Microfluidic 3D Hepatic Cultures Integrated with a Droplet-based Bioanalysis Unit

Abstract

A common challenge in microfluidic cell cultures has to do with analysis of cell function without replacing a significant fraction of the culture volume and disturbing local concentration gradients of signals. To address this challenge, we developed a microfluidic cell culture device with an integrated bioanalysis unit to enable on-chip analysis of picoliter volumes of cell-conditioned media. The culture module consisted of an array of 140 microwells with a diameter of 300 μm which were made low-binding to promote organization of cells into 3D spheroids. The bioanalysis module contained a droplet generator unit, 15 micromechanical valves and reservoirs loaded with reagents. Each 0.8 nL droplet contained an aliquot of conditioned media mixed with assay reagents. The use of microvalves allowed us to load enzymatic assay and immunoassay into sequentially generated droplets for detection of glucose and albumin, respectively. As a biological application of the microfluidic device, we evaluated hormonal stimulation and glucose consumption of hepatic spheroids. To mimic physiological processes occurring during feeding and fasting, hepatic spheroids were exposed to pancreatic hormones, insulin or glucagon. The droplet-based bioanalysis module was used to measure uptake or release of glucose upon hormonal stimulation. In the future, we intend to use this microfluidic device to mimic and measure pathophysiological processes associated with hepatic insulin resistance and diabetes in the context of metabolic syndrome.

1. INTRODUCTION

Regulation of glucose levels in the body is accomplished by complex interactions between multiple organs including pancreas and liver. Food ingestion causes glucose levels to rise and triggers production of insulin by the pancreas. Insulin is then delivered into the liver via portal circulation and stimulates hepatocytes to take up and store excess glucose. Fasting, on the other hand, triggers pancreatic production of glucagon which in turn stimulates the liver to release glucose into blood circulation (Klover and Mooney, 2004; Petersen et al., 2017). The ability of the liver to control glucose levels is affected by injury (e.g. lipotoxicity) and, as result, liver diseases such as non-alcoholic steatohepatitis (NASH) co-occur with diabetes (Parthasarathy et al., 2020). The tools available to study the pathophysiological processes underlying diabetes and NASH remain limited. These processes involve rapid changes in hormone and glucose levels which are difficult to quantify using standard laboratory equipment. In this study, we integrated bioanalytical capabilities into a microfluidic liver (hepatocyte) cell culture system to enable dynamic monitoring of glucose levels upon hormonal stimulation.

Some of the early microfluidic hepatocyte cultures were developed by Kane et al., who combined micropatterned co-cultures of hepatocytes and fibroblasts with oxygenation strategies to demonstrate two weeks of stable culture (Kane et al., 2006). A different device was described by Park et al., who seeded hepatocytes into grooved channels to protect cells from shear stress generated during perfusion (Park et al., 2006). Other studies focused on recapitulating liver sinusoid composition and interactions (Blake et al., 2010; Du et al., 2017; Lee et al., 2007; Prodanov et al., 2016), and inducing liver zonation in microfluidic devices (Bulutoglu et al., 2019; Kang et al., 2018). Unlike the above mentioned studies that relied on perfusion, our laboratory has been interested in microfluidic hepatocyte cultures in the absence of active flow (Choi et al., 2020; de Hoyos‐Vega et al., 2023; Haque et al., 2016). We observed enhanced hepatic phenotype in such devices and determined that rapid accumulation of hepato-inductive autocrine signals, including hepatocyte growth factor (HGF), played an important role.

While microfluidic devices offer numerous advantages for cultivation of liver and other cell types, they also present challenges. One challenge is the analysis of cell function indicators using minimal (microliter) volumes of media available in such devices. A solution to this challenge may be to send cell-conditioned medium from a microfluidic cell culture device to a different bioanalytical device. For example, microfluidic bioreactors were developed to culture hepatocyte spheroids with conditioned media transferred to electrochemical sensing platforms for measurements of glucose/lactate (Bavli et al., 2016) or transferrin/albumin (Riahi et al., 2016). Another microfluidic detection scheme involved perfusing insulin into hepatocyte cultures and then detecting glucose levels in a separate droplet microfluidic device connected by tubing (Adams et al., 2019). Our lab has previously developed a microvalve-enabled microfluidic device for handling sub-nanoliter volumes and performing multiplexed biochemical analysis (Cedillo-Alcantar et al., 2019). This device worked by mixing aliquots of sample with reagents and then converting this mixture into a train of 0.8 nL droplets. While this bioanalysis unit could be positioned downstream of the microfluidic cell culture chamber and connected by tubing, the dead volume in this connection was similar to the volume of the cell culture chamber (4.25 μL). Thus, each measurement required a total volume exchange in the cell culture chamber. Given the importantance played by accumulating exogenous signals in shaping phenotype of cells in microfluidic cultures (Choi et al., 2020; de Hoyos‐Vega et al., 2023; Fattahi et al., 2023; Haque et al., 2016), it is important to design systems for sampling of conditioned media with minimal dilution and disturbance to accumulated signals. Such design criterion may only be achieved by minimizing the dead volume of interconnections between cell culture and bioanalysis units. We, therefore, wanted to develop a monolith device integrating both cell culture and analysis capabilities.

There have been multiple reports describing integration of biosensors alongside cells in microfluidic devices (Zhou et al., 2015b). Our team described placing enzyme-functionalized electrodes into microfluidic devices with hepatocytes to monitor oxidative stress produced by alcohol injury (Matharu et al., 2013). Another microfluidic device developed by our group incorporated electrochemical aptasensors to continuously monitor secretion of TGF-β1 (Matharu et al., 2014), and to study how two cell populations interacted via this signal (Zhou et al., 2015a). Other examples of devices integrating cell culture and analysis capabilities includes systems for continuous monitoring of insulin secretion from the islets of Langerhans using an electrophoresis-based immunoassay (Dishinger et al., 2009; Lu et al., 2018) or a fluorescence anisotropy immunoassay (Adablah et al., 2020; Schrell et al., 2017). Easley and co-workers developed devices which integrate peristaltic pumps and pneumatic microvalves to automate the measurement of adipose (Hu et al., 2020) or endocrine (Li et al., 2018) tissue secretion dynamics using droplet microfluidics. While incredibly useful for monitoring cell secretory activity, these devices were used in experiments lasting 24 to 48 h and were not designed for longer-term culture.

