Immunocompetent High-Throughput Gut-on-Chip Model for Intestinal Microbes—Host Interaction Studies

Abstract

The intestinal microbiota plays a crucial role in maintaining epithelial barrier integrity, while its impairment and the resulting inflammation contribute to numerous human pathologies. To preserve intestinal homeostasis, various probiotics are being developed; however, their selection and validation require accessible yet physiologically relevant models. We recently established a high-throughput Gut-on-Chip model comprising human epithelial (Caco-2) cells and peripheral blood mononuclear cells (PBMCs), demonstrating epithelial barrier disruption and pro-inflammatory cytokine secretion upon inflammation induction. The present study aimed to evaluate the feasibility of co-culturing anaerobic members of the human intestinal microbiota within this model and to assess their effects on inflammation-induced epithelial damage. We successfully co-cultured five intestinal anaerobic bacterial species in direct contact with the epithelial monolayer for two days. As proof of concept, we demonstrate that live Bacteroides thetaiotaomicron and its supernatant preserve epithelial barrier integrity and attenuate CCL2 secretion by Caco-2 cells. In contrast, Clostridium scindens did not prevent epithelial damage but suppressed CCL20 secretion, revealing a promising target for future studies. By recapitulating some of the key aspects of intestinal inflammation, we suggest that the current Gut-on-Chip model has potential as an easy-to-use platform for screening next-generation probiotics and live biotherapeutics with homeostatic and immunomodulatory properties.

1. Introduction

The intestinal mucosa surface separates the host from the myriads of microorganisms found in the intestinal lumen. This barrier consists primarily of epithelial cells tightly connected thanks to different junction proteins, thus restricting the infiltration of external antigens between the cells [1,2]. Disruption of this barrier, referred to as the “leaky gut”, and the resulting inflammation contribute to the onset or progression of numerous human pathologies. These include gastrointestinal disorders such as inflammatory bowel disease, irritable bowel syndrome, and celiac disease, as well as extraintestinal conditions like rheumatoid arthritis, non-alcoholic fatty liver disease, diabetes, and Parkinson’s disease [3,4]. Intensive research in recent decades unraveled the key role of the intestinal microbiota in the maintenance of the epithelial barrier homeostasis, while changes in the microbiome (also called “dysbiosis”) are often associated with a variety of autoimmune, metabolic, and neurodegenerative diseases [5,6,7]. On the other hand, some microbial structural components and metabolites were demonstrated to strengthen the intestinal epithelial barrier through modification of the expression and spatial organization of junction proteins of intestinal epithelial cells [8,9] either directly or via the modulation of cytokines secretion (e.g., IFNγ, TNFα, IL-1β, IL-6 and others) by epithelial and immune cells [10,11,12,13]. Therefore, “classical” probiotics and postbiotics, as well as next-generation probiotics such as live biotherapeutic products (LBPs), are extensively developed to aid in maintaining the intestinal epithelial barrier and dampening excessive and chronic inflammation [14,15]. However, while numerous product candidates pass through pre-clinical development that includes both in vitro and in vivo models, only ~15% of these candidates in the field of inflammatory diseases succeed in clinical trials and receive approval [16]. Specifically, for IBDs, the use of adequate pre-clinical models that take into account microbiota–host interactions was identified as one of the major targets to improve the success rate of novel treatment development [17]. In this light, microfluidics-based models offer several advantages over “standard” (one cell type-based) in vitro assays as well as over animal models including the following:

• The use of flow, which mimics the physiological conditions in the intestine and also exerts shear stress on epithelial cells, promoting their differentiation [18,19].
• The possibility of combining several cell types in one assay (e.g., epithelial and immune cells).
• The 3D spatial organization of the tissues.
• The use of human-derived host cells.

