Modeling hyperosmotic blood–brain barrier opening within human tissue-engineered in vitro brain microvessels

Abstract

As the majority of therapeutic agents do not cross the blood–brain barrier (BBB), transient BBB opening (BBBO) is one strategy to enable delivery into the brain for effective treatment of CNS disease. Intra-arterial infusion of the hyperosmotic agent mannitol reversibly opens the BBB; however, widespread clinical use has been limited due to the variability in outcomes. The current model for mannitol-induced BBBO assumes a transient but homogeneous increase in permeability; however, the details are poorly understood. To elucidate the mechanism of hyperosmotic opening at the cellular level, we developed a tissue-engineered microvessel model using stem cell-derived human brain microvascular endothelial cells (BMECs) perturbed with clinically relevant mannitol doses. This model recapitulates physiological shear stress, barrier function, microvessel geometry, and cell-matrix interactions. Using live-cell imaging, we show that mannitol results in dose-dependent and spatially heterogeneous increases in paracellular permeability through the formation of transient focal leaks. Additionally, we find that the degree of BBB opening and subsequent recovery is modulated by treatment with basic fibroblast growth factor. These results show that tissue-engineered BBB models can provide insight into the mechanisms of BBBO and hence improve the reproducibility of hyperosmotic therapies for treatment of CNS disease.

Introduction

The human blood–brain barrier (BBB) restricts delivery of therapeutics to the brain for treatment of CNS disease.¹ Tight junctions (TJs) between adjacent brain microvascular endothelial cells (BMECs) restrict paracellular transport of solutes, while expression of efflux pumps regulates transcellular transport.² Therefore, drug therapies are largely limited to small lipophilic molecules that have moderate to high rates of passive transport across BMEC membranes. However, reversible disruption of TJs can be exploited to transiently increase bioavailability of a broad range of therapeutics into the brain.^3,4

Intra-arterial (IA) infusion of hyperosmotic agents, such as mannitol, is one approach for transient BBB opening (BBBO).⁵ Drug delivery techniques based on IA administration have distinct advantages including high regional drug concentration, high spatial selectivity, rapid onset of action, and limited systemic dose.⁶ Systemic intravenous (IV) administration of mannitol is widely used in the clinic to reduce cerebral edema; however, IA injection at a sufficiently high concentration (close to the solubility limit of around 1.4 M) causes BMEC shrinkage which is sufficient to induce transient BBBO.^5,7,8 This technique has been used for several decades in both preclinical models and clinical studies to improve delivery of chemotherapeutics, stem cells, and gene vectors.^9–16 The reproducibility of BBBO can be difficult to control due to dilution of the injected dose by collateral blood supply in the Circle of Willis, as well as more distally in microcirculation.^17,18 Advances in image-guided insertion of catheters beyond the Circle of Willis into specific locations in the cerebrovasculature have renewed interest in hyperosmotic BBBO for delivery of therapeutic and diagnostic agents.^15,19–21

Although hyperosmotic therapies have been studied for more than 100 years,²² the mechanism of action is not fully elucidated. The current model of hyperosmotic BBBO assumes that water loss from BMECs into the capillary lumen induces vasodilation, resulting in a homogeneous increase in BBB permeability.⁵ Much of what we know about hyperosmotic therapies relies on magnetic resonance imaging (MRI) and positron emission tomography (PET).¹⁹ However, these techniques are not well suited for interrogating mechanisms at the cellular level. Here, we report on real-time imaging of hyperosmotic BBBO on the scale of individual cells in a tissue-engineered microvessel model of the cerebrovasculature incorporating stem cell-derived human BMECs (dhBMECs).^23–25 From fluorescence imaging of different molecular weight solutes following exposure to different mannitol doses, we assess spatial and temporal changes in barrier function, while from phase contrast imaging we assess changes in BMEC structure. The key observations are: (1) mannitol causes transient focal leaks that result in a significant increase in the overall permeability, (2) the focal leaks occur at small (1–2 µm) sub-cellular disruptions in the endothelium, (3) the global TJ network is unaffected by mannitol, and (4) the increase in permeability is due solely to paracellular transport. We establish the timeline of events following hyperosmotic BBBO, including spatially heterogenous increases in paracellular permeability and BMEC vacuolation. Finally, we show that bFGF has an important role in modulating the susceptibility to BBBO and recovery. Our results highlight the ability to evaluate drug delivery mechanisms using a tissue-engineered microvessel model of the human BBB.

Materials and methods

MRI imaging of BBBO in mice

Animal experiments were performed in accordance with guidelines for the care and use of laboratory animals approved by the Institutional Animal Care and Use Committee of Johns Hopkins University, designed in compliance with the Animal Welfare Act regulations and Public Health Service (PHS) policy, and were performed in accordance with ARRIVE guidelines. For details of animal procedures see Supplemental Text and Figure 1.

BBB microvessel fabrication

Human brain microvascular endothelial cells (dhBMECs) were differentiated from induced pluripotent stem cells (iPSCs) as previously reported.^23,26 The BC1 iPSC line derived from a healthy individual was used for all experiments.²⁷ BC1 dhBMECs show transendothelial electrical resistance (TEER) values above 1500 Ω cm² in two-dimensional culture, when cultured in media supplemented with retinoic acid.²⁶

BBB microvessels were fabricated as previously reported.^23,24 The microfluidic device consists of a 150 µm diameter and 1 cm long microvessel lined with BC1 dhBMECs in a genipin-cross-linked type I collagen hydrogel contained in a polydimethylsiloxane housing (Figure 2(a)). Inlet and outlet fluid reservoirs are connected to a flow loop to maintain a shear stress of approximately 4 dyne cm⁻² (Figure 2(b)).^23,24 Functional BBB microvessels with physiological barrier function are formed two days following seeding dhBMECs in the microfluidic device (Figure 2(c)).^23,24 After establishing barrier function (defined as day 0), timelapse imaging experiments were initiated (Figure 2(d)).

Figure 2.

