Human Intestinal Morphogenesis Controlled by Transepithelial Morphogen Gradient and Flow-Dependent Physical Cues in a Microengineered Gut-on-a-Chip.

We leveraged a human gut-on-a-chip (Gut Chip) microdevice that enables independent control of fluid flow and mechanical deformations to explore how physical cues and morphogen gradients influence intestinal morphogenesis. Both human intestinal Caco-2 and intestinal organoid-derived primary epithelial cells formed three-dimensional (3D) villi-like microarchitecture when exposed to apical and basal fluid flow; however, 3D morphogenesis did not occur and preformed villi-like structure involuted when basal flow was ceased. When cells were cultured in static Transwells, similar morphogenesis could be induced by removing or diluting the basal medium. Computational simulations and experimental studies revealed that the establishment of a transepithelial gradient of the Wnt antagonist Dickkopf-1 and flow-induced regulation of the Frizzled-9 receptor mediate the histogenesis. Computational simulations also predicted spatial growth patterns of 3D epithelial morphology observed experimentally in the Gut Chip. A microengineered Gut Chip may be useful for studies analyzing stem cell biology and tissue development.

Introduction

Understanding how local gradients of morphogens and their antagonists regulate tissue development remains a fundamental question in developmental biology (Logan and Nusse, 2004, Petersen and Reddien, 2009, Zallen, 2007), and elucidation of these mechanisms could lead to new approaches to stem cell engineering and regenerative medicine (Clevers et al., 2014). Intestinal development is a classic example whereby intestinal villus morphogenesis is known to be controlled by polarized gradients of the morphogens, but the precise physical mechanism remains unknown. For example, Wnt ligands produced by Paneth and mesenchymal cells (Farin et al., 2012) stimulate intestinal epithelial proliferation, whereas bone morphogenetic protein represses Wnt signaling and promotes cytodifferentiation as well as apoptosis as they move vertically along the crypt-villus axis (Biswas et al., 2015, Martini et al., 2017, Sato and Clevers, 2013). However, Wnt antagonists, such as Dickkopf-1 (DKK-1), are known to inhibit Wnt/β-catenin signaling (Bafico et al., 2001), and it remains unclear how all of these factors interact with each other to maintain intestinal homeostasis in vivo.

It has not been possible to dissect the mechanism by which the morphological three-dimensional (3D) formation of epithelium occurs in the human intestine under controlled conditions because it is difficult to recreate the localized gradient of morphogens and their antagonists in conventional static cell culture models. Intestinal organoids derived from intestinal crypts or single intestinal stem cells have been used to study crypt regeneration and crypt-epithelial domain formation in vitro (Farin et al., 2016, Sato et al., 2009, Sato et al., 2011). However, the localized morphogen gradients that drive crypt formation are randomly organized in organoid cultures, and thus, it is impossible to dissect the molecular and biophysical mechanisms that orchestrate the regulated morphogenesis. Therefore, there is a critical need for physiological tissue models that can control spatiotemporal gradients of morphogens and their antagonists with a defined developmental axis in a human organ-relevant context.

Human Organ-on-a-Chip (Organ Chip) technology, which involves the development of microfluidic cell culture devices that recreate the physical and biochemical microenvironment of key functional units of living human organs, offers an alternative approach to study intestinal structure and function. We previously described a Gut Chip device lined by an intact monolayer of human Caco-2 intestinal epithelium, which spontaneously forms intestinal villi-like 3D structures when cultured under continuous flow and cyclic peristalsis-like mechanical deformations (Kim et al., 2012, Kim and Ingber, 2013). These microengineered villi-like epithelial cells recreate all four differentiated cell types of the small intestine (absorptive, goblet, enteroendocrine, and Paneth) and contain proliferative cells limited to their basal crypts. This 3D epithelium also exhibits physiological migration of proliferative cells from the crypt to the villus tip, formation of a specialized apical brush border, augmented barrier function, increased drug-metabolizing cytochrome P450 activity, and enhanced mucus production relative to static cultures (Kim et al., 2012, Kim and Ingber, 2013). In addition, the microfluidic Gut Chip model has been used to co-culture anaerobic commensal or pathogenic gut microbiome with living human intestinal epithelium for extended periods and to recapitulate the pathophysiology of intestinal inflammation and small intestinal bacterial overgrowth in vitro (Kim et al., 2016, Shin et al., 2019). The genome-wide transcriptome analysis confirmed that Caco-2 cells also exhibit a highly differentiated intestinal epithelial phenotype similar to that shown by the normal human ileum when cultured in the Gut Chip (Kim et al., 2016), even though the Caco-2 cells were originally isolated from human colorectal cancer that show truncating mutations in adenomatous polyposis coli (APC) tumor suppressor and β-catenin proteins (De Bosscher and Nicolas, 2004, Ilyas et al., 1997). By leveraging the Gut Chip, we also identified that the epithelial barrier dysfunction is the culprit trigger that initiates the onset of intestinal inflammation under complex host-microbiome cross talk (Shin and Kim, 2018). Formation of villi-like structures by Caco-2 cells also was previously observed by another group (Pusch et al., 2011), although their structure and function were not fully characterized. Thus, the mechanism of this epithelial morphogenesis remains unknown.

The Gut Chip is a two-channel microfluidic device that contains human intestinal epithelial cells cultured on one surface of a porous membrane that separates the channels, which makes it possible to independently control the fluid flow in each channel and to establish molecular gradients across the epithelium. As Wnt signaling is known to mediate intestinal villus morphogenesis, and Caco-2 cells secrete both Wnt molecules (Munemitsu et al., 1995, Voloshanenko et al., 2013) and the Wnt-antagonist DKK-1 glycoprotein (Koch et al., 2009, Saaf et al., 2007), we explored whether the human Gut Chip can be used to analyze how gradients of Wnt agonists and antagonists interplay to promote intestinal morphogenesis under controlled conditions in vitro. Also, we extended this work to the Gut Chips lined by primary intestinal epithelial cells originated from biopsy-derived intestinal organoids to confirm the physiological relevance of our findings.

