Lab on a Chip

View Article Online

PAPER
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

View Journal

Simulating drug concentrations in PDMS

Cite this: DOI: 10.1039/d1lc00348h
microfluidic organ chips†
Jennifer Grant, a Alican Özkan, a Crystal Oh,a Gautam Mahajan, a

Rachelle Prantil-Bauna and Donald E. Ingber *abc

Microfluidic organ-on-a-chip (Organ Chip) cell culture devices are often fabricated using
polydimethylsiloxane (PDMS) because it is biocompatible, transparent, elastomeric, and oxygen permeable;
however, hydrophobic small molecules can absorb to PDMS, which makes it challenging to predict drug
responses. Here, we describe a combined simulation and experimental approach to predict the spatial and
temporal concentration profile of a drug under continuous dosing in a PDMS Organ Chip containing two
parallel channels separated by a porous membrane that is lined with cultured cells, without prior
knowledge of its log P value. First, a three-dimensional finite element model of drug loss into the chip was
developed that incorporates absorption, adsorption, convection, and diffusion, which simulates changes in
drug levels over time and space as a function of potential PDMS diffusion coefficients and log P values. By
then experimentally measuring the diffusivity of the compound in PDMS and determining its partition
coefficient through mass spectrometric analysis of the drug concentration in the channel outflow, it is
possible to estimate the effective log P range of the compound. The diffusion and partition coefficients
were experimentally derived for the antimalarial drug and potential SARS-CoV-2 therapeutic, amodiaquine,
and incorporated into the model to quantitatively estimate the drug-specific concentration profile over
Received 21st April 2021, time measured in human lung airway chips lined with bronchial epithelium interfaced with pulmonary
Accepted 26th July 2021
microvascular endothelium. The same strategy can be applied to any device geometry, surface treatment,
DOI: 10.1039/d1lc00348h
or in vitro microfluidic model to simulate the spatial and temporal gradient of a drug in 3D without prior
knowledge of the partition coefficient or the rate of diffusion in PDMS. Thus, this approach may expand
rsc.li/loc the use of PDMS Organ Chip devices for various forms of drug testing.

Introduction elastomeric polymer.3,11 However, PDMS can strongly adsorb

some hydrophobic small molecules from solution as well as
Organs-on-chips (Organ Chips) are microfluidic culture accumulate these molecules into its bulk volume. Together,
devices lined by living human cells that reconstitute human adsorption and absorption can decrease the effective dosing
organ-level structures as well as physiology and concentration, making it difficult to predict actual drug
pathophysiology. Their ability to model many features of a concentrations experienced by the cells. Thus, development of
functional human organ, such as the response to viral a method to overcome this limitation could benefit the entire
infection,1,2 bacterial infection,3,4 disease states,5–9 and Organ Chip field.
clinically relevant drug exposure pharmacokinetic (PK) The logarithm of the octanol–water partition coefficient
profiles,10 make Organ Chips particularly useful for drug (log P) is frequently used to approximate the extent of PDMS
development and toxicity studies. Organ Chips are often absorption, where compounds with high log P have a greater
fabricated out of poly-dimethylsiloxane (PDMS) because it is tendency to bind to or be absorbed by PDMS. Partition
an oxygen permeable, transparent, biocompatible, and coefficients have been experimentally determined using
analytical techniques such as fluorescence,12 UV-vis,13 or IR
a
Wyss Institute for Biologically Inspired Engineering, Harvard University, Boston, spectroscopy,13 and computationally predicted with
MA 02115, USA. E-mail: don.ingber@wyss.harvard.edu quantitative structure–activity relationship (QSAR)14,15 or
b
Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard molecular dynamics (MD)16 models. However, the partition
University, Cambridge, MA 02138, USA
c
coefficient alone cannot predict the extent of binding, and
Vascular Biology Program and Department of Surgery, Harvard Medical School
and Boston Children's Hospital, Boston, MA 02115, USA
past work suggests that it should be combined with the
† Electronic supplementary information (ESI) available. See DOI: 10.1039/ topological polar surface area (TPSA)12,17 or the number of
d1lc00348h H-bond donors13 to improve the accuracy of bioavailability

This journal is © The Royal Society of Chemistry 2021 Lab Chip

 View Article Online

Paper Lab on a Chip

predictions in PDMS-based microfluidic devices. Although which can bias the extent of absorption. Thus, there is a clear
physiochemical descriptors are useful for describing the need for a more comprehensive model that simulates three-
extent of compound loss into PDMS devices, they cannot be dimensional (3D) drug concentrations in chips lined with
used to quantitatively estimate drug loss because it is a cultured cells and incorporates all physical processes that
dynamic process that depends strongly on a variety of factors contribute to drug loss. This is necessary to quantitatively
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

