A 3D printed microfluidic perfusion device for multicellular spheroid cultures

All chemicals and reagents were purchased from Sigma-Aldrich Pte Ltd, Singapore unless otherwise stated.

CFD simulation was performed using Autodesk CFD (Autodesk, USA) using the available commercial packages. Simulations with different mesh sizes were iterated until the difference in the estimated flow rate between two successive mesh scales was less than 5% (table S1, figure S1 is available online at stacks.iop.org/BF/9/045005/mmedia) [29]. The viscosity of the medium was assumed to be the same as that of water (0.001 003 Pa s). Flow velocity and wall shear stress at the cell culture chambers were determined at hydrostatic pressure generated when the height difference between the inlet and outlet media reservoirs were set at 3, 6, 9 and 12 mm.

The device was designed as a two-part construct comprising of a top layer and the bottom mounting base. The top layer consisted of the cell culture chamber, perfusion and seeding channels and connecting Luer interfaces, while the bottom mounting base contained a viewing window for light microscopy. 3D models of the device were designed using SOLIDWORKS Standard (SolidWorks Corp., USA). The models were then exported in .stl format for the printing of the device. Both models were orientated such that the side containing the micro-scale features were printed along the x–y plane. Two commonly used modes of fabrication were explored in this study: PolyJet printing and SLA. For PolyJet printing, the device was printed using Objet260 Connex3 Printer (Stratasys, USA) using a biocompatible PolyJet photopolymer (VeroClear-RGD810) with a z-resolution of 18 μm. The PolyJet-printed parts were immersed in 0.2% NaOH solution for 1 h followed by sonication in DI water for 1 min to remove the support material (SUP705, iSQUARED2, Switzerland). The parts were then washed by copious amount of DI water. The whole printing process of the microfluidic devices took approximately 5 h. SLA-printed models were fabricated via a commercial printer (Proto Labs, USA) using 3D Systems Accura® 60 (polycarbonate (PC) mixture) with a z-resolution of 50.8 μm. The SLA-printed parts were then subjected to 2 h of UV curing at 60 °C.

GFP-labeled parental and tdTomato-labeled metastatic patient-derived OSCC cell lines were derived from patient-derived xenograft (PDX) models (patient-HN137) that were used to expand the original tumor material. The PDX tumors were dissociated using 4 mg ml⁻¹ Collagenase type IV (ThermoFisher Scientific, USA) in cell culture media containing Dulbecco's Modified Eagle's Medium (DMEM)/F12, at 37 °C. Cells were washed 3 times in phosphate buffered saline (PBS) (ThermoFisher Scientific, USA); strained through 70 μm cell strainers (Falcon, cat. no. 352350), and plated in RPMI (ThermoFisher Scientific, USA), 10% fetal bovine serum (FBS) (Biowest, France), and 1% penicillin-streptomycin (ThermoFisher Scientific, USA). Cells were incubated with 5% CO₂ at 37 °C. Primary cell colonies appear within 2–3 weeks, which were then serially passaged to expand the cell line. Primary cell lines were then transduced with lentivirus containing GFP or tdTomato-expressing plasmids (cloned into pLL3.7 and pLV vector, respectively) to generate 'red' or 'green' fluorescently labeled cell lines.

Both parental and metastatic HN137 cell lines were cultured and maintained in RPMI 1640 medium (Life Technologies, Singapore) supplemented with 10% FBS and Penicillin-Streptomycin (100 μg ml⁻¹). HepG2 cells (ATCC, USA) were maintained in DMEM high glucose medium (Life Technologies, Singapore) supplemented with 10% FBS and Penicillin-Streptomycin (100 μg ml⁻¹).

All cells used in this study were harvested as single cells with 0.125% Trypsin-EDTA solution (Life Technologies, Singapore) for 3 min at 37 °C, centrifuged at 1000 rpm for 5 min and resuspended in their respective culture media. To induce 3D spheroid formation, the harvested cells were seeded at fixed density of 1–3 × 10⁵ cells per well into Aggrewell400™ plates (StemCell Technologies, Canada), spun down at 60 g for 3 min and incubated for 24 h at 37 °C, 5% CO₂. After 24 h of culture in their respective culture medium, the spheroids were harvested and collected into a 15 ml tube for cell seeding.

