Chemicals, culture media, and supplements(3-Aminopropyl) tiethoxysilane (APTES), bovine serum albumin (BSA), dextran from Leuconostoc spp., ethylenediaminetetraacetic acid (EDTA), fetal bovine serum (FBS), glutaraldehyde, human plasma, fibronectin, were from Sigma-Aldrich (St. Louis, MO, USA). The following buffers, culture media, and supplements were all from Gibco (Grand Island, NY, USA): Dulbecco’s Modified Eagle Medium (DMEM), Dulbecco’s Phosphate-Buffered Saline (PBS 1×), AIM-V medium, FluoroBrite DMEM, 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), minimum essential medium amino acids solution (EAA), and non-essential amino acids solution (NEAA); penicillin/streptomycin, sodium pyruvate, Collagen-I Bovine 5 mg/ml. L-Alanyl-L-Glutamine (L-Ala/L-Glu) was obtained from Bioswisstec AG (Schaffhausen, Switzerland). Ficoll-Paque PLUS density gradient media was from GE Healthcare (Uppsala, Sweden). The human NK cell isolation kit and LS columns were purchased from Miltenyi Biotec (Bergisch Gladbach, Germany); human recombinant interleukin 2 (IL-2) was from Novartis Pharma (Proleukin, Switzerland), 7-Aminoactinomycin D (7-AAD) from BD Biosciences (La Jolla, CA, USA); DELFIA EuTDA Cytotoxicity Reagents from PerkinElmer (Waltham, MA, USA). Hoechst 33342 solution (20 mM) and CellTrace Yellow Cell Proliferation Kit (Invitrogen), 16% formaldehyde solution (FA), methanol-free, were purchased from Thermo Fisher Scientific (Waltham, MA, USA). Draq7 (0.3mM) was from BioStatus (Shepshed, UK). Recombinant porcine TNF (pTNF) and human TNF (hTNF) was from R&D Systems (Minneapolis, MN, USA), and SYLGARD 184 Silicone Elastomer Kit from Dow Inc. (Midland, MI, USA). Details in Supplementary Table S1.Endothelial cells culturePrimary porcine aortic endothelial cells (PAECs), wild-type, kindly provided by Eckhard Wolf (Ludwig-Maximilians-University of Munich). Cells were isolated mechanically from porcine thoracic aorta obtained from female and male German landrace outbred pigs32 upon euthanasia after terminal experiments performed by different groups at LMU Munich, complying with the 3R principles. Cells were cultured in DMEM medium supplemented with 10% heat-inactivated FBS, 2 mM L-Ala/L-Glu, 1 mM sodium-pyruvate, 20 mM HEPES, 1× EAA and 1× NEAA, 100 U/ml penicillin, and 100 µg/ml streptomycin at 37ºC, 5% CO2. For live cell imaging, 16,000 PAECs per well were seeded in 96-well µ-Plates with black walls and optically clear flat bottom (Ibidi, Fitchburg, WI, USA) in phenol red-free FluoroBrite DMEM supplemented as described above. As indicated PAECs were treated with 100 ng/ml pTNF overnight.The human endothelial cell line EA.hy926 was kindly provided by Prof. Michelangelo Foti (University of Geneva) and cultured in the same media as PAECs. For live cell imaging, 16,000 EA.hy926 cells per well were seeded in 96-well micro-plates with black walls and optically clear flat bottom (Ibidi, Fitchburg, WI, USA) in phenol red-free FluoroBrite DMEM supplemented as described above. As indicated EA.hy926 cells were treated with 100 ng/ml hTNF overnight.Isolation and stimulation of human NK cellsHuman NK cells were isolated from healthy donors’ buffy coats obtained from the Blood Transfusion Centre of the University Hospitals Geneva, with approval from the ethical committee of our institution (CER18-00552). Blood samples were collected upon receiving an informed consent and immediately anonymized. The experimental protocols were conducted according to Swiss Federal regulations and the guidelines of the University of Geneva. Peripheral blood mononuclear cells (PBMC) were separated by density gradient centrifugation in Ficoll-Paque PLUS at 900×g, 20 min at room temperature, without brake. Once collected, cells were washed twice in PBS-EDTA by centrifugation at 200×g for 10 min. Thereafter, PBMC were used for NK cell purification by negative selection using NK isolation kits according to the manufacturer’s instructions. Flow cytometry analysis demonstrated 90–95% purity of CD3–CD56+CD16+ NK