Here, we co-cultured Caco-2, human PBMCs, and several intestinal bacteria in a previously described high-throughput Gut-on-Chip model [20,21,22], which allows rapid maturation of the epithelial monolayer and easy monitoring of its barrier integrity as well as cytokine sampling and microscopic analysis. In this model E. coli LPS-stimulated PBMCs secrete a variety of pro-inflammatory cytokines that compromise the integrity of the epithelial barrier formed by Caco-2 cells [20]. The capacity of two bacterial strains to reduce the inflammation and protect the epithelial barrier was then tested. We suggest that the current Gut-on-Chip model recapitulates some of the key features of intestinal inflammation and enables easy-to-use screening of novel (bio)therapeutics in a high-throughput manner.

2. Materials and Methods

2.1. PBMC

Human peripheral blood samples were collected at the French Blood Establishment (Établissement Français du Sang (EFS), Rungis, France) which is responsible for the collection and distribution of blood in France. All donors provided informed consent to EFS. This sample collection does not require any ethical committee submission, in accordance with current regulations for products supplied by the EFS.

PBMCs from six human blood samples collected in lithium heparin tubes (Becton Dickinson and Co., Franklin Lakes, NJ, USA) were isolated using standard procedure within 12 h of collection as previously discussed in [20]. PBMCs were kept in liquid nitrogen until use and then thawed on the day of experiment, centrifuged at 450× g for 10 min, and set to 10⁶ cells/mL in RPMI 1640 medium supplemented with 10% heat inactivated fetal bovine serum (HI FBS) and 1% penicillin/streptomycin (Sigma-Aldrich, St. Louis, MO, USA).

2.2. Intestinal Epithelial Cells

Caco-2 cells (TC-7 clone, Sigma-Aldrich, St. Louis, MO, USA) were cultured in complete Minimum Essential Medium (MEM, Gibco, Thermo Fisher Scientific, Inc., Waltham, MA, USA) supplemented with 10% HI FBS (Gibco, Waltham, MA, USA), 1% sodium pyruvate (Gibco, Waltham, MA, USA), 1% non-essential amino acids (NEAA, Gibco, Waltham, MA, USA), and 1% penicillin/streptomycin (Sigma-Aldrich, St. Louis, MO, USA). Cultures were maintained at 37 °C in a humidified atmosphere containing 5% CO₂ and used between passages 2 and 12.

2.3. OrganoPlate Seeding

Caco-2 culture in the 3-lane OrganoPlate^® (Mimetas B.V., Oegstgeest, The Netherlands) was performed as previously reported in [21]. In short, 2 × 10⁴ Caco-2 cells were seeded on top of a 4 mg/mL Collagen I gel (Collagen-I rat tail, R&D systems, Minneapolis, MN, USA) and allowed to attach for 4 h (37 °C, 5% CO₂). Caco-2 medium was then added to the inlets and outlets, and bidirectional flow was created by placing the plate on an interval rocker (Perfusion Rocker, Mimetas B.V., Oegstgeest, The Netherlands) switching between a +7° and −7° inclination every 8 min (37 °C, 5% CO₂). The medium was changed 4 days after the seeding. When indicated, 5 days after Caco-2 seeding, the medium in the opposite channel was exchanged with 10⁵ PBMCs in supplemented RPMI (without antibiotics), with (or without) LPS mix of three previously communicated strains of E. coli (Sigma-Aldrich, St. Louis, MO, USA, cat. nos. L2637, 3012 and 3137, [23]) at 100 ng/mL each.

2.4. Bacteria Co-Culture in OrganoPlate

Bacterial strains (

Table 1. Bacterial strains used in the study.

