|Home | About | Journals | Submit | Contact Us | Français|
In vitro 3D cancer models that provide a more accurate representation of disease in vivo are urgently needed to improve our understanding of cancer pathology and to develop better cancer therapies. However, development of 3D models that are based on manual ejection of cells from micropipettes suffer from inherent limitations such as poor control over cell density, limited repeatability, low throughput, and, in the case of coculture models, lack of reproducible control over spatial distance between cell types (e.g., cancer and stromal cells). In this study, we build on a recently introduced 3D model in which human ovarian cancer (OVCAR-5) cells overlaid on Matrigel™ spontaneously form multicellular acini. We introduce a high-throughput automated cell printing system to bioprint a 3D coculture model using cancer cells and normal fibroblasts micropatterned on Matrigel™. Two cell types were patterned within a spatially controlled microenvironment (e.g., cell density, cell-cell distance) in a high-throughput and reproducible manner; both cell types remained viable during printing and continued to proliferate following patterning. This approach enables the miniaturization of an established macro-scale 3D culture model and would allow systematic investigation into the multiple unknown regulatory feedback mechanisms between tumor and stromal cells and provide a tool for high-throughput drug screening.
The majority of ovarian cancer patients are diagnosed after the disease has metastasized to local and distant sites . The survival rate remains low (approx. 30%) mostly due to residual disease, which does not fully respond to the consecutive chemotherapy treatment . It is clear that new treatments are urgently needed to improve the clinical management of this deadly disease. Inadeuate progress in this area is partly due to poor predictive capability of traditional cell culture models for evaluation of therapeutic efficacy. This situation points to the need for in vitro cancer models that more accurately emulate human disease in vivo for high-throughput assessment of new therapies.A variety of models for in vitro cancer growth and treatment have made significant advances toward mimicking in vivo tumor architecture and growth behavior as compared to cells grown on flat 2D tissue culture substrates.
In vivo, tumors are complex tissues composed of, in the case of carcinomas, stromal cells such as fibroblasts [3, 4] and endothelial cells [5, 6]. Fibroblasts are among the most important stromal signaling partners found within the various forms of human carcinomas . Fibroblasts are associated with cancer cells at all stages of cancer progression, and have been linked to different tumor growth and metastasis activities. For example, fibroblasts could play an important role in promoting ovarian cancer growth and progression . Fibroblasts are thus a key determinant in the malignant progression of cancer and represent an important target for cancer therapies . Therefore, platforms to study the effect of stromal cells on cancer are urgently needed.
It has been traditionally difficult to pattern cocultures of cells. The manual methods do not provide control over cell position within a biologically relevant scale, which ranges from tens to hundreds of micrometers. The newer approaches to create such microcontrol over patterning of multiple cell types include stencils  and microelectro-mechanical system (MEMS)  approaches. Approaches based on stencils are limited by fabrication challenges . In addition, it is difficult to control the adherence and response of different cell types based on stencil material properties. Some of the MEMS approaches use clickwise manual attachment of gears on the stencils to control cell-to-cell distance . However, this platform requires microfabrication techniques such as deep reactive ion etching (DRIE) or chemical etching on silicon, which are expensive and suffer from low cell adhesion and imaging problems since silicon is not transparent. Such techniques are also low throughput since pieces have to be manually assembled. Here, we introduce a simple method that only involves the patterning of multiple cell types to predetermined locations.
In the present study, we utilized an innovative cell printing approach that overcomes inherent limitations of manually ejecting cells via a pipette, which enabled us to pattern ovarian cancer cells and normal fibroblasts at fixed spatial separations on a Matrigel substrate. In the 3D patterned overlay coculture model developed here, based on a previously described manually pipetted model, cells spontaneously form 3D acinar structures after adhering to the Matrigel surface [13, 14].
Currently, cancer biology experiments mostly use pipette-patterned large gel structures, where a few thousands of cells are plated in tens of microliters of gels with a low patterning precision capability of a few millimeters. The low control precision in placement of these distinct cell types has restricted their use in systematic investigations of such crucial tumor-stroma interactions where resolution of tens of microns in the placement of distinct populations is needed. To address these challenges with manual pipette ejection of cells, possible technologies include cell printing [15–21], negative dielectrophoresis  and microfluidic patterning technologies [23, 24].
