Primary human cerebral cortex microvascular endothelial cells (HCCMECs, Cell Systems: ACBRI 376) were cultured in endothelial cell medium (ScienCell Research Laboratories) and grown in 37°C at 5% CO₂. Human aortic endothelial cells (HAECs, passage 4, Lonza: CC-2535) were cultured in EGM-2^TM BulletKit^TM medium (Lonza, Switzerland) under the same environmental conditions. The medium was changed every 2–3 days of culture. Unless stated otherwise, experiments were performed at least 2 days after subculture between passages 7–9.

Commercially purchased human HDL (Kalen Biomedical LLC) was labeled with Alexa Fluor 568 NHS Ester (ThermoFisher Scientific) following the manufacturer's protocol. Following conjugation, the reaction was dialyzed against 3x1L sterile 1x PBS over 3 days at 4°C using Slide-A-Lyzer^TM Dialysis Cassettes, 10K MWCO, 0.5 mL (ThermoFisher Scientific) where dialysis buffer was changed after each day. The efficiency of Alexa Fluor labeling was verified by running dialyzed Alexa Fluor labeled HDL (AF568-HDL) on an SDS-PAGE to compare the amount of Alexa Fluor labeled ApoA1 (approximately 25 kDa) to free Alexa Fluor. Additionally, to ensure the HDL particle remained intact after Alexa Fluor labeling, the size of AF568-HDL was compared with unlabeled HDL by running on an agarose gel followed by lipid staining with Sudan Black B (Sigma-Aldrich).

A codon optimized cDNA encoding the human ApoA1 was synthesized with flanking restriction enzymes for subcloning by Integrated DNA Technologies DNA (Coralville, Iowa). The open reading frame was subcloned into the pET28b vector (Novagen) This recombinant ApoA1 was expressed in endotoxin-deficient ClearColi® BL21 (DE3) cells (Lucigen) and purified as previously described (Ryan et al., 2003), using a Talon resin (Clontech) (see Supplemental Figure 1).

Total internal reflection fluorescence (TIRF) microscopy videos were taken of HCCMECs seeded on 25 mm glass coverslips in a cell chamber on the Olympus cell TIRF Motorized Multicolor TIRF module mounted on an Olympus 1X81 microscope (Olympus, Hamburg, Germany). Samples were imaged using a 150x/1.45 NA objective with 561 nm excitation lasers and Volocity acquisition software. The penetration depth or the distance into the basal end of the cells that are excited was set at 110 nm. For each cell, 150 TIRF images were taken at a frame rate of 6.67 per second for a constant duration of approximately 22 s. At least 8 randomly selected fields were imaged in each experimental replicate. Unless stated otherwise 2.5 μg/mL AF568-HDL and nuclear stain NucBlue® Live ReadyProbes® (Molecular Probes, Oregon, United States) were added and incubated 10 min at 4°C. Cells were washed twice with PBS(+) after which HCCMECs were returned to fresh warm HPMI; the cell chamber was placed on a heated stage (37°C) stage and after 2 min images were taken. For competition with unlabeled HDL, cells were pre-chilled for 5 min at 4°C. 40x excess of unlabeled HDL was added with 2.5 μg/mL of the AF568-HDL and allowed to bind for 10 min at 4°C after which the coverslip was washed. Dyngo4a was added to the cells for 30 min at 37°C prior to HDL binding in the cold; Dyngo4a was left in the binding and imaging step. To inhibit SNARE-mediated exocytosis, 2.5 μg/mL AF568-HDL was allowed to bind for 10 min at 4°C after which the coverslip was washed and 2 μM N-ethylmaleimide (NEM) was added only during the imaging step.

