The human colon adenocarcinoma cell line Caco-2 (European Collection of Authenticated Cell Cultures (0,9,042,001, ECACC) were maintained in Eagle’s Minimum Essential Medium (EMEM) (30–2003, ATCC) supplemented with 10% fetal bovine serum (FBS) (S1810-500, Biowest), 1% MEM non-essential amino acids (NEAA) (M7145, Sigma) and 1% penicillin/streptomycin (P/S) (P0781, Sigma). The clone HT29-MTX-E12 generated from the parental human colon adenocarcinoma cells (12,040,401, ECACC), HT29, were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) (D5796, Sigma) supplemented with 10% FBS, 1% NEAA and 1% P/S. Both cell lines were maintained at 37°C, 5% CO₂ in a humidified incubator. Caco-2 cells with passage numbers between 47–61 and HT29-MTX-E12 cells between passage numbers 50–70 were used for all experiments. Cells were seeded for experiments after reaching 80–90% confluency in parental flasks. For coculture experiments, both cell lines were cultured in Caco-2 cell growth media.

For all experiments the three-lane OrganoPlate (4,004–400-B, Mimetas BV) with 40 individual chips was used, where each chip consists of a seeding channel, an ECM-channel and a perfusion channel (220 µm height x 300 µm width). Cell seeding into the OrganoPlate was performed as previously described (Trietsch et al., 2017). In brief, a 4 mg ml⁻¹ Collagen 1 (3,447–020–01, R&D Systems) ECM-gel supplemented with 100 mM HEPES (H0887, Sigma) and 3.7 mg ml⁻¹ NaHCO₃ was prepared. The ECM-mixture was dispensed into the ECM-channel (2 µL) and placed in the incubator for 30 min to polymerize. After polymerization 30 µL Hanks’ Balanced Salt solution (HBSS) was added to the ECM-channel to avoid gel drying. Next, cells were detached with 0.25% trypsin EDTA and diluted to a final concentration of 1 × 10⁷ cells mL⁻¹. For coculture of Caco-2 and HT29-MTX E12 cells, a ratio of 6:1 was used. The upper channel of the chip was filled with 2 μL cell suspension (a total of 20 × 10³ cells/channel) and the OrganoPlate was placed vertically inside the incubator for 3 h to allow cell attachment. Thereafter, 50 µL medium was added to the seeding channel (inlet and outlet) and the perfusion channel (inlet and outlet). The OrganoPlate was incubated horizontally for 4 days on an interval rocker (± 7° angle, 8 min rocking interval) at 37°C, 5% CO₂ resulting in a formation of a tubule inside the chip. The medium was replaced every 2–3 days.

Caco-2 cells were seeded at a final cell density of 5 × 10⁴ cells cm⁻² in a 24-well format Transwell system with a surface area of 0.33 cm² (CLS3413, Corning). The cells were grown in EMEM with 10% FBS and 1% P/S over 19–21 days and the medium was replaced every 2–3 days.

The barrier integrity of cells seeded in the Transwell system was monitored routinely using an EVOM2 system equipped with an EndOhm-6G cup for measurements TEER (World Precision Instruments, Sarasota, FL). Raw resistance data were translated into TEER using the equation: T E E R [ Ω × c m 2 ] = T E E R _ R A W [ Ω ] × s u r f a c e   a r e a [ c m 2 ] (1)

Caco-2 monolayers with a TEER above 1,500 Ω*cm² were selected for further drug absorption studies.

Spinning disc (SD) confocal microscopy was performed on a Nikon Ti2 inverted microscope equipped with a Yokogawa CSU-W1 module and a Photometrics Prime 95B sCMOS camera was used for all experiments (excl. Image-based TAT transport). The set-up includes four excitation lasers 405/488/561/638 nm with 442/42, 520/28 BrightLine HC filters and 600/50 ET, 700/75 ET bandpass filters. For this work, a set of CFI Plan Apochromat Lambda objectives were used: ×4/0.2, 20x/0.75, 60x/1.4 and 100x/1.45. For image-based transport experiments, epi-fluorescence microscopy was performed using a Nikon Eclipse Ti2 microscope equipped with an LED source (CoolLED), a dual-band filter cube (AHF, Excitation 455–489 nm, 557–588 nm; Emission 506–542 nm, 605–660 nm) and a Photometrics Evolve 512 electron-multiplying charge-coupled device camera was used (10x N.A. 0.30).

The BI of the mono- and coculture systems was analyzed on day 1, 2 and 4. The Medium in the seeding channel was replaced with EMEM containing 0.5 mg ml⁻¹ 4.4 kDa TD (T1037, Sigma) and placed on an interval rocker for 15 min in the incubator. After incubation, the plate was imaged using a SD confocal microscope at ×4 magnification. Inhomogeneous cell coverage in the seeding channel over during the cultivation leads to an inhomogeneous fluorescence intensity profile. To ensure an accurate comparison of BI values the whole channel (Seeding and ECM channel) was selected for measuring the intensity.

