Physicochemical regulation of endothelial sproutingin a 3-D microfluidic angiogenesis model

2.1. Endothelial cell culture

Primary human umbilical vein endothelial cells (HUVECs) were purchased from Lonza and expanded in growth media consisting of BioWhittaker® medium 199 (M199; Lonza, Walkersville, MD) with endothelial cell growth supplement (ECGS; Millipore Co., Billerica, MA), 0.05% [v/v] heparin (10,000 U/mL stock, Sigma-Aldrich, St Louis, MO), 1.3% [v/v] L-glutamine (200 mM, Glutamax-1, Gibco), 20% [v/v] FBS, and 1% [v/v] PS.

2.2. Fabrication of 3-D microfluidic gradient device

A three-channel geometry was chosen for gradient generation²¹ (Fig. 1a). Channel dimensions were 100 × 100 μm in cross section, with 350 μm edge-to-edge spacing between parallel channels. The central microchannel was 3.7 mm in length. A silicon master was fabricated using standard lithographic processing and surface-treated with perfluorooctyltrichlorosilane (Alfa Aesar, Ward Hill, MA), prior to casting a negative Sylgard® 184 polydimethylsiloxane (PDMS; Dow Corning Co., Midland, MI) stamp for collagen micromolding. Milled plexiglass pieces (overall size 40×40 mm) were fabricated in order to define the overall geometry of the collagen scaffold (Fig. 1b) and provide feeding reservoirs to micropatterned channels. To promote collagen adhesion, surfaces to contact collagen were treated with 1% polyethyleneimine (PEI, Aldrich Chemical, St. Louis, MO) and 0.1% glutaraldehyde (GA, Fisher Scientific, Fair Lawn, NJ) followed by rinsing with de-ionized water. Rat tail-derived, acetic acid-extracted collagen (1% wt.) was prepared as previously reported²², and injection-molded into the plexiglass piece against the PDMS stamp. Stainless steel pins were used to mold the inlet and outlet ports. Following collagen cross-linking at 37°C for 30 min, the stamp was removed and channels were sealed with a collagen-coated (∼100 μm) glass coverslip in the plexiglass frame (Fig. 1c).

2.3. Microfluidic endothelial cell culture

HUVECs (up to passage 4) were seeded in the central channel of the three-channel microfluidic device. To this end, culture media was aspirated from all six reservoirs, and HUVEC suspension added to the center channel reservoirs (10 +/- 3 μL of a 2×10⁶ cell/mL suspension to each reservoir). Complete media aspiration from side channel reservoirs and close balancing of cell seeding volumes resulted in a packed channel and high cell seeding density (>1000 cell/mm² for confluent layers). After allowing cells to adhere, channels were perfused at 1 μl/min (∼1 dyne/cm²) with growth media using gravity-driven flow or a syringe pump (Fig. 1c).

2.4. Generation and characterization of VEGF gradients

Gradient experiments were carried out for 24 hrs on confluent endothelialized channels, after 48 hrs of culture under the above-described conditions. Gradient media consisted of growth media plus 1% [v/v] L-ascorbic acid (50 μg/mL; Acros Organics, Morris Plains, NJ) and 0.16% [v/v] tetradecanoyl phorbol acetate (TPA; 50 ng/mL; Cell Signaling Technology Inc., Danvers, MA). TPA, a reagent commonly used in angiogenesis assays, was included as it mimics endothelial cell activation typical of the tumor vasculature^22,23 Gradient media in the source channel was supplemented with 75 ng/mL VEGF-165 (R&D Systems, Minneapolis, MN) to generate a VEGF gradient of ∼90 ng/mL/mm across the cell-coated channel, which has been previously shown to achieve maximum sprout alignment¹¹. 250 μM 40 KDa FITC-dextran was also added to source media to verify the presence of a chemical gradient at the experimental endpoint. Growth media from input and output reservoirs of all three channels was analyzed for VEGF-165 content via enzyme linked immunosorbent assay (ELISA, R&D Systems, Minneapolis, MN) according to manufacturer's instructions.