Results

Basal Fluid Flow Is Crucial for Intestinal Epithelial Morphogenesis

The Gut Chip is a microfluidic cell culture device composed of transparent silicone polymer (polydimethylsiloxane) that contains two apposed hollow microchannels separated by a flexible, extracellular matrix (ECM)-coated, porous membrane (Huh et al., 2013, Kim et al., 2012). The channels are lined on each side by hollow chambers that are exposed to cyclic vacuum to repeatedly strain and relax the porous membrane, thereby mimicking peristalsis-like deformations (Figure 1A). As previously demonstrated (Kim et al., 2012), when human Caco-2 intestinal epithelial cells are cultured on the upper surface of the porous flexible membrane and exposed to physiological fluid flow (30 μL/h; 0.02 dyne/cm2) and cyclic mechanical deformations (10% in cell strain; 0.15 Hz in frequency), these epithelial cells spontaneously undergo 3D intestinal morphogenesis (Figure 1B) with finger-like projections extending vertically up to ∼300 μm in height after 5 to 7 days of culture (Figure S1). Importantly, human primary intestinal organoid-derived epithelial cells also form similar structures when cultured in the Gut Chip under physiological flow and motions (Figure 1C), as demonstrated previously (Kasendra et al., 2018). On the other hand, a Caco-2 cell monolayer maintained its planar form even when analyzed for up to 8 weeks of culture (Figure 1D). Previous studies suggested that fluid flow is more important than mechanical deformations for induction of 3D morphogenesis in this system (Kim et al., 2012), and when we repeated these studies with or without cyclic mechanical strain, we confirmed that the 3D intestinal histogenesis occurs under both conditions (Figure 1E). Thus, we then focused on how perfusing fluid flow similar to that observed within the lumen of the living intestine while flowing medium below to mimic vascular or interstitial flows that exist in vivo (Granger, 1981), which might influence this developmental process.

Figure 1

A Human Gut Chip Model of Intestinal Morphogenesis

(A) A schematic of the microfluidic Gut Chip containing villi-like intestinal epithelial cells adherent to the upper surface of the flexible, porous, ECM-coated membrane in the top channel of the Gut Chip (light blue and orange arrows indicate flow directions in the upper and lower microchannels, respectively; white arrows indicate mechanical deformations through application of cyclic suction).

(B) Differential interference contrast (DIC) microscopic top-down views at low (top left) and high magnification (top right) and a fluorescence microscopic cross-sectional view highlighting the nuclei (bottom; DAPI) of intestinal villi-like microarchitecture that formed spontaneously when the Caco-2 epithelium was exposed to continuous flow (30 μL/h) in both channels and cyclic mechanical strain (10%; 0.15 Hz) for approximately 100 h.

(C) Human primary organoid-derived 3D epithelial growth in a Gut Chip. DIC microscopic top-down views at low (top left) and high magnification (top right) and a vertical cross-sectional view of 3D epithelial layer fluorescently visualized the plasma membrane (bottom).

(D) A schematic diagram (left) and a phase contrast view (right) of a planar monolayer of Caco-2 cells cultured in a static Transwell insert for 8 weeks.

(E) Phase contrast views showing the villi-like morphogenesis after 96 h regardless of the absence (−Str) or the presence (+Str) of cyclic mechanical strain under continuous flow in both microchannels (30 μL/h).

(F) A schematic diagram (left; an arrow indicating the fluid flow) and a phase contrast view (right) of Caco-2 cells grown on a single-channel microfluidic device without mechanical strain in the presence of flow and apical shear stress (0.02 dyne/cm2) for 150 h. A white arrow indicates a dome formed in the cell monolayer.

(G) A schematic diagram (left) and a phase contrast view of the epithelium in a Gut Chip in which human microvascular endothelial cells (“Endo”) were pre-cultured on the opposite side of the membrane from the Caco-2 intestinal epithelial cells in the lower channel to block access of fluid shear stress (30 μL/h) to the basolateral surface of the epithelium without mechanical deformations. The schematic diagram depicts the experimental setup at the point of the co-culture that both endothelium and epithelium independently formed a monolayer.

Scale bars, 50 μm.

We first explored whether apical shear stress due to luminal fluid flow above the epithelium is responsible for induction of the epithelial morphogenesis. To test the effect of apical shear stress independent of the basal fluid shear generated in the lower channel, we cultured the intestinal epithelium without mechanical deformation in a single channel microfluidic device. Even though these cells were cultured under fluid flow and experienced apical shear stress, they did not show 3D morphology as in the two-channel microfluidic Gut Chip (Figure 1F). Instead, they grew as an epithelial monolayer with occasional epithelial domes being observed, as previously described in other static Caco-2 cell cultures (Ramond et al., 1985). Another possibility is that application of fluid shear stress to the basal surface of the Caco-2 cells could drive formation of the villi-like 3D structure given that the central membrane contains multiple large pores (10 μm in diameter). However, when we co-cultured the epithelial cells with capillary endothelial cells pre-grown on the surface of the porous membrane in the lower microchannel to eliminate the direct basal mechanical signal (Figure S2), Caco-2 cells continued to form 3D morphology in the upper channel in the absence of mechanical deformations (Figure 1G). Thus, application of fluid shear stress to neither the apical nor the basal surfaces of the intestinal epithelium is responsible for inducing 3D villi-like morphogenesis.