including absorption, convection, diffusion, medium estimate drug concentrations inside the chip throughout an
composition, dosing time, cell type, and surface treatment, in exposure period and thereby guide dosing strategies, evaluate
addition to adsorption. toxicities, and analyze pharmacokinetic properties.
A more comprehensive understanding of drug loss can be Here, we describe a combined experimental and
obtained by combining computational models and computational approach that simulates spatial and temporal
experimental data acquired directly from the PDMS Organ drug concentration profiles in 3D under continuous dosing in
Chip devices used for in vitro culture. A one-dimensional (1D) two-channel, microfluidic, Organ Chips lined with cultured
simulation model has been developed that predicts drug loss living cells. This strategy involves the development of a finite
in a PDMS device by combining computational fluid element simulation of drug absorption in a PDMS Organ Chip
dynamics (CFD) with drug-PDMS binding kinetics, which as well as experimental quantification of the diffusion and
suggested that drug loss can be minimized by increasing the partition coefficients of the drug. The computational model is
flow rate, decreasing channel height, or decreasing channel then used to estimate drug concentrations that the cells
width.13 A PDMS adsorption compartment was incorporated experience at any time within the microfluidic channels of the
into a mathematical pharmacokinetic/pharmacodynamic (PK/ chip. We applied this method towards simulating levels of the
PD) model of terfenadine and fexofenadine in a fluidically antimalarial drug amodiaquine when administered continuously
connected heart-on-a-chip and liver-on-a-chip system.18 These under flow in human lung airway chips, which recently resulted
approaches demonstrate the potential to estimate the in identification of this drug as a potential SARS-CoV-2
concentration of the drug in microfluidic culture devices by therapeutic.1 This strategy can estimate compound loss due to
combining experimentally derived constants with PDMS absorption in any device composition, and thus, it should
computational simulations of physical processes. However, help to improve experimental design and analysis of dose–
they do not provide a way to simulate how drug response efficacy and toxicity studies in PDMS Organ Chips.
concentrations vary over time and space within microfluidic
Organ Chip devices that contain two parallel channels Results
separated by a membrane lined by living cells, which have
been shown to faithfully recapitulate the physiology and Computational modeling of drug concentrations in two-
pathophysiology of many different human organs.10,19 Two- channel PDMS Organ Chips
channel microfluidic Organ Chips can recapitulate dynamic The mass transport of a drug in a microfluidic Organ Chip is
pharmacokinetic (PK) profiles that are observed in vivo controlled by diffusion, convection, dissolution, dose, device
because the drug is perfused through endothelium-lined geometry, temperature, and pressure.21 We carried out 3D
vascular channels and physiologically relevant tissue-tissue simulations that incorporate each of these factors to
interactions permit analysis of clinically relevant drug fluxes understand the drug distribution in the microfluidic
between compartments. We have previously shown that PK channels. The Organ Chip we model is a commercially
parameters can be estimated in a vascularized human bone available PDMS microfluidic culture device that contains two
marrow-on-a-chip (BM chip) by fitting a three-compartment parallel fluidic channels separated by a porous PDMS
PK model to LC-MS/MS-quantified drug concentrations from membrane, which is lined by living epithelium on one side
the chip outflows.20 It is possible to increase the accuracy of and endothelium on the other (Fig. 1a). A SOLIDWORKS
the PK model and in vivo extrapolations by accounting for rendering of the entire PDMS Organ Chip geometry was
drug loss due to PDMS absorption. A PK model of a human- imported into COMSOL Multiphysics software and fluidic
body-on-chips system incorporated drug loss due to PDMS domains were assigned to the apical and basal channels
absorption by performing mass spectrometric analysis of the (Fig. 1b). The height of the epithelium and endothelium (3.9
drug concentration in the outflows of blank chips and μm and 1 μm, respectively) was experimentally determined
incorporating the fractional loss into the model.10 Although from microscopic cross-sections of human lung large airway
this model quantitatively predicted PK parameters for orally chips lined by primary human bronchial epithelium
and intravenously administered drugs, it cannot be used to interfaced with primary human pulmonary microvascular
accurately simulate the spatial and temporal concentration of endothelium and incorporated into the total height of the
the drug inside the chip throughout the dosing period central membrane; the membrane pores that represent 3% of
because the model did not consider diffusion of the drug the total membrane area were removed to increase
into bulk PDMS and it simplified the geometry of the Organ computational efficiency.
Chip into two-dimensional (2D) domains. Additionally, the We used the human lung airway chip as a starting point
loss due to PDMS absorption was performed by mass to develop this modeling approach because we had found
spectrometric analysis of blank chips without cultured cells, that the antimalarial drug, amodiaquine, significantly

Lab Chip This journal is © The Royal Society of Chemistry 2021

 View Article Online

Lab on a Chip Paper

partition coefficient was implemented for all medium-PDMS

interfaces in the Organ Chip geometry. After adsorption
onto the wall surface, the drug diffuses into the bulk
polymer down a concentration gradient at a rate described
by Fick's second law of diffusion. In our model, transport of
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

this small hydrophobic molecule through the porous

membrane and living tissue-tissue interface was assumed to
be passive and simplified to follow Fick's second law. We
simulated continuous perfusion of a small molecule with a
size similar to that of amodiaquine (∼400 Da) at 1.24 μM,
under continuous flow (60 μL h−1) through the basal
vascular channel to mimic systemic dosing after oral
Fig. 1 Design of a two-channel human Organ Chip. a) Side-view administration; the drug was also maintained under static
schematic of the microfluidic chip with epithelium lining the top of the conditions in the apical channel, as carried out in the
horizontal membrane that separates the upper and lower channels,
published study.1 A time-dependent simulation was
interfaced with endothelium that is cultured on its basal surface. Drug
is dosed at a concentration (ci) in the endothelium-lined channel under performed for 12 h with the temperature, flow rate,
continuous perfusion to mimic systemic distribution after oral atmospheric pressure, diffusion coefficient of drug in
administration (image generated with BioRender). b) The entire Organ medium, diffusion coefficient of drug in PDMS, and
Chip geometry was modeled in COMSOL. The Organ Chip has two partition coefficient being held constant.
central fluidic channels (blue apical channel and red basal channel)
First, we explored the effect of varying the partition
that are separated by a porous membrane. The fluidic channels are
positioned between two full height vacuum chambers (grey), although coefficient on the distribution of the small molecule when
they were not used to apply cyclic strain during the culture of these flowed through the basal channel of the Organ Chip for 12 h.
large airway cells. The two channels and vacuum chambers all have The simulation was performed with P = 100 and 400 (log P = 2
inlets and outlets that vertically rise to the upper surface of the chip. c) and 2.6, respectively) and Dpdms = 1 × 10−13 m2 s−1, which is the
Side-view image of the simulated Organ Chip with the inset showing
approximated diffusion coefficient for a compound with
the apical channel, the central membrane lined with epithelium on its
upper surface and endothelium on its underside, and the basal fluidic molecular weight ∼400 Da.24–26 3D surface heat maps of the
channel. Arrows indicate the flow direction. drug concentration in the apical and basal channel show how
the drug accumulates in both fluidic channels throughout the
simulated dosing period (Fig. 2a). 2D heat maps of a vertical
inhibits infection of the epithelium by a pseudotyped SARS- cross section through the center of the chip in the xz axis show
CoV-2 virus when flowed through the vascular channel of the that absorption to the channel wall generates a concentration
device at a clinically relevant dose of 1.24 μM (the maximum gradient that radially extends outwards from the center of the
concentration or Cmax of amodiaquine in human blood), and basal channel with lower concentrations along its lateral
its ability to inhibit infections by native SARS-CoV-2 was sidewalls (Fig. 2b). The concentration gradient in the basal
confirmed in vitro and in animal studies.1 Amodiaquine is channel is anisotropic because the 90° corners generate a non-
known to participate in hydrophobic interactions with lipids uniform velocity distribution (Fig. S1†). These models reveal
and proteins,22,23 and thus, there was a good chance that it that the drug with P = 400 takes more time to saturate the basal
might adsorb to the walls of PDMS channels. So, we wanted channel because it rapidly absorbs into the channel walls. At
to determine the effective drug concentration that the cells both simulated P values, drug enters the membrane from the
were exposed to in this microfluidic chip under the basal channel and begins to diffuse into the apical channel
experimental conditions we utilized; however, we did not medium (Fig. 2b and S2†). 2D heat maps through the center of
know the log P for this drug in the Organ Chip environment the basal channel along the xy plane show that a concentration
with cultured cells and extracellular matrix. Thus, to fully gradient also exists along the fluidic stream because adsorption
characterize the concentration of amodiaquine in the lung depletes the drug from the medium and lowers the downstream
airway chip, we set out to develop a combined experimental concentration along the length of the channel (Fig. 2c).
and computational approach that simulates the Together, these data emphasize that drug absorption and
concentration of the drug in the chip over time. adsorption are spatially and temporally dependent processes
A drug must first adsorb to the channel wall before it and both influence the concentration of drug dose exposure
enters into the bulk PDMS. Dissolution of a drug from the experienced in this in vitro chip model. Thus, applying a single
cell culture medium onto the PDMS wall is described using constant to describe drug loss in an Organ Chip will not capture
the partition coefficient (P): these dynamic concentration distributions and may inaccurately
cpdms describe the dosing window.
P¼ (1) Next, we explored the effect of Dpdms and P on the
cmed
concentration of the drug in the basal channel when
where cpdms is the concentration of drug in PDMS and cmed continuously dosed at 1.24 μM under 60 μL h−1 flow for 12
is the concentration of drug in cell culture medium. The h. We varied Dpdms from 1 × 10−12 m2 s−1 to 1 × 10−15 m2

