Oral mucositis on a chip: modeling induction by chemo- and radiation treatments and recovery

Abstract

Oral mucositis (OM) is a debilitating complication affecting roughly 70% of head and neck cancer patients receiving chemotherapy and/or radiation treatment. No broadly effective preventative treatment for OM exists. Therefore, an in vitro model of cancer treatment-induced OM would aid studies into possible origins of the pathology and future drug targets to ameliorate it. In this study, we present a microfluidic oral mucosa triculture tissue construct consisting of a keratinocyte layer attached to a subepithelial fibroblast and endothelial cell-embedded collagen gel. To address the typically low stability of mucosal constructs in microfluidics, ruthenium-catalyzed photocrosslinking was implemented to strengthen the collagen gel and prevent the invasion of kerotinacytes, thus maintaining tissue construct geometry and oral mucosa barrier function for over 18 days of culture. Next, the OM chip was exposed to cisplatin (day 10) and damaging radiation (day 11, +/−cisplatin at day 10), mimicking damage from cancer therapy. Damage to and then recovery of the tissue layers and function were observed over days 11–18. Therefore, several important features of oral mucositis induction and resolution were modeled in microfluidic culture. The oral mucositis model on a chip allows for more sophisticated studies into mechanisms of OM and potential treatments.

Graphical Abstract

Oral mucositis is a serious side effect of cancer therapy with limited treatment options. A triculture of human gingival keratinocytes, fibroblasts, and microdermal vascular endothelial cells in a microfluidic chip represents a platform to test the induction and resolution of mucositis. Cisplatin and radiation treaments cause damage to the mucosa constructs, followed by recovery.

1. Introduction

Each year in the United States, approximately 60,000 patients are administered in outpatient oncology clinics with diagnosed head and neck cancers.[1, 2] Roughly two-thirds of these cancer patients develop oral mucositis (OM) as a significant off-target consequence of chemotherapy and radiation treatment. Common symptoms associated with OM are a burning sensation and confluent ulceration of the oral mucosa, often causing extreme pain, compromised nutrition and oral hygiene, and elevated risk for local and systemic infection.[3] Furthermore, OM frequently leads to longer hospital stays, insertion of feeding tubes, high rates of narcotic painkiller prescriptions, and even cessation of chemotherapy.[4–6] Risk factors associated with OM can be classified into two main groups: patient-related and treatment-related risks.[7] In the former group, age, female gender, nutritional status, smoking history, and pre-existing oral conditions are common critical factors contributing to the incidence and severity of mucositis.[8–10] In the latter group, mucositis incidence varies depending on the type of cancer therapy the patient received. Particularly, mucositis incidence occurs in roughly 80% patients that receive high-dose chemotherapy before bone marrow transplantation, 30–60% in head and neck cancer patients receiving radiation and even greater if chemotherapy is also given.[11] Palifermin is the only restorative therapy for oral mucositis approved by the USA Food and Drug Administration (FDA). However, palifermin, which is recombinant human keratinocyte growth factor, is only recommended for mucositis patients undergoing hematopoietic stem cell transplantation.[12, 13] New anti-mucositis treatments are being developed and tested. Therefore, effective techniques for screening anti-mucositis therapeutic strategies are in great need to evaluate potential therapeutic drug candidates.

Appropriate in vitro and in vivo models are needed to advance knowledge of OM and treatment mechanisms. There are three main conventional OM modeling platforms: (i) in vitro two-dimensional (2D) cell culture, (ii) engineered oral mucosa equivalents, and (iii) in vivo animal models. Monoculture of oral keratinocytes on well-plates was used to test the effects of putative treatments for chemotherapy-induced OM.[14] Researchers constructed in vitro three-dimensional (3D) oral mucosa equivalents based on acellular human dermis to examine the effects of radiation-induced OM on oral cell motile behavior, metabolism, and other relevant factors.[15, 16] Finally, many animal-based OM models were developed to investigate pathogenesis and toxicity of chemo- or radiation therapy induced mucositis as well as to explore therapeutic effects of new anti-mucositis preventive treatments.[17–20] In spite of demonstrated utility, existing OM models possess certain drawbacks. First, 2D cell culture cannot mimic the complex geometry, biochemistry and functionality of native tissues. [21] Further, 3D mucosal equivalents are restricted in size and complexity by a lack of vascularization and perfusion into the center of the constructs. It is challenging to repeatedly assess 3D tissue constructs during culture, making evaluation of tissue damage, remodeling and repair associated with OM difficult.[22] Lastly, animal models are often expensive and time-consuming, while results may not reflect the behavior of human cells.[23] An alternative platform that allows straightforward and longitudinal assessments of human mucosal constructs with the tissue geometry and functionality of native oral mucosal tissue would address these challenges.

Tissue chips model the biology, physiology, and functionality of native tissues in ways difficult to achieve with conventional cell culture models.[24] In particular, the interactions of cells in tissues-on-a-chip with their physiological environment can be simulated by regulating key microfluidic parameters in static and dynamic culture.[25, 26] Further, the entire tissue construct is visible under a microscope, with tissue geometry and cell morphology able to be tracked over many days without compromising aseptic culture. These features of microfluidic cultures enable higher complexity, controllability and reproducibility as compared to current 2D in vitro models, and easier evaluation than with 3D in vitro models. Numerous studies have employed tissue chips for disease modeling and drug screening applications. For instance, a liver-on-a-chip model was developed for drug toxicity testing.[27] Several tumor-on-a-chip models were built and improved for understanding tumor growth mechanisms, testing anticancer drugs, and developing cancer therapies.[28] In terms of dental research, a tooth-on-a-chip was developed to enable direct observation of pulp cells’ responses to dental materials.[29] A high-throughput dental biofilm model was proposed to investigate biochemical conditions associating with biofilm growth,[30] and an oralmucosa-on-a-chip was developed to evaluate the sensitivity and layer-specific interactions of oral cells with bacteria and dental materials.[31, 32] Nevertheless, no microfluidic-based cancer therapies-induced mucositis models are yet available. Costly drug development for mucositis prevention and treatment could be made cheaper and more efficient by testing drug candidates for mucositis on a microfluidic mucositis model, as part of early in vitro drug evaluation. A relevant preclinical model of cancer therapies-induced OM that aids in improving drug screening success and lowering the costs of expensive animal models is of significant interest.

