Effect of flow on targeting and penetration of angiopep-decorated nanoparticles in a microfluidic model blood-brain barrier

The blood-brain barrier (BBB) limits transport of nanoparticles from the circulation to the brain parenchyma. Angiopep-2, a peptide which functions as a brain transport vector, can be coupled to nanoparticles in order to facilitate binding and internalization by brain endothelial cells (ECs), and subsequent BBB penetration. This multi-step process may be affected by blood flow over brain ECs, as flow influences endothelial cell phenotype as well as interactions of nanoparticles with ECs. In the present study a microfluidic BBB model was constructed to evaluate binding and internalization by brain ECs, as well as BBB penetration of Angiopep-2 coupled liposomes (Ang2-Liposomes) in static and flow conditions. Ang2 conjugation to liposomes markedly improved binding relative to unconjugated liposomes. Ang2-Liposomes bound and were internalized efficiently by brain endothelial cells after static incubation or with 1 dyne/cm2 of fluid shear stress (FSS), while binding was reduced at a FSS of 6 dyne/cm2. Penetration of the model microfluidic BBB by Ang2-Liposomes was higher at a FSS of 1 dyne/cm2 and 6 dyne/cm2 than with static incubation. Analysis of barrier function and control experiments for receptor-mediated penetration provided insight into the magnitude of transcellular versus paracellular transport at each tested FSS. Overall, the results demonstrate that flow impacted the binding and BBB penetration of Ang2-functionalized nanoparticles. This highlights the relevance of the local flow environment for in vitro modeling of the performance of nanoparticles functionalized with BBB penetrating ligands.


Introduction
The blood-brain barrier (BBB) is a unique interface in the brain that tightly regulates the flow of ions, nutrients, and metabolites between flowing blood and brain tissue [1][2][3][4][5][6]. Brain endothelial cells (ECs) form the primary physical barrier to mass transport across the BBB due to the presence of intercellular tight junction complexes that restrict paracellular transport. Transcellular transport is restricted by limited pinocytosis, and regulated by membrane transporters and receptors that selectively transport substances across brain ECs [7,8] increase hydraulic conductivity, [49,50,52], and enhance cell turnover, [45,53], potentially opening the paracellular route to solutes (e.g. dextran, [44], ldl, [54], albumin [55]) or perhaps even nanoparticles. Flow may also alter EC phenotype in ways that promote BBB penetration via receptor-mediated transcytosis. Binding to brain ECs (the first step of receptor-mediated transcytosis) can be enhanced by FSS [56][57][58][59]. This is because FSS can increase expression of cell membrane-bound receptors (e.g. cell adhesion molecules, [56,57], low density lipoproteins (ldl), [58,59]) that can be targeted by ligand-functionalized nanoparticles [32,60]. Internalization of nanoparticles by ECs can also be affected by flow-induced changes to EC phenotype. While adaption of ECs to chronic flow reduces internalization of polymer nanoparticles targeted to PECAM-1 or ICAM-1 [30,61], acute exposure to flow can enhance internalization of PECAM-1-targeted nanoparticles [30,62]. Internalization of solutes by ECs (e.g. ldl, acetylated ldl, horseradish peroxidase) is also modulated by flow [54,63,64]. In addition, FSS increases the density of caveolae, a mechanosensitive domain which mediates transcytosis, on the luminal plasma membrane of ECs [65][66][67]. Overall, these studies suggest that flow may increase or decrease BBB penetration of Ang2-functionalized nanoparticles, depending upon the magnitude and exposure time of the generated shear stress. BBB penetration of Ang2-functionalized nanoparticles has been observed in static in vitro BBB models as well as in vivo where sustained flow is present, but the effect of acute flow has not been previously investigated [13,[15][16][17][20][21][22][23].
To advance our understanding of this phenomenon, a microfluidic BBB model was constructed. Microfluidic models are increasingly utilized for preclinical drug development with the aim of improving the attrition rate of drugs from the laboratory to clinical trials [68,69]. These systems provide microscale control of fluid flows, tissue architecture, biochemical stimuli (e.g growth factors, cytokines), and biomechanical stimuli (e.g. fluid shear stress, cyclic stretch) in order to simulate the tissue microenvironment with high spatial and temporal precision [68,69]. Simulating the tissue microenvironment may be particularly relevant for in vitro BBB modeling. The microenvironment, moreso than intrinsic characteristics of brain ECs, is thought to be critical to promoting a BBB phenotype [5,7,70]. Current microfluidic BBB models do not fully replicate conditions in vivo, but are powerful tools which can simulate many aspects of the BBB microenvironment in physiological or pathophysiological conditions [71]. Microfluidic BBB models have simulated luminal FSS [72][73][74][75][76][77], interstitial fluid flow [78], cyclic strain [78,79], 3D extracellular matrix [74,80,81], capillary-like geometry [80,82,83], co-cultures with other cell types of the neurovascular unit [73-77, 81, 84], and optimization of fluid-to-tissue ratio [74,85]. Microfluidic models have been utilized to investigate hemodynamic effects on binding of nanoparticles [31,37,39,40,86]. Compartmentalized microfluidic BBB models incorporating flow can be utilized to evaluate BBB penetration and monitor barrier phenotype with fluorescence microscopy [72,73,75,77,79,83,85,87,88].
In the present study, brain ECs were grown in the upper microchannel of a compartmentalized microfluidic device to establish a model BBB. Brain ECs were grown on a topographicallypatterned substrate, as flow and topography can work synergistically to influence cell phenotype [89]. Ang2-liposomes were then incubated in the upper microchannel in static fluid or subjected to acute flow in order to test effects on binding to brain ECs and BBB penetration.