Cells seeded in the OrganoPlate were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) (D8537, Sigma) for 15 min, washed with PBS and permeabilized for 10 min in 0.3% Triton-X (T8787, Sigma). Subsequently, cells were washed with 4% FBS/PBS and incubated for 30 min in blocking solution 3% Bovine Serum Albumin (BSA)/PBS followed by incubation with primary antibody for 60 min at room temperature. Cells were then washed with 4% FBS/PBS and incubated for 30 min with the secondary antibody. Primary antibodies against Rabbit-a-Zonula occludens-1 (ZO-1) (61–7,300, Thermo Fischer, 1:200), Rabbit-a-Mucin 2 (Muc2) (PA5-21329, Thermo Fischer, 1:200), Mouse-a-Ezrin (610,602, BD Transduction, 1:100), Mouse-a-Multidrug Resistance Protein 2 (MRP2) (SC-59608, Santa Cruz, 100 µg ml-1), Mouse-a-Breast Cancer Resistance Protein (BCRP) (ab3380, Abcam, 1:50) were used. For the secondary antibody staining, Goat Anti-Rabbit IgG (H + L) Alexa Fluor 488 (ab150077, Abcam, 1:200) and Goat anti-Mouse IgG (H + L) Alexa Fluor 647 (A-21236, Thermo Fisher, 1:200) were used. Following antibody staining cells were washed three times with 4% FBS/PBS and nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI) (D1306, Thermo Fisher, 1:1,000) for 5 min. Cells were kept in PBS at 4°C and imaged using the previously described SD confocal microscope. All protein stainings were repeated at least three times in mono-and coculture tubules. Images in Figure 2 display representative immunostained cell-tubules.

The coculture tubules were fixed by filling all three channels of the chip system with a solution of 2% glutaraldehyde, 4% paraformaldehyde in 0.1 M Na cacodylate buffer at pH 7.4. After fixation for 30 min at room temperature, the plate was stored at 4°C to await further processing. Each channel was washed three times with 0.1 M Na cacodylate buffer and stained with 1% osmium tetroxide in a 0.1 M Na-cacodylate buffer for 2 h at room temperature. After staining with osmium tetroxide, the channels were washed three times in milliQ water and stained with a 1% uranyl acetate solution overnight. The following day, the channels were washed twice in milliQ water followed by dehydration in increasing concentrations of ethanol, 50, 70, 95% and twice 100% with 20 min between solvent exchanges, to ensure complete dehydration. The channels were further dehydrated three times in acetonitrile for 10 min each before the start of resin infiltration. The channels were infiltrated with a 1:1 solution of TAAB 812 medium hardness embedding resin and acetonitrile. After 1 h, the channels were re-infiltrated with the same 1:1 solution to ensure complete infiltration. After one more hour, the channels were infiltrated with a 2:1 resin:acetonitrile solution overnight. The next morning the channels were re-infiltrated with the 2:1 solution for 6 h. The channels were then infiltrated with pure resin overnight. The following morning the channels were again re-infiltrated with pure resin to ensure complete infiltration. After 6 hours the plate was cured at 60°C for 72 h.

After the resin was completely cured, individual channel setups were cut from the larger plate for further processing. The samples were placed in a concentrated hydrofluoric acid solution, 48 wt%, to dissolve the glass substrate. After 45 min, the sample was removed from the hydrofluoric acid and rinsed with a saturated calcium chloride solution, 45 wt%. The sample was then rinsed with milliQ water. The now exposed channel was cut from the remaining device and re-embedded in TAAB resin which was cured overnight at 60°C. The resin block was trimmed with a razor blade, exposing a surface perpendicular to the channel. Thin sections, 120 nm thick, were cut using a Leica EM UC7 ultramicrotome and placed on a nickel slot grid coated with carbon-stabilized formvar. TEM was performed utilizing a FEI Tecnai T20 G2 TEM located at the Center for Electron Nanoscopy at the Technical University of Denmark, and images were acquired using TVIPS XF416 CCD camera.

The coculture tubules were fixed by filling all three channels of the chip system with a solution of 2% glutaraldehyde, 4% paraformaldehyde in 0.1 M Na cacodylate buffer at pH 7.4. After fixation for 30 min at room temperature, the plate was stored at 4°C to await further processing. Each channel was washed three times with 0.1 M Na cacodylate buffer and stained with 1% osmium tetroxide in a 0.1 M Na cacodylate buffer for 2 h at room temperature. After staining with osmium tetroxide, the channels were washed three times in milliQ water and stained with a 1% uranyl acetate solution overnight. The following day, the channels were washed twice in milliQ water followed by dehydration in increasing concentrations of ethanol, 50, 70, 95% and twice 100% with 20 min between solvent exchanges, to ensure complete dehydration. The channels were further dehydrated three times in acetonitrile for 10 min each before the start of resin infiltration. The channels were infiltrated with a 1:1 solution of TAAB 812 medium hardness embedding resin and acetonitrile. After 1 h, the channels were re-infiltrated with the same 1:1 solution to ensure complete infiltration. After one more hour, the channels were infiltrated with a 2:1 resin:acetonitrile solution overnight. The next morning the channels were re-infiltrated with the 2:1 solution for 6 h. The channels were then infiltrated with pure resin overnight. The following morning the channels were again re-infiltrated with pure resin to ensure complete infiltration. After 6 hours the plate was cured at 60°C for 72 h.

