This paper presents a method to construct and operate a low-cost, multichannel perfusion cell culture system for measuring the dynamics of secretion and absorption rates of solutes in cellular processes. The system can also expose cells to dynamic stimulus profiles.
Certain cell and tissue functions operate within the dynamic time scale of minutes to hours that are poorly resolved by conventional culture systems. This work has developed a low-cost perfusion bioreactor system that allows culture medium to be continuously perfused into a cell culture module and fractionated in a downstream module to measure dynamics on this scale. The system is constructed almost entirely from commercially available parts and can be parallelized to conduct independent experiments in conventional multi-well cell culture plates simultaneously. This video article demonstrates how to assemble the base setup, which requires only a single multichannel syringe pump and a modified fraction collector to perfuse up to six cultures in parallel. Useful variants on the modular design are also presented that allow for controlled stimulation dynamics, such as solute pulses or pharmacokinetic-like profiles. Importantly, as solute signals travel through the system, they are distorted due to solute dispersion. Furthermore, a method for measuring the residence time distributions (RTDs) of the components of the perfusion setup with a tracer using MATLAB is described. RTDs are useful to calculate how solute signals are distorted by the flow in the multi-compartment system. This system is highly robust and reproducible, so basic researchers can easily adopt it without the need for specialized fabrication facilities.
Many important biological processes occur in cell and tissue cultures on the timescale of minutes to hours1,2,3. While some of these phenomena may be observed and recorded in an automated fashion using time-lapse microscopy4, bioluminescence1, or other methods, experiments involving the collection of culture supernatant samples for chemical analysis are often performed manually in static cell cultures. Manual sampling limits the feasibility of certain studies due to the inconvenience of frequent or after-hours sampling timepoints. Further shortcomings of static culture methods include experiments involving controlled, transient exposures to chemical stimuli. In static cultures, stimuli must be added and removed manually, and stimulus profiles are limited to step changes over time, while medium changes also add and remove other medium components, which can affect cells in an uncontrolled manner5. Fluidic systems can overcome these challenges, but existing devices pose other challenges. Microfluidic devices come with the prohibitive costs of specialized equipment and training to produce and use, require microanalytical methods to process samples, and cells are difficult to recover from the devices after perfusion6. Few macrofluidic systems have been created for the types of experiments described here7,8,9,10, and they are built of multiple custom parts made in-house and require multiple pumps or fraction collectors. Furthermore, the authors are not aware of any commercially available macrofluidic perfusion cell culture systems other than stirred tank bioreactors for suspension culture, which are useful for biomanufacturing, though are not designed for modeling and studying physiology.
The authors previously reported on the design of a low-cost perfusion bioreactor system composed almost entirely of commercially available parts11. The base version of the system enables multiple cultures in a well plate to be kept in a CO2 incubator and continuously perfused with medium from a syringe pump, while the effluent medium streams from the cultures are automatically fractionated into samples over time using a fraction collector with a custom modification. Thus, this system enables automated sampling of culture medium supernatant and continuous solute input to the cultures over time. The system is macrofluidic and modular and can be easily modified to meet the needs of novel experiment designs.
The overall goal of the method presented here is to construct, characterize, and use a perfusion cell culture system that enables experiments in which the secretion or absorption rates of substances by cells over time is measured, and/or cells are exposed to precise, transient solute signals. This video article explains how to assemble the base setup, which is capable of perfusing up to six cell cultures simultaneously using a single syringe pump and modified fraction collector. Two useful variants on the base system that make use of additional pumps and parts to allow for experiments that expose cells to transient solute concentration signals, including brief pulses and pharmacokinetic-like profiles12, are also presented, shown in Figure 1.
Figure 1: Three variations on the perfusion system design. (Top) The basic perfusion system. (Middle) The perfusion system with a stopcock for multiple medium sources. (Bottom) The perfusion system with a stirred tank to mimic a well-mixed volume of distribution. Please click here to view a larger version of this figure.