Our objective was to develop a microfluidic device capable of multiday cultivation of cells and on-chip analysis of cell function. The design of the microfluidic 3D culture chamber was adapted from simpler microfluidic devices that allowed to maintain functional hepatocyte spheroids for up to four weeks (Choi et al., 2020; Fattahi et al., 2023). This design, shown in Figure 1, contained an array of microwells for culturing cell spheroids connected to media reservoirs via transport channels. Our device also integrated a droplet-based bioanalysis unit that was used to sample 0.8 nL aliquots of conditioned media for on-chip glucose measurements. Placement of the bioanalysis module next to cells ensured that analyte was not diluted and that only 0.25% of the culture volume (5 nL of 2 μL) was used for each measurement. To demonstrate the utility of this device, hepatic spheroid cultures were stimulated with insulin or glucagon while measuring glucose consumption or release.

Figure 1: Capabilities and use of the microfluidic device developed in this study.

The automated microfluidic device offers the following capabilities: 1) Formation and culture of hepatic spheroids. Single cells organized into spheroids in the microwells of the cell culture chamber. Hepatic spheroids were injured by lipid loading. 2) Rapid exchange of media in the cell culture chamber. This capability was used to introduce pancreatic hormones. 3) Sampling sub-nanoliter volumes of conditioned media using a droplet generator for measuring glucose levels upon hormonal stimulation. This device is equipped with micromechanical valves to enable flow metering, media exchange and droplet generation. We leveraged this microfluidic device to evaluate changes in glucose metabolism of hepatic spheroids in response to stimulation with pancreatic hormones, glucagon and insulin.

2. MATERIALS AND METHODS

2.1. Reagents

The following reagents were purchased from Sigma-Aldrich: 4-aminoantipyrine (4-AAP, A4382), propylene glycol monomethyl ether acetate (PGMEA, 484431), hexamethyldisilane (HMDS, 440191), chlorotrimethylsilane (386529), pluronic F-127 (P2443), glucagon (G2044), D-(+)-glucose (G7021), glucose oxidase from Aspergillus niger (GOx, G2133), β-nicotinamide adenine dinucleotide sodium salt (NAD+, N0632), sodium L-lactate (L7022), palmitic acid (P0500), Oil Red O (O0625), resazurin sodium salt (R7017), and tetramethylrhodamine isothiocyanate (TRITC)–dextran 20 kDa (73766). The following reagents were purchased from Thermo Fischer Scientific: Dulbecco’s modified eagle medium (DMEM, MT-10–013-CV), penicillin-streptomycin (15140122), Gibco fetal bovine serum (FBS, 10437028), 1X phosphate-buffered saline (PBS, 21–031-CV), and LIVE/DEAD^™ viability/cytotoxicity kit (L3224). Diaphorase (DPR, ICN15084301) and horseradish peroxidase (HRP, ICN19537325) were obtained from MP Biomedical. N-ethyl-N-(2-hydroxy-3- sulfopropyl)-3-methoxyaniline sodium salt (ADOS, OC01–10) was purchased from Dojindo Molecular Technology. The PicoProbe^™ LDH-Cytotoxicity Fluorometric assay kit (K314) and Albumin ELISA kit (E110–125) were purchased from BioVision and Bethyl Lab, respectively. Polydimethylsiloxane (PDMS) was acquired from Ellsworth as Sylgard 184 elastomer kit (2065622). SU-8 2025, SU-8 2050 and SU-8 2100 negative photoresists and SU-8 developer were acquired from Kayaku Advanced Materials. AZ 50XT positive photoresist and AZ-400K developer were obtained from MicroChem. Insulin (Novolin N, 100 units/mL) was purchased from Novo Nordisk. HFE-7500 3M^™ Novec^™ Engineered fluid (051243) was purchased from Oakwood Chemical. Pico-SURF^™ 1 (2% in Novec^™ 7500) (#F005) was purchased from Sphere Fluidics. Aquapel glass treatment was purchased from PGW Auto Glass, LLC.

2.2. Device fabrication

Three master molds were fabricated by photolithography, one for each layer of the microfluidic device: flow, control and microwells. Designs for the master molds were created in AutoCAD (v2020 Student Edition, Autodesk). The control and microwell molds consisted of a single layer of negative photoresist, SU-8 2025 and SU-8 2100, respectively. For the control mold, a pristine silicon wafer (452, University Wafer MS) was treated with oxygen plasma for 3 min in a plasma etcher (YES-G500, Yield Engineering Systems, Inc), and a 50 μm layer of photoresist was spin coated at 1400 rpm on top of it. The wafer was soft baked at 65 °C for 5 min and 95 °C for 10 min. The corresponding design was exposed using a micro pattern generator (μPG 101, Heidelberg) equipped with a UV laser (375 nm, 70 mW), followed by a post bake at 65 °C for 2 min and 95 °C for 7 min. Afterwards, the mold was developed by soaking in SU-8 developer until all undesired photoresist was removed (~5 min). A hard-bake step was performed by placing the mold on a hot plate for 15 min at 160 °C, to finally exposed to chlorotrimethylsilane vapor in a closed chamber for 1 h. The microwell mold was fabricated similarly: SU-8 2100 was spin coated at 900 rpm to obtain a layer of 300 μm, and the soft bake consisted of 65 °C for 20 min and 95 °C for 120 min. The mold was exposed using the μPG 101 and post baked 65 °C for 5 min and 95 °C for 30 min. Developing, hard bake, and exposure to chlorotrimethylsilane were performed similarly.