The focus of this paper is the creation of well-patterned biological acini structures using a printing method. This work addresses unique challenges in the sense that the cells have to be both viable and form 3D functional acini units. As an in vitro model of ovarian cancer, we demonstrate a cell patterning platform that prints two types of cells onto Matrigel, where the acini formed using this platform post-pattering recapitulate the morphology and growth kinetics previously reported for cells deposited by manual pipetting.We present the technique and its biological proof of concept; it represents an elegant approach for miniaturization of an established macro-scale 3D culture model. The novelty of this work comes from: (1) the combination of state-of-the-art cell biopatterning with cancer biology to build in vitro 3D ovarian cancer models at high throughput that are capable of recreating distinct in vitro cancer models under well-defined and reproducible conditions; and (2) the potential of the technology as a tool for delving into coculture interaction between cancer and stromal cells that may be otherwise overshadowed by manual handling and culturing methods.
NIH: OVCAR-5 cells (an epithelial human ovarian cancer cell line) were obtained from Thomas Hamilton (Fox Chase Cancer Institute, Philadelphia, PA). MRC-5 cells (normal human fibroblast cell line) were obtained from ATCC. Both cells lines were cultured in a CO2 water-jacketed incubator at 37°C, 5% CO2 (Forma Scientific, Model 3110) and passaged under sterile conditions. The cell media were prepared by mixing 500 mL 1× RPMI (RPMI 1640 with L-glutamine, Cellgro, 10-040-CV) for OVCAR-5 or 1× MEM(MEM with Earle’s Salts and L-glutamine, Cellgro, 10-010-CV) for MRC-5 with 50 mL fetal bovine serum (FBS, Gibco, 10439-024) and 5 mL 1% penicillin-streptomycin (pen/strep, Sigma, P4333) through a sterile filter (500 mL Express Plus 0.22-μm membrane, Millipore, 5179-SCGPU05RE). After confluence, the cells were collected through trypsinization and centrifugation. Initial cell concentration of cell suspensions was calculated using 0.4% Trypan blue solution (for use with the Countess” automated cell counter, Invitrogen, T10282) and a hemacytometer (Hausser Scientific, 1483). The required cell concentrations (1 × 106, 2 × 106, 5 × 106 and10 × 106 cells/mL) were prepared by diluting with the cell medium.
A bed of growth factor-reduced (GFR) Matrigel™ (BD, 354230) was prepared prior to cell printing on each culture dish (35 mm in diameter, MatTek Corporation, P35G-1.0-20-C). GFR Matrigel™ was thawed overnight at 4°C on ice and was kept cool on ice before use. GFR Matrigel™ (150 μL) was added to the central glass bottom portion of each dish to produce a ~250-μm-thick basement membrane for overlay cell printing. All pipettes, tips, and tubes were pre-cooled on ice to prevent premature gelation of Matrigel™.
We have already developed and described a cell biopatterning system [25, 26]. Briefly, the system consists of an automated micrometer-resolution xyz stage (Precision Linear Stage, Newmark systems, NLS4) controlled by a stage controller (Newmark Systems, DMC-21×3) and nanoliter dispensing valves (Solenoid valve ejector, TechElan, G100-150300NJ) controlled by a pulse generator (Hewlett Packard, 8112A) (Fig. 1). The entire setup is enclosed within a sterile hood allowing long-term cultures of up to 3 weeks for patterned ovarian cancer model constructs. Here, two ejectors were used, one for ejecting OVCAR-5 and the other for ejecting MRC-5. The ejectors were connected to pressured nitrogen gas through a syringe reservoir.We used a valve ejector with a wide nozzle (with nozzle diameter 150 μm) to minimize local shear force created within the droplet during generation. This offers high post-ejecting cell viability. The stage and dispensing subsystems were synchronized and programmed. The cell suspensions of OVCAR-5 and MRC-5 were pipetted into two 10-mL syringe reservoirs, respectively.The valve opening duration and gas pressure were regulated to control the droplet size. Individual cell-encapsulating droplets were patterned at pre-determined positions on the GFR Matrigel™-coated glass-bottom culture dish. The dishes with patterned cells were then cultured with 2% GFR Matrigel™ media. Before and after cell patterning, the ejector was washed out with 70% ethanol and deionized water to sterilize the ejector.
The size of each droplet was determined by ejecting droplets into a petri dish (60 × 15 mm) filled with liquid nitrogen. The frozen droplets were imaged using a microscope (Eclipse TE-2000 U, Nikon). Droplet diameters were obtained by fitting circles around each droplet image using ImageJ (NIH, Bethesda, MD).To assess the number of cells in individual ejected droplets, we used constant valve opening duration (60 μs) and gas (nitrogen) pressure (34.5 kPa) for constant droplet size. The valve opening time and the gas pressure determine the size of the generated droplets and how fast the droplets are ejected [25–27]. We then stained the ejected cells with DAPI (dilactate, Sigma-Aldrich, D9564) for counting the number of cells in individual droplets. Both bright-field and florescent images of the ejected cell-encapsulating droplets were viewed under the microscope (Nikon TE2000) immediately after printing. The numbers of cells were counted using ImageJ.
