Synergistic action of diclofenac with endotoxin-mediated inflammation exacerbates intestinal injury in vitro

Abstract

Intestinal homeostasis is tightly regulated by the orchestrated actions of a multitude of cell types, including enterocytes, goblet cells and immune cells. Disruption of intestinal barrier function can increase susceptibility to pathogen invasion and destabilize commensal microbial-epithelial-immune interaction, manifesting in various intestinal and systemic pathologies. However, a quantitative understanding of how these cell types communicate and collectively contribute to tissue function in health and disease is lacking.

Here, we utilized a human intestinal epithelial-dendritic cell model and multivariate analysis of secreted factors to investigate the cellular crosstalk in response to physiological and/or pathological cues (e.g., endotoxin, non-steroidal anti-inflammation drug (NSAID)).

Specifically, we demonstrated that treatment with diclofenac (DCF), an NSAID commonly used to treat inflammation associated with acute infection and other conditions, globally suppressed cytokine secretion when dosed in isolation. However, the disruption of barrier function induced by DCF allowed for luminal lipopolysaccharide (LPS) translocation and activation of resident immune cells that overrode the anti-inflammatory influence of DCF. DCF-facilitated inflammation in the presence of LPS was in part mediated by upregulation of macrophage migration inhibitory factor (MIF), an important regulator of innate immunity. However, while neutralization of MIF activity normalized inflammation, it did not lead to intestinal healing. Our data suggest that systems-wide suppression of inflammation alone is insufficient to achieve mucosal healing, especially in the presence of DCF, the target of which, the COX-prostaglandin pathway, is central to mucosal homeostasis. Indeed, DCF removal post-injury enabled partial recovery of intestinal epithelium functions, and this recovery phase was associated with upregulation of a subset of cytokines and chemokines, implicating their potential contribution to intestinal healing. The results highlight the utility of an intestinal model capturing immune function, coupled with multivariate analysis, in understanding molecular mechanisms governing response to microbial factors, supporting application in studying host-pathogen interactions.

Graphical Abstract

A human intestinal epithelial-dendritic cell model was utilized to investigate cellular crosstalk in response to physiological and/or pathological cues (e.g., lipopolysaccharide (LPS) and the non-steroidal anti-inflammation drug (NSAID) diclofenac (DCF). Disruption of barrier function induced by DCF allowed for LPS translocation and activation of resident immune cells that overrode the anti-inflammatory influence of DCF.

Intestinal homeostasis is dictated by the regulated interactions between intestinal epithelial cells, immune cells, and the microbiome. The intestinal epithelium functions as a selective barrier that moderates the exposure to microbial stimulation, thereby maintaining mucosal immune homeostasis. Perturbations to barrier integrity, due to genetic and/or environmental factors, can increase susceptibility to pathogen invasion and alter the interaction between the commensal microbiome and the mucosal immune system ¹. For instance, genetic polymorphisms in junctional proteins are risks factors for development of intestinal pathologies ². Diverse environmental factors, such as high-fat diet ³, food allergy ⁴ and drugs ⁵ have been shown to modulate gut barrier integrity. Barrier defects and systemic dissemination of microbial-derived antigens, metabolites, and inflammatory mediators, have been linked to extra-intestinal disease manifestations, including various liver pathologies, Alzheimer’s disease, and autistic spectrum disorders ^{1, 6}.

While ‘leaky gut’ syndrome has been recognized as a hallmark of many intestinal and systemic inflammatory disorders, the causal progression from barrier disruption to disease development at the mechanistic level is not well understood ^{1, 7}. Consequently, there are no approved therapeutics that target barrier repair specifically ⁷.

Chronic consumption of non-steroidal anti-inflammatory drugs (NSAIDs), which are commonly used to treat pain and inflammation associated with acute infection as well as other inflammatory conditions, are known to cause varying degrees of intestinal injury in over 70% of patients ⁸. In select cases, NSAID intake has been reported to predispose inflammatory bowel disease (IBD) onset, exacerbate disease severity or cause relapse in patients with quiescent IBD ^9–11. NSAIDs have been shown to worsen experimental colitis in animal models ¹¹. Despite the widespread use of NSAIDs and their well-documented intestinal toxicity, there is a poor mechanistic understanding of NSAID-induced enteropathology, which is multifactorial, likely involving perturbations to multiple cell types including epithelial and immune cells.