The flow mold comprised a combination of positive and negative photoresists. First, a silicon wafer was treated with oxygen plasma for 3 min and baked at 200 °C. A solution consisting of HMDS and PGMEA at a 1:4 ratio was spin coated on the wafer at 3000 rpm followed by baking at 120 °C for 3 min. Then, a layer of 70 μm of positive photoresists was deposited on the wafer by spin coating at 750 rpm. The soft bake consisted of 1 min at 95 °C and 5 min at 115 °C. The negative of the corresponding design was exposed using the μPG 101, followed by developing using AZ 400K. A reflow step to create rounded channels was done by placing the wafer on a hot plate at 60 °C and increasing temperature to 120 °C with a ramp of 2 °C/min. Temperature was held for 30 min and then increased to 200 °C with a ramp of 3 °C/min, maintaining the temperature for 2 h. The wafer was cooled down on the hot plate to room temperature. Afterwards, the mold was treated with oxygen plasma for 30 s and a 100 μm layer of negative resist was coated on top of the positive resist structures (SU-8 2050 at 1400 rpm). The mold was then processed in the same fashion as the control mold, with the only difference that soft bake and post exposure bake consisted of 65 °C for 5 min and 95 °C for 30 min, and 65 °C for 5 min and 95 °C for 10 min, respectively. All molds were placed in 150 mm Petri dishes until used.

The PDMS replica devices were fabricated by multilayer soft lithography. For the flow and control layers, a PDMS base and curing agent were mixed at 20:1 and 5:1 ratio in a planetary centrifuge (ARE-310, Thinky), respectively. 30 g of PDMS mix (5:1) was poured on the control mold and degassed on a vacuum desiccator for 10 min. 10 g of PDMS mix (20:1) was spin coated on the flow mold at 600 rpm to create a PDMS layer of ~150 μm. Both molds were baked in an oven at 80 °C for 20 min at the same time. Control mold devices were peeled off by the edges and inlets punched. Each control PDMS slab was manually aligned on top of the flow mold with the thin PDMS layer using a stereoscope. The flow mold with both flow and control PDMS layers was baked at 80 °C for 90 min to bond the layers together. Then, the two-layer device was peeled off and flow inlets/outlets punched. For the microwells layer, 20 g of PDMS at a 10:1 ratio was poured onto the mold, degassed for 10 min and baked at 80 °C for 1 h. Then, each PDMS device was peeled off from the mold and kept. For bonding all PDMS layers together, 5 g of curing agent was spin coated on a pristine silicon wafer at 3000 rpm for 20 s. The microchannels side of each flow/control PDMS slab was wetted in the curing agent and aligned on top of each of the microwells PDMS slab, with the microchannels facing the microwells. The three-layer device was placed in a Petri dish, two glass cloning cylinders were bonded with uncured PDMS to the cell seeding inlets and then baked for 30 min at 80 °C to fully bond all PDMS layers and glass cylinders together.

2.3. Device automation

To automate droplet assays in the microfluidic device, a PC custom-made graphic user interface (GUI) was created using LabVIEW (Community Edition, National Instruments). The GUI interfaced the PC with a microcontroller that controlled an individual 3-way solenoid valve for each one of the 13 microvalves in the microfluidic device. The microvalves were connected to the solenoid valves through medical-grade tubing, and could be switch between an ON (pressurized, 30 psi) and OFF (not pressurized, 0 psi) state. To automate droplet assays, the GUI was programmed with a feature where the number of droplets, size of droplet (through valve opening time), inlet media, and assay type could be selected. Necessary microvalves are actuated based on the input parameters and droplets are generated.

For droplet generation, the incubation channel was first washed with 70% ethanol and dried out. Then, 100 μL of Aquapel were injected into the incubation channel. Once the Aquapel had dried out, the oil phase (1% Pico-Surf in HFE-7500) was continuously injected into the droplet generation channel at 0.2 psi and the microvalve connecting the media/reagents channel to the droplet generation channel was kept activated. Then, the connecting microvalve was deactivated and the media/reagents solution was co-flowed into the droplet generation channel at 0.5 psi, stopping the oil phase during the droplet generation time previously determined. Next, the connecting microvalve was activated again, reestablishing the oil phase flow and moving the droplet further into the droplet microchannel. The same process was repeated for the determined number of droplets per assay.

2.4. Assessing analyte dilution during droplet generation

We compared analyte dilution for two scenarios: 1) two microfluidic devices, one for cell culture, another for droplet analysis, connected via tubing, and 2) a monolith device with integrated cell culture and analysis chambers. Dextran-rhodamine (DTR) (MW 20 kDa) was used as a model analyte. It was first loaded into the cell culture chamber at 0.5 mg/mL and then moved into a droplet generating device or module by infusing 1X PBS into the device. Droplets were generated continuously, and images were collected for every 30 droplets using fluorescence microscopy (IX83, Olympus, Japan). Images were analyzed using ImageJ to compare the dilution of the fluorescent signals over time between both systems.

2.5. Cell culture in the microfluidic device

Human hepatocarcinoma cells (HepG2) were used in these experiments. First, all channels and culture chambers of a microfluidic device were treated overnight with 1% of pluronic F-127 in 1X PBS to prevent adhesion of cells. Cells were then seeded into microfluidic devices using a cell concentration of 4×10⁶ cells/mL. The cell suspension was introduced into one media reservoir while the other reservoir was empty. Cells flowed into the chamber as media levels equilibrated between the reservoirs and sedimented into the microwells. Excess cells were washed out by exchanging the cell suspension with fresh media. Hepatic cells formed spheroids within 2 days and were typically cultured for a total of 7 days. Cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplement with 25 mM glucose, fetal bovine serum (10% in volume ratio, Invitrogen) and penicillin and streptomycin (200 mg/mL, Invitrogen) at 37 °C and 5% CO₂. Media was exchanged daily.

