The human gastrointestinal (GI) tract is host to a microbiome community of trillions of bacteria, diversified by around a thousand unique species.1, 2 These commensal microbes play critical roles in nutrient metabolism, immune training and prevention of opportunistic infections.3, 4 Imbalance in the GI bacterial consortium is now implicated in the pathophysiology of several human diseases, including inflammatory bowel disorders,5 type 2 diabetes,6 and certain cancers.7 The gut microbiome also impacts distal systems, including the brain where neurologic processes are altered via communication through the gut-brain axis.8, 9 Yet, despite its link to human health, our understanding of the functional properties of the gut microbiome, and how gut ecology influences human physiology, remains incomplete.
This gap in knowledge is largely due to a lack of facile and accessible tools suitable to studying, and controllably manipulating, complex host-mucus-microbe interactions.10 Although gnotobiotic mice remain the gold standard, these expensive models are inaccessible for many researchers and do not always faithfully replicate human disease. In addition, it remains difficult to carefully shape microbiome composition and control mucus composition in the in vivo environment.11, 12 Tractable ex vivo models of the GI microbiome therefore represent powerful tools to study host-microbiota interactions in a multiparametric fashion at the cellular and molecular level.
Towards this goal, the development of gastrointestinal organoids – in vitro models of human intestinal epithelium – have expanded opportunities to study disease and probe microbe-host interactions ex vivo.13, 14 While these are valuable additions to the modeling spectrum, three-dimensional systems are expensive, require significant technical training, and can be limited by access to primary material, making them difficult to implement in high throughput applications. Monolayer cultures can address these limitations but are currently unable to faithfully replicate the colonic mucus, which is composed of inner and outer mucus layers that possess varied structural morphologies and densities.
Here, we report a self-assembling gel that can be engineered to replicate the diverse mechanical, structural, and biochemical profiles of colonic mucus. The material is generated from the interfacial organization of a non-natural amino acid at fluorous-water interphases to form a viscous, double layer, colloid of fibrillar and coacervate assemblies. We demonstrate this synthetic platform can be readily coated with a variety of mucin proteins and directly incorporated into multicellular systems to create customizable ex vivo microbiome organoids. We show that the mechanical, structural, and biochemical properties can be independently tailored to create designer multicellular systems amenable to high throughput applications. As an exemplary demonstration, we incorporate these materials into a gastrointestinal permeability assay and demonstrate their ability to model oral bioavailability of macromolecular compounds.
Fabrication and Molecular Characterization of Synthetic Mucus
While screening a surfactant library to prepare fluorous emulsions, we discovered the ability of the non-natural amino acid, Fmoc-pentafluoro-L-phenylalanine (Fmoc-FF), to form viscous, mucus-like, gels at liquid-liquid phase separated interfaces (Fig. 1a). The material is generated by pipetting a solution of perfluorodecalin (PFD) containing Fmoc-FF (20 mmol/L) into saline, leading to the spontaneous supramolecular organization of the amino acid at the fluorous-water interface. This rapid and irreversible step produces a dense coacervate gel, from which a fibrillar network evolves into the aqueous fraction when incubated at 37°C (Fig. 1b). This double layered structure mimics the typical morphology of colonic mucus, which is characterized by a dense inner layer that adheres to epithelial cells and a diffuse fibrillar outer layer that harbors commensal bacteria (Fig. 1c). To further visualize the bilayer composition of our synthetic mucus, the lipophilic dye nile red was dissolved in the PFD solvent before gel fabrication. Thioflavin T (ThT), a dye that displays enhanced fluorescence upon binding to amyloid structures,15 was added to the aqueous fraction to label Fmoc-FF assembled fibrils. Representative fluorescent micrographs shown in Fig. 1d demonstrate that the bottom layer of the material is a tightly packed network of PFD droplets, from which Fmoc-FF assembly initiates at the fluorous-water interface. Co-localization of nile red and ThT signals suggests these interfacial structures are amorphous hydrophobic oligomers and/or protofibrils, representing the nascent stages of Fmoc-FF assembly. Conversely, the upper layer is a diffuse collection of extended fibers characterized only by ThT fluorescence, with a marked absence of nile red (Fig. 1d, top). This suggests these are mature, organized fibrils that have evolved from rearrangment of the disorganized structures templated in the lower coacervate layer. Parallel electron microscopy, shown in Fig. 1e, further demonstrates that the less dense upper layer is an interpenetrating system of Fmoc-FF self-assembled fibrils. The cohesivity of this layer is imparted via non-covalent entanglement of fibers, creating a porous mesh-like architecture.
