High-Throughput Magnetic Actuation Platform for Evaluating the Effect of Mechanical Force on 3D Tumor Microenvironment

Abstract

Accurately replicating and analyzing cellular responses to mechanical cues is vital for exploring metastatic disease progression. However, many of the existing in vitro platforms for applying mechanical stimulation seed cells on synthetic substrates. To better recapitulate physiological conditions, a novel actuating platform is developed with the ability to apply tensile strain on cells at various amplitudes and frequencies in a high-throughput multi-well culture plate using a physiologically-relevant substrate. Suspending fibrillar fibronectin across the body of the magnetic actuator provides a matrix representative of early metastasis for 3D cell culture that is not reliant on a synthetic substrate. This platform enables the culturing and analysis of various cell types in an environment that mimics the dynamic stretching of lung tissue during normal respiration. Metabolic activity, YAP activation, and morphology of breast cancer cells are analyzed within one week of cyclic stretching or static culture. Further, matrix degradation is significantly reduced in breast cancer cell lines with metastatic potential after actuation. These new findings demonstrate a clear suppressive cellular response due to cyclic stretching that has implications for a mechanical role in the dormancy and reactivation of disseminated breast cancer cells to macrometastases.

Graphical Abstract

1. Introduction

Metastatic disease is the leading cause of cancer-related mortalities that culminates with the outgrowth of a secondary tumor in a distal tissue.^{[1, 2]} Each step of the metastatic cascade involves changes in the mechanical forces of the tumor microenvironment that affect cell fate. In primary breast cancer (BC), the cancerous tissue is five times stiffer than the surrounding mammary gland.^[3–5] Tissue stiffening enables cells to bypass the contact inhibition of proliferation and continue dividing beyond the typical physical limits of a tissue.^[3] The activation of Yes Associated Protein/Transcriptional Activator with PDZ binding motif (YAP/TAZ) partially achieves the bypass of contact inhibition.^[3] However, changes in the mechanical properties of tissues are not the only changes that occur during tumor progression. A currently under-explored area of the metastatic cascade is the effect that dynamic forces native to the distant organs have on early disseminated tumor cells.^{[3, 6–9]} The two most common sites of BC metastasis are the bones and the lungs, which natively undergo cyclic compressive and tensile stresses, respectively.^{[10, 11]} The effects of this native mechanical stimulation on the phenotypic and genetic changes of disseminated tumor cells are not yet known.

To study this phenomenon, we developed a novel biomimetic lung model. This device utilizes magnetic actuation of suspended fibronectin (FN) fibrils to dynamically stretch cells in a 3D biologically-relevant manner. FN is a major extracellular matrix component that transiently increases in the lungs during premetastatic niche formation, in part, to recruit disseminated tumor cells.^[12–17] The majority of commercial platforms that incorporate cyclic mechanical stimulation can only be applied to cells in 2D, where cell response to biophysical and biochemical cues can deviate from in vivo behaviors.^[18–20] Also, these technologies often require specialized culturing plates and/or bulky pneumatic connections, resulting in cumbersome experimental conditions.^[21,22] To overcome the limitations encountered by commercial systems, many research groups use silicone elastomer that can be made into an array of individually addressable elements with high strain rates.^[23–25] Other cell stretching modalities include electromagnetically actuated structures and