Modular microfluidic systems cast from 3D-printed molds for imaging leukocyte adherence to differentially treated endothelial cultures

Microfluidic systems are very useful for in vitro studies of interactions between blood cells and vascular endothelial cells under flow, and several commercial solutions exist. However, the availability of customizable, user-designed devices is largely restricted to researchers with expertise in photolithography and access to clean room facilities. Here we describe a strategy for producing tailor-made modular microfluidic systems, cast in PDMS from 3D-printed molds, to facilitate studies of leukocyte adherence to endothelial cells. A dual-chamber barrier module was optimized for culturing two endothelial cell populations, separated by a 250 μm wide dividing wall, on a glass slide. In proof-of-principle experiments one endothelial population was activated by TNFα, while the other served as an internal control. The barrier module was thereafter replaced with a microfluidic flow module, enclosing both endothelial populations in a common channel. A suspension of fluorescently-labeled leukocytes was then perfused through the flow module and leukocyte interactions with control and TNFα-treated endothelial populations were monitored in the same field of view. Time-lapse microscopy analysis confirmed the preferential attachment of leukocytes to the TNFα-activated endothelial cells. We conclude that the functionality of these modular microfluidic systems makes it possible to seed and differentially activate adherent cell types, and conduct controlled side-by-side analysis of their capacity to interact with cells in suspension under flow. Furthermore, we outline a number of practical considerations and solutions associated with connecting and switching between the microfluidic modules, and the advantages of simultaneously and symmetrically analyzing control and experimental conditions in such a microfluidic system.


molds
To analyze the difference between the intended and actual z-height of PDMS structures cast from 3D printed molds a mold containing 9 cuboid structures with heights increasing in steps of 100 µm from 100 µm to 900 µm was designed and drawn using Fusion 360 software (Fig.   S3a). Three replicates of the mold were 3D printed using the same FormLabs printer used to print the barrier and flow modules. PDMS chips were cast from the mold and three crosssections (approximately 1 mm thick) were cut through the length of each chip using a microtome blade. Each cross-section was then mounted onto a glass slide and images of each of the PDMS wells (100-900 µm z-axis dimensions) were captured using the DIC function on a confocal microscope (LSM 700, Zeiss, Jena, Germany). The images were imported to ImageJ where the line tool was used to manually measure the actual height of each of the PDMS wells (Fig. S3b insert and c). The average of the heights measured for each well from each of the three cross-sections was determined for each mold. The average ± standard deviation for each well from the three replicate molds was plotted against the intended height, and a linear regression analysis was performed using Prism 7 software (Fig. S3d).

Image analysis of leukocyte deceleration on HUVECs in the flow module
Time-lapse images of Celltracker green leukocyte fluorescence was captured in the flow module at 5 s intervals using a 20x objective. An example of leukocyte deceleration and attachment to HUVECs was selected (Fig. S5a). In ImageJ the time-lapse sequence of interest was isolated as a substack and the x-and y-coordinates for the front edge of the leukocyte of interest determined by manually defining ROIs using the multi-point tool (Fig. S5b). This permitted the velocity of the leukcoyte from one frame to the next to be calculated as the product of the distance travelled divided by the time interval between frames (Fig. S5c). The trajectory of the leukocyte in the time-lapse sequence of interest was also represented as a temporal color-coded projection using the ImageJ hyperstack function (Fig. S5b 0-30 s).