Species	Strain
Bacteroides thetaiotaomicron	DSM 72079
Akkermansia muciniphila	DSM 22959
Christensenella minuta	DSM 22607
Roseburia intestinalis	DSM 14610
Clostridium scindens	DSM 5676
Adlercreutzia equolifaciens	DSM 19450

) were cultivated first on previously described mGAM-CRIM agar plates [24] for 48 h in the BACTRON600 (Sheldon Manufacturing, Inc., Cornelius, OR, USA) anaerobic chamber at 37 °C. Then a single colony of each strain was inoculated in 5 mL mGAM-CRIM liquid broth and incubated in the chamber for 24 additional hours. The next day, bacterial cultures were centrifuged at 10,000× g for 10 min and the supernatants were filtered through a 0.22 µm filter (Millex-GS, Millipore, Darmstadt, Germany). To eliminate the residuals of the supernatant, bacteria pellets were washed twice in PBS, by repeating resuspension and centrifugation steps, and then set to an optical density (600 nm) of 1 in a MEM medium supplemented with 1% sodium pyruvate (Gibco, Waltham, MA, USA), 1% NEAA (Gibco, Waltham, MA, USA), and 1X Insulin-Transferrin-Selenium-Ethanolamine solution (ITS; Gibco, Waltham, MA, USA), called here “MEM-co-culture medium”. Of note, this medium is free of antibiotics and serum to avoid interference with bacterial growth, while the addition of ITS maintains the mammalian cell viability in the absence of serum. On day 5, upon Caco-2 seeding, and simultaneously with PBMCs (+/− E. coli LPS) administration in the bottom channel, the medium in the top channel was replaced by bacteria suspension in MEM-co-culture medium or with sterile medium supplemented with filtered bacteria supernatant at 30% final concentration. OrganoPlates were then incubated in hypoxic conditions: 4% O₂ and 5% CO₂ at 37 °C. At indicated time points, 10 µL samples of medium from the top channels were taken, serially diluted in PBS, plated on mGAM-CRIM plates, and incubated for 48 h in the anaerobic chamber for colony forming units (CFU) determination.

2.5. Cytokine Secretion

For cytokine analysis the medium from each channel was collected and stored at −20 °C for analysis using the Luminex method. The concentrations of CCL2, CCL20, CCL28, IFNγ, IL-2, CXCL8 (IL-8), TNFα, CCL3, CCL25, CXCL10, IL-1β, IL-6, and IL-10 were quantified using the Bio-Plex^® 200 system (Bio-Rad, Hercules, CA, USA) with a multiplex assay (R&D Systems, Minneapolis, MN, USA) following the manufacturer’s instructions. Only the results of quantitatively detected cytokines are presented. IL-8 concentration was assessed using the ELISA method according to the manufacturer protocol (R&D Systems).

2.6. TEER Measurements

Transepithelial electrical resistance (TEER) was measured as previously described by Beaurivage et al. [21] and presented in resistance units (Ω·cm²).

2.7. Immunohistochemistry and Microscopy

At the end point of experiments, all cell cultures were fixed in 4% paraformaldehyde (VWR 100504-858) in DPBS +Ca/+Mg (14040-083; Gibco, Waltham, MA, USA) for 30 min. Cell membranes were permeabilized with a 0.1% Triton-X-100 solution in DPBS for 20 min and blocked with a 2% bovine serum albumin (BSA, A2153; Sigma-Aldrich St. Louis, MO, USA) in DPBS for over 1 h. Primary antibodies against ZO-1 (Invitrogen 61-7300, AB_138452, Carlsbad, CA, USA) were then diluted in a 2% BSA solution in DPBS and incubated overnight at 4 °C. Secondary antibodies (Invitrogen A-21429, AB_2535850), DAPI (Invitrogen, D1306), and Phalloidin (Sigma-Aldrich, St. Louis, MO, USA; P5282) were then also diluted in 2% BSA and incubated for 2 h. Cell cultures were rinsed with DPBS prior to imaging. All steps, except when specified otherwise, were performed on chips and at room temperature. Confocal immunostaining images were acquired using a spinning disk microscope (Andor BC34 CF) using a 20X magnification air objective (NA 0.8). Z-stacks were acquired over a range of 200 µm with a Z resolution of 0.3 µm.