Microfluidic device structure and fabrication
The microfluidic device structure consists of S-shaped microchannels where the region of overlap is separated by a porous membrane patterned with 1 micron wide pits as described previously [90]. Briefly, polydimethylsiloxane (PDMS) microchannels were molded from SU-8 wafers and bonded to a polycarbonate membrane using a silicone adhesive, followed by oxygen plasma treatment, bonding to a glass coverslip, sterilization with 70% ethanol and autoclave, and washes with phosphate buffered saline (PBS). The upper channel dimensions are 5 mm x 0.5 mm x 0.1 mm in length x width x height, and the lower channel dimensions are 5 mm x 0.25 mm x 0.1 mm (surface area of the region of overlap is 1.25 mm 2 ). Fluidic ports were used to introduce liposomes or dextran into the upper channel, and to collect permeated solutes from the lower channel.
For experiments in microfluidic devices, the topographically-patterned polycarbonate membrane was treated with a solution of 100 μg/ml of fibronectin and 100 μg/ml collagen IV in PBS for 15-24 hours at 37˚C prior to bEnd.3 cell seeding. bEnd.3 cells were allowed to attach for 2 hrs, and then channels were aspirated to remove unattached cells. The cells were cultured over 3-6 days with daily replacement of the medium prior to experiments.

TEER measurement
TEER of confluent monolayers was measured after 4 days culture using Ag/AgCl sintered wire electrodes with 0.20-mm diameter (A-M systems). Electrodes were added to the fluidic ports of the device and connected to an in-house device which adapted the current to a range measurable with an epithelial voltohmeter (EVOM2, WPI). The measured resistance of devices was subtracted from the resistance with no cells present, and multiplied by the surface area of the monolayer (2.5 mm 2 ) to calculate TEER.

Immunofluorescence imaging of tight junction proteins
After exposure to FSS for 2 hours, bEnd.3 monolayers were washed with HBSS containing calcium and magnesium at 37˚C to remove medium and cell debris, fixed with 3.7% paraformaldehyde (PFA)/phosphate buffered saline (PBS), washed again to remove excess PFA, blocked for 45 min with 3% BSA/PBS, and permeabilized with 0.3% triton for 10 min. Primary antibody for Claudin-5 (5 µg/ml, rabbit polyclonal, ThermoFisher) was added to the upper channel for 2-3 hours at room temperature. The cells were then washed with PBS, and Alexa-Fluor 488 anti-rabbit IgG was used to label the primary antibody (1:400) for 30 min at room temperature, followed by DAPI for 5 min (40 μg/ml) as a counterstain for cell nuclei.