2.6. Calibration of in-droplet bioassays

For each assay, a reagent solution was prepared and then mixed in a 1:1 ratio with control solutions inside the microfluidic device to create droplet bioreactors. For glucose detection, we performed an enzymatic assay, and the reagent solution was prepared using GOx (70 U/mL), HRP (117 U/mL), ADOS (3.6 mM), and 4-AAP (3.1 mM) dissolved in 1X PBS PBS (Han et al., 2014). Eight different glucose control solutions were prepared in culture media (at concentrations of 0, 0.46, 0.93, 1.58, 3.75, 7.5 and 10 mM) to create a calibration curve. In the enzymatic assay, glucose was broken down by glucose oxidase (GOx) producing H₂O₂, which in turn became oxidized by horseradish peroxidase (HRP). In the process, chromogens 4-AAP and ADOS were reduced, generating a magenta-colored product. The intensity of the magenta color was directly proportional to the glucose concentration.

A microbead-based fluorescent immunoassay was used for albumin detection. Streptavidin-decorated microbeads were incubated in 50 μg/mL of anti-albumin antibodies (Abs) for functionalization. For performing in-droplet immunoassays, a suspension containing 7×10⁶ particles/mL of microbeads was mixed with Alexa-546 labeled anti-albumin detection antibodies (at 3 μg/mL) in 1X PBS and added 25% PEG 35 kDa and 25% of OptiPrep as densifiers. Control albumin solutions with known concentrations (100, 50, 25, 12.5, 6.25, and 0 ng/mL) were mixed inside the device with the suspension containing microbeads and detection Abs to construct a calibration curve.

To create both calibration curves, the control solutions were introduced into the device through the media inlet for 10 min to fill the culture chamber. The reagent solutions were then injected into the reagent inlets. For each analyte concentration, five droplets were generated and incubated for 5 min for the glucose assay and 15 min for the albumin assay. Micrographs of the droplets were acquired and analyzed to generate calibration curves.

2.7. Assessing glucose metabolism of hepatic cells

Hepatic spheroids were cultured in the microfluidic device for five days. Spheroids were then conditioned for 1 h with culture media supplemented with 10 mM glucose. Next, the culture chamber was washed with fresh conditioned media (10 mM glucose) for 10 min. Immediately after, 5 droplets were generated by mixing culture media from the chamber with glucose assay reagents. These droplets were used to establish a baseline for the glucose detection. After 5 minutes of incubation, micrographs of the droplets were collected and analyzed. Subsequently, droplets were generated every 2 h to obtain a glucose consumption curve. At the end of the experiment, the spheroids were exposed to calcein and ethidium homodimer to assess cell viability. Live/Dead assay was used per manufacturer’s instructions.

For hormonal stimulation, hepatic spheroids were preconditioned with media containing 5 mM glucose for 1 h. Hormonal stimulation experiments were carried out in media without FBS to exclude the possibility of insulin and glucagon presence in the serum. Next, the culture chamber was perfused with fresh media supplemented with 5 mM glucose and 3 mM insulin or 2 nM glucagon for 10 min. 5 droplets were then generated to establish a glucose baseline (5 mM). Additional measurements were made by generating droplets every 30 min over the course of 3 h. Each session of glucose measurement consisted of the following steps. 1) 5 droplets were generated for 1 min. 2) Flow was stopped, and in-droplet biochemical assay was allowed to develop over the course of 5 min. 3) Images were obtained using an inverted microscope and analyzed with MATLAB to obtain intensity values. Mean intensity and standard deviation of in-droplet signals were plotted using Prism (v8, GraphPad).

2.8. Lipid loading into hepatic cells and assessment of glucose metabolism

Hepatic cells were seeded into a microfluidic device as described above and cultured for three days to promote spheroid formation. Subsequently, spheroids were exposed to 800 μM palmitate (PA) for 48 h. To prepare PA-containing media, 80 mM of PA in isopropanol was mixed with hepatic media (see above) to achieve a final concentration of 800 μM. The media also contained 2% bovine serum albumin (BSA). After lipid injury experiments, hepatic spheroids were retrieved from microfluidic devices, fixed with paraformaldehyde, cryosectioned (8 μm thick slices) and stained with Oil Red O. This step was used to confirm lipid loading of hepatic cells.

Additional assays included LDH for cytotoxicity and albumin ELISA for hepatic function. All assays were carried out in accordance with manufacturer’s instructions. Optical density measurements were made using an UV/Vis spectrophotometer (Synergy H1, BioTek).

When assessing the glucose metabolism of injured cells, hepatic spheroids in a microfluidic device were preconditioned for 1h in DMEM at 5 mM glucose without FBS. After this preconditioning, we commenced the hormone stimulation and glucose analysis experiment in the microfluidic device as described above.

2.9. Image Analysis

We utilized a custom-made MATLAB script for image analysis. For the microbead-based fluorescent immunoassay, the images were initially converted to grayscale, and a region of interest (ROI) containing a droplet was defined. Then, circles with radii ranging from 2 to 5 pixels, corresponding to the size of the microbeads, were identified, and the mean intensity was extracted. We plotted the mean intensity derived from these circles against the respective albumin concentrations to create the calibration curve. For the colorimetric enzymatic assay, the images were transformed into CMYK. Similarly, a ROI containing a droplet was defined, and the intensity of a rectangle of 180 × 70 px in the middle of the droplet was measured for magenta intensity. For the glucose calibration curve, we plotted the magenta intensity from the images as a function of the corresponding glucose concentrations.