cells. NK cells were cultured at 1 × 106 cells/ml for 5–9 days in AIM-V medium supplemented with heat-inactivated plasma, 20 mM HEPES, 1 mM sodium-pyruvate, 2 mM L-Glu/L-Ala, 1× NEAA and 1× EAA, 100 U/ml penicillin and 100 µg/ml streptomycin at 37ºC, 5% CO2. On day one, 200 IU/ml IL-2 was added, thereafter, 50 IU/ml IL-2 every two days of culture.Characterization of NK cells by flow cytometryCell surface staining was performed on freshly isolated, and IL-2 activated NK cells at 4°C in the dark for 25 min with fluorophore-conjugated monoclonal antibodies directed against CD16, CD25, CD56, CD57, CD69, CD94, NKp30, NKp44, NKp46, CD158e, NKG2A, NKG2D (Supplementary Table S2). After washing, cell pellets were resuspended in 300 µl of staining buffer and labeled with 7-AAD for dead cell exclusion. Attune NxT was used for the sample acquisition and FlowJo (version 10) for data analysis.Cytotoxicity assay using standard release assayNon-radioactive DELFIA EuTDA Cytotoxicity Reagents (PerkinElmer, Waltham, MA, USA) were used for NK cell-mediated cytotoxicity assays according to the manufacturer’s instructions. Briefly, target cells (PAECs) were loaded for 20 min with fluorescence-enhancing ligand BATDA, washed three times with DMEM at 200×g, 4 min. NK and target cells were then plated in round-bottom 96-well plates at different effector-to-target (E:T) ratios (10:1, 5:1, and 1:1). Triplicates were prepared for each condition. Maximum release was obtained using labeled target cells that were completely lysed with Lysis Buffer (PerkinElmer), while spontaneous release was measured from labeled target cells in the absence of effector cells. After 2 h of incubation at 5% CO2 and 37 °C, the supernatants were mixed with 200 µl Europium salt solution to form a chelate and measured in a time-resolved fluorometer (EnVision 2014 Multilabel reader, PerkinElmer). Cytotoxicity was calculated with formulas provided by the manufacturer as a percentage (%) of specific lysis.3D microfluidic system for flow cytotoxicity assayThe microfluidic channels were made as described previously30, using polydimethylsiloxane (PDMS, from Silicone Elastomer kit), and mold needles (0.55 mm diameter) to create a cylindrical channel simulating small vessels. Inlet and outlet holes were created with a 2 mm biopsy punch. After oxygen plasma activation, the channels were treated sequentially with APTES, glutaraldehyde, fibronectin (50 µg/ml), and collagen-I (100 µg/ml). Ultimately, PAECs were seeded twice in the microchannels at a concentration of 1–2 × 106 cells/ml. First, 50 µl of cell suspension was added and incubated without flow for 1 h. The chip was then inverted, and another 50 µl of PAEC suspension was added, followed by overnight incubation at 37 °C to allow cell adhesion and monolayer formation. The endothelial cells reached confluence one day after cell seeding. The chip was subsequently connected to the peristaltic pump (Ismatec, Glattburg, Switzerland) and a reservoir using silicon tubing (MaagTechnic, Dübendorf, Switzerland). The culture medium was replaced with 4 ml of flow medium composed of DMEM-10 with 4% dextran and 1% BSA to reach a viscosity corresponding to 2.1 dyn/m2×seq at 37 °C. Laminar sheer stress was fixed at 2.0 dyn/cm2 using the peristaltic pump and kept for an additional 48 h at 37ºC, 5% CO2 before proceeding with 3D cytotoxicity assay. These flow conditions resemble those of arterial branches with a diameters ranging from 600 to 3000 μm33 and blood-like viscosity; however, the shear stress was approximately five times lower in order to avoid detachment of the endothelial cells from the channels.Live-cell imaging cytotoxicity assays in static 2D and 3D microfluidic systemPAECs’ nuclei were stained with 2 µg/ml Hoechst for 1 h in the incubator and washed with PBS once using an automatic dispenser BioTek EL406 (Agilent, Santa Clara, CA, USA). NK cells were labeled with cell trace yellow (CTY), 0.2 µl per million cells for 20 min in PBS 1× at 37ºC. Once washed, NK cells were placed