Liposome binding and internalization by brain ECs
Liposome binding and internalization was visualized with fluorescence imaging and quantified with fluorescence spectroscopy. For fluorescence imaging experiments, bEnd.3 cells were incubated with Ang2-Liposomes for varying times/concentrations in static fluid or in the presence of FSS (detailed in figure legends), washed to remove unbound liposomes, fixed with 2% PFA, stained with DAPI, imaged with a scanning confocal microscope (Olympus FV1000), and processed using FIJI software. To quantify liposome binding, bEnd.3 cells were grown directly in cell culture plates. Following incubation with Ang2-liposomes, the cells were lysed with 1%Triton/1M NaOH and fluorescence detected with a SpectraMax M5 plate reader (ex/em 560/595 nm) which was converted to liposome concentration using a calibration curve (S1 Fig). For the competition experiment, a rabbit polyclonal anti-mouse LRP1 antibody (MyBioSource) was preincubated with bEnd.3 cells for 45 minutes at 37˚C, then co-incubated with Ang2-liposomes for 60 minutes at 4˚C.
To assess internalization, Z-stacks of bEnd.3 cells (0.6-0.9 microns per slice) were acquired through the entire cell, and assembled into 3 dimensional images using FIJI software.

Permeability of dextrans and liposomes in microfluidic devices
Permeability of 4 kDa, 20 kDa, and 500 kDa FITC-dextran (Sigma-Aldrich) and rhodaminelabeled liposomes was assessed by measuring tracer flux across the brain EC cell barrier. Devices were prepared as described in the previous section prior to addition of the tracer. The tracer was then added to the luminal channel (0.05-2 mM FITC-dextran incubated statically for 15-20 min or 120-720 pM liposomes incubated in the presence or absence of flow for 120 minutes). For the competition experiment with antibody for LRP1, anti-LRP was preincubated in devices for 45 min prior to adding Ang2-liposomes with co-incubated anti-LRP (33 μg/ml) for 2 hrs of static incubation. After this time period, a sample from the abluminal channel was collected in 100 μl PBS, and fluorescence intensity was measured using a microplate reader (FITC dextrans: ex/em 488/525 nm, Liposomes: ex/em 560/595 nm). Known masses of FITCdextran or liposomes were measured under the same excitation/emission settings and sample volume (100 μl PBS) to generate a calibration curve (S1 and S2 Figs), which was used to convert fluorescence signal intensity to mass of dextran or number of liposomes. The permeability coefficient was calculated using the following equation: where PS is the permeability, ΔC L is the change in the lower channel concentration, V U is the volume of the upper channel, C U is the initial concentration of the upper channel, A is the membrane surface area, and Δt is the change in time. The apparent permeability coefficient, P app , was then calculated as: where PS cells is the permeability across the brain EC barrier and porous membrane, while PS no cells is the permeability across the porous membrane alone.

Statistics
Data were calculated as the mean ± standard error of the mean (SEM). Statistical significance was determined by a Student's t-test when only two conditions were present, and one-way ANOVA with post hoc Tukey's test when > 2 conditions were present where p < 0.05 was depicted with one symbol in graphs, p < 0.01 was depicted with two symbols, and p < 0.001 was depicted with 3 symbols.

Results
Liposomes were used as a model nanoparticle for functionalization with Ang2. Ang2-Liposomes were fabricated to a final diameter of 80-95 nm, and contained rhodamine in the lipid shell for detection purposes, PEGylated lipid commonly used to extend circulation time in vivo, and Angiopep-2 conjugated to a fraction of the PEGylated lipid ( Fig 1A). The microfluidic device structure consisted of two overlapping channels separated by a patterned porous membrane with murine brain ECs (bEnd.3) grown to confluence in the upper channel ( Fig 1B  and 1C). The device was used to assess binding and internalization of Ang2-Liposomes by brain ECs via fluorescence microscopy, and penetration of the brain EC barrier via fluorescence spectroscopy (Fig 1B). During the experiment the upper channel was opened to incubate Ang2-Liposomes with flow or in static fluid, while the lower channel was closed and kept static to limit advection between the upper and lower channels (Fig 1C). After the experiment the upper channel was closed, and the lower channel was opened to collect samples for analysis.