Initial symptoms of OM usually appear a few days after beginning cancer therapy, progressing to ulceration by 7–14 days.[3] Therefore, an OM model should sustain culture through at least three weeks. Nevertheless, long-term cultivation of mammalian cells in microfluidics remains a major challenge due to construct remodeling and breakdown.[33] Remodeling and mechanical failure of the tissue construct are two major culprits that shorten the lifespan of tissue chips.[34, 35] In previous microfluidic mucosal models, keratinocyte invasion and fibroblast contraction of the collagen hydrogel destroy the layered organization of the construct after 7 days in culture. The epithelial barrier is no longer functional in the chip after these remodeling events.[31, 32] One strategy to overcome tissue remodeling is to enhance the mechanical strength of the tissue construct through crosslinking, but toxicity of the crosslinking regimen has been of major concern. Some chemical crosslinkers such as glutaraldehyde are toxic during and after application, so not suitable for tissues on chips. Alternatively, photocrosslinking of the collagen network in tissue chips with living cells is possible, but long exposure time and/or ultraviolet light may also induce toxicity. On the other hand, Ruthenium-based photocrosslinking is rapid, controllable, operates at longer wavelengths, and should limit toxicity.[36, 37]

The aims of this study are (1) to develop a microfluidic-based oral mucosa triculture tissue platform with enhanced mechanical strength and high viability for long-term culture, and (2) to model cancer therapies-induced OM and recovery in the developed triculture platform (Figure 1). This triculture model consisted of human gingival keratinocytes (GIE), human gingival fibroblasts (HGF), and human dermal microvascular endothelial cells (HMEC), closely resembling the configuration of native oral mucosa tissue, depicted in Figure 1A. The triculture was built in a single layer polydimethylsiloxane (PDMS) microfluidic device that consisted of an apical layer of GIE attached to the basement membrane in a side channel that was bound to a collagen gel with mixed HGF and HMEC cells embedded. The collagen layer resides in the central channel in fluidic contact with two parallel side channels through interconnecting pores (Figure 1(B–C)). The structural integrity of this model was enhanced using a ruthenium-catalyzed photocrosslinking method,[36, 37] which allowed for long-term cultivation of at least 21 days. Importantly, keratinocyte invasion and collagen gel contraction were minimized, thus maintaining the native layer-mimicking configuration of the mucosa-ona-chip. Next, OM experiments were conducted with the focus on damage and recovery of the triculture exposed to cancer therapies (Figure 1(D–E)). The effects of cancer therapies-induced OM on the tissue chips were evaluated by assessing impairment of epithelial barrier function, cellular cytotoxicity, cellular morphology, and network damage. We demonstrated a direct observation of epithelial ulceration due to cancer therapies induced OM within a one-to-three-day window, followed by tissue recovery 7 days after treatment(s). These findings suggest that this triculture model is a promising platform for more in-depth studies of mucositis and potential anti-mucositis approaches.

Figure 1.

Design of the oral mucosa triculture tissue construct and summary of results. (A) A schematic illustration of an oral mucosa in cross-section. (B) A three-channel microfluidic device under phase contrast microscopy showing the culture area for oral mucosa triculture construct. (C) Illustration of the oral mucosa triculture tissue construct from the blue dashed box in (A-B). (D) A representative micrograph of the construct with intact mucosal barrier function evidenced by the inhibition of F-dextran penetration before being exposed to cancer therapies (chemo- and/or radiation therapy) at Day 10 and/or Day 11. (E) Post-chemo- and radiation treatment, the mucosal barrier function was impaired, causing clear F-dextran penetration into the tissue construct. The white arrows showed the clusters of dead GIEs while the red arrows pointed out the damaged barrier layer. (F) The barrier function was regained after several days with the green arrows indicating the regrowing barrier layer.

2. Method

2.1. Microfluidic platform fabrication

Each polydimethylsiloxane (PDMS) chip consisted of eight sets of three-channel devices for the fabrication of up to eight oral mucosa triculture tissue constructs per chip. The dimension of each device is 1300 μm W x 20 mm L x 50 μm H. The tissue triculture construct was fabricated in the central area of each device where three channels were separated by twelve 100 μm x 50 μm PDMS pillars and connected through fourteen 50 μm x 50 μm square apertures (Figure 1(B)). The mold for the microfluidic platform was manufactured on a 4″ silicon wafer using conventional photolithography technique with negative photoresist SU-8 3035. The PDMS chip was constructed using soft-lithography by mixing Sylgard 184 and its curing agent (Ellsworth Adhesives, NY, USA) at a ratio of 10:1, degassed, poured on top of the patterned silicon mold, and cured at 65°C for 4 h.[38, 39] Next, the cured PDMS was peeled off from the mold and cut into a set of eight devices per chip. The end of each microchannel was punched for inputs and outputs. The cured PDMS chip was then bound to a cleaned glass slide using a Plasma Cleaner PDC-32G (Harrick Plasma, NY, USA). The bonded chips were then heated at 120°C for 24 h to restore the hydrophobicity of PDMS. The PDMS chip was sterilized with UV light for at least 30 min before triculture tissue assembly.

2.2. Cell culture and oral mucosa triculture tissue model fabrication

Three immortalized human cell lines used in this study were human gingival keratinocytes (GIE), human gingival fibroblast (HGF), and human dermal microvascular endothelial cells (HMEC). The HGF [40, 41] and GIE [42] cell lines were purchased from Applied Biological Materials, Inc. (abm, BC, Canada) while HMEC was obtained from ATCC (MD, USA). The media used to feed triculture construct, named triculture media, comprised of Prigrow III (abm)/Prigrow IV (abm)/MCDB 131 (Gibco, MD, USA) at a 1:1:1 ratio and supplemented with 1 μg/mL hydrocortisone (Sigma-Aldrich, MO, USA), 10 mM L-glutamine (ATCC, MD, USA), 10 ng/mL epidermal growth factor (ThermoFisher Scientific, MA, USA), and 20% fetal bovine serum (ATCC, MD, USA). To enhance the strength of the oral mucosa triculture tissue construct, a ruthenium-catalyzed photocrosslinking method was employed.[36] A Ruthenium visible light photocrosslinking kit, consisting of ruthenium (Ru) and sodium persulfate (SPS) photo-initiator powders, was purchased from Advanced Biomatrix (CA, USA). Stock solutions of Ru and SPS were prepared separately in 1X phosphate saline buffer (PBS, Sigma-Aldrich, MO, USA) according to the company’s guidelines. A working solution of Ru/SPS was obtained by diluting stock solutions in PBS to yield a final concentration of 0.37 mg/mL Ru and 1.19 mg/mL SPS prepared fresh for each experiment.

Figure 2 shows the general assembly process of the oral mucosa triculture construct and the overall experimental timeline of the current study. On the first day of the assembly process, assigned as Day (−1), two HGF and HMEC cell lines were harvested from culture flasks by trypsinization. Next, HGFs and HMECs were mixed with 7 mg/mL collagen type I (rat tail, 10.45 mg/mL, Corning, MA, USA) at the final density of 5,000 HGFs/μL and 5,000 HMECs/μL. The mixture was gently injected into the central channel of each device using a micropipette. The chip was then incubated at 37°C for 30 min allowing collagen to polymerize. The two side channels were then filled with triculture media and cultured overnight at 37°C, 5% CO₂. On the next day, Day 0, the HGF/HMEC-embedded collagen gel was incubated with a working solution of Ru/SPS for 1 min, before being exposed to blue light (Wolezek, UK, intensity of 13.1 mW/cm² at λ=400 nm) from a 50 mm distance (Figure 2(B)(ii)). Various photocrosslinking durations, including 10, 20, 30, and 45s, were tested. To further mimic the physiology of native oral mucosa tissue where the epithelial layer adheres to a basement membrane (made of collagen IV, laminins, and other proteoglycans), collagen IV was deposited to the apical layer of the collagen gel by coating the left-side channel at 10 μg/mL (R&D Systems, MN, USA) and 37°C for 1 h. The side channel was then rinsed with media to remove residual collagen IV, leaving a thin layer of collagen IV at the interface of the HGF/HMEC-collagen gel and the side channel. Figure 2(B)(iii) shows the fabricated collagen IV layer using fluorescein-labeled collagen IV (Flabeled collagen IV, AnaSpec, CA, USA). The collagen IV layer remained visible during triculture (Figure 2(B)(iv)). Following this, GIEs were harvested from the culture flask, adjusted to a density of 2,000 cells/μL, and injected gently into the left-side channel. The chip was then vertically tilted to allow GIE attachment to the collagen matrix through the interconnecting pores. The device was further incubated at 37°C for 45 min, before the unbound GIEs were rinsed away. The chips were fed daily with triculture media using the tension-driven flow method [31, 32] and were cultured for up to 18–21 days. To maintain the humidity and prevent the drying of media in the devices, each chip was placed in a petri dish filled with sterile water droplets. Figure 2C shows that after the initial assembly period, the chips were cultivated for 10 more days before being exposed to cancer therapies at Day 10 and/or Day 11. The OM damage peaked around Day 12/13, followed by a recovery period until Day 18. Day 18 was chosen as the end of the OM modeling experiment as the keratinocyte layer was typically confluent again in the side channel by that time.