The device was assembled by mounting the top part of the device onto a 170 μm thick PDMS membrane on top of a 1 mm thick glass slide placed into the viewing window of the mounting base, and securing the parts with M5 stainless steel screws. The PDMS membrane was made by casting uncured PDMS (Sylgard 184, Dow Corning, USA) between two aluminum plates which were lined with transparency sheets (SureMark, Singapore) and separated by 170 μm glass spacers. The PDMS sheets were cured at 70 °C for 4 h before demolding. All the inlet and the outlet of the devices were fitted with the 1-way stopcocks with Luer connections (Cole-Parmer, USA). Prior to the device assembly, all parts were sterilized by autoclaving at 121 °C for 20 min except for the stopcocks and the 3D printed parts, which were sterilized with 70% ethanol for 1 h. Prior to the seeding, the device was passivated with 0.2% bovine serum albumin (BSA) in PBS for 1 h at room temperature, and flushed with cell culture media using a syringe pump (KD Scientific, USA) at 0.3 ml hr⁻¹.

Cell seeding into the 3D printed microfluidic devices was initiated by withdrawing spheroids from the seeding inlet at a flow rate of 0.1–0.15 ml hr⁻¹. Throughout the seeding process, the perfusion inlet and outlet were kept closed. Once the cell culture chamber was filled with spheroids, the seeding inlet and outlet were closed, respectively. 3 ml medium dispensing reservoirs (Nordson EFD, USA) were attached to the perfusion inlet and outlet, and the device was then transferred into a sterile polypropylene container and placed into a 37 °C 5% CO₂ incubator before medium flow was initiated. Medium replenishment was performed every 24 h by manually removing medium from the outlet reservoir and topping up the inlet reservoir.

The labeling of viable and necrotic HepG2 cells was performed with LIVE/DEAD® Viability/Cytotoxicity Kit (ThermoFisher Scientific, USA). 2 μM Calcein AM and 4 μM EthD-1 was prepared in culture medium and perfused at 0.08 ml hr⁻¹ using a syringe pump for 1 h. The devices were then washed with culture medium at 0.08 ml hr⁻¹ for 30 min before fluorescence imaging.

Fluorescence images of GFP or dTomato-labeled HN137 OSCC spheroids and Calcein AM/EthD-1 labeled HepG2 spheroids were obtained using fluorescence microscope (Nikon, Japan) with CoolLED pE-2 LED-excitation light source (CoolLED, UK) at wavelength of 470 nm for the parental cell line and 565 nm for the metastatic cell line.

The Cytochrome P450 1A1 (CYP1A1) metabolic activity of HepG2 spheroids was determined by using the fluorescent 7-ethoxy-resorufin-O-deethylase assay. A working substrate solution containing 10 μM 7-ethoxy-resorufin and 80 μM Dicuramol in DMEM medium was perfused through the microfluidic device at 0.08 ml hr⁻¹ for 6 h inside a 37 °C, 5% CO₂ incubator. Subsequently, the device was perfused with fresh culture medium for 1 h before fluorescence imaging using at an excitation wavelength of 520 nm.

CYP3A4 activity of the HepG2 spheroids was measured using the Vivid^® CYP450 Blue Screening Kit (Thermo Fischer, USA). A working substrate solution prepared according to manufacturer's protocol was perfused through the microfluidic device at 0.08 ml hr⁻¹ for 6 h inside a 37 °C, 5% CO₂ incubator, before the device was flushed with fresh culture medium for 1 h. The fluorescent products were imaged using a fluorescence microscope at an excitation wavelength of 415 nm. The CYP1A1 and CYP3A4 activities in static 2D HepG2 cultures were determined by incubating the respective working substrate solutions described above for 6 h, washing with fresh culture medium before fluorescence imaging.

The 3D printed microfluidic perfusion device was designed with performing organotypic cell culture and biological assays in mind [4]. The device was built as two separate pieces. The top layer comprised of the cell culture chamber, perfusion channels as well as connecting Luer interfaces, which was aligned and clamped to a bottom mounting base (figures 1(A) and (B)). Instead of printing the device in a single build, the two-piece design of the 3D printed microfluidic device (figures 1(A) and (B)) will allow users to disassemble the device to retrieve tissue samples much more easily than a uni-body device. In addition, this design facilitated the removal of support materials used in PolyJet printing from the microfluidic network as compared to prototyping the device in a single build (data not shown), which also allows the 3D printed device to be reused provided that the repeating cleaning and sterilization process do not degrade the printed materials. Since common materials such as acrylonitrile butadiene styrene and PC used in 3D printing are not fully optically transparent [8, 10, 30], we also designed the mounting base to include a viewing window, where a microscope slide can be incorporated to allow for light and fluorescence microscopy to be performed on the biological samples. A thin PDMS membrane was used as a gasket to provide proper sealing of the microfluidic channels when the top and mounting base were assembled (figure 1(A)).