Blinded and automated quantification of the transcytotic events was performed using a tracking algorithm for MATLAB as previously reported (Armstrong et al., 2015; Azizi et al., 2015). Briefly, the scripts correct the image for noise and local background using a Gaussian filter. Then putative vesicles are identified based on size (9–36 pixels², XY dimension, 73.5 nm/pixel), aspect ratio (>0.2; the ratio of the minor axis to the major axis) and intensity (threshold of 10% above mean image intensity). The tracking algorithm then tracks each moving vesicle based on a maximum-probability assessment of how closely those potential tracks resemble free and super-diffusive Brownian diffusion. The resulting tracks are analyzed for the duration of the vesicle being stationary in the TIRF field (vesicle docking), the speed of vesicular movement and the degrees to which the particles' movements deviates from free Brownian diffusion (γ). Vesicles undergoing fusion with the plasma membrane are identified as those having a γ significantly less than that of an equivalent model population undergoing Brownian diffusion, typically 0 < γ < 0.873, and which undergo a decrease in fluorescence signal over the last two time points of their tracks equivalent to a drop of at least 2.5 standard deviations of vesicular intensity over the entire period the vesicle has been tracked.

Confluent endothelial cells in a 6 cm dish were washed twice with cold PBS(+) and then scraped into 0.5 mL of cold PBS(+). The cells were collected by centrifugation for 5 min at 2,000 rpm, 4°C. Next, cells were resuspended and lysed in 100 μL of TN buffer (25 mM Tris pH 7.5, 150 mM NaCl, 10% Sucrose, 1% TX-100, 1 mM DTT, supplemented with protease inhibitors) for 30 min with shaking. The lysate was transferred to a 1 mL Wheaton 33 low extractable borosilicate glass mortar of the tapered tissue grinder with a clearance of 0.1–0.15 mm (Wheaton, USA–item #: 358133) along with 700 μL of 60% Optiprep™ (Axis-Shield, Oslo Norway). This solution was homogenized with 20–25 strokes with the steel rod attached PTFE pestle of the tissue grinder on ice and then transferred to the bottom of a ½ × 2 inch ultracentrifuge tube (Beckman Coulter, USA). Equal volumes of 40, 30, and 20% Optiprep™ (in TN buffer) were then layered on top in decreasing percentage. This was then subjected to centrifugation at 28,000 rpm for 18 h at 4°C in a Beckman Coulter L8-70 M Ultracentrifuge using a SW55Ti rotor. A total of 8 fractions (2 fractions per 10% change in Optiprep™) were collected from this density gradient and subsequently the protein was concentrated using methanol:chloroform precipitation. One fraction from each percentage Optiprep™ was then analyzed through western blot. For western blots, primary antibodies against the VE-Cadherin (Santa Cruz: sc-6458), cav-1 (Santa Cruz: sc-894), and SR-B1 (Novus Biological: NB400-104) were incubated with the membrane overnight at 4°C with shaking. Horseradish peroxidase-conjugated secondary against the appropriate species was then added for 1 h at room temperature and the blot was then developed.

To measure nitric oxide production, 5.2 μM 4-Amino-5-Methylamino-2′,7′-Difluorofluorescein (DAF-FM) Diacetate (Molecular Probes, Oregon, United States) made up in HBSS was added to HCCMECs and incubated at 37°C for 30 min. DAF-FM was aspirated and the cells were washed once with HBSS and resuspended with HBSS supplemented with NucBlue® Live ReadyProbes®. The fluorescent signal was captured using the ImageXpress Widefield High-Content Analysis System (Molecular Devices, California, United States) at 10x magnification in the DAPI and FITC channel. The average integrated intensity per cell was measured and calculated by the MetaXpress software (Molecular Devices).

Statistical analysis was performed using the GraphPad Prism software (GraphPad Prism 5.0; GraphPad Software Inc., La Jolla, CA, USA). Student's t-tests and one sample t-tests (GraphPad, La Jolla, CA, USA) were used to determine the significance of raw or normalized data (respectively). Data are presented as mean ± SEM. All experiments contained >10 technical replicates and were performed at least three times on different batches of cells. Post-hoc power analysis was performed using the Statistical Power Calculator (Decision Support Systems, Fort Worth, Texas) with a 5% alpha error level. All experiments had statistical power of 90-100%.

Primary human cerebral cortex microvascular endothelial cells (HCCMECs) were tested to see if their physiological properties are maintained when grown in vitro. They exhibited typical cobblestone morphology by phase contrast microscopy and formed monolayers with few gaps when grown to confluency (not shown). HCCMECs express junctional proteins VE-Cadherin and ZO-1 which enriched circumferentially at the plasma membrane where contacts are made with neighboring cells (Figure 1A). This suggests that HCCMECs are able to form cell-cell contacts essential for their barrier function.