Next, we explored the possibility that the presence of continuous fluid flow might remove secreted molecules, such as Wnt antagonists, which have been reported to suppress villi formation in the past in vitro and in vivo studies (Kuhnert et al., 2004, Pinto et al., 2003). Interestingly, when we flowed medium simultaneously through both the upper and lower channels (Figure 2A, left), or through the lower channel alone while maintaining a static epithelial cell culture in the upper channel (Figure 2A, middle), 3D morphogenesis progressed normally; however, it took approximately 1.5 times longer to form villi when fluid was only flowed through the basal channel. In contrast, when the experimental protocol was reversed and fluid was only flowed above the epithelium in the upper channel, this epithelial morphogenesis was completely inhibited (Figure 2A, right).

Figure 2

Basal Flow and Removal of Polarized Secreted Molecules from the Epithelium Are Necessary for Intestinal Morphogenesis

(A) Diagrams (top) and phase contrast images (bottom) of villi-like epithelium when culture medium flowed (30 μL/h) through both upper and lower microchannels (left) or only through the basal channel with no flow in the apical microchannel (middle), but not when fluid flow was stopped in the basal channel with flow continuing in the top channel (right).

(B) Schematics (top) and DIC views (bottom) showing that a planar Caco-2 monolayer cultured in a Transwell for 3 weeks can be induced to form villi-like protrusion (white arrow) by transferring the Transwell insert to a hybrid microfluidic device and applying constant flow (30 μL/h) in the basal chamber for the next 48 h (middle) or to a larger static culture well containing excess medium (70 mL) to dilute factors contained within the basal chamber for 120 h to sufficiently diffuse out basolaterally released secretomes (right). The intestinal epithelium remains as a planar monolayer in the static Transwell even after culture for up to 6 weeks (left).

(C) A monolayer of human intestinal organoid-derived epithelium pre-cultured in a Transwell insert underwent 3D morphogenesis when setup was transferred into the hybrid microfluidic device (left), whereas the same organoid-derived epithelium maintains a planar monolayer in the static condition (right). Inset schematics show the experimental setups. Cyan, F-actin; white, nuclei.

Scale bars, 50 μm.

One explanation for these observations is that the intestinal epithelial cells might secrete a certain type of inhibitory factor in a polarized manner causing it to concentrate in the lower basal channel and thereby feedback via basal membrane receptors to inhibit villi-like epithelial growth. To verify this hypothesis, we collected the conditioned medium from the basolateral side of Caco-2 cells grown for 3 days in Transwells, and then flowed this conditioned medium into the lower microchannel of the Gut Chip while fresh culture medium was flowed through the apical channel. Surprisingly, the introduction of the basally collected conditioned medium completely inhibited the 3D morphogenesis of Caco-2 cells in the Gut Chip (Figure S3). Moreover, when we cultured the Caco-2 cells on both sides of the same porous membrane, the formation of the villi-like structure was also suppressed in both microchannels (Figure S4). Taken together, these results suggest that the Caco-2 intestinal epithelial cells may secrete inhibitory factors basally that potentially antagonize the 3D morphogenesis of Caco-2 cells.

To further investigate this mechanism, we designed a hybrid device (Figure S5) that holds a Transwell insert and basally is in contact with a lower microfluidic channel that continuously removes epithelial secretomes released into the basal chamber of the Transwell. After Caco-2 cells were grown as a planar monolayer under the static condition for 3 weeks in a Transwell (Figure 2B, left), the Transwell setup was transferred to the hybrid microfluidic device. While the Caco-2 monolayer was maintained without flow apically, the medium was continuously flowed in and out (30 μL/h) through the basal microfluidic chamber, where we observed a rapid formation of 3D morphogenesis within 48 h (Figure 2B, middle). Moreover, this structural formation was similarly induced in the static Transwell inserts by simply diluting the basal medium by >100-fold in volume (Figure 2B, right), which was accomplished by placing the Transwell insert (0.33 cm2 in surface area) in a larger culture dish containing 70 mL of static basal culture medium for 120 h. Importantly, we also replicated this 3D histogenesis response using primary intestinal epithelial cells derived from human intestinal organoids (Figure 2C, left), whereas a 2D monolayer of primary epithelium was maintained under static conditions (Figure 2C, right). Thus, the removal of inhibitory factors released basally by the polarized intestinal epithelium can rapidly (<2 days) trigger intestinal morphogenesis. We also observed that the cessation of basolateral flow induced the loss of pre-formed 3D epithelial microstructure within 4 days (Figure 3A). Furthermore, the number of proliferative cells labeled with Ki67 was significantly decreased under basolateral cessation of flow (<7%) compared with the control (>50%) (Figure 3B). However, a live/dead staining assay revealed that the number of dead epithelium was negligible when the basolateral flow was stopped (Figure 3C), suggesting that the loss of villi-like morphology was caused by the reduced proliferation rather than the cell death without a loss of epithelial barrier function (Figure S9B, “Control” vs. “BL ceased”).

Figure 3

Cessation of Basal Flow Decreases Population of Proliferative Cells but Maintains Cell Viability

(A) A schematic (left) and phase contrast micrographs showing that the Caco-2 villi-like structure pre-formed within a Gut Chip device for ∼100 h (“Before”) lost the microarchitecture when basal flow in the lower channel was ceased for 90 h (“After”).

(B) Visualization of the proliferative Ki67-positive Caco-2 cells (magenta) in the absence (“BL ceased”) or the presence (“Control”) of basolateral flow in the Gut Chips (left, middle). Quantification of the Ki67-positive cells normalized by the total number of nucleated cells (gray) (right, N = 5).