Previously, in vitro studies using either human primary or transformed intestinal cells implicated certain mechanisms of intestinal injury, which involved mitochondrial toxicity as a result of uncoupling of oxidative phosphorylation, reactive oxygen species (ROS) stress and reduced ATP production ^{5, 12}. However, the focus of these studies was on epithelial cell response to NSAIDs, as the culture models employed do not capture the reciprocal interactions between epithelial cells and immune cells. These interactions are likely significant in driving disease phenotype in vivo. It is known that germ-free mice are resistant to NSAID-induced intestinal toxicity and that antibiotic treatment reduces toxicity ¹³. The impact of the microbiome on NSAID-induced toxicity, coupled with the function of immune cells in microbiome-host cross-talk, suggests the gut immune system is significant in NSAID-induced toxicity. Further, the known impact of NSAIDs on intestinal permeability suggests specifically that immune cell response to bacterial factors able to pass a compromised intestinal barrier could play a role.

To better understand the NSAID-induced microbiome-immune crosstalk, we utilized an in vitro tri-culture of Caco2, HT29-MTX and monocyte-derived dendritic cells for the systematic examination of epithelial-immune interactions in response to various stimuli. Using a model NSAID, we showed DCF broadly suppressed cytokine release, but that in combination with luminal endotoxin, increased permeability promoted immune activation and intestinal inflammation. Multivariate analysis of secreted factors revealed distinct molecular profiles corresponding to divergent epithelial responses and identified upregulation of macrophage migration inhibitory factor ¹⁴ as a key contributor of inflammation. However, inhibition of MIF alone did not lead to full epithelial healing, indicating that a complex network of cytokines contribute both to epithelial damage and recovery. This analysis provides important insight into the cellular crosstalk that drives barrier dysfunction to disease progression and provides a framework for examining key contributors to intestinal injury and healing.

Results

Characterization of a multicellular intestinal culture

The intestine represents not only a physical barrier, but also a chemical and an immune barrier, integrated to protect the host against pathogen invasion and a variety of environmental insults. The intestinal defense system can be broken down into 3 distinct compartments: 1) the mucus layer, 2) the epithelial barrier including intercellular tight junctions, and 3) the immune component ¹⁵. We set out to capture these 3 aspects of intestinal function in a Transwell^™-based gut model. Briefly, we cultured a 9:1 mixture of enterocytes (Caco2, parental clone) and mucus-secreting goblet-like cells (HT20-MTX) on the apical surface ¹⁶, and dendritic cells, derived from in vitro differentiation of human monocytes, on the basolateral side of the Transwell membrane ^17–19 (Figure 1). We demonstrated functional maintenance of barrier integrity using transepithelial electrical resistance (TEER, Figure 2A) of epithelial/immune co-culture in a defined, serum-free media formulation, which is important for the elucidation of soluble factor-mediated intercellular communication.

Figure 1: Overview of experimental timeline and design.

The gut culture comprised a tri-culture of absorptive enterocyte (Caco2), secretory goblet-like cell (HT29-MTX) and monocyte-derived dendritic cells. Intestinal response to various perturbations (± diclofenac, ± LPS and drug withdrawal) were determined by measurement of secreted factors and various functional and phenotypic metrics. The diagrams on the left represent the different treatment conditions and the corresponding physiological and pathophysiological contexts of interests. Diagram 1 features the baseline condition for the tri-culture system, mimicking a “sterile” intestinal epithelium in the absence of any microbial stimulation. Diagram 2 features a simplified host-microbial interaction under homeostasis, where an intact intestinal epithelium can tolerate the presence of luminal endotoxin (LPS). Diagram 3 represents the scenario where DCF, in isolation, can increase intestinal permeability. Diagram 4 represents the scenario where DCF-induced barrier disruption leads to LPS translocation, resulting in immune cell activation and exacerbated epithelial damage.

Figure 2: Evaluation of intestinal phenotypes and functions in response to LPS and/or diclofenac treatment.

A) Barrier function was assessed by TEER measurement. B) LPS translocation was evaluated in gut cultures with intact barrier, with increased permeability induced by diclofenac, and following diclofenac withdrawal. C) Enterocyte injury was evaluated by FABP2 release. D) Gross cell death was evaluated by loss of DNA from gut cultures. E) Enterocyte maturation was evaluated by intracellular FABP2 level. F) Goblet cell maturation was evaluated by intracellular mucin level.

(C-F) panels show fold-change data normalized to the control conditions. Legend indicates the data from 3 independent experiments with immune cells derived from different human donors, with N=2–4 replicates per condition. The yellow shading indicates the conditions that were exposed to DCF±LPS for 4 days followed by DCF withdrawal for the subsequent 2 days. For experiment 2, DCF withdrawal treatment arm was not done due to insufficient number of cells. The mean and 95% confidence interval for each condition was plotted. Pair-wise comparisons were performed using Sidak’s multiple comparisons tests to determine significance among treatment arms. ****P<0.0001, ***P<0.001, **P<0.01, *P<0.05, †<0.06

Gut culture responded appropriately to physiological and pathological perturbations

To demonstrate the ability to recapitulate physiologically and pathologically relevant intestinal functions, we first challenged the gut cultures apically with lipopolysaccharide (LPS) to mimic the presence of luminal endotoxin.