in phenol-red free FluoroBrite DMEM-10 containing 1 µl/ml of Draq7, to evaluate cell viability, and added to target cells at a 1:1 E:T ratio.In the static 2D assays, NK cells seeded in the 96-well plates were centrifuged at 300×g for 1 min in order to reach the same microscope focal plane as PAECs. An automated spinning disc microscope (ImageXpress Micro Confocal High-Content Imaging System, 10× objective, Supplementary Table S3) was used to capture video time-lapse images of the interactions of NK cells with PAEC monolayers during 120 min at 5 min intervals. Three technical replicates with four regions per replicate were acquired. Time-lapse videos were created by MetaXpress. Analysis of dead and apoptotic cells was done using Fiji, ImageJ2 2.9.0 software (Supplementary Table S4). Microscopic image domains were split into segments to detect individual cells (cell segmentation) based on Hoechst staining signals and nuclei morphology. The percentage of total dead cells was determined according to the equation \(\:Total\:dead\:cells\:\left(\%\right)=\left(\frac{{{Live\:cells}_{\left(t0\right)}-Live\:cells}_{\left(t120\right)}}{{Live\:cells}_{\left(t0\right)}}\right)\times\:100\), where t0 and t120 correspond to the counting at time zero and at the endpoint (120 min), respectively. The counts of apoptotic cells included apoptotic bodies with intact membranes (intensified blue signals due to DNA condensation), and late apoptotic cells with Draq7+ signal. Thus, cell death through apoptosis was calculated by the equation \(\:Apoptotic\:cells\:\left(\%\right)=\left(\frac{{Apoptotic\:cells}_{{t}_{120}}}{{Live\:cells}_{{t}_{0}}}\right)\times\:100\). Lastly, necrotic cells were counted manually based on their distinct features consisting of normal nuclei shape with Draq7+ signals, but without intensified blue signals, and percentages were calculated as for apoptotic cells.In the 3D microfluidic system, after 48 h under flow conditions, the channels were checked under the microscope before adding NK cells for quality control. If the endothelial layer was not 100% confluent, this channel was not used for NK cell perfusion experiments. For both wild-type PAECs and human endothelial EA.hy926 cells, we did not observe any cell detachment before or after the perfusion for 48 h at 2 dyn/cm2 shear stress. Image acquisition was performed with a Leica DMi8 wide-field microscope at 10× magnification to capture the entire chip (1 cm) connected to the reservoir (Supplementary Table S3). 4 ml of flow medium FluoroBrite DMEM − 10 containing 2 × 106 CTY-labeled NK cells with 1 µl/ml Draq7 added were used per channel and injected into the 15 ml reservoir tubes. The peristaltic pump perfused the NK cell solution through the system for 2 h. Images were acquired at one focal plane – no z-stacks with the thickness of 130 μm from the 10× objective lens. The recording one area was of 0.1 cm length per channel continuously for 40 min. Images captured the adhesion of NK cells to the endothelial cell monolayer under flow, as well as all dead PAECs of the whole channel at t0 and t120 min for endpoint analysis. The number of attached NK cells and Draq7+ PAECs were quantified using Fiji, ImageJ2 2.9.0 software.Automatic detection and tracking of cell bright centersA Python program was developed for the automatic detection of NK cell bright centers and their trajectories. A series of 16-bit monochrome images were first loaded. A median filter with kernel size of one pixel was applied to remove single-pixel high-frequency noise. Instead of using traditional segmentation and centroid-finding methods, each individual image frame was processed to locate gaussian-like intensity distributions, which effectively mapped the location of each cell bright center. The “locate” module from the trackpy Python library was used to accomplish this task, which is an implementation of the Crocker-Grier centroid finding algorithm34. A set of parameters were experimentally defined to reach the optimum detection results. The