(C) Viability of Caco-2 epithelium cultured in the absence (“BL ceased”) or the presence (“Control”) of basolateral flow in the Gut Chips, assessed by staining with Calcein AM (Live, green) and ethidium homodimer-1 (Dead, red), and quantification of cell viability (right, N = 10). The basal flow was ceased for 48 h after 3D villi-like structure was formed in the Gut Chips by culturing for ∼100 h.

N.S., not significant, **p < 0.001; scale bars, 50 μm.

Wnt Antagonists Suppress the On-Chip Morphogenesis of an Intestinal Epithelium

Past works have shown that the canonical Wnt signaling pathway promotes villus morphogenesis in the embryonic intestine (Peifer and Polakis, 2000, Pinto et al., 2003) and human organoids (Ootani et al., 2009) via autocrine regulation. To explore whether Wnt signaling mediates the 3D histogenesis in the human Gut Chip, we, respectively, added human recombinant Wnt antagonists including DKK-1 (rDKK-1), Wnt inhibitory factor 1 (rWIF-1), secreted frizzled-related protein 1 (rsFRP-1), and Soggy-1/DKK-like 1 (rSoggy-1/DKKL-1) (Kawano and Kypta, 2003) to the culture medium that was perfused to the lower microchannel of the Gut Chips, which had pre-formed villi-like epithelium (Figure 4A, “Control”). As expected, all of these Wnt antagonists induced the loss of 3D morphology within 48 h of the exposure (Figures 4A and S6), and the percentage of epithelial surface that exhibited the morphologically blunted lesion was significantly higher in the cultures treated with each of the Wnt antagonists compared with the control (Figure 4A, bottom).

Figure 4

Wnt Antagonists Inhibit Intestinal Villous Morphogenesis

(A) Phase contrast views (top) and a graph (bottom) showing that addition of human recombinant versions of rDKK-1, rWIF-1, rsFRP-1, or rSoggy-1/DKKL-1 proteins (all at 100 ng/mL) to the medium flowing through the basal channel of the Gut Chips containing well-developed Caco-2 villi-like epithelium (“Control”) resulted in the loss of 3D microarchitecture (N = 10) within 48 h.

(B) Secreted amount of DKK-1 from a Caco-2 monolayer cultured in a Transwell for 48 h (N = 3).

(C) Phase contrast views (left) and a graph (right; N = 10) showing that addition of increasing concentrations of rDKK-1 (0, 100, and 500 ng/mL) resulted in a dose-dependent suppression of the 3D epithelial morphogenesis. The culture medium containing rDKK-1 was perfused into the basolateral microchannel at 72 h since the seeding. The overall time course of villus morphology is provided in Figure S7.

(D) Phase contrast views (left) and a graph (right; N = 10) showing that the inhibitory effect of rDKK-1 (500 ng/mL) was suppressed by addition of a blocking anti-DKK-1 antibody (20 μg/mL). The rDKK-1 and anti-DKK-1 antibody were treated to villi-like epithelium for 48 h.

Scale bars, 50 μm; *p < 0.001, **p < 0.05.

We then selected the most potent and well-characterized antagonist, DKK-1 (Aguilera and Munoz, 2007, Gonzalez-Sancho et al., 2005), to further investigate the mechanism of inhibition. First, we confirmed that the Caco-2 cells cultured in a Transwell secrete approximately 5.3-fold more DKK-1 (p < 0.001) into the basal chamber than into the apical side (Figure 4B), which shows a good agreement with our observation in Figures 2 and 3. Addition of rDKK-1 to the basal channel resulted in a statistically significant (p < 0.05), dose-dependent reduction of the height of 3D epithelium (Figures 4C and S7). Moreover, when we analyzed the same location in the chip over time, we found that, although the presence of the rDKK-1 antagonist for 48 h resulted in the loss of villi-like structure (Figure S8A), removal of rDKK-1 resulted in the rapid restoration of epithelial 3D growth within 24 h (Figure S8B). However, the villi-like microarchitecture was involuted once again when we resumed rDKK-1 treatment for an additional day (Figure S8C). We also confirmed that the inhibition of 3D epithelial morphogenesis by rDKK-1 can be successfully suppressed by the co-treatment of an anti-DKK-1 monoclonal antibody (Figure 4D) for neutralizing the antagonistic function of DKK-1. We also confirmed a decreased population of Ki67-positive proliferative cells when the rDKK-1 was treated to Caco-2 villous epithelium for 48 h (Figure S9A), whereas the barrier integrity was well maintained (Figure S9B). Furthermore, when we added the same anti-DKK-1 antibodies to the Caco-2 monolayers grown in static Transwells (Figure S10), the average height of the cell monolayer significantly increased compared with the control (Figure S10C; p < 0.0001). Interestingly, nucleated cells were observed beginning to extend above the surface of the planar epithelial monolayer (Figure S10B, a zoomed-in inset), suggesting that the addition of anti-DKK-1 antibody contributed to the initiation of morphogenesis.

Computer Simulation Predicts Transepithelial Morphogen Gradient

We then built a multi-physics, finite element model of the Gut Chip to better understand how polarized secretion of Wnt antagonists and generation of the gradient of a Wnt inhibitor may contribute to the spatiotemporal control of epithelial 3D growth patterns. This simplified computational model assumes that DKK-1 is the most relevant and potent morphogen antagonist and computes the concentration of DKK-1 and Wnt within the geometry of the Gut Chip, taking into account the relative production rate of both molecules by the intestinal epithelium, diffusion through the medium, and convection due to the fluid flow. We estimated the diffusion coefficient of DKK-1 to be two orders of magnitude greater than that of Wnt based on the past work (Sick et al., 2006), which we set at 9.3×10−11 and 6.9×10−13 cm2/sec, respectively. The production rate of DKK-1 (421 pg/106 cells/h) by the Caco-2 cells was applied from the previous study (Koch et al., 2009), where we postulated the same production rate for Wnt because binding of Wnt to its receptors stimulates production of both itself and DKK-1 (Sick et al., 2006). Quantitation of Caco-2 cell numbers in the Gut Chip revealed that the epithelial cell layer contains ∼5.0×105 cells/chip (∼4.5×106 cells/cm2).

