Effect of Fluidic Configuration on PK Studies: Flow-through or Recirculating Tissue Chips

We investigated the theoretical effects of fluidic configurations, recirculation vs flow-through, on drug depletion kinetics in tissue chips. Using quantitative systems pharmacology (QSP)–based algorithms, two hypothetical tissue chips are assumed to have identical hepatic activity and tissue culture chamber dimensions, but one with flow-through configuration and the other with recirculation. After administering 1 µM drug to each configuration, drug depletion kinetics were simulated, and drug profiles plotted for samples collected from media reservoirs (Fig. 1). In flow-through configuration, the drug concentration in the media reservoir increases and plateaus within 2 h. The magnitude of the plateau value is inversely proportional to clearance rate, i.e., the higher the intrinsic clearance rate, the lower the plateau value. In recirculation configuration, the simulated drug profiles were a function of mono-exponential decay. Drug profiles with a range of clearance values were simulated and an arbitrary low threshold value, which is a minimum of 10% parent drug depletion, was assumed to estimate the lowest detection limits in each configuration for comparison. The lowest clearance value in the flow-through system was 5 µL/min/million cells while in recirculation for 8-day drug incubation; the value that results in at least a 10% drug depletion was 0.08 µL/min/million cells. This demonstrates the sensitivity of recirculating tissue chips to observe wide clearance range and its importance to study kinetics of both low- and high-clearance compounds.

Fig. 1

Modeling the effect of recirculation on drug metabolism studies. Schematic of a flow-through and b recirculation chip design along with the kinetic profile simulated for 1 μM parent drug in milli-fluidic LTC without and with recirculation for three hypothetical compounds spanning a wide range of clearances over 6 h and 192 h respectively

Recirculating Milli-fluidic LTC and LTC Controllers

We developed a recirculating milli-fluidic liver tissue chip with an on-board pumping system (Fig. 2a). The milli-fluidic design incorporates larger tissue size, media volume, and overall channel dimensions than microfluidic chips. LTC has a 1-cm2 cell culture area to accommodate 200,000–250,000 PHH in monolayer for measurable drug metabolism and tissue-based analysis and 1.5–2 mL media volume to allow repeated sampling for drug quantification. Additionally, this design allows multi-scale (media- and tissue-based) PK characterization on a single chip to generate reproducible kinetic and endpoint data. The chip footprint mirrors a standard microtiter plate, making it compatible with standard laboratory equipment.

Fig. 2

Liver tissue chip (LTC) platform and biomimetic design description. a Architecture and key features of the liver tissue chip. b Features of four-plex controller with LTC connected to flow. c Schematic of recirculation pathway with d top view of the LTC. e Isometric and front sectional view of oxygenation chamber to illustrate media being reoxygenated in the chamber where sampling is carried out during experiment. Computational fluid dynamics (CFD) simulations of LTC culture chamber representing f a uniform shear profile throughout the tissue achieved by the design parameters in the culture chamber. g The simulation of O2 tension in liver culture chamber that are within the physiologically relevant levels

LTC has two chambers: a cell culture chamber designed for PHH cultures (mono- or co-cultures) and an oxygenation chamber to re-oxygenate O2-depleted medium from the culture chamber that are connected through closed-loop channels and on-board pump system allowing media recirculation (Fig. 2b, c, d). Here, the oxygenation chamber allows passive gas exchange between ambient air and air in the oxygenation chamber through an O2-permeable silicone membrane. As such, there is no contact between the membrane and recirculating media (Fig. 2e), thus avoiding silicone-related NSB of drugs. LTC is designed to mitigate bubble formation issue, which is a challenge in microfluidic systems that has detrimental effects on cells (12,13,14). Additionally, the recirculating medium can be accessed with a standard single-channel pipette via the sampling port located on the lid of the oxygenation chamber allowing collection of multiple time points samples, e.g., seven time-points of 50 µL sample volume. The culture chamber is designed with a removable lid that allows direct access to the cell compartment during seeding, longitudinal experiments, and endpoint assays.

The media recirculation was driven by an on-board piezoelectric diaphragm pump eliminating the need for external pumps, reservoirs, and tubing connections. The pump is actuated by an on-board flow sensor that measures the flowrate in real-time and is actively controlled by a four-plex controller which is connected via a ribbon cable (Fig. 2b). The controller regulates the recirculation with a continuous feedback loop between the pump and flow sensor to maintain constant and accurate flow. Each chip is controlled independently, and the flow rates can be set with the touch-screen panel. This is a stand-alone and user-friendly platform that does not require any additional external support or computer connection and up to four controllers were used in an incubator.