Three-dimensional human in vitro model of hyperosmotic BBB opening. (a) Schematic illustrations of tissue-engineered iPSC-derived BBB microvessels including: (i) side view, (ii) front view, and (iii) three-dimensional view. (b) Microvessels are continually perfused at ∼4 dyne cm⁻² shear stress via gravity-driven flow reservoirs. (c) Phase contrast images of microvessel following seeding (day −2), microvessel formation (day 0) and 48 h later (day 2). (d) Experimental timeline over four days. (e) Imaging protocol for studying human BBB opening. Mannitol and fluorescein are added to the upper reservoir for 2, 5 or 10 min. Next Lucifer yellow and 10 kDa dextran are added to the upper reservoir for two hours. Fluorescein is re-administered for another hour. (f) Timelapse images of: (i) phase, (ii) fluorescein, (iii) Lucifer yellow, and (iv) 10 kDa dextran are shown for a representative microvessel dosed with 1.4 M mannitol for 5 min. (g–h) Representative fluorescence images of fluorescein during 2-min and 5-min mannitol doses. Fluorescein focal leaks are widely observed during 5-min exposure. White arrows denote the sites of focal leaks.

Microvessels were perfused with media (denoted as BBB microvessel media): human endothelial cell serum-free media (Life Technologies) supplemented with 1% human platelet poor derived serum (Sigma), 1% penicillin–streptomycin (Life Technologies), 400 μM db-cAMP (Sigma), 20 μM phosphodiesterase inhibitor Ro-20-1724 (Calbiochem), 3% 70 kDa dextran (Sigma), and 10 μM ROCK inhibitor Y27632 (ATCC, only supplemented for first 24 hours after seeding). Where noted, 20 ng mL⁻¹ recombinant human basic fibroblast growth factor (R&D systems; 146 aa, bFGF) was supplemented for either the initial 24 h of device perfusion (day −2 to day −1) or for 48 h following imaging experiments (day 0 to day 2) (Supplemental Figure 1(a)).

Live-cell imaging of BBB opening

The experimental design and timeline of dosing and assessment of barrier function during live-cell imaging are summarized in Supplemental Figure 1. Microvessels were maintained at 37℃ and 5% CO₂ in a live-cell chamber, and imaged on an inverted microscope (Nikon Eclipse TiE). Epifluorescence illumination was provided by an X-Cite 120LEDBoost (Excelitas Technologies). On day 0 (two days after seeding microvessels), media was removed from the upper reservoir and replaced with 1.4 M mannitol (Sigma) and 5 μM Sodium fluorescein (Sigma) prepared in BBB microvessel media for 2, 5, or 10 min depending on the mannitol dose. After dosing, the upper reservoir was replaced with BBB microvessel media containing 200 μM Lucifer yellow (CH dilithium salt; LY) (Sigma) and 2 μM Alexa Fluor647-conjugated 10 kDa dextran (Thermo Fisher) (Figure 2(e), Supplemental Figure 1(b)). A NIS Elements (Nikon) imaging protocol was initiated immediately following addition of Lucifer yellow and 10 kDa dextran to the upper reservoir. Phase contrast and fluorescence images were acquired every 2 min for 2 h (61 frames). At every time point, six images were collected (Figure 2(f)): (1) a phase contrast image of the top of the microvessel, (2–5) phase contrast and fluorescence images of the microvessel midplane, and (6) a phase contrast image of the bottom of the microvessel. To independently excite and collect the emission from each fluorophore, three filter cubes were used: Chroma 39008 for Lucifer yellow (20 ms exposure), Chroma 49003 for fluorescein (50 ms exposure), and Chroma 41008 for dextran (200 ms exposure). The total image area was approximately 8.2 mm × 0.67 mm, corresponding to 10 adjacent frames using a 10 × objective.

Mannitol (MW 182.2) dosing was indirectly monitored using co-administered fluorescein (MW 376.3 Da), a small fluorescent compound that does not exhibit intracellular accumulation. Fluorescein begins filling the microvessel several minutes after imaging is initiated due to transit within 40 cm long tubing, which connects the upper reservoir and microvessel (Supplemental Figure 1(c)). Since convective flux is proportional to fluid velocity and not dependent on the physicochemical properties of a solute, it is reasonable to assume that mannitol and fluorescein are transported at the same rate; therefore, the fluorescein intensity can be used as a proxy to identify the duration and maximum of the mannitol dose. After 2 h of imaging, microvessels were re-perfused with fluorescein for 1 h to monitor recovery of barrier function.

Spatial permeability analysis

Analysis is summarized in Supplemental Figure 2. Large images were sectioned into 10 regions of interest (ROIs), each with dimensions of 820 μm × 670 μm (Supplemental Figure 2(a)). Time course fluorescence intensity profiles of Lucifer yellow were obtained using ImageJ (Supplemental Figure 2(b)). Microvessel permeability (cm s⁻¹) was calculated as P = (d/4)(1/ΔI)(dI/dt), where d is the vessel diameter, ΔI is the increase in fluorescence intensity upon luminal filling, and (dI/dt)₀ is the rate of increase in fluorescence intensity as the solute is transported across the endothelium into the surrounding matrix (Supplemental Figure 2(c)).^28,29 The rate of increase of fluorescence intensity, from which the permeability is calculated, was determined from a linear least squares fit over 20 min following luminal filling. Lucifer yellow permeability was calculated for each of the 10 ROIs, with the minimum across all ROIs reported as P_min(LY). Five adjacent ROIs with the lowest mean permeability were used to analyze spatial heterogeneity of BBBO; ROIs outside this range were discarded from analysis as they were not buffered from slow interstitial leakage of dye into the hydrogel from the inlet and outlet, which artificially increases permeability values.