We first performed the simulation to explore if the accumulated DKK-1 can form a stable gradient in the lower microchannel of a simplified 2D representation of the Gut Chip (Figure S11). Under static flow conditions (i.e., 0 μL/h), DKK-1 was simply accumulated in the basal channel; however, as flow rate is increased up to 30 μL/h, the simulation predicts that a gradient of DKK-1 molecules will be generated with lower concentrations at the bottom of the basal channel and toward the inlet. Furthermore, when the flow rate was increased above 50 μL/h, the model predicted that DKK-1 levels would fall to almost zero in the lower channel.

We then used a more complex 3D model to analyze the concentration of DKK-1 along the length and width of a perfused channel in the Gut Chip, assuming that the polarized cells are secreting DKK-1 from their basolateral surface. This model predicted that the concentration of DKK-1 near the inlet of the microchannel would be relatively low, whereas the level of DKK-1 will increase by 10-fold or more near the channel outlet (Figures 5A and S12). Interestingly, when we experimentally analyzed villus growth at 80 h in regions corresponding to the upstream, middle, and downstream positions in the microchannel (Figure 5A), we observed that there was a gradient of pseudo-villous growth that closely mirrored the pattern predicted by the computational model. For example, vigorous 3D morphogenesis was observed in the upstream region, whereas the least formation of the villi-like structure was found in the downstream portion in the microchannel (Figure 5B). The 3D computational model also predicted a parabolic distribution of DKK-1 accumulation at the surface of the porous ECM-coated membrane, with lower concentrations in the center of the channel and higher concentrations near the sidewalls of the channel (Figures 5A and S12). This result is attributed to the parabolic profile of laminar flow in the microchannel, where the linear flow velocity will be the highest in the center, therefore washing away DKK-1 at a higher rate, with DKK-1 accumulation where the flow is the lowest near the channel walls. In fact, we experimentally confirmed that the growth of 3D epithelium exhibited a similar pattern on-chip, in which both the height and abundance of the villi-like structure are higher in the middle of the channel compared with its sides near the wall where the inhibitor levels are lower (Figure 5C).

Figure 5

The Spatiotemporal Pattern of Villi-like Epithelial Structure Is Governed by the Gradient of Morphogen Antagonists

(A) A computational simulation based on the results shown in Figure S11 displaying the spatial pattern of DKK-1 concentrations at the bottom surface of the membrane under the flow (30 μL/h) mapped on a schematic of a Gut Chip, indicating that it will form a parabolic gradient and exhibit a gradient of concentration levels with lowest near the inlet and highest downstream near the outlet. Color bar, the scaled range of DKK-1 concentrations (unit, ×10−9, mol/m3). An inset at the right shows the structure of a Gut Chip, where the orange, green, and blue boxes overlaid on the device design at the right indicate the location of up-, mid-, and downstream snapshots of the microchannel shown in (B); red arrows indicate the flow direction.

(B) Stitched DIC images showing that growth patterns of the Caco-2 villi-like epithelium along the longitudinal gradient in the corresponding up-, mid-, and downstream regions of the Gut Chip as indicated by colored squares in (A) at 80 h.

(C) Vertical cross-sectional (upper) and top (middle) views of immunofluorescence images of the Caco-2 epithelium grown on-chip for 80 h at mid-stream region shown in (A) and stained with a fluorescent membrane dye (green) to highlight the surfaces of the cells and laid over the DIC image. A graph at the bottom shows the distribution of epithelial height across the width of the microfluidic channel at the mid-stream position designated with a green box in (A–C). N = 5; scale bars, 100 μm.

Interestingly, we experimentally observed that the intermediate flow rate regime (70–120 μL/h) resulted in faster growth of taller villi-like epithelial structure per unit time compared with flow rates below 30 μL/h or above 120 μL/h, which produced significantly slower morphogenesis (Figures 6A and S13). We then compared effects of three representative flow rates (30, 100, and 200 μL/h) in the Gut Chips versus static Transwell cultures at 48 h and performed quantitative real-time polymerase chain reactions (qPCR) targeting 92 human Wnt-related genes. The qPCR results revealed that only three genes, G protein-coupled receptor (GPCR) frizzled 9 (FZD9), Myc (MYC), and lymphoid enhancer-binding factor 1 (LEF1), were significantly (p < 0.05) upregulated in the Gut Chips under flow compared with Transwells. Consistent with the predictions from the computational model, the Wnt receptor FZD9 exhibited the highest and the most significant (p < 0.01) upregulation (∼66-fold increase) at a flow rate of 100 μL/h, whereas cells in the chips exposed to lower or higher flow rates (30 or 200 μL/h, respectively) exhibited less increment in FZD9 (∼49- and ∼36-fold, respectively) (Figure 6B). However, there was no significant difference in MYC and LEF1 regardless of the flow rates. Immunofluorescence confocal microscopy showed that the expression level of FZD9 was significantly (p < 0.001) higher in the fluidic than in the static culture condition in both Caco-2 (Figure 6C) and organoid-derived epithelium (Figure S14). A computational simulation accounting for the production rates of DKK-1 and Wnt molecules at the basolateral membrane revealed that the relative ratio of DKK-1 and Wnt at the steady state is almost constant at flow rates greater than ∼30 μL/h, whereas the ratio exponentially increases as flow rates approach static conditions (i.e., 0 μL/h) (Figure 6D). This result suggested that the flow-dependent 3D morphogenesis of Caco-2 cells is predominantly orchestrated by the FZD9 receptor because the DKK-1/Wnt ratio was almost constant regardless of the flow rate at >30 μL/h.