2.5. Immunostaining

Staining reagents were delivered through all three channels by adding small volumes of solution to each inlet reservoir. A 1 hr PBS rinse was carried out following formalin fixation and after both primary and secondary antibody incubations. HUVECs were initially permeabilized and blocked with a mixture of Triton-X and BSA diluted in PBS for 30 minutes at room temperature, and then incubated overnight at 4°C with mouse anti-human CD31 (1:200, Invitrogen Co., Carlsbad, CA). Alexa Fluor® 488 goat anti-mouse IgG (1:500, Invitrogen Co., Carlsbad, CA) secondary antibody was added in the presence of DAPI (1:5,000) and Alexa Fluor® 568 phalloidin (1:100, both from Invitrogen Co., Carlsbad, CA) and incubated for 1 hr at room temperature.

2.6. Confocal microscopy and image analysis

Confocal microscopy (LSM 710; Carl Zeiss, Inc., Thornwood, NY) was used to capture z-stack images of endothelial cell-coated, immunostained channels. Images were taken with a 25× water immersion objective, with 3.75 μm z-step size. 11 frames were captured as a tile image, to record the entire microchannel length. Images were viewed and analyzed with ImageJ (NIH), Volocity (PerkinElmer, Waltham, MA) and Matlab (Mathworks, Natick, MA).

Confocal z-stacks were quantified for cell density and sprouting frequency using ImageJ and Matlab. Stacks were re-sliced with 2.5 μm steps along the axial channel direction using ImageJ. The fluorescence intensity of the actin channel from the cross-sectional image was used for analysis of invasion frequency and direction. Top and bottom channel walls were analyzed for local nuclear density because of the higher image resolution for these planes, as compared to sidewall images, which were reconstructed from confocal z-stacks. Channels were binned into ten sections along a 2.5 μm portion of channel length. Nuclear density (number of nuclei per area, for 50 μm wide region of top and bottom channel walls) was quantified in Matlab for each bin, as was total invasion pixel intensity from either the full channel height, or a 25 μm tall central portion of sidewalls, and either towards the VEGF source or sink. To quantify sprouting intensity, channel sidewalls were identified in cross-sectional images via intensity thresholding, and cellular actin extending beyond the channel boundary was quantified by summing the projected pixel intensity with a local bin, normalized to the bin intensity threshold used for wall identification. To quantify corner effects on cellular density, channel cross-sections were projected (average intensity) along a direction 45 degrees rotated from the y-axis (defined along the channel height) and projected pixel intensity was plotted.

2.7. Simulation of VEGF distribution

To simulate the concentration profile of VEGF throughout a cross-section of the entire construct, a 2-D mathematical model was defined. The construct included source, drain (endothelialized) and sink channels of dimensions 100 μm × 50 μm, similar to the microfluidic channel geometry following sealing with a collagen layer. The 2-D steady state concentration profile was calculated from Fick's Second Law using the finite difference method. Boundary conditions were defined by a constant concentration of VEGF (100 AU) at the source and a constant consumption rate at the walls of the drain (10 AU per unit time per channel, based on relative VEGF concentration measurements of culture media in the input and output reservoirs, data not shown). The concentrations were visualized across the construct using contour plots in MATLAB (The Mathworks, Natick, MA).

2.8. Simulation of endothelial cell secreted IL-8 distribution

To simulate the concentration profile of cell-secreted interleukin-8 (IL-8) in the vessel vicinity, a 2-D model was developed similar to that used to calculate VEGF profile. A uniform cell density of 1,400 cell/mm² was assumed to approximate the experimentally measured range of cell densities. An IL-8 secretion rate of 0.0035 pg/cell/24 hrs was experimentally determined with 2-D cultured HUVECs using ELISA, and incorporated into the model.