The process of HDL transcytosis across endothelial cells in vivo occurs through a polarized manner from the apical to the basolateral end. This was tested in HCCMECs in vitro by labeling the apical membrane of HCCMECs on glass coverslips with elderberry bark lectin, (Armstrong et al., 2015) binding Alexa 568-labeled HDL (AF568-HDL) in the cold (4°C) and monitoring its intracellular localization over time. At time zero, most of the HDL (approximately 75%) was near the apical membrane whereas after 30 min, only 33% of total HDL remained near the apical membrane (p < 0.0001, n = 3); the remainder was intracellular and located at or approaching the basal membrane (Figures 1B–D). Together, these data suggest that HCCMEC maintain their polarized character in vitro and constitute a suitable model to study HDL transcytosis.

We began by focusing on the initial stage of HDL transcytosis, namely its internalization. DiI (1,1′-dioctadecyl- 3,3,3′,3′-tetramethylindocarbocyanine perchlorate) labeled HDL (DiI-HDL) was incubated with HCCMECs for 10 min at 37°C, leading to the appearance of intracellular red punctae (Figure 2A). The addition of 40-fold excess unlabeled HDL largely abrogated the fluorescent signal, consistent with competition for a cell-surface receptor. Similarly, the addition of 40-fold excess by mass of recombinant ApoA1 also inhibited HDL internalization (Figures 2A,B, 83% decrease compared to control). Accordingly, pre-treatment with the dynamin GTPase inhibitor Dyngo4a almost entirely blocked DiI-HDL internalization (Figures 2C,D, 92% decrease compared to labeled HDL alone). As dynamin is necessary for both clathrin- and caveola-mediated endocytosis (Oh et al., 1998), as well as other modes of internalization (Conner and Schmid, 2003), we depleted HCCMECs of clathrin heavy chain by siRNA transfection (approximately 70% depletion by western blot, Figure 2F); this led to no significant change in HDL internalization (Figure 2E). Recent studies have demonstrated that the presence of Mfsd2a largely abrogates caveola formation in brain microvascular endothelial cells (Andreone et al., 2017). To determine if the low-levels of caveolae could still be important in the uptake process we decided to test whether HDL is internalized through caveola-mediated endocytosis. In these experiments significant Cav-1 depletion (≈71% depletion by Western blot, Figure 2H) had no impact on HDL internalization (Figure 2G). Curiously, treatment of HCCMECs with the cholesterol-binding compound nystatin significantly decreased the amount of internalized HDL (reduction of about 81%; see Figures 2C,D). Thus, HDL internalization uses a dynamin- and cholesterol-dependent pathway distinct from clathrin and caveola.

As alluded to earlier, relatively little is known about the molecular mechanisms of HDL transcytosis due to the limitations of both EM and transwell studies. In particular, transwells are often confounded by paracellular leak and poor transfection efficiency (particularly of primary cells), while EM studies are often descriptive rather than quantitative. To circumvent these issues, we developed a method of quantifying transcytosis by single endothelial cells in a confluent monolayer (Armstrong et al., 2015; Azizi et al., 2015) by detecting the exocytosis of internalized fluorescently-labeled HDL through total internal reflection fluorescence microscopy (TIRFM) (see Schematic, Figure 3A; Steyer et al., 1997). Quantification is performed in a blinded manner using an automated MATLAB script. The TIRF assay obviates issues such as low transfection efficiency since transcytosis can be examined in individual cells within the monolayer. Importantly, this assay avoids the edge-effects of cells seeded on a transwell and is not affected by paracellular leak since it focuses on individual cells in the confluent monolayer.

AF568-HDL was added to the apical surface of the endothelial monolayer and could be detected entering the evanescent field and undergoing exocytosis (Figure 3B). The addition of 40-fold excess unlabeled HDL significantly decreased the number of fusion events (Figure 3C), again suggesting competition for a receptor. Pre-treatment with Dyngo4A or preventing exocytosis by adding N-ethylmaleimide (NEM) - an inhibitor of SNARE complex disassembly and recycling (Söllner et al., 1993) - both significantly decreased HDL transcytosis to the basal membrane (Figure 3D). Finally, a dose-response curve indicated that HDL transcytosis was saturable (Figure 3E) consistent with it being a receptor-mediated uptake pathway.