This journal is © The Royal Society of Chemistry 2021 Lab Chip

 View Article Online

Paper Lab on a Chip

Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

Fig. 2 Heat maps of the drug concentration in the chip over time show that loss due to adsorption and absorption is a spatially and temporally
dynamic process. a) 3D surface heat maps of the concentration of drug in the fluidic channels of the human lung airway chip viewed from above
after 3, 6, and 12 h with P = 100 and Dpdms = 1 × 10−13 m2 s−1. Drug is continuously flowed into the basal channel and the apical channel is held
static. Arrow indicates the flow direction. b) 2D heat maps of drug concentration through a vertical cross section in the center of the chip along
the xz plane at 3, 6, and 12 h for P = 100 and 400, and Dpdms = 1 × 10−13 m2 s−1. The heat maps show the concentration of drug in the apical and
basal channel medium in the vertical chip cross section. The drug concentration radially extends outwards from the center of the channel with the
lowest concentrations near the lateral side walls and begins to enter the apical channel fluid over time. Drugs with lower P values require less time
to reach the center of the channel. c) 2D heat maps of drug concentration through a horizontal cross section in the center of the basal channel
along the xy plane at 3, 6, and 12 h for P = 100 and 400 and Dpdms = 1 × 10−13 m2 s−1. Arrows indicate the flow direction.

s−1 and observed that compounds with low PDMS diffusivity work complements past studies which concluded that drug
(Dpdms ≤ 1 × 10−13 m2 s−1) require less time to saturate in loss in PDMS can be significant for log P > 2.7, but total
the fluidic channel because the compound accumulates in loss also depends on diffusivity, convection, and dosing
bulk PDMS near the PDMS-medium interface (Fig. 3a). time.27 Therefore, drug loss should be simulated and
Conversely, more drug loss is observed for compounds with evaluated under the same conditions as used in the in vitro
high PDMS diffusivity (Dpdms > 1 × 10−13 m2 s−1) because model of interest.
the compound travels down its concentration gradient faster With the understanding that P and Dpdms both
and does not accumulate near the wall. We also varied P contribute to drug loss, we developed a computational and
from 25–600 while keeping Dpdms constant at 1 × 10−13 m2 experimental strategy to solve for both values and simulate
s−1 and observed that drugs with higher P values (P > 200 = the loss of a drug in the chip over time using amodiaquine
log P > 2.3) show significant depletion in the basal channel (1.24 μM perfused through the basal channel of the human
(Fig. 3b). As expected, drugs with high partition coefficients lung airway chip at a flow rate = 60 μL h−1) as a test case.
adsorb more rapidly to the channel walls and deplete from Our strategy incorporates three steps: 1) experimentally
the medium at a higher rate. The simulations emphasize solve for Dpdms of drug, 2) experimentally determine the
that in vitro models with the same flow rate and dosing concentration of drug in the basal outflow, and 3) vary P in
concentration can experience dramatically different the computational simulation to fit the experimental basal
compound loss depending on the value of P and Dpdms. Our outflow concentrations.

Lab Chip This journal is © The Royal Society of Chemistry 2021

 View Article Online

Lab on a Chip Paper

Fick's second law to the experimental data by varying the

value of Dpdms and minimizing the sum of squares (Fig. 4b
).27,29 We calculated the diffusion coefficient in four regions
of the Organ Chip—two regions in the apical channel and
two regions in the basal channel—and found that the
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

location in the chip does not affect the value of Dpdms. The
diffusion coefficient that we calculate, Dpdms = 3.8 ± 4.5 ×
10−13 m2 s−1, agrees with previously reported values for
rhodamine B (Dpdms = 1.9 ± 0.5 × 10−13 m2 s−1) and Cy3
(Dpdms = 6.0 ± 2.8 × 10−13 m2 s−1), which are small molecules
with molecular weights <1000 Da.27