Apically administered LPS (25 μg/mL) minimally affected paracellular permeability (TEER) and did not result in LPS translocation to the basal compartment, indicating maintenance of an intact and functional barrier in response to the physiological presence of luminal endotoxin (Figure 2A, B). Fatty acid binding protein 2 (FABP2) is an enterocyte-specific intracellular protein. Leakage of FABP2 into the media due to compromised cell membrane indicates enterocyte death. We measured no change in FABP2 release in the presence of LPS, indicating low Caco2 cell death (Figure 2C). Interestingly, although apical LPS marginally affected most of the metrics examined, intracellular mucin level was significantly enhanced (Figure 2D), suggesting that luminal LPS stimulation can activate goblet-like cells, possibly as a compensatory mechanism to prevent bacterial invasion. Together, these data demonstrated the ability of the gut culture to respond appropriately to luminal endotoxin in an adaptive rather than pathological manner.

We next challenged the gut cultures apically with diclofenac, a NSAID with known gastrointestinal toxicity. Upon addition of 930 μM DCF, barrier integrity was significantly compromised (~70% reduction in TEER, Figure 2A) and enterocyte cell death was observed (Figure 2C). DCF-induced toxicity affected both enterocyte- and goblet cell-specific maturation markers as evidenced by reduction in intracellular mucin and FABP2 level (Figure 2D–E).

Inflammation exacerbates DCF-induced epithelial cell toxicity

Although DCF treatment alone recapitulated increased intestinal permeability in our in vitro model, clinical manifestation of DCF-related intestinal injury is a complex phenomenon involving dysregulation of multiple aspects of gut function. As the epithelial monolayer functions to limit exposure to luminal toxins, loss of barrier integrity can result in endotoxin translocation and immune activation, thereby potentiating iterative tissue damage. To evaluate the contribution of endotoxin-mediated inflammation to diclofenac-induced toxicity, we treated the gut cultures apically with a combination of DCF and LPS. Barrier integrity (Figure 2A) as well as goblet-like cell maturation marker intracellular mucin (Figure 2D) and enterocyte-specific maturation marker intracellular FABP2 (Figure 2E) were not further affected in the DCF + LPS conditions. However, co-dosing of LPS and DCF resulted in a synergistic increase in media FABP2 level (Figure 2C) and a decrease in total cell number as measured by DNA content (Figure 2F), likely due to further epithelial damage arising from activation of dendritic cells. Indeed, LPS level in basal compartment was significantly increased in the presence of DCF (Figure 2B), indicating a leaky barrier.

Cytokine analysis reveals divergent immune signatures

To evaluate immune activation, media in the basal compartment was sampled and analyzed for various soluble factors using multiplex bead-based ELISA assays. Unsupervised pair-wise hierarchical clustering of the cytokine and chemokine production profile revealed distinct clusters of responses corresponding to each treatment arm (Figure 3). Overall, cytokine landscape was largely unchanged between the control (‘sterile’ gut) and LPS (apical or luminal treatment) alone conditions, as they cluster closely together. Apical LPS treatment induced a subtle increase in cytokine levels, representing controlled response to a stimulus. DCF-treatment led to global suppression of baseline secretion of cytokines/chemokines, consistent with its role shift in the soluble milieu, with marked upregulation of a large number of pro-inflammatory mediators (Figure 3). It appears that LPS translocation and activation of dendritic cells can override the anti-inflammatory influence of DCF.

Figure 3: Multivariate analysis of inflammatory response in the basal compartment of the gut model.

A) Hierarchical clustering of soluble factor secretion under various experimental conditions. The concentration for each analyte was normalized by control and z-score transformed for each independent experiment. The cytokine measurements correspond to the phenotypic and functional data presented in Figure 2, which reflect three independent experiments with immune cells derived from different human donors, with N=2–4 replicates per treatment condition. B) Principal component analysis of the cytokine response. Score plot shows the segregation of different treatment conditions, where PC1 separated conditions with or without inflammation and PC2 separated conditions with or without diclofenac treatment. C) Loading plot shows the correlation of different secreted factors with PC1 and PC2, respectively.