Zoom In Zoom Out Reset image size

The 3D printed device consisted of a duplex of perfusion culture units. Each unit had a cell culture chamber that was fed by an orthogonal pair of microfluidic channels for cell seeding and culture media perfusion, respectively (figure 1(D)). Cells introduced via the seeding inlet channel will be physically entrapped by a series of micro-structures arranged in a circular configuration in the central region of the cell culture chamber (figure 1(D)). The cell trapping mechanism relied on designing the gap size between the micro-structures to be smaller than the size of a single cell or cell aggregate, which has been widely reported by various groups [3, 25, 31]. After cell seeding, culture medium was perfused through a separate channel network placed orthogonally to the seeding channels. A pump-free perfusion system was employed to culture the cells immobilized in the microfluidic device as previously reported [29]. A constant medium flow rate was sustained by a pair of 3 ml media reservoirs, which were secured onto the printed inlets and outlets at a specified height difference and orientated horizontally (figure 1(C)). This setup allows the maintenance of the liquid meniscus in the inlet and outlet reservoirs at a given height difference to generate a constant hydrostatic pressure to drive medium flow across the microfluidic channel (figure 4(E)) [28, 29]. Hence, the entire 3D printed microfluidic perfusion culture device can be placed in a sterile secondary container inside a 37 °C, CO₂ incubator and operate as a standalone device independent of pumps and tubing.

The 3D printed device was sterilized by rinsing the channels with 70% ethanol solution for 1 h. We performed profiling studies on the 3D printed micro-structures before and after treatment with 70% ethanol. Since there is no significant no significant changes in the printed features, the ethanol-based sterilization process is employed to ensure the sterility of the microfluidic perfusion device (figure S2).

We investigated the suitability of SLA and PolyJet technologies, two of the most common modes of 3D printing [32–35], for printing microfluidic devices with micro-structures. The manufacturer specifications reported similar printing resolutions in the x–y (85 μm for PolyJet and 63.5 μm for SLA) and z direction (18 μm in layer thickness for PolyJet and 50.8 μm for SLA). To identify the fabrication limits of the two technologies, we first designed and printed a prototype consisting of 200 μm high square pillars with lateral (x–y) dimensions of 130, 250, 500 and 750 μm (figure 2(A)), and measured the printed structures to estimate the printing errors in x–y directions (figures 2(B), (C)). It was observed that in general, SLA can print structures with better fidelity to the designed features (less than 50% error), which was in concordance to findings by Shallen et al [32]. The difference in the printing fidelity of both modes of 3D printing was particularly apparent at smaller length scales. With SLA, we could print 130 μm pillars with an error rate of 43% ± 3% (figures 2(A), (B)), while PolyJet could not replicate a discernable structure at nominal lengths of 250 and 130 μm (figures 2(A) and (C)). As the feature length increased to 700 μm, PolyJet seemed to be able to generate structures that have similar shape but the surface roughness highly distorted the shape of the printed structure (figure 2(A)). This may be attributed to the fact that PolyJet printing deposits a sacrificial material in embedded or overhanging structures [35, 36], and interactions between the printed and sacrificial materials may have contributed to increased surface roughness of the side walls [37]. Moreover, incomplete removal of the sacrificial material may also increase the surface roughness. SLA, while being able to produce clean-cut structures, also had limitations in generating structures closely similar to the design. This was especially obvious at the range of 130–250 μm for the given SLA system used to prototype the device, where the corners of the micro-structures were rounded off (figure 2(A)). When we printed the designed microfluidic parts with both SLA and PolyJet, we observed similar results. SLA could print the 750 μm wide × 200 μm deep microfluidic channels and the circular cell culture chamber (750 μm in diameter) with better accuracies and smoother surfaces than PolyJet (figure S3). Hence, we selected the SLA-printed microfluidic devices for subsequent cell culture experiments.

Zoom In Zoom Out Reset image size

We observed that the printing error was not only dependent on the nominal dimensions of the structures but also their geometries. The nominal widths of the cell immobilization and guiding micro-structures were both designed to be at the range of 100–150 μm in order to be within the diffusional limit of oxygen to ensure sufficient mass transport to the spheroids [38]. Using the experimentally fitted curve generated from the printed prototype, we predicted the printing error to be approximately 68% and 57.7% at a nominal width of 100 μm and 150 μm, respectively (figure 2(B)). When the cell immobilization micro-structures were designed as an array of equilateral triangles with equal nominal length and width of 150 μm (μ-Struct 1) (figure 3(Ai)), the printing errors for the feature length and width were 51.7% ± 2.3% to 56.9% ± 3%, respectively (figure 3(C)). This was in agreement with the predicted error based on a square structure. However, when the cell immobilization micro-structure was designed as an array of arcs (μ-Struct 2), where the nominal length was increased to 820 μm while maintaining the nominal width to be 150 μm (figure 3(Aii)), we found that the printing error for both nominal dimensions was significantly decreased to 11.9% ± 1.1% (length) and 9.1% ± 3% (width) (figure 3(C)). Similar observations were made when comparing the different guiding structures, which were designed as parallelograms where the nominal lengths were larger than the nominal widths (figure 3(B)). The printing errors of μ-Struct 3 measured at 25.2% ± 2% (length) and at 27% ± 2.4% (width), while those of μ-Struct 4 measured at 5% ± 2% (length) and 12% ± 2% (width) (figure 3(C)). In both cases, although the printing errors of the feature length were slightly higher than predicted (25.2% at 400 μm and 5.1% at 820 μm), the printing errors of the smaller feature widths were significantly smaller than predicted. This observation showed that nominal dimensions alone cannot be used predict print fidelity.

