΔM/M_control in VG at different labeling durations. Visual observation suggested that longer labeling durations generally led to higher ΔM/M_control signals. Fitting of ΔM/M_control signals into Equation (3) yielded a dispersion function with an averaged full-width-of-half-maximum (FWHM) of 806.4±65.4 ms. Based on this FWHM value, simulations suggested that a labeling duration of 1200 ms with a PLD of 100 ms can yield 96% of the maximal possible signal, which is considered a trade-off between sensitivity, scan duration, and SAR. A labeling duration of 1200 ms was used for later studies.

Figure 2.

The effects of labeling duration on VG ASL signal. (A-C) present the control, labeled, and difference images for varying labeling durations. (D) shows the ΔM/M_control signal in VG as a function of labeling duration after averaging across mice (N=5). Error bar stands for standard error across mice.

3.3. Study 3

Figure 3 demonstrates a complete dataset of BBB permeability measurement in mice. Pair-wise subtraction between the control and labeled pCASL images yielded a difference image (Figure 3A), which was then normalized by the M₀ image (Figure 3B) to obtain the percentage signals (Figure 3C). The multi-VENC PC dataset along with a reference GRE image without velocity encoding (Figure 3D) was used to fit for voxel-wise blood fraction (β). The ΔM/M₀ signal at VG, BAT_vein, and ROI-averaged blood fraction ($\overline{\beta }$) are used to estimate the water extraction, E. Blood flows over the four major feeding arteries (LICA, RICA, LVA, and RVA) are quantified to provide CBF information (Figure 3F), which was combined with E to calculate PS following Equation (1).

Figure 3.

A complete dataset for the measurement of BBB permeability. (A) shows the control, labeled, and difference images collected at the mid-sagittal plane. (B) presents the M₀ scan. (C) presents the ΔM/M₀ image. (D) shows the complex-difference (CD) images of multi-VENC PC MRI along with a GRE image at the same location. (E) shows the fitting to estimate voxel-wise blood fraction (β). (F) shows the CD images and flow maps of the four feeding arteries (LICA, RICA, LVA, and RVA) by reference to a TOF maximum intensity projection (MIP) image. In calculation, ΔM/M₀ signal at VG, BATvein, and $\overline{\beta }$ were used to estimate E, which was then combined with CBF to calculate PS according to the Renkin-Crone model.

Figures 4A–4D show three repetitions of CBF, ΔM/M₀, E, and PS measurements for each mouse. Linear mixed-effect model revealed significant increases over repetitions in CBF (P=0.0004) and ΔM/M₀ signal (P=0.012), a significant decrease in E (P=0.009), but no significant change in PS (P=0.38). There was a significant negative correlation between CBF and E (Figure 4E, y=−0.17x+116.82, R²=0.61, P=0.0006), suggesting that the variations in CBF and E have a physiological origin. The averaged coefficient-of-variation (CoV) values were 6.2±1.7%, 8.6±3.0%, 4.9±1.7%, and 6.1±1.2% for CBF, ΔM/M₀, E, and PS, respectively (Figure 4F).

Figure 4.

Summary and reproducibility of BBB permeability in mice. (A-D) show CBF, ΔM/M₀ in VG, E, and PS across repetitions. (E) shows a scatter plot between CBF and E across all mice and all repetitions. Different color denotes different mouse. (F) shows the coefficient of variation (CoV) of different measurements. Error bar stands for standard error across mice.

PS values obtained with an assumed BAT_vein were consistent with those using individual BAT_vein (Figure 5, y=1.00x, R²=0.94, P<0.0001), suggesting that the use of group-averaged BATvein can provide similar estimations of PS.

Figure 5.

A scatter plot between PS values estimated with an assumed BAT_vein and those with individually measured BAT_vein.

In the post-processing procedure described above, ten voxels with the highest signal intensities in an ROI covering VG were selected to calculate the ΔM/M₀ signal. We have compared the PS results obtained with different numbers of voxels (i.e., 8, 10, and 12 voxels) and did not find a significant difference (ANOVA: P=0.93). The differences in the PS values were 2.2% between processing with 8 and 10 voxels and 2.3% between processing with 10 and 12 voxels. These results suggest that the PS estimates were minimally dependent on the choice of voxel number in the ROI.