Computational Investigation of Optimal Recirculation Flowrate in LTC

High oxygen consumption and low shear tolerance are two essential characteristics of primary hepatocyte cultures (15,16,17,18). In LTC, the culture chamber is designed to provide uniform shear stress (Fig. 2f) by continuous perfusion of the hepatic culture at appropriate flowrate. As the O2 in media entering the cell chamber is consumed by cells, the oxygen-depleted media leaving the chamber gets reoxygenated in the oxygenation chamber through gas exchange. These phenomena were modeled with computational fluid dynamics (CFD) simulations to determine optimal flowrate to bring sufficient O2 to the culture, while maintaining low shear (<0.005 dyne/cm2). The calculated O2 tension on the hepatic tissue is similar to physiologically relevant O2 concentrations observed in the human liver (45–50 mmHg in periportal zone 1 and 15–20 mmHg in perivenous zone 3) (19, 20). CFD-based O2 concentration estimates for LTC were 50–60 mmHg at the inlet and 15–25 mmHg at the outlet (Fig. 2g). The experimental validation of on-chip O2 concentration calculated by CFD will be valuable in future research. Through simulations and empirical experiments, 2 mL/h flow rate was determined to achieve sufficient oxygenation, appropriate nutrient delivery to the hepatic tissue, efficient removal of cellular waste, optimal shear stress, and uniform drug distribution in LTC. In addition, the features in chamber inlet minimize the shear from high flow rates to hepatocyte tolerable shear range that helps to maintain the improved morphology and metabolic activity during long-term culture.

Non-biological Characterization of LTC

In PDMS microfluidic chips, NSB drastically increases with increasing lipophilicity (18), and the absorption increases with the hydrophobicity and topological polar surface area of the small molecule (21, 22). In contrast, COC is a relatively hydrophilic material vs. PDMS with a higher surface energy. To validate minimal NSB in LTC, small molecule compounds at 0.1–1 µM initial concentrations were recirculated on fully assembled tissue chips without hepatic tissue for 24 h; samples were collected at 0 and 24 h and quantified by LC-MS/MS. Twenty-one compounds of varying ionic state and lipophilicity (logP) ranging from −1.04 to 6.08 were tested, and near complete recovery of neutral and basic compounds was achieved, while the high logP acidic drugs (bosentan, pitavastatin, fluvastatin, and repaglinide) was recovered >60% (Fig. 3).

Fig. 3

Non-specific binding (NSB) characterization of liver tissue chip. NSB of various small molecules were determined by incubating 0.1–1 μM drug concentration for 24 h under recirculating flow without hepatic tissue. LTC exhibited very minimal NSB, where 17 out of 21 drugs were recovered over 85% of their initial concentration

Evaporation can be a concern for drug studies, usually in micro-bioreactor-type tissue chips in which the media is exposed to ambient air (23,24,25). In such systems, continuous perfusion may contribute to higher evaporation which becomes critical for long-term drug incubations, especially when the evaporation rate is in similar order of magnitude to clearance rates (26). Due to COC’s material property, the evaporative media losses evaluated over a week with no media replenishment were negligible <0.4% per day or <2.8% over a week (Figure S1).

Biological Characterization of LTC

To demonstrate metabolically active long-term hepatic cultures and polarized morphology, which are essential for PK applications, PHH were seeded as a SCH with collagen I and fibronectin as underlay and Matrigel® GFR as overlay to allow polarized hepatocytes for maintenance of differentiated morphology, stabilized metabolic activity, and bile canaliculi formation. After a day, the culture was maintained under continuous recirculation (2 mL/h) for the rest of the experiment and performed partial media change every 2–3 days to allow media conditioning (27,28,29). Live imaging revealed that the differentiated morphology of characteristic cobblestone pattern maintained for at least 15 days (Fig. 4a). LTC was live stained for formation of bile canaliculi using CMFDA (Fig. 4b), which was observed at the junction of cells confirming polarization. IHC staining showed canalicular membrane expressing multidrug resistance–associated protein 2 (MRP2), an apical bile acid transporter. The tissue stained positive in cytoplasm for CYP3A4, and epithelial cytoskeleton marker, CK18, which is highly concentrated in hepatocytes (30, 31) and E-cadherin, an epithelial marker expressed in the adjacent cell boundaries of differentiated hepatocytes as a cobblestone pattern. 594-Phalloid stained the high-density F-actin filaments that are arranged in the apical side of hepatocytes and localized in the bile canaliculi bordering hepatocytes.