Simulating the drug concentration in a two-channel Organ

Chip
With knowledge of the approximate Dpdms value, we
estimated P by continuously measuring the concentration of
Fig. 3 Effect of Dpdms and P on the concentration of the drug in the
the amodiaquine drug in the outflow over time and fitting
center of the basal channel over time. a) Simulated drug
concentrations at different Dpdms values with P held constant (P =
the simulation to match the experimental data. Human lung
100). b) Simulated drug concentrations at different P values with Dpdms airway chips lined with living bronchial epithelium interfaced
held constant (Dpdms = 1 × 10−13 m2 s−1). with pulmonary microvascular endothelium were perfused
with 1.24 μM amodiaquine in the vascular channel for 12 h
at 60 μL h−1 and the basal outflow was collected every 3 h.
Experimentally solving for the diffusion coefficient Quantification of amodiaquine in these effluent samples
using mass spectrometry reveals that its concentration
Using amodiaquine to test the approach, the diffusivity of gradually increases over the dosing period, but does not
the drug was determined by correlating Fick's second law to reach Cmax, because the drug continues to adsorb to the
the fluorescence distribution of a surrogate compound inside channel walls (Fig. 5). We solved for a range of P values that
an Organ Chip. The diffusivity of a compound in a solid fit the experimental data by simulating the concentration of
polymer is inversely proportional to the molecular size.27,28 amodiaquine in the basal channel outflow at steps of P = 5
Fluorescein isothiocyanate (FITC) was selected as a surrogate and terminating the simulation when the concentration
compound to amodiaquine because it is similar in molecular exceeded the experimental error at all time points. The
weight, structure, and hydrophobicity (Table S1†), and FITC simulated basal outflow values fit the experimental data
absorption into the bulk PDMS of the Organ Chip can be when P = 20–60 (log P = 1.3–1.8). The log P range that we
quantitatively and spatially analyzed using fluorescence calculate is lower than the log P for amodiaquine reported in
microscopy. We flowed FITC into the Organ Chip at 1.24 μM, DrugBank (log P = 3.7),30 because the channels of the human
the Cmax of amodiaquine, and observed entry into the bulk lung airway chip are coated with extracellular matrix and
polymer through the PDMS wall (Fig. 4a). The diffusion cells that effectively reduce the hydrophobicity and decrease
coefficient was obtained by fitting the analytical solution of drug adsorption.

Fig. 4 Calculating the diffusion coefficient in PDMS. a) Fluorescence microscopic image of FITC in the basal channel (left) at its interface with the
channel wall of an Organ Chip after 6 h of continuous flow (bar, 500 μm). Fluorescence is observed extending outward from the surface of the
channel wall into the bulk PDMS at the right. b) Plot of the normalized fluorescence intensity in bulk PDMS at a distance away from the channel
wall. These data, which were acquired from fluorescence images in the center of the apical channel near the device inlet, show that the
fluorescence intensity gradually decreases away from the channel wall and the loss is proportional to Dpdms. The value of Dpdms is determined by
fitting the analytical solution of Fick's second law to the experimental data by varying Dpdms and minimizing the sum of squares (with R 2 = 0.99).

This journal is © The Royal Society of Chemistry 2021 Lab Chip

 View Article Online

Paper Lab on a Chip

profile inside the basal channel at earlier time points show a

significant depletion zone near the channel walls
(Fig. 6b and c). The drug becomes more uniformly
distributed across the channel over time because the drug
approaches a saturation limit near the PDMS wall. These
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

results reveal that use of a pretreatment step would allow the

drug concentration to reach near-saturation before
introducing a perturbation (e.g. viral or bacterial infection)
and that dosing the drug at a higher concentration could
Fig. 5 Plot showing the experimentally determined concentrations of
compensate for the loss due to adsorption and absorption,
amodiaquine in the basal effluent measured over 12 h and the
simulated amodiaquine concentration at P = 20 and P = 60. The and hence allow the in vitro model to more accurately
simulated amodiaquine concentrations from P = 20–60 lie within the approach the clinical Cmax. The ability to recapitulate the
range of the experimental data. The simulations were performed using clinical Cmax permits pharmacokinetic modeling of the
the experimentally determined Dpdms value. absorption and distribution phases after IV bolus
administration.

Next, we used the calculated values of Dpdms = 3.8 ± 4.5 × Discussion

10−13 m2 s−1 and P = 20–60 to simulate the spatiotemporal
concentration of amodiaquine in human lung airway chips The combined experimental-computational strategy described
when administered continuously in the basal channel for 72 here provides a way to quantitatively simulate the spatial and
h used in our past experiment on SARS-CoV-2.1 The temporal gradient of a drug in 3D within a two-channel
simulated concentration profile indicates that the level of PDMS Organ Chip without prior knowledge of the partition
amodiaquine inside the basal channel sharply rises during coefficient or the rate of diffusion in PDMS. Importantly, this
the first 5–10 h of dosing and continues increasing more approach addresses a longstanding limitation associated with
gradually to reach 0.82–1.08 μM after 72 h (Fig. 6a and S3†). the use of microfluidic culture devices, which are frequently
Heat maps of the estimated amodiaquine concentration composed of PDMS, to more accurately inform dosing studies

Fig. 6 Amodiaquine concentration in the human lung airway chip over a 72 h dosing period. a) Plot of the simulated amodiaquine concentration
in the center of the basal channel over 72 h at Dpdms = 3.8 × 10−13 m2 s−1 and P = 20–60 (grey). b) 2D heat maps of a horizontal cross section
through the center of the basal channel along the xy plane at different times with P = 40 and Dpdms = 3.8 × 10−13 m2 s−1. A depletion zone is
observed near the channel walls at earlier time points. c) 2D heat maps of a vertical cross section through center of the chip along the xz plane at
different times with P = 40 Dpdms = 3.8 × 10−13 m2 s−1 showing that the drug evenly fills the lower channel of the chip by approximately 48 h.

Lab Chip This journal is © The Royal Society of Chemistry 2021

 View Article Online

Lab on a Chip Paper

and define toxicity thresholds. We simulated drug loss in 3D points, a region of drug depletion forms near the channel
and show that it is a spatially and temporally dynamic wall and the concentration decreases along the direction of
process that depends on the physicochemical properties of fluid flow, but over time, the concentration becomes more
the drug and the experimental Organ Chip model. By homogenously distributed throughout the channel. Taken
experimentally measuring the diffusion coefficient of the together, these data indicate that drug loss occurs in two
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