Unsupervised principal component analysis was performed to identify co-correlations among these secreted factors that reflect divergent immunological outcomes. The first two principal components captured over 69.5% and 17.2% of the variability in the cytokine data, respectively. Principal component 1 separated conditions with (i.e. co-treatment groups with increased levels of multiple inflammatory cytokines) and without inflammation, while principal component 2 segregated the presence and absence of diclofenac treatment (Figure 3B–C). The loading plot (Figure 3C) reveals the various cytokines and chemokines that are differentially enriched under the different contexts including baseline, perturbed and recovery phase of DCF induced toxicity. Consistent with hierarchical clustering results and phenotypic readouts, luminal LPS treatment co-localized closely with the control conditions, which indicates the maintenance of homeostatic intercellular communication in the presence of an intact barrier (Figure 3). The slight right shift of the LPS conditions on PC1 indicates a controlled response to luminal endotoxin (Figure 3B). This effect is modest in comparison to the robust inflammation induced by DCF+LPS, which dominated the cytokine responses as evidenced by the number of cytokines/chemokines with large positive PC1 values. It is important to highlight that the DCF-induced inflammatory response has a unique cytokine signature, as other means of barrier disruption (i.e. with EDTA+LPS) yielded a qualitatively different profile that is reflected by differing position on the score plot (Figure S1).

Inflammation attenuated epithelial recovery following drug withdrawal

Clinically, it is known that the NSAID-induced intestinal damage can heal naturally upon termination of treatment ⁸. In order to assess the molecular mediators associated with this process, we withdrew DCF in select conditions to examine the changes in cytokine response during healing. Removal of DCF after 4 days of treatment led to partial recovery of barrier function (Figure 2A). Interestingly, while the combined treatment of DCF+LPS did not cause a further drop in TEER compared to DCF treatment alone, the barrier recovery following drug withdrawal in the co-dosed conditions was hampered compared to the DCF only treatment (Figure 2A), suggesting that inflammation could have caused additional epithelial damage not captured by TEER measurement or that residual inflammation could delay wound healing. Furthermore, we observed improvement in terms of decreased epithelial cell death and increased DNA content (Figure 2C, D), suggesting that both reduced cellular injury and increased proliferation (though not measured explicitly) might be at play during the recovery process. Although barrier function was restored to varying degrees after DCF washout, the continued depression of intracellular FABP2 and mucin level indicated lack of differentiation/maturation and full healing (Figure 2E, F). This is not entirely surprising given the short time period investigated and the absence of certain key cellular interactions from the stromal compartment (e.g., cytokine secretion from stromal myofibroblasts).

The changes in the cytokine landscape during the recovery phase were captured by the PCA score plot, as the recovery conditions (DCF➔Ctl, DCF+LPS ➔Ctl, DCF+LPS➔LPS) moved from the negative PC2 space towards the positive PC2 space, indicating a decreased contribution from DCF and a partial reversal towards the baseline states (Figure 3B). The partial restoration of barrier function correlated with reduced LPS translocation (Figure 2B) and concomitant sharp decrease in the levels of TNFα, IL-6, sTNF-R2 and MIF (Figure 3A and C), implicating their potential involvement in driving intestinal inflammation and injury. The fast decline of these cytokines following DCF washout and barrier restoration may indicate a protective mechanism to limit cytokine-induced toxicity/tissue damage. While TNF and IL-6 are vital for rapid host defense in protective innate immunity, they are also potent drivers of intestinal pathologies ²⁰. For example, TNF and IL-6 have been shown to disrupt barrier integrity and tight junction formation in polarized epitheliums ^{21, 22}.

While some cytokines and chemokines were reduced following DCF withdrawal, many others persisted or even increased in production (Figure 3A). This is also reflected in the score and loading plot (Figure 3 B, C), where the recovery conditions DCF+LPS ➔Ctl, DCF+LPS➔LPS still had a highly positive score on the PC1 axis. Cytokines and chemokines are pleiotropic and their functions are context-specific. Sustained production of these soluble factors might be attributed to not only residual inflammation, but also a compensatory response that involves anti-inflammatory processes (e.g., IL-10) and promotion of epithelial healing (Figure 3A, C). Fractalkine, for example, has been shown to stimulate proliferation of T84 intestinal cell lines in vitro ²³.

DCF-induced MIF secretion promotes immune activation

DCF treatment alone induced a global suppression of most secreted factors with the exception of MIF, which was enhanced (Figure 3A). Co-treatment with DCF and LPS led to further enhancement of MIF production. MIF is expressed by both immune and non-immune cells across many tissues, in particular mucosal barrier epithelia that interface with the external environment ²⁴. The constitutive expression and intracellular storage of MIF enable its rapid response to an infection without de novo synthesis ²⁴. MIF can potentiate inflammation via several mechanisms. MIF interacts with other cytokines, such as TNFα or IFNγ, to amplify inflammation in autocrine and paracrine loops ^{24, 25}. Further, it can antagonize the anti-inflammatory activities of steroid hormones (i.e. glucocorticoids) to promote inflammation during primary immune response ²⁶. MIF has been shown to modulate TLR4 expression, as MIF-deficient macrophages were reported to be hyposensitive to LPS stimulation ²⁴.