Fig. 4

Biological characterization of LTC with morphology imaging and biochemical assays. a Long-term hepatocyte morphology imaged using bright field microscope on days 3, 7, 11, and 15 showing phenotypic cobblestone morphology. b Immunocytochemistry staining on day 7 showing presence of E-cadherin, F-Actin, CYP3A4, CK-18, MRP-2, and bile canaliculi using CMFDA staining. c Albumin and d urea levels were assessed in LTC daily from days 2 to 15 showing sustained functional activity of the hepatic culture

Albumin and urea production rates, which are commonly used to evaluate in vitro functionality of hepatocyte cultures, were measured throughout the 15-day culture period (Fig. 4c, d). The hepatic culture in LTC has comparable albumin and urea production to estimated human liver production (37–105 μg/day/million cells for albumin production and 56–159 μg/day/million for urea production) (7). The initial increase in albumin levels can be attributed to gradual remodeling of the ECM in the tissue microenvironment stimulated by recirculation of the conditioned media. While the albumin production rate steadily increased, the urea production rate decreased slowly over 15 days. Gene expression studies (Figure S2) also showed a progressive increase in albumin mRNA levels (>2.5-fold) in the 15-day culture, following similar trends as albumin production.

The metabolic activity of hepatocyte culture was quantified by assessing CYP isoform-specific metabolite formation at days 3, 8, and 13 using CYP-probe cocktail (Table S1). CYP1A2, CYP2C9, CYP2C19, CYP2D6, and CYP3A4 retained or increased activity for up to 13 days of culture (Fig. 5a). The quantified drug metabolizing Phase 1 enzymes (CYP Isoforms, CES1 and CES2), Phase 2 (UGTs, GST), and transporter (SLCs, ABCs) genes using RT-PCR maintained their expression levels up to 15 days, with varying degrees of fold-change (Fig. 5b). LTC was validated using an additional donor to demonstrate the platform’s robustness in long-term culture characterization, which also resulted in sustained functional (albumin and urea production) and CYP activity levels (Figure S3, S4).

Fig. 5

Activity and gene expression levels of cytochrome P450 isoforms in liver tissue chip. a The relative activity levels and stability of CYP1A2, CYP3A4, CYP2C9, CYP2D6, and CYP2C19 CYP isoforms were measured by cocktail CYP probe substrate analytical assay. b Fold change expression levels of PK-relevant genes were measured by TaqMan RT-PCR on days 7, 11, and 15 relative to day3

To demonstrate the biological reproducibility of LTC, we quantified the coefficient of variance (CV%) for albumin and urea levels in four independent experiments on days 3, 8, and 10 (Figure S5a, b). Compared to current MPS models, albumin and urea data show < 28.5% variability across chips and <29.5% between experiments and operator, leading to a robust and reproduceable platform (Figure S5c).

Here, we demonstrated that LTC’s PHH culture maintained their polarized morphology and functional markers comparable to in vivo for at least 15 days, while the long-term culture retained metabolic activity measured as both drug-metabolizing enzyme levels and gene expression profiles of PK-relevant genes

Effect of Culture Age and Drug Incubation Window on Intrinsic Clearance Estimation in LTC

Intrinsic clearance values were evaluated by incubating cocktails of four high-clearance compounds (midazolam, propranolol, diclofenac, and dextromethorphan) on days 3, 5, or 7 of culture for 72 h and two low-clearance compounds (alprazolam and tolbutamide) on days 3 and 7 for 120 h. High-clearance drugs were significantly depleted over 72 h, whereas low-clearance drugs took up to 120 h for 20% and 50% parent drug depletion, respectively (Fig. 6). The intrinsic metabolic clearance (\(C_\left(u\right)}\)) values for each drug and drug incubation window (Table I) show high reproducibility with average CV% of 18% and no statistically significant differences in variance or AUC for any compounds. However, \(_}\) for diclofenac (p = 0.052) and propranolol (p = 0.11) slightly decreased across culture windows which may be related to changes in CYP2C9 activity levels and gene expression over time (Fig. 5a, b) and other related metabolizing enzymes. Overall, LTC produces very consistent drug depletion profiles and \(_}\) estimates with minimal chip-to-chip and day-to-day variability. LTC predicted clinical clearance values for slow cleared drugs (alprazolam and tolbutamide) were more accurate than clinical predictions obtained from static culture. For the faster cleared drug (Diclofenac), clearance values obtained either from LTC or static cultures were similar (see Table S9). Thus, chips can be potentially used for repeated drug studies for low and high-clearance compounds at least for 12 days of culture and has flexibility regarding the culture window used for drug depletion studies of high and low cleared drugs.

Fig. 6

Effects of culture age and drug incubation on intrinsic clearance studies in liver tissue chips. Both high clearance drugs (a Midazolam, b Propranolol, c Diclofenac, and d Dextromethorphan) and low clearance drugs (e Alprazolam and f Tolbutamide) were evaluated on LTC at different incubation window for 3 and 5 days of incubation period, respectively

Table I Intrinsic Metabolic Clearances (\(_\) [μL/min/million cells]) Calculated from LTC with Different Incubation WindowsDrug Interaction Study on LTC Using Rifampicin-Mediated Induction and Itraconazole-Mediated Inhibition

For induction, LTCs were dosed with 10 mM rifampicin on day 3 and incubated for 3 days. Induction increased metabolic activity of CYP2C9, CYP3A4, and CYP2C19, while CYP2D6 activity was unchanged (Fig. 7a). Additionally, mRNA analysis showed differential effects of induction on various Phase 1 drug-metabolizing enzymes, conjugative UGT enzymes, and efflux Transporters (ABCs) (Fig. 7b).