Focal leaks induced by mannitol were manually counted in ImageJ (Supplemental Figure 2(d)). Focal leaks appear as radial extravascular plumes of fluorescence due to diffusion of dye into the hydrogel from the leakage site. Focal leaks were counted over the initial dosing of mannitol and over the 1 h fluorescein re-administration. In the 2 h following dosing, the x-position along microvessels (from 0 to 8,200 μm) and time of appearance of individual 10 kDa dextran focal leaks were recorded. The total number of 10 kDa dextran focal leaks during 2 h following dosing was normalized to microvessel length (# cm⁻¹). To assess penetration of 10 kDa dextran into the surrounding matrix, the fluorescence intensity was determined in the five adjacent ROIs with the lowest mean permeability excluding the microvessel lumen, corresponding to two-thirds of the imaging frame (total area = 1.83 mm²). Within this region, the normalized cumulative and instantaneous change in fluorescence intensity was plotted as a proxy for drug extravasation.

Phase contrast analysis

Structural changes observed after exposure to mannitol (i.e. intracellular vacuoles) were counted in ImageJ using time-lapse phase contrast images at the microvessel poles cropped to an area of 20 cells; the vacuole count was then normalized to the number of cells (# cell⁻¹). Vacuole quantification was verified by comparison to the corresponding fluorescence images following treatment with 2 μM Calcein AM (ThermoFisher), which stains all intracellular compartments besides vacuoles (Supplemental Figure 3(a)). Vacuole density did not vary substantially along the x-axis of microvessels, justifying sampling of a single region containing 20 cells (Supplemental Figure 3(b)).

To probe vacuole localization and function, four fluorescence assays were used: 2 μM Calcein AM for live cells, 4 μM ethidium homodimer-1 (ThermoFisher) for dead/dying cells, 2 μg mL⁻¹ Hoechst 33342 (ThermoFisher) for nuclei, and pHrodo™ Green AM Intracellular pH Indicator (ThermoFisher) for intracellular pH (following manufacturer suggested protocols). For these experiments, microvessels were: (1) washed with live-cell imaging solution (LCIS; ThermoFisher) for 5 min, (2) stained at 37 ℃ for 30 min, (3) washed with LCIS for 5 min, and (4) imaged as previously described.

Cell loss events were recorded from time-lapse imaging at the microvessel midplane. These events are visible as BMECs balling-up and detaching from the endothelium. Events were normalized to the length of microvessel, based on analysis within representative ∼465 μm long microvessel image sections. Cell density was determined by manual counting in a 32,000 μm² area at the pole region of a microvessel (∼100 cells).

Statistical analysis

All statistical analysis was performed using Prism ver. 8 (GraphPad). Metrics are presented as mean ± standard deviation (SD). The principle statistical tests used were a Student's unpaired t-test (two-tailed with unequal variance) for comparison of two groups, and an analysis of variance (ANOVA) for comparison of three or more groups. Reported p-values were multiplicity adjusted using a Tukey test. Differences were considered statistically significant for p < 0.05, with the following thresholds: *p < 0.05, **p < 0.01, ***p < 0.001. To determine the time course of experimental metrics, the beginning of the mannitol dose is designated as t = 0 and the half-time represents the time following dosing at which the metric reaches half its maximum value.

Results

Hyperosmotic BBB opening in mice

To establish a clinically relevant comparison for in vitro experiments, in vivo measurements of BBBO following mannitol dosing were performed in a mouse model. Contrast enhanced T1 MRI, a standard technique for monitoring BBB status both clinically and in animal models,¹⁹ was used to imaging of BBBO following dosing with mannitol. At hemodynamically safe IA infusion rates, consistent BBBO occurs in subcortical brain structures (primarily the hippocampus), while the cerebral cortex and the contralateral brain regions are not reproducibly opened (Figure 1(a)). The Gadoteridol (Gd) contrast enhancement after mannitol dosing (compared to the value to before dosing) and normalized to the unaffected brain ROI, shows robust opening in the ipsilateral hippocampus ( > 2-fold increase) compared to other regions (p < 0.01 for each comparison) (Figure 1(b)). Due to the relatively low spatial resolution of MRI, opening appears to result in a uniform increase in permeability. Additionally, high spatial resolution using multiphoton microscopy is not possible in mice as BBBO is not reproductively observed in superficial brain regions.¹⁹

Figure 1.

Hyperosmotic BBB opening in mice. (a) T1 magnetic resonance (MR) images were acquired before 1-min 1.4 M mannitol infusion and 5 min after subsequent Gadoteridol (Gd) injection. (b) The ratio of Gd contrast post- and pre-mannitol exposure (normalized to ROI 1). The ipsilateral hippocampus displays Gd enhancement, while other regions do not. (c) T1 images during and after a 1-min 1.4 M mannitol injection with Gd. (d) Dynamics of normalized Gd contrast across ROIs; only the ipsilateral hippocampus (ROI 4) displays significant enhancement, which occurs rapidly following cessation of mannitol infusion. Pre/Post T1 (n = 4 mice), dynamic T1 (n = 1 mouse).

A 1-min injection of mannitol, along with Gd, during real-time T1 MRI demonstrates the temporal and spatial signature of BBBO in the mouse brain (Figure 1(c)). Analysis of the ipsilateral hippocampus (ROI 4) shows that BBBO occurs shortly after cessation of mannitol infusion (less than one minute later) (Figure 1(d)). Additionally, while MRI enables non-invasive assessment of BBBO, precise calculations of vascular permeability are complicated because gadolinium is a contrast agent rather than a tracer, and T1 relaxivity-derived measures lack precision.³⁰

Mannitol induces spatially heterogenous and dose-dependent increases in solute permeability in a tissue-engineered BBB model

To study the mechanisms of hyperosmotic BBBO, we introduced mannitol into a tissue-engineered model of the BBB with a single 1 cm long microvessel that mimics the dimensions (∼150 µm diameter) and wall shear stress (4 dyne cm⁻²) of a post-capillary venule (Figure 2(a) and (b)).²³ Two days after seeding stem cell-derived human BMECs into genipin-crosslinked collagen channels (day 0), the formation of a confluent monolayer (Figure 2(c)) results in physiological barrier function, as previously reported.²⁴ Microvessels were perfused with mannitol and the tracer fluorescein for 2 or 5 min. Following dosing, the microvessels were then perfused with the fluorescent probes Lucifer yellow and 10 kDa dextran (Figure 2(d) and (e)). During this process, phase contrast microscopy was used to monitor the status of the endothelium, and epifluorescence imaging was used to visualize the transport of fluorescein, Lucifer yellow, and dextran out of the lumen and into the surrounding matrix (shown for a representative 5-min mannitol dose in Figure 2(f)).

