Whole tissue robotic interface system for oral drug formulation development

Abstract

For nearly four decades cancer-derived cell line monolayers have served as the recognized standard for the modeling of gastrointestinal (GI) absorption and have been widely used as a tool for oral drug development. However, they show limited in vivo predictability. We have developed approaches to cultivate porcine GI tissue and enable it to function ex vivo for prolonged periods. We then created an interface design that can achieve fully automated high-throughput interrogation of whole segments of the GI tract. This GI Tract-Tissue Robotic Interface System (GI-TRIS) demonstrated high predictive capacity of human oral drug absorption (Spearman correlation coefficient of 0.906 vs 0.302 for Caco-2 cell-based systems) while allowing a sample throughput of several thousand samples per day in a fully automated robotic facility. To examine the capacity of the GI-TRIS, we analyzed the intestinal absorption of 2930 formulations with the peptide drug oxytocin resulting in the discovery of an enhancer that resulted in an 11.3-fold increase in oral bioavailability of oxytocin in vivo in a large animal model while no disruption of the intestinal tissue was observed. In sum, the GI-TRIS system has the potential to transform oral drug formulation development and introduces the GI-TRIS concept as a pre-clinical strategy for a wide range of applications.

Introduction

Patients and physicians prefer the oral route for drug delivery; however, poor drug absorption in the small intestine presents a significant challenge in drug formulation development^1,2. Currently, there are no high-throughput screening assays enabling interrogation of the small intestine, which is required to predict in vivo intestinal absorption accurately. Many attempts have been made to produce culture systems that mimic normal intestinal epithelial growth and differentiation^3–5. The biggest obstacle has been the rapid initiation of apoptosis within a few hours after intestinal cells are removed from the basement membrane and underlying stroma^6,7. Therefore, tumor-derived cancer cell lines from the colon or intestine such as Caco-2 that are easier to culture compared to primary intestinal cells remain the current standard for in vitro absorption modeling along with artificial membrane systems^8–11. However, their predictability of actual human absorption remains limited^11–13. Organoids developed from intestinal stem cells have shown significant promise in modeling the GI tract including the villus-like architecture^14,15. However, these spherical, micrometer-sized systems present conformational and cost challenges due to their shape and requirement for specific media and supplements respectively^16–18. Recently, technologies using 3-dimensional in vitro cultivation of intestinal organoids demonstrated the ability to form monolayers resembling the native intestinal tissue structure^19–21. While these systems seem very promising for a wide range of applications, the ability to model the complex tissue architecture and drug absorption predictability requires further development. We believe that using a high throughput screening system based on fully intact and functional ex vivo cultured gastrointestinal tissue offers an alternative approach. Based on recent observations related to the requirement for an intact stroma for supporting ex vivo cultivation of murine intestinal epithelium^7,22, we hypothesized that an intact stromal layer could enable maximal intestinal tissue viability and function over an extended period ex vivo. We, therefore, focused on a porcine model due to its recognized similarities to human anatomy and dimensions of tissue that could enable high-throughput scaling^23,24. Furthermore, since the pig genome was sequenced, new opportunities for biological analysis and genetic modification have increased the potential of this model for a wide range of applications²⁵. We developed a culture system to maintain large intestinal tissue explants from pigs ex vivo for an extended period. We then combined these tissue explants with a custom designed interface system. This system, referred to as GI-TRIS, enables high-throughput evaluation of drug transport. We demonstrated the capabilities of the system to enable systematic high-throughput oral drug formulation screening and validated this in vivo in a large animal model.

Results

Ex vivo cultivation system for intestinal tissue

The viability of intestinal explants from pigs was systematically tested and found to be dependent on specific media compositions (Fig. 1a, Fig. S1–2, quantification in Fig. S3) as well as the presence of the stroma layer (Fig. S7). Furthermore, the age of the animal (Fig. S4) influenced long term viability. Explants derived from younger animals without removal of the stroma layer cultivated in a liquid-air interface using advanced DMEM/F-12 were used for further analysis. The villi-crypt topography of intestinal tissue explants cultured ex vivo was analyzed by scanning electron microscopy (SEM) confirming stable structures for approximately 3 weeks with mucus resembling structures (Fig. 1b, Fig. S4). Using the same cultivation conditions, we also analyzed the viability of other ex vivo cultured gastrointestinal segments (Fig. S8). Confocal analysis of sectioned intestinal tissue explants that were incubated for up to 10 weeks ex vivo revealed intact cells up to that time although morphologic changes were observed after 3 weeks. Stromal layer removal prior to ex vivo explant cultivation resulted in the loss of the epithelial layer viability within 5 days of culture, supporting the necessary role of the stromal layer in ex vivo cultivation of a large mammal’s surface intestinal epithelium (Fig. 1c). Intact villi and crypts were successfully isolated from the intestinal epithelium after long term ex vivo culture (3 and 10 weeks). The isolation of intact crypts after ex vivo explant cultivation was found to be dependent on the presence of the stoma layer (Fig. 1d). In order to investigate changes in the intestinal epithelium induced by ex vivo cultivation, the expression level of a broad range of cell type markers and drug transporters in fresh tissue and tissue cultured for 7 days ex vivo was investigated by rtPCR analysis (Fig. 1e). Based on cell type-specific marker expression analysis, various cell types were found to be present in freshly isolated tissue and tissue cultured 7 days ex vivo including: Intestinal stem cells (Leucine-rich repeat-containing G-protein coupled receptor 5 (LGR-5) and Olfactomedin 4 (OLFM-4)), , enterocytes (Villin), and enteroendocrine cells (Glucagon-like peptide-1 (GLP-1) and Gastric inhibitory polypeptide (GIP)), Paneth cells (Lysozyme 1) as well as neuronal cells (Nestin and Synaptophysin (SYP)). Expression level of Mucin 2 (Muc-2) appears to be almost absent after 7 days compared to freshly isolated tissue indicating possible disappearance of goblet cells. In addition, we have analyzed markers for differentiated intestinal epithelium using caudal type homeobox 2 (CDX-2), Keratin 20 (KRT20), Fatty Acid Binding Protein 1 (FABP-1) as well as cell-cell adhesion markers of the intestinal epithelium (Zonula Occludens Protein 2 (ZO-2)) and E-Cadherin as well as various Claudins). We also observed continuous expression of Wnt3a and R-Spondin-1 which are soluble ligands of the canonical Wnt/β-Catenin pathway and reported to play an important role in the maintenance of intestinal stem cell function and viability^7,26.Furthermore, an extensive list of transporters were found to be expressed in the intestinal explants in freshly isolated as well at 7 days ex vivo cultured small intestinal segments: Canalicular multispecific organic anion transporter 2 (ABCC3), P-glycoprotein 1 (MDR-1), Multidrug resistance-associated protein 2 (MRP-2), Peptide transporter 1 (PEPT-1), Breast Cancer Resistance Protein (BCRP), Organic Cation Transporter 1 (OCT-1) and Organic solute transporter subunit alpha (OST-α) and Monocarboxylate transporter 1 (MCT-1). In addition, the expression of small intestinal cytochromes P-450 (CYP3A4) was also investigated. Protein expression of the above-mentioned markers and drug transporters (where antibodies were available) was investigated by western blot analysis confirming their expression in both fresh and cultured tissue (Fig. 1f). Importantly, as in the expression analysis, qualitative analysis of the western blot bands showed no clear change in the protein concentration of various intestinal drug transporters (Fig. 1f). In addition, ex vivo cultured tissue maintained a constant level of Wnt3a secretion according to western blot analysis even up to 3 weeks of ex vivo culture further supporting the presence of Wnt/β-Catenin signaling even during prolonged ex vivo culturing (Fig. 1g). Sectioned intestinal tissue explants were analyzed by immunohistochemistry (Fig. 1h). The Vimentin-mesenchymal cells in the lamina propria were observed by Vimentin staining in both the freshly isolated and 7 day ex vivo cultured tissue. Furthermore, we found similar results in CDX-2 staining in 7 day ex vivo cultured tissue compared to freshly isolated tissue with a characteristic accumulation of CDX-2 signal in the nucleus of the differentiated intestinal epithelium. Other mature intestinal epithelium markers (FABP-1 and KRT20) were found to be specific to the intestinal epithelium but showed various signal intensities between freshly excised and ex vivo cultured tissue. KRT20 shows increased expression after 7 days ex vivo cultivation compared to freshly isolated tissue, which is in line with the rtPCR analysis in Fig. 1e. The reason for these differences is unclear. Importantly, E-cadherin staining reveals intact cell-cell adhesions in tissue explants after 7 days ex vivo culture tissue with no visible difference compared to freshly isolated tissue.

Figure 1.