Figure 2.

General assembly and experimental processes of the oral mucosa triculture on a chip. (A) An illustration of ruthenium-catalyzed photocrosslinking of the collagen gel using blue light (λ=400–450 nm). When being exposed to blue light, Ru(II) and persulfate turned into Ru(III) and sulfate radical. Ru(III) then oxidized tyrosine into tyrosyl radicals, which generated bonds with nearby tyrosine proteins to form di-tyrosine crosslinks. (B) Sequence images of the general assembly process of the oral mucosa triculture tissue on a chip: (i) the subepithelial HGF/HMEC-collagen layer after 24 h of coculture (Day 0); (ii) the PDMS chip containing HGF/HMEC-collagen construct perfused with Ru/SPS mixture before being exposed to blue light; (iii) a basement membrane protein coating (collagen IV) followed by the loading of GIEs onto the apical channel using gravity-assisted settling of GIEs on top of subepithelial layer. (C) Experimental timeline showed the triculture construct fabrication process, followed by a maturation period. The matured oral mucosa triculture constructs were then exposed to cancer therapies at Day 10 and 11 and maintained in culture until Day 18 to demonstrate the periods of damage and recovery of oral mucosa functional behavior. The dashed blue, red, and green boxes highlight Day 10 (initiation Day), 12 (peak or near-peak damage day), and 18 (recovery day).

2.3. Characterization of triculture model

2.3.1. Cytotoxicity assessment

The cytotoxicity of photocrosslinking was investigated using the LIVE/DEAD^® Viability/Cytotoxicity Kit, ThermoFisher Scientific (MA, USA), to stain HGF and HMEC cells. Five groups were examined: a no photocrosslinking control (noXL), and photocrosslinking for 10s, 20s, 30s, and 45s (XL10s, XL20s, XL30s, and XL45s, respectively). The staining procedure was conducted according to the manufacturer’s guidelines. In brief, the side channels of each device were rinsed thoroughly with serum-free media. A staining mixture containing 8 μM ethidium homodimer-1 and 2.67 μM calcein acetoxymethyl ester in PBS was then introduced to both side channels. The chips were incubated in the dark at room temperature (RT) for 1 h before imaging. Dead cells were counted based on cell morphology.[32] The live signal of HGFs/HMECs under different photocrosslinking conditions was measured as follows.[31] The total area of the central channel was selected as a region-of-interest (ROI). The area fraction of live signal within each ROI was quantified with ImageJ (MD, USA). The live area was obtained by multiplying the total area with the live signal area fraction. The live area of the noXL group (control group) was set as 100% and the live areas of other groups were normalized to that control group. Experiments were performed with n=7–8 per group and repeated in two independent trials.

2.3.2. Collagen gel area

The contraction of HGF/HMEC-embedded collagen gels among noXL and XL groups was monitored over a 10-day co-culture period. The total collagen gel area within the central channel at Day 0 was set as 100% for normalization among all groups. Normalized collagen area was assessed from one trial with n=3–4 replicates per group.

2.3.3. Keratinocytes invasion and growth assessment

The invasion of GIE among noXL and XL groups was assessed during 21-day monoculture. Collagen gels at 7 mg/mL without embedded cells were assembled into the central channel of each device at Day (−1), followed by photocrosslinking, basement membrane coating and GIE culture at Day 0 as reported above. The invasion percentage of GIE was calculated by dividing the area invaded by GIEs within the central channel at a specific time point with the total ROI’s area of the central channel as depicted in Figure 4(A)(i). Experiments were performed with n=4–7 constructs/group and repeated in two independent trials. Meanwhile, the growth characteristics of GIE cells were also monitored through 10-day triculture. Two groups including noXL and XL20s were selected for this experiment. The growth of GIE cells was defined as the coverage percentage of GIE within the apical lumen channel and was calculated by dividing the area occupied by GIE within the side channel at specific time points with the total ROI area of the side channel at Day 0 as shown in Figure 5(B)(i). Experiments were performed with n=4–7 constructs/group and repeated in two independent trials.

Figure 4.

Photocrosslinking minimized GIEs invasion. (A) Representative micrographs of GIE monoculture showing invasion progression of GIEs into the uncrosslinked (noXL) and 20-second crosslinked (XL20s) collagen gels at Day 0, 1, 4, 10, and 21. The dashed red box in A(i) indicated the total area of the central channel for invasion percentage measurement. The orange arrows indicated the progressive invasion of GIEs into collagen gel while the orange dashed area highlighted the invasion area of GIEs at specific time points. The invasion percentages of GIEs were calculated by dividing the invasion area (orange dashed area) with the central channel’s total area (red dashed box) as shown in (A). (B) The GIE invasion percentages (i) among noXL and XL20s gels and (ii) among XL10s, XL20s, XL30s, and XL45s gels (mean ± SEM). Results of pairwise comparisons are indicated by * (p<0.05), ** (p<0.01), and † (p<0.001).

Figure 5.

The structural integrity and functionality of triculture was enhanced by photocrosslinking. Representative micrographs of oral mucosa triculture tissue construct (A) without and (B) with photocrosslinking at Day 0, 4, and 10. The dashed white box in B-(i) indicated the total area of the apical lumen channel for GIE coverage measurement, while the green dashed areas in (A-B)-(ii) highlighted the coverage areas of GIEs at Day 4. The coverage percentage was quantified by dividing the coverage areas by the total area of apical lumen channel. The orange arrows indicated the progressive invasion of GIEs, and the orange dashed area highlighted the invasion area of GIEs into HGF/HMEC-collagen gel in the central channel. (C) The coverage percentage of GIEs over a 10-day period within the noXL and XL20s triculture groups (mean ± SEM). (D) The F-dextran permeability within the noXL and XL20s groups to test the mucosal barrier function of oral mucosa triculture construct over the first 10 days of experimental period (mean ± SD). The F-dextran permeability was determined by averaging the percentages of F-dextran penetrated through the seven interconnecting pores as indicated in cyan boxes in A-(iii). Results of pairwise comparisons are indicated by * (p<0.05), ** (p<0.01), and † (p<0.001).