Characterization of microfluidic BBB model in static fluid
We first confirmed formation of a functional barrier in the microfluidic BBB model after 3-6 days static culture (i.e. the time frame when experiments of Ang2-Liposome penetration were initiated). Confluent monolayers formed with an average transendothelial electrical resistance (TEER) of 172 ± 8.5 ohm x cm 2 ( Fig 1D). This value correlates with barrier formation in microfluidic BBB models utilizing bEnd.3 cells, whereas baseline TEER is in the range of 30-50 ohm x cm 2 [72,83,92]. An inverse relationship between permeability of solutes and solute size was observed as permeability of FITC-labeled dextrans of sizes 4 kDa, 20 kDa, and 500 kDa was 0.48 x 10 −6 cm/s, 0.35 x 10 −6 cm/s, and 0.09 x 10 −6 cm/s, respectively (Fig 1D), although only the difference between 20 kDa and 500 kDa was statistically significant. Taken together, TEER and the size-selectivity of dextran permeability supported that a functional barrier had formed.

Binding and internalization of Ang2-liposomes by brain ECs in static fluid or in the presence of flow
We first confirmed the ability of Ang2 to facilitate binding of liposomes to brain ECs in static fluid. Functionalization of liposomes with Ang2 resulted in significant binding to bEnd.3 cells compared with non-functionalized liposomes (Fig 2A and 2B). After incubation for 45 minutes at 37˚C (cell binding and internalization occurs concurrently at this temperature), Ang2-liposomes bound bEnd.3 cells at 1244 ± 183 liposomes/cell, an increase of 5.1-fold over binding of non-functionalized liposomes (243 ± 171 liposomes/cell). When incubated for 90 minutes, binding of Ang2-liposomes increased to 2334 ± 489 liposomes/cell, an increase of 13.8-fold over non-functionalized liposomes (169 ± 31 liposomes/cell). Binding of Ang2-liposomes to bEnd.3 cells was inhibited by an antibody recognizing LRP1 which reduced binding of Ang2-liposomes by 59% (Fig 2C), supporting that binding was mediated by interactions between Ang2 and the LRP1 receptor. Binding of Ang2-liposomes was concentration-dependent, saturable, and with avidity in the range of ligand-functionalized nanoparticles [32,93], providing further evidence that binding of Ang2-Liposomes to brain ECs was mediated by ligand-receptor interactions (Fig 2D). We next tested the effect of FSS on binding of Ang2-Liposomes. At a concentration of 120 pM Ang2-liposomes for each condition, binding was reduced when incubated at 6 dyne/cm 2 relative to 1 dyne/cm 2 or in static fluid. This suggested that while Ang2-Liposomes bound brain ECs efficiently in static fluid or low FSS, nanoparticle detachment forces introduced by FSS overcame the avidity of Ang2-Liposomes at higher FSS.
Next the internalization of Ang2-liposomes by brain ECs was evaluated by confocal fluorescence microscopy. Z-stacks indicated that Ang2-liposomes were internalized with perinuclear localization observed at 37˚C (endocytosis was active, Fig 3A i-ii, whereas Ang2-liposomes were localized mainly at the cell surface at 4˚C (endocytosis was inactive, Fig 3A iii-iv). To estimate magnitude and efficiency of internalization in static fluid, binding of Ang2-liposomes over time at 4˚C vs 37˚C was tested (Fig 3B). Binding of Ang2-liposomes was 2.9-fold greater at 37 o C than at 4 o C after 45-minute incubation (1245 ± 262 liposomes/cell vs. 435 ± 84 liposomes /cell), and 6.9-fold greater after 90-minute incubation (3900 ± 372 liposomes/cell vs. 567 ± 123 liposomes/cell). This suggested that most Ang2-liposomes were internalized by brain ECs rather than remaining on the cell surface, and were internalized relatively rapidly. Having established that Ang2-Liposomes were internalized in static fluid, we next evaluated internalization in the presence of FSS. After 2 hours of incubation at 37˚C, internalized Ang2-Liposomes were observed at 1 dyne/cm 2 and 6 dyne/cm 2 (Fig 3C). Cells with perinuclear localization of Ang2-Liposomes (e.g. as displayed in Fig 3A and 3C) were selected for analysis via z-stacks. Perinuclear localization was frequently observed at 1 dyne/cm 2 as in static condition, and was relatively rare at 6 dyne/cm 2 , likely due to reduced binding at 6 dyne/cm 2 . Overall, these studies suggest that FSS significantly impacted binding levels relative to incubation in static fluid, while it remained possible for Ang2-Liposomes to internalize in the presence of FSS.