Fluorescein provides a proxy for the time-dependent concentration of mannitol in the microvessel and enables visualization of BBBO during and immediately after (∼10 min) mannitol dosing. Fluorescein focal leaks (white arrows in Figure 2(g) and (h)) develop during dosing, indicating that BBBO occurs rapidly following exposure to mannitol. The densities of focal leaks during dosing were 0.24 ± 0.54 cm⁻¹ and 5.59 ± 4.35 cm⁻¹ for 2-min and 5-min mannitol doses, respectively (Supplemental Figure 4(a) and (b)).

Following dosing with mannitol, Lucifer yellow (MW 444 Da) and 10 kDa dextran were introduced into the microvessel to assess barrier function over the subsequent 2 h. Mannitol induces dose-dependent changes in permeability (Figure 3(a) and (b), Supplemental Video 1 and 2). To assess the spatial heterogeneity in Lucifer yellow leakage, we determined the permeabilities in five adjacent segments (820 µm in width) along the length of the microvessels over 2 h after dosing with mannitol (Figure 3(c)). Under baseline conditions (no mannitol), the Lucifer yellow permeability along the length of the microvessel varies by less than 68% between five adjacent ROIs. In contrast, there is substantial spatial variation in Lucifer yellow permeability following 2- and 5-min mannitol doses (on average 228%). This variation is due to the appearance of focal leaks in different segments. The minimum Lucifer yellow permeability across ROIs was 1.59 ± 0.87 × 10⁻⁷ cm s⁻¹ (control), 8.27 ± 8.66 × 10⁻⁷ cm s⁻¹ (2-min mannitol), and 28.8 ± 18.4 × 10⁻⁷ cm s⁻¹ (5-min mannitol). 5-min mannitol doses result in significantly higher Lucifer yellow permeability compared to the control and 2-min mannitol (p = 0.004 and p = 0.033, respectively) (Figure 3(d)). The maximum Lucifer yellow segment permeability (5-min dose) was about 8 × 10⁻⁶ cm s⁻¹, 50-fold higher than the average baseline permeability (control). This increase in permeability represents the upper limit for transiently enhancing drug delivery. Additionally, mannitol dosing resulted in increased variation in the Lucifer yellow permeability, with a 10-fold higher standard deviation compared to the controls (Figure 3(d)).

Figure 3.

Spatially heterogeneous dose-dependent hyperosmotic BBB opening. Representative images at 30, 60, and 90 min for (a) 10 kDa dextran and (b) Lucifer yellow under baseline conditions (no mannitol; control) or after 2-min or 5-min bolus doses of 1.4 M mannitol. White arrows denote sites of focal leaks. The onset of the mannitol dose begins approximately 2 min after injection into the reservoir. Following the mannitol dose, the microvessels are perfused with Lucifer yellow and 10 kDa dextran for 2 h. (c) The spatial dependence of Lucifer yellow permeability along microvessels (ROI = 825 µm segment) is highly heterogeneous following mannitol exposure. (d) Minimum Lucifer yellow permeability (i.e. the segment with the lowest permeability) for each condition. (e) Focal leaks are uniformly distributed along the length of the microvessels. (f) Focal leak density for each condition. (g) Cumulative 10 kDa dextran focal leaks and (h) normalized cumulative and instantaneous 10 kDa dextran extravasation following a 5-min mannitol dose. The dotted line represents the mean, while upper and lower solid lines represent the mean ± SD; time is normalized to onset of mannitol dosing at t = 2 min. Control (n = 6 microvessels), 2-min mannitol dose (n = 5 microvessels), and 5-min mannitol dose (n = 5 microvessels).

Increases in Lucifer yellow permeability result from discrete disruptions in barrier function at sites of focal leaks. From analysis of 10 kDa dextran images, we found that focal leaks were located randomly across the entire length of microvessels (Figure 3(e)). Over the 2-h imaging period immediately after mannitol dosing, control microvessels displayed no focal leaks, 2-min mannitol microvessels displayed ∼3 focal leaks cm⁻¹, and 5-min mannitol microvessels displayed ∼18 focal leaks cm⁻¹ (Figure 3(f)). In subsequent experiments, we focus on 5-min doses.

To determine the time course of BBBO, the number of 10 kDa dextran focal leaks and the penetration of the dye into the matrix were analyzed as a function of time (Figure 3(g) and (h)). New focal leaks are not observed after ∼60 min following onset of a 5-min mannitol dose (Figure 3(g)). Additionally, extravasation of 10 kDa dextran occurs rapidly; half of all extravasation occurs within 30 min following onset of mannitol administration (Figure 3(h)), while no extravasation is observed in control microvessels (Figure 3(a)), as previously reported.²³

To assess microvessel recovery, fluorescein was re-administered after the initial 2 h imaging window for an additional hour (Figure 2(e)). For 2-min mannitol doses, fluorescein focal leaks were rare during dosing (0.24 ± 0.54 cm⁻¹) and none were observed during fluorescein re-administration (Supplemental Figure 4(c) and (d)), indicating that BBBO is reversed and normal barrier function is re-established. For a 5-min mannitol dose, while fluorescein focal leaks were observed (5.59 ± 4.35 cm⁻¹) during initial dosing, 2 h later focal leaks were rare (0.91 ± 1.82 cm⁻¹ over 1 h) and not observed in most microvessels (75%) (Supplemental Figure 4(c) and (d)). After fluorescein re-administration, only 4% of all focal leaks observed during 10 kDa dextran perfusion displayed persistent opening, indicating general reversal of BBBO within 2 h of dosing.