Figure 6

Mediation of Intestinal Villi-like Morphogenesis by the Wnt FZD9 Receptor

(A) A graph showing profiles of epithelial height at various flow rates (15, 30, 70, 100, 150, and 200 μL/h) over time. The highlighted region indicates the optimal flow rate regime that maximizes the formation of villi-like morphology (Error bars are smaller than the symbol size).

(B) A graph showing fold increase in expression of FZD9, MYC, and LEF1 genes in Caco-2 intestinal epithelial cells grown in the Gut Chip at different flow rates (30, 100, and 200 μL/h) compared with the static Transwell (*p < 0.05; N.S., not significant).

(C) Expression of FZD9 on Caco-2 villi-like epithelium in fluidic and static culture conditions, visualized by immunofluorescence imaging and 3D reconstruction of z-stacked images (left). Caco-2 cells were cultured in a Gut Chip under 30 μL/h in the fluidic condition or a 96-well plate for 100 h as a static condition. The intensity of each condition was quantitated using Image J (right) (***p < 0.0001; N = 5).

(D) Mathematical prediction of the ratio of DKK-1 and Wnt molecules in the Gut Chip probed at the surface of the porous membrane within the lower channel for a range of flow rates.

(E) Phase contrast microscopic views (top) and a graph (bottom) show that addition of blocking anti-FZD9 antibodies (“+Anti-FZD9 Ab”; 20 μg/mL) successfully inhibited the formation of a 3D villi-like epithelial layer and resulted in the significant decrease in epithelial height after 24 h when compared with control. (*p < 0.01; scale bars, 50 μm).

To confirm if FZD9 is a critical receptor for intestinal morphogenesis, anti-FZD9 blocking antibodies were added to the villi-like epithelium pre-established in the Gut Chip. We found that the infusion of anti-FZD9 antibodies (20 μg/mL) into both the apical and basal microchannels of the Gut Chip substantially altered the 3D morphology (Figure 6E). Epithelial height was 122.5 ± 4.2 μm in the control group, whereas it was significantly (p < 0.05) reduced to 75.8 ± 2.7 μm when FZD9 receptors were blocked (Figure 6E). Taken together, our finding suggests that FZD9 is a key receptor that mediates control of intestinal morphogenesis through its interactions with Wnt and DKK-1 in the microengineered model.

Discussion

Our mechanistic study that leverages a microphysiological Organ Chip uncovers the molecular basis of a developmental morphogenic process in vitro that is governed by complex cellular signaling. The Gut Chip enabled separate access to the apical lumen and basal abluminal compartments of this engineered intestinal epithelium, as well as precise independent control over fluid flow rates, molecular components, and, hence, transepithelial gradients while allowing high-resolution microscopic imaging. By manipulating biophysical and biochemical cues in the Gut Chip, we discovered that the Wnt antagonist DKK-1 is secreted in a polarized basolateral direction and that its removal by fluid flow in the basolateral microchannel is a crucial factor that directly triggers intestinal 3D morphogenesis in this in vitro model using Caco-2 as well as the primary organoid-derived epithelial cells. We also discovered that the Wnt receptor FZD9 mediates this morphogenic response, where its expression level is dependent on the flow rate and correlates with epithelial differentiation. Experimental results that we obtained were verified with the computational modeling designed to understand the molecular distribution and dynamics of secretory DKK-1 and Wnt in the Gut Chip. By using a simple computational simulation, we successfully explained the morphogenic patterns of epithelial growth inside the microfluidic channel reminiscent of the intestinal development.

Our past finding that human Caco-2 intestinal epithelial cells spontaneously undergo villus morphogenesis and small-intestine-specific cytodifferentiation and histogenesis in the microfluidic Gut Chip (Kim et al., 2012, Kim and Ingber, 2013) was surprising because Caco-2 cells cultured under static conditions, or even under microfluidic flow, did not exhibit this response in prior studies. The results of the current study now explain this disparity because those past studies did not include basolateral flow (Gao et al., 2013, Imura et al., 2009). However, although fluid flow was found to be more critical than peristalsis-like deformations in triggering the formation of villi-like epithelial microarchitecture, we found that direct application of fluid shear stress to the epithelial cells is not sufficient to induce the morphogenesis in the Gut Chip. This observation suggested that fluid flow might influence histogenesis by altering the delivery or removal of soluble signaling factors. Thus, we explored whether fluid-flow-dependent changes in delivery of Wnt and removal of Wnt-antagonistic molecules such as WIF, sFRP-1, DKKL-1, and DKK-1 modulate structural changes. WIF-1 binds directly to Wnt proteins and prevents the initiation of the Wnt-signaling pathway (Malinauskas et al., 2011). sFRP-1 also binds to Wnt proteins to inhibit Wnt signaling as an antagonist (Bovolenta et al., 2008). DKKL-1, also known as Soggy-1, is a homologue to the DKK family proteins. However, DKKL-1 does not affect canonical signaling of the Wnt/β-catenin pathway, which acts differently from DKK-1 protein (Yan et al., 2012). Regardless of their specific mechanism, all of these Wnt antagonists involuted the preformed Caco-2 villi-like structure. We further studied how the removal of Wnt antagonists mediates the intestinal morphogenesis by using DKK-1, which have been previously implicated in control of intestinal morphogenesis (Crosnier et al., 2006, Ootani et al., 2009) and are known to be produced by Caco-2 epithelial cells (Koch et al., 2009, Voloshanenko et al., 2013).

