Design of the chip
The chip consisted of two layers (Fig. 2a). The lower layer was a perfusion chamber for the nutrient solution. The upper layer consisted of a cell chamber that contained plant cells. A porous membrane separated both layers (Fig. 2c). This allowed keeping the cells in place and enabling nutrient supply and mass exchange by the flow of nutrient solution. This set-up enabled the modular combination of several chips with a common metabolic perfusion flow. In principle, plant cells of the first cell chamber can secrete metabolites to the perfusion flow, which reach the plant cells in the next modules. Thus, signals and metabolites can migrate from module to module. Insertion and removal of cell was possible through two openings with a diameter of 5 mm in the cell chamber, whereas the perfusion chamber had one inlet and one outlet to perfuse with nutrient solution (Fig. 2b).
The modular and user-friendly set-up of the chips was possible through tube fittings N210-9 (Value Plastics dba Nordson MEDICAL, USA) with an inner diameter of 1.6 mm inner diameter as fluidic connectors as described elsewhere (Finkbeiner et al. 2019).
The dimensions of the chip corresponded to a standard microscope slide (25 mm × 75 mm), and a cell-chamber height of 1 mm. These dimensions provided short diffusion paths of the nutrients or products but sufficient space for the cells that are around 40–50 µm in height, at an average length of 250 µm (Kreppenhofer 2013; Maisch and Nick 2007). The volume was set to 800 µl offering direct comparison with experiments in 48-well titre plates. To avoid adhesion of air bubbles in corners of the cell chamber, the design was elliptical with a large area to facilitate the simultaneous observation of many cells. The total height of the chip was 4.2 mm to provide sufficient working distance for microscopic observation.
To achieve the combination of high transparency, biocompatibility, and hardness, we used the thermoplastic polymer polycarbonate (PC) Makrolon® GP clear 099 (Bayer Material Science AG, Germany) as material for the chip housing (Domininghaus 2012). The membrane had to be transparent and made up of pores small enough to retain the cells in the cell chamber. Additionally, it had to be thin as well and sufficiently porous to sustain short diffusion paths for nutrients and products. Therefore, we chose a polyethylene terephthalate (PET) membrane PET5020030 (Sterlitech Corporation, USA) with a thickness of 19 µm and a pore diameter of 5.0 µm.
Computational fluid dynamics
To validate the suitability of the membrane, we modelled the mass transfer of sucrose as nutrient from the perfusion chamber through the porous membrane into the cell chamber using a simulation software (COMSOL Multiphysics®, COMSOL Inc., Sweden). We simulated the membrane as porous filter (Chung et al. 2014) rather than as a porous medium (Loskill et al. 2017; Vereshchagina et al. 2013). Due to the long computation times, we restricted the simulation to 2D using the module ‘transport of diluted species’. Furthermore, we assumed a ‘laminar flow’ model with a Reynolds number of 2.37 × 10−7 deriving from a medium flow velocity of 2.84 × 10−4 m s−1, a channel height of 0.50 mm, a canal width of 2.35 mm, and a flow rate of 20 \(\upmu \mathrm{l}{\cdot \min}^{-1}\) .
The geometry was integrated into the model (COMSOL Multiphysics®) which allowed calculating the pore distance Lp–p according to Chung et al. (2014) as
$${L}_{p-p}=\sqrt{\frac{{\pi r}^{2}}{\mathrm{sin}\left(60\right)\cdot \varepsilon }}$$
(1)
with a membrane porosity ε that depends on the number of pores nPores. We could determine pore number by counting a membrane area of 230 μm × 143 µm (= 3.29 × 10−2 mm2) using SEM micrographs arriving at an average of 116.7 pores was received, leading to a pore distance of Lp–p = 18.09 µm under the consideration of vertical cylindrical pores compared to Lp–p = 16.99 µm specified by the manufacturer.