3.4. Study 4

Figure 6 summarizes the results of physiological measurements in the zQ175 HD model. Heterozygous zQ175 HD mice showed a higher E (69.7±2.4%, Figure 6A, P=0.026) and PS (318.1±17.1 ml/100g/min, Figure 6B, P=0.040) when compared to WT mice (E=59.9±3.2%, PS=260.9±18.9 ml/100g/min), suggesting a higher BBB permeability. There was not a significant difference in brain volume (Figure 6C, P=0.72), baseline CBF (Figure 6D, P=0.26), or BAT_vein (Figure 6E, P=0.43) between HD and WT mice.

Figure 6.

Comparison of physiological parameters between HD and WT mice. (A-E) show the box plots comparing HD and WT mice in terms of E, PS, brain volume, CBF, and BAT_vein, respectively.

Figure 7 presents the levels of several key tight junction proteins in the zQ175 HD model. As shown in the immunofluorescent staining images (Figure 7A), the zQ175 HD mice exhibited lower Claudin-5 signals in the brain than those in the age-matched control WT mice. Alterations in the three major tight-junction proteins (i.e., ZO-1, Occludin, and Claudin-5) were further verified with the Western blot (Figures 7B–7E). There were significant alterations in ZO-1 (Figure 7C, P=0.037), Occludin (Figure 7D, P=0.048), and Claudin 5 (Figure 7E, P<0.001).

Figure 7.

Altered tight junction proteins in the zQ175 HD model. (A) Representative immunofluorescence images of Claudin-5 staining in the striatum of indicated wild type (WT) and zQ175 mice. (B) shows the Western blot images of tight-junction proteins, i.e., ZO-1, Occludin, and Claudin-5, and a loading control protein β-actin. (C, D and E) present the quantification of protein levels of ZO-1, Occludin, and Claudin-5 between WT and zQ175 mice, respectively.

4. DISCUSSION

We developed an ASL-based MRI method for the non-invasive assessment of BBB permeability to water in mice. We demonstrated that quantitative measurement of permeability surface-area product (PS) in the unit of ml/100g/min can be obtained in less than 6 minutes. We optimized the imaging parameters and benchmarked its reproducibility. An initial application to a mouse model of Huntington’s disease demonstrated the sensitivity of the proposed method in detecting BBB disruption in pathological conditions.

BBB leakage has been implicated in many neurological diseases (Dickie et al., 2019; Haruwaka et al., 2019; Li et al., 2015; Lin et al., 2021b; Lu et al., 2018; Soon et al., 2007). However, in vivo detection of BBB damages, especially subtle ones, has been a challenging task in clinical imaging. Positron emission tomography (PET) was an early method for BBB imaging (Pozzilli et al., 1988). However, the exposure to ionizing radiation along with the need for special equipment, e.g., cyclotron, has limited its application. MRI methods based on the intravenous injection of contrast agent (e.g., gadolinium) have been employed to detect the T₁-relaxation effects of contrast agent molecules once leaked into the extravascular space. Advanced modeling of the DCE MRI data can yield estimates of the volume transfer rate (i.e., K^trans), providing a quantitative index of BBB leakage (Dickie et al., 2019; Heye et al., 2016; Montagne et al., 2015; Thrippleton et al., 2019; Wardlaw et al., 2016). Moreover, the shutter-speed pharmacokinetic model was developed for DCE MRI (Li et al., 2005b; Yankeelov et al., 2003) with successful applications to disease studies. The shutter-speed DCE MRI revealed reduced transcapillary water exchanges in white matter lesions and normal appearing gray matter of patients with progressive multiple sclerosis (Tagge et al., 2021) and in glioblastoma (Rooney et al., 2015). Non-contrast MRI methods for the assessment of BBB permeability have recently emerged and garnered a strong interest in the field. These methods utilize the water molecule as an endogenous tracer and exploit the transverse relaxation time (Ohene et al., 2019), apparent diffusion coefficient (Shao et al., 2019), or spatial locations of water (Lin et al., 2018) to separate the extracellular from the intracellular signal, thereby providing a measurement of BBB integrity. WEPCAST MRI is one such method. It works by determining the relative fraction of labeled water that was extracted into the brain tissue as opposed to those that remained in the vein (Lin et al., 2021a; Lin et al., 2018; Lin et al., 2021b). The present work used a similar principle. However, it was modified for BBB imaging on animal scanners in that the phase-contrast bipolar gradients were not included in our sequence. This is because: 1) the inclusion of bipolar gradients will increase the TE of the sequence, but $T_2^{*}$ of venous blood at 11.7T is very short (~8.9ms) (Wei et al., 2018); 2) water extraction fraction is much lower and venous ASL signal is much higher in rodents (ΔM/M₀ of ~10% in mice vs. ΔM/M₀ of ~0.5% in human), thus the venous signal is already differentiable from the tissue without the bipolar gradients. Water exchanges have been measured in rats using contrast agent-based methods (Dickie et al., 2021; Dickie et al., 2019; Schwarzbauer et al., 1997). In mice, other than the reports of multi-TE ASL methods (Ohene et al., 2021; Ohene et al., 2019), the present technique is one of a few studies to measure water BBB permeability.