Endothelium response to mannitol dosing

From phase contrast images (Figure 2(f)), mannitol dosing induced morphological changes including BMEC shrinkage, thinning of the endothelium, and vacuolation (Supplemental Video 3). Upon mannitol dosing, vacuoles formed in the dhBMECs (Figure 4(a), Supplemental Video 3); their density increased rapidly following the onset of mannitol dosing and reached a plateau 30 min later (Figure 4(b)). The formation of vacuoles was dose-dependent; 2-min doses resulted in 6.7 ± 2.1 cell⁻¹, while 5-min doses resulted in significantly higher vacuolation of 10.6 ± 1.9 cell⁻¹ (p = 0.005) (Figure 4(c)). Vacuoles were 3.8 ± 1.5 μm in diameter within microvessels exposed to 5-min mannitol (Supplemental Figure 3(c)), excluding the possibility that these structures are lysosomes or endosomes.³¹

Figure 4.

Hyperosmotic vacuole formation and tight junction disruption in BBB microvessels. Vacuolation of endothelial cells is dose-dependent: (a) representative time course images of bottom plane of microvessel upon exposure to mannitol. t = 0 min represents addition of Lucifer yellow and 10 kDa dextran to the upper reservoir, after which the onset of mannitol dose occurred 8 or 6 min later for the 2-min or 5-min mannitol doses shown, respectively. Red arrows denote examples of vacuoles visible from phase contrast microscopy. (b) Time course of vacuole formation, (c) maximum vacuoles across doses. (d–e) 10 kDa dextran does not accumulate intracellularly following mannitol exposure: (d) representative 10 kDa dextran and phase contrast images, (e) intensity projection of 10 kDa dextran along the dashed white line in subsequent subpanel. (f–g) Live-dead stains (Calcien AM, ethidium homodimer-1) and percentage of ethidium homodimer-1 positive cells before and after mannitol exposure (n = 3 microvessels). (h) Cumulative cell loss over 2 h for each condition. (i) Confocal imaging at the midplane of a microvessel following a 5-min mannitol dose shows a focal leak (about 1–2 µm) of 10 kDa dextran across the endothelium; 50 min later, the endothelium is intact showing that the focal leak has closed. (j) Epifluorescence image of a 10 kDa dextran focal leak at the midplane of a microvessel following a 5-min mannitol dose. Concurrent epifluorescence imaging of tight junctions (ZO1) shows no global changes in TJ localization during hyperosmotic BBB opening. For quantification of vacuoles and cell loss: control (n = 6 microvessels), 2-min mannitol dose (n = 5 microvessels), and 5-min mannitol dose (n = 5 microvessels). All other presented images (d–j) were following a 5-min mannitol dose.

To gain insight into vacuole localization, microvessels were stained with Calcein AM and DAPI. Interestingly, the cytoplasmic stain Calcien AM was found to distinguish cytoplasm and nucleus from vacuoles. Vacuoles were distributed heterogeneously within dhBMECs including within cell nuclei, within the cytoplasm, or at the boundary of cell nuclei (Supplemental Figure 4(d)). Additionally, BMEC nuclei appeared distorted following mannitol treatment, likely due to abrupt changes in cell shape which alters nuclear shape via the actin cytoskeleton.³²

Vacuole formation could occur via multiple mechanisms. Dextran remained excluded from intracellular compartments during and after mannitol dosing (Figure 4(d) and (e)), indicating that vacuoles did not form as plasma membrane invaginations. Dextran did not accumulate within the cytoplasm or vacuoles in the BMECs at any point during our experiments, implying that mannitol induces changes in paracellular permeability but not transcellular permeability for compounds with similar or higher molecular weight. To further confirm that mannitol does not induce changes in membrane integrity, microvessels were stained before, or 30 min after, a 5-min mannitol dose. Most BMECs remained viable following mannitol treatment (Calcein AM), and although ethidium homodimer-1 positive cells were slightly more prevalent, this was an exceedingly small population (∼0.06 %) (Figure 4(f) and (g)). Next, we hypothesized that intracellular vacuoles were formed in response to an increase in hydrogen ion concentration (decrease in pH) resulting from the rapid loss of water. However, there was no change in intensity or localization of a pH indicator dye (Supplemental Figure 3(e)). The dye was co-localized with DAPI, suggesting that nuclei are the most acidic cell compartment under baseline conditions. However, it is difficult to definitively conclude that vacuoles do not have a lower pH as indicator dyes could be restricted from entry into vacuoles, cleaved in the cytoplasm, or self-quenched at high concentrations. Vacuoles have been observed in vivo following mannitol treatment, but were not correlated with BBBO.³³

Vacuole formation reduced the definition of dhBMECs in phase contrast imaging, making it difficult to measure cell loss events.²³ However, from timelapse imaging at the microvessel midplane, loss of dhBMECs from the endothelium is observed following mannitol dosing (Supplemental Video 3). Cell loss is dose-dependent, with events observed ∼6 or ∼2 times more frequently following 5-min doses, compared to control and 2-min doses, respectively (Figure 4(i)). Increased apoptosis has previously been observed following hyperosmotic stress in other cell types.^34,35

Characterization of focal leaks

Confocal images during perfusion with 10 kDa dextran shown that the focal leaks are associated with small gaps (1–2 µm) in the endothelium (Figure 4(i)). Images above and below the z-plane (0.4 μm thickness) show no leakage (Figure 4(i), Supplemental Figure 5). There is no evidence of intracellular accumulation of 10 kDa dextran, indicating that focal leaks are paracellular. Healing of the focal leak is evident from exclusion of 10 kDa dextran from the endothelium 50 min later (Figure 4(i)).