Characterization of ex vivo cultured intestinal tissue. (a) LiveDead analysis of intestinal tissue explants cultured in different media compositions. Green = Viable cells, Red = Dead cells. Scale bar = 200 μm. ETOH Control = tissue treated with 70% ethanol for 24 hours (negative control). Results were repeated with 3 independent tissue batches from different animals. (b) Scanning electron microscopy (SEM) analysis of ex vivo cultured intestinal epithelium. Scale bar = 200 μm. (c) Confocal analysis of sections of ex vivo incubated intestinal tissue with the stroma layer left intact or removed. The sections were stained with DAPI (cell nucleus, blue), Phalloidin (F-actin, green) Wheat Germ Agglutinin (plasma membrane, red), LysoTracker® (lysozyme, purple). Scale bar = 200 μm. (d) Confocal microscopy analysis of intestinal villi and crypts isolated from fresh intestinal tissue or tissue incubated for 3 or 10 weeks with stroma layer intact or 10 days without stroma layer. Tissue was stained for DAPI (blue) and Phalloidin (green). Scale bar = 500 μm. Results shown in (b)-(d) were performed with one tissue batch. (e) Gene expression analysis of porcine intestinal tissue that was incubated for 7 days ex vivo or freshly isolated at day zero. (f) Western blot analysis of lysates of intestinal tissue cultured for 0 days (fresh) or 7 days ex vivo (g) Western blot analysis of secreted Wnt3a of tissue explants. Results shown in (e)-(g) were repeated with 2 independent tissue batches in total. (h) Immunohistochemical analysis of sections of tissue that was incubated for 7 days ex vivo or freshly isolated. Brown color shows antibody signal; blue color is the counterstain. Scale bar = 100 μm. Results shown were performed with one tissue batch. (i) Protease, Thioreductase, P450 CYP3A4, UGT activity of tissue cultured for 0 (fresh tissue) or 7 days ex vivo. Results represent 3 independent experiments (n=3) for protease activity assay; 3 independent experiments performed in duplicate for CYP3A4 assay; 5 and 4 independent experiments with 1 replicate for Thioreductase activity assay for day 0 and day 7 respectively; 4 and 5 independent experiments performed in duplicate for UGT activity assay for day 0 and day 7 respectively. For all graphs in (i) Error bars show S.D. Line shows mean value. Unpaired two-tailed t-test was used as statistical method. *** equals p < <0.0001 with 95% confidence intervals 0.03699 to 0.05647. For (a-d) and (h) one image was taken per condition.

Wnt3a was also found to be present in the intestinal crypts as observed by histological analysis. The tissue stains Masson’s Trichromes as well as Alcian Blue Periodic Acid Schiffs were used to analyze the overall tissue architecture of freshly isolated and ex vivo cultured tissue. Bioactivity analysis of thioredoxin reductase, which is a ubiquitous enzyme involved in many cellular processes such as cell growth, and protection against oxidative stress, showed no significant differences between fresh tissue lysates or tissue cultured ex vivo for up to 7 days. Also, no significant difference was found in protease activity between fresh tissue and tissue cultured ex vivo for 7 days. Furthermore, we confirmed the activity of the metabolizing enzymes Cytochrome P450 3A4 (CYP3A4) and Uridine 5’-diphospho-glucuronosyltransferase (UGT). Interestingly, the activity appears to fluctuate between freshly excised and 7 days ex vivo cultured tissue (Fig. 1i).

GI-TRIS technology development

Drug formulation scientists have been using animal-derived intestinal tissue for decades to evaluate drug transport including the everted gut sac, the Ussing- and Franz Diffusion chambers and modified systems incorporating these technologes^27–30. However, so far there is no system available using intestinal tissue that enables high-throughput screening. The possibility of culturing large areas of pig intestinal explants for an extended period without any significant change in tissue viability and architecture enables possibilities for in vitro assay development, in particular for high-throughput screening experiments as well as assays that require long-term incubation. To interrogate the explants in a high-throughput manner, we designed an interface device for intestinal drug transport measurement combined with long-term tissue culturing capability (Fig. 2). We systematically assessed a broad range of different designs and materials for potential interfacing systems and developed a system that maintains tissue viability and low sample variability while enabling rapid assembly (Fig. 2a). The design consists of an upper device that compartmentalizes the intestinal tissue in a 96 multiwell plate format. The tissue forms the bottom of the multiwell plate and is sealed off around each of the 96 wells with an additional device underneath the tissue (detailed design drawings in Fig S10). For transport or transport experiments this device design can be mounted on any commercially available 96 multiwell plate filled with reaction buffer. This system is enclosed by a case that enables adjustable pressure to maintain the system in position. Fig. 2a shows a schematic of the entire device before (upper image) and after assembly (lower image). After adding a drug formulation to the upper plate, the concentration difference between the upper and lower plate is measured over time to calculate the rate of transport of the drug through the explanted intestinal tissue. The design, geometry and compression force of the device was systematically optimized by transport and well-to-well leakage experiments using porcine-derived jejunum tissue (Fig. 2). Development of the interface between upper and lower segments of the device included identification of optimal well sizes of the upper plate (Fig. 2b) as well as well sizes of the lower device (Fig. 2c). Based on these experiments, optimal design and geometry were chosen (specific dimensions shown in Fig S10). Besides the device geometry, the compression of the system was optimized in order to avoid well-to-well leakage. To understand the effect of mechanical pressure on the tissue, the tissue-device interaction in the GI-TRIS system was modeled using finite element analysis. The results revealed a non-uniform strain distribution in the tissue due to the mechanical contacts. The tissue area located within each well, however, was clearly less mechanically affected compared to the tissue surrounding each well (Fig. 2d). Based on this we defined a range of pressures applied to the system and tested the well-to-well leakage when conducting transport experiments with a model drug (FITC) (Fig. 2e). We then tested the experimental variability of transport experiments using fluorescently conjugated dextrans with different molecular weights. (Fig. 2f). Based on these results we chose a force of 5N to compress the interface system. After optimizing the GI-TRIS geometry and compression force, we then conducted transport experiments with model compounds across a range of molecular weights and with intestinal tissue from different animals, different regions within the jejunum, and different incubation times in order to understand the variability between different animal batches as well as within the jejunum (Fig. 3). Combining a large number of individual measurements from 6 different animals in one data set resulted in a coefficient of variation between 27-43% depending on the model drug analyzed (Fig. 3a). Using various fluorescently labelled model compounds, we demonstrate that the device can be used for serial sampling for kinetic drug transport experiments. (Fig. 3b). Transport experiments with cultivated porcine-derived jejunum in a liquid-air interface using advanced DMEM/F-12 did not change the permeability of various model compounds over the duration of 7 days. (Fig. 3c). Using the same system, we also tested the transport of model drugs with other segments from the gastrointestinal tract including stomach, duodenum and colon (Fig. 3d–e). This GI-TRIS system can also be applied to a 384-well plate setup with a coefficient of variation of 33% and 32% for transport measurements with FITC and FITC-Dextran 4 kDa respectively (Fig. 3f–g). Expression analysis using the same markers as in Fig. 1e confirmed that ex vivo cultivation porcine-derived jejunum tissue was unchanged between standard cultivation shown in Fig. 1 and cultivation in the GI TRIS device (Fig. S11).

Figure 2.

GI-TRIS interface device development. (a) Schematic of the 96-well plate device set-up used as interface design for the intestinal tissue culture system for high-throughput assays. The design consists of an upper device that compartmentalizes the intestinal tissue in a 96 multiwell plate format. The tissue forms the bottom of the multiwell plate and is sealed off around each of the 96 wells with a lower device underneath the tissue. This system is enclosed by a case that enables adjustable pressure to maintain the system in position. (b-d) Graph of well-to-well leakage experiments using FITC as a model drug. The % leakage is defined by amount of FITC found in wells filled with PBS that are adjacent to well filled with FITC. The leakage as a function of different design parameters was tested (varying upper plate well sizes (b, right panel); lower plate hole sizes (c, right panel), and compression forces(e)). For (b) graph shows n = 71, 69 and 60 for 3 mm, 6 mm and 9 mm respectively. For (c) graph shows n = 51, 72, 66, 71 and 84. For (e) n= 131, 116 and 123 for 20N, 5N and 0N respectively. For (b), (c) and (e) red line shows mean. (d) Computational model of strain on tissue as a function of tissue displacement in interface design. (f) Effect of tissue compression on transport of FITC and FITC conjugated dextran with different molecular weight (4, 20, 70 kDa) in the GI-TRIS device. Sample size is as follows: FITC permeability n = 86, 92 and 90 for 20N, 5N and 0N respectively; FITC-Dextran 4 kDa permeability n = 96 for 20N, 5N and 0N respectively; FITC-Dextran 20 kDa permeability n = 96 for 20N, 5N and 0N respectively; FITC-Dextran 70 kDa permeability n = 95, 90 and 77 for 20N, 5N and 0N respectively. Red line shows mean. (g) Schematic illustration of the cross section of the interface design in 96 well plate format.

Fig. 3. GI-TRIS transport variability validation.