compound into PDMS and determining its partition phases: the first phase is dominated by adsorption to the
coefficient by measuring its levels in the channel outflow channel wall and the second is dominated by diffusion
using mass spectrometry, we are able to determine the through bulk PDMS. This type of modeling can be useful for
effective log P range of the compound. We applied this retrospective analysis of results of past studies, and to assist
strategy to estimate the concentration of amodiaquine in a in the design of future Organ Chip experiments. For example,
human lung airway chip after 72 h of continuous dosing and our results suggest that a pretreatment step that spans the
show that the concentration reduces by 13–34% in the center adsorption-dominating phase would allow a drug to reach
of the basal channel, which affects analysis of prior results near-saturation in the channel before applying a
and design of future studies with this drug. perturbation, and thus, amodiaquine should be flowed
Previous work has shown that PDMS can strongly bias through the basal channel of the human lung airway chip for
drug dosing studies and influence cell behavior. For example, approximately 24 h before infection with the virus in future
when tested using a multiple myeloma cell line, the IC50 for studies. The analysis also indicates that we might have
two cancer drugs (bortezomib and carfilzomib) was shown to slightly underestimated the antiviral potency of amodiaquine
be 4.3-fold greater in a microfluidic device composed of in our past study1 as the inhibition we observed was obtained
PDMS compared to polystyrene and cyclic olefin polymer.31 with a dose that was approximately 25% lower than the
Similarly, human embryonic kidney (HEK) cells responded to clinical Cmax we originally targeted.
fluoxetine (Prozac) treatment when cultured in polystyrene, We recognize that individually calculating the D and P
but not PDMS microchannels.32 To demonstrate the effect of values for each compound in a large drug library would be
PDMS on cell behavior, one study observed a reduction in very time-consuming and costly. The computational-
CD38 expression on human umbilical cord blood-derived experimental approach described here can be used in the
CD34+ cells cultured on PDMS due to the loss of all-trans future to refine existing QSPR34,35 and QSAR36 models and
retinoic acid into the PDMS.33 Therefore, understanding and permit in silico predictions of the diffusion and partition
quantifying drug loss in PDMS microfluidic culture systems coefficients of a drug in the Organ Chip. QSPR and QSAR
is critical because it can shift dose–response curves, bias models use descriptors that represent the molecular structure
toxicity thresholds, and alter cell behavior. and properties of a compound to predict an unknown activity
In a previous study, the loss of drug due to PDMS or property,37–39 and experimental determination of D and P
absorption was incorporated into a PK model of multiple values for a training set of compounds can help to refine
fluidically linked Organ Chips by experimentally measuring QSPR and QSAR models, respectively. For example, the
the drug in the channel outflows of blank chips and diffusivity of a training set of diverse fluorescent compounds
calculating the fractional loss.10 In this approach, the may be experimentally determined and used to build a QSPR
geometry of the Organ Chip was simplified into three discrete model that is predictive of the experimental data. The
regions and modeled in 2D, and diffusion of the drug into descriptors in the QSPR model can then be iterated to
bulk PDMS was not considered. Our approach advances this maximize the correlation between experimental and
work by constructing a 3D model that matches the exact predicted diffusivities. This strategy would circumvent the
geometry of the Organ Chip and considers all physical need to use a fluorescent surrogate compound because the
processes that contribute to drug loss (e.g., convection, diffusivity of the actual drug will be derived from the QSPR
diffusion, adsorption, and absorption). This improved 3D model. Similarly, a QSAR model can be iterated using P
model can simulate the concentration of a drug at any values derived from our computational-experimental
location inside the Organ Chip throughout the dosing period. approach for a representative group of molecules. We note
Knowledge of the concentration and distribution of the drug that mass spectrometric analysis of the drug concentration in
inside the chip allows for the optimization of dosing the outflow of the Organ Chip should still be performed for
protocols to ensure that the drug concentration in the chip each drug and chip design to verify that the simulated
approaches a clinically relevant dose (e.g., Cmax) before outflow concentration matches the actual concentration
applying a perturbation. Furthermore, we describe a strategy present in the medium.
to calculate the effective diffusivity and range of log P values It is important to note that another potential limitation of
that are specific to a drug under continuous dosing in the our approach is that the PDMS membrane, epithelium, and
in vitro model with cultured cells. endothelium are modeled as a homogenous PDMS block,
Using amodiaquine administration in the human lung and we assume that drug passively diffuses through the block
airway chip as a test case, we analyzed its distribution in 3D at a rate described by Fick's second law. The accuracy of the
and observed spatial variations in concentration throughout predicted range of log P values may be improved in the future
the fluidic channel that varied with time. At earlier time by measuring accumulation of the drug in the cells and

This journal is © The Royal Society of Chemistry 2021 Lab Chip

 View Article Online

Paper Lab on a Chip

apparent drug permeability (Papp), and incorporating this and endothelium (1.0 μm) was experimentally determined
information into the model. The concentration of drug in the from cross-sections of human airway chips. The epithelium
basal channel outflow can then be combined with the and endothelium layers were combined with the membrane
simulation of drug loss and fit to a PK model to estimate the to create a single geometric domain. The membrane pores
rate and extent of distribution.20 Papp can be determined by were removed to increase computational efficiency.
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

flowing a tracer molecule into the basal channel and The drug flow was simulated using the Navier–Stokes
calculating the fraction of molecule that entered the apical equations for incompressible flow coupled with continuity
channel.1,10,40 Furthermore, known metabolic parameters can law, and assuming incompressible flow and constant
also be incorporated into the CFD simulation, which has viscosity across the domain:
been previously incorporated into a PK model to predict
∂u
terfenadine-fexofenadine in a connected heart-liver ρ ¼ ∇P þ μ∇2 u (2)
∂t
microfluidic system.18,41
Organ Chips and other microfluidic culture models are ∇·u = 0 (3)
commonly fabricated with PDMS, but compound loss can bias
results and affect cell behavior. In this study, we described a where is ∇ the gradient operator and ∇2 is the square of
combined experimental and computational strategy to simulate vector Laplacian, P is pressure, u is velocity, μ is viscosity,
the concentration of a drug inside a two-channel microfluidic and ρ is the density of medium. Apical and basal inlet flow
Organ Chip composed of PDMS under continuous vascular rates were set as 0 μL h−1 and 60 μL h−1, respectively, for the
dosing that considers convection, diffusion, and dissolution of flow simulations. The outlet boundary condition was set to
the drug in the chip. Using this strategy, drug concentration zero gauge pressure and a no-slip boundary condition was set
can be estimated over a multi-day dosing period without prior at the channel walls.
knowledge of the partition or diffusion coefficient. The 3D Fick's second law was applied to model drug transport in
computational simulations of drug loss emphasize that an culture medium (eqn (4)) and PDMS (eqn (5)):
understanding of spatial and temporal drug loss is necessary to
∂cmed
establish dosing guidelines. In addition, a single coefficient ρ − ∇·ðDmed ·∇cmed Þ þ u·∇cmed ¼ 0 (4)
∂t
derived from the literature or a chemical database is likely not
applicable for an in vitro microfluidic model because drug loss
∂cpdms