Due to dispersion and diffusion within the flow, the solute signals become distorted or "smeared" as they travel through the flow system. This distortion can be quantified through the use of residence time distributions (RTDs)13. This article explains how to perform tracer experiments on components of the perfusion system (Figure 2), and provides MATLAB scripts to generate RTDs from measured data. A detailed explanation of this analysis can be found in the authors' previous paper11. Additional MATLAB scripts fit appropriate functions to the RTDs and extract physical parameters, and perform signal convolution using RTDs to predict how solute signal input by the user will propagate and distort through the perfusion system14.
Figure 2: Residence time distributions. The RTDs of flow system components, such as this length of tubing, are measured by inputting a pulse of tracer to the system and measuring how it "smears" by the time it exits into the collected fractions. This figure has been modified from Erickson et al.11. Please click here to view a larger version of this figure.
1. Prepare parts for well plate perfusion
2. Laser-cut the multi-head dispenser and attach it to a fraction collector
3. Measure component RTDs and perform signal convolution
4. Set up the basic perfusion system with cells in a well plate
5. Set up the perfusion system with a stopcock for multiple medium sources
6. Set up the perfusion system with a stirred tank to mimic pharmacokinetics
Figure 3: The multi-head dispenser. Design for the laser-cut multi-head dispenser. This figure has been modified from Erickson et al.11. Please click here to view a larger version of this figure.
The perfusion system with multiple medium sources from section 5 of the protocol was used to measure the expression dynamics of a reporter gene driven by the nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) transcription factor in human embryonic kidney 293 (HEK293) cells in response to a 1.5 h pulse of tumor necrosis factor alpha (TNF-α). HEK293 cells were stably transduced using lentiviral vectors with a gene construct containing Gaussia luciferase (GLuc), driven by a promoter called NFKB containing the NF-κB response element to create NFKB-GLuc HEK293 cells, which were sorted via fluorescence-activated cells sorting (FACS) to isolate transduced cells. Sorted cells were cryopreserved until use.
For each perfused culture, the perfusion setup included an upstream section of 1 m of tubing leading to a plugged 48-well plate containing the cells, followed by 1 m of downstream tubing leading to the multi-head fraction collector, with a flow rate of 0.5 mL/h (0.0083 mL/min). The RTDs of the 1 m tubing alone and of the full setup (1 m tube + 48-well plate + 1 m tube) were measured using a flow rate of 0.5 mL/h with a 20 min pulse of tracer. The background solution was deionized water, with a tracer solution of blue food dye diluted in deionized water, and fractions were collected at a rate of 1 fraction/h using the multi-head dispenser. Tracer concentrations in 100 μL samples of the fractions were measured in a plate reader via absorbance at 628 nm.
The tracer concentration data were processed in MATLAB to produce RTDs for both setups. First, the RTDs of the two setups were computed from the data using the RTD_from_Data.m script. The RTD data points of the 1 m tube and the full setup are shown in Figure 4.
Figure 4: Residence time distributions. (Left) RTD for a 1 m tube at a flow rate of 0.5 mL/h (0.0083 mL/min). (Right) RTD of the full perfusion setup (1 m tube + 48-well plate + 1 m tube) at a flow rate of 0.5 mL/h. Please click here to view a larger version of this figure.
The RTD of tubing alone is fit well by the axial dispersion function, which is to be expected based on existing literature13. Multiple pieces of tubing in series are also fit well by a single axial dispersion model, and adding the 48-well plate in-line causes negligible deviation from this model, so the 1 m tube alone and the 1 m tube + 48-well plate + 1 m tube setup were each expected to be well fit by an axial dispersion model. It is common for poor model fits to occur when the initial guesses for the model parameters are too far from their actual values. To demonstrate this, first, the axial dispersion model was fit to the 1 m tube data using the Fit_RTD_Function.m script, using Pe = 10 and tau = 30 min as the initial guesses for the function parameters. Figure 5 shows that these guesses resulted in poor fitting models, plotted alongside the RTD data. The guesses were changed to Pe = 10 and tau = 300 min, which resulted in the good model fits shown in Figure 5, as tau should be approximately equal to the perfusion system volume divided by its volumetric flow rate. The fit parameters for the 1 m tube are Pe = 154.3, tau = 317.2 min, while the parameters for the whole system are Pe = 396.5, tau = 596.5 min.