(a) Variability analysis of jejunum permeability of different model drugs (n = 529, 348, 398, 249 and 273 from 6 different animal batches for FITC, FITC-Dextran 4 kDa, FITC-Dextran 20 kDa, FITC-Dextran 70 kDa, Alexa488-Oxytocin and Alexa488-Insulin. Red line shows mean. (b) Jejunum permeability time lapse analysis of model drugs during 4 hours incubation (n = 94) (blue to red = increasing density of overlapping curves). (c) Permeability of model drugs using porcine jejunum tissue incubated ex vivo prior to experiment (from 3 different animal batches). N= 273, 286, 286 and 278 for FITC 1 day, 2 days, 3 days and 7 days respectively; n = 288 for Alexa488-Oxytocin; n = 192, 191 and 186 for FITC-Dextran 70 kDa for 1 day 3 days and 7 days respectively. (d) Hematoxylin and eosin stain of sections of porcine derived colon, duodenum and stomach tissue. Section (I) indicates mucosa and (II) submucosa and underlining muscle layer. Scale bar = 400 μm. Experiment was performed once. (e) Variability analysis of permeability of the model drugs FITC (left), and FITC-Dextran 4kDA (right) (both n = 96). (f) Schematic of the GI-TRIS interface device in 384-well plate set-up (g) Jejunum permeability variability analysis of model drugs in 384-well plate device. N = 298 and n = 297 for FITC and FITC-Dextran 4 kDa respectively. Red line in all graphs shows mean.

GI-TRIS system intestinal absorption validation

The Food and Drug Administration (FDA), recommends using model drugs with well-established human clinical pharmacokinetic data to validate the in vivo predictability of in vitro intestinal transport test systems³¹. Following this guidance, we analyzed the transport of 60 model drugs from the 4 Biopharmaceutical Classification System (BCS) classes on the GI-TRIS system (16 BCS class I, 13 BCS class II, 15 BCS class III, 12 BCS class IV drugs, 4 dextran-based control substances) (Fig. 4a) with the optimized parameters described in Fig 2. The results combine transport data from 6 different animals to analyze the batch-to-batch variability. Importantly, this drug panel was systematically assembled to represent the same distribution and variability in drug properties as to the entire list of the FDA approved drugs for oral administration (Fig. S13).

Figure 4.

In vivo absorption predictability analysis using model drugs. (a) GI-TRIS transport analysis of 60 model drugs. The box plot shows the calculated apparent permeability (P_app) of 6 independent experiments performed in duplicate. Color code indicates reported human absorption of drugs (blue= high, green = moderate, purple = low). Red color shows controls. Bars show minima and maxima, box shows 25% and 75% percentiles, square shows mean, x shows 1% and 99% value. (b) Intestinal transport time lapse analysis of various model drugs known to have either slow, moderate or rapid intestinal absorption to assess human absorption kinetics predictability using the GI-TRIS system. The graphs show 12 individual time lapses (n = 12) over the time course of 4 hours. (c) GI-TRIS transport analysis of 38 model drugs in presence of native intestinal fluid. The box plot shows the calculated apparent permeability of model drugs in intestinal fluid (P^Mucus) relative to the apparent permeability in PBS (P^PBS) intestinal fluid (P_app) of 3 independent experiments independent experiments. Bars show minima and maxima, box shows 25% and 75% percentiles, square shows mean, x shows 1% and 99% value. (d) Measured fluorescent intensity in the GI-TRIS receiver chamber of a substrate of MDR-1 indicating substrate permeability in presence of various concentrations of the MDR-1 inhibitor verapamil.

For quantitative detection of each model drug a spectrophotometric detection method and a calibration curve was established (Fig. S9). For intestinal absorption predictability analysis, we have used jejunum segments in the GI-TRIS system; however, we also report similar results using other parts of the small intestine (Fig. S14). Analysis of absorption kinetics of model drugs using the GI-TRIS system reveals the predictive capacity of reported human absorption kinetics (slow, moderate and rapid absorption) (Fig. 4b). Another important aspect of drug transport assays is the ability to model drug-food interactions since food can have a substantial effect on intestinal absorption³². Unfortunately, the Caco-2 Transwell transport assay is not compatible with native intestinal fluid including digested food, mucus, and bacteria³³. However, the GI-TRIS system allows transport experiments in native intestinal fluid as demonstrated by multiple independent experiments with a range of different small molecular model drugs (Fig. 4c). The results reveal modulation of intestinal absorption of most analyzed drugs in native intestinal fluid compared to normal transport medium underlying the importance of analyzing drug absorption in the presence and absence of food. Furthermore, we have found that the system can be used for MDR-1 drug transporter inhibitor studies using a fluorescent MDR-1 specific substrate. MDR-1 is a drug efflux transporter affecting absorption of a wide range of drugs. We observe clear dose-dependent effects in intestinal uptake of the fluorescent substrate using the established MDR-1 inhibitor verapamil (Fig. 4d).

The intestinal transport data obtained by the GI-TRIS system (shown in Fig. 4a) was compared to intestinal absorption in humans based on previously published data for 55 drugs with human absorption data available (Fig. 5a, Supplementary Table 1). We report an excellent correlation between the average intestinal transport values obtained by the GI-TRIS system and the reported human data (Spearman correlation coefficient: 0.906) demonstrating near perfect in vivo predictability. For comparison purposes, we compared the in vivo predictability of the GI-TRIS system to the Caco-2 Transwell assay, which is one of the most commonly used in vitro intestinal transport assay systems. Comparing the average Caco-2 permeability values for each drug against the human absorption data resulted in a Spearman correlation coefficient of 0.302 (Fig. 5b). In order to conduct an unbiased comparison, Caco-2 permeability values used for the correlation analysis were based on a systematic literature analysis of Caco-2 Transwell drug permeability for each drug in our drug panel (Fig. 5c, overview of analysis shown in Fig. S12, individual values and publications listed in Supplementary Table 2). Interestingly, the results revealed drug-specific variability that cannot be fully explained by variation in sample size (Fig. 5c) or the experimental parameters of the Caco-2 Transwell drug permeability experiment (Supplementary Table 3). It has been suggested that inherent genomic instability of the tumor derived Caco-2 cell line could increase variability in transport caused by variable expression levels of drug transporters¹². Indeed, the average coefficient of variation (CV) is approximately two-fold higher in drugs actively absorbed compared to the passively absorption ones (Fig. 5d). In contrast, the CV across 6 independent animals in the GI-TRIS system is similar between active and passively absorbed drugs, suggesting that GI-TRIS maintains consistent expression of drug transporters and avoids the issues with genomic instability observed with Caco-2 culture.

Figure 5.

In vivo absorption predictability comparison to Caco-2 Transwell system. (a) Correlation of the average P_app obtained from the GI-TRIS transport analysis with the human intestinal absorption data from literature. (b) Correlation of the average P_app obtained from the Caco-2 Transwell literature analysis with the human intestinal absorption data from literature as in (a). (c) Box plot of the Caco-2 Transwell apparent permeability (P_app) obtained by systematic literature analysis. Red bubble size shows number of individual studies found per drug (sample size), blue bubble size shows the coefficient of variation (CV) of the data. Bars show minima and maxima, diamond box shows 25% and 75% percentiles, square shows mean, x shows 1% and 99% value. N (from left to right): 16, 7, 22, 54, 5, 1, 16, 14, 1, 1, 11, 4, 9, 5, 1, 6, 7, 29, 4, 10, 13, 43, 5, 23, 10, 1, 3, 6, 4, 43, 1, 18, 3, 9, 11, 11, 60, 22, 1, 29, 7, 24, 2, 15, 14, 6, 38, 10, 3, 14, 18, 15, 2, 1. (d) Coefficient of variation (CV) of the Caco-2 Transwell P_app values as in (c) and the GI-TRIS P_app values for the model drugs categorized according to active or passive mode of human intestinal absorption. N = 14 and 16 for active and passive transport Caco-2 respectively and n= 26 and 27 for active and passive transport GI-TRIS. Red line shows mean.

(Fig. 5d). In contrast, the CV across 6 independent animals in the GI-TRIS system is similar between active and passively absorbed drugs supporting the hypothesis that the genomically unstable background appears to increase the variability of actively transported drugs (Fig. 5d). Drug-Transporter interactions reported in the literature were systematically analyzed to classify the drugs as active or passively absorbed (Table S4). For comparison purposes, we have also used computational analysis techniques to predict intestinal absorption of model drugs namely admetSAR, the model employed by drugbank.ca, as well as prediction based on Lipinski’s “rule of five” violations³⁴. For the set of model drugs analyzed both methods showed any correlation to human absorption data suggesting very low predictive capabilities of these systems (Fig. S15 and Supplementary Table 4).