To calculate the diffusion coefficient of sucrose in water, we used the Stokes–Einstein equation:
$${D}_{\mathrm{Sucrose}}=\frac{{k}_{\mathrm{B}}{T}_{0}}{6\pi {\eta }_{\mathrm{Sucrose},\mathrm{ Water}}{R}_{\mathrm{Sucrose}}}$$
(2)
with a value of 5.5 × 10−10 m as molecular radius of sucrose (Kashima and Imai 2017). The viscosity ηSucrose, Water of a 30 g l−1 sucrose-water solution was determined in triple measurements with a rheometer (RheoStress 300, ThermoHaake®, Germany) with a conical titanium plate (60 mm diameter, 1°, 52 μm gap) leading to a value for ηSucrose, Water of 1.0277 mPa s. Since the Boltzmann constant is kB = 1.38 × 10−23 \(\mathrm{J}\cdot{\mathrm{K}^{-1}}\), we get for room temperature (T0 = 294.15 K) an estimated diffusion coefficient DSucrose = 3.81 × 10−10 m2 s−1, which is in accordance with the published record (Kashima and Imai 2017; Ueadaira and Uedaira 1969). Finally, we calculated the concentration maps using a fine two-dimensional grid by means of the backward differentiation formula solver at time intervals of 0.5 s over a total calculation interval of 250 s.
Fabrication of the chip parts
We fabricated the chip from semi-finished polycarbonate plates of 3 mm thickness by hot embossing with a brass moulding tool (i-sys Mikro- und Feinwerktechnik GmbH, Germany) for fast and reliable replication of the chip parts (Maisch et al. 2016; Worgull 2009). To facilitate the demoulding process, we adjusted a demoulding angle of 3° at all vertical edges in the brass tool and used a polished chrome counter plate to ensure transparency of the chip parts. The membrane and the woven fabric were prepared using a laser cutter VLS3.50 (Universal Laser Systems, USA).
Assembly technology
We assembled the microfluidic bioreactor by ultrasonic welding (PS DIALOG digital control, Herrmann Ultraschalltechnik GmbH & Co. KG, Germany) in a two-step process. This procedure allows integrating the tube fittings and the membrane in plane, reducing total chip height to 4.2 mm.
As first step, we welded the membrane directly on the chip housing with the cell chamber by means of the woven PET fabric (Fig. 3a). We generated a defined first welding seam, using a structured titanium horn (Herrmann Ultraschalltechnik GmbH & Co. KG, Germany) of 400 µm in height with a rounded energy director (ED), and a self-transformation of 1:1.7, in combination with a 1:1.5 booster. Due to the circular cross section of the fabric fibres, the point contacts of the fabric on the polymer membrane acted like additional miniaturised EDs. This procedure enables ultrasonic welding of very thin polymer sheets, because weldability decreases with decreasing thickness (Potente 2004; Rotheiser 2009). Furthermore, it allows for more homogeneous and gentle welding of the membrane. To avoid damage from fast oscillations of the thin layer, we used a comparatively low force of 200 N and a total amplitude of 13.3 µm to weld membrane and mesh (Grewell et al. 2003).
In the second step, we welded the perfusion chamber onto the membrane-sealed cell chamber (Fig. 3b, c), which sealed the whole system to the outside. In the same step, we integrated the tube fittings as described elsewhere (Finkbeiner et al. 2019). To avoid further tension on the membrane, the second welding seam was placed with an offset of 0.28 mm and an overlap of 0.07 mm in relation to the first welding seam. Height and width of the ED were 400 µm with a 200-µm radius edge rounding and was surrounded by drainages on both sides for excess melt. For this step, we used a flat horn (Herrmann Ultraschalltechnik GmbH & Co. KG, Germany) with a self-transformation of 1:1.5 in combination with a 1:2.5 booster resulting in a total amplitude of 21.9 µm. The clamping force was 500 N to compensate the slope due to the tube fittings, and the welding force was larger than in the first step (300 N). We verified the welding seam of the first welding step by SEM (SEM Supra 60VP, Zeiss, Germany) and, in addition, analysed membrane planarity with a digital stereo microscope (KEYENCE VHX-6000, KEYENCE Corporation, USA).