depends on the relationship between the physicochemical ρpdms − ∇· Dpdms ·∇cpdms ¼ 0 (5)
∂t
properties of the drug, the duration of dosing, and the design
of the Organ Chip device used for therapeutic testing. This where Dmed is the diffusivity of drug in cell culture medium,
methodology can be further strengthened and validated by Dpdms is the diffusivity of drug in PDMS, cmed is the
evaluating a group of chemically diverse compounds in the concentration of drug in medium, cpdms is the concentration
future. Importantly, we anticipate that our model will translate of drug in PDMS, and ρpdms is the density of PDMS. The use
towards any compound that can be quantitatively measured of mass transport equations to describe the flux of drug
using mass spectrometry because it does not rely on previously permits straightforward application of this method to a range
reported log P values. Our strategy may also increase the of drug concentrations.
usefulness of Organ Chips in clinical research programs by The partition coefficient was implemented in COMSOL
offering a way to more accurately plan dosing strategies and Multiphysics by applying the following pointwise constraint
predict drug pharmacokinetics, therapeutic windows, and expression at all PDMS-medium and PDMS-cell interfaces:
toxicities in a human organ-relevant context in vitro.
cpdms − P × cmed (6)
Methods
where test(cpdms) – test(cmed) was implemented in the
Computational simulations constraint force expression field to ensure continuous flux
A finite element simulation of drug concentration in the over the phase boundaries. A zero flux boundary condition
human lung airway chip was developed in COMSOL was applied on the walls and ports of the vacuum channels. A
Multiphysics 5.5 (COMSOL, Inc.). A SOLIDWORKS custom meshing sequence was carried out to increase the
(Solidworks 2017) 3D rendering of the microfluidic device mesh resolution near PDMS-medium boundaries, within the
was imported into COMSOL. The Organ Chip is a (PDMS) fluidic domain, and near corners of the fluidic domain. The
microfluidic culture device (37.2 mm long × 16.2 mm wide) laminar flow and diffusion of dilute species modules were
that contains parallel fluidic channels (15.5 mm long × 1 mm coupled and a time-dependent simulation was performed for
wide) separated by a porous PDMS membrane (50 μm thick; 0–12 h with 0.5 h time steps or for 0–72 h with 0.5 h time steps.
7 μm pores). The fluidic component consists of two adjacent The concentration of drug in the basal outflow was
parallel microchannels (apical, 1 mm wide × 1 mm high; calculated by placing a boundary probe on the basal outlet
basal, 1 mm wide × 0.2 mm high; length of overlapping and programming the probe to report the average drug
channels, 16.7 mm). The height of the epithelium (3.9 μm) concentration. Similarly, the concentration of drug in the

Lab Chip This journal is © The Royal Society of Chemistry 2021

 View Article Online

Lab on a Chip Paper

center of the basal channel was determined by placing a interface (ALI) was established after 5–7 days. After
point probe in the middle of the basal channel and establishment of ALI, the apical channel medium was
programming the probe to output the concentration of the removed from the human lung airway chip and the basal
drug. We also simulated the concentration in the basal medium was replaced with PneumaCult-ALI medium
channel without dosing 1.24 μM statically in the apical (PneumaCult™-ALI Medium, StemCell Technologies, 05001)
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

channel and observed no effect on the concentration of drug. supplemented with 1% penicillin–streptomycin, 10 ng mL−1
Parameters used in the simulation are provided in Table S2.† of VEGF and 1 μg mL−1 ascorbic acid. The apical channel of
the chips were flushed for 2 minutes at 1000 μL h−1. After
flushing, the flow rate in the apical channel was set to 0 μL
Cell culture h−1 and the Zoë was programmed to “Air”. Medium flowed
This work was reviewed by the Harvard Faculty of Medicine through the basal channel at 30 μL h−1. The apical surface of
Office of Regulatory Affairs and Research Compliance. The the epithelium was rinsed once weekly with PBS (−/−) to
Institutional Review Board (IRB) of the Harvard Faculty of remove cellular debris and mucus. Drug was dosed into
Medicine determined this work to be not human subjects human lung airway chips after 12 days of ALI culture.
research as defined by DHHS or FDA regulations. All
experiments were performed in compliance with relevant Diffusion coefficient (Dpdms) determination
laws, federal and University guidelines. The Organ Chip A blank Organ Chip (CHIP-S1 Stretchable Chip, RE00001024
culture protocol is based on the methods described in Si Basic Research Kit; Emulate, Inc) was equilibrated with PBS
et al. with minor modifications.1 Microfluidic Organ Chips (−/−) and placed inside a microscope chamber at 37 °C and
were purchased from Emulate, Inc (CHIP-S1 Stretchable 100% humidity to match the environment of a cell culture
Chip, RE00001024 Basic Research Kit). The Organ Chips have incubator. A 1.24 μM solution of FITC (Sigma Cat# 46950) in
two fluidic microchannels that are separated by a membrane PBS was introduced into the apical and basal channel at 60
containing 7 μm diameter pores. To prepare the chips for cell μL h−1 for 7 h with a syringe pump (Harvard Apparatus). Four
culture, the surface of the fluidic channels and membrane regions of the chip (two in the apical channel and two in the
were activated with ER1 + ER2 (Emulate, Inc) in the presence basal channel) were imaged after 6 h of flow at 10×
of UV light (Nailstar, NS-01-US) for 20 minutes. The channels magnification (Zeiss Axio Observer Z1.2). The fluorescence
were washed with ER2 buffer, followed by PBS −/−. The intensity across the microfluidic channel was quantified
activated fluidic channels were coated with 0.5 mg mL−1 using ImageJ and normalized to the dosing concentration.
collagen type IV human placenta (Sigma Aldrich, C5533-5MG) The diffusion coefficient of FITC in PDMS was determined by
in the cell culture incubator overnight at 37 °C, 100% applying the analytical solution of Fick's second law:
humidity and 5% CO2. The coating solution was aspirated
!
from the chip the following day. x
Human lung microvascular endothelial cells (HMVEC-L) Cðx; tÞ ¼ Ci;p erfc pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (7)
4Dpdms t
were sourced from LONZA (CC-2427) and cultured in EGM-
MV2 medium supplemented with growth factors and 1%
penicillin–streptomycin (Gibco,15 140 122) according to the where Ci,p is the concentration of the drug on the PDMS side
manufacturer's instructions (EGM-MV2 Kit, Promocell, C- of the medium-PDMS interface, x is the distance away from
22121). Endothelial cells were seeded onto the top and the wall into bulk PDMS, Dpdms is the diffusion coefficient of
bottom of the basal channel at 2 × 107 cells mL−1, passage 2 the drug in PDMS, t is the elapsed time, and C(x, t) is the
(p2), and incubated statically for 2 h at 37 °C. Primary normal concentration of the compound. C(x, t) was fit to the
human bronchial epithelial (NHBE) cells were sourced from experimentally determined concentration profile by varying
LONZA (CC-2540). The epithelial cells were cultured and Dpdms and minimizing the sum of squared residuals.27,29 The
expanded in Airway Growth Medium (Airway Epithelial diffusion coefficient was calculated from n = 3 experimental
Growth Medium Kit, Promocell, C-21160) supplemented with replicates and R 2 > 0.98 for each replicate.
1% penicillin–streptomycin. NHBEs were seeded into the
apical channel of the Organ Chips at a density of 2.5 × 106 Amodiaquine dosing
cells mL−1 at passage 2 (p2) and incubated for 4 h under Amodiaquine dihydrochloride dihydrate (cat. #A2799; Sigma-
static conditions at 37 °C. Aldrich) was dissolved in dimethyl sulfoxide (DMSO) to a stock
The next day, the chips were inserted into Pod Portable concentration of 10 mM and diluted in ALI medium to the
Modules (RE00001024 Basic Research Kit; Emulate, Inc) and reported Cmax in human blood (1.24 μM). Amodiaquine was
placed in a Zoë Culture Module (Emulate, Inc). EGM-MV2 perfused through the vascular channel of human lung airway
and Airway Growth Medium was added to the basal and chips at 60 μL h−1 while the airway channel was statically
apical inlet reservoirs, respectively. The Zoë was programmed treated with the same concentration of drug. Medium from the
to flow at 30 μL h−1 medium into the apical and basal inlet and outlet of the vascular channel was collected every 3 h
channel and the chips were maintained cells were for 12 h and frozen at −80 °C until LC-MS/MS analysis. Mass
maintained at 37 °C, 100% humidity and 5% CO2. Air–liquid spectrometry was performed by PureHoney Technologies