After the resin was completely cured, individual channel setups were cut from the larger plate for further processing. The samples were placed in a concentrated hydrofluoric acid solution, 48 wt%, to dissolve the glass substrate. After 45 min, the sample was removed from the hydrofluoric acid and rinsed with a saturated calcium chloride solution, 45 wt%. The sample was then rinsed with milliQ water. The now exposed channel was cut from the remaining device and re-embedded in TAAB resin which was cured overnight at 60°C. The resin block was trimmed with a razor blade, exposing a surface perpendicular to the channel. Thin sections, 120 nm thick, were cut using a Leica EM UC7 ultramicrotome and placed on a nickel slot grid coated with carbon-stabilized formvar. TEM was performed utilizing a FEI Tecnai T20 G2 TEM located at the Center for Electron Nanoscopy at the Technical University of Denmark, and images were acquired using TVIPS XF416 CCD camera.

Leak tight tubules in the OrganoPlate were fixed with 4% PFA for 15 min and washed afterwards three times with PBS. The cell tubules were acidified with 3% acidic acid and stained with Alcian blue (B8438, Sigma) for 30 min at room temperature. Next, cells were washed three times with PBS for 5 min and bright-field images were acquired with a light microscope.

Calcein-AM is a substrate of the P-gp efflux transporter. After internalization, it fluoresces until it is pumped out the cells (see Supplementary Figure S3). All channels in the OrganoPlate were washed with Opti-buffer containing Phenolred-free medium (DMEM) and HBSS (H6648, Sigma, Ratio of 1 DMEM: 3 HBSS). For detection of Calcein-AM efflux, the Opti-buffer in the seeding channel was replaced with Calcein- AM (C3099, 10 µM in 0.1% DMSO). Inhibition of the P-gp transporter was achieved by pre-incubation with Verapamil (V4629, Sigma, 50 µM in 0.5% Methanol) in the cell-seeding channel. The OrganoPlate was incubated for 60 min on an interval rocker at 37°C with 10 µM Calcein-AM or 50 µM Verapamil +10 µM Calcein-AM. To ensure that the presence of neither 0.5% methanol nor 0.1% DMSO affected the cross membrane partitioning of calcein-AM, both ± verapamil experiments contained the same amount of these compounds. Following incubation, all solutions in all inlets/outlets were replaced with ice cold stopping solution (20 µM Verapamil and nuclei stain Hoechst (H1399, Thermo Fischer, 10 μg ml⁻¹). After 15 min of incubation the cells were imaged immediately by SD confocal microscopy using 405 nm (442/42 BrightLine HC emission filter) and 488 nm (520/28 BrightLine HC emission filter) excitation lasers using a ×20 air objective. Analyzing tubule intensity was performed in Fiji, and the whole tubule was selected as the region of interest (ROI) and the fluorescence intensities for both, calcein and nuclei were extracted. The ratio of calcein and nuclei fluorescence intensity was calculated, representing the intracellular fluorescence of the cell.

Aminopeptidase-N activity of cells was determined using 1.5 μM l-Alanine 4-nitroanilide hydrochloride (A4N) (A9325, Sigma) as a substrate. Serum-free medium was used for the bottom flow channel and the medium in the seeding channel was replaced with A4N. The OrganoPlate was incubated on the interval rocker for 2 h at 37°C, 5% CO₂. After incubation, the solution in the inlet and outlets of the seeding channels were collected and transferred to a 96-well plate. The absorbance of the cleaved product 4-nitroanaline was measured at 405 nm in a plate reader (Tecan, Switzerland).

All drugs were purchased from Sigma Aldrich. Cells seeded in Transwells and OrganoPlate were washed with transport buffer (TB; 10 mM HANKS/HBSS). For the apical side, the pH of TB was adjusted to 6.5 (TB-apical) and for the basolateral side the pH was set to 7.4 (TB-basal). All drugs were diluted to a final concentration of 10 µM in 0.1% DMSO. The cultured monolayers were exposed to the drug for 2 h at 37°C, 5% CO₂. Control cells were exposed to TB without drugs but supplemented with a final concentration of 0.1% DMSO on the apical side. The OrganoPlate was placed for the total 2 h of drug incubation under constant flow, whereas the Transwells were incubated for 2 h under static conditions according to standard protocols (Artursson and Karlsson, 1991; Pontier et al., 2001; Béduneau et al., 2014; Twarog et al., 2020). For the permeability quantification of each drug, samples were analyzed using LC-MS and correlated to a standard curve. As a control for barrier integrity, 100 µM Lucifer Yellow (LY) were added to all samples (Ayehunie et al., 2018).