2.9. Statistical analysis

In analyzing directional bias in VEGF gradient experiments, only the highly invasive devices, referred to as type ii devices in the figures (N = 3) were included; such devices displayed a large number of invasions, as compared to the type i devices which exhibited few or no invasions (N = 8). Invasions directed either towards the source or sink channel were quantified in Matlab, and differences were analyzed with Student's t-test. For one of these type ii devices, sprout intensity versus local nuclei density was plotted and statistically evaluated by linear regression. Statistical analyses were performed in Graphpad Prism.

3.1. Microfluidic endothelial cell culture

3-D collagen-embedded microchannels were generated whose geometry deviated slightly from the lithographically defined dimensions (approx. 50 × 100 μm vs. 100 × 100 μm) due to sealing with the collagen-coated glass coverslip. Endothelial cells seeded into the central microchannel quickly adhered to collagen (10 min), and within 48 hrs of culture in endothelial cell growth media formed a confluent layer of cells coating all four walls (Fig. 1d). Without any further stimulation, this monolayer was stable with no evident invasions. Cell viability was ensured via perfusion of media through the channel at 1 μL/min; cells cultured under static conditions within microchannels either did not reach confluence, or delaminated from channel walls after an initial growth period (data not shown). Applied average flow rates of ∼1 μL/min corresponded to an average shear stress²⁵ of ∼1 dyne/cm². Although experimental shear stresses in these experiments were an order of magnitude or more lower than physiological values for similarly sized blood vessels^24,25, this flow rate ensured a media residence time within the cell channel that was significantly shorter than the calculated time for cellular oxygen consumption (∼1 s vs. 60 s), providing an explanation for reduced cell viability at significantly slower flow rates.

3.2. Gradient generation and vessel barrier function

Generation of a chemical gradient across the endothelialized microchannel was accomplished using two parallel side channels that served as a source and a sink for the factor of interest. Using this setup, gradient formation and steady-state stability were first tested using fluorescently-labeled 40 kDa dextran, approximating the molecular weight of VEGF. Image analysis of fluorescence intensity across the 3-channel device suggested that a steady-state gradient of FITC-dextran was established within three hours of initial delivery (data not shown) and was maintained for up to three days of cell culture (Fig. 2a). Consistent with these results, analysis of VEGF transport using ELISA of input and output media after 12 hrs of media flow confirmed that excess VEGF delivered via the source channel diffused across the microvessel to reach the sink channel at low, but detectable concentrations. Importantly, no detectable levels of VEGF were measured in the output reservoir of the endothelialized center channel (Fig. 2b), consistent with a diffusion barrier provided by the endothelial vessel as reported in our previous paper²⁰ We note that flow rate in the side channels was ∼20% slower than in the central channel due to the increased resistance inherent to these slightly longer channels. This lower flow rate did not compromise the ability of the sink channel to evacuate solute; the output VEGF concentration was measured to be ∼0.8% the source concentration. Together, these results suggest that gradient formation was driven by diffusion of VEGF through the collagen hydrogel to the sink channel rather than leakage into the center channel. Furthermore, mathematical modeling (Fig. 2c) confirmed the generation of relevant gradients¹¹ as cells on the source side of the cell channel were exposed to approximately twice the concentration of VEGF as compared to cells on the sink side, with a similar enhancement of VEGF concentration in the vicinity of cell corners as compared to central side positions (Fig. 2c, inset).