Figure 5: Examples of good and poor model fits for an RTD. An axial dispersion model was fit to the RTD data for a 1 m tube at a flow rate of 0.5 mL/h (0.0083 mL/min). (Left) The initial guesses for the model parameters input to the MATLAB script were too far from their true values, causing the script to return a poor model fit. (Right) Better parameter values were input to the script, resulting in a good model fit. Please click here to view a larger version of this figure.
Signal convolution was performed using these fit RTD functions in the Signal_Convolution.m script to predict how a 1.5 h pulse of medium containing 10 ng/mL TNF-α at the inlet of the upstream tubing would propagate through the system. The input signal was first convolved with the 1 m tubing RTD to determine what TNF-α signal would enter the well plate. The input signal was also convolved with the RTD of the 1 m tube + 48-well plate + 1 m tube to determine what the TNF-α signal would be at the outlet of the system in the collected fractions, where it will appear alongside the corresponding GLuc response from the cells. Figure 6 shows the input TNF- α pulse signal is shown alongside the predicted signals at the well inlet and at the system outlet.
Figure 6: Predicting TNF-α concentration signals in the perfusion system. TNF-α-containing medium was infused into the perfusion system at 10 ng/mL for the first 90 min of operation, represented by the blue rectangular signal. (Left) By convolving the input signal with the fit axial dispersion model of the 1 m tube RTD, the TNF-α concentration signal at the inlet to the 48-well plate was predicted. (Right) Similarly, using the RTD model for the whole system, the TNF-α signal at the outlet of the system was predicted. Please click here to view a larger version of this figure.
NFKB-GLuc HEK293 cells were thawed and plated in a 48-well plate at 1 x 104 cells/well in 0.4 mL of Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS), and were incubated for 1 day at 37 °C with 5% CO2, after which the steps in section 5 of the protocol above were followed to set up the perfusion system with a stopcock for multiple medium sources. Four streams were set up to perfuse four wells. One multichannel syringe pump was loaded with four full 60 mL syringes containing DMEM as medium 1. A second syringe pump was loaded with 5 mL syringes containing medium 2, two of which contained plain DMEM as a control, and two containing DMEM with 10 ng/mL TNF-α as the stimulus. At the start of the experiment, the 5 mL syringes containing TNF-α or the control medium were dispensed at 0.5 mL/h for 1.5 h, after which the stopcocks were switched and the 60 mL syringes containing DMEM were dispensed for the next 40 h. At 24 h and at the end of the perfusion, the fractions from the four streams, which were dispensed into microcentrifuge tubes, were labeled and frozen at -80 °C, and the fraction collector was reset.
After the perfusion, all samples were thawed, and their GLuc concentrations were measured using a bioluminescence assay, described previously11,16. The luminescence values were converted to luminescence/h rates by dividing the luminescence of each fraction sample by the time duration of that fraction, and were plotted as a time series for each of the four streams. The time of each GLuc measurement was the midpoint of the time interval during which that fraction was collected, minus the mean residence time (tau = 317 min) of the downstream tubing of the perfusion system. The TNF-α signal predicted to be contained within the fractions was also shifted in time by the mean residence time of the downstream tubing and was overlaid on the GLuc data, revealing the stimulus-response dynamics NF-κB-driven gene expression, shown in Figure 7.
Figure 7: Stimulus-response dynamics measured using the perfusion system. The perfusion system with two medium sources was used to transiently expose two cell cultures to a pulse of TNF-α with two unexposed cultures as controls. The HEK293 cells were engineered to secrete GLuc driven by a promoter containing the NF-κB response element, which was collected in the effluent fractions. The experiment reveals a dramatic expression increase driven by NF-κB following TNF-α exposure. Please click here to view a larger version of this figure.