Fig. 7

Effects of rifampicin induction and itraconazole inhibition on CYP enzyme activity and mRNA expression of PK relevant gene. a Cocktail CYP probe substrate analytical assay and b mRNA expression levels as measured by TaqMan RT-PCR showing rifampicin-mediated induction of multiple CYP enzyme activity and expression levels in LTC. The data is expressed as average fold change expression levels of the induced group activity levels are relative to the DMSO vehicle control group. c Itraconazole-mediated inhibition effects measured using P450-Glo™ CYP3A4 assay showing average fold difference compared to DMSO vehicle control group

To inhibit CYP3A4 activity, LTCs were dosed with 3 mM itraconazole on day 3 and incubated for 72 h. The enzymatic activity of CYP3A4 was inhibited 70% compared to DMSO control (Fig. 7c).

Estimation of Intrinsic Clearance Values of Small Molecule Drugs and IVIVC

A range of compounds with varying clearance values, physicochemical properties, and metabolizing enzyme-specificity were assayed in LTC to estimate on-chip clearance values (Table S10). Observed intrinsic clearances spanned 3 orders of magnitude from 0.15 μL/min/million cells for rosuvastatin to 115 μL/min/million cells for repaglinide (Fig. 8). Diclofenac and S-warfarin, both metabolized by CYP2C9, showed intrinsic clearances of 81.8 μL/min/million cells and 0.82 μL/min/million cells, and raloxifene and zidovudine, both metabolized by UGTs, showed intrinsic clearances of 70.6 μL/min/million cells and 1.94 μL/min/million cells, showing differential clearance values for both CYP and UGT substrates. For all drugs, the standard deviation across chips was small compared to the mean (13% median CV; 16% mean CV, excluding rosuvastatin, which had 440% CV and the lowest CLint value).

Fig. 8

Unbound intrinsic clearance (\(_(u)}\)) for 16 compounds estimated from LTC. \(_(u)}\) of 16 compounds spanning 3 orders of magnitude from 0.15 to 115 μL/min/million cells estimated from LTC with error bars showing standard deviations across chips. The bar color indicates whether the compound is an acid (orange), base (blue), or neutral (gray)

To evaluate the clinical relevance of the clearance values, in vitro intrinsic clearance values from LTC were scaled to predicted human clearance values using human hepatocellularity and the parallel tube model (PT, Table S11). Figure 9 a shows the IVIVC of intrinsic clearance of ten drugs for which clinical intrinsic clearance values can be estimated. IVIVC of these drugs showed average absolute fold error (AAFE) of 2.7, and 40% of the predictions were within 2-fold of clinically derived values and were particularly well-predicted for drugs with low intrinsic clearance.

Fig. 9

In vitro in vivo correlation (IVIVC) of LTC predicted clinical clearance estimates to clinically observed values for a intrinsic metabolic clearance (\(_}\)) and b total hepatic clearance (\(_\)). LTC estimates of human \(_}\) were obtained by scaling the observed in vitro LTC \(_}\) according to human hepatocellularity. LTC estimates of human \(_\) were obtained by further application of the parallel tube model. c The table shows average absolute fold errors calculated across drugs with different clearances shown in 9B. The in vivo human hepatic clearance for drugs are classified as low clearance (\(_)}\) < 5 ml/min/kg), medium clearance (5 < \(_)}\)< 15 ml/min/kg), and high clearance (\(_)}\)> 15 ml/min/kg)

After applying the PT model, which accounts for blood flow and protein binding, the in vivo hepatic clearance (CLH) due to metabolism is better predicted (Fig. 9b). For the 12 drugs cleared primarily hepatically, the LTC-predicted in vivo hepatic clearance values with AAFE of 1.7, and 83% of the predictions were within 2-fold of clinical values and comparable results were obtained using the well-stirred model (1.8 AAFE; see Table S10). The in vivo hepatic clearance was best estimated for drugs with low-clearance (S-warfarin, tolbutamide, alprazolam, and theophylline). AAFE for low-, medium-, and high-clearance drugs were 1.1, 1.7, and 3.7, respectively (Fig. 9c). Most high-cleared compounds were predicted within twofold of clinical values except zidovudine and dextromethorphan. Additionally, for few compounds with detectable NSB (>10% depletion over 24 h), a computational correction was applied to account for drug depletion due to NSB, which had minimal impact on the estimate: with 16% on average shift with NSB correction. These results demonstrate that the LTC generates highly reproducible drug depletion kinetics and accurate estimates of clinical clearance values for diverse set of compounds.