2.8. Statistical Analysis

TEER measurements: Repeated measures (RM) two-way or one-way Anova analysis was performed comparing each condition to a control condition decided for each type of experiment. Only statistically significant different results were considered when the p-value was lower than 0.05 and presented with asterisks.

IL-8 ELISA: Ratio paired t test was performed comparing each condition to a control condition. Only statistically significant different results were considered when the p-value was lower than 0.05 and presented with asterisks.

Multiplex cytokine analysis: The analysis was performed as previously described [20]. Briefly, each assay was analyzed individually, comparing the expression in one condition with a control using a non-parametric Wilcoxon test while the p-values were adjusted using the Benjamini–Hochberg approach [25]. When indicated, the log₂-fold change was calculated as the log₂ ratio of mean expression levels between the two conditions that was considered significantly different if an adjusted p-value < 0.05.

3. Results

A tubule-shaped monolayer of Caco-2 cells within one of the channels (“Lumen” channel) was prepared within each chip of the microfluidic device called OrganoPlate^®, as previously demonstrated by Beaurivage et al. [21]. Once the epithelial barrier (assessed by TEER) was established, six bacterial species (Table 1) commonly inhabiting the human intestine were introduced within the Lumen channel of the chips to survive in co-culture with the epithelial cells (Figure 1A). Since these bacteria are anaerobic, experiments were conducted at 4% oxygen, which better reflects the physiological values ranging from ∼6% in the vascularized submucosa to 1–2% near the crypt–lumen interface [26]. Except for Roseburia intestinalis, all five other species maintained a concentration of viable units above 10⁶ CFU/chip (Figure 1B) for 48 h. At the same time, no decrease in epithelial barrier efficiency (Figure 1D) and no significant increase in secretion of the main pro-inflammatory cytokine IL-8 [27,28] were detected (Figure 1C), suggesting that epithelial cells suffered no damage or inflammatory stress.

Encouraged by these observations, we tested the capacity of some of these bacterial species to mitigate inflammation-induced epithelial damage. We decided to proceed with two species that were proposed for live biotherapeutic development. The first is B. thetaiotaomicron, whose anti-inflammatory and epithelial barrier-supporting features were demonstrated both in vitro and in vivo [29,30,31,32]. The second is C. scindens, which was not reported to possess such abilities but is capable of secondary bile acid production, preventing the emergence of important human pathogen Clostridioides (Clostridium) difficile [33,34] and playing a significant role in steroid metabolism in humans [35]. To mimic epithelial damage mediated by the pro-inflammatory activity of immune cells, we introduced human PBMCs into the opposite (“Basal”) channel of each chip and stimulated them with a lipopolysaccharide (LPS) from E. coli, as we reported previously [20]. Simultaneously with the addition of PBMCs to the Basal channel, washed B. thetaiotaomicron, C. scindens, or their filtered spent culture medium from overnight cultures at 30% vol/vol were co-incubated in the Lumen channels (Figure 2A). Similarly to the experiment without PBMC, B. thetaiotaomicron viability remained roughly unchanged during the two days of the co-culture experiment, while the CFU of C. scindens rapidly declined (Figure 2B). Intriguingly, while LPS stimulation of PBMCs provoked a 41% decrease in TEER of Caco-2 (from 144.0 ± 25.89 to 84.83 ± 4.22 Ω·cm²) at 24 h post stimulation, co-culture with B. thetaiotaomicron or co-incubation with its supernatant completely protected the Caco-2 cells from the LPS-induced damage (Figure 2C). On the other hand, the spent medium of C. scindens, but not the bacteria themselves, demonstrated moderate mitigation of the epithelial barrier damage (123.0 ± 18.33 Ω·cm²) compared to the control medium (84.83 ± 4.22 Ω·cm²). The overall pattern of effects on TEER remained similar at both time points although the effect of bacteria supernatants fell below the statistical significance threshold at 48 h due to increased variability of the results.