Effect of vacuum bonding on the height of the flow module
To assess the effect of evacuating the vacuum grid on the geometries of the flow module the volume of the flow chamber was imaged with the vacuum applied (on) and switched off. To visualize the flow chamber volume fluorescent microspheres (FluoSpheres carboxylate (580/605), Thermo Fisher Scientific) with a diameter of 2 µm were suspended in a 0.4 % solution of agarose (Top Vision Agarose, Thermo Fisher Scientific) dissolved in water. The solution was drawn into a flow module mounted on a cover glass where it polymerized as an easily deformable soft gel (Fig. S6a). The device was transferred to a confocal microscope stage (LSM 700, Zeiss, Jena, Germany) and the center of the flow module was visualized with a 20X objective using Zen imaging software (Zeiss). With no vacuum applied a z-stack from the bottom to the top of the flow chamber (as determined by the lowest and highest position in which fluorescent spheres were detected) was acquired. The pinhole was set to achieve an optical section thickness of 1 µm. The vacuum was turned on and the same zstack acquisition settings were applied to collect a second stack of the flow chamber. The fluorescent microspheres in the z-stack acquired with the vacuum off were pseudo-colored red, and those in the z-stack acquired with the vacuum on were pseudo-colored green. Using ImageJ image analysis software the z-stacks of the x, y planes were projected into a single image, and the spatial overlap between images of microspheres acquired with the vacuum off and on were compared by merging the two images into one composite image (Fig. S6b). To assess the spatial distribution of microspheres along the z-axis of the flow module the zstacks were resliced using ImageJ such that the z, x axis were presented in a y-stack, which was projected into a single z, x plane. The projected z, x stacks of microspheres imaged with the vacuum off and on were merged to create a composite image which allowed for the vacuum-mediated displacement of microspheres to be visualized (Fig. S6c). Individual microspheres in the upper (n=35) and lower (n=32) positions of the z, x projected stack acquired with the vacuum off were identified as regions of interest (ROIs), and the position of the same microspheres were identified as ROIs in the z, x projected stack acquired with the vacuum on. The co-ordinates for each of the ROIs were exported to Excel and the formula for the distance between two-points (Ö(x 2 -x 1 ) 2 +(y 2 -y 1 ) 2 ) was applied to calculate the displacement of microspheres as a result of evacuating the vacuum grid. The displacement for each microsphere analyzed in the upper and lower part is presented as a percentage of the channel height (200 µm) using a scatter dot plot prepared using Prism 7 software (Fig. S6d).

Image analysis of leukocyte distribution in the flow module
The final time-lapse images of Celltracker green leukocyte fluorescence, from three independent adherence experiments (also analyzed in Fig. 5

Laminar flow properties and analysis of potential transverse diffusion in the flow module
To assess the flow patterns within the flow module a black dye was diluted in water to yield a suspension containing easily distinguishable dye particles. The dye suspension was drawn into the flow module and filled the chamber (Fig. S8a). The flow module was placed on the stage of the Axiovert 200M microscope (Zeiss) and the central position across the midline of the chamber was viewed using the live feed from the AxioVision imaging software, which was recorded in real-time using screen capture software. As for the leukocyte experiments, a withdrawal flow with a rate of 4 µl/min was applied to the chamber. The screen captured videos of dye particles passing through the flow module were imported to ImageJ as image stacks and individual particles were tracked using the Manual Tracking ImageJ plugin over 100 frames. The complete track (dot-and-line) for each tracked particle was recorded and presented overlaid on the first frame of the image sequence (Fig. S8b), and for clearer visualization were also presented on a black background (Fig. S8c).
To determine the risk of transverse diffusion in the flow module (i.e. whereby solutes from one side of the module would diffuse to the opposite side) a finite elements model of the flow module was constructed in COMSOL to simulate the Ctrl and TNFa-treated HUVEC populations. The model aimed to simulate a hypothetical situation where a factor such as TNFa was constantly released at a fixed concentration from the TNFa treated HUVECs.
This scenario was modeled to depict the release of TNFa bound to the surface of the treated endothelium or the induced release of a similar cytokine. A flow rate of 4 µl/min was applied to the model and the diffusion coefficient for the simulated TNFa was obtained from a mathematical approximation published by Goodhill et al. 1 After 25 mins of simulated flow in the model (i.e. the same duration as the leukocyte experiments) the average concentration of TNFa on the Ctrl and TNFa sides of the model were calculated (Fig. S8d). This demonstrated that 98.16% of the simulated factor remained on the side of the chamber from which it was released, while a minimal degree of transverse diffusion resulted in the opposite side of the chamber (simulating cells not releasing this hypothetical factor) receiving an average of 0.16% of the simulated factor. The flow speeds at the withdrawal rate of 4 µl/min were also modeled within the flow module (Fig. S8e).