To further investigate the mechanism of TJ disruption, we performed live-cell imaging of BBB microvessels where the dhBMECs were differentiated from iPSCs with fluorescently labeled zona occludens-1 (ZO1). Following mannitol administration, the global TJ structure is not perturbed (Figure 4(j)). Direct visualization of TJ disruption at the site of a focal leak is not possible due to limited z-resolution at the microvessel midplane where the focal leaks are visualized. However, from analysis of the images at the top and bottom of the microvessels, we can show that there is no disruption in ZO1 within about 125 μm (∼5 cell lengths) of a focal leak (Supplemental Figure 6).

bFGF pre-treatment attenuates BBB opening

To assess the role of bFGF in modulating BBB function, microvessels were treated with 20 ng mL⁻¹ bFGF during the initial 24 h after cell seeding (Figure 5(a)). Control microvessels (no mannitol) pre-treated with bFGF displayed similar Lucifer yellow permeability on day 0 compared to control microvessels (p = 0.449), although there was an increase in the number of vacuoles (p = 0.003) (Figure 5(b)). Pre-treatment with bFGF increases the transendothelial electrical resistance (TEER) of iPSC-derived BMEC monolayers within 2D transwells (p = 0.021, Supplemental Figure 7(a)). However, increased TEER was not associated with increases in endothelial cell density (p = 0.750), total TJ fluorescence intensity (p = 0.755), or junctional localization (Supplemental Figure 7(b) to (d)). Microvessels pre-treated with bFGF and then exposed to a 5-min mannitol dose (on day 0) were less susceptible to BBBO (Figure 5(c) and (d)). The minimum Lucifer yellow permeability and the presence of focal leaks were reduced with bFGF pre-treatment (p = 0.029 and 0.036, respectively) compared to control microvessels (Figure 5(d)). For example, following a 5-min mannitol dose, the minimum permeability was 3.02 ± 2.14 × 10⁻⁷ cm s⁻¹, 10-fold lower than with no bFGF treatment. Similarly, the number of focal leaks decreased from 18 ± 12 cm⁻¹ to 1.8 ± 1.6 cm⁻¹ following pre-treatment with bFGF. The maximum number of vacuoles per cell was slightly higher with bFGF pre-treatment, although not statistically significant (p = 0.113) (Figure 5(d)). However, microvessels pretreated with bFGF and exposed to a 10-min mannitol dose displayed BBBO (Supplemental Figure 8).

Figure 5.

Basic fibroblast growth factor (bFGF) pre-treatment mitigates BBB opening. (a) Summary of 20 ng mL⁻¹ bFGF treatment. (b) Comparison of the minimum Lucifer yellow permeability, the number of focal leaks, and the number of vacuoles following a 5-min mannitol dose with (n = 3) and without bFGF pre-treatment with (n = 6). (c) Representative images of 10 kDa dextran and Lucifer yellow in bFGF pre-treated microvessels following a 5-min mannitol dose showing no focal leaks. (d) Comparison of the minimum Lucifer yellow permeability, the number of focal leaks, and the number of vacuoles following a 5-min mannitol dose with (n = 4) and without bFGF pre-treatment with (n = 5).

bFGF post-treatment enhances BBB stability 48 h after mannitol dosing

Endogenous bFGF circulates in human blood at concentrations in serum below ∼10 pg mL⁻¹.³⁶ To aid in recovery and repair following injuries, administration of exogenous bFGF or other growth factors has been explored.^37–40 Treatment with hyperosmotic agents perturbs endothelial homeostasis and causes significant stress to BMECs. Previously, we showed that focal leaks 2 h following a 5-min mannitol dose were rare events, indicating recovery of barrier function (Supplemental Figure 4(c) and (d)). To investigate the long-term response to mannitol stress, we investigated barrier function 48 h following treatment with and without bFGF continuously administered starting 2 h after mannitol dosing (Figure 6(a)). Without bFGF post-treatment, BBB microvessels displayed variable barrier function 48 h following a 5-min mannitol dose (Figure 6(b)). The minimum Lucifer yellow permeability decreased 48 h following dosing (day 2) and was approximately three times higher than the baseline control although not statistically significant (p = 0.360) (Figure 6(c)). While widespread disruption of the monolayer was not observed 48 h following dosing, some microvessels did display leakage of 10 kDa dextran indicating instability of barrier function (Figure 6(d)). With bFGF post-treatment, all microvessels displayed stable barrier function following hyperosmotic BBBO (Figure 6(f)). The minimum Lucifer yellow permeability 48 h following 5-min mannitol dosing was 3.42 ± 1.74 × 10⁻⁷ cm s⁻¹, approaching physiological values, and similar to controls (no mannitol, no bFGF) (p = 0.132) (Figure 6(g)). In addition, with bFGF post-treatment, no focal leaks were observed, representing a significant decrease from the response immediately after mannitol dosing (p = 0.047) (Figure 6(h)). The recovery of dhBMEC vacuolation was not dependent on bFGF post-treatment; under both conditions, vacuolation decreased by ∼10-fold 48 h following 5-min mannitol dosing (p = 0.007 and 0.013, for control and bFGF post-treatment microvessels respectively) (Figure 6(e) and (i)). Administration of bFGF after mannitol dosing was associated with a small, non-significant increase in cell density, possibility due to increased BMEC proliferation (p = 0.326, Supplemental Figure 7(e)).

Figure 6.

bFGF post-treatment promotes BBB repair following hyperosmotic opening. (a) Microvessels were continuously perfused with 20 ng mL⁻¹ bFGF following a 5-min mannitol dose. (b) Representative images of 10 kDa dextran and Lucifer yellow on day 2 following a 5-min mannitol. Focal leaks are denoted by white arrows. (c–e) Comparison of minimum Lucifer yellow permeability, the number of focal leaks, and the number of vacuoles after a 5-min mannitol dose, and on day 2 following dosing (n = 3). (f) Representative images of 10 kDa dextran and Lucifer yellow on day 2 following a 5-min mannitol dose in bFGF post-treated microvessels. No focal leaks are observed. (g–i) Comparison of minimum Lucifer yellow, the number of focal leaks, and the number of vacuoles after a 5-min mannitol dose, and on day 2 following dosing in microvessels post-treated with bFGF (n = 3).