The induction of MIF following DCF treatment is intriguing given the opposing pro-inflammatory effect of MIF and the anti-inflammatory property of DCF. While MIF has not been previously associated with DCF-induced enteropathology, its role in driving indomethacin (another NSAID)-related gastric injury has been reported in animal models. In fact, MIF-deficient mice were protected from developing indomethacin-induced gastric ulcers as well as dextran sodium sulfate (DSS)-induced colitis ^{27, 28}. We hypothesized that MIF may act as an amplifier of inflammation under a DCF-induced immunosuppressive environment. DCF-induced MIF production can synergize with LPS-induced TNFα or IFNγ produced by dendritic cells to potentiate intestinal inflammation, thereby overriding the immune suppressive effects of DCF. To test the role of MIF as a perpetuator of inflammation, we neutralized MIF activity in cultures treated with DCF + LPS with a small molecular inhibitor, ISO-1. ISO-1 treatment dampened global cytokine secretion (Figure 4A–B), but did not lead to reversal of epithelial damage (barrier integrity and cell death) (Figure 4C–D). Collectively, our data seem to suggest that suppression of MIF and inflammation were not sufficient to drive epithelial recovery. As noted above, in the conditions where DCF was removed after 4 days of treatment and partial restoration of monolayer functions were observed, many cytokines and chemokines were actually increased, suggesting that they might play an active role in restoring barrier function. It appears that strategies in addition to immune suppression are needed to achieve epithelial healing.

Figure 4: Effect of MIF inhibition on cytokine response and intestinal function.

A) Hierarchical clustering of soluble factor profile from Figure 3 was updated with results from the ISO-1 experiment. The concentration for each analyte was normalized by control and z-score transformed for each independent experiment. B) Principal component analysis of the cytokine response. Score plot shows the segregation of different treatment conditions, where PC1 separated conditions with or without inflammation and PC2 separated conditions with or without diclofenac treatment. C) Effect of ISO-1 on enterocyte health measured by FABP2 release. D) Effect of ISO-1 on barrier integrity measured by TEER. Replicates N=3–4 per condition. Pair-wise comparisons were performed using Sidak’s multiple comparisons tests to determine significance among treatment arms. ****P<0.0001, ***P<0.001, **P<0.01, *P<0.05

Moreover, it is recognized that the therapeutic and toxicity effect of NSAIDs are tied given that COX1/2-dependent prostaglandin signaling, the target of NSAIDs, is a central driver of inflammation as well as mucosal maintenance and healing. COX-2 dependent prostaglandin production plays a dual role in potentiating inflammation and promoting wound healing. Prostaglandin signaling plays a key role in the normal turnover of intestinal epithelial cells and epithelial healing in response to injury ^29–32. In fact, DCF has been shown to delay mucosal healing in human volunteers ³³. Therefore, it is not entirely surprising that MIF inhibition alone, in the presence of DCF, did not lead to recovery of barrier function, because suppression of COX-2 signaling can inhibit epithelial proliferation ¹⁴. Genetic deletion of COX-2 in a DSS injury model led to reduced epithelial proliferation, which could be reversed by exogenous prostaglandin E2 administration ³⁴. Therefore, attenuation of inflammation alone in the presence of DCF is not permissive to epithelial proliferation and healing.

Discussion

NSAIDs are a class of drugs often used to treat inflammation and pain in some medical conditions, including acute infection. However, the use of NSAIDs is often accompanied by adverse side effects such as increased intestinal permeability, which has been associated with multiple diseases, likely due in part to increased bacterial translocation ³⁵. Given the wide usage of NSAIDs, an improved understanding of NSAID induced intestinal toxicity can have a broad impact in the prevention of undesired side effects and development of serious inflammatory complications in high-risk individuals.

In this study, we implemented an intestinal epithelial-immune tri-culture to study the cellular and molecular players in the development of DCF-induced gut toxicity. Our results showed that in the presence of an intact barrier, the tri-culture model could tolerate luminal endotoxin as modeled by LPS. However, when the barrier was compromised by DCF, endotoxin translocation overrode the anti-inflammatory effect of DCF and triggered activation of dendritic cells, leading to epithelial cell death and barrier destruction. Our data show that neutralization of inflammation alone via MIF inhibition was insufficient to promote epithelial recovery. COX2-dependent prostaglandin secretion is critical to the normal maintenance of mucosal homeostasis, by regulating epithelial cell turnover and mucus secretion⁸, as well as mucosal healing by stimulating epithelial cell proliferation and differentiation. This is reflected by the fact that DCF treatment alone adversely impacted the enterocyte- and goblet cell-specific maturation markers (Figure 2E and F). Therefore, it is not surprising that in the continued presence of DCF, inhibition of MIF activity and blockade of inflammation did not result in epithelial recovery.