This journal is © The Royal Society of Chemistry 2021 Lab Chip

 View Article Online

Paper Lab on a Chip

(Billerica, MA) using a RapidFire-MS/MS system. The References

RapidFire-MS/MS system consisted of an Agilent RapidFire 300
interfaced to a Sciex API4000 triple quadrupole mass 1 L. Si, H. Bai, M. Rodas, W. Cao, C. Y. Oh and A. Jiang,
spectrometer with Analyst v.1.6 data acquisition software. et al., Nat. Biomed. Eng., 2021, DOI: 10.1038/s41551-021-
Quantitative analysis was performed with Agilent RapidFire 00718-9.
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

Integrator software. Standard curves were generated with 0–4 2 H. Tang, Y. Abouleila, L. Si, A. M. Ortega-Prieto, C. L.
μM amodiaquine and all samples were run in triplicate. The Mummery, D. E. Ingber and A. Mashaghi, Trends Microbiol.,
concentration of amodiaquine in the basal channel outflow 2020, 28, 934–946.
was normalized to the concentration in the basal inlet to 3 D. Huh, B. D. Matthews, A. Mammoto, M. Montoya-Zavala,
account for adsorption of the drug to the Pod walls that occurs H. Y. Hsin and D. E. Ingber, Science, 2010, 328, 1662–1668.
within the first 6 hours of dosing. 4 A. Tovaglieri, A. Sontheimer-Phelps, A. Geirnaert, R. Prantil-
Baun, D. M. Camacho and D. B. Chou, et al., Microbiome,
Partition coefficient (P) determination 2019, 7, 43.
The concentration of amodiaquine in the basal outflow was 5 S. Jalili-Firoozinezhad, R. Prantil-Baun, A. Jiang, R. Potla, T.
computationally simulated using the experimentally determined Mammoto and J. C. Weaver, et al., Cell Death Discovery,
2018, 9, 223.
Dpdms (Dpdms = 3.8 ± 4.5 × 10−13 ms ). A boundary probe was
2