2.10. Statistical Analysis

A minimum of three replicates (n > 3) were analyzed and averaged for each experimental group. All data was presented as mean and standard deviation (mean ± SD). Statistical significance was determined using a two-tailed unpaired t-test analysis with p < 0.05 or ** p< 0.01. GraphPad Prism (ver. 7; GraphPad Software) was used to statistically analyze the data.

3. RESULTS AND DISCUSSION

Our paper describes a microfluidic device integrating 3D cell cultures with a droplet-based bioanalysis module for sampling and analysis of cell-conditioned media. This device was employed for hormonal stimulation and measurement of glucose metabolism in hepatic cultures.

3.1. Design and operation principle of the microfluidic device

As shown in Figure 2A, our microfluidic device consisted of an array of 140 microwells for 3D cell culture integrated with a droplet-generator for sampling and analysis of conditioned media. Microvalves were integrated into the device to 1) control composition of media entering the cell culture chamber (See Figure 2B), 2) isolate the cell culture chamber from the media reservoirs during sampling, 3) co-flow conditioned media with assay reagents, and 4) generate droplets for analysis of media composition (See Figure 2C). This device allowed us to rapidly alternate between 4 media inputs and 3 reagent inputs (See Supplementary Video 1). Droplet generation was accomplished by co-flowing conditioned media and assay reagents at the same flow rate (~38.8 μL/min) to achieve a 1:1 ratio of sample to reagent volume. It should be noted that our device was different from other droplet generators where droplet size is solely controlled by the flow rate of aqueous and oil streams. As shown in Figure S1, our droplet generator contained a micromechanical valve (labeled as valve 1) at the aqueous/oil junction which allowed us to control droplet volume by opening and closing of the valve, as well as by adjusting the flow rates. This feature of our droplet generator is highlighted in Figure S1 which shows that keeping the valve open for opening t0.1, 0.15, 0.2, and 0.25 s generated droplets of 0.4, 0.8, 1, and 1.4 nL, respectively. For this study, we chose to work with smaller, ~0.8 nL droplets. We generated 3 to 5 droplets for each time point, striking a balance between minimal dilution of media and sufficient number of replicates. After droplets were generated, the flow was stopped and droplets were stored in the incubation channel to give time for the assay to develop. The enzymatic assay for glucose produced detectable signals within 1 min while the immunoassay for albumin detection required ~15 min.

Figure 2: Droplet generation in the microfluidic device.

(A) Photograph of the microfluidic device showing a cell culture unit (blue) and a bioanalysis unit (yellow). Automation is accomplished with microvalves (red) that allow to isolate the culture chamber from the media reservoirs, control fluid flow and generate droplets. Scale bar: 5 mm. (B) Sequence of micrographs showing exchange of media inputs into the culture chamber. Scale bar: 500 μm. (C) Droplets were generated by co-flowing solutions from the culture chamber and reagent inlets towards the droplet generation unit. The sequence and pattern of microvalve actuation was used to direct the co-flow into the droplet generator. Scale bar: 500 μm.

We previously used a standalone droplet-based bioanalysis microfluidic device connected by tubing to a standalone cell culture microfluidic device (Cedillo-Alcantar et al., 2019). In the present paper, we hypothesized that combining the droplet generator and cell culture units in the same microfluidic device would significantly decrease analyte dilution during sampling. To test it, we set up an experiment, Figure 3A, where dilutions of a model fluorescent analyte (dextran-rhodamine or DTR) during droplet generation/sampling was compared between the two-device system with tubing interconnection and the monolith device developed in this paper. For the first scenario, we kept tubing diameter and length to a minimum (0.5 mm and 16 mm), for a volume of 3.24 μL, which was similar to that of the cell culture chamber (4.25 μL). The cell culture device was first filled with DTR solution. Then, the process of droplet generation was commenced by flowing 1X PBS into the cell culture chamber and pushing a fluorescent solution into the droplet generator. Droplet size and generation rate were controlled by the microvalve at the oil/water junction and by the flow rate and were kept the same for both scenarios. As can be seen from Figure 3B, the fluorescence intensity of the first droplet in the system of two interconnected devices was ~2-fold lower compared to fluorescence intensity at the source, in the cell culture chamber. This is expected because in the case of two interconnected devices, droplet generation begins by injecting PBS into the cell culture device and then gradually displacing the fluorescent dextran solution into the droplet generator. Dilution of signal occurs because the volume of Tygon tubing connecting the two devices is 3.2 μL and similar to the volume of the fluorescent dextran being displaced toward the droplet generator (4.3 μL). In the case of the monolith device, fluorescent dextran was displaced with minimal dilution and allowed to generate 500 droplet (0.8 nL each) with stably high fluorescence intensity.

Figure 3: Characterizing sample dilution in the device integrating cell culture and bioanalysis units.

(A) Dilution during sampling was compared for two scenarios: 1) a cell culture device connected by tubing to a droplet-based bioanalysis unit, and 2) a monolith device with integrated bioanalysis and cell culture units. (B) A DTR solution was loaded into the cell culture chamber and then converted into droplets in the bioanalysis unit.

Fluorescence images were acquired for 300 to 800 consecutive droplets. Scale bar: 200 μm. (C) Analysis of fluorescence intensity in droplets demonstrates stably high DTR concentration for ~500 consecutive droplets in a monolith device. For the interconnection of two devices, DTR signal was 2-fold lower initially and dropped off precipitously after 50 droplets.

Importantly, in-droplet fluorescence intensity remained relatively constant for up to 500 droplets in the monolith device while it dropped precipitously in the connected system (see Figure 3C). The latter observation was due to dilution of the fluorescence molecules in the connection tubing. This experiment highlights the benefit of integrating the droplet generator into the cell culture system in terms of minimizing analyte dilution and permitting continued sampling.