To test whether the protective effect is associated with immunomodulation, we analyzed the cytokine profile in both channels at the end of the 48 h of the experiment. As expected, LPS stimulation of PBMCs induced a sharp increase in concentration of most tested cytokines, which was more profound in the Basal channel where the PBMCs resided (Figure 3A,B and Figure S1). Interestingly, bacteria and their supernatants had very limited and mostly statistically insignificant effect on the cytokine profiles in the Basal channel (Figure 3C and Figure S1). On the other hand, the concentrations of CCL2 and CCL20 chemokines in the Lumen channel were strongly altered (Figure 3D and Figure S1). Specifically, B. thetaiotaomicron and its spent medium decreased the CCL2 concentration by 4.53- (p = 7.63 × 10⁻⁶) and 5.76- (p = 7.63 × 10⁻⁶) fold, respectively, while only the bacteria, but not the supernatant, reduced the concentration of CCL20 by 5.85- (p = 1.53 × 10⁻⁵) fold (Table S1). Co-culture of C. scindens with the Caco-2 cells provoked an even more profound decrease—14.10- (p = 7.63 × 10⁻⁶) fold—in the concentration of CCL20, however this tendency was reverted when only the spend medium of C. scindens was used, provoking a 2.3- (p = 7.63 × 10⁻⁶) fold increase in this cytokine. Accordingly, the concentration of two pro-inflammatory cytokines IL-6 and IL-8 showed very moderate—1.10- and 1.56-fold, respectively—but still statistically significant (p = 6.57 × 10⁻³ and p = 8.03 × 10⁻³) increase when only the supernatant of C. scindens was added to the Lumen channel.

Since the efficiency of the epithelial barrier is maintained mainly by tight junction proteins [1,2], we investigated the spatial organization of the key component of this complex—Zonula Occludin-1 (ZO-1)—using immunofluorescence microscopy (Andor Technology Ltd, Belfast, Northern Ireland). Indeed, LPS stimulation of PBMCs altered the homogeneous distribution of ZO-1 along Caco-2 cell edges and also impacted the general morphology of the epithelial monolayer as evidenced by the filamentous actin (F-Actin) staining (Figure 4). In accordance with TEER measurements, co-culturing B. thetaiotaomicron in the Lumen channel mitigated the alteration of ZO-1 and F-actin spatial distribution. The spent medium of B. thetaiotaomicron, on the other hand, provoked a dramatic effect on the organization of both markers: ZO-1 often appeared in patches along the cell edges, while F-actin seemed to be concentrated almost exclusively at the cell edges (cortex) in contrast to a more homogeneous distribution seen in other experimental conditions. The morphology of epithelial cells upon incubation with C. scindens or its supernatant looked similar to the condition with LPS-stimulated PBMCs without any additional treatment.

4. Discussion

To co-culture anaerobic bacteria with host cells in vitro, researchers usually use sophisticated models involving pumps and/or anaerobic chambers that allow control of oxygen tension in different compartments of the microfluidic device [36,37,38,39,40,41]. The current model demonstrates the feasibility of prolonged co-culturing of some anaerobic bacteria in an easy and high-throughput model simply by reducing the oxygen concentration to the physiologically relevant value of 4%. It is also possible that the rapid oxygen consumption by epithelial cells creates a niche in their close vicinity with an even lower oxygen concentration, allowing anaerobic bacteria to survive.