PS measured from the present study was 260.9±18.9 ml/100g/min. This was higher than human PS values of approximately 133.6 ml/100g/min (Lin et al., 2022). A previous study using a multi-TE ASL method evaluated water exchange rate (K_w) in mice and found that K_w = 136/min (Ohene et al., 2021). Since PS = K_w · CBV (CBV denotes cerebral blood volume), assuming that CBV is 2–5 ml/100g (Chugh et al., 2009; Verant et al., 2007), this corresponds to a PS value of 272–680 ml/100g/min, which is in a general agreement with the PS values measured in the present study. The water extraction fraction, E, obtained in the present study (E=60±3%) is similar to the previous report in rats (E=67±13%) (Takagi et al., 1987), but is lower than that in rhesus monkey (~90%) (Eichling et al., 1974) or human (91%) (Lin et al., 2022). According to the Renkin-Crone model (Equation 1), part of the reason for the lower E is attributed to a higher CBF, which is in turn due to two factors. One is that CBF in mice is intrinsically higher than that in humans; the other is that the anesthetic agent used, isoflurane, is vasodilative and will increase CBF (Li et al., 2013). Indeed, a negative correlation between CBF and E was observed (Figure 4E), which was consistent with findings in previous reports from rhesus monkeys utilizing H₂¹⁵O as a tracer (Eichling et al., 1974; Herscovitch et al., 1987). A plausible physiological underpinning of this observation is that, with the increase in brain perfusion, transit time through the capillary bed will become shorter, thus hindering the water extraction into the tissue. From a technical point-of-view, since the MRI signal intensity of our method is proportional to 1-E, the sensitivity of our technique tends to be higher in mice than in humans.

This study also measured BAT_vein, which is the time it takes for the blood to travel from major arteries (e.g., internal carotid arteries) to major veins (e.g., the great vein of Galen). To our knowledge, this is the first time that BAT_vein was reported in mice. The measured value, 691 ms, is about twice the value of BAT to tissue (370 ms). For comparison, in humans, BAT to tissue was 1000–2000 ms (white matter had a longer value than grey matter) (Alsop et al., 2015) and BAT_vein was approximately 4000 ms (Lin et al., 2018). Note that a shorter BAT_vein will present another advantage of BBB MRI in mice compared to that in humans, as there is less T₁-related signal decay.

In general, the MRI signal in our method will benefit from long LD (Buxton et al., 1998). However, the SAR issue may become a major concern at prolonged labeling durations. Therefore, a tradeoff needs to be reached between increasing SNR and controlling SAR. In this study, we recommend the use of a labeling duration of 1200 ms and a PLD of 100 ms to observe the ASL signal at VG in mice. Note that, when a different parameter set is desired, the general relationship between LD, PLD, and BAT_vein should be PLD = BAT_vein – 0.5LD, in order to ensure that the measurement is made when the center of the bolus is at the targeted vein. ASL signals in the veins follow a different model from that in the tissue. Most importantly, ASL signals in tissue accumulate as longer labeling bolus is used and more labeled spins arrive. In contrast, in the veins, longer labeling duration does not yield more ASL signals because the labeled spins do not accumulate, as oppose to the situation in the tissue.

Reproducibility studies of the proposed technique revealed that the CoV of the measures is less than 10%, which is considered excellent for a physiological MRI technique. A time dependence of the measured values was also observed. This is attributed to the effects of the temporal accumulation of isoflurane (Wei et al., 2022). As the experiment goes on, CBF gradually increased due to the animal entering a deeper anesthesia level. On the other hand, E decreased accordingly. It appears that the changes in CBF and E cancel each other out, and PS did not show a time dependence. Therefore, it may be that PS measurement is less sensitive to the anesthetic depth. Interestingly, the maintenance of PS value was not only seen under elevated CBF states (due to anesthetic agents as in the present study), but also under reduced CBF states as reported by a recent human study (Lin et al., 2022) administering caffeine challenge.