Cell viability was assessed using a live/dead viability staining kit (Live/Dead Viability/Cytotoxicity Kit for mammalian cells, Invitrogen, L3224). Pre-ejection cell viability was measured in samples taken directly from the cell solutions as a control. Post-ejection cell viability was measured from ejected droplets at both 4 h and 3 days post printing to investigate the effects of cell ejection process and different cell patterns. Basically, the dishes containing the patterned cells were washed with PBS, stained for 10 min at 37°C with live/dead staining solution and then washed with PBS again prior to imaging under a fluorescent microscope (Nikon Eclipse TE-2000 U).
The patterned cells were monitored for 15 days using an inverted microscope (Zeiss Axiovert at the Wellman Center for Photomedicine) to acquire longitudinal dark-field microscopy images. Image data were processed at high throughput using custom MATLAB scripts (Mathworks, Natick, MA, USA) as reported earlier . Basically, the images were first thresholded, made binary and then segmented to identify individual acini, which were then used to calculate size distributions and size change with time in temporally sequenced directories of dark-field image data.
To confirm the 3D acinar structure formed by patterned cancer cells, two-photon imaging (Olympus FV1000 MPE at the Wellman Center for Photomedicine) of endogenous fluorescent species using 750 nm excitation as previously described  was used to image printed 3D acini.
We first characterized the spatial patterning precision through investigating individual droplet placement and inter-droplet distance. The droplet deposition variation was 4.9 μm and 18 μm in the distal and proximal directions, respectively, as reported in our earlier study . The difference between programmed distance (Dprogram) and actual printed distance of droplets after patterning (Dactual) was <3.5% (Fig. 2a).The droplet ejection directionality determined this patterning variation. To measure the droplet size, droplets were ejected into liquid nitrogen (LN2) using the method previously described  (see Supporting information, Fig. S1).The droplet diameter in LN2 was measured under a microscope and the average size was 510 ± 26 μm (mean ± SD, n=51) as shown in Fig. 2b. When the droplets were ejected onto a substrate (i.e., Matrigel™), they spread out after landing (Fig. S1a).
Control over the number of cells in droplets can be achieved by changing the droplet size or the initial loading cell concentration. In this study, we used constant droplet size and only adjusted the cell concentration in the cell/medium mixture. Four different cell concentrations were used (1 × 106, 2 × 106, 5 × 106 and 10 × 106 cells/mL).The relationships between the number of cells per droplet and the cell concentration for OVCAR-5 and MRC-5 are shown in Fig. 2c, which clearly shows that the number of cells per droplet increases with increasing cell loading concentration. OVCAR-5 and MRC-5 were stained and compared under bright-field and UV light (see Supplementary Fig. S2). Mean ± SD at each concentration were 9 ± 1, 23 ± 3, 52 ± 2, 92 ± 5 cells/droplet (n=10) at 1 × 106, 2 × 106, 5 × 106 and 10 × 106 cells/mL for OVCAR-5, and 9 ± 1, 19 ± 2, 48 ± 7.3, 115 ± 10 cells/droplet (n=10) at 1 × 106, 2 × 106, 5 × 106 and 10 × 106 cells/mL for MRC-5.We chose a concentration of 52 ± 2 cells/droplet for acini growth experiments in culture.This cell concentration gave an average cell-to-cell distance of ~50 μm within the droplet and there is small variation (± 4%) of the number of encapsulated cells between two different cell types. These results prove that we can effectively control the initial cell density in the created constructs. Each cell-encapsulating droplet has on average the same number of cells.This is important because the number of cells per droplet determines the average cell-to-cell distance. Cell viability was assessed for three different cases, i.e., OVCAR-5 only, MRC-5 only, and cocultures of OVCAR5 and MRC-5 (Figs. 2d-f).The average cell viability relative to the culture control were 100%, 96.2%, 100%, 100% for only OVCAR-5, MRC-5 only, OVCAR5 in coculture and MRC-5 in coculture at 4 h post patterning, indicating that the ejection process did not have a detrimental effect on the cell viability. At 72 h post patterning, the coculture of patterned cancer cells and fibroblasts did not show any dead cells, whereas the OVCAR5 and MRC-5 viabilities were 93.8% and 90.1%, respectively.
It takes less than 100 μs (60 μs valve opening) to generate one droplet. With our current platform (two ejectors), we were able to generate 20 000 droplets/min for each cell type. With on average 100 cells per droplet, this platform can pattern 2 million OVCAR-5 cells and fibroblasts per minute.