Studies in aortic endothelial cells have implicated SR-BI in HDL transcytosis (Rohrer et al., 2009); we chose to focus our initial experiments on this receptor. SR-BI is a bifunctional protein with the ability to function as both a scavenger receptor as well as a channel allowing for the bi-directional movement of cholesterol between the plasma membrane and lipoproteins (Neculai et al., 2013). However, its function as an endocytic receptor is less well characterized (Fruhwürth et al., 2013) especially in the cerebral circulation.

To begin to investigate the participation of SR-BI in HDL internalization, the localization of SR-BI was determined by immunofluorescence. Endogenous SR-BI was enriched at the cell surface and on intracellular vesicles. Internalized AF568-HDL co-localized with endogenous SR-BI to the degree that was significantly higher than expected by chance (Manders Correlation Coefficient for HDL to SR-BI = 0.247; p < 0.0001, Figure 4A). The fact that co-localization is partial is perhaps unsurprising given that other receptors are known to bind HDL (Calvo et al., 1997, 1998). Additionally, cell-surface SR-BI was also observed to be internalized over time in HCCMECs (Figure 4B), mimicking the behavior of surface bound HDL (Figures 1B–D, section Cell Culture). Unexpectedly, in contrast to previous reports of SR-BI residing in lipid rafts in human microvascular endothelial cells (Uittenbogaard et al., 2000), this does not appear to be the case in human brain endothelial cells as SR-BI co-fractionated with the non-lipid raft plasma membrane marker VE-Cadherin (Figure 4C). This is likely due in part to the lack of caveolae in HCCMECs.

To test the functional role of SR-BI in HDL internalization, HCCMECs were pretreated with a blocking scFv to SR-BI which decreased AF568-HDL internalization by ≈57% (Figure 5A) while depletion of SR-BI by siRNA (assessed through immunofluorescence, see Figure 5D) significantly attenuated HDL internalization by ≈40% (Figures 5B,C). Together, these experiments strongly implicate SR-BI in the internalization of HDL by primary cerebral microvascular endothelial cells in a clathrin- and caveolae-independent manner.

In the liver, the adaptor protein PDZ-containing 1 (PDZK1) is essential for the stability of SR-BI at the plasma membrane (Kocher et al., 2003). Furthermore, in bovine aortic endothelial cells, HDL binding to SR-BI stimulates activation of endothelial nitric oxide synthase and NO production in a PDZK1-dependent manner (Zhu et al., 2008). To determine whether PDZK1 is also involved in HDL internalization upon its binding to SR-BI in brain microvascular endothelial cells, the expression levels of PDZK1 were assessed by Western blot analysis. HCCMEC expressed almost undetectable levels of PDZK1 (Figure 6A); this was supported by qPCR demonstrating essentially no PDZK1 mRNA in HCCMEC (data not shown). This suggests that PDZK1 is very unlikely to play a major role in HDL uptake or transcytosis in HCCMECs. In support of this notion, overexpression of PDZK1 in HCCMECs had no significant impact on DiI-HDL internalization (Figures 6B,C).

The dispensability of PDZK1 for HDL uptake and transcytosis brought into question the role of downstream eNOS activation and nitric oxide (NO) production. The release of exogenous NO using the donor spermine NONOate - which was verified by using the sensitive intracellular NO sensor 4-Amino-5-Methylamino-2′,7′-Difluorofluorescein (DAF-FM) Diacetate (Figure 7A; Nakatsubo et al., 1998) - had no effect on HDL internalization (Figure 7B). Next, basal NO production was significantly reduced by the classic eNOS inhibitor L-NNA (Figure 7C); this led to an increase in AF568-HDL internalization by ≈2.5-fold (Figures 7D,E). L-NNA reduced basal NO production in human aortic endothelial cells to a similar degree as in the brain endothelial cells (Figure 7C), but this increased AF568-HDL internalization by only ≈40% (Figures 7D,E). These results suggest that there may be different mechanisms involved in HDL internalization by different vascular beds. Future studies will be needed to better understand the regulation post-endocytosis of the receptor and the fate of the ligand.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.