Samples and standards were analyzed with a Shimadzu Nexera X2/Prominence HPLC (Shimadzu Europe, Duisburg, Germany) and ESI micrOTOF-Q III (Bruker Daltonics, Bremen, Germany) LC-MS setup. The LC was performed by injection of the analyte (5 µL) on a Poroshell 120 SB-C8 column, 2.7 µm, 2.1 × 50 mm (Agilent, Santa Clara, CA, United States) followed by elution with a linear gradient of MeCN and 2.5 mM NH₄OH in water with 0.1% formic acid (from 0 to 100% over 9 min) at a flow rate of 0.4 ml min⁻¹ (a detailed description of the HPLC setup and gradient is reported in Supplementary Table S1). The chromatographic front (1.75 min) was diverted to waste while the remaining run was injected into the ion source. A calibration solution consisting of 2.5 mM NaOH, 2.25 mM formic acid in 90% i-PrOH/water was injected into the ion source between 1.75 and 1.85 min at a flowrate of 30 μL h⁻¹ for internal calibration of the spectra. MS analysis was performed in positive mode in the range 50–2,000 m z ⁻¹ at a rate of 2 Hz. The MS settings used for analysis of the various drugs are listed in the Supplementary Table S2. The concentration determination of each individual drug was used to calculate the apparent permeability of a compound (apical to basolateral) according to the equation (Lennernäs et al., 1997): P a p p = d Q d t ∗ 1 / A C 0 (2)

Where dQ/dt is the flux of drug across the cell monolayer per time [µM sec⁻¹], A is the surface area of the monolayer exposed to the drug (cm²) and C₀ is the initial concentration of the drug.

A two-tailed non-parametric Spearman’s correlation function was applied to determine the Spearman correlation coefficient R (von Erlach et al., 2020). The R-value defines the correlation between the quantified P _app of the drugs in the used model with the known human absorption values, a value closer to one represents a stronger correlation. Furthermore, a four-parametric logistic model curve fit was applied (Sjöberg et al., 2013) to better visualize the correlation between the quantified P _app values and the known human absorption.

Coculture tubules were used on day 4 after seeding and after confirming their leak-tight BI. Varying concentrations of TAT (24 μM; 2 µM) were applied to the coculture-tubules for capturing the internalization. Here, the cells were counterstained with Hoechst (H1399, Thermo Fischer, 10 μg ml⁻¹) and incubated for 1 h at 37°C, 5% CO₂, washed with PBS and images were acquired using SD confocal microscopy with 488 nm laser and 520/28 BrightLine HC filters for detecting FITC-labelled TAT and the 405 nm laser and 442/42 BrightLine HC filter for acquiring Hoechst. The cell tubules exposed to TAT/INS (24 µM/50 μM, IS1-AF647-1, Nanocs) were stained with above mentioned cellular dyes and CellMask-green (C37608, Thermo Fisher, 1:500) for 30 min on ice after 3 h of incubation. The transport of TAT was tracked over 2 h, acquiring time lapse images every 30 s using epi-fluorescence microscopy (see Imaging modalities). Two different concentrations of TAT (24 μM; 2 µM) or TAT/INS were applied to the coculture chips and imaged. To simultaneously monitor the barrier integrity TD (0.5 mg ml⁻¹) was co-applied to the chips and measured in parallel to TAT. For the quantification of both TAT, INS and TD transport across the cell barrier, calibration curves were established. The rate of transport was extracted by a linear fit of each individual curve (Supplementary Figures S7, S8). In-house developed MATLAB based macro was used for analysis.

Coculture tubules were exposed to SNAC (20 mM or 40 mM) in combination with semaglutide-Cy3 (5 µM) and FITC-dextran (46,944, Sigma Aldrich, 0.5 mg ml⁻¹) and imaged over approx. 4 h as described above.

To develop an imaging compatible intestinal in vitro model we utilized the OrganoPlate, constituting a 348-Microwell set up with 40 individual chips and a 150 µm thick glass bottom (Trietsch et al., 2017; Gijzen et al., 2020). The chips are constructed with a seeding channel, an extracellular matrix (ECM) channel and a (fluidic) perfusion channel (Figures 1A,B). To enable the use of the chip as a model for studying intestinal transport, we first ensured that an intact and tight cellular barrier was formed. We seeded either pure Caco-2 cells (Monoculture) or a 6:1 ratio of Caco-2:HT29-MTX E12 cells (Coculture), the latter in an attempt to increase the biological relevance through the presence of mucus-producing goblet cells. For both systems, we facilitated the formation of a differentiated epithelial monolayer by inducing constant shear stress of 0.13 Pa and a mean flow rate of 2.02 μL min⁻¹ using a pump free perfusion system for the total culturing period of only 4 days (Schutgens et al., 2019). To evaluate the barrier integrity (BI) we then applied a 0.5 mg ml⁻¹ solution of 4.4 kDa Tetramethylrhodamine isothiocyanate (TRITC)-dextran (TD) in the seeding channel and used confocal microscopy to image the TD distribution in the chip. If no cells were added to the chip the TD intensity spread through the whole chip, a distribution also seen for both mono- and coculture systems after 1 day of barrier formation (Figure 1C). After 2 days we recorded a partial decrease in the TD intensity in the ECM and medium channels, while after 4 days both mono- and coculture chips displayed complete restriction of TD intensity to the seeding channel, demonstrating that a tight barrier had formed (Figure 1C, Day 4). In comparison, culturing of HT29-MTX E12 cells as a monoculture did not result in the formation of a tubules structure capable of restricting TD diffusion (Supplementary Figure S1). To quantitatively verify the presence of a tight barrier after 4 days of incubation, we determined the BI value by extracting the integrated intensity inside a region of interest in the ECM- and seeding channels (Figure 1C). We calculated the BI value for each chip as F_ECM/F_cell. This allowed us to ascertain cell barrier tightness, with values ranging from 1 (leaky) - 0 (tight), noting that BI values below 0.4 have previously been defined to represent a tight barrier (Trietsch et al., 2017) (Figure 1D). On day one post cell seeding we quantified BI_Mono = 0.95 ± 0.07 and BI_Co = 0.96 ± 0.05, constituting leaky cell barriers. After 2 days in culture BI_Mono = 0.32 ± 0.08 and BI_Co = 0.43 ± 0.05 values were calculated, representing a semi-tight cell barrier (Figure 1D, Day 2). On day four post seeding, the low TD intensity in the receiving channels lead to BI_Mono = 0.05 ± 0.01 and BI_Co = 0.07 ± 0.01 values, verifying that for both systems a tight barrier was established (Figure 1D, Day 4).