LPS stimulation of PBMCs in the current Gut-on-Chip model upregulated several cytokines that play a crucial role in modulating the intestinal epithelial barrier (Figure 3). These include chemokines that control the recruitment of immune cells to the inflamed tissue, such as CCL-20, CCL2 (MCP-1), and CCL3 (MIP-1α), as well as cytokines regulating the activation status of these cells: IL-1β, IL-8, IL-6, TNFα, and IL-10 [11,42]. Administration of live bacteria or of their filtered supernatants in the Lumen channel mostly did not modulate the increase in cytokine concentration in the opposite Basal channel, whose main source is PBMCs (Figure 3C). In the Lumen channel, however, the increase in CCL2 concentration was reverted by B. thetaiotaomicron and its supernatant, and the increase in CCL20 was reverted by both bacterial strains (Figure 3D). As no modulation of these two chemokines was observed in the Basal channel, the impact of the bacteria is most probably mediated via modulation of cytokine secretion from the Caco-2 cells. The decrease in CCL2, but not in CCL20, correlates with the improvement in the epithelial barrier function assessed by TEER (Figure 2B), suggesting that CCL2 plays a key role in epithelial barrier damage in this model. In agreement with this observation, elevated CCL2 concentrations were detected in intestinal specimens of patients suffering from inflammatory bowel diseases (IBD) in several studies [43,44,45]. While C. scindens did not significantly improve epithelial barrier integrity, its pronounced impact on CCL20 secretion, a central chemokine in intestinal inflammation, underscores its relevance for future studies as a biotherapeutic in the autoimmune disease area [46,47]. However, the opposite effect of C. scindens supernatant on CCL20, as well as a mild increase in IL-8 concentration in the Lumen channel, suggest ambivalent interaction of this species with intestinal epithelium. The rapid decline of cultivable bacteria number (~2 logs CFU/day) might be one of the reasons for this ambiguous phenotype of C. scindens under the current model conditions.

On the cell morphology level, B. thetaiotaomicron co-culture with epithelial cells seems to prevent the inflammation-induced decrease in ZO-1 localization at the cell edges (Figure 4), which aligns well with a previous communication about the positive effect of this species on ZO-1 mRNA expression [31]. The fact that the bacteria supernatant alone also promoted the protection of the epithelial barrier function suggests that the effect of B. thetaiotaomicron is mediated by a secreted factor. Indeed, tryptophan metabolites, including indole derivatives, produced by this bacterium were demonstrated to enhance epithelial barrier function [30]. However, the effect of the supernatant on F-actin and ZO-1 organization looked strikingly different from the one provoked by the B. thetaiotaomicron co-culture, although both treatments protected the epithelial barrier from inflammation-induced deterioration. While the patchy ZO-1 staining is not usually associated with a strong epithelial barrier, we observed a noticeable enhancement of cortical actin cytoskeleton, which is also critical for the integrity of the gut barrier [48] and might explain the positive effect of B. thetaiotaomicron supernatant after all. In general, the different results obtained with live bacteria and its supernatant suggest that B. thetaiotaomicron has more than one way of impacting epithelial homeostasis. In fact, at least one cell-associated factor of B. thetaiotaomicron—pirin-like protein was demonstrated to reduce pro-inflammatory NF-κB signaling in Caco-2 cells [29]. More investigations are needed to elucidate the complete picture of the effect of this species.

5. Conclusions

Previously, we have set up the Gut-on-Chip inflammatory model using epithelial cell line and human PBMCs and demonstrated the homeostatic effect of probiotic supernatants [20]. Besides enterocytes and circulating immune cells, the intestinal tissue comprises a variety of cell types, including secretory epithelial cells, intraepithelial and tissue-resident immune cells. Therefore, the current Gut-on-Chip model cannot fully replicate the entire complexity of the inflammatory process in vivo. Nevertheless, it recapitulates selected key features of intestinal inflammation: epithelial barrier damage and pro-inflammatory cytokine secretion. Furthermore, in the present study we showed that some anaerobic bacteria from the human intestine can be co-cultured in this model. As a proof of concept, we screened six anaerobic species and demonstrated the protective effect of B. thetaiotaomicron on the epithelial barrier against inflammation-mediated damage, associated with CCL2 secretion prevention and modulation of ZO-1 and F-actin spatial organization, while also flagging C. scindens as a promising candidate for future studies. In addition, established in the commercially available OrganoPlate platform, the model is easily standardizable and scalable to higher-throughput applications.