Importantly, we should note that DCF in vitro is not cleared as rapidly as in vivo due to the absence of the hepatic metabolism, such that DCF residence time in vitro is considerably higher, which may impact the observed healing response. To address this, future work may include connecting a gut and a liver model using a microfluidic device to better recapitulate DCF pharmacokinetics ^{17, 19}. A linked gut-liver system can be used to study the systemic effect of DCF-induced intestinal inflammation and DCF metabolism in driving co-morbidities in downstream organs, such as idiosyncratic hepatotoxicity. Other future considerations for improving the existing model may include the use of primary intestinal cells from gut organoids and the inclusion of other cell types (e.g., such as stromal cells) that are relevant for modeling mucosal healing, as well as commensal and pathogenic microbes. Indeed, results obtained should be interpreted in light of the fact that the presence of other immune cell types such as T cells, both during differentiation and while subjecting cells to injury/LPS exposure, could modulate the severity of barrier disruption and dendritic cell activation. Dendritic cells were chosen in this intestinal model as they are the sentinel cells at mucosal sites and are known to bridge innate and adaptive immunity via downstream interactions with T-cells; thus, their inclusion offers the potential for capturing a greater array of immune functions in vitro, although the DC-T-cell interactions are out of scope in the current study. Thus, while the current model only captured limited aspects of the full range of immune response in vivo due to the inclusion of only innate immune dendritic cells, the cytokine profiles indicate many chemokines are produced which might provide insight into potential of secondary response.

More broadly, our work highlights that intestinal pathologies associated with increased permeability generally involve 3 common phases. First, a perturbation that triggers epithelial barrier dysfunction, followed by leakage of intestinal content into the lamina propria, and the activation of resident immune cells. These 3 phases occur in a positive feedback manner that iteratively weakens the barrier. Experimental models modifying components of these interactions have shown their significance in controlling disease, including genetic knockouts impacting immune functions and junctional proteins (epithelial), and germ-free mice ^36–38. Intestinal injury is multi-faceted, making it difficult to discriminate the contribution from different cell types. Our system can facilitate deconstruction of this phenomenon in vitro by providing a means to interrogate different parts of this process in a controlled manner.

Conclusion

The utility of a simple intestinal model incorporating an innate immune component and integrated multivariate cytokine analysis for providing molecular mechanistic insight into the interplay of bacterial factors and intestinal damage in response to NSAIDS was demonstrated. A paradoxical inflammatory response results from dosing of diclofenac in the presence of LPS in this in vitro model, as a result of LPS translocation and immune activation. This inflammatory response is mediated by MIF, and inactivation of MIF is sufficient to suppress the inflammatory response but does not result in barrier recovery, likely due to the impact of diclofenac on COX-2 dependent prostaglandin production, which is important in maintenance of intestinal homeostasis. This integrated experimental approach can be applied to provide insight into molecular mechanisms underlying host-pathogen interactions.

Methods:

Epithelial-immune tri-culture preparation

The epithelial-immune tri-culture protocol was adapted from previous studies ^16–19, as described below.

Epithelial cell preparation and seeding

Caco2 (Parental clone, ATCC) and HT29-MTX (Sigma) cell lines were used. Both cell lines were passaged twice post-thawing before their use in Transwell seeding. The cell lines were maintained in DMEM (Gibco^™ 11965–092) supplemented with 10% Fetal Bovine Serum (FBS, Atlanta Biologicals S11150, heat inactivated at 57°C for 30 minutes), 1x GlutaMax (Gibco^™ 35050–061), 1x Non-Essential Amino Acids (NEAA, Gibco^™ 11140–050), and 1% P/S (Gibco^™ 15140–148).

The apical (0.5 mL) and basal side (1.5 mL) of the Transwell membrane were coated with 50μg/mL Collagen Type I (Corning 354236) overnight at 4°C. Prior to seeding, the inserts were rinsed with PBS^−/− to remove unbound protein. Caco2 at 70–80% confluence and HT29-MTX at 80–90% confluence were trypsinized (Gibco^™ 25200–056) and mechanically broken up into single cells and seeded at 9:1 ratio of Caco2:HT29-MTX onto 12-well Transwell inserts (0.4μm pore size, Costar 3460) at a density of 10⁵ cells/cm². Seeding media contained 10% heat-inactivated FBS, 1x GlutaMax and 1% P/S in Advanced DMEM (Gibco^™ 12491–015). One day after seeding, the apical media was replaced to remove any unattached cells. The apical and basal compartments were fed with 0.5 mL and 1.5 mL, respectively, of seeding medium every 2–3 days. After 1 week, the cultures were switched to a serum-free medium by replacing FBS with Insulin-TransferrinSodium Selenite (ITS) cocktail (Roche 11074547001).