Another important feature of our microfluidic device is the ability to dynamically change the composition of cell culture media. The device is equipped with 4 reagent inputs for delivering compounds into the cell culture chamber (see Figure 4A). Microvalve actuation is used to reconfigure the microfluidic device from culture mode, where the cell culture chamber is connected to media reservoirs, to sampling mode, where chemical inputs are introduced into the culture chamber and droplets may be generated. Figure 4A shows microvalves that are deactivated (light red) to introduce media components into the culture chamber. The use of automated microvalves allows one to switch between media inputs and to fully exchange the composition of the cell culture chamber within 10 min (see Figure 4B). Microvalves are computer-controlled and may be used to alternate between media inputs in an automated manner, controlling frequency of exchanges and residence time for each solution. Figure 4C shows profiles for different frequency of media exchanges, 2.5 and 9 cycles with different solution concentrations in the 4 h time window. These data highlight our ability to create complex stimulation profiles in a repeatable manner. Furthermore, the culture chamber may be stimulated in a region-specific manner as shown in Figure 4D where equal parts of the cell culture chamber are exposed to red and green fluorescence with negligible mixing. In summary, integration of 3D cultures with automation allows us to precisely define chemical stimuli. This capability is leveraged in the paper to stimulate hepatic cells with insulin and glucagon while measuring changes in glucose metabolism.

Figure 4: Demonstrating automated sample exchange capabilities of the microfluidic device.

(A) Schematic illustration of the culture chamber being perfused with fluorescent molecules (DTR) through media inputs. Note that the microvalves are actuated to isolate volume of the cell culture chamber from media reservoirs. (B) Sequence of images demonstrating solution exchange in the culture chamber. The chamber was filled with DTR and then flushed with 1X PBS in three cycles. The fluorescence intensity profile shows repeatability of solution exchange across three cycles. The fluorescence intensity was measured inside a given microwell. Numbers in the graphs correspond to micrograph sequence. Scale bar: 500 μm. (C) Fluorescence intensity profiles of low and high frequency stimulation cycles. Left panel: the low frequency stimulation shows changing of 4 different solutions of DTR (0, 1, 2, and 2 mg/mL) every 15 min. Right panel: high frequency stimulation where 3 different DTR solutions were exchanged with different periodicity: 0 mg/mL was perfused for 5 min, 1 mg/mL for 15 min, and 3 mg/mL for 10 min. Fluorescence intensities were measured in a microwell (red curve) and on the headspace of the cell culture chamber (blue curve). (D) Parallel injection of stimulation solutions. Left panel: fluorescence micrograph of the culture chamber being simultaneously perfused with FITC and DTR solutions. Right panel: fluorescence intensity profiles from a cell culture chamber containing FITC (green) and DTR (red). Note limited mixing between the colors. Scale bar: 500 μm.

3.2. Characterizing in-droplet assays for glucose and albumin.

The biological goal of this study was to investigate the response of normal and injured liver (hepatic) cells to stimulation with pancreatic hormones, insulin and glucagon. This stimulation was expected to modulate glucose concentrations in the extracellular space. We therefore wanted to establish the ability to detect changes in glucose concentrations in the microfluidic device. The liver has multiple roles, where one of them is the production of serum proteins, such as albumin. Therefore, albumin synthesis is a well-established indicator of hepatic health and function (de Hoyos-Vega et al., 2021) which motivated us to develop an on-chip assay for this analyte.

We adapted a commercial enzymatic glucose assay, shown in Figure 5A, for use in droplets by increasing the concentration of HRP and ADOS, and thus improving the intensity (absorbance) of the signal (Han et al., 2014). To obtain a calibration curve, we prepared different glucose concentrations (0–10 mM) that were introduced sequentially into the microfluidic device via media inputs, and then mixed with assay reagents in the droplet generator unit. We used an incubation time of 5 minutes which was deemed sufficient for the enzymatic reaction (Cedillo-Alcantar et al., 2019). As shown in Figure 5B, the increased intensity of the magenta color in droplets correlated with an increasing concentration of glucose, allowing us to construct a calibration curve for this assay. This experiment revealed that an in-droplet glucose assay had a limit of detection of 0.034 mM, with a linear range extending to 10 mM.

Figure 5: Characterizing in-droplet glucose and albumin assay.

Assay for glucose (A) and albumin (C) detection carried out in droplets. Microfluidic devices were filled with media containing 1 to 10 mM glucose (B) or 0 to 100 ng/mL albumin (D). Aliquots of the standard solutions were mixed with glucose or albumin assay reagents and converted into droplets. As seen in the images, magenta color and fluorescent intensity correlated to glucose and albumin concentration respectively. Scale bar: 200 μm. (B, D) Calibration curves demonstrating linear relationship for glucose/albumin and intensity. The limit of detection for glucose was 0.034 mM, and for albumin was 8.2 ng/mL, as determined by the three standard deviations above the blank. Data is represented as mean ±SD of 5 droplets generated at each concentration. (E) Representative micrographs showing detection of glucose (10 mM) and albumin (100 ng/mL) in alternating droplets. Scale bar: 500 μm.

To highlight the high temporal resolution of our device, we set up an experiment to monitor the signal as a function of time for different glucose concentrations (Figure S2A). While our device allows to generate droplets every 10 sec, we chose to produce droplets every minute for this proof-of-concept experiment. As may be appreciated from Figure S2B, the in-droplet enzymatic assay produces a detectable signal (magenta color) within one minute of droplet generation, highlighting the ability for rapid measurements afforded by the device.

The droplet-based bioanalysis unit is not limited to enzymatic assays and can also be configured to quantitate protein secretion. To demonstrate this point, we set up a one-step microbead fluorescence immunoassay (see Figure 5C), where beads fluoresce in the presence of analyte due to the assembly of fluorescent detection Abs (Son et al., 2016). We incorporated several defined albumin concentrations into droplets, together with microbeads and detection Abs. As shown in Figure 5D, on-bead fluorescence increased with albumin concentration, with a detection limit of 8.2 ng/mL, and linear range extending to 100 ng/mL.

