Human mammary epithelial cell line (HMT-3522, also known as S1; Public Health England) was cultured in mammary epithelium growth medium (MEGM) (Lonza). Human primary mammary fibroblasts (HMF; ScienCell) were grown in fibroblast medium containing 2% fetal bovine serum (FBS), 1% fibroblast growth supplement, and 1% penicillin/streptomycin (ScienCell). The DCIS cell line (MCF10-DCIS.com; Asterand) was cultured in DMEM/F12 (50:50) supplemented with 2mM L-glutamine, 5% horse serum, and 1% penicillin/streptomycin (Invitrogen).

To achieve endogenous fluorescence, lentiviruses expressing red fluorescent protein (RFP), cyan fluorescent protein (CFP), and green fluorescent protein (GFP) were transduced into the mammary epithelial cells (S1), mammary fibroblasts (HMF), and DCIS cells (DCIS.com), respectively. 10⁶–10⁷ transduction units/ml of lentivirus was used and incubated in the culture media for 6–10 hours in the presence of 8 μg/ml polybrene. Three days of transduction were followed by treatment of culture media with puromycin (10 ng/ml) for 2 weeks, and cells expressing the fluorescent proteins were selected.

To generate DCIS spheroids, DCIS cells were seeded into a 96-well hanging drop plate (Perfecta 3D hanging drop plate, 3D Biomatrix) and cultured for three days. The average size of the spheroids was calculated by measuring the diameters of spheres in a well.

To mimic the 3D structural organization of the human mammary duct, we designed a compartmentalized microdevice consisting of upper and lower microchannels separated by a thin ECM-derived membrane that mimics the basement membrane (Fig. 1B). In this design, the upper microchannel recreates the ductal lumen in vivo and enables continuous flow of culture media required for growing and maintaining mammary epithelial cells and DCIS spheroids on the ECM membrane. A stromal layer impregnated with mammary fibroblasts is formed on the lower side of the ECM membrane and perfused with culture medium through the lower microchannel to mimic the vascular compartment of capillaries in mammary stroma in vivo.

The upper and lower microchannel layers were fabricated using soft lithography. Briefly, poly(dimethylsiloxane) (PDMS; Sylgard, Dow Corning) pre-polymer was mixed with a curing agent (10:1 w/w ratio of PDMS to curing agent) and cast against a mold containing photolithographically prepared microchannel features. The size of the cell culture chamber was 1 mm (width) × 3 mm (length) × 200 μm (height) (Supplementary Fig. 1A).

A vitrified collagen membrane was used as a biomimetic basement membrane that served to separate the upper and lower microchannels. Preparation of the vitrified collagen membrane began by mixing type I collagen (9.46 mg/ml; BD Biosciences) with 10X DMEM (Sigma-Aldrich), 1M HEPES (HEPES, Gibco), distilled water, and 1N NaOH. The final collagen concentration was 4.0 mg/ml, and pH was neutralized to 7.2–7.4. 14 μl of the mixture solution was then placed over the cell culture chamber of the lower PDMS layer and incubated at 37°C in 100% humidity for 1 hour to induce gelation (Supplementary Fig. 1B). Subsequently, the collagen hydrogel was dried at room temperature for at least 48 hours to produce a thin membrane. Since this resulting membrane was anchored to the large flat surface surrounding the recessed microchannel, it remained firmly attached to the lower PDMS slab. For bonding, a hand-held corona treater (Harrick plasma) was used to treat the surfaces of the upper and lower PDMS layers. For the lower layer containing the collagen membrane, care was taken to treat only the edges of the PDMS slab and to avoid direct exposure of the membrane to corona. Finally, the two layers were brought in contact and incubated at 60 °C for 3 hours to achieve permanent bonding (Supplementary Fig. 1B). After bonding, the membrane was visually inspected using a microscope to ensure its integrity without structural failure.

Prior to cell seeding, a fully assembled microdevice was sterilized by ultraviolet irradiation. To model the stromal tissue in the mammary duct, we formed a thin fibroblast-laden collagen gel layer attached to the vitrified membrane in the lower microchannel (Figs. 2B, 2C). As the first step to accomplish this, the lower microchannel was filled with 2% bovine serum albumin (BSA; Sigma-Aldrich) and incubated for 1 hour at 37°C to prevent binding of the collagen gel to the channel surfaces. Upon aspiration of BSA, a collagen precursor solution (3.0 mg/ml) mixed with human mammary fibroblasts (5×10⁶ cells/ml) was injected into the lower microchannel and incubated for 2 hours to permit collagen gelation (Fig. 2B). For this step, type I collagen was used to represent the most abundant ECM component in the mammary stroma. Subsequently, HMF medium was perfused into the upper channel by a syringe pump (New Era Pump System) at a volumetric flow rate of 60 μl/hour. The fibroblast-embedded collagen gel was maintained for 3 days with continuous medium perfusion. During this period, traction forces generated by the fibroblasts resulted in contraction and subsequent detachment of the cell-laden gel from the bottom surface of the lower microchannel, creating a continuous gap along the channel length that permitted flow of culture media (Fig. 2C). Consistent gel contraction during this procedure was achieved by controlling parameters such as cell seeding density, collagen concentration, and polymerization conditions characterized by time, temperature, and humidity.