Our study revealed that the reduction of basolaterally secreted DKK-1, potentially other Wnt antagonizing molecules as well, is the crucial trigger that orchestrates the epithelial histogenesis in the Gut Chip as well as in the hybrid microfluidic device. When we removed secreted DKK-1 under either fluidic or diffusion-based conditions, 3D morphogenesis of intestinal epithelial cells was promoted. This approach was verified even in the presence of an endothelial layer at the opposite side of the porous membrane in the Gut Chip. It is noted that the endothelial layer we cultured in this experiment has a high permeability to large molecules via transcytosis (Fung et al., 2018, Mehta and Malik, 2006). Therefore, DKK-1 that are basolaterally secreted by the Caco-2 cells can be readily transported through the endothelial layer, which potentially leads to the epithelial morphogenesis.

In the normal intestinal epithelium, Wnt molecules are mainly localized on the external cell surface where they activate downstream Wnt signaling, and diffusion of Wnt molecules to adjacent epithelial cells is negligible (Farin et al., 2016). Caco-2 cells have shown a continuous activation of Wnt/β-catenin signaling because of truncating mutations in APC and β-catenin (Voloshanenko et al., 2013). Interestingly, autologous secretion of DKK-1 by Caco-2 cells concomitantly occurs in conventional Caco-2 cultures at about 42.1 pg/105 cells/h (Koch et al., 2009, Takahashi et al., 2010). However, it is notable that the presence of DKK-1 does not completely block the proliferation of Caco-2 cells (Koch et al., 2009), which has been confirmed in our previous and current studies. As the inhibitory DKK-1 molecules are freely secreted and bind to membrane receptors such as low-density lipoprotein receptor-related protein (LRP) 5/6 (Bafico et al., 2001), these observations are consistent with the mechanism we uncovered here, in which removal of the secreted Wnt inhibitor DKK-1 can directly initiate intestinal morphogenesis in Caco-2 epithelium. Furthermore, the presence of DKK-1 can induce disruption of intestinal villi in the in vivo mouse models (Kuhnert et al., 2004), which was also replicated in our current in vitro study (Figure 4). Although the averaged molecular weight of Wnt family members (38–42 kDa) (Gavin et al., 1990) is larger than DKK-1 (28.7 kDa) (Aguilera and Munoz, 2007), our computational model suggests that, as a consequence of production and secretion by Caco-2 cells, the ratio of DKK-1 and Wnt molecules at the steady state is almost constant at flow rates higher than 30 μL/h, suggesting that the ratio of DKK-1/Wnt is independent of flow rate. Thus, the flow-dependent profile of secreted Wnt antagonist molecules, such as DKK-1, is likely a key feature that controls the intestinal 3D morphogenesis in the Gut Chip.

We found that the cessation of basolateral fluid flow in the Gut Chip can also induce the loss of preformed villi-like microarchitecture. We hypothesized that the disappearance of 3D epithelium under the cessation of basolateral flow might result from either the increased death of cells or the decreased proliferation of cells. We discovered that the cessation of basal flow significantly decreased the number of Ki67-positive proliferative cells, whereas the control group that experienced continuous basolateral flow maintained proliferative population more than 10-fold. However, neither the cell viability nor the barrier function was compromised in response to both the cessation of basal flow and treatment of rDKK-1, suggesting that the maintenance of proliferative cell population may be a crucial element to sustain the intestinal epithelial morphogenesis and its microarchitecture over time.

The directional secretion of DKK-1 in a polarized monolayer of the human intestinal epithelium has not been reported, although secretion and antagonism of DKK-1 in Wnt/β-catenin signaling is a common developmental feature in vertebrates (Farin et al., 2012, Glinka et al., 1998). Furthermore, we also verified the 3D morphogenesis mechanistically using primary epithelial cells obtained from human intestinal organoids in the same Gut Chip as well as in the hybrid microfluidic device. Since the organoid-derived epithelial culture is constitutively supported by the high level of Wnt, R-Spondin, and Noggin (Saxena et al., 2015), it was evident that the removal of morphogen antagonists by fluid flow in the basolateral side is critical for the intestinal morphogenesis in this primary cell culture model. Although the regeneration of 3D microarchitecture of organoid-derived epithelium was previously reported using human small intestinal organoids (Kasendra et al., 2018), it has not been clear which factor triggers this epithelial morphogenesis.

The wave of rostral-to-caudal (oral-to-anal) formation of the villi during the intestine development has been long recognized (Johnson, 1910, Kammeraad, 1942, Walton et al., 2012). This proximal-to-distal wave of development results in a progressive decrease in the height of villi along the length of the intestine from duodenum to jejunum, ileum, and colon (Walton et al., 2012). Thus, the temporal growth pattern observed in the present study in which villi-like structure first emerged in the proximal (upstream) region of the Gut Chip near the inlet where their heights are also the longest, and then they progressively shorten toward the outlet, is remarkably reminiscent of what is observed during vertebrate intestine development (Spence et al., 2011). However, our results are different from those of the past in vivo study, which suggested that Hedgehog (HH)-dependent intestinal mesenchymal cell clusters orchestrate the patterning and generation of villi within the adjacent intestinal epithelium (Walton et al., 2012). Thus, it will be interesting to explore whether the Wnt signaling we discovered occurs in vivo and, if so, how it interplays with HH signaling-mediated epithelial-mesenchymal interactions. Mesenchymal cells (e.g., intestinal fibroblasts) that produce tissue-specific morphogens (Powell et al., 2011) could be potentially integrated into the Gut Chip to explore this mechanism in vitro in future studies.