Another important feature of our device is the ability to detect both analytes in distinct droplets in parallel. To highlight this feature, we loaded an enzymatic glucose assay and microbead immunoassay for albumin into droplets generated sequentially in the bioanalysis unit. The enzymatic assay and microbead immunoassay reagents were combined with 10 nM glucose and 100 ng/mL albumin. Figure 5E shows a train of droplets in the collection channel where alternating droplets are reporting on glucose and albumin concentration in the solution. The current design allows us to load up to three different assays into distinct droplets. The number of assays may be increased further in the future with a straightforward redesign of the microfluidic device.

3.3. Characterizing changes in glucose metabolism upon hormonal stimulation of hepatic cells

The ability to culture cells and rapidly sample conditioned media afforded by our device is useful for monitoring dynamic physiological processes, for example those occurring in the liver upon stimulation with pancreatic hormones. Hepatocytes are the main epithelial cells in the liver and are responsible for multiple functions including storage and release of glucose upon stimulation with pancreatic hormones. This is a rapid process (tens of minutes) where the liver receives insulin from the pancreas after food intake and upregulates programmes of glucose uptake and storage (see Figure 6A) (Rui, 2014). Conversely, delivery of glucagon from the pancreas during fasting triggers the conversion of glycogen into glucose via glycogenolysis and results in release of glucose by the liver (Klover and Mooney, 2004). Importantly, changes in hormone and glucose levels track with feeding and fasting cycles and change every 4 to 8 h. Our automated microfluidic device is particularly amenable for characterizing such dynamic processes.

Figure 6: Hormonal modulation of glucose metabolism in microfluidic hepatic cultures.

(A) Diagrams describing simplified mechanism of glucose consumption and release in response to pancreatic hormones, glucagon and insulin. Cartoons show expected glucose levels for the three scenarios tested in the paper. Higher intensity of magenta color in a droplet signifies higher level of glucose. (B) Measuring glucose consumption in hepatic spheroid cultures over the course of 10 h. In-droplet color intensity was converted to glucose concentrations using a calibration curve. Live/dead staining confirming that cells were viable at the end of the experiment. Live and dead cells produced green and red fluorescence respectively. Scale bar: 200 μm and 100 μm for droplets and hepatic spheroids respectively. (C) Glucose consumption of hepatic cells in the absence (control) and in the presence of insulin. Cells were exposed to 3 μM insulin for 3 h with droplets generated every 30 min. (D) Hepatic cultures were stimulated with 2 nM glucagon for 3 h with droplet generation and glucose detection performed every 30 min.

Data is represented as mean ±SD of 5 droplets generated at each time point. B-C were conducted in different devices.

Cell seeding in the microfluidic device was accomplished by opening the microvalves to allow the communication between media reservoirs and cell culture chamber via transport channels. During cell seeding, the microvalves controlling media inputs and the droplet generator unit were closed. Hepatic cells (HepG2) introduced into the device settled into microwells and aggregated into spheroids within 2 days of seeding (Figure S3). Compared to HepG2 spheroids in commercial 3D plates, microfluidic spheroids had similar proliferation rates but formed faster (2 days instead of 4, data not shown).

The HepG2 cells used in this study retain a differentiated hepatic phenotype and are frequently used to model liver functions in vitro (Arzumanian et al., 2021). These cells are known to be responsive to pancreatic hormones, including insulin (Iyer et al., 2010). Hormonal stimulation experiments were typically performed 5 days after seeding cells into devices. Glucose consumption by hepatic spheroids was monitored over the course of 10 h by generating 5 droplets every 2 h with in-droplet glucose assay taking 5 min to complete. We note that the device was configured in sampling mode with the cell culture chamber disconnected from the media reservoir. This configuration confined hepatic cells to a volume of 7.25 μL. As shown in Figure 6B, glucose levels in the culture chamber dropped rapidly from 10 to 2 mM in the first 6 h, with a glucose consumption rate of 1.25 mM/h. Subsequently, the consumption rate decreased to 0.31 mM/h for the remaining 4 h. This may be explained by cells switching from glycolysis to a different energy metabolism programme, for example gluconeogenesis, where lactate, pyruvate and amino acids are used as substrates for oxidative phosphorylation (Jose et al., 2011; Rui, 2014). Importantly, hepatic spheroids remained viable during a 10 h period, as shown by Figure 6B (see inset) and Figure S4.

Prior to insulin stimulation, we found it necessary to precondition hepatic cells in 5 mM glucose (compared to 10 mM used by us typically) for 1 h to mimic the preprandial state and equilibrate intracellular glucose. In parallel experiments, we compared glucose consumption by hepatic cells in the presence or absence of 3 μM insulin. As it can be seen from Figure 6C and Figure S5A, glucose consumption rates were similar for hepatic cells with and without insulin for the first 30 min, but the rates diverged thereafter, with insulin-stimulated cells consuming glucose at a higher rate (0.72 vs. 1.05 mM/h for the 30 min to 180 min segment). This delay in glucose consumption may be attributed to the time required for the cell to inhibit glycogenolysis and activate glycogen synthesis machinery.

During fasting, glucose levels in the blood stream decrease, stimulating the pancreas to release glucagon. This hormone is then delivered to the liver via the portal vein and triggers a programme of glycogenolysis in hepatocytes, whereby glycogen is converted into glucose and then released into the blood stream (see Figure 6A) (Klover and Mooney, 2004). We mimicked this process by exposing hepatic cultures in our microfluidic device to 2 nM glucagon in DMEM with 5 mM glucose. Figure 6D shows that hepatic cells responded to glucagon by secreting glucose, which went up two-fold within 30 min of stimulation. Subsequently, glucose was consumed at a rate exceeding that of control cells that did not receive glucagon. Importantly, stimulation with a lower concentration of glucagon (1 nM) produced a similar pattern of glucose release followed by reuptake but at lower levels. (see Figure S5B). Similar patterns of glucose release and reuptake after glucagon stimulation have been reported in the literature previously with possible explanations for this behavior being - desensitization of glucagon receptor function, inhibitory effects of hyperglycemia on hepatic glucose production, and stimulatory effects of hyperglycemia on hepatic glucose uptake (Cherrington et al., 1981; Leturque et al., 2009). Overall, experiments outlined in Figure 6 are important because they describe hepatic responses to pancreatic hormones that mirror physiological pancreas-liver interactions.