Fully automated screening for oxytocin oral formulation development on the GI-TRIS

To conduct fully automated screening experiments that include robotic handling, we have developed an additional GI-TRIS design that uses custom designed plates in standard well plate formats that confine intestinal tissue in sealed wells by magnetic compression (Supplementary Fig. 30). The weight, dimension, and shape of these plates were specifically designed to fully interface with a robotic screening technology including a liquid handling station, automated well plate storage station, well plate reader, and 6-axis industrial robot that enables automated transfer of GI-TRIS plates to each station (full design drawings are shown in Fig. S16). The system could also be used with an automated incubator to conduct transport studies at 37 degrees Celsius as opposed to room temperature. Automated transport screens were conducted with model drugs to establish variability and reproducibility (Fig. S17) and to measure the effect of the inner muscle layer on drug transport (Fig. S18). This system also enables detection over time without disassembling the device and can, therefore, be used for kinetic absorption evaluations. To examine the potential of the GI-TRIS system for oral formulation development, we conducted a large-scale absorption screen for the peptide drug oxytocin with a total number of 2930 co-excipient formulations in aqueous solution from diverse chemical groups. In order to overcome the poor oral bioavailability of oxytocin³⁵, formulations that enhance intestinal absorption would be required. To date, no effective oral oxytocin absorption enhancer has been reported. Our screening result revealed that a large range of chemicals could modulate intestinal absorption of Alexa488-conjugated oxytocin substantially. The results of the fold change of Alexa488-oxytocin intestinal transport compared to the non-formulated Alexa488-oxytocin control are summarized as a Heatmap graph (Fig. 6). Furthermore, we analyzed the effect of specific excipient combination by a network analysis (Fig. S19) as well as by analyzing the Pearson correlation coefficient (Fig. S20). Additional validation experiments were performed for the initial screening of hits including measurement of the excipient alone to ensure that the accumulated excipient in the receiver chamber was not interfering with the fluorescence signal of oxytocin (Fig. S22). Furthermore, using a sub-panel of formulations, the correlation between fluorescence detection of Alexa-488 conjugated oxytocin and ELISA detection of unlabeled oxytocin was investigated in order to study the effect of drug degradation, which is not detectable by fluorescence detection (Fig. 7a). Alexa-488 fluorophore conjugation could potentially also affect the formulation dependent intestinal transport. Therefore, we also investigated the intestinal transport of co-formulations using either Alexa-488-oxytocin or native oxytocin by oxytocin ELISA detection (Fig. 7a). The correlation with the actual formulation names is shown in Fig. S21. The polyethyleneimine (800 Da, end-capped) excipient was tested with other model drugs including Alexa488-teicoplainin, Alexa488-carbetocin, Alexa488-insulin and appeared to enhance the transport of some of these across the intestinal explants (Fig. 7b, other polyethyleneimine variants were also tested Fig. S23). Interestingly a limited increase in intestinal transport was observed for the oxytocin analogue carbetocin when formulated with polyethyleneimine (Fig. 7b). Polyethyleneimine (800 Da, end-capped) consistently increased the intestinal transport of Alexa-488-oxytocin several fold in a concentration-dependent manner (Fig. 7c) and showed no detectable cell toxicity over 2 hours of incubation (Fig. S24). In vivo pharmacokinetic studies of the identified absorption enhancer were performed in pigs. Orally administered formulations were compared against non-formulated oxytocin as well as the formulation without oxytocin. In order to calculate the absolute oral bioavailability, pharmacokinetic experiments with intravenously administered oxytocin were performed (Fig. 7d). The absolute oral bioavailability of oxytocin alone was found to be 0.58% while the oxytocin-polyethyleneimine (800 Da, end-capped) formulation resulted in average oral bioavailability of 6.54% showing an 11.3-fold enhancement compared to the non-formulated oxytocin control (Fig. 7d, Fig. S25). In order to investigate the mechanism of absorption enhancement and to analyze potential local effects of the formulation on the small intestine histologically, a surgical procedure coupled with custom-made devices was developed that enables controlled exposure of the formulation on a defined area of intestinal tissue in vivo in pigs (Fig. 7e). Obtained in vivo biopsy samples from the small intestine exposed to the liquid formulation were subsequently analyzed histologically. E-cadherin stained sections of these biopsy samples were used in order to analyze the disruption of cell-cell adhesions qualitatively (Fig. 7f). Intestinal epithelium exposed to mixtures of oxytocin-polyethyleneimine (800 Da, end-capped) appeared to show intact cell-cell adhesions. There is no quantitative assessment that enables comparison to non-formulated oxytocin but the results indicate that the maintenance of oxytocin-polyethyleneimine (800 Da, end-capped) formulation does not result in complete disruption of cell-cell adhesions in the small intestinal epithelium. Furthermore, this formulation was not observed to induce significant loss of intestinal epithelial cell barrier function in the well-established Caco-2 Transwell system measured by Transepithelial Electrical Resistance (TEER) (Fig. S29).

Figure 6.

GI-TRIS intestinal transport screening of Alexa488-oxytocin formulations. Heatmap of the data obtained by GI-TRIS transport screening of Alexa488-oxytocin formulated with 2930 formulations that are combinations of 2 excipients. The color code indicates fold change compared to unformulated control (red = negative change, black = no change, green = 3-fold increase, blue more than 3-fold increase). Experiment was performed once.

Figure 7.

Validation experiments and In vivo pharmacokinetic and histological analysis of discovered oxytocin formulations. (a) Heatmap showing the fold change of intestinal absorption of a panel of formulations with Alexa488-oxytocin or native oxytocin analyzed by different readouts. The results are averages of 3 independent experiments. (b) Transport analysis of PEI formulations with various small-or macromolecular model drugs. Red line shows mean, error bars are showing standard deviation. Results are from 3 independent experiment performed in duplicate 2 experiments and for 1 experiment with 1 replicate. (c) Dose-dependent experiment of polyethyleneimine (PEI) used as Alexa488-oxytocin transport enhancer. Red line shows mean. For (b-c) results show 3 independent experiments. (d) Area Under the Curve of oxytocin plasma concentration over a time period of 2 hours (AUC_0-2h) of PEI-oxytocin, oxytocin only, PEI only and IV administered oxytocin based on in vivo pig pharmacokinetic experiments (n = 4 for IV and PEI only, n = 5 for PEI-Oxytocin, n = 5 for Oxytocin only). Oxytocin dosed for per oral (PO) administration was 10IU and 2IU for IV administration. Red line shows mean and error bars standard deviation. ** equals p = 0.0140 using an unpaired two-tailed t test. Confidence intervals: −109010 to −16558. (e) Image showing the surgical procedure and device set up that was used in order to expose liquid oxytocin formulations to the small intestinal tissue in vivo in pigs for systematic histological assessment of biopsy samples. (f) E-Cadherin stained sections of tissue biopsy samples obtained from the small intestine after the surgical procedure shown in (e) and subsequently imaged via epifluorescence microscopy. Scale bar = 20 μm. DAPI = blue, e-cadherin staining = red. Experiment was repeated once. One image obtained per condition.

Discussion and Conclusion

Our extensive biological characterization of the intestinal GI-TRIS system, supports that under the developed procurement, isolation and cultivation techniques, cell viability of several weeks can be achieved (Fig. 1a–d). Clear histological changes can, however, be observed beyond 1 week of ex vivo culture (Fig S6). The ex vivo viability of the intestinal tissue was found to be dependent on cultivation conditions, medium composition, and presence of the underlying stroma as well as animal age (Fig. 1a–d, Fig. S5). Based on the extensive characterization of 7 day ex vivo cultured intestinal explants we observe changes of certain cell markers at the protein and mRNA level compared to freshly isolated tissue indicating that the ex vivo cultivation potentially affects various aspects of intestinal tissue morphology and function (Fig. 1e–f). However, overall the data suggest preservation of the differentiated intestinal epithelium including its barrier function as well as the metabolizing enzymes, drug transporters, and lamina propria. MUC-2 expression analysis suggests that ex vivo cultivation adversely affects goblet cells. Interestingly, the expression analysis suggests the preservation of neuronal components as well as intestinal stem cells during ex vivo cultivation. The latter is further supported by the observed continuous expression and secretion throughout ex vivo cultivation of the soluble factors Wnt3a and R-Spondin-1 which have previously been shown to be required for maintaining intestinal stem cell viability and function^7,26. Given that these factors are secreted from the stromal layer⁷ and removal of this layer has shown a dramatic reduction of ex vivo viability (Fig. 1c–d), we hypothesize that the prolonged ex vivo cultivation is at least partly caused by ex vivo maintenance of intestinal stem cells. Activity assays suggest higher enzyme activity of cytochrome P450 3A4 in 7-day ex vivo cultured compared to freshly isolated tissue (Fig. 1k). This difference cannot be explained by the gene expression level of cytochrome P450 3A4 and might be caused by other factors not related to gene expression. Further approaches to extend the time limit of ex vivo cultivation could include mechanical stimulation and integration of a blood supply via a transport system. We anticipate the widespread application of the presented intestinal ex vivo cultivation system beyond drug absorption analysis. In particular it could be used for studies that require functional and viable tissue for extended periods including modulation of gene expression as well as potential analysis of tissue toxicity and enteroendocrine functions. Furthermore, while using freshly isolated tissue for transport studies is an option in certain cases, ex vivo cultivation of porcine GI tissue enables to conduct large screening experiments which require longer viability of the tissue since it is not feasible to perform these experiments in a matter of hours. Previous efforts to develop systems for high throughput assessment of drug permeability are based on using artificial membranes in a multi-well plate format^36,37. While these systems show greatly increased sample throughput compared to the Caco-2 Transwell system, they are a cellular and unable to model active transport. Therefore Caco-2 Transwell systems have remained the industry standard. In this work, we report 2 different gastrointestinal tissue interface designs described in Fig. S10 and S16 that support high throughput screening of fully intact intestinal tissue. The first design presented can be combined with standard commercially available well plates and can be easily transferred between different assay plates. The second device is designed not to be transferred between assay plates but can be used directly in a spectrophotometer without disassembly and is fully compatible with a robotic system. Importantly, this enables fully automated screening experiments including measuring absorption kinetics with a sample throughput of several thousand samples per day when a suitable animal procurement network, as well as a customized automated robotic facility, are available. We have performed extensive validation experiments in order to analyze the variability, well leakage and reproducibility of permeability measurements using different design parameters of the GI-TRIS interface system including the compression pressure applied as well as the presence of the inner muscle layer (Fig. 2 as well as Fig. S17 and S18). While ex vivo cultivation of intestinal tissue did not change absorption measurements using various model drugs, it is conceivable that the changes induced by ex vivo cultivation changes drug transport beyond the model drugs measured.