Ang-2 liposome penetration of brain ECs in the presence of FSS
We next evaluated penetration of the microfluidic BBB model by Ang2-liposomes as a function of FSS. Ang2-liposome penetration was observed in static fluid, and in the presence of FSS (Fig 4A). Penetration of Ang2-liposomes was 1.6±0.6 x 10 −7 cm/s at 1 dyne/cm 2 and 2.1 ±0.7 x 10 −7 cm/s at 6 dyne/cm 2 , an enhancement of 2.7-fold and 3.5-fold relative to penetration in static fluid (0.6±0.1 x 10 −8 cm/s). Functionalization of liposomes with Ang2 enhanced  Fig 4B). Ang2-Liposomes did not appear to affect barrier function, as Ang2-Liposomes co-incubated with dextran in static fluid did not affect dextran permeability, whereas co-incubation of dextran with trypsin significantly enhanced dextran penetration (Fig 4C). Competitive inhibition with anti-LRP1 antibody in static fluid reduced the penetration of Ang2-liposomes by 37%, although this reduction was not statistically significant (p = 0.12, Fig 4D).
A key question of interest was the distribution of tight junction proteins after culture of brain ECs on pit-patterned substrate and acute exposure to FSS. As shown in Fig 5, the expression of Claudin-5 appeared primarily perinuclear in brain EC cultures after 2 hours of flow exposure (i.e. 6 days static incubation of brain ECs on pit-patterned membranes, followed by 2 hours of flow, Fig 5). This was in contrast to expression of Claudin-5 at intercellular borders with brain ECs cultured in cell culture plates (6 days static incubation). The observation suggested that tight junctions were not fully intact during the flow exposure, which may have influenced the penetration of nanoparticles through the model BBB.