initial central detection diameter (initial approximation) was defined as 31 pixels, the minimum integrated intensity under a single distribution was defined as 9,000, the minimum separation between bright centers was 15 pixels, and the gaussian blurring filter kernel size was three pixels. The coordinates (x and y position in the unit of pixels) from each detected cell bright center were stored for each image frame as tabular data. Coordinates were transformed to µm by using the calibration constant of 0.684 μm/pixel. Once the positions were determined for each frame, trajectories were built by using the “link” module from the trackpy Python library, which is an implementation of the Crocker-Grier linking algorithm34. Due to the high number of detected bright centers and the large range of motion, the linking algorithm was run on an adaptive search mode, which could adaptively change the size of the linking trajectory if needed, with the goal of accelerating convergence. A maximum search radius of 80 pixels and an adaptive stop value of 30 pixels were defined. In case of complex situations, the search radius could be automatically limited to the stop value, reducing the search possibilities and computing time. The linking process grouped each detected bright center and assigned an identifier in the form of a sequential number. The trajectory information was stored as tabular data, in such a way that spatial and temporal information could be extracted during the trajectory analysis stage.Analysis of NK cell trajectoriesTo assure a density of NK cells in the optical field suitable for analysis of cell trajectories in the 2D cytotoxicity assays, the E:T ratio was reduced to 1:20. The trajectories of single NK cells were tracked and classified as large, intermediate, and small based on the total effective displacement for each cell, also referred to as the trajectory diameter (µm). Trajectory diameter was defined as the distance between the position of the cell detected in the first and last frame (µm). Three classes were categorized based on the trajectory diameter (small < 10, intermediate 10–100 and large > 100 μm) for further analysis of trajectory length-to-diameter ratio, average speed (µm/min) and arrest coefficient. Arrest coefficient was defined as the percentage of time that the cell remained in arrest based on the threshold on instantaneous speed (0.2 μm/min)21. Tracking of NK cells was performed by analyzing every frame (frame rate: 4 min) of 120 min recording with a program built in Python. Data were analyzed from a heterogeneous population of at least 1,500 NK cells in nine different regions of three technical replicates.Immunofluorescence staining and imaging of 2D and 3D systemsFor immunofluorescence staining of the 2D system after the live-cell imaging session, cells were fixed in 4% buffered formaldehyde for 20 min at room temperature, followed by eight washes with DPBS, and blocked overnight at 4ºC using 3% BSA in DPBS. Next, the samples were incubated overnight at 4ºC with properly titrated antibodies (Supplementary Table S2), followed by secondary antibodies for 25 min at room temperature in the dark. An ImageXpress Confocal microscope was used for acquiring images at 10× and 20× (Supplementary Table S3).For the 3D microfluidics system, the channels were disconnected from the perfusion system, washed three times with DPBS and fixed with freshly prepared 4% formaldehyde, washed, and blocked. Antibody staining was performed following the same protocol as described above for the 2D system. Finally, the chip was washed and incubated for 15–30 min with Hoechst (1 µg/ml). Images were acquired using a confocal Leica SP8 microscope at 20×; 3D images were created by Imaris software (Supplementary Table S3, S4).Statistical analysisMeans of two groups were compared using paired and unpaired t-test (indicated in the figure legends). Error bars show ± SD. Differences were considered significant when p ≤ 0.05. The analysis was performed using Prism GraphPad.