We initially expected that the growth rate of villi-like epithelium would proportionally increase as a function of flow rate because higher flow rates should remove Wnt antagonists more efficiently. However, among six different flow rates in a range of 15-200 μL/h, the epithelial growth we observed experimentally was slower at flow rates >100 μL/h compared with the intermediate flow rate regime. This discrepancy suggested that there may be an additional factor that orchestrates the epithelial morphogenesis independently of constitutive competition between Wnt agonists and antagonists. Our qPCR analysis revealed that the expression of the Wnt receptor FZD9 significantly increased among 92 human Wnt-related genes at the intermediate-flow-rate regime compared with lower or higher rates, which correlated directly with the degree of morphogenesis. Immunofluorescence imaging revealed that the fluid flow significantly increased the expression of FZD9 receptor in both Caco-2 and human organoid-derived epithelium, suggesting that FZD9 expression is dependent on fluid flow. Moreover, blocking the upstream signaling by neutralizing FZD9 remarkably reduced the formation of villi-like structures in Caco-2 cells. We revealed that the expression of FZD9 is regulated by the fluid flow, and it represents a morphogenetic control mechanism. The function of FZD9 is poorly understood compared with other well-characterized FZD receptors (Bafico et al., 1999). Thus the use of the microfluidic human Gut Chip may help to mechanistically unravel the role of FZD9 in development and morphogenesis in the human intestine.

Intestinal morphogenesis has been promoted in vitro previously by culturing intestinal organoid-derived epithelial cells in the presence of Wnt (Farin et al., 2012, Sato et al., 2011), or Wnt-producing Paneth-like cells (Sato et al., 2011) or mesenchymal cells (Valenta et al., 2016). Interestingly, the cell-elaborated Wnt factors, which normally contribute to generation of localized Wnt gradients in the intestinal crypt microenvironment, induce formation of crypt-like protrusion with intervening villus domains within the crenulated organoids, whereas addition of soluble Wnt often promotes the formation of round spheroids that fail to undergo villus morphogenesis (Farin et al., 2012, Sato et al., 2009). Thus, organoids require the presence of live Wnt-producing cells to create spatiotemporal gradients of Wnt (and possibly its inhibitors as well) that are required for histogenesis, which is consistent with the importance of mesenchymal clusters observed during intestinal development in vivo (Walton et al., 2012). In contrast, Caco-2 epithelial cells, which were originally isolated from a tumor, secrete Wnt molecules constitutively and exhibit autocrine activation of FZD9 receptors, which may contribute to their stem cell-like behaviors as well as their ability to undergo 3D morphogenesis in the absence of mesenchymal cells. However, similar observation when primary organoid-derived human intestinal epithelial cells were cultured on-chip suggests that the removal of Wnt antagonists and the flow-dependent expression of Wnt receptors may play a pivotal role in the control of intestinal morphogenesis. Furthermore, when the organoid-derived epithelium is cultured in static, intestinal morphogenesis does not occur regardless of the presence of Wnt, R-spondin, noggin, and various growth factors (Ettayebi et al., 2016, Noel et al., 2017). Thus, our finding in this study and the prior reports strongly suggest that the demonstration of epithelial 3D morphogenesis may be predominantly driven by the removal of morphogen antagonists rather than by the addition of morphogen or the origin of cell source (e.g., cancerous vs. normal).

The current study provides an exceptional example showing how spatial control of morphogen antagonists can orchestrate epithelial morphogenesis by establishing asymmetry across the epithelial cells and sustaining differences between their apical versus basolateral microenvironments during tissue growth. This finding is consistent with the past observation that spatiotemporal morphogen gradients and control of extracellular morphogen antagonists are as important as the type of morphogen for control of development (Kawano and Kypta, 2003). Although the Wnt gradient in vivo has been well characterized (Scoville et al., 2008), the gradient profile of DKK-1 in vivo has been insufficiently discussed (Du et al., 2013), suggesting that the power of the Organ Chip technology is that we can identify the complex mechanism by precisely controlling gradients and independently varying potential morphogenic parameters one at a time, in both time and space. This ability to control directional flow rates, fluid shear stresses, mechanical deformations, and asymmetric stimulation of the apical versus the basolateral side of a developing epithelium cannot be easily achieved in any other culture system or animal model. Thus, Organ Chips may offer a compelling in vitro tool to decipher cellular, molecular, and biophysical mechanisms of developmental control that underlie histogenesis of the intestine, as well as other epithelial tissues.

In summary, we discovered that human intestinal morphogenesis is controlled by a transepithelial gradient of the Wnt antagonist DKK-1 and flow-dependent induction of the Wnt FZD9 receptor using a microfluidic Gut Chip as a model system. DKK-1 is secreted asymmetrically across the epithelium resulting in higher concentrations in the basal compartment, where the presence of basal fluid flow removes this inhibitor and promotes intestinal epithelial morphogenesis. The expression of FZD9 varies with flow rate, and the location and height of Caco-2 intestinal villi-like epithelium scale directly with its expression levels. This microfluidic experimental platform can be further expanded to incorporate other cell types (e.g., mesenchymal cells) to explore how they contribute to this morphogenic response, in addition to exploring other unknown questions in developmental biology that involve the establishment of chemical gradients or variations in the local physical microenvironment.

Limitations of the Study

Caco-2 intestinal epithelium formed in the Gut Chip might not sufficiently recapitulate the normal physiology observed in vivo. This study was performed in the absence of other cell types such as mesenchymal cells or vascular components that may support epithelial morphogenesis in the intestinal microenvironment. For instance, myofibroblasts in the lamina propria area are known to produce morphogens and interact with the intestinal epithelium to control Wnt signaling (Roulis and Flavell, 2016). Therefore, incorporation of other tissue-specific cell types can further improve the accuracy of the Gut Chip model to study morphogenesis of intestinal epithelium in vitro. In addition, identification of cross talk between DKK-1 and FZD9 remains as a future study.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.