To confirm the polarization and differentiation of the cell-tubules into an epithelial monolayer we relied on conventional staining strategies for characteristic phenotypical epithelial cell markers (Anderson et al., 1989; Sambuy et al., 2005; Englund et al., 2006). The unique ability of the microfluidic platform of immunostaining and imaging directly in the system bypasses the need for cell monolayer extraction necessary in traditional in vitro TW models. Both culture systems revealed a homogenous expression of the tight junction protein Zonula occludens-1 (ZO-1) along the whole tubule, demonstrating its differentiation into a compact epithelial cellular network (Figure 2A). We next determined the apicobasal membrane orientation of the cell-tubules by immunostaining the known apically expressed brush border protein ezrin (Cao et al., 2014). Performing z-stack imaging allowed us to investigate ezrin staining in vertical cross-section of the cell tubule (Figure 2B, top). For all individual cells, ezrin staining was completely restricted to the cellular membrane facing the tubule lumen, as determined by a distinct fluorescence signal of ezrin above the cell nucleus (Figure 2B). This demonstrates that the cellular tubules are differentiated into monolayers of correctly polarized epithelial cells. Next, we identified two major intestinal efflux transporters of the ABC-family, the Breast Cancer Resistance Protein (BCRP) and the Multi Drug Resistant Protein (MRP2), which are key regulators of drug localization and thus essential for the in vitro platform accurately predicting intestinal transport (Giacomini et al., 2010; Billat et al., 2017). By staining for BCRP and MRP2 we observed the expected expression profile along the whole monolayer tubules, strengthening the conclusion that the cells are fully differentiated (Figures 2C,D). All apicobasal membrane orientation assessments described above were performed at least three times in both mono- and coculture tubules finding equivalent expression and localization of ZO-1, ezrin, MRP2 or BCRP for both systems (Supplementary Figure S2). Overall, the in-chip immunostaining revealed the formation of a fully differentiated and polarized epithelial tubule.

To evaluate if introducing the goblet cells in the coculture induced the expected mucus production we performed various imaging-based assays to evaluate the mucus expression in both the mono- and coculture system. First, we studied the presence and expression profile of unspecific mucins by staining the cultured cell-tubules with Alcian blue, which binds to all acidic mucins (Figure 2E, Supplementary Figure S2) (Béduneau et al., 2014). The coculture tubules showed a notable higher mucus expression compared to the monoculture tubule (Gijzen et al., 2020). This was especially observed along the ECM-interface, highlighting the need for an ECM support to promote adequate cell differentiation. To verify the presence of Muc2, which is the small intestinal specific mucus type produced by goblet cells (Schneider et al., 2018), we visualized Muc2 by confocal imaging of immunostained cell-tubules. The staining for Muc2 proteins showed a strong and homogenous expression in coculture tubules but was completely absent in monoculture tubules (Figure 2F). Again, it was primarily the cells supported by the ECM that displayed high expression of Muc2 in coculture tubules. Thus, the introduction of the goblet-like cells in coculture systems leads to strong and homogeneous mucus production in contrast to Caco-2 cells in monoculture. This addition increases the biological relevance of the epithelial monolayer system (Karlsson et al., 1993; Hilgendorf et al., 2000; Doherty and Charman, 2002) and permits its use for studying how the mucus layer affects drug transport across the intestinal cell barrier, by comparing the mono and coculture tubules.