Zoom In Zoom Out Reset image size

Flow induced shear stress has a significant impact on cellular phenotype and functions [39, 40]. Therefore, it is imperative that we ensure that the spheroids were not subjected to excessive shear stress during device operation. CFD simulation was performed to estimate the wall shear stresses for optimizing flow rates during spheroid seeding and perfusion culture. The 3D spheroids were withdrawn from a reservoir attached to the seeding inlet (figure 4(A)), where they will be entrapped by the cell immobilizing micro-structures in the culture chamber (figure 4(B)). At the designated seeding flow rates (0.1–0.15 ml hr⁻¹), the maximum walls shear stress was around 0.8 dynes cm⁻² (figure 4(B)), which was below reported deleterious shear stress magnitudes [39, 40]. Perfusion culture in the 3D printed device relied on gravitational-induced flow, which was dependent on the height difference between the horizontally-orientated media reservoirs attached to the perfusion inlet and outlet of the device (figure 4(C)). The horizontal orientation of media reservoirs allowed the liquid meniscus to maintain the fluid height difference between the perfusion inlet and outlet throughout the culture period, generating constant perfusion flow rate [28, 29]. We used CFD simulation to estimate the flow velocities (figure 4(E)) and corresponding shear stresses (figures 4(D), (F)) at the cell immobilization micro-structures as a function of height difference between the inlet and outlet media. At a height difference of 3 mm between the perfusion inlet and outlet, the flow velocity was estimated to be 15 μm s⁻¹, translating to a perfusion flow rate of 0.01 ml hr⁻¹. The corresponding shear stress at this flow rate was found to be 0.0025 Pa and 0.88 Pa at the cell compartment and perfusion compartment, respectively, which falls within the suitable range for hepatocyte culture [39, 40]. This condition was employed throughout this study for consistency.

Zoom In Zoom Out Reset image size

We selected the SLA-printed microfluidic device containing μ-Struct 2 and μ-Struct 4 to demonstrate its biological utility in culturing multicellular spheroids. First, patient-derived parental and metastatic OSCC tumor spheroids (figure 5(A)) were employed to investigate the remodeling and maintenance of the 3D tumor spheroids in the device. By tuning the spheroid size to be >100 μm in diameter (figures 5(B), (C)), the cell immobilizing micro-structure array can trap and retain both parental and metastatic OSCC tumor spheroids in the cell culture compartment with high efficiency (figures 5(D), (E)). Under continuous perfusion, it was shown that the pump-free 3D printed microfluidic device can sustain the viability of both parent and metastatic OSCC tumor spheroids for 48 h (figures 5(F) and (G)).

Zoom In Zoom Out Reset image size

Next, we investigated the capability of the 3D printed microfluidic device in maintaining hepatocyte spheroid viability and functionality by incorporating preformed HepG2 spheroids (figure 6(A)) into the device. Similar to the OSCC tumor spheroids, the HepG2 spheroids can be efficiently trapped by the 3D printed micro-structures when the spheroid size was configured to be >100 μm in diameter (figure 6(B)). After perfusion culture, we observed that the HepG2 spheroids were able to retain their 3D morphology (figure 6(C)) as well as a high degree of viability as indicated by live-dead staining (figure 6(D)). We further assessed the Cytochrome P540 metabolic functions of the HepG2 spheroids after 72 h of perfusion culture as compared to conventional 2D cultures. The HepG2 spheroids in the 3D printed microfluidic device exhibited positive CYP1A1 and CYP3A4 metabolic activities as indicated by the generation of fluorogenic metabolites (figures 6(E), (F)). Both the CYP1A1 and CYP3A4 metabolic activities of HepG2 spheroids cultured in the 3D printed device were significantly higher (Student's t-test, p < 0.05) than those of static 2D controls (figures 6(G), (H)), which were consistent with previous reports that 3D perfusion cultures enhances the liver-specific functions of hepatocytes [41].

Zoom In Zoom Out Reset image size