Immune cell preparation and seeding

The immune component of the tri-culture consisted of monocyte-derived dendritic cells. It is noted that while tissue-resident dendritic cells would be ideal to include in the intestinal model, monocyte-derived cells were used because they are more readily available in large numbers and can be differentiated from cryopreserved cell sources. Peripheral blood mononuclear cells (PBMCs) were either purchased as frozen stocks (Zenbio) or processed from Leukopak (STEMCELL Technologies, 70500) and stored in liquid nitrogen. For each experiment, PBMCs were thawed and monocytes were isolated using the EasySep Human Monocyte Enrichment Kit (STEMCELL Technologies, 19058). Dendritic cells were obtained from differentiating monocytes in Advanced RPMI medium (Gibco^™ 12633–012) supplemented with 1x GlutaMax, 1% P/S, 50 ng/mL GM-CSF (Biolegend 572903), 35 ng/mL IL4 (Biolegend 574004) and 10 nM Retinoic acid (Sigma R2625). After 7 days of differentiation (at day 18–20 of Caco2:HT29-MTX monolayer maturation), immature dendritic cells were gently lifted using Accutase (Gibco^™ A11105–01) and seeded onto the basal side of the inverted gut Transwells for 2 hours to allow for adhesion. After immune cell seeding, the cultures were fed with an apical medium containing DMEM (Gibco^™ 11965–092) containing 1x ITS, 1xNEAA, 1xGlutaMax and 1%P/S and a basal medium containing Advanced DMEM (Gibco^™ 12491–015) supplemented with 1x GlutaMax, 1% P/S, 1x ITS, and 1.48 mL of BSA stock (Sigma A9576–50ML). One day after dendritic cell seeding, barrier integrity was assessed. Tri-cultures with transepithelial electrical resistance (TEER) values of ~200 Ohm*cm² were suitable for subsequent drug dosing experiments.

Dosing with Diclofenac and/or LPS

During dosing phase, the tri-cultures were dosed apically with either 930 μM of diclofenac (Sigma D6899–10G) and/or 25 μg/mL LPS (Sigma L4391–1MG, LPS from Escherichia coli O111:B4) dissolved in the serum-free apical medium. The basal compartments were fed with basal medium. The dosing was repeated every 2 days for a total of 6 days. At day 4 post-dosing, diclofenac was removed from some conditions to investigate the effect of drug withdrawal and recovery. Specifically, apical DCF or DCF+LPS dosing solutions were replaced with apical media or apical medium + LPS during the recovery phase. The experimental design in Figure 1 was repeated 3 independent times with immune cells derived from 3 separate human donors, N=2–4 replicates were used for each treatment condition per experiment.

MIF inhibition

To inhibit MIF activity, the dosing solution DCF+LPS was supplemented with 50 μM of ISO-1 (EMD Millipore, 82602–252), dosed every 2 days for a total of 6 days. Untreated, DCF+LPS, and DCF+LPS withdrawal conditions were repeated in the experiment for direct comparison. N=3–4 replicates were used for each treatment condition.

Dosing with EDTA and/or LPS

To disrupt barrier function with a different stimulus, 3 mM of Ethylenediaminetetraacetic acid (EDTA) with or without LPS was dosed apically to the tri-culture for 2 days. Similar molecular and functional endpoints were collected and analyzed. Untreated conditions were repeated in the experiment for direct comparison. N=4 replicates were used for each treatment condition.

TEER measurement

Barrier function was determined by measuring TEER using the Endohm-12 chamber and the EVOM2™ resistance meter (World Precision Instruments). Inserts and Endohm were kept warm on a hot plate set to 40 °C and measurement was performed while keeping the cultures warm. Final results were obtained by multiplying the raw readings by the surface area of the 12-well Transwell (1.12 cm²). TEER values of ~ 200 Ohm*cm² were considered acceptable.

Cell lysis

Briefly, cell monolayers were rinsed with ice-cold PBS twice to remove dead cells. Transwell inserts were stored in −80 °C for later processing. The cell lysis buffer contained 50 mM Tris, 10% glycerol, 150 mM NaCl, 1% NP40, pH 7.5. Before lysis, the lysis buffer was supplemented with 1:100 protease inhibitor (Sigma P8340). Transwell plates were thawed on ice for 5 mins after removal from −80 °C. 0.5 mL of lysis buffer was added manually, the cell monolayer was scraped off with a pipette tip, and the lysate was transferred to an Eppendorf tube. The cell lysate was vortexed vigorously for 30 seconds. Eppendorf tubes containing cell lysate were put on a rocker at 4 °C for 1 hour to allow further lysis. After 1 hour, each sample was vortexed again for ~30 seconds and then centrifuged at 10,000 xg at 4 °C for 10 minutes. The supernatant was then transferred to a clean pre-chilled Eppendorf tube. The lysate was used for intracellular FABP, mucin and DNA determination.