To further corroborate the correct polarization of the cell monolayer we visualized cell morphology in detail by extracting the coculture tubule samples and imaging them using transmission electron microscopy (TEM) (Figure 3). Cell tubules were fixed, stained, embedded and sectioned into 120 nm thick slices after 4 days of culture before the tubule sections of the cell-ECM interface were imaged. The micrographs of the tubule sections displayed a distinct and homogenous formation of microvilli along the whole tubule, restricted only on the side facing the lumen opposite to the ECM, clearly illustrating the development of a well-defined apical side (Figure 3A). The basal side of the cells, which is attached to the ECM did not show any morphological changes. Additionally, each individual cell expressed a dense network of microvilli and thus a membrane surface area expansion typically seen for differentiated intestinal epithelial (Figure 3B). Having this characteristic cellular morphology in an in vitro assay is pivotal for its in vivo absorption predictability as the cell membrane surface area is a strong regulator of the total drug absorption. Thus, the generated cell tubules fully differentiated into epithelial cells evidenced by their apicobasal characteristic membrane morphologies.

To establish a physiological relevant small intestinal in vitro platform for drug transport studies the presence and activity of proteolytic enzymes and drug transporters are crucial (Gan and Thakker, 1997; Langguth et al., 1997; Estudante et al., 2013). A highly relevant group of intestinal metabolic enzymes are the aminopeptidases, therefore we evaluated the presence of the brush border aminopeptidase N by employing a standard absorbance based assay, measuring the enzymatic cleavage product 4-Nitroanilide (Figure 4A) (Kim et al., 2012). Both mono- and coculture tubules showed a significant increase in measured 4-Nitroanilide compared to the cell-free control, confirming the presence of functional aminopeptidase N in both mono- and coculture tubules after 4 days of culture.

After having verified the presence of intestinal efflux transporters (Figures 2C,D) we wanted to ensure that these also constituted a functional efflux transport machinery, therefore we examined the function of the intestinal P-glycoprotein (P-gp) efflux transporter by following the change in intracellular calcein upon inhibition with Verapamil (Figures 4B,C, Supplementary Figure S3). We added 10 μM of calcein to the tubular lumen and incubated for 60 min before imaging the whole tubule (Figure 4B, active). The relative intracellular fluorescence (I_relative) was then determined as the ratio of the calcein and DAPI intracellular fluorescence intensities for the whole tubule. For both the mono- and coculture systems we found similar low I_relative values for mono- and coculture tubules, due to the continuous efflux of calcein mediated by the active P-gp transporter (Figure 4C, light grey (mono) and light red (coculture)). Next, we repeated the experiment while adding the P-gp inhibitor Verapamil leading to a significant increase in I_relative values for both the mono- and coculture systems (Figure 4B, Inhibited; Figure 4C, dark grey (mono) and dark red (coculture)). This demonstrated the presence of functional P-gp transporters, representing a key component of the dynamic efflux machinery known to affect drug transport and thus vital for creating an in vitro model with high in vivo transport predictability.

To validate the chip platform for drug transport studies, we correlated the absorption of twelve small model drugs in both the mono- and coculture tubules to their known human absorption values. To ensure a complete evaluation across the whole range of known human absorption values, as recommended by the Food and Drug Administration (FDA), we selected model drugs from all Biopharmaceutical Classification System (BCS) classes with quantifiable transport (Supplementary Table S4) (Amidon et al., 1995; Niazi, 2019). For comparison, we performed the same correlative analysis for a classical TW system (Lipka and Amidon, 1999), which was set up following standard protocols (see experimental section). To quantify the transport, we added the drugs to either chips or TW and extracted both the apical and basolateral fractions after 2 h (Supplementary Figure S4, Supplementary Tables S3, S4). The concentration in all samples was determined using liquid chromatography mass spectrometry (LC-MS) and internal standards (see experimental section, Supplementary Tables S1, S2). The apparent permeability (P _app ) of each drug was calculated as described in the method section and summarized in Supplementary Table S4. Comparing the TW and the monoculture tubules systems we quantified ranges of P _app values from 1.20 × 10⁻⁶ cm s⁻¹ for Erythromycin to 243.41 × 10⁻⁶ cm s⁻¹ for Warfarin in TW and 1.33 × 10⁻⁶ cm s⁻¹ for Erythromycin to 213.99 × 10⁻⁶ cm s⁻¹ for Carbamazepine in the chip monoculture. We correlated the obtained permeability data of each system to the known human absorption values (Cheng et al., 2007; Skolnik et al., 2010; Takenaka et al., 2016; Youhanna and Lauschke, 2021) by plotting the P _app values against the fraction absorbed in humans (FA) and extracted the Spearman correlation coefficient (SCC-R) (von Erlach et al., 2020). This statistic represents a measure of how well an in vitro model recapitulates the in vivo transport and thus allows for comparison between models (Figures 5A,B) (Li et al., 2007; Skolnik et al., 2010; Takenaka et al., 2016). We determined a SCC-R of 0.77 for the TW system, indicating a strong correlation to in vivo transport. In addition, the transport in monoculture tubules showed a similar SCC-R to TW with 0.88. This demonstrates that the monoculture is a qualified platform for drug transport studies, with an ability to predict in vivo intestinal transport at least on the level of the widely employed TW system, additionally having the unique ability to be directly compatible with advanced imaging modalities.