Discussion
Nanoparticles functionalized with Ang2 are under development to improve treatment of diseases affecting the brain [13][14][15][16][17][18][19][20][21][22]. Understanding how the BBB microenvironment affects binding and BBB penetration is critical to optimizing performance of Ang2-functionalized nanoparticles in (patho)physiological conditions. Luminal fluid flow over brain ECs is an aspect of the BBB microenvironment which merits investigation in this context, as flow impacts nanoparticle interactions with ECs, and modulates EC phenotype [28-30, 35, 38-40]. In the present study, a microfluidic BBB model was utilized to examine the effect of flow on binding and BBB penetration of Ang2-functionalized liposomes.
Binding of Ang2-Liposomes to brain ECs was efficient in static fluid or at low FSS (1 dyne/ cm 2 ), but was inhibited at higher FSS of 6 dyne/cm 2 (Fig 2E). This bears consideration for the design of Ang2-functionalized nanoparticles, as physiological FSS in brain capillaries ranges from approximately 5-23 dyne/cm 2 [24,94]. Designing Ang2-nanoparticles with higher avidity via the multivalent effect could help withstand nanoparticle detachment forces imparted by the presence of flow. However, increasing avidity beyond a certain threshold can reduce BBB penetration of ligand-functionalized nanoparticles [14,95]. It may be necessary to find a balance by designing avidity to enable binding to brain ECs in the presence of FSS, and to maximize subsequent BBB penetration. Other nanocarrier characteristics (e.g. size, shape) can also influence binding and subcellular transport, [96], and may aid in optimizing BBB penetration. On the other hand, lowering avidity of Ang2-nanoparticles could in theory be exploited to preferentially target vasculature where low FSS is present. Relatively few studies have examined the impact of flow on binding of ligand-functionalized nanoparticles <100 nm in size, and this size regime is suited for BBB penetration.
Penetration of the BBB model by Ang2-Liposomes was enhanced in the presence of flow relative to static incubation. At 1 dyne/ cm 2 , the results collectively suggest that penetration of the BBB model was primarily via receptor-mediated transport. This is because binding and internalization of Ang2-Liposomes by brain ECs remained efficient (Fig 2E, Fig 3C), penetration of non-functionalized liposomes was considerably less than Ang2-Liposomes (Fig 4B), Flow effects on targeting and BBB penetration of Ang2-nanoparticles and Ang2-Liposomes did not compromise barrier integrity (Fig 4C). Perinuclear localization of Claudin-5 after 2 hours of flow at 1 dyne/cm 2 suggested that at least partial disassembly of tight junctions had occurred (Fig 5), although the relative contribution of brain EC culture on pit-patterned membrane or exposure to flow was not assessed. BBB integrity is maintained in the absence of Claudin-5 for solutes larger than 800 daltons, [97], which supports that paracellular transport of Ang2-liposomes (80-95 nm in size) may have been restricted at 1 dyne/cm 2 despite lack of Claudin-5 at cell junctions. In addition, tight junction disassembly is thought to precede disassembly of adherens junctions [45]. Thus, adherens junctions may have remained intact despite perinuclear localization of Claudin-5. It is tempting to speculate that flowinduced upregulation of endocytic signaling enhanced penetration of Ang2-Liposomes via receptor-mediated transcytosis. In support of this, acute FSS of 1-15 dyne/cm 2 applied to static ECs enhances fluid phase endocytosis of horseradish peroxidase in a magnitude-dependent manner over 2 hours [63]. Additionally, transcytosis of Ang2-functionalized nanoparticles has been reported to occur via caveolae, [41], and acute FSS can increase presence of caveolae at the plasma membrane of ECs [65][66][67]. However, detailed mechanistic studies of Ang2-liposome penetration are needed to substantiate this hypothesis.
In the case of Ang2-Liposome penetration at 6 dyne/cm 2 , opening of the paracellular route appears more likely. This is because binding to brain ECs was reduced relative to static condition (Fig 2E), while penetration of the BBB model was enhanced (Fig 4A). It is possible that flow at 6 dyne/cm 2 induced paracellular opening to a greater extent than at 1 dyne/cm 2 . Acute changes in FSS can induce transient disassembly of junction complexes, [49][50][51], and increase endothelial cell turnover, [45,53], both of which are characteristic of leaky EC junctions [45]. These changes can be time or magnitude dependent, supporting that FSS of 6 dyne/cm 2 may have enhanced paracellular transport of Ang2-Liposomes to a greater extent than 1 dyne/cm 2 in the present study.
A limitation of our study is the use of immortalized brain ECs and lack of co-culture with other cells of the neurovascular unit [92]. bEnd.3 cell monocultures were used in the present study due to their durability, demonstrated barrier function in published microfluidic BBB models, [72,73,83], and expression of LRP1 for testing of Ang2-Liposomes [22]. However, coculture models and models incorporating primary cells better reflect in vivo physiology as the phenotype of brain ECs incorporates signaling input from other cells of the NVU [7,92]. LRP1 expression and accessibility, as well as endocytic signaling could in theory be affected by signaling from astrocytes or pericytes.

Conclusion
Understanding the role of the BBB microenvironment on the targeting and penetration nanoparticles is critical to developing more effective treatments for diseases affecting the brain. Flow impacts nanoparticle-EC interactions and can alter EC phenotype in a manner which influences binding and BBB penetration of Ang2-functionalized nanoparticles. The present study demonstrated that flow modulates binding to brain ECs and BBB penetration of Ang2-functionalized nanoparticles, highlighting the need for in vitro BBB models which replicate the local flow environment. Further investigation of these findings in more physiologically representative in vitro models is warranted. Future studies may focus on tuning nanoparticle characteristics (e.g. Ang2 valency) to enable binding in the presence of flow, while maximizing BBB penetration.