The presence of accumulated GLuc in the cultures at the beginning of the perfusion caused the apparent drop in GLuc signal from the first to the second fraction in all four curves. This common artifact can be avoided in the perfusion system by changing the medium in the cultures immediately before beginning perfusions. The two control cultures which did not receive TNF-α maintained a low GLuc secretion rate that gradually increased over the duration of the experiment. This result may reflect the growth of the cell population. The stimulated cultures initially had the same GLuc secretion rate as the controls and then dramatically increased secretion as NF-κB-driven expression is activated in response to a TNF-α pulse. GLuc secretion peaks approximately 9 h after the first arrival of the TNF-α, after which it declines with exponential decay.
Supplemental File 1: CAD file: Multi-head_Dispenser.DFX: This file contains the model for the multi-head dispenser which can be laser cut and used to modify fraction collectors. Please click here to download this File.
Supplemental File 2: example_tracer_data.xlsx. Contains example tracer experiment data to be read by the RTD_from_Data.m MATLAB script. Can also be used as a template for new data. Please click here to download this File.
Supplemental File 3: MATLAB file: RTD_from_Data.m. This script reads tracer experiment data from a .xlsx file and returns t and Et vectors representing the RTD. Please click here to download this File.
Supplemental File 4: MATLAB file: Fit_RTD_Function.m. This script loads RTD vectors from the RTD_from_Data.m script and returns a fit model for the RTD. The type of model is chosen by the user. Please click here to download this File.
Supplemental File 5: MATLAB file: Signal_Convolution.m. This script takes a user-defined solute signal and an RTD as inputs and predicts how the signal will be changed after passing through the flow system with that RTD. Please click here to download this File.
Supplemental File 6: MATLAB file: Stirred_Tank_Fit.m. This script fits a CSTR RTD curve to pharmacokinetic data input by the user and returns the tank volume required to produce the RTD for the given flow rate. Please click here to download this File.
This work describes the assembly and operation of a perfusion cell culture system with multiple medium sources demonstrated with a specific example in which the dynamics of NF-κB-driven gene expression in response to a transient pulse of TNF-α were measured. The RTDs of the perfusion system components were measured and modeled, and signal convolution was used to predict both the exposure of the cells to the TNF-α pulse and the TNF-α distribution in the collected effluent medium fractions. The cells were exposed to the pulse and fractions were collected for 40 h, after which the GLuc was measured in the fractions and plotted alongside the predicted TNF-α signal to reveal the stimulation-response dynamics.
This perfusion method is not without shortcomings. Notably, the system carries sterility risk due to the critical brief step in which the tubing is connected to the well plate after setting up within the incubator. In the authors' experience, contamination is quite rare; nevertheless, contamination risk can be mitigated in the following ways. The first is to perform all cell handling procedures and sterile connections in a biosafety cabinet. An additional precaution that can be considered to improve sterility is the inclusion of a 0.22 μm syringe filter in-line at the inlet of the cell cultures. The use of sterile filtration will trap bacteria or larger microorganisms present in the medium before they enter the cell culture, though it increases the pressure requirements to drive flow and may clog due to fouling of the membrane. The outgoing samples are also collected in open environments at the risk of contamination. Collection of samples in an enclosure system, ideally with air filtration, for the fractionation module would eliminate this issue in the future system builds.