Discussion

MRI assessment of BBB status in animal models does not provide sufficient spatial resolution for studying the mechanisms of BBBO. Here, we utilize time lapse imaging to visualize hyperosmotic BBBO in a tissue-engineered microvessel model with human stem cell-derived BMECs.^23,24 In the current model for transient hyperosmotic BBBO, removal of water induces vasodilation, and above a critical threshold (related to the product of osmolarity and injection duration), vasodilation causes sufficient mechanical stress to disrupt TJs, resulting in increased paracellular transport.^5,7,8 This model usually assumes a time-dependent but spatially homogeneous increase in permeability.⁵

Mannitol induces spatially heterogeneous increases in paracellular permeability

The key observations from imaging mannitol-induced BBBO in microvessels are: (1) mannitol causes transient focal leaks that result in a significant increase in the overall permeability, (2) the focal leaks occur at small (1–2 µm) sub-cellular disruptions in the endothelium, (3) the global TJ network is unaffected by mannitol, and (4) the increase in permeability is due solely to paracellular transport. Together these results suggest that focal leaks are formed by disruption of TJs between adjacent cells.

A 1 cm long, 150 μm diameter microvessel consists of ∼8000 cells (average cell area is 625 μm²), with each cell having ∼6 neighbors,²³ corresponding to a total of 48,000 unique cell–cell junctions. Therefore, the observation of 18 focal leaks (assuming that focal leaks occur when a single cell–cell junction is disrupted) in response to a 5-min mannitol dose corresponds to a failure of 0.04% of TJs or ∼0.25% of all dhBMECs. This suggests that there is a distribution of TJ strengths and it is the weakest TJs that are the first to be disrupted during mannitol dosing. Consistent with this hypothesis, spatial heterogeneity in endothelial barrier function was reported in 2D monolayers in response to histamine treatment⁴¹ utilizing total internal reflection fluorescence (TIRF) microscopy.

Under baseline conditions, the permeability of 10 kDa dextran is below the detection limit and there is no appreciable intracellular accumulation indicating negligible paracellular and transcellular transport. Following mannitol dosing, the permeability of 10 kDa dextran increases via focal leaks but there is no intracellular accumulation or widespread BMEC membrane permeabilization (i.e. low ethidium homodimer-1 fraction), consistent with negligible transcellular transport. Despite the low incidence of TJ failure, dramatic increases in permeability are still observed. For example, the minimum Lucifer yellow segment permeability increases 20-fold over 20 min following a 5-min mannitol dose. This increase matches previous measurements of baseline Lucifer yellow permeability within human umbilical vein endothelial cell microvessels (∼4 × 10⁻⁶ cm s⁻¹).²³ These results suggest that transient BBBO following mannitol dosing is associated with a small number of focal leaks that are responsible for rapid extravasation of solutes into the surrounding matrix. We note that mechanisms of BBBO and response to mannitol may be different in other conditions, such as trauma and stroke.^42–44

Timeline of hyperosmotic BBB opening, drug delivery and recovery

To further elucidate the timeline of opening and recovery (Figure 7(a) and (b)), we normalized experimental metrics to the onset of a 5-min mannitol dose (Figure 7(c) and (d)). Fluorescein focal leaks appear during mannitol dosing, suggesting that changes in BBB permeability can occur almost instantaneously, matching observations from dynamic T1 MRI (Figure 1(d)). The emergence of dhBMEC vacuolation and the decrease in microvessel diameter show delayed responses compared to focal leaks, with half-times of 6.5 and 12 min, respectively, after dose onset. This suggests that vacuolation and changes in microvessel diameter are not direct mediators of BBBO.

Figure 7.

Mechanisms of hyperosmotic BBB opening, drug delivery, and recovery. (a) Cross-sectional view of BBB microvessels during and after mannitol dosing, with (b) zoom-in on tight junctions (TJs) between adjacent BMECs. (c) Mechanistic timeline: fluorescein focal leaks first emerge during dosing followed by vacuoles and 10 kDa dextran focal leaks after dosing. Drug delivery is proportional to focal leak density and persistence. Short-term recovery is visible as cessation of fluorescein focal leaks, while in the long-term mannitol can compromise stability. TJ strengthening (bFGF pre-treatment) mitigates BBB opening, while BMEC proliferation (bFGF post-treatment) promotes BBB opening. (d) Time course of events over the initial 20 min of imaging and (e) over 2 h of imaging following a 5-min mannitol dose (n = 5 microvessels). All events are normalized to the 5-min dose in which fluorescein fluorescence peaks, corresponding to the vertical grey bar from 2 to 7 min. Half-times (τ) are reported with respect to onset of dose (t = 2 min). The dotted line represents the mean, while upper and lower solid lines represent the mean plus or minus one standard deviation.

Drug delivery, based on extravasation of 10 kDa dextran, is proportional to focal leak density and persistence. The cumulative number of focal leaks for 5-minute mannitol doses displays a half-time of 20 min after dose onset (Figure 7(e)). The extravasation of 10 kDa dextran occurs over a longer period (τ = 30 min), due to the persistence of focal leaks (Figure 7(e), Supplemental Figure 9). The persistence of focal leaks following a 5-min mannitol dose spanned an order of magnitude, from less than 2 min to more than 20 min (Supplemental Figure 9). Determining the persistence time was complicated due to impingement of dextran from neighboring focal leaks. Cell loss is significantly delayed from the onset of focal leaks and leakage of fluorescent probes into the matrix (τ = 60 min), supporting the hypothesis that focal leaks are associated with TJ disruption and not cell loss from the endothelium (Figure 7(e)). Over 2 h following a 5-min dose ∼2% of all BMECs are lost from the microvessel, likely due to induced apoptosis from hyperosmotic stress.^34,35

10 kDa dextran extravasation into the ECM (a proxy for drug delivery into the brain) is dramatically increased by mannitol dosing. In contrast, in the absence of mannitol dosing, there is no detectable leakage of dextran from the microvessel. Techniques to modulate transcellular permeability, including efflux pump inhibition, have a more modest potential to alter CNS drug penetration. For example, p-glycoprotein inhibition in vivo⁴⁵ and in vitro²³ modulates efflux pump substrate permeability by only 2- to 3-fold. Thus, hyperosmotic methods for BBBO have a distinct advantage for delivery of large molecular weight compounds that are not efflux substrates.