6 K. H. Benam, R. Villenave, C. Lucchesi, A. Varone, C. Hubeau

placed on the basal outlet and programmed to output the
and H.-H. Lee, et al., Nat. Methods, 2016, 13, 151–157.
average drug concentration. We first performed a course
7 Q. Zhou, D. Patel, T. Kwa, A. Haque, Z. Matharu and G.
estimate of the P value range that fit the LC-MS/MS data. We
Stybayeva, et al., Lab Chip, 2015, 15, 4467–4478.
performed a time-dependent simulation for 12 h with 0.5 h time
8 S. Lu, F. Cuzzucoli, J. Jiang, L.-G. Liang, Y. Wang and M.
steps at P = 25, 50, 100, and 200 and found that the basal
Kong, et al., Lab Chip, 2018, 18, 3379–3392.
outflow concentrations at P∼15–100 approximate the LC-MS/MS
9 D. Huh, D. C. Leslie, B. D. Matthews, J. P. Fraser, S. Jurek
data. We solved for the range of P that fit the experimental data
and G. A. Hamilton, et al., Sci. Transl. Med., 2012, 4,
by simulating the concentration of amodiaquine at steps of P =
159ra147.
5 and terminating the simulation when the concentration
10 A. Herland, B. M. Maoz, D. Das, M. R. Somayaji, R. Prantil-
exceeded the experimental error at all time points. We
Baun and R. Novak, et al., Nat. Biomed. Eng., 2020, 4,
determined the lower and upper boundary of P by initiating the
421–436.
stepwise simulation at P = 15 and P = 100, respectively. P = 20–
11 H. Jung Kim, D. Huh, G. Hamilton and D. E. Ingber, Lab
60 fit the experimental LC-MS/MS data (log P = 1.3–1.8).
Chip, 2012, 12, 2165–2174.
12 J. D. Wang, N. J. Douville, S. Takayama and M. ElSayed, Ann.
Statistical analysis
Biomed. Eng., 2012, 40, 1862–1873.
All experimental results are expressed as means ± standard 13 A. W. Auner, K. M. Tasneem, D. A. Markov, L. J. McCawley
deviation (SD). Each experiment was conducted with a and M. S. Hutson, Lab Chip, 2019, 19, 864–874.
sample size of n = 3 chips per condition. 14 A. Cherkasov, E. N. Muratov, D. Fourches, A. Varnek, I. I.
Baskin and M. Cronin, et al., J. Med. Chem., 2014, 57,
Author contributions 4977–5010.
15 T. Zhu, W. Chen, H. Cheng, Y. Wang and R. P. Singh,
J. G. and D. E. I. conceived this study. J. G., A. O., and C. O.
Ecotoxicol. Environ. Saf., 2019, 182, 109374.
performed and analyzed experiments and simulations with
16 N. M. Garrido, A. J. Queimada, M. Jorge, E. A. Macedo and
other authors assisting with data analysis. R. P. designed and
I. G. Economou, J. Chem. Theory Comput., 2009, 5,
executed drug dosing experiments. R. P. and D. E. I. managed
2436–2446.
the project progress. J. G. and D. E. I. wrote the manuscript
17 B. J. van Meer, H. de Vries, K. S. A. Firth, J. van Weerd,
with all authors providing feedback.
L. G. J. Tertoolen and H. B. J. Karperien, et al., Biochem.
Biophys. Res. Commun., 2017, 482, 323–328.
Conflicts of interest 18 C. W. McAleer, A. Pointon, C. J. Long, R. L. Brighton, B. D.
D. E. I. holds equity in Emulate Inc. and is a member of its Wilkin and L. R. Bridges, et al., Sci. Rep., 2019, 9, 9619.
board of directors and chairs its scientific advisory board. 19 B. Zhang, A. Korolj, B. F. L. Lai and M. Radisic, Nat. Rev.
Mater., 2018, 3, 257–278.
Acknowledgements 20 D. B. Chou, V. Frismantas, Y. Milton, R. David, P. Pop-
Damkov and D. Ferguson, et al., Nat. Biomed. Eng., 2020, 4,
This work was supported by DARPA under Cooperative 394–406.
Agreements (W911NF-16-C-0050 and HR0011-19-2-0008/AWD- 21 J. S. Kuo and D. T. Chiu, Annu. Rev. Anal. Chem., 2011, 4,
102840-612, NIH (NCATS 1-UH3-HL-141797-01 to D.E.I.), and 275–296.
the Wyss Institute for Biologically Inspired Engineering at 22 R. P. Barroso, L. G. M. Basso and A. J. Costa-Filho, Chem.
Harvard University. Phys. Lipids, 2015, 186, 68–78.

Lab Chip This journal is © The Royal Society of Chemistry 2021

 View Article Online

Lab on a Chip Paper

23 S. Salentin, M. F. Adasme, J. C. Heinrich, V. J. Haupt, 32 X. Su, E. W. K. Young, H. A. S. Underkofler, T. J. Kamp, C. T.

S. Daminelli and Y. Zhang, et al., Sci. Rep., 2017, 7, January and D. J. Beebe, J. Biomol. Screening, 2011, 16,
11401. 101–111.
24 P. Buchwald, Theor. Biol. Med. Modell., 2011, 8, 20. 33 K. Futrega, J. Yu, J. W. Jones, M. A. Kane, W. B. Lott and K.
25 J. Wang and T. Hou, J. Comput. Chem., 2011, 32, Atkinson, et al., Lab Chip, 2016, 16, 1473–1483.
Published on 27 July 2021. Downloaded by Thueringer Universitats Landesbibliothek Jena on 8/25/2021 8:06:33 AM.

3505–3519. 34 A. Belles, C. Franke, C. Alary, Y. Aminot and J. W. Readman,

26 J. Järnefelt, T. Laurent and R. Rigler, FEBS Lett., 1988, 242, Environ. Toxicol. Chem., 2018, 37, 1291–1300.
129–133. 35 A. M. Lampic, D. Mackay and J. M. Parnis, Mol. Inf.,
27 V. S. Shirure and S. C. George, Lab Chip, 2017, 17, 681–690. 2019, 38, 1800110.
28 G. Kaur, J. Grewal, K. Jyoti, U. K. Jain, R. Chandra and J. 36 J. Plante and S. Werner, Aust. J. Chem., 2018, 10, 61.
Madan, in Drug Targeting and Stimuli Sensitive Drug Delivery 37 Danishuddin and A. U. Khan, Drug Discovery Today,
Systems, ed. A. M. Grumezescu, William Andrew Publishing, 2016, 21, 1291–1302.
2018, pp. 567–626. 38 I. Ponzoni, V. Sebastián-Pérez, M. J. Martínez, C. Roca, C. De
29 C. L. Walsh, B. M. Babin, R. W. Kasinskas, J. A. Foster, la Cruz Pérez and F. Cravero, et al., Sci Rep., 2019, 9, 9102.
M. J. McGarry and N. S. Forbes, Lab Chip, 2009, 9, 39 M. Karelson, V. S. Lobanov and A. R. Katritzky, Chem. Rev.,
545–554. 1996, 96, 1027–1044.
30 D. S. Wishart, C. Knox, A. C. Guo, S. Shrivastava, M. 40 A. Sontheimer-Phelps, D. B. Chou, A. Tovaglieri, T. C.
Hassanali, P. Stothard, Z. Chang and J. Woolsey, Nucleic Ferrante, T. Duckworth and C. Fadel, et al., CMGH Cell. Mol.
Acids Res., 2006, 34(Database issue), D668-72. Gastroenterol. Hepatol., 2019, 9(3), 507–526.
31 T. A. Moore, P. Brodersen and E. W. K. Young, Anal. Chem., 41 C. Oleaga, A. Riu, S. Rothemund, A. Lavado, C. W. McAleer
2017, 89, 11391–11398. and C. J. Long, et al., Biomaterials, 2018, 182, 176–190.

This journal is © The Royal Society of Chemistry 2021 Lab Chip