Next, we investigated the nature and rate of molecular organization of Fmoc-FF assemblies using the ThT dye. The multilamellar appearance of ThT-stained fibers suggests they are composed of β-sheet like plates that organize through n→π* stacking of Fmoc-FF residues (Fig. 2a).16, 17 This organization is strongly induced by the presence of the phase-separated PFD droplets, as demonstrated by the attenuated ability of Fmoc-FF to form fibers when PFD is absent in the saline solution (Fig. 2b). Here, the PFD-aqueous interface serves to rapidly template the assembly of Fmoc-FF monomers into a stable gel, from which fibers continue to grow and evolve over several days (Fig. 2c).
To further investigate this assertion, and specifically isolate fluorine-fluorine driven effects, we evaluated the assembly of non-fluorinated Fmoc-L-phenylalanine (Fmoc-F) under similar conditions (Fig. 2d). Results show this fluorine-deficient analogue is not capable of forming viscous fibrillar gels and instead generates colloidal emulsions, as indicated by the combination of high optical density (Fig. 2e) and low ThT fluorescence (Fig. 2f). Prior studies from our group suggest these divergent assembly pathways result from the propensity for perfluorocarbon-water systems to preferentially direct J-aggregate formation of Fmoc-FF, and not the non-fluorinated Fmoc-F analogue.17 This leads to Fmoc-FF being uniquely organized into anti-parallel arrangements, where fluorenyl moieties form alternate β-sheets to create π-stacked pairs with interleaved fluorinated phenyl rings. Propagation of these stacked assemblies likely yields the observed fibrils.17 Given the unique hierarchical organization of Fmoc-FF, induced by the presence of the perfluorinated phenyl ring and perfluorocarbon droplet interface, we hereafter refer to the assembled material as fluorine-assisted mucus surrogate, or FAMS.
To prepare the surface of FAMS for bacterial attachment we next coated the fibrils with mucin proteins (Fig. 3a). Here, simple addition of a protein solution to pre-formed FAMS led to rapid and robust fibril adsorption, producing a protein surface coating that was stable to multiple washes with media. Optimization of the coating procedure was done using two model fluorescent proteins, GFP and Cy5-BSA. Fluorescence microscopy demonstrated these proteins non-covalently decorate the materials and are retained after multiple washing (Fig. 3b, c). Similar studies were then performed using Cy5-labeled bovine submaxillary mucin (BSM) and porcine gastric mucin (PGM). As expected, addition of BSM lead to uniform coating of the material network (Fig. 3d), with additional SEM imaging demonstrating a relatively smooth surface topography to BSM-coated FAMS (FAMSBSM, Fig. 3e). This resembled the topology of reconstituted mucus prepared from the BSM protein stock (Fig. 3f), suggesting the mucin assembles into cohesive sheets that envelop the FAMS fibrillar network. Coating with PGM was similarly successful (Fig. 3g), although the surface morphology was more irregular (Fig. 3h) due to the uneven assembly of PGM itself (Fig. 3i). SEM imaging confirmed the morphology of BSM and PGM coatings was consistent across multiple length scales (Supporting Figs. 1 and 2). Taken together, our data indicates that addition of mucins does not disrupt the integrity of the FAMS super-structure, producing a mucin-rich double layer material that approximates the structural and biochemical characteristics of colonic mucus.