It has previously been shown that the utilization of Caco-2 monocultures for drug transport studies suffers from 1) underestimation of transport through the paracellular route due to a high density of tight junctions and 2) an overestimation of passively absorbed drugs due to the lack of mucus (Hilgendorf et al., 2000; Pontier et al., 2001; Béduneau et al., 2014). This underscores the importance of developing more biological complex in vitro models like the coculture tubules described in this work. Like for the TW and monoculture tubules, we next measured the transport of the twelve drugs in the coculture platform using LC-MS and quantified the P _app (Supplementary Table S4). The detected ranges of P _app values for the coculture system were from 0.99 × 10⁻⁶ cm s⁻¹ for Erythromycin to 300.39 × 10⁻⁶ cm s⁻¹ for Warfarin. Plotting the FA against the P _app for the coculture revealed a SCC-R of 0.89, which is overall similar to the values found for the monoculture tubules and TW (Figure 5C). Thus, while the coculture system did not directly improve the in vivo predictability as compared to the TW system it is encouraging that this more physiologically relevant model can be established in a platform compatible with advanced imaging. Also, to further compare the mono- and coculture tubule system and delineate the effect of introducing the goblet cells we performed a direct comparison of P _app (monoculture) versus P _app (coculture) for each individual drug (Figure 5D). No systematic trend towards elevated or diminished transport for one system in particular, was evident (Figure 5D), however, knowing the transport route employed by the individual drugs allowed us to selectively plot the relative difference in average P _app , using a Bland-Altman comparison analysis (Altman and Bland, 1983). For all four paracellular transported drugs we saw increased relative transport for the coculture (red) versus the monoculture (black), up to 27% seen for Terbutaline (Figure 5E, top). This corroborates the believed increase in the monolayer permeability introduced when adding another cell line into the Caco-2 monolayer, directly increasing paracellular drug transport (Hilgendorf et al., 2000). Plotting the same relative difference in average P _app for the transcellular drugs, we did not see any systematic change towards elevated or diminished transport for either the mono- (black) or the coculture (red) tubules (Figure 5E, bottom, Supplementary Table S5). This suggests that the presence of a mucus layer only had a minute effect on the transcellular transport. However, we caution on underestimating the importance of the mucus layer based on experiments with only small molecule drugs. We imagine that for larger peptide and protein drugs, the presence of a mucus layer might strongly affect the transport rate and thus give a more biological accurate measure of their transport across an intestinal cell layer. Based on this, we choose to continue with the coculture tubule for subsequent transport studies of biologics.

We next used the developed and verified coculture tubules to use imaging-based live cell assays to untangle the uptake and transport mechanism of biologics. To validate the platform for quantitative image based intestinal transport studies, we first employed the FITC-labeled transcriptional activator peptide in HIV (TAT), an arginine-rich cell-penetrating peptide (CPP) extensively used to deliver therapeutic proteins or peptides across cellular barriers (Morris et al., 2001; Kristensen and Nielsen, 2016; Guo et al., 2019). While the ability of TAT to cross cellular-membranes and barriers are uniformly reported, the transport mechanism of TAT remains debated, potentially due to previously studies being restricted to fixed cell samples or undifferentiated single cell experiments (Frankel and Pabo, 1988; Brock, 2014). Here, the unique imaging capabilities of the cell-tubule platform allowed us to quantify, for the first time, TAT transport across a differentiated cellular barrier using high-temporal live cell imaging. We first investigated how the TAT concentration affected its transport mechanism by applying either a low (2 µM) or a high (24 µM) concentration of TAT to the cell-tubules and imaged the coculture chips after 1 h of incubation. We detected a clear concentration dependent difference in the intracellular localization of TAT, with the low TAT concentration system displaying a punctate distribution of TAT inside the cells, suggesting endosomal uptake (Figure 6A). In contrast, the high TAT concentration (24 µM) system displayed a diffuse TAT intensity in the cytoplasm, indicating uptake predominantly via transduction (cell membrane accumulation and destabilization) into the cells (Figure 6A). These results corroborate previous studies on undifferentiated single cell models, showing a concentration dependence of the cellular internalization mechanisms of TAT (Tünnemann et al., 2006, 2008; Duchardt et al., 2007; Brock, 2014). Here, we expand this to a fully differentiated coculture system demonstrating that concentrations differences down to a factor of 12 can lead to distinctly different uptake mechanisms of TAT.