We compared the growth and development of our patterned 3D cultures with previously characterized growth properties of ovarian 3D acini formed from manually pipetted cells [13, 14]. Two-photon images show representative 3D acinar structure 7 days post patterning (Fig. 3). This 3D structure qualitatively recapitulates the micronodular feature of ovarian cancer that was observed in pipetted models [13, 14] and in vivo [2, 29–31].
We also characterized the kinetics of acini growth by tracking the patterned cancer cells for up to 15 days after patterning. The size (calculated as the cross-sectional area from the 2D microscopy image) and number of the 3D acini at multiple time points were quantified (Figs. 4a–c) using a previously described custom-developed MATLAB program . The size distribution of the 3D acini changes significantly with culture time (Fig. 4a and Table 1). We observed that the distribution of acinar sizes started from a small range (~100–500 μm2 at day 1), but the range consistently broadened with culture time, as indicated by the increased number of acini with larger size (days 5, 9, 15) (Fig. 4a).This may be due to the combined effect of cell proliferation and acini fusion via migration and coalescence [13, 14]. This heterogeneity in acinar size is consistent with that observed in manually pipetted models [13, 14] and in vivo [30, 31].We also observed a significant fraction in the 200–400 μm2 size range at all time points, and another fraction that developed more rapidly into larger structures over time. This overall growth pattern is consistent with that previously reported for 3D growth resulting from manually pipetted cells [13, 14]. Figure 4b show the change in number of acini as a function of number of initial cells per droplet. More acini were obtained with higher initial number of cells per droplet. However, the number decreases continuously with culture time and gets to a stable state after day 9 independent of the initial cell concentration.These observations agree with the results previously reported for 3-D acini formed from manually pipetted cells [13, 14].The average size of ovarian cancer acini increased exponentially with culture time (Fig. 4c). The ability to obtain uniformsized acini is of great importance for cancer study, e.g., the drug response of different sized acini. Using our current method, multiple acini were formed in individual printed droplet.To decrease the number of acini per droplet, one approach could be to decrease the droplet size or the cell concentration in the ejection reservoir. Another approach might be to use non-adhesive microwell arrays, which have been utilized to form uniform-sized embryoid bodies (EB) [32–35].
In this study, we micropatterned ovarian cancer cells (OVCAR-5) and fibroblasts (MRC-5) with spatial control. We characterized the biopatterned OVCAR-5 and MRC-5 for the number of cells per droplet, droplet size and cell viability. We also investigated printed acini growth kinetics such as change in acini size and number over time.The results show that both OVCAR-5 and MRC-5 can be ejected with controlled number of cells per droplet maintaining high viability. Microprinted OVCAR-5 remained viable and proliferated in Matrigel forming 3D acinar structures. The acinar growth kinetics in the patterned model resemble that of 3D acini formed from cells originally ejected by manual pipetting, and similarly recapitulate features of ovarian cancer micronodules in vivo.
The developed model system with spatial patterning of cell types and the initial cell density control in patterned constructs would enable physiologically relevant ovarian cancer coculture models to be created for a better understanding of ovarian cancer biology and improved clinical therapies. As a future application, such a platform could be used to build in vitro disease models in which various cell types are required to be placed with precise spatial control. This would allow systematic investigation of the many unknown regulatory feedback mechanisms between cells in a well-defined 3D environment, e.g. tumor and stromal cells for cancer. In addition, the model constructs fabricated at high throughput using this platform could be used in high-throughput screening of drug and treatment responses for reliable statistical analysis,reducing the testing costs and supporting alternative physiological models to animal testing.
U.D. would like to thank the Randolph Hearst Foundation and the Department of Medicine, Brigham and Women’s Hospital for the Young Investigators in Medicine Award.We thank Jie (Jenny) Zhao, Director of the Photopathology core of the Wellman Center for Photomedicine for her assistance with multiphoton fluorescence imaging. I.R. gratefully acknowledges support from the Wellman Center for Photomedicine in the form of a Wellman Fellowship. We thank Drs. Conor Evans and Adnan Abu-Yousif of the Wellman Center for several useful discussions of this work. This work was performed at both the Bio-Acoustic MEMS in Medicine (BAMM) Laboratories at the Center for Bioengineering, Brigham and Women’s Hospital and Harvard Medical School and at the Wellman Center for Photomedicine at the Massachusetts General Hospital and Harvard Medical School. Funding was provided by the National Institutes of Health, 5R01CA119388-03 (to T.H.), P01CA084203-06 (to T.H.) and R21 (EB007707). UD and FX were partially supported by the Center for Integration of Medicine and Innovative Technology (CIMIT) under US Army Medical Research Acquisition Activity Co-operative Agreement – New Development Grant.
The authors have declared no conflict of interest.