Mechanical Analysis of Mucin-Functionalized FAMS
A key checkpoint for mimicry of GI colonic mucus is material viscoelasticity, as the ability of native mucus to flow over long loading periods is necessary for the movement of solids during peristalsis. Mucus viscosity also has important implications in shaping microbial behavior and contributes to disease. For example, H. pylori can alter environmental pH to reduce the viscoelasticity of gastric mucus, thereby compromising the integrity of its barrier function.18 Similarly, several intestinal microbial pathogens secret proteases that degrade MUC2 to modulate mucus viscosity during pathogenesis,19 leading to pro-inflammatory contact between gut flora and immune cells.20
Recent rheologic studies show healthy human colonic mucus is composed of 1.3–1.9 wt% (13–19 mg/mL) mucin solids, yielding a dynamic viscosity of 150–250 mPa*s.21 Using this benchmark, our first mechanical characterization step was to assess the baseline viscosity for the mucins themselves when reconstituted in saline (Fig. 4a-c). Rheological dynamic time sweep measurements of reconstituted BSM showed a dynamic viscosity of 1–228 mPa*s as mucin concentration was increased from 3–50 mg/mL, resembling the rheological performance of native human colonic mucus.21 PGM was less viscous, reaching an average dynamic viscosity of 6 mPa*s at the highest tested concentration. We next tested the viscosity of native FAMS without mucin coating. Results in Fig. 4d demonstrate a prolonged stress relaxation response of the viscoelastic material over the first 40 seconds of loading, reaching a plateau dynamic viscosity of approximately 150 mPa*s. This behavior is similar to the long stress-relaxation response reported for porcine gastric mucus gels.22 Mucus viscoelasticity results from reversible, non-covalent interactions between mucin components, enabling solid-like responses over short loading periods and flow behavior on longer time scales. Interestingly, while FAMS was able to replicate the viscoelastic nature of native mucus, we did not observe this same flow behavior for reconstituted BSM and PGM (Figs. 4a,b). This is likely due to the compositional simplicity of these solutions, which do not replicate the varied biomolecular constituents and gradient structural morphology of native gut mucus. With these benchmarks established, we next tested the mechanical performance of FAMS coated with BSM (Fig. 4e) and PGM (Fig. 4f) mixtures. Rheological measurements demonstrate that, although the relaxation behavior of native FAMS is retained, the initial dynamic viscosity magnitude is greater when the gels are coated with mucins than without. Still, mucin-coated FAMS were found to maintain a ~ 100 mPa*s viscosity plateau regardless of the loaded mucin concentration (Fig. 4g).
Given the sensitivity of GI mucus to changes in environmental ionic strength and pH, we next tested these parameters on the viscosity of FAMS. Increasing the total salt concentration of the phosphate buffered saline environment from 0–750 mM led to a corresponding increase in FAMS viscosity (Fig. 4h, i). This is converse to native mucus, where increasing ion strength generally correlates to reduced viscosity due to polyelectrolyte charge-shielding.23 In our case, screening of the Fmoc-FF anionic charge by salt may alter its solubility and shift its kinetic equilibrium in favor of fibrillar assembly, thereby producing a more cohesive and viscous FAMS gel. Changing solution pH also led to variable FAMS viscoelastic responses (Fig. 4k). Here, average dynamic viscosity increased from 83 mPa*s at pH 5.5 to 231 mPa*s at pH 6.5, before declining as the solution became more basic. This bears resemblance to native mucus, which exhibits decreasing viscosity as environmental pH transitions from neutral to weakly alkaline.24 The ability of FAMS to mechanically respond to its environment is most likely regulated by the protonation state of Fmoc-FF’s carboxylic acid. This is supported by studies on various Fmoc-Phe derivatives demonstrating that changes in ionic strength and pH modulate Coulombic repulsion between anionic charged amino acids to alter their assembly propensities.16, 25 In sum, our results show that the viscoelastic behavior of FAMS can be modulated by environmental conditions to create mucus analogues with customizable rheologic properties that match native colonic mucus.