3.4. Assessing effects of lipotoxic injury on hormonal responsiveness of hepatic cells.

The liver plays a major role in lipid metabolism and may be injured by exposure to a high fat diet, resulting in hepatic steatosis (Parthasarathy et al., 2020). Prolonged exposure to a high fat diet may lead to non-alcoholic steatohepatitis (NASH) – a disease characterized by high levels of oxidative stress, liver inflammation, fibrosis, and cell death. There is also considerable clinical evidence pointing to the co-occurrence of NASH and diabetes, suggesting that pancreas-liver crosstalk is important for both diseases (Noureddin and Rinella, 2015). One pathological mechanism that connects NASH and diabetes is increased hepatic insulin resistance, where hepatocytes experience lipotoxicity and are less capable of responding to insulin by taking up glucose. Because of this, postprandial glucose levels remain high and continue to stimulate pancreatic insulin production, which, over time, may lead to pancreatic dysfunction and Type 2 diabetes (Gastaldelli and Cusi, 2019).

We wanted to explore the effects of lipotoxic injury on hepatic insulin resistance. Hepatic cells in a microfluidic device were exposed to palmitate (PA), a free fatty acid, over the course of 48 h to induce lipotoxicity (Hirsova et al., 2016). The extent of injury was assessed using an LDH assay and live/dead staining for cytotoxicity, as well as Oil Red O staining for lipid accumulation in hepatic spheroids. Figure 7A highlights the fact that 2 days of injury was associated with only moderate levels of cytotoxicity, while resulting in significant lipid uptake. The fact that injured hepatic cells remained viable and functional is underscored by the LDH assay and live/dead staining in Figure 7B-C and Figure S6 and albumin synthesis in Figure 7D. Hormonal stimulation experiments, shown in Figure 7E were carried out as described in the previous section. Cells were preconditioned in 5 mM glucose and exposed to 3 μM insulin over the course of 3 h while making glucose measurements every 30 min. Results in Figure 7E highlight the fact that injured cells do not consume glucose and do not respond to insulin. Note that control hepatic spheroids had a similar glucose consumption in the absence of insulin as in Figure 6C. Given the fact that PA-injured spheroids are viable (see Figure 7F and Figure S6 for live/dead staining) and functional (as evidenced by continued production of albumin), they appear to have switched from glucose to an alternative source of energy. There is, for example, evidence of lipid loading causing hepatocytes to switch to oxidation of fatty acids as a primary energy source (Holland et al., 2008). Additional evidence suggests that upon uptake by a hepatocyte, palmitate may be converted to ceramide. This in turn may block glycolysis and glycogen synthesis by inhibiting Akt/PKB signaling (Konstantynowicz-Nowicka et al., 2015; Zhu et al., 2023). Figure 7G summarizes mechanisms that may explain lack of insulin sensitivity and glucose uptake in hepatic cultures injured by palmitate.

Figure 7: Using a microfluidic automated device to assess effects of lipotoxicity on glucose metabolism in hepatic spheroids.

(A) Lipid accumulation after 48 h exposure with 0 and 800 μM PA was confirmed by staining for lipid droplets in hepatic spheroids using Oil Red O. Scale bar: 20 μm. (B) The cytotoxicity of PA loading was confirmed by LDH assay. (C) The hepatic function was assessed by albumin ELISA and was demonstrated to be similar for injured and normal hepatic spheroids. (D) Live/dead staining confirming viability on day 4. Scale bar: 100 μm. (E) Assessing glucose metabolism of lipid-loaded spheroids in the presence and absence of insulin. Cells were exposed to 3 μM insulin for 3 h with droplets generated every 30 min. (F) Live/dead staining confirming viability before and after the glucose measurement experiment. Scale bar: 100 μm. (G) Mechanism of palmitate-induced insulin resistance in hepatocytes. Cartoon below the diagram describes glucose levels observed after lipid injury and stimulation with pancreatic hormones. Higher intensity of magenta color in a droplet signifies higher level of glucose.

4. CONCLUSIONS

In this paper, we described a microfluidic device that combines a cell spheroid culture unit with a droplet generator for sampling of conditioned media and analysis of cell function. The use of computer-controlled pneumatic microvalves integrated with cell culture and bioanalysis units enabled: 1) generating droplets with high temporal resolution (every 10 s), 2) sampling media in the culture chamber without dilution of the culture volume (5 droplets used for analysis at each time point constituted 0.25% of the culture chamber volume), 3) performing enzymatic assay and immunoassay in sequentially generated droplets, and 4) dynamically switching inputs in the cell culture chamber. The droplets were used as ‘mobile’ nano-wells to carry out detection of glucose and albumin.

We demonstrated a biological application of this device by dynamically measuring hepatic uptake and release of glucose in response to stimulation with the pancreatic hormones, insulin and glucagon. Such experiments are challenging to perform using standard cell culture and analysis tools. Our results have physiological/biological significance as they provide hepatic glucose consumption rates in response to stimulation with insulin and glucagon and demonstrate that lipid injury affects glucose metabolism and responsiveness to these hormones. In the future, our device may enable quantitative physiology of hormonal stimulation, may integrate pancreatic cells in addition to hepatocytes, and may be used to test poly-pills targeting liver and pancreas.