Quality control of the fabricated chip
To examine the tightness of the chips, we connected the tube fittings of the perfusion chamber to compressed nitrogen, while the cell chamber openings were closed using 3 M Polyester Film Tape 851 (3 M Deutschland GmbH, Germany). Subsequently, we immersed the whole chip in a beaker with water and tested for tightness as described elsewhere (Finkbeiner et al. 2019). In addition, we recorded pressure over flow rates of individual chips, or of two chips connected in series to validate a proper and leakage-free operation. For this purpose, we connected the chips to a syringe pump PHD ULTRA™ (Harvard Apparatus, USA) with a syringe containing water, steadily increasing flow rate to 30 \(\mathrm{ml}{\cdot \min}^{-1}\) and visually checking for potential leakage of the chips.
Use of the microfluidic chip to detect quorum sensing in BY-2
As a proof of concept for an interaction between two physiologically different cell populations, we used quorum sensing in BY-2. In suspension, the cells stop proliferating at high dilutions but resume proliferation in response to cell-free medium from proliferating cells growing at sufficient density. To simulate this phenomenon technically, we coupled two chips into a microfluidic series, using stationary cells at day 7 after subcultivation. Two serially coupled chips were combined with cells in different dilutions (1:300 or 1:30 of a stationary culture) in different combinations (Fig. 6): Either donor chamber and recipient chamber at 1:300, both at 1:30, or the donor chamber at 1:30 and recipient chamber at 1:300. The set-up was circular, with a reservoir flask, feeding the two serial chips with MS medium and collecting the liquid again, after it had passed through the chip. The total volume was 30 ml, such that in steady state, the effective solution of the cells in the chamber would be multiplied by a factor of 30 ml (total volume) by 2 × 0.8 ml (the loaded volume in the two chambers). In the 1:30 set-up, this would correspond to a 560-fold dilution of a stationary culture, in the 1:300 set-up, accordingly 5600-fold. The cell cultures were cultivated using a modified Murashige-Skoog (Duchefa Biochemie, Netherlands) medium for 3 days, such that the cells could enter the proliferation phase (Huang et al. 2017). Four days later, at the end of the culture cycle, we removed the cells from the chip and determined their density in a haemocytometer (Paul Marienfeld GmbH & Co. KG, Germany). Data represent mean and standard error from three independent experimental series, with 5 technical replications per experiment. If the dense donor culture would produce the quorum-sensing factor, this should be manifested as a re-initiated proliferation of the diluted recipient cells. We connected the two chips by Tygon® tubing with an inner diameter of 1.6 mm (Reichelt Chemietechnik GmbH & Co., Germany) and perfused with culture medium at a flow rate of 20 \(\upmu \mathrm{l}{\cdot \min}^{-1}\) in a circular flow (to facilitate accumulation of the quorum-sensing factor) driven by a 4-channel Ismatec REGLO Digital peristaltic pump (Cole-Parmer GmbH, Germany).
Use of the microfluidic chip to generate metabolic synergy in Catharanthus roseus
We cultivated suspension cell strains of Catharanthus roseus (L.) G. Don (C1 and C4) originating from seeds of Catharanthus roseus plants in fresh and autoclaved growth medium containing Gamborg B5 salts (3.21 g l−1), sucrose (30 g l−1) and 2,4-D (5 mM), adjusted to pH 5.6. We subcultured cells weekly, by inoculating 3 g (fresh weight) of filtered cells into the growth medium (50 ml) in 250 ml polycarbonate Erlenmeyer flasks with filter caps (Corning GmbH, Kaiserslautern, Germany). The cells were incubated at 26 °C in the dark on a gyratory platform shaker (Heidolph Instruments GmbH, Germany) at 120 rpm. A volume of 800 µl suspension for each strain was loaded to a chip and combined in a circular flow at a flow rate of 30 \(\upmu \mathrm{l}{\cdot \min}^{-1}\) as described above after elicitation with 100 MeJA (Duchefa Biochemie (Haarlem, Netherlands). After 2 weeks, the supernatant was frozen at − 20 °C and lyophilised for 3 days. After dissolving