Huntington’s disease is an autosomal-dominant inherited monogenic disease leading to progressive neurodegeneration (McColgan and Tabrizi, 2018) that presents with motor, cognitive, and psychiatric symptoms. Given the genetic origin, early interventions are desired to delay or prevent the pathological progression of HD and improve treatment outcomes (Mestre et al., 2009). Reliable and sensitive biomarkers are the key to demonstrating the progressive feature and monitoring therapeutic efficacy in both preclinical and clinical studies. Claudin-5 is a key endothelial-specific component of the tight junction strand, particularly in brain endothelial cells and its role is to selectively decrease the permeability to ions. ZO-1 plays an important role in connecting transmembrane proteins to skeleton proteins and interact directly with most of the transmembrane proteins like Occludin and claudins. The significantly reduced ZO-1 and Claudin-5 proteins in the zQ175 HD mice are consistent with the notion of increased BBB permeability in this mouse model. Since Occludin is not only present in endothelial cells but is also expressed in large quantities in cells that have very active metabolism, such as astrocytes, the increased Occludin may reflect a result of astrogliosis which is occurring in HD brain. Taken together, the decreased levels in key tight-junction proteins, which form the structural basis of BBB integrity, provided a direct support for the imaging findings (i.e., elevated E and PS). The proposed ASL-based measurement of BBB permeability to water may facilitate the mechanistic understanding of HD pathogenesis or the therapeutic development by providing a non-contrast MRI measure.

VG primarily drains the deep gray matter (hippocampus, thalamus, hypothalamus, and basal ganglia) and periventricular white matter (Mancini et al., 2015). Therefore, the ASL measurement at VG will be sensitive to BBB alteration in deep brain regions, e.g., in the HD model where striatum was known to be the most vulnerable region (Liu et al., 2021). We point out that, although the vein of Galen does not drain all the arterial blood (De Vis et al., 2018), the unit-volume CBF measured with PC-MRI can still be used for the measurement of BBB permeability associated with the vein of Galen, because the different signal in Eq. (2) has been normalized to the local M₀ signal. Our proposed method, which provides a non-contrast measurement for BBB permeability to water in mice, may open important avenues for studying various biological processes and mechanisms related to BBB function. For example, it can be applied to AQP4 deficient mouse model to probe whether and to what extent the AQP4 channel affects the BBB permeability to water, or applied to trace the potential BBB permeability changes after Na⁺/K⁺ ATPase manipulations (e.g., deslanoside to inhibit Na⁺/K⁺ ATPase). Our method can also be employed for real-time monitoring of the BBB opening induced by interventions, e.g., high-intensity focused ultrasound (HIFU), which attracted broad interests in recent years as a means to enhance the delivery of drugs to brain tumors (Han et al., 2017).

A limitation of the proposed method is the lack of regional BBB permeability information. Further technical development targeting water extraction fraction in small veins will prove useful in evaluating regional BBB damage in brain diseases. Note, however, that many diseases affect the brain diffusely. Thus, a global BBB technique can still find utility in clinical applications, as demonstrated in the HD study in this report. Our proposed method will benefit from the utilization of techniques with sensitivity enhancement, which could shorten scan duration and provide the opportunity to study BBB function with a higher temporal resolution. Enhanced sensitivity will allow for higher spatial resolution, which will facilitate the identification and quantification of small veins, e.g., superior sagittal sinus, to provide more specific BBB permeability measurement for veins draining different regions. In current study, only the intrasession reproducibility has been performed. Further systematic validation focusing on intersession reproducibility of the proposed method will prove helpful.

5. Conclusions

We developed a quantitative MRI method to evaluate BBB permeability to water in mice. This technique is reproducible and does not require the use of contrast agents. The proposed method may open new avenues for preclinical research on pathophysiological mechanisms of brain diseases and therapeutic trials in animal models.

Highlights.

• An MRI method was developed to assess BBB permeability in mice.

• BBB permeability to water can be measured in 6 min without contrast agents.

• BBB breakdown was observed in a mouse model of Huntington’s disease.

Data and code availability statement

Experimental data and processing scripts for the present study are available through reasonable request to the corresponding author upon approval from the Export Control & Facility Security Office of the Johns Hopkins University.