The human in vivo drug absorption predictability was investigated using model drugs with known human absorption. Human absorption values are based on a fixed drug dose that was used in human testing. It is possible that the drug absorption changes depending on the dose administered. We hypothesize that the observed difference in performance concerning the human absorption correlation of the GI-TRIS system compared to the Caco-2 Transwell system stems from the organization and architecture differences between the two systems as well as the genomic instability influencing the expression of drug transporters. Previous studies support our observation of a high discrepancy in drug transport between intestinal tissue and Caco-2 monolayers^38,39. In addition, there is evidence that these differences could also be caused by the tumorigenic origin and variation in clonality of Caco-2 cells^40,41. Primary intestinal epithelial cells show very different transport compared to Caco-2 cells and other tumorigenic cell lines⁴² and have previously been found to show much greater correlation to human data using a panel of model drugs⁴³. Additionally, it was previously reported that tight junctions, relevant for paracellular transport, are clearly different in the Caco-2 Transwell model compared to native human jejunum⁴⁴. Furthermore, Caco-2 cells are unable to accurately model drug transporter based drug uptake as well as the drug metabolism that takes place in intestinal tissue^42,45. Given the high variability in Caco-2 permeability data and in vivo correlation reported in literature⁴⁶ we have conducted a systematic meta-analysis and averaged all values obtained for each model drugs in order to ensure an unbiased comparison. It should be emphasized that for certain model drugs some laboratories have reported clearly better in vivo correlation using the Caco-2 Transwell system compared to the results in our meta-analysis^10,47. Our emphasis, however, was to ensure a systematic and unbiased analysis of the currently available data in the literature. While we were unable to find a correlation between high variability of reported Caco-2 drug permeability values with the sample number of differences in the experimental procedure, we observed higher variability in actively transported compared to passively transported drugs indicating variation in expression levels of drug transporters in Caco-2 cells. We have confirmed the presence of MDR-1 in ex vivo cultured tissue (Fig. 1e, 1f and 1g) and shown that the GI-TRIS system can be used for MDR-1 drug inhibitor studies (Fig. 4d) suggesting further applications such as using the GI-TRIS to knock-down MDR-1 for studying MDR-1 mediated drug absorption. Similarly, the ability to model CYP mediated drug metabolism could be an interesting application of the GI-TRIS system to investigate and validate further.

The present work focuses on an initial characterization of the GI-TRIS system and its application for high throughput drug transport studies with specific focus on absorption enhancement of peptides. We believe the GI-TRIS system could also be a valuable tool for solubility and dissolution optimization of poorly soluble drugs across the GI tract due to its ability to tolerate native GI fluids as well as high local drug concentrations due to its mucus layer and native GI tissue architecture compared to traditional cell-based models.

We applied the GI-TRIS system for systematic formulation screening of oral absorption enhancers for Alexa488 labelled oxytocin to demonstrate the potential to conduct fully automated robotic intestinal absorption screens with a high-throughput. To date, we have limited systematic understanding of how chemicals can affect intestinal absorption. By measuring the intestinal absorption of 2930 Alexa-488-oxytocin co-formulations (Fig. 6), a broad range of previously unidentified chemicals were found to affect oxytocin drug absorption substantially. The identified absorption enhancers show a diverse range of chemical properties indicating that the observed absorption enhancement is caused by various mechanisms based on complex interactions between drug, excipient and tissue and there is a wealth of chemicals that could potentially be used for intestinal absorption enhancement that have yet to be explored. Considering that excipients could potentially damage the mounted intestinal tissue and cause increased drug permeability, introducing an internal permeability control would be a helpful improvement for subsequent studies. Furthermore, given the ability of ex vivo cultivation, the GI-TRIS assay could potentially be used to conduct simulations transport as well as toxicity analysis in order to select for excipient combinations that increase transport but show no toxic effect on tissue. Similarly, genetic modification of drug transporters would enable to assess the effect of excipients on the drug transporters.

Based on the screening results we selected a specific polyethyleneimine polymer as an oxytocin absorption enhancer for further validation in vivo in pigs and confirmed 11.3 fold enhancement in oxytocin bioavailability compared to the non-formulated control. PEI is a well-known transfection reagent⁴⁸ but has not yet been reported to act as a peptide absorption enhancer. We demonstrate that PEI-oxytocin formulations do not disrupt the intestinal epithelium in vivo, which could be an important safety concern. Future work aiming to develop and test solid dosage forms of oxytocin as well as detailed mechanistic analysis of intestinal absorption enhancement could pave the way towards a potential drug candidate. Overall, we anticipate the capacity to interrogate the GI tract in a high-throughput system could radically accelerate the development of therapeutics and moreover expand our understanding of this environment.

METHODS

Tissue Dissection and Cultivation

All animal tissue procedures were conducted in accordance with protocols approved by the Massachusetts Institute of Technology Committee on Animal Care. Small intestinal tissue was isolated from freshly procured intact gastrointestinal tracts from pigs from selected local slaughterhouses. A stretch of the tissue was cut out of the GI tract and dissected longitudinally. Depending on the experiment, the tissue was either left with all layers intact, or the outer muscle layer and serosa were removed. For standard transport experiments jejunal tissue, defined as 50 cm away from the pylorus, was used. The tissue was washed in a series of saline solutions supplemented with 5% Antibiotic-Antimycotic solution (Cat. nb.15240062, Thermo Fisher Scientific) under sterile conditions. The tissue was then either mounted on the GI-TRIS device or kept in cell strainers (Falcon™ Cell Strainers, Mesh size: 100 μm, Thermo Fisher Scientific) depending on the experiment. For cultivation, the following media and supplements were used: Dulbecco’s Modified Eagle Medium (DMEM) high glucose (Lifetechnologies, cat. no. 11965084), DMEM, high glucose, HEPES (Lifetechnologies, cat. no. 12430054), DMEM, high glucose, no glutamine (Lifetechnologies, cat. no. 11960044), DMEM, high glucose, pyruvate, no glutamine (Lifetechnologies, cat. no. 10313021), Advanced DMEM/F-12 (Lifetechnologies, cat. no. 12634028), MEM Non-Essential Amino Acids Solution (Lifetechnologies, cat. no. 11140050), EGF Recombinant Human Protein (Lifetechnologies, cat. no. PHG0311), Fetal Bovine Serum, certified, US origin (Lifetechnologies, cat. no. 16000044). It was determined that the optimal media combination was Advanced DMEM/F-12 media. Tissue was cultured in this serum-free cell culture media in a sealed polystyrene container that was then stored in an incubator kept at 37°C no supplement gas was provided. We have conducted a set of experiments with fixed time points (Fig. 1e–h) as well as longer term experiments monitoring the tissue daily with analysis performed at varying time points. Time points with the greatest differences are represented in Fig. 1a–d. For biological characterization, intestinal crypts and villi were isolated based on a previously published protocol⁴⁹. For cytotoxicity studies of PEI formulations, HT29-MTX-E12 cells were purchased from European Collection of Authenticated Cell Cultures (ECACC) (Cat. Nb. 12040401) and cultured under standard cultivation conditions (37°C, 5% CO₂) in DMEM high glucose pyruvate (Lifetechnologies, cat. no. 11995-065) with 1% Gibco MEM Non-Essential Amino Acid Solution (Lifetechnologies, Cat # 11140-050), 1% Pen/Strep (Lifetechnologies, Cat # 15140122), 10% FBS (heat inactivated) (Lifetechnologies, Cat # 10082-147). C2BBe1 [clone of Caco-2] cells were purchased from ATCC (ATCC® CRL-2102™) and cultured under standard cultivation conditions (37°C, 5% CO₂) in DMEM high glucose pyruvate (Lifetechnologies, cat. no. 11995-065) with 1% Human Transferrin-insulin-Selenium (ITS-G) 100x (Lifetechnologies, Cat # 41400-045), 1% Pen/Strep (Lifetechnologies, Cat # 15140122), 10% FBS (heat inactivated) (Lifetechnologies, Cat # 10082-147). All cells tested negative for mycoplasma contamination.