Development of Synthetic Microbiome Organoids
Microbial integration into FAMS was initially investigated using a stably expressing GFP-E. coli strain to aid visualization. Although E. coli was able to bind to the surface of uncoated FAMS, its attachment was relatively poor as indicated by low cellular fluorescence (Fig. 5a). Conversely, mucin coatings led to significant E. coli colonization on, and within, the fibrillar scaffold, forming dense microbial communities (Fig. 5b-d, Supporting Fig. 3). Scanning electron microscopy confirmed that the microbes don’t simply reside at the surface of the material, but integrate within the fibrillar mesh (Fig. 5e,f, Supporting Fig. 4).
Next, growth studies evaluated the proliferation of E. coli seeded onto the FAMS materials after a 24 hour incubation (Fig. 5g). These assays were performed in sterile PBS to minimize environmental nutrients, and thereby allow us to isolate the effects of FAMS coatings on E. coli growth trends. Results show the mucin-coated FAMS formulations (e.g., FAMSPGM, FAMSBSM) supported logarithmic growth of colonizing E. coli, with PGM coatings leading to more rapid microbial growth than BSM. Control FAMS prepared with BSA, which serves as a non-mucin protein control (FAMSBSA), and the naked material alone (FAMS) showed a significantly blunted growth profile. This suggests that E. coli can utilize the loaded mucins as a nutrient source to support robust colonization and growth within the coated FAMS materials.
While these results are encouraging, E. coli is considered a gastrointestinal pathobiont and so we next tested the canonical probiotic commensals L. acidophilus and L. rhamnosus (Fig. 5g). Visual microscopic inspection of inoculated FAMS gels showed these anaerobic strains more deeply integrated within the synthetic mucus bulk relative to E. coli, likely to minimize their exposure to oxygen in solution. Although the media used for these studies contains an L-cysteine reducing agent, the solution is not completely anoxic. As a result, we found that L. acidophilus and L. rhamnosus were difficult to extract from the entangled fibers during plating assays, leading to an apparent decline in the measured CFU/mL over 24 hours (Fig. 5g, lower left plots). However, both strains showed a parallel increase in the size of colonies formed on the plated media (Fig. 5g, lower right plots), suggesting cohesion of plated bacteria by FAMS fibers. This indicates that the reduction in plated CFU/mL for these strains may be due, in part, to association of the bacteria with the FAMS network, thereby inhibiting its transfer to the agar surface and creating larger seed colonies. In sum, our results suggest that gastrointestinal pathobionts and commensals can successfully integrate into FAMS and colonize the synthetic mucus network to create microbial communities.
Encouraged by these results, we set out to develop a simple, rapid, and low-cost fabrication protocol to generate multilayer microbiome organoids suitable for high throughput screening applications (Fig. 6a). We began by creating a colorectal epithelial layer using human Caco2 cells cultured for ≥ 19 days on transwell inserts. Appearance of tight junctions between cells in the monolayer, as indicated by Occludin staining (Supporting Fig. 5), confirmed the formation of an organized epithelial interface. We then added the FAMSPGM mucus analogue and inoculated with E. coli to create the final microbiome model. Orthographic microscopy images shown in Fig. 6b demonstrate the three-dimensional layering of each component in the synthetic microbiome. Like native colonic mucosae, FAMS was adhered to the surface of colorectal cells (Fig. 6c) and was permeated throughout the z-plane by colonizing microbes (Fig. 6d). Surprisingly, adherence of the Caco2 cell monolayer to FAMS was rapid, leading to sufficient transfer/migration of the cells with/into the fibrillar assembles to remove them from the transwell surface (Fig. 6e). Additional imaging studies showed this adhesion was significant after ≥ 1 hour of incubation, and that FAMS-adhered Caco2 cells remained viable and metabolically active (Fig. 6f,g). Although we cannot conclusively rule out some level of Caco2 cell death in these models, our data strongly supports the assertion that Caco2-FAMS-microbe mixtures form a tightly integrated system that reproduces many of the practical structural and morphologic features of native colonic mucus.