the lyophilisate with 1 ml of MetOH and ultrasonication for 2 min (amplitude 100%, 0.5 s pulse) using a high-efficiency ultrasound device (UP 100H, Hielscher Ultrasonics GmbH, Teltow, Germany). We removed all particulate matter by spinning the samples down for 10 min with 10,000 × g at 25 °C and filtering the supernatant through a 0.45-μm needle-type Chromafil PET-20/15 MS filter (Macherey–Nagel GmbH & Co. KG, Düren, Germany) into the autosampler vials (WIC4200, WICOM Germany GmbH, Heppenheim, Germany). Individual stock solutions of the alkaloid standards such as catharanthine, tabersonine, vindoline, vinblastine and vincristine were prepared at a concentration of 1 mg/ml in MetOH. These stock solutions and the alkaloid extracts were stored at − 20 °C for further analysis. For sensitive qualitative analysis by liquid chromatography-mass spectrometry (HPLC–DAD-ESI–MS/MS) of vinca alkaloids, we used a LXQ Linear Ion Trap MSn system (Thermo Fisher Scientific, Waltham, MA, USA) equipped with a Finnigan Surveyor HPLC–PDA. The extracts were separated on a Phenomenex Luna C18 column (4.6 mm × 250 mm, 5 μM particle size) with a gradient of 10 mM ammonium acetate, pH 6.0 (solvent A) and LC-grade MetOH (solvent B) as mobile phase using a flow rate of 500 \(\upmu \mathrm{l}{\cdot \min}^{-1}\) . The eluent profile (% of solvent A/% of solvent B) was 0–5 min using a linear gradient from 30:70 to 10:90 and 5–23 min with an elution gradient from 10:90 to 30:70. We detected masses using an ion trap mass spectrometer coupled with electrospray ionisation, operating in a positive mode at a spray voltage of 4 kV, a capillary voltage of 33 V, a capillary temperature of 350 °C and a tube-lens voltage to 70 V. The full mass scan covered the range from m/z 100 to 1000.
Use of the microfluidic chip to detect secreted fungal phytotoxins
To test whether the chip allows detecting the interaction between cells from different species through soluble factors, we combined tobacco BY-2 cells and the wood-decaying fungus Neofusicoccum parvum. To create a suspension culture of this fungus and to stimulate the release of phytotoxins, we transferred a plug of fungal mycelia (growing on potato dextrose agar) into liquid MS medium and maintained the suspension at biweekly intervals. The two life forms were growing in two separate microfluidic chips, connected with Tygon S3™ tubing (Saint-Gobain Performance Plastics, France). For perfusion, we used the standard BY-2 cultivation medium (Huang et al. 2017) at a flow rate of 30 \(\upmu \mathrm{l}{\cdot \min}^{-1}\) driven by a 4-channel Ismatec REGLO Digital peristaltic pump (Cole-Parmer GmbH, Germany). Here, we used a unidirectional flow from a source flask through the chips into a waste receptacle. The cell chamber of the first chip contained 800 µl of a N. parvum suspension culture, the cell chamber of the second chip 800 µl of tobacco BY-2 cells. We used the N. parvum cells at day 1 post-subcultivation, the tobacco BY-2 cells at the day of subcultivation. After filling the cell chambers, we closed their openings with biocompatible tape (3 M Polyester Film Tape 851, 3 M Deutschland GmbH, Germany) and incubated the set-up for two additional days. To evaluate the experiment, we extracted the BY-2 cells from the cell chamber and quantified viability by the fluorescence diacetate assay (Widholm 1972). Briefly, we added 1% of a 50-mg ml−1 stock (in acetone) FDA dye and viewed directly. Living cells are fluorescent green, since cytoplasmic esterases cleave the non-fluorescent FDA into the fluorescent product fluorescein. Dead cells lacking this enzyme activity emit no fluorescence. The FDA signal was examined by an AxioImager Z.1 microscope (Zeiss, Jena, Germany), using the filter set 38 HE (excitation: 470 nm, beamsplitter: 495 nm and emission: 525 nm, Zeiss). To detect potential mortality due to the chip itself, we scored viability in a negative control where tobacco BY-2 cells were perfused omitting flow through fungal cells. Data represent three biological replicates counting 550 to 4000 cells per replicate using the analyse particle routine of the ImageJ software (NIH, USA, ver.1.53c, https://imagej.nih.gov/ij/).