Reagents

Human oxytocin (synthetic, O3251-5000IU, Sigma), human insulin (recombinant, Cat. no. I2643-25MG, Sigma), teicoplanin (recombinant, Cat. no. T0578, Sigma), and carbetocin acetate (synthetic, SML0748, Sigma) were labelled using A-20000 Alexa Fluor® 488 NHS Ester (Succinimidyl Ester) labelling kit prior to usage. Label IT® RNAi Delivery Control Cy®3 was purchased from Mirus Bio. Additionally, the following model drugs were all purchased from Sigma: antipyrine, beta carotene, danazol, verapamil, ivermectin, metropolol, naproxen, oseltamivir phosphate, memantine, entecavir monohydrate, emtricitabine, ergotamine D-tartrate, labetalol, ketoprofen, desipramine, moxifloxacin, carbamazepine, atorvastatin, domperidone, piroxicam, ibuprofen, theophyline, propanolol, mesalamine, caffeine, phenytoin, valacyclovir, coumarin, doxycycline, metformin, fluvastatin, terbutaline, warfarin, indomethacin, acyclovir, chlorpheniramine, saquinavir, rosuvastatin, quinine, quinidine, furosemide, ranitidine, chlorotetracycline, dihydroergotamin-tartrate salt, amiloride, omeprazole, atenolol, famotidine, curcumin, FITC Dextran 4 kDa, FITC Dextran 20 kDa, FITC Dextran 70 kDa, and fluorescein. The following chemicals that were tested as formulation excipients were all purchased from Sigma: δ-decalactone, 2-Phospho-L-ascorbic acid trisodium salt, 4-arm PEG, 8-arm PEG, 4-(dimethylamino)pyridine, 6-O-Palmitoyl-L-ascorbic acid, acesulfame K, adipic acid, agar, agarose, albumin (bovine serum), alginic acid sodium salt (brown algae), alginic acid sodium salt (kelp), alpha cyclodextrin, bacitracin, B-alanine, B-Cyclodextrin, BD PuraMatrix Peptide Hydrogel, bentonite, caffeine, Carbopol 934, carboxymethylcellulose, Carnuba wax No. 1 yellow, castor oil, cellulose acetate, cellulose acetate phalate, cellulose acetate propionate, chitin (from shrimp shells), chitosan (high and medium molecular weight), cholesterol, citric acid, corn oil, cottonseed oil, cysteamine, d (+)-mannose, d(−)fructose, d(+)glucose, d(+)trehalose dihydrate, dextran, dextrose, d-lacititol, d-leucine, dl-tartaric acid, d-mannitol, d-Methionine, d-Tryptophan, Dynasan 118 Microfine, edetate dissodium, ethylenedinitrilol-tetracetic acid disodium salt, l(+)arabinose, laponite, l-arginine, l-ascorbic acid, l-cysteine hydrochloride monohydrate, lecithin, l-histidine, locust bean gum from ceratonia siliqua seeds, l-phenylaline, l-proline, l-threonine, meglomine, miglyol 812, Mineral Oil, Mowiol 10-98, Mowiol 18-88, Mowiol 4-98, Mowiol 56-98, Mowiol 8-88, mucin (porcine stomach), neocuproine, parrafin wax, peanut oil, Poloxamer 407, pepsin from porcine gastric mucosa, Pluronic F-127, Pluronic F-68, Pluronic P85, poly(dimethylsiloxane), bis(3-aminopropyl) terminated, poly(DL-lactide-co-glycolide), poly(l-lactide co-caprolactone co-glycolide), poly(propylene glycol) diglycidyl ether, poly(dimethylsiloxane)-graft polyacrylates, poly(ethylene glycol) bis(amine), poly(ethylene-co-glycidyl methacrylate), poly(ethylene-co-vinyl-acetate), poly(Lactide-co-Glycolide) acid (PLGA), poly(methyl methacrylate-co-methacrylic acid), poly(propylene glycol) diglycidyl ether, poly(tert-butyl acrylate-co-ethyl acrylate-co-methacrylic acid), poly(vinyl acetate), poly(vinyl alcohol), poly(vinyl alcohol), poly[dimethylsiloxane-co-[3-[2-(2-hydroxyethoxy)ethoxy]-propyl]methylsiloxane], poly[dimethylsiloxane-co-methyl(3-hydroxypropyl) siloxane] graft-poly(ethylenegylcol) methyl ether, polyacrylic acid, polyacrylic acid, polyaniline, polycaprolactone (PCL) , polycaprolactone, polycaprolactone triol, polyethylene glycol 3350 Da, polyethylene glycol 400 Da, polyethylene glycol methylether, polyethylene gylcol 10 kDa, polyethylene gylcol 35 kDa, polyethylene gylcol 500 kDa, polyethylene gylcol 800 Da, polyethylenimine, polyoxyethylene (20 kDa), sorbitan monooleate (Tween 80), polystrene beads (200 nm), polystrene thiol terminated, polyvinyl chloride-vinyl acetate, polyvinylpyrrolidone, polyvinylpyrrolidone K90, propyl gallate, riboflavin, riboflavin 5′-monophosphate sodium salt, SDS, sebacic acid, Sesame Oil, Sigma 7-9 (Tris base), silica gel, sodium glycholate, sodium glycochenodeoxycholate, sodium glycocholate hydrate, sodium hyaluronate, sodium taurocholate hydrate, Soluplus, soybean oil, Span 80, starch, soluble sucrose, sucrose ultra, Synperonic F108, talc, tauchloric acid, taurochenodeoxycholate, taurodeoxycholate, tetraglycol, thioflavin T, tragacanth, triacetin, tristearin, TritonX100, Tween 28-LQ-(AP), Tween20, uridine, vanillin, Vegetable Oil, vitamin B12, xantham gum from Xanthomonas Campestris, xylitol, y-decalactone, Zonyl FSO-100 fluorosurfactant, α-tocopherol, ε-caprolactam, ε-caprolactone, ω-pentadecalactone, gelatin, gelatin from bovine skin, type B, gelatin from cold water fish skin, gelatin from porcine skin, type A, glycerin, glycine, glycocholic acid, guar, heparin sodium salt from porcine intestinal mucosa, hydroxyapatide nanoparticles (200 nm), hydroxypropylmethylcellulose phalate, influenza hemagglutinin (HA) peptide, iron (III) oxide, Koliphor® EL, Kollicoat® SR 30D, Kollidon® 25, Kollidon® VA 64, Kollidon® 12PF, Kollidon® P188, Kollidon® SR, Kollidon® V67, Kollidon®P407, Kollidon®RH40, EUDRAGIT® E PO, EUDRAGIT® E100, EUDRAGIT® NM 30D, EUDRAGIT® RL PO, EUDRAGIT® S100, EUDRAGIT® L 100-55, EUDRAGIT® RS PO. For immunolabelling, the following primary antibodies were used in 1:200 dilutions: CDX2 (Rabbit polyclonal, Cell Signaling, Cat. number 3977, Lot number 1), E-Cadherin (Rabbit polyclonal, Abcam, Cat number ab15148, Lot number GR248948-1), Claudin-1 (Rabbit polyclonal, Biorbyt, Cat. number Orb10447, Lot number A0936), Rabbit Cell Signaling, Vimentin (Rabbit monoclonal D21H3, Cell Signaling, Cat. number 5741, Lot number 1), FABP1 (Rabbit monoclonal D2A3X, Cell Signaling, Cat. number 13368, Lot number 1), GLP-1 (Rabbit monoclonal 10, ThermoFisher, Cat. Number ABS 033-10-02, Lot number 08161712), Lgr5/GPR49 (Rabbit polyclonal, Novus Biologicals, Cat. number NBP1-28904, Lot number D-1), Wnt3a (Rabbit polyclonal, Abcam, Cat. Number ab28472, Lot number GR135162-3), Villin (Rabbit polyclonal, Thermo Fisher, Cat. Number A5-22072), MUC2 (Rabbit polyclonal, Thermo Fisher, Cat. Number PA5-103083, Lot number QK2116329), MDR-1 (Rabbit polyclonal, Novus Biologicals, Cat. number NB100-80870, Lot number 43176), CYP3A4 (Rabbit monoclonal D9U6N, Cell Signaling, Cat. number 13384, Lot number 1), R-spondin1 (Mouse monoclonal 422407, R&D Systems, Cat. number MAB4658), Chromogranin A (Rabbit polyclonal, Abcam, Cat number ab45179, Lot number GR3217043-1), Lysozyme (Rabbit polyclonal, Abcam, Cat number ab2408), Ost-alpha (Rabbit polyclonal, Origene, Cat. number TA338900, Lot number QC26750), OLFM4 (Rabbit polyclonal, Abcam, Cat number ab188812), OCT-1 (Rabbit monoclonal EPR16570, Abcam, Cat. number ab178869, Lot number GR243064-3), SNAT2 (Rabbit polyclonal, Origene, Cat. number TA315521, Lot number GR217181-1), Keratin 20 (Rabbit monoclonal D9Z1Z, Cell Signaling, Cat. number 13063, Lot number 1). Additionally, Wheat Germ Agglutinin, Alexa Fluor® 594 Conjugate, L12492, LysoTracker® Deep Red and R37112 ActinRed™ 555 ReadyProbes® Reagent DAPI (4′,6-Diamidino-2-Phenylindole, Dihydrochloride) were all purchased from Lifetechnologies.