Next we used the platform to study the barrier transport kinetics of TAT. We performed whole-chip real-time live cell imaging, to quantify the concentration of TAT transported across the coculture cell layer and reaching the receiving channel (Figure 6B). Here, we took advantage of the high temporal resolution and strong detection sensitivity offered by the imaging-compatible chip system. The same two concentrations of TAT (2 μM, 24 µM) were added to coculture tubules along with a 0.5 mg ml⁻¹ concentration of TD, allowing us to simultaneously track barrier integrity over time. Both the FITC and TRITC channels were recorded for each tubular system every 30 s for a total of 2.5 h. The well-plate format of the microfluidic platform allowed us to mark the position of numerous coculture tubules and use automated multi-spot imaging to perform several experiments in parallel. A calibration experiment where known concentrations of TAT or TD were added to a chip without cells allowed us to convert the measured intensities to actual concentration in μM and mg mL⁻¹ respectively (Supplementary Figure S5). A TAT surface intensity plot of the seeding, ECM and perfusion channels after 130 min of incubation showed a clear intensity gradient in the ECM-channel for both concentrations, demonstrating that TAT is capable of crossing the tubular cell layer and migrate through the ECM (Figure 6B). To quantify the transport of TAT we integrated the intensity within the ECM facing the receiving channel in the region displayed in Figure 6B (red square). We then used the calibration control to convert this intensity to concentration, which we plotted as a function of time for both the low- and high TAT concentration systems (Figure 6C). For both systems, we record a lag time of approx. 30 min showing a high noise to signal ratio, indicating the detection limit of the individual molecules before a significant signal above the background was detected. Subsequently, both systems reach a linear regime displaying a steady increase in transported peptide. Thus, to decouple the transport efficiency from the initial difference in concentration we quantified the transport rate for the linear regime for both the low and high TAT systems (Supplementary Figures S6, S7). Calculating the average transport rate for all experiments revealed a 54-fold higher rate for the high versus the low TAT system (Figure 6D). To elucidate how this affected the overall amount of transported TAT we again normalized for the initial difference in concentration by converting the quantified TAT concentration after 2.5 h to a percentage of the initially added concentration of TAT. Doing this, we quantified average relative transport percentages of 0.5 ± 0.1% for low TAT and 2.0 ± 0.4% for high TAT, demonstrating a 4.4 fold higher relative amount of TAT transported in the high versus low TAT concentration system (Figure 6E). Finally, the influence of TAT on the barrier integrity was displayed using the acquired TD time series, showing an intact barrier indicated by an overall transport of TD below 5%. However, the sensitivity of the assay allows to detect a TAT concentration dependent difference in the TD transport profile. In the high TAT system a steady increase of TD was observed, but only a minimal transport of TD was detected in the low TAT system (Figure 6F). This TAT concentration dependent differences in the barrier interaction, further support that TAT can use two very different transport pathways, governed by the initial TAT concentration. Additionally, our data demonstrate that the transduction pathway employed by TAT at higher concentrations is vastly more efficient than the endocytic pathway employed at lower concentrations.

Insulin is considered to be at the forefront of oral drug delivery, but despite immense efforts, no insulin-based oral formulation has reached the market, primarily due to extremely low bioavailability, originating both from low stability of free insulin in the harsh gastro-intestinal environment and an extremely low intrinsically cross barrier transport efficiency (Goldberg and Gomez-Orellana, 2003). Next, we took advantage of the subcellular resolution of the platform to study 1) the transport of AlexaFluor647-labeled insulin (INS) alone or 2) the transport of INS when using TAT as a vehicle for the attempted delivery across the cell barrier. In the latter case, we relied on the well-established strategy of electrostatic CPP and peptide complexation (Chen et al., 2017; Guo et al., 2019) and thus simultaneously added both TAT and INS to the tubule system. Additionally we added TD allowing us to track, in real-time, the transport of both the drug delivery vehicle (TAT) and the cargo (INS), while also monitoring how these affected the barrier integrity (TD). We detected no INS transport across the cell barrier independent of whether or not TAT was present, whereas a clear transport of TAT was again detected (Figure 7A). Also here, the barrier integrity was not compromised significantly in chips with and without TAT (Supplementary Figure S8). To ascertain the lack of INS transport we investigated its fate along the transport pathway after 3 h. We described the distribution of internalized INS by live cell imaging of the tubules showing that INS was indeed internalized into the enterocyte cell layer even without TAT (Figure 7B). Additionally, to investigate the intracellular we changed the magnification from ×20 to 100x for live cell imaging, allowing us to elucidate that for both INS and INS/TAT, the INS was localized inside the cytosol in punctual patterns, indicating endosomal uptake (Figure 7C). This suggests that the main limiting factor for the INS delivery across the cellular barrier in our setup was an inability to escape endosomes and/or be moved across the basolateral membrane.

Additionally, we attempted to use the model to study other pharmacological relevant transport mixtures including the permeation enhancer SNAC and the glucagon-like peptide 1 analog semaglutide (SG) used in the recently FDA approved treatment for type II diabetes (Buckley et al., 2018). However, from tracking the transport of SG-Cy3 and the BI by TD, in combination with a direct visual inspection of the cell layer in the bright field channel, we concluded that before any significant semaglutide-Cy3 transport across the barrier could be detected, SNAC had caused irreversible damage to the cell layer (Supplementary Figure S9). This illustrate the substantial impact on barrier morphology and integrity imposed by SNAC and suggests that its in vivo feasibility strongly relies on the regenerative nature of the epithelium cell layer. Overall, these findings illustrate the immense potential of employing imaging modality compatible in vitro intestinal barrier models, offering the ability to simultaneously perform detailed studies on uptake mechanisms and high-quality quantitative kinetic transport measurements and with that obtaining valuable mechanistic insights.