3.3. Invasion response to VEGF gradient

To test invasion response to VEGF gradients, the endothelial lining of the center channel, as established by culture in growth media for 48 hours, was exposed to VEGF gradient conditions for 24 hrs. Unexpectedly, this treatment resulted in two types of responses. In the first response type, vessels remained largely stable, exhibiting few sprouts into the surrounding collagen (Fig. 3a-i). In the second response type, application of a VEGF gradient resulted in the formation of abundant endothelial sprouts (Fig. 3a-ii) that were up to ∼100 μm in length, typically remained connected to the parent cell layer, and consistently contained multiple nuclei. Sprouts emanated from all vessel walls and corners (Fig. 3b-ii). These differences in invasion behavior were not related to geometrical variations between type i and type ii devices as similar channel cross sectional areas and thus flow and shear rates exhibited very different invasion behaviors (Fig. 3b). Quantification of invasion frequency by normalizing the sum of invasion pixel intensity over a 2.5 mm center region of the vessel revealed no statistically significant directionality (N=3) of invasions towards the VEGF source channel for the second response type (Fig. 3c); this lack of bias was unexpected based on previous reports of directional sprouting¹⁷ in 3-D gradient devices and was reproducible for multiple experiments.

Because multiple invasion response patterns were observed to result from seemingly identical treatment conditions, control experiments were conducted in which vessels were exposed to gradient media lacking VEGF. Under these nominally uniform culture conditions, vessels similarly exhibited two types of behavior. In the first class, vessels remained stable with no sprouts (Fig. 4a-i), whereas a large number of sprouts were observed in the second class (Fig. 4a-ii). However unlike in the VEGF gradient experiments, sprouts were observed to originate preferentially from channel corners (Fig. 4b). Culture of endothelial cells in open channels (i.e., channels not sealed by a second collagen layer) that were fully immersed in culture media also demonstrated 3-fold invasion enhancement at channel corners relative to sides confirming the relevance of this trend independent of flow conditions (data not shown).

3.4. Invasion response to variations in cell density

Enhancement of invasions at vessel corners led us to hypothesize that cell density may play a role in determining invasion behavior. Cell density analysis indicated an increase in nuclear density in the vicinity of vessel corners as revealed by projecting a 45 degree cross section rotation and plotting average pixel intensity across this projection for the DAPI fluorescence (Fig. 5a). Lower corners, i.e., corners embedded deeper within the collagen scaffolds, similarly exhibited increased average pixel intensity relative to sides, albeit at reduced levels due to depth-dependent reduction of optical signal in confocal imaging (Fig. 5a).

As changes in cell number and geometry can affect cell behavior due to varied autocrine signaling²⁶, we next assessed the chemical environment generated by endothelial cell-secreted factors. While ELISA analysis of HUVEC secretion revealed no detectable levels of VEGF (data not shown), IL-8 – an endothelial cell-secreted signaling molecule that may modulate sprouting in an autocrine manner²⁷ - was released at 0.0035 pg/cell/24 hrs. Using finite element modeling of IL-8, we demonstrated steeper gradients of this factor in the vicinity of vessel corners (Fig. 5b). This enhancement was due to direct geometrical effects as well as increased cell numbers in the corners, and it was consistent with augmented invasion observed at these locations. These changes may be compounded by the presence of exogenous VEGF (mathematically predicted to be 5% of the maximum concentration near endothelial cells facing the sink channel [inset Fig. 2c]) that is known to up-regulate IL-8²⁸. These results motivated a re-quantification of invasions for the type 2 response VEGF gradient experiments (N=3) to assess if removal of corner effects would reveal a directional bias towards the VEGF source in sprouts originating on the flat channel walls. While this approach revealed a nominal 2.2-fold increase in source–to-sink invasion ratio, this trend was not statistically significant (p=0.16). However, quantification of endothelial cell density for all experiments indicated that inter- as well as intra-vessel variations in cell density may regulate endothelial cell sprouting. Specifically, plotting total invasion frequency per bin along the channel length (source + sink, for each of ten bins) versus average wall nuclear density for a highly invasive VEGF gradient vessel (Fig. 5c-i) revealed a statistically significant non-zero slope (P<0.01). This suggested that cell density and invasion frequency are locally correlated within a single vessel. Furthermore, analysis of multiple devices confirmed that invasion frequency represents a function of cell density (Fig. 5c-ii). Together, these data suggest both intra- and inter-vessel correlation of cell density and sprouting.