RT-PCR Analysis

The expression level of various intestinal cell markers, drug transporters and metabolizing enzymes of ex vivo cultured porcine derived intestinal tissue was determined through RT-PCR analysis and detected by agarose electrophoresis. Briefly, total RNA from each tissue sample was extracted and purified with Quick-RNA plus™ (Zymo Research) followed with reverse transcription into cDNA by High-Capacity cDNA reverse transcription kit (ThermoFisher Scientific). Target genes were amplified by designed primer (Sequences listed in Supplementary Table S6) and Platinum DNA polymerase (ThermoFisher Scientific). The expression level of each target was measured with 1.5% agarose gel and detected with Bio-Rad Image suit.

Western blot Analysis

Tissue was cut in pieces of around 30 mg, snap frozen in liquid nitrogen and stored at −80°C. For tissue lysis, frozen tissue was washed with 1 volume of PBS (chilled) followed by 2 volumes of freshly prepared RIPA Lysis and Extraction Buffer (Cell Signal) with protease inhibitors (Halt™ Protease Inhibitor Cocktail, ThermoFisher). Afterward, tissue was lysed with a hand motor for 3-5 cycles of 30 seconds mixing followed by 30 seconds of cooling on ice until tissue was completely homogenized. Afterwards, the lysate was centrifuged at 2000 rcf for 30 minutes at 4°C and the resulting supernatant transferred into a new vial. The total protein concentration of the lysate was then analyzed by a BCA assay (Pierce™ BCA Protein Assay Kit, ThermoFisher) according to the manufacturer protocol. For SDS-PAGE, each set of tissue protein (~500 μg in RIPA buffer with protease inhibitor, pH 7.5) was mixed concentrated (“2X”) Laemmli sample buffer (Bio-Rad) that contained β-mercaptoethanol. The samples were heated at 95°C for 5 minutes and then run in 12% polyacrylamide gel containing 3.5 mM SDS at 120 V for 90-120 minutes in SDS-Tris-Glycine buffer, pH 8.0. Proteins were transferred onto the methanol activated PVDF membrane under 200 mA for 1-3 hours. After 5% BSA blocking, the membrane was incubated with primary antibody (1:200 as the working concentration) at 4°C overnight, followed with incubation with anti-mouse (1:3000, abcam) or anti-rabbit (1:2000, abcam) secondary antibody at room temperature for 3 hours. The target protein was detected with a Bio-Rad imager according to the manufacturer protocol. In order to compare different samples western blot analysis was performed on the same gel and the same protein concentration of the different samples were used.

Bioactivity Assays

Thioredoxin Reductase Activity Assay Kit (Cat. no. 68AT-ThioRed-S100, RayBiotech) and Protease Activity Assay Kit (Cat. no. 68AT-Protease-S100, RayBiotech) were performed according to manufacturer instructions. Lysates from tissue were obtained according to previous section. Prior to assay, the protein concentration was analyzed by a BCA assay (Pierce™ BCA Protein Assay Kit, ThermoFisher) according to the manufacture protocol. AlamarBlue® Cell Viability assay (Cat. no. DAL1100, Lifetechnologies) as well as LIVE/DEAD® Viability/Cytotoxicity Kit, for mammalian cells (Cat. nb. L3224, Lifetechnologies) were performed for intact tissue explants according to manufacturer protocol using intestinal tissue treated with 70% (v/v) ETOH (ACS reagent, ≥99.5%, Sigma) in sterile filtered deionized water for 24 hours as negative control.

Pharmacokinetic Analysis in Porcine model

All animal procedures were conducted in accordance with protocols approved by the Massachusetts Institute of Technology Committee on Animal Care. For tissue experiments, fresh tissue was obtained from local slaughterhouses within 20 minutes of euthanization. Sample size was guided by prior proof-of-concept studies in the area of gastrointestinal drug delivery and electronics^50–52. For in vivo drug delivery studies female Yorkshire pigs between 45 and 50 kg in weight were used. Before every experiment, the animals were fasted overnight. On the day of the procedure the morning feed was held. The animals were sedated with an intramuscular injection of telazol (tileramine/zolazepam) 5 mg/kg, xylazine 2 mg/kg and atropine 0.04 mg/kg. The duodenum was accessed endoscopically and the formulation directly delivered to the duodenum. Serial blood sampling from the peripheral veins was performed. For oxytocin serum quantification, an oxytocin ELISA Kit (Cat. no. ab133050, Abcam) was used according to the manufacturer’s protocol. Pharmacokinetics data was analyzed by using GraphPad Prism. The bioavailability was calculated based on the following equation:

$$Bioavailability\left(\%\right)=\frac{SampleAU{C}_{0-2h}/Dose}{IVAU{C}_{0-2h}/Dose}*100$$

Scanning Electron Microscopy (SEM) Analysis

Tissue explants were fixed in 4% (v/w) formalin in PBS for 2 days at 4°C. After that, samples were washed five times with deionized water and dehydrated through a graded ethanol (Sigma, ACS reagent 99.5%) series two times for each concentration (20, 30, 40, 50 70, 80, 90, 100, 100, 100% (v/v)) for 2 minutes in each solution. After dehydration, the samples were immersed in a hexamethyldisilazane (Sigma) solution overnight. Due to the volatile nature of hexamethyldisilazane, the solution evaporates overnight resulting in dry tissue pieces. The morphology of the fabricated surfaces was observed using a JEOL 5600LV SEM. Before visualization under SEM, all samples were sputter-coated with carbon using the Hummer 6.2 Sputter Coating System. Samples were cut to be under 0.5 cm² in area and fixed to the aluminum stubs by a double-sided adhesive carbon conductive tape.

Immunohistochemical staining

Isolated villi and crypts were fixed with 4% (v/w) formalin in PBS for 30 minutes at room temperature, washed with PBS, permeabilized with 0.25% (v/v) Triton-X-100/PBS for 2 minutes, washed with PBS and then blocked with 4% (w/v) bovine serum albumin in PBS for 1 hour. Primary and secondary antibodies were incubated in blocking buffer for 2 hours at room temperature or at 4C overnight. The stained cells were then mounted on a cover slide using ProLong® Diamond Antifade Mountant (Thermo Fisher Scientific). Tissue explants were fixed in 4% (v/w) formalin in PBS for 2 days at 4°C. Then dehydration and paraffin embedding was performed followed by tissue sectioning. For the resulting paraffin embedded tissue slides, dewaxing and antigen retrieval was conducted according to standard protocols followed by staining procedure.

Microscopy Analysis

Light microscopy analysis of histology slides was conducted using an EVOS FL Cell Imaging System with 10x or 20x air objectives. Fluorescent samples were analyzed using an Ultra-Fast Spectral Scanning Confocal Microscope (Nikon A1R) with a Galvano scanner and 20x air or 60x oil immersion objectives. Resulting raw images were analyzed with NIS-Elements C software and ImageJ. If needed the brightness and contrast of images was adjusted. This was done consistent for the entire set of images in the same experiment. No further image processing was applied.

GI-TRIS transport experiments

For the GI-TRIS system assembly, freshly isolated intestinal tissue from the jejunum was prepared according to the described tissue dissection procedure and mounted on the manufactured interface design with a generic 96-well plate as receiver plate (Corning® 96-well plates, clear bottom, Corning) or a UV transparent plates UV transparent 96-well plates (Corning® Brand 96-Well UV Plates, Corning). The jejunum was identified as the region of the small intestine approximately 50 cm after the pylorus. The difference between the jejunum and ileum was determined based on anatomical location, the structural differences of the tissue, differences in blood supply, fat deposition, and presence of lymphoid tissue. For the system in 384-well plate format, we used a 384 microplate with glass bottom as receiver plate (Greiner Sensoplate™ glass bottom multiwell plates, Sigma). For transport experiments an automated dispenser (EL406 Combination Washer Dispenser, BioTek Instruments) was used to fill receiver 96-well plates with PBS solution (transport buffer) to achieve a convex surface of the PBS liquid at the top of each well of the 96 well plate. This avoids air bubble formation that can interfere with drug transport studies. Jejunum tissue was then carefully moved over the GI-TRIS bottom plate the apical side facing upwards. Afterwards, the GI-TRIS top plate was mounted onto the receiver well plate and secured. If tissue was not used immediately for transport experiments, the tissue was cultured ex vivo for an extended period, the receiver well was filled with advanced DMEM/F-12 media and donor well left empty to maintain a liquid/air cultivation. Prior to conducting the transport experiment, the tissue was washed and the liquid in the receiver well replaced with PBS (transport buffer). Formulation samples were prepared using a liquid handling station (Evo 150 liquid handling deck, Tecan) that followed a protocol to mix the pre-prepared excipient plate (see Excipient preparation section) 10 times, pipette the appropriate amount of excipient and dispense into 96-well plate with the appropriate amount of oxytocin already in it. It would then mix the excipient/oxytocin formulations 60 times, pipette the appropriate amount, and dispense onto the GI-TRIS device. After the appropriate incubation time, the GI-TRIS device was removed from the receiver plate and a microplate reader (Infinite® M1000 PRO, Tecan) was used for spectrophotometric analysis of the receiver plate. All experiments, including sample incubation, were performed at room temperature. Optionally, a fully integrated and automated system can be used. This system integrates the liquid handling station, dispenser and microplate reader as well as a microwell plate hotel (Peak Analysis & Automation) by using a 6-axis industrial robot (Staubli). Informed by prior published transport experimental data⁵² utilizing pig derived gastrointestinal tissue in Franz diffusion cells 3-6 replicates were performed for all FDA-approved small molecule transport evaluations. For the oxytocin transport experiments an initial screen with an n=1 was performed given the high number of samples. Secondary confirmatory screens were performed in duplicates in 3 independent experiments from different animal batches. The apparent Permeability (P_app) values were calculated according to previous reports⁵³ using the following equation:

$$P_{app}=\frac{V}{A*C_0}*\frac{\Delta C_R}{\Delta t}$$

where V is the volume in the receiver chamber, A is the tissue surface area, C₀ is the initial concentration in the donor chamber, and ΔC_R is the concentration increase in the receiver chamber in the incubation time Δt. Experiments were conducted during an incubation timeframe of 1-2 hours unless otherwise noted. The initial concentration was 0.05 mg/ml except for model drug transport experiments in Fig. 3a and 3c where concentration was increased to 1 mg/ml in order to achieve detectable concentration values in the receiver chamber. For the following drugs even, higher concentration were used: Atropine (10 mg/ml), Ciprofloxacin (50 mg/ml), Famotidine (50 mg/ml), Hydrochlorothiazide (50 mg/ml), Nadolol (50 mg/ml), Cefpodoxime (50 mg/ml) and Methotrexate (10 mg/ml). The sample volume used was 0.07 ml for the 96-well plate format and 0.04 ml for the 384-well plate format.

GI-TRIS leakage experiments

The well-to-well leakage was tested using fluorescein isothiocyanate (FITC) and different interface designs including optimal well sizes of the upper plate, well sizes of the lower plate, well angles and pressure applied. FITC was added at a concentration of 1 mg/ml to isolated donor wells in the 96-well plate and PBS alone was added to the surrounding wells. The device was incubated for 6 hours at room temperature after which all wells of the receiver plate were read with a microplate reader for the presence of FITC. Percent leakage was calculated by dividing the background subtracted fluorescence signal detected in the receiver chamber by the background subtracted signal detected in the receiver chamber of the surrounding wells that contain PBS.

Data collection computational analysis

For the meta-analysis of Caco-2 permeability data reported in literature in vitro Caco-2 permeability (LogP_app) values for model drugs were obtained through a thorough literature search of published studies (see overview of process illustrated in Fig. S12). A PubMed search was performed on 10/28/2016 using the search script: “(CACO-2[Text Word]) AND (absorption[Text Word] OR permeability[Text Word] OR transport[Text Word]) AND (DRUGX[Text Word])”. DRUGX was substituted for the name of each model drug used in the study. If a model drug is known by multiple alternate names, each name was used for a Pubmed Search. Then each publication was carefully analyzed manually to obtain relevant Caco-2 Transwell permeability data. All P_app values found for the relevant compounds were converted to cm/s*10⁻⁶. Supplementary Table 2 shows an overview of all values obtained by this search. For reported P_app values clearly below or above the average, we performed a systematic analysis of the experimental parameters that could affect P_app values in these studies (Supplementary Table 3). Drug-transporter interactions reported in literature were obtained by the DrugBank database⁵⁴. We have manually checked the reported literature and removed studies that report no experimental evidence for the specific drug- interactions. We have also excluded drug-transport interactions of transporters not expressed in the small intestine. Human in vivo drug absorption data for the model drugs tested was obtained from previously published results as shown in Supplementary Table S1. The absorption kinetics classification we have used the qualitative description of slow, moderate or rapid absorption based on FDA packaging information. For comparison of delivery-relevant physicochemical properties of approved drugs and model drug library, structures for 1857 approved drugs were extracted from Drugbank 5.0 in SMILES format. Model drug structures were manually extracted from PubChem. Properties were calculated using the RDKit (http://www.rdkit.org) and were visualized in Python (2.7.6). Statistical tests were calculated in Python (ks_2samp). Intestinal absorption prediction based on Human intestinal absorption (HIA) classification was performed using the admetSAR webserver (http://lmmd.ecust.edu.cn:8000/predict) from SMILES structures extracted from PubChem. Lipinski “rule of five” violations were derived from calculated physicochemical properties in RDkit (http://www.rdkit.org, ExactMW, NumLipinskiHBD, NumLipinskiHBA, SlogP) for all model drugs.

Mechanical Modeling

Numerical analysis of the tissue-well plate interactions was performed using the finite element package COMSOL Multiphysics/Structural (COMSOL 5.2, Stockholm, Sweden). Due to the symmetry, a single well-plate system with periodic boundary conditions on the tissue was considered. The tissue behavior was captured using an isotropic near incompressible hyperelastic (neo-Hookean) model with shear modulus of μ = 3160 Pa, and κ/μ = 50, where κ is the bulk modulus of the tissue. Since the plates are much stiffer than the tissue, both upper and lower plates are considered as rigid. The tissue is modeled by a refined mesh of linear hexagonal elements. The contact between the tissue and the plates was modeled using a penalty technique.

Device Manufacturing

GI-TRIS interface device assembly: The interface apparatus consists of a standard 96-well plate (such as Corning® 96-well plates, clear bottom, Corning), a thin middle plate, and an upper load plate. Intestinal tissue is placed over through-holes in the middle plate. The upper load plate is placed onto the tissue. This compresses the tissue onto the middle plate and around the through-holes. Based on the pressure maintained by the upper plate, a seal is created. Several methods of fabrication were used in the prototyping phases of this project. Equipment used included a 3D printer (Stratasys Objet30 Pro), water jet (OMAX MicroMax), and laser cutter (Universal VLS6.60). The following methods are detailed per piece of the GI-TRIS interface device. For the upper load plate, varying diameter posts from 3 mm to 5 mm were printed on a 96-well format using a 3D printer (Objet30 Pro). Sizes ranged between 2 mm and 3 mm for the 384-well format. Supplementary weights were added during testing to compensate for the 3D printer polymer’s light weight. The final devices were manufactured using aluminum alloy. The aluminum plate was provided by Proto Labs Inc. through direct metal laser sintering (DMLS). The tissue placed on the GI-TRIS middle plate device is slightly recessed into each well by forces from the upper plate in both the 96- and 384-well plate formats. The thickness, rigidity, and diameter of through-holes of the middle plate were explored to optimize this condition, thereby reducing well-to-well leakage, described previously. Several materials were used including aluminum and acrylic because of their rigidity and machinability. The aluminum plates were water jet cut (MicroMax) while the acrylic sheets were laser cut (VLS6.60). Plate thicknesses explored were between 1 mm and 2 mm. The diameter of the middle plate is machined to be larger than the diameter of the upper load plate so the tissue can rest in between the upper and middle plate. Several diameters were explored ranging from 6.5 mm to 8 mm. For magnet-based compression plates, holes identical to standard 6,12,24,48,96,384,1536 well plate design were manufactured by using a laser cutter using acrylic sheets with 1 cm thickness (McMaster-Carr). A recess on the longer sides was milled by the laser to separate plates by hand and allow robotic arm to grab and one corner is rounded off to know well plate orientation (see detailed design in Fig S16). Either black, white or translucent acrylic was used depending on the final assay read out. Nickel plated, axially magnetized N52 grade magnets (2.28 lb force/magnet), were embedded in both plates and enabled tissue compression in between to ensure tight assembly for robotic handling and no well-to-well leakage. The magnets needed to be positioned at the outer edge as well as in the middle of the plate with 9 magnets per plate exerting a total force of 20.52 lbs. The holes of the plates were sealed by adhesive and optically clear Microseal ‘C’ Film (Biorad MSC1001). For absorption experiments, the bottom of the 2 well plate system was prefilled with media or transport buffer. Then dissected intestinal tissue was carefully placed on top without creating any air bubbles that would obstruct the transport. Then the upper plate was placed on top. The magnetic force immediately aligns the plates and maintains the position of the set up without any further requirements. Multiple plates were then stacked and were held in position by the magnetic force from adjacent plates.

Statistical Analysis

Correlation of human absorption compared to P_app of GT-TRIS or P_app of Caco-2 was performed by a two-tailed non-parametric Spearman correlation function. Significance of AUC values from oxytocin pharmacokinetic in vivo between oxytocin-PEI formulation and oxytocin control was analyzed via one-tailed t-test assuming equal variance (exact p-value = 0.007). Significance of Fig. 1i was calculated via one-tailed t-test assuming equal variance. Correlation matrix for formulation screening analysis was calculated using two-tailed Pearson correlation function. The statistical test for comparison of delivery-relevant physicochemical properties of approved drugs and model drug library was calculated in Python (ks_2samp).

Data Availability Statement

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.