Based on the time-dependent responses, we propose that mannitol induces shrinkage of BMECs which increases the tensile stress on TJs (Figure 7(a) and (b), Supplemental Figure 10). Cell shrinkage continues to a point where TJs between adjacent cells are compromised. Assuming a distribution of TJ strength, the weakest TJs fail first resulting in transient focal leaks and a local increase in paracellular permeability. Microvessels exposed to low doses of mannitol do not reliably open due to insufficient cell volume loss, while bFGF pre-treatment strengthens TJs preventing BBBO. Long-term failure of barrier function is likely due to accumulated hyperosmotic stress on BMECs, which can be recovered by promoting cell proliferation with bFGF post-treatment.

Baseline permeability is not a predictor of susceptibility to BBB opening

Pre-treatment with bFGF does not alter baseline permeability of microvessels, but does increases the mannitol dose required to induce BBBO. This suggests that bFGF shifts the distribution of TJ strengths so that longer doses are required to exert sufficient stress on the weakest TJs to initiate opening. In the context of the brain, it is possible that some regions are more resistant to opening due to differences in TJ expression, even though the permeability to small molecules is independent of location. For example, the cerebral cortex does not exhibit significant BBBO following hyperosmotic therapy in mice.¹⁹ It is not well understood if this is due to heterogeneity of TJ strength⁴⁶ or hemodynamics and the mannitol distribution compared to other brain regions. In vitro, treatment with bFGF upregulates primary hBMECs TJ proteins.³⁸ Additionally, while bFGF is not required to achieve physiological barrier function (low permeability), as shown here and previously,²³ its removal during seeding of transwells with dhBMECs reduces transendothelial electrical resistance (TEER) (Supplementary Figure 7). Thus, bFGF is likely able to shift the distribution of TJ strengths so that a 10-min mannitol dose is required to induce disruption. Importantly, these results imply that the culture conditions of in vitro BBB models impact interpretation of BBBO efficacy.

Long-term recovery to BBB opening is promoted by bFGF

Fluorescein focal leaks were not observed 2 h after a 2-min mannitol dose and were uncommon (25% of microvessels) following a 5-min dose, indicating that barrier function has largely recovered within 2 h. However, the hyperosmotic stress on dhBMECs resulted in inconsistent barrier function 48 h following dosing in the absence of exogenous growth factors. Focal leaks and increased permeabilities were observed 48 h after 5-min mannitol doses. In previous work, BBB microvessels not exposed to mannitol displayed no focal leaks and stable permeability over six days after seeding.²³ Post-treatment of microvessels with bFGF promoted recovery from hyperosmotic stress in the 48 h following dosing. A possible explanation for the influence of bFGF is that it promotes cell growth, and increases cell density, thereby reducing the stress on the dhBMECs in response to hyperosmotic stress. Although, we measure a small increase in cell density, it is not significant (p = 0.326, Supplemental Figure 7(e)). However, only a small increase in cell density may be sufficient to reduce the stress on TJs.

In vivo studies of the duration of osmotic BBBO have reported conflicting timelines.^17,18,47 These inconsistencies likely result from differences in species, anesthetic agents, infusion rates, doses, tracers, and imaging techniques, among other factors. Across 38 human subjects, good-to-excellent BBB disruption was obtained in ∼75% of patients, while the remaining 25% displayed poor-to-moderate disruption.¹⁸ This study reported that the BBB remained open for at least 40 min after osmotic exposure and returned to baseline after 6 to 8 h following induction of good-to-excellent BBBO.¹⁸ We find that BBBO is generally reversible within 2 h. Additionally, our results suggest that mannitol treatment induces stress in dhBMECs that, for larger doses, manifests two days later as disruptions in barrier function, but which can be recovered by bFGF treatment. bFGF exerts a protective effect on BBB damage following traumatic brain injury and intracerebral hemorrhage in mice.^38,39 While bFGF treatment is not required to reverse hyperosmotic therapies in vivo, supporting cells (i.e. pericytes⁴⁸ or astrocytes⁴⁹) may play a role in recovery.

Model advantages and limitations

Our tissue-engineered microvessel model mimics key components of the human BBB, including physiological permeability, cylindrical geometry, cell-matrix interactions, and shear stress.²⁵ Microvessels formed from dhBMECs exhibit low permeability of small molecules, consistent with physiological BBB function in animal models.^50,51 Previous in vitro studies have relied on modeling hyperosmotic BBBO within two-dimensional microfluidic systems⁵² or hollow-fiber-based platforms,^53,54 where direct visualization of cell behavior is challenging. There are two main limitations to our model for studying hyperosmotic BBBO. (1) Microvessels are comprised of only BMECs. Recent evidence suggests that other cellular components of the BBB are not necessary to maintain physiological permeability in vitro,^25,55 but may play a critical role in the response to injury or stress as observed in vivo,^56,57 which is not captured here. (2) Microvessels mimic features of brain post-capillary venules but are larger than brain capillaries (typically 8–10 μm). The influence of diameter on BBBO susceptibility is unknown. These differences may explain discrepancies in mannitol doses required to initiate BBB opening.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by DTRA (HDTRA1-15-1-0046) and NIH (NINDS R01NS106008). RML acknowledges a National Science Foundation Graduate Research Fellowship under Grant No. DGE1746891, JD acknowledges support from the Nanotechnology for Cancer Research training program, PW acknowledges support from NIH (NINDS R01NS091110).

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Authors' contributions

RML and PCS wrote the paper. RML, JGD, MBS, and PCS designed the in vitro experiments. RML, JGD, and MBS performed the experiments and analyzed the data. CC and PW designed, conducted, and analyzed in vivo experiments. All authors reviewed and approved the manuscript.

Supplemental material

Supplemental material for this article is available online.

Data availability statement

The raw/processed data required to reproduce these findings are available from the corresponding author on reasonable request.