The relative ease with which these FAMS-enabled model colonic microbiomes can be constructed highlights their potential for screening applications. As an exemplary demonstration, we tested the permeation of low (4kDa) and high (70kDa) molecular weight dextran dyes through the FAMS generated model mucosae (Fig. 6i). Permeability assays showed that both the 4kDa and 70kDa markers were unable to diffuse across control, unmodified, Caco2 monolayers over the 24-hour incubation period (see grey circles in Fig. 6i). Conversely, Caco2 monolayers layered with FAMS, either uncoated (FAMS) or functionalized with PGM (FAMSPGM), showed an increase in cumulative dextran basolateral diffusion at the 2-hour incubation time point, which then generally increased monotonically with time. The notable exception was for the complete Caco2-FAMSPGM-E. coli mixture, which showed a decline in 4kDa basolateral fluorescence between the 12- and 24-hour measurement time points (see left plot in Fig. 6i). We ascribe this reduction to a metabolism of the dextran dye by E. coli at these longer time intervals. Several gastrointestinal bacteria have been reported to metabolize dextran as a nutrient source;26, 27 although we were unable to find a specific reference for the E. coli strain used in these assays (101-1). The 70kDa dextran did not show the same decline in fluorescence during the 12–24 hour interval, suggesting it’s higher molecular weight inhibited enzymatic processing.
In vivo pharmacokinetic studies report that 4kDa dextran permeates the gut and enters systemic circulation as early as 15 minutes after oral gavage, achieving maximum plasma concentration at 1–4 hours; dependent on mouse strain.28 The same study described slowed gastrointestinal transit kinetics for 70kDa dextrans compared to 4kDa markers, with serum bioavailability presumably similarly delayed. Taken in context with our data (Fig. 6i), FAMS-generated microbiome organoids appear significantly better at mimicking the in vivo pharmacokinetics of dextran than unfunctionalized Caco2 monolayers, which currently are considered the gold standard for ex vivo drug permeability assays.29 This is most likely due to mechano-chemical cross-talk between FAMS, E. Coli and Caco2 cells, leading to a more permissive mucosal interface that may better replicate the in vivo environment.
Outlook
Here, we exploit privileged fluorine-fluorine interactions to template the assembly of a fluorinated amino acid at liquid-liquid phase separated interfaces. The resultant supramolecular matter (fluorine-assisted mucus surrogate, FAMS) is a viscous gel that mimics the double layer architecture of native human colonic mucus. Rheological studies show FAMS resembles the viscoelastic properties of colonic mucus, and can be tuned in its mechanical performance via controlling environmental ionic strength and pH. Added mucin proteins rapidly adhere to the surface of the fibrillar network to simulate the proteinaceous profile of mucus without disrupting the viscoelastic properties of FAMS. Together, this generates a colonic mucus surrogate that can be inoculated with gastrointestinal pathobionts or commensals, and added to human colorectal epithelium, to generate a multicellular synthetic microbiome. We envision these materials will provide a simple, robust, and tractable tool to study commensal biology and microbe-mucus-host interactions ex vivo. The addition of leukocytes may expand this platform to advance our understanding of microbe-immune cell interactions in the gut. Since FAMS utilizes inexpensive and commercially available building blocks, are generated using simple protocols that do not require specialized equipment, and can be integrated with established epithelial models, these materials may have utility in high throughput screening campaigns. As an exemplary application, we show that FAMS can be readily incorporated into standard Caco2-based permeability assays to enhance the prediction of in vivo drug adsorption. Because these FAMS-enabled models can be developed without significant cost or training they represent a complementary alternative to microfluidic and stem cell-based organoid approaches that are currently limited in their broad utility due to complexity and cost.