After 3 days of culture, the upper side of the vitrified membrane was coated with a mixture of Matrigel (2 mg/ml) and fibronectin (0.5 mg/ml) for 2 hours. Matrigel, which contains laminin and type IV collagen, was advantageous for more faithfully recapitulating the key composition of the basement membrane of the mammary ductal epithelium in vivo. Human mammary epithelial cells (7×10⁶ cells/ml) were then introduced into the upper microchannel and allowed to attach to the membrane surface for 2 hours. During microfluidic co-culture, the epithelial cells and fibroblasts were maintained by flowing MEGM and HMF culture media through the upper (40 μl/hour) and lower (30 μl/hour) microchannels, respectively (Fig. 2E). These flow rates were determined empirically by identifying the optimal rates of medium flow at which the cells remained viable for extended culture periods and no significant structural changes to the collagen stromal layer were induced by fluid flow. The flow rates used in our study were also within the physiological range of interstitial flow (2–70 μl/hour)²¹, which plays an important role in the survival of the mammary cells.

DCIS spheroids were collected from the 96-well hanging drop plate and centrifuged. The spheroids were then reconstituted in MEGM medium at 5 spheroids/μl and introduced into the upper microchannel. This step was followed by incubation of the microdevice without media perfusion for at least 8 hours to enable adhesion of the spheroids to the mammary epithelium (Fig. 2F). After firm attachment of the spheroids, flow of culture medium was resumed in the upper microchannel at a volumetric flow rate of 40 μl/hour.

We used a commercially available colorimetric lactate dehydrogenase (LDH) cytotoxicity detection kit (Takara Biotechnology) to evaluate cell viability in our model. Perfused media from the upper and lower microchannels were collected separately and assayed using the detection kit to measure LDH release from the mammary epithelial cells and fibroblasts, respectively. The measured LDH levels were then converted to percent cytotoxicity according to the instruction provided by the manufacturer. Finally, percent cell viability was determined by subtracting the percent cytotoxicity from 100%.

For drug testing studies, the cells were treated with paclitaxel (Sigma Aldrich) at a concentration of 20 nM, which was injected into the lower microchannel for 24 hours. This particular dose was chosen based on clinically relevant doses previously used in in vitro studies of paclitaxel efficacy²². Analysis of cytotoxicity was carried out using the aforementioned method based on a commercial LDH detection kit to calculate the percentage of cells that exhibited cytotoxic responses to paclitaxel. To assess the relative contribution of LDH production by DCIS cells and normal mammary epithelial cells, we analyzed the levels of paclitaxel-induced LDH release from normal epithelial cells in control microdevices that excluded DCIS spheroids from the upper microchannel while maintaining the lower stromal layer unchanged. The cytotoxic response of our DCIS-on-a-chip model to paclitaxel was also compared to that measured in another control group that did not receive paclitaxel treatment. Additionally, changes in the projected area of DCIS spheroids due to drug treatment were analyzed by using ImageJ (NIH) to evaluate the therapeutic effect of paclitaxel on the suppression of tumor growth.

The microstructure of the vitrified collagen membrane was imaged using scanning electron microscopy (SEM). Prior to imaging, membrane samples were fixed with 2% formaldehyde for 10 minutes. SEM images were acquired using a JSM 7410F scanning electron microscope (JEOL Ltd.) with a Gatan cooled charge-coupled device camera.

Statistical significance was determined using a two-tailed Student’s t-test for at least three independent experiments and are presented as the mean ± standard deviation (std.). P-values less than 0.05 were considered to be statistically significant and indicated with asterisks (**).

Our microengineering techniques allowed for the production and prolonged maintenance of the human mammary ductal epithelium and a human mammary fibroblast-containing 3D collagen matrix separated by a thin biomimetic basement membrane (Fig. 3A). The collagen vitrification process conducted in this study produced thin ECM membranes with an average thickness of 20.9 μm that consisted of a dense network of collagen fibers (Fig. 3A). The vitrified membrane remained attached to the lower microchannel, and its structural integrity was maintained during corona treatment and subsequent channel bonding. When buffer solutions were flowed into the channels prior to cell seeding, the membrane was observed to become hydrated rapidly without exhibiting significant structural changes. The hydrated vitrified membrane served as a stable barrier during injection of collagen gel solution into the lower microchannel.