2.3.4. Mucosal barrier function

Rhodamine B isothiocyanate–dextran (F-dextran, molecular weight of 70 kDa, Millipore Sigma, MA, USA) was used to assess the mucosal barrier function of oral mucosa triculture constructs. A stock solution of 10 mg/mL F-dextran was prepared in PBS, filter sterilized, and diluted in triculture media to yield the final concentration of 1 mg/mL. The apical lumen channel was filled with F-dextran/media solution and the chips were incubated for at least 30 min before imaging. The F-dextran permeability was determined by averaging the percentages of F-dextran signal on the collagen gel side of the seven 50 μm x 50 μm square transchannel pores as indicated in Figure 5(A)(iii), taken as a percentage of the maximal signal in the image, averaged from a region of interest always in the upper portion of the apical lumen. The F-dextran permeability experiments were conducted in one trial with n=4–6 constructs/group among noXL and XL groups (only XL20s was chosen for this experiment) to determine the effects of photocrosslinking on triculture functionality.

2.3.5. Epifluorescence and immunofluorescence staining

To visualize filamentous actin (F-actin), the chips were stained with ActinGreen^™ 488 Ready Probes (ActinGreen, Invitrogen, MA, USA) for F-actin and propidium iodide (PI, Invitrogen, MA, USA) for nuclei. The microfluidic chips were fixed in 3.7% paraformaldehyde for 20 min, then permeabilized in 0.1% Triton-X/PBS for 30 min. A staining mixture containing ActinGreen (2 drops per 1 mL) and PI (1:3,000 dilution) in 0.1% Triton-X/PBS was introduced to both side channels. The chips were incubated in the dark at RT for 1 h before rinsing with PBS and imaging.

To differentiate GIEs, HGFs, and HMECs within triculture constructs, the cells were labelled with specific markers. All primary antibodies, secondary antibodies, and normal goat serum (NGS) were supplied by Abcam (Cambridge, UK) or Invitrogen (MA, USA). Bovine serum albumin (BSA) was purchased from ThermoFisher Scientific (MA, USA). GIEs were labelled by rabbit anti-CK19 primary antibody-conjugated to a green-fluorescence probe AlexaFluor^® 488 (ab192643, Abcam), HGFs were labelled by rabbit anti-S100AA primary antibody-conjugated to a redfluorescence probe AlexaFluor^® 647 (ab196168, Abcam), and HMECs were labelled by mouse anti-CD31 (14–0319-82, Invitrogen) primary antibody. Mouse anti-CD31 was then labelled by goat anti-mouse IgG H&L conjugated to a blue fluorescence probe AlexaFluor^® 405 (ab175660, Abcam). The immunostaining procedure was adjusted according to the manufacturer’s recommendations. In general, oral mucosa triculture constructs were fixed in 3.7% paraformaldehyde for 20 mins at RT, then rinsed with PBS 1X twice. The constructs were permeabilized with 0.1% Triton-X/PBS solution at RT for 30 min before blocking with 10% NGS/1% BSA/0.1% Triton-X/PBS at RT for 1 h. Primary antibodies were prepared in 1% BSA/0.1% Triton-X/PBS at 1:25 dilution for anti-CK19, 1:25 dilution for anti-S100A4, and 1:25 dilution for anti-CD31. The chips were rinsed with PBS twice before being incubated with primary antibodies at 4°C for 72 h. All channels were rinsed twice with PBS to remove excessive primary antibodies. Next, the chips were incubated with a secondary antibody of 1:50 dilution in 1% BSA/0.1% Triton-X/PBS at 4 °C for 72 h. The chips were finally rinsed with PBS twice before imaging.

2.4. Oral mucosa maturation and mucositis model

The maturity of oral mucosa triculture constructs was determined by the growth and barrier function of the GIE layer. As the GIE layer reached to more than 80% coverage at Day 8–10, the barrier function of oral mucosa triculture was established (see in the later section). Day 10 was set as the initiation point for mucositis modeling experiments, when the oral mucosa triculture constructs were exposed to chemotherapy and/or radiation therapy. Cisplatin was purchased from Millipore Sigma (MA, USA) while a box fitted with UV LEDs, used as a radiation source in this study, was obtained from Niels Ryberg LLC (FL, USA).

2.4.1. Cancer therapies treatment simulations on the chip

Cisplatin treatment is associated with oral mucositis. Therefore, a stock solution of 0.5 mg/mL in PBS was prepared, filter-sterilized, and diluted in triculture media to yield a final concentration of 35.7 μM, which had been reported to induce OM.[43–46] For radiation and chemoradiation treatments, UV exposure was chosen to model radiation-treatment effects. Ultraviolet light that emits the wavelength of 260 nm, the maximum absorbance of nucleic acids (DNA or RNA),[47] is an ionizing source safe to use in laboratory settings, with easy to adjust radiation intensity and exposure time. The induction of oral mucositis was carried out according to the timeline depicted in Figure 2C. Briefly, at Day 10, a cisplatin/media solution was administered to the basal lumen (the right-side channel) of the oral mucosa triculture construct. The chip was then cultured for a subsequent 24 h. At Day 11, nutrient media was changed, and side channels were rinsed thoroughly to remove residual cisplatin. For radiation therapy, at Day 11, the chip was placed in the UV box with the intensity of 40–45 mW/cm² at λ=260 nm and exposed to UV light for 30s. Chemoradiation treatment was conducted by combining chemo- (17.9 μM) and radiation (30s) exposures sequentially. At Day 10, the chip was treated with chemotherapy for 24 h, followed by intense rinsing and exposure to UV light. Afterwards, the chip was maintained in culture and monitored daily until Day 18.

2.4.2. Cell damage and viability assessment

Lactate dehydrogenase (LDH) is a stable cytoplasmic enzyme released upon plasma membrane damage.[48] Here, the damage to cells of cancer therapies applied to the tissue chips was evaluated by measuring the LDH activity in cell culture media after different cancer treatments and the notreatment control at Day 10, 12 and 18. The LDH assay was conducted using a CytoTox 96^® non-radioactive cytotoxicity kit (Promega, WI, USA) followed the manufacturer’s guidelines. Conditioned media for the assay was collected in two independent trials with four replicates per condition per trial. Cell culture media of 12.5 μL from each sample was transferred to a 384-well plate and mixed 1:1 with reagent for the assay. The optical density was then recorded at 490 nm using a Tecan Infinite F Plex microplate reader (Tecan Systems Inc., CA, USA). The pre-treatment LDH level at Day 10 was set to 1, and all subsequent culture timepoints were normalized to the pre-treatment LDH level. The effect of the treatments on cell viability in the chips was assessed with the Live/Dead assay at Day 10, 12 and 18 following the steps described above. The dead GIE and HGF/HMEC cells were counted from epifluorescence images confirmed by cell morphology from co-localized phase contrast image. The total area of the apical side channel, outlined in Figure 5B(i), was chosen to be the ROI for live area measurement of GIE cells. Meanwhile, three randomly selected ROIs of 250 × 250 μm² were selected within the central channel for live area measurement of HGF/HMEC. The area fraction of live signal within each selected ROI was obtained as described above. Since the total ROI area of GIE was 6.15 times larger than the total ROI area of HGF/HMECs, the live area of HGF/HMECs was multiplied by 6.15 to have an equivalent comparison between the live area of GIE versus HGF/HMECs under the same treatment, as indicated in the graphs below. Experiments were performed with n=5–6 per group and repeated in two independent trials.