Furthermore, signals secreted by cells are distorted according to the RTD of the downstream tubing before they are fractionated and measured. Thus, measured signals are smeared versions of those secreted by cells. Theoretically, this smearing may be undone by performing deconvolution on the measured signal to remove the effect of the downstream RTD14, but it is difficult to apply this numerical technique successfully to nonideal data as signal noise becomes greatly amplified. However, many biological dynamics of interest unfold on the timescale of hours to tens of hours, and thus are affected very little by the smearing effects of tubing RTDs, being shifted in time but changed little in shape. The RTD measurement method shown here may also be improved by using shorter tracer pulses and using background solutions and tracer substances that match those which will be used in cell culture experiments. However, the pulse times suggested here have been shown to give identical RTDs to all shorter pulse times and therefore are suitable for RTD measurements, but as pulse times are increased, the RTDs change more significantly and so should not be used. The authors have found previously that both food dye and GLuc have very similar RTDs to each other in both water and culture medium11, so the molecular species is unlikely to change the RTD much. Finally, no difference in RTDs was observed for food dye in water when measured at 20 °C vs. 37 °C, or in straight tubing vs. highly coiled tubing11. These data, and RTD data for a range of perfusion system geometries and flow rates, can be found in the authors' previous publication11, along with a detailed explanation of the operations performed by the RTD analysis script in this paper. Further details about RTD analysis theory13 and examples of applications of RTDs to bioreactor systems17,18,19,20,21 can be found in the provided references.
This versatile method can be used to measure the secretion or absorption rates of soluble substances in cell cultures, and transient solute signals can be delivered to the cultures by switching upstream medium sources. Simple signals such as solute pulses or step changes can be delivered simply by switching the inlet valve. Pharmacokinetic-like signals can be produced by placing a stirred tank with the appropriate liquid volume in-line upstream of the culture and delivering a solute pulse through it. Other signals can be created by applying different combinations of pulses, step changes, and components with novel RTDs to distort the pulses into desired shapes. Existing static cell culture methods are unable to continuously collect medium samples in an automated fashion and cannot be used to expose cells to smoothly varying solute signals. Microfluidic systems are expensive and require expertise, and no commercially available macrofluidic systems exist for this class of experiments. The low-cost, simple system can be adopted by labs to study natural cell behavior, impulse responses, the effects of different transient solute profiles that may mimic drugs or endogenous solute signal dynamics, and can be reconfigured to meet the needs of novel experiments. Future applications of this technique and those currently underway include: measuring the secretion rate dynamics of cytokines and exosomes from an ex vivo cell therapy; assessing the effect of circadian clock time on cellular response to drugs; measuring the growth rate and metabolism of bacterial biofilms; producing dose-response curves for adeno-associated virus (AAV) vectors in vitro; measuring the dynamic production rate of lentiviral vectors from producer cells; and incorporating pharmacokinetic chemotherapy drug profiles into personalized medicine biopsies for precision cancer therapy.
The authors have nothing to disclose.
This research was conducted with support under Grant Nos. R01EB012521, R01EB028782, and T32 GM008339 from the National Institutes of Health.
18 Gauge 1 1/2- in Disposable Probe Needle For Use With Syringes and Dispensing Machines | Grainger | 5FVK2 | |
293T Cells | ATCC | CRL-3216 | HEK 293T cells used in the Representative Results experiment. |
96-Well Clear Bottom Plates, Corning | VWR | 89091-010 | Plates for measuring dye concentrations in RTD experiments and GLuc in representative results experiment. |