In the formation of the membrane-bound stromal layer in the lower microchannel, BSA surface coating was critical for preventing unwanted adhesion of collagen matrix to the bottom channel floor during continuous contraction of the collagen gel. In the absence of BSA coating, this contraction led to the detachment of the collagen gel layer from the vitrified membrane and its shrinkage to the bottom surface of the lower microchannel (Supplementary Fig. 2). Three days of microfluidic culture and concomitant gel contraction produced a stromal layer with an average thickness of approximately 150 μm that remained attached to the vitrified membrane. This process also resulted in the formation of a gap between the contracted gel and the bottom floor of the lower microchannel that provided fluidic access for perfusion of fibroblast medium during co-culture with epithelial cells. Reduction in the concentration of serum in the fibroblast culture medium inhibited further shrinkage of the gel layer.

During the course of culture, mammary fibroblasts embedded in the collagen gel began to spread within 24 hours after seeding and gradually became elongated to exhibit their characteristic spindle-shaped morphology (Fig. 3A). Human mammary epithelial cells seeded into the upper microchannel were observed to establish firm attachment to the vitrified membrane surface and proliferate to form a confluent monolayer over a period of 24 hours. Cross-sectional images of the tissue constructs showed a dense mammary epithelium overlying the 3D stromal layer containing uniformly distributed fibroblasts (Fig. 3B). As shown in Fig. 3C, microfluidic culture conditions in our device permitted prolonged co-culture of the epithelial cells and fibroblasts without significant viability loss, with over 85% of the cells maintained alive throughout the 1-week culture period. Flow of culture medium through the gap in the lower channel created by gel contraction was effective for maintaining the fibroblasts embedded in the 3D matrix scaffold and did not induce undesirable changes in the structural integrity of the stromal tissue during long-term culture.

For optimal injection and handling of DCIS spheroids in the upper microchannel with a height of 200 μm, we examined the relationship between the number of seeded cells and the diameter of resulting spheroids in standard hanging drop culture after 3 days of culture (Supplementary Fig. 3). Spheroids having a diameter of approximately 150 μm were selected to mimic early-stage DCIS lesions. Under static conditions without the flow of culture medium, the spheroids injected into the upper channel successfully adhered to the epithelial cell surface within 10 hours after seeding. When flow was resumed, the DCIS spheroids remained adherent on the surface and gradually became integrated into the epithelium over the course of two days (Fig. 4A). Confocal analysis of the epithelium revealed that the spheroids became flattened and positioned themselves on the same horizontal plane as the epithelial cells, suggesting full incorporation of DCIS into the epithelial layer (Fig. 4B). Importantly, the vast majority of the DCIS cells remained in their original spheroids without invading the surrounding epithelial tissue and the vitrified membrane, and disintegration of the spheroids due to cell migration was not observed throughout the entire culture period. Prolonged culture in the microfluidic device led to the enlargement of the spheroids, indicating the ability of our model to support proliferation of the DCIS cells. In addition, DCIS spheroids did not induce any significant alterations in the morphology of the underlying fibroblasts (Fig. 4A).

Next, we evaluated the potential of our disease model as a drug screening platform using a clinical anticancer drug, paclitaxel. To simulate physiological delivery of intravenously administered paclitaxel from the vasculature in the mammary stroma to DCIS lesions, we generated a continuous flow of paclitaxel solution through the lower microchannel. As shown in Fig. 5A, during the course of drug treatment, cellular production of LDH, as represented by percent cytotoxicity, gradually increased and doubled at day 4 (DCIS+S1 Paclitaxel), indicating a paclitaxel-induced statistically significant increase in cytotoxicity compared to paclitaxel untreated group (DCIS+S1; p<0.05). Interestingly, when paclitaxel was introduced into control devices without DCIS spheroids (S1 Paclitaxel) at the same dose and duration, cytotoxicity did not increase and statistically significant difference between DCIS+S1 and S1 Paclitaxel was not observed (Fig. 5A). This is presumably due to substantially higher sensitivity of the cells in the DCIS spheroids to the drug designed to selectively inhibit the assembly of mitotic spindles and cell division in actively proliferating cancer cells.

To quantitatively assess the anti-proliferative effects of paclitaxel, we measured changes in the projected area of the DCIS spheroids during drug treatment. In the untreated group, continuous proliferation of DCIS cells led to significant enlargement of the spheroids (p<0.05; Fig. 5B) and a nearly threefold increase in their size (Fig. 5C). Although mixed populations of DCIS cells and normal mammary epithelial cells were observed along the immediate periphery of the spheroids, it appeared that the proliferating DCIS cells continuously displaced their surrounding epithelial cells and managed to maintain the boundary between the two distinct populations during expansion of the spheroids. However, this growth was confined in the epithelial compartment and did not result in tumor cell invasion into the underlying stroma. In contrast, the diameter of the DCIS spheroids exposed to paclitaxel remained unchanged or slightly decreased, demonstrating the efficacy of paclitaxel in arresting tumor cell proliferation and preventing the growth of DCIS lesions (Fig. 5A&C). As was the case with the untreated group, the stromal layer remained intact without DCIS cell invasion during and following paclitaxel treatment.