2.4.3. Mucosal barrier function impairment and keratinocytes layer damage

The disruption in mucosal barrier function post-exposure to cancer therapies was evaluated by the changes in F-dextran permeability of oral mucosa triculture constructs. Meanwhile, the damage to the GIE layer caused by cancer therapies was characterized by the coverage percentage of GIE within the left-side channel as reported above. The F-dextran permeability and GIE coverage percentage were examined daily in four groups, untreated, chemo treated, radiation treated, and chemoradiation treated, from Day 10 to Day 18. Experiments were performed on n=6–8 and repeated in two independent trials.

2.5. Microscopy and image analysis

Phase contrast, epifluorescence, and immunofluorescence signals of the oral mucosa triculture constructs were captured using an inverted microscope (Accu-Scope EXI-310, Ludesco LLC, MD, USA) equipped with a fluorescence light source (LM-75 PhotoFluor, 89 North, VT, USA). Appropriate filters were used to image specific fluorescence-labeled antibodies and markers. The TexasRed filter (emission and excitation filters of 604–644 and 542–582 nm, respectively) was used for F-dextran permeability experiments. Image processing and data analysis were performed using ImageJ with a Fiji image processing package (NIH, USA). Multiple fluorescence tiles of each sample were stitched using Pairwise stitching or the Grid/Collection stitching package in Fiji.[49]

2.6. Statistical analysis

Experiments were performed on n=4–8 replicates and repeated in two independent trials unless specified elsewhere. The means and standard error of the means (SEM) or standard deviation (SD) for each set of data were calculated. The data was displayed as mean ± SEM (used for three or more trials) or mean ± SD, specified in figure captions. After checking ANOVA assumptions (normality, independence, homoscedasticity), the effects of treatments and time on the various experimental endpoints were evaluated using two-factor ANOVAs. Then, if any factor (treatment or time) was significant, one-factor ANOVAs were performed at each level of time (i.e., day of culture). Finally, Tukey’s post hoc test was used to compare between groups at each timepoint, or Dunnett’s test was used to compare treatment groups to control or later timepoints to a pre-treatment timepoint. Software used was SPSS (version 27.0, SPSS Inc., IL, USA). The level of significance was set at 0.05.

3. Results

3.1. Effects of photocrosslinking on cell viability and collagen gel contraction

Photocrosslinking of the collagen hydrogel caused lower viability, more dead cells, and less collagen gel contraction over 10 days culture, dependent on crosslinking time (Figure 3). Without crosslinking, most cells were viable, with many dead cells apparent at exposure times of 30 seconds and longer (Figure 3(A)) Cell viability was significantly affected by photocrosslinking time (ANOVA, F=9.0, p<0.001), with 30- and 45-seconds exposure leading to 60±12% and 34±6% viability, respectively, significantly lower than no crosslinking (Figure 3(B)). Dead cell counts were affected by photocrosslinking time (ANOVA, F=5.7, p<0.001), with more dead cells resulting from 45 seconds light exposure than the uncrosslinked controls, 49±32 compared to 5.8±1.5 dead cells, respectively (Figure 3(C)). While the uncrosslinked collagen gel contracted significantly (ANOVA, F=34.8, p<0.001), at a significant magnitude beginning by day 3 and by 45% at day 10 of culture, collagen constructs crosslinked for 10–45 seconds had no significant contraction over 10 days (Figure 3(D)).

Figure 3.

The effects of photocrosslinking duration on HGF/HMEC-embedded collagen gel coculture. (A) Representative live/dead stained epifluorescence images of HGF/HMEC coculture post photocrosslinking for different durations such as 10, 20, 30, and 45s (named XL10s, XL20s, XL30s, and XL45s, respectively) compared to the control group (noXL). (B-C) The normalized viability and dead cell counts within each condition (mean ± SEM). (D) The normalized collagen gel area of each condition over a 10-day period (mean ± SD). Results of pairwise comparisons are indicated by * (p<0.05), ** (p<0.01), and † (p<0.001).

3.2. Photocrosslinking prevention of keratinocyte invasion

To determine effects of photocrosslinking on keratinocyte invasion in the chip, GIE monoculture was performed in a side channel adjacent to a collagen gel in the central channel. Invasion occurred as early as Day 1 for the uncrosslinked gel but began at Day 10 for the constructs photocrosslinked for 20 seconds (Figure 4(A)). There were significant effects of photocrosslinking duration and culture time on invasion (Figure 4(B), ANOVA, F=31.9, p<0.001). After 21 days of culture, on average, 91±7% of the uncrosslinked construct was invaded by GIE cells at Day 21 compared to 12±2% in the constructs crosslinked for 20 seconds (Figure 4(B)(i)), and 5±1% of the constructs crosslinked for 45 seconds (Figure 4(B)(ii)).

Photocrosslinking of the collagen gel before introduction and attachment of keratinocytes within the microfluidic construct might affect GIE cell adhesion and proliferation. To test this, the initial adhesion of GIE cells to collagen gels with varied photocrosslinking durations (noXL, XL10s, XL20s, XL30s, and XL45s) was investigated by comparing the pore fill percentage in these constructs. In microscope images, there were somewhat fewer GIE cells attached to some of the XL45s constructs compared to other groups, reflective of more variation between constructs in that group (Figure S1(A)). Further, roughly 80% of pores were filled among five tested groups, with no statistically significant difference between groups (Figure S1(B)). Any differences in initial adhesion of GIE cells to the collagen gels did not significantly affect proliferation. Adherent keratinocytes in all groups proliferated until they became confluent at around Day 10 for each condition tested (Figure S1(C)).

3.3. Photocrosslinking enhanced triculture structural integrity and functionality

Next, we investigated the impact of photocrosslinking on maintenance of cell layers and epithelial barrier function of the triculture constructs. Photocrosslinking duration of 20 seconds was selected as this was the minimum time that supported low collagen gel invasion/contraction and high fibroblast/endothelial cell viability. Without crosslinking, keratinocytes grew into the collagen gel by day 4, and barrier function was non-existent by Day 10 (Figure 5(A)). In comparison, 20 seconds of photocrosslinking limited invasion and largely prevented F-dextran transport across the keratinocyte layer by Day 10 (Figure 5(B)). By Day 10, keratinocytes reached >80% confluence in the apical channel (Figure 5(C)). There were significant effects of photocrosslinking time and culture duration on F-dextran permeability (ANOVA, F=85.0, p<0.001). A plateau was reached at Days 8–10, corresponding to F-dextran permeability of 90% in the uncrosslinked constructs and 10% in the photocrosslinked constructs (Figure 5(D)), demonstrating stable barrier function and a functionally mature tissue construct by this metric after 10 days of culture.