FABP quantification

FABP levels (apical, basal and intracellular) were measured in a 96-well high-binding plate (Costar #2592) using the FABP2/I-FABP ELISA DuoSet kit (R&D Systems #DY3078, plus WA126, DY994, DY995, DY999). Samples were diluted in reagent diluent (#DY995), if needed. Plate was read using a SpectraMax M3/M2e. Extracellular FABP release was calculated by multiplying the FABP concentration in the apical and basal compartments by their respective volumes, 0.5 mL and 1.5 mL. Then, the extracellular FABP level was normalized to the total FABP level (extracellular + intracellular) to obtain the %FABP released. Finally, %FABP release for all the treatment conditions were normalized to the corresponding untreated controls for each independent experiment. Statistical analysis was performed on the entire dataset. To assess enterocyte maturation, intracellular FABP level was normalized to DNA content for each sample.

DNA quantification

DNA was quantified using Quant-IT PicoGreen dsDNA Kit (Life Technologies P7589). Plate was read using a SpectraMax M3/M2e.

LPS quantification

LPS was quantified using Pierce^™ LAL Chromogenic Endotoxin Quantitation Kit (Thermo Scientific 88282). Apically treated samples were diluted 1:500,000 in steps no greater than 1:100 and basal samples were diluted 1:100–500. Dilutions were made using endotoxin-free water (supplied in kit) and Protein LoBind Tubes (Eppendorf 022431081, 022431064), vortexing for 1 minute after each dilution. Plate was read using a SpectraMax M3/M2e.

Mucin quantification

The mucin quantification protocol was modified from ³⁹. Intracellular mucin level in the lysate was quantified against a standard of mucin (Sigma M3895) dissolved in lysis buffer. Samples and standards were incubated in a 96-well plate in a 3:1 ratio of sample to Alcian Blue solution (Richard Allen Scientific) for two hours. Then, plates were centrifuged at 1640g for 30 minutes at room temperature. Supernatant was removed by inverting and blotting the plates onto dry paper towels. Plates were rinsed twice with a wash buffer (40% (v/v) of ethanol and 60% (v/v) of 0.1M sodium acetate buffer containing 25mM MgCl₂ at pH 5.8), with a 10-minute centrifugation step after each rinse. After second centrifugation, supernatant was removed again and samples were dissolved in 10% SDS in distilled water. Samples were re-suspended thoroughly by pipetting and agitation on a shaker. If bubbles formed during re-suspension, bubbles were popped using a needle and plates were centrifuged at 1000g for 5 minutes prior to absorbance measurement on a Spectramax M3/M2e at 620nm.

Multiplex cytokine/chemokine quantification

Cytokine and chemokine levels were measured using multiplex cytokine assays, including the Pro Human Inflammation Panel 24-plex and Human Chemokine Panel 40-plex (Bio-Rad Laboratories, Inc., Hercules, CA, USA). Briefly, media samples from the basal side of the Transwell model were collected in low-binding tubes and spun down at 10,000 xg for 5 minutes to remove cell debris. The supernatant was aliquoted and stored at −80 °C. Samples were measured at multiple dilutions to ensure they were within the linear dynamic range of the assay. BSA was supplemented to the samples to achieve a final concentration of 5 mg/mL to minimize non-specific binding. Protein standards were prepared in the same media and serially diluted to generate an 8-point standard curve. Assays were prepared according to manufacturer’s protocol and run on a Bio-Plex 3D Suspension Array System (Bio-Rad Laboratories, Inc.). Data were collected using the xPONENT for FLEXMAP 3D software, version 4.2 (Luminex Corporation, Austin, TX, USA). The concentration of each analyte was determined from a standard curve, which was generated by fitting a 5-parameter logistic regression of mean fluorescence on known concentrations of each analyte (Bio-Plex Manager software). Analytes that were below the limits of quantification and those that are outside of the dynamic range were excluded from further analysis.

Statistical analysis

Statistical analysis was performed using GraphPad Prism version 7.03 Windows (GraphPad, La Jolla, CA).

Multivariate analysis

Hierarchical clustering and principal component analysis was perform using Matlab R2017a software (Mathworks, Natick MA). The cytokine and chemokine data were normalized to the respective untreated controls, mean-centered and variance scaled individually for each experiment and then combined prior to performing clustering and principal component analysis.