BD Disposable Syringes with Luer-Lok Tips, 5 mL | Fisher Scientific | 14-829-45 | |
BioFrac Fraction Collector | Bio-Rad | 7410002 | Fraction collector that can be used for a single stream, or modified using our method to enable collection from multiple streams. |
Clear High-Strength UV-Resistant Acrylic 12" x 12" x 1/8" | McMaster-Carr | 4615T93 | This sheet is cut using a laser cutter according to the DXF file in the supplemental materials to produce the multi-head dispenser that can be attached to the BioFrac fraction collector. |
Coelenterazine native | NanoLight Technology | 303 | Substrate used in Gaussia luciferase bioluminescence assay in representative results. |
Corning Costar TC-Treated Multiple Well Plates, size 48 wells, polystyrene plate, flat bottom wells | Millipore Sigma | CLS3548 | Used to grow and perfuse 293T cells in representative results. |
Corning Costar Flat Bottom Cell Culture Plates, size 12 wells | Fisher Scientific | 720081 | Can be plugged and used as a stirred tank to produce pharmacokinetic profiles in perfusion. Can also contain cells for perfusion. |
DMEM, high glucose | ThermoFisher Scientific | 11965126 | |
Epilog Zing 24 Laser | Cutting Edge Systems | Epilog Zing 24 | Laser cutter used to produce multi-head dispenser from acrylic sheet. Other laser cutters may be used. |
Fisherbrand Sterile Syringes for Single Use, Luer-Lock, 20 mL | Fisher Scientific | 14-955-460 | |
Fisherbrand Sterile Syringes for Single Use, Luer-Lock, 60 mL | Fisher Scientific | 14-955-461 | |
Fisherbrand Premium Microcentrifuge Tubes: 1.5mL | Fisher Scientific | 05-408-129 | Microcentrifuge tubes for collecting fractions. |
Fisherbrand Round Bottom Disposable Borosilicate Glass Tubes with Plain End | Fisher Scientific | 14-961-26 | Glass tubes for collecting fractions. |
Fisherbrand SureOne Micropoint Pipette Tips, Universal Fit, Non-Filtered | Fisher Scientific | 2707410 | 300 ul pipette tips that best fit the multi-head dispenser and tubing to act as dispensing tips. |
Gibco DPBS, powder, no calcium, no magnesium | Fisher Scientific | 21600010 | Phosphate buffered saline. |
Labline 4625 Titer Shaker | Marshall Scientific | Labline 4625 Titer Shaker | Orbital shaker used to keep stirred tanks mixed. |
Masterflex Fitting, Polycarbonate, Four-Way Stopcock, Male Luer Lock, Non-Sterile; 10/PK | Cole-Parmer | EW-30600-04 | Used to join multiple inlet streams for RTD experiments and cell culture experiments. |
Masterflex Fitting, Polycarbonate, Straight, Female Luer x Cap; 25/PK | Masterflex | UX-45501-28 | |
Masterflex Fitting, Polypropylene, Straight, Female Luer to Hosebarb Adapters, 1/16" | Cole-Parmer | EW-45508-00 | |
Masterflex Fitting, Polypropylene, Straight, Male Luer Lock to Hosebarb Adapter, 1/16" ID | Cole-Parmer | EW-45518-00 | |
Masterflex Fitting, Polypropylene, Straight, Male Luer Lock to Plug Adapter; 25/PK | Masterflex | EW-30800-30 | |
Masterflex L/S Precision Pump Tubing, Platinum-Cured Silicone, L/S 14; 25 ft | Masterflex | EW-96410-14 | |
MATLAB | MathWorks | R2019b | Version R2019b. Newer versions may also be used. Some older versions may work. |
NE-1600 Six Channel Programmable Syringe Pump | New Era Pump Systems | NE-1600 | |
Rack Set F1 | Bio-Rad | 7410010 | Racks to hold collecting tubes in the fraction collector. |
Recombinant Human TNF-alpha (HEK293-expressed) Protein, CF | Bio-Techne | 10291-TA-020 | Cytokine used to stimulate 293T cells in representative results. |
Saint Gobain Solid Stoppers, Versilic Silicone, Size: 00, Bottom 10.5mm | Saint Gobain | DX263015-50 | Fits 48-well plates. |
Saint Gobain Solid Stoppers, Versilic Silicone, Size: 4 Bottom 21mm | Saint Gobain | DX263027-10 | Fits 12-well plates. |
Sodium Hydroxide, 10.0 N Aqueous Solution APHA; 1 L | Spectrum Chemicals | S-395-1LT | |
SolidWorks | Dassault Systems | SolidWorks | CAD software used to create the multi-head dispenser DXF file. |
Varioskan LUX multimode microplate reader | ThermoFisher Scientific | VL0000D0 | Plate reader. |
Wilton Color Right Performance Color System Base Refill, Blue | Michaels | 10404779 | Blue food dye containing Brilliant Blue FCF, used as a tracer in RTD experiments. Absorbance spectrum peaks at 628 nm. |