It is important to note that photocrosslinking not only supported the functionality but also maintained the layer-specific configuration of the proposed triculture. Figure S2 shows a time-lapse comparison of the triculture construct morphologies with and without photocrosslinking of the subepithelial layer. At Day 2, keratinocytes started to invade into the subepithelial layer of the noXL construct, which disrupted the tissue layers by Day 6 (Figure S2(A)(ii–iv)). Meanwhile, the layer morphology of the XL construct (XL20s) was strongly improved, leaving an intact epithelial-subepithelial interface (Figure S2(B)(ii–iv)). The layered organization of cells in the XL construct resembled the native oral mucosa in cross-section, with polygonal keratinocytes adjacent to the subepithelial layer containing fibroblasts and endothelial cells.

Cell appearance during construct fabrication was assessed by phase contrast microscopy and biomarkers during culture by immunostaining (Figure 6). At Day (−1), the subepithelial layer of HGF/HMEC in collagen gel was assembled into the central channel (Figure 6(A)(i)). The next day, the coculture construct was photocrosslinked for 20s, with no significant difference in morphology observed (Figure 6(A)(ii–iii)). Afterwards, keratinocytes were loaded into the apical side channel and adhered well onto the subepithelial layer, completing the triculture construct assembly (Figure 6(A)(iv)). The epithelial layer grew thicker and more confluent over time, but little proliferation was observed in the subepithelial layer (Figure 6(B)(i–iii). At Day 3, the construct was stained with appropriate biomarkers confirming the locations of the three cell types (Figure 6(B)(iv)). In the apical layer, GIE (green) in polygon shapes were tightly connected and accumulated the most at the epithelial-subepithelial interface, forming a thin barrier layer. In the subepithelial layer, elongated HGFs (red) and stellate/spindle shaped HMECs (cyan) were evenly distributed within the collagen matrices. In our triculture model of oral mucosa, HGFs and HMECs were embedded in a hydrogel composed of type I collagen, the major component of the lamina propria extracellular matrix (labeled subepithelium in Fig. 1A). Further, the GIEs strongly adhered to a thin layer of type IV collagen placed at the type I collagen gel interface and confirmed by epifluorescence microscopy of fluorescently labeled collagen IV. This layer was designed to resemble the basement membrane between native epithelial and subepithelial mucosal layers (Fig. 1A). We reasoned that natural extracellular matrix components and layer-by-layer assembly would create a tissue construct that more closely resembles native oral mucosal tissue, with increased relevance for disease modeling.

Figure 6.

Fabrication and development of oral mucosa triculture construct that was photocrosslinked for 20s. (A) Assembly process of oral mucosa triculture tissue construct starting with (i) HGF/HMEC assembly on Day (1) to form the tissue construct (ii) before and (iii) after photocrosslinking (XL) for 20s on Day 0. Next, a thin layer of collagen IV was coated onto the subepithelial layer, followed by (iv) GIEs loading to complete the triculture assembly process on Day 0. (B)-(i-iii) Time-lapse micrographs showed the growth of oral mucosa triculture construct on Day 1, 2, and 3. (B)-(iv) Immunostaining micrograph of the oral mucosa triculture construct on Day 3 with GIEs in green, HGFs in red, and HMECs in cyan.

3.4. Demonstration of cancer treatment-induced oral mucositis and recovery

Simulated cancer treatments induced significant cell damage and death in both epithelial and subepithelial layers, with more damage in epithelium. Epithelial damage and ulceration were tracked in the first 24 hours after treatment (Figure 7). Visible F-dextran transport across the keratinocyte layer was apparent by 24 hours after cisplatin exposure (Figure 7(A)), radiation treatment (Figure 7(B)), and both (Figure 7(C)). Treatment (ANOVA, F=5.2, p<0.01) and post-exposure time (ANOVA, F=4.1, p<0.01) significantly affected keratinocyte coverage (Figure 7(D)). Similarly, treatment (ANOVA, F=8.1, p<0.001) and post-exposure time (ANOVA, F=39.1, p<0.001) significantly affected F-dextran permeability (Figure 7(E)). Keratinocyte coverage was lower after 24 hours in the chemoradiation treatment only (Figure 7(D)). F-dextran permeability was higher by 8 hours after radiation and chemoradiation treatment, and 24 hours after cisplatin exposure (Figure 7(E)).

Figure 7.

Responses of oral mucosa triculture constructs within 24 h of cancer treatments. Representative micrographs reveal F-dextran transport consistent with disruption of barrier function post-exposure to (A) cisplatin, (B) radiation, and (C) chemoradiation (combined) treatments. (D) The disruption of epithelial layer coverage post-exposure from 0 hours (Day 11) to 24 hours (Day 12). (E) Higher F-dextran permeability among oral mucosa triculture constructs exposed to different cancer treatments from 0 hours (Day 11) to 24 hours (Day 12) (mean ± SEM). The white dashed box in (B)-(i) indicate the total area of the apical lumen channel spanning transchannel pores for GIE coverage measurement. The green dashed areas in (B)-(ii-iv) highlight the measured coverage areas of GIEs at specific timepoints. The coverage percentage was determined by dividing the coverage areas (green dashed areas) by the total area of the apical lumen channel spanning the transchannel pores (white dashed box). The F-dextran permeability of each construct was determined by averaging the signal in regions of interest defined by the cyan boxes in (B)-(i), taken as a percentage of signal in the apical lumen. Results of pairwise comparisons are indicated by * (p<0.05), ** (p<0.01), and † (p<0.001).

Recovery followed chemo- and radiation treatment-induced damage in the oral mucosa triculture constructs. Epithelial damage and F-dextran permeability were higher by Day 12, and mostly restored, dependent on type of treatment, by Day 18 (Figure 8). Representative combined phase contrast/epifluorescence micrographs before (Day 10) and after cisplatin treatment on Days 13, 16, and 18 reveal a transiently higher F-dextran transport across the keratinocyte layer, visibly resolved by Day 18 (Figure 8(A)). Before treatment, the keratinocyte layer was confluent, while after treatment ulcers in the keratinocyte layer were visible. Cancer treatment (ANOVA, F=60.1, p<0.001), post-exposure culture time (ANOVA, F=6.1, p<0.001), and their interaction (ANOVA, F=2.1, p<0.01) significantly affected keratinocyte coverage (Figure 8(B)). Meanwhile, F-dextran permeability was dependent on cancer treatment (ANOVA, F=56.9, p<0.001), post-exposure culture time (ANOVA, F=24.7, p<0.001), and their interaction (ANOVA, F=4.3, p<0.001) (Figure 8(C)). Compared to pre-exposure, chemoradiation treatment significantly lowered keratinocyte coverage by Day 12, while radiation treatment significantly lowered coverage by Day 13. Minimum coverages of 56±8%, 43±7%, and 26±9% resulted for cisplatin, radiation, and combined treatment exposed cultures, respectively. In contrast, keratinocyte coverage was significantly higher by Day 12 (Figure 8(B)). The time course of barrier dysfunction depended on treatment type. Higher F-dextran permeability resulted from cisplatin exposure over Days 13–15; from radiation exposure over Days 12–16, and combined exposure over Days 12–18 (Figure 8(C)). At Day 10, F-dextran permeability was roughly 10% for the three groups. For the untreated group, permeability ranged from 5–12% over the entire time-course. In contrast, F-dextran permeability at Day 12 (24-h-post-treatment) was higher at 43±13%, 48±7%, and 59±3% for cisplatin, radiation, and combined treatments, respectively. F-dextran permeability peaked at Day 13 (67±10%), Day 14 (77±8%), and Day 14 (86±4%), for cisplatin, radiation, and combined treatments, respectively. By 18, barrier function was restored to baseline except for the combined treatment constructs, that still had permeability of 47±13%.

Figure 8.

Cancer therapies-impaired mucosal barrier function and subsequent recovery. (A) Representative micrographs of oral mucosa triculture construct at specific timepoints starting from Day 10. At Day 10, GIEs covered almost all the apical lumen channel (green dashed area) and the F-dextran permeability was low as shown in (A)-(i). The cisplatin/media was administered to the basal lumen channel for 24 h. Post-chemotherapy, the epithelial was severely damaged and a significant amount of F-dextran penetrated the lamina propria at Day 13 (A)-(ii). The triculture construct was maintained in culture until Day 18 to allow for recovery. The GIE coverage gradually increased and mucosal barrier function slowly regained during the recovery phase (A)-(iii-iv). (B) The coverage percentage of GIEs underwent the same treatment over the experimental period. (C) The F-dextran permeability during the damage and recovery phases within untreated and individual cancer therapy-treated samples. Data in (B, C) is expressed as mean ± SEM. Results of pairwise comparisons are indicated by * (p<0.05), ** (p<0.01), and † (p<0.001).

Markers of cell damage, cell death, and viability were also altered by simulated cancer treatments, and returned to baseline after 8 days (Figure 9). Released LDH levels were significantly affected by post-exposure culture time (ANOVA, F=48.3, p<0.001) and interaction between treatment and post-exposure culture time (ANOVA, F=5.0, p<0.001) (Figure 9(A)). For all treatment groups, LDH levels were higher at Day 12 by approximately 1.4 times over baseline but returned to baseline by Day 18. Culture time post-exposure (ANOVA, F=4.2, p<0.05) and interaction between treatment and culture time post-exposure (ANOVA, F=3.1, p<0.05) significantly affected dead cells within epithelial (Figure 9(B)(i)) and subepithelial layers (ANOVA, F=20.2, p<0.001 and F=10.4, p<0.001, respectively) (Figure 9(B(ii))). At Day 12, the number of dead keratinocytes more than doubled for all treatment groups compared to baseline but returned to baseline by Day 18. At Day 12, the number of dead HGF/HMEC cells was approximately 100 dead cells for all treated groups, higher than that at Day 10. At Day 18, the number of HGF/HMEC dead cells returned to the Day 10 baseline as well. Treatment (ANOVA, F=3.9, p<0.05) and culture time postexposure (ANOVA, F=7.9, p=0.001) significantly affected viable epithelial area in the chip constructs (Figure 9(C)(i)). Similarly, treatment (ANOVA, F=12.6, p<0.001) and culture time post-exposure (ANOVA, F=6.5, p<0.01) significantly affected viable subepithelial area in the chip constructs (Figure 9(C)(ii)). At Day 12, combined cisplatin and radiation treatment lowered the viable keratinocyte area by half compared to the Day 10 baseline. By Day 18, viable keratinocyte area recovered and was not different from baseline, although the trend in growth appeared be stunted compared to constructs never exposed to simulated cancer treatments. At Day 12, all simulated cancer treatments lowered viable HGF/HMEC area compared to the Day 10 baseline. At Day 18, the viable area of HGF/HMEC from radiation and chemoradiation-treated constructs was only 194±27 (x1,000 μm²), lower than that of the baseline (222±23 (x1,000 μm²)).

Figure 9.

The cytotoxic effects of cancer treatments on triculture tissue constructs evaluated before exposure (Day 10), 1–2 days after exposure (Day 12), and 8 days after exposure (Day 18). (A) LDH activity, (B) dead cell count and (C) live area of (i) GIE and (ii) HGF/HMEC cells from chips among four groups: untreated, chemotherapy, radiation therapy, and chemoradiation therapy (mean ± SEM). Results of pairwise comparisons are indicated by * (p<0.05), ** (p<0.01), and † (p<0.001), with Tukey test comparisons indicated by horizontal bars connecting the compared groups.

3.5. Other effects of cancer treatments on oral mucosa constructs

Noticeably, cancer therapies also altered the morphology of epithelial cells. Time-lapse images of the epithelial layer in the constructs reveal altered morphology of keratinocytes following the cancer treatments (Figure S3). After 24 hours of cisplatin exposure, the epithelial layer was slightly disrupted with a few clusters of dead cells spotted (Figure S3(B, D)(ii)). At Day 12 and 13, more dead cells and thinner keratinocyte layers were observed within treated constructs (Figure S3(A–D)(iii–iv)). Noticeably, the cells also underwent morphological changes from smooth, polygonal cells in the untreated group to enlarged, irregularly-shaped cells in the treated groups (Figure S3(A–D)(iv)). The keratinocyte layer thickness grew in the following days for chemo- and radiation-treated groups, although the morphological changes remained (Figure S3(B, C)(v–vi)). Meanwhile, the epithelial layer in the chemoradiation group barely recovered from the treatment and the cells’ morphology remained irregular (Figure S3D(v–vi)).

To visualize the impact of cancer therapies-induced OM on the subepithelial layer, constructs were stained with PI and ActinGreen. Epifluorescence micrographs from constructs at Day 12 and Day 18 are presented in Figure S4. For the untreated group, HGFs and HMECs were organized in contiguous cell-cell networks at Day 12 and even denser networks at Day 18 as the cells became confluent (Figure S4A). In contrast, for cancer therapy-treated groups, the HGF/HMEC cytoskeleton networks were disrupted at Day 12, with the most damage in chemoradiation-treated constructs (Figure S4(B–D)(i)). After 7 days of tissue recovery, organized subepithelial cell networks were reestablished within the chemo- and radiation-treated constructs (Figure S4(B, C)(ii)), but not the chemoradiation-treated constructs (Figure S4D(ii)).

4. Discussion

Oral mucositis associated with chemo- and/or radiation therapy causes extreme pain, difficulty eating, weight loss, systemic infections and other serious complications in cancer patients.[5] Epithelial damage and ulceration are associated with damage to subepithelial fibroblasts and endothelial cells, resulting in impaired basic oral functions.[50, 51] The initiation phase of mucositis caused by a series of molecular signals occurs rapidly within seconds post-insult,[52] developing over days into confluent and progressive pathology. The triculture mucosa on a chip utilizes all three main structural cells implicated in mucositis—keratinocytes, fibroblasts, and endothelial cells—and was fabricated in a geometry such that morphological alterations and optical signals from the epithelial and sub-epithelial layers are instantly and simultaneously accessible with conventional wide-field light microscopy. Layer morphology and barrier function were established after 10 days of chip culture. Cell and tissue layer alterations were apparent hours to days after exposure on the chip to simulated cancer treatments, with synergistic effects on the induction of damage and recovery by cisplatin and radiation treatments. Barrier function and high cell viability were restored in the chip constructs by Day 18 in most cases, with some lasting alterations to cell morphology. This data, in total, establishes an in vitro microfluidic-based platform to model oral mucositis. The effects of cancer therapies on the individual cell layers and the overall construct functional properties are easy to track via simple microscopy and biochemical endpoints.

Photocrosslinking with Ruthenium and blue light was the first innovation for chip culture that allowed functional maturation of the oral mucosa construct, damage-simulating mucositis, and recovery. Uncrosslinked mucosa-on-a-chip constructs previously failed after 7 days of culture, due to keratinocyte invasion and collagen gel contraction,[31, 32] confirmed in this study (Figures 3(D), 4). Blue light (400–450 nm) catalyzes di-tyrosine crosslink formation in collagen through Ru(III) and a sulfate radical [53] (Figure 2(A)), with minimum reported cytotoxicity to embedded cells.[54–56] In the mucosa chip cytotoxicity is minimal with 20 seconds or less blue light exposure (Figure 3(A–C)), and 20 seconds is sufficient to inhibit gel contraction (Figure 3(D)) and keratinocyte invasion (Figure 4). Indeed, maturation of keratinocyte layer barrier function was only possible in the chip with photocrosslinking (Figure 5). Previous microfluidic mucosa constructs inserted a cell impermeable membrane or barrier between epithelial and sub-epithelial layers.[57–60] Photocrosslinking is a viable alternative to synthetic membranes that are not present in native mucosa tissue. Ruthenium crosslinking has been employed to strengthen the mechanical properties of a wide range of biopolymers including collagen.[36, 37, 61, 62] Compared to other crosslinking approaches, Ruthenium crosslinking requires no ultraviolet light, thus minimizing phototoxicity, and appears to be less cytotoxic than many chemical crosslinkers.[36, 63, 64] Several reports indicate ruthenium photocrosslinking is fast, efficient, and requires no chemical conjugation (e.g., methacrylation), comparing favourably to Irgacure 2959, Lithium phenyl-2,4,6trimethylbenzoylphosphinate, Eosin-Y, and riboflavin.[55, 65, 66] However, Ruthenium crosslinking in the presence of cells must still be optimized, as exposure to excess persulfate is toxic to cells.[56, 67] Controlling photocrosslinking time to be as short as possible and rapid rinsing of the constructs are important steps to minimize toxicity. After rinsing, long-term toxicity is minimal, unlike glutaraldehyde crosslinking.[68]

Besides photocrosslinking, the second innovation to establish an oral mucositis model on a chip was to observe mucosal damage and recovery caused by simulated cancer treatments. Within 8 hours of some treatments, F-dextran permeability is higher (Figure 7(C)). By 24 hours keratinocyte coverage is lower as lesions begin to form (Figure 7(B)), with the most severe alterations 48–72 hours post-exposure (Figure 8). By Day 18 (7 days post-exposure), F-dextran permeability and keratinocyte coverage were restored except in the chemoradiation-treated constructs (Figure 8). This pattern of damage by Day 12 (24 hours post-exposure) and partial-to-total restoration by Day 18 (7 days post-exposure) was also apparent by released LDH, numbers of dead cells, and live cell area (Figure 9). Interestingly, keratinocytes returned more completely to baseline numbers of dead cells and viable cell area than HGF/HMECs (Figure 9(B, C)). Palifermin, currently used to treat mucositis in cancer patients, would presumably act primarily on the epithelial cell layer and would not strongly influence sub-epithelial cell types. The morphological changes of keratinocytes post-treatments also suggest cellular degeneration persisted even after epithelial ulceration had recovered (Figure S3). Subepithelial cell cytoskeletal alterations also persisted over the experimental period (Figure S4). In the literature, mucosal toxicity varied widely between cancer patients, possibly due differences in treatment type, dose, frequency, modality of administration, and factors intrinsic to the patient.[5, 69] Also, mucositis was associated with signal secreted from the submucosa and interactions with the external environment.[3, 70] Therefore, a combination of treatments involving growth factors or protective agents that target keratinocytes, fibroblasts and endothelial cells may prove more effective against oral mucositis than any single treatment.

Advantages to the presented triculture oral mucosa on a chip exist despite limitations. On advantage is the presence of endothelial cells within the current triculture model. Recent studies emphasized the importance of endothelial injury in determining the severity of cancer therapy-induced mucosal damage.[71] Other cell types, particularly immune cells, could be added to the chip. Further, vascularized networks were not developed in the present chip constructs, although this could be pursued in the future. The horizontal layout of the triculture chip is a second key advantage of the tissue platform. In situ observation and assessment of cellular responses to cancer therapies was facile and rapid with conventional widefield microscopy.[31, 32] This layout is particularly well-suited to live-cell studies with incubated microscopy, to observe timedependent processes such as cell death, apoptosis, and proliferation, which will be pursued in future studies. A third advantage was the extended cultivation period (18 days and more) allowed by photocrosslinking. Eighteen days was enough time for the complete maturation of the triculture construct by the metric of barrier function, and modeling experiments focused on the initiation of mucosal damage and recovery. Other evaluations are possible besides the cytotoxicity, viability, cell damage, and microscopy metrics presented in this study, and other treatments besides cisplatin and radiation. Molecular alterations will be explored in future studies and other treatments (doxorubicin, and x-ray radiation) will also be evaluated. Potential therapeutic strategies against oral mucositis will also be tested on the chip. Such potential treatments should be carefully assessed for effects on cancer cell lines in vitro, before preclinical testing on animal models is considered. Platform improvements are possible, including air-liquid interface, dynamic culture with media flow, and addition of immune and microbial cells commonly found in the oral environment.

While no in vitro model completely recapitulates native tissue and tissue pathology, functional metrics partially establish the clinical relevance of oral mucositis modeled on a chip. F-dextran permeability, low at Day 10, high 24–72 hours post-exposure to cancer treatments, and mostly restored by Day 18, was one functional metric relevant to initial damage found in oral mucositis. This metric is superior to transepithelial electrical resistance on a chip due to variable resistance of the PDMS material in the chip.[32] Low permeability was associated with keratinocyte coverage of more than 80%. In comparison, epithelial barrier function was disrupted in uncrosslinked constructs due to the keratinocyte invasion (Figure 5). Secondly, the development of lesions in the keratinocyte layer following cancer treatments on the chip is consistent with ulceration found in mucositis. Thirdly, the subepithelial configuration of HGFs/HMECs in the collagen gel of the chip constructs mimicked the physiology of native oral mucosa tissue where fibroblasts and vascular endothelial cells reside in close proximity in the lamina propria.[72] Furthermore, previous studies suggested that co-culture of fibroblasts and endothelial cells promote and stabilize vessel networks,.[73, 74] although these were not observed in the present study. Therefore, the triculture mucosa on a chip represents a basic in vitro model that mimics some aspects of initial mucositis damage and also exhibits recovery.

5. Conclusions

This study presents a novel triculture platform resembling human oral mucosa that is robust and stable for multiple weeks of culture. Photocrosslinking was biocompatible and significantly enhanced the structural integrity of the triculture construct. Epithelial damage and ulceration caused by cancer therapies resembled early oral mucositis and the subsequent recovery was easily tracked in situ through widefield microscopy in a single field-of-view, as well as through compatible biochemical assays. The triculture platform provides a physiologically relevant tissue model that can be further developed to test therapeutic treatment strategies for mucosal pathologies and provide mechanistic insights into mucosal function and dysfunction in a cost-effective manner. Future studies will explore the molecular mechanisms driving oral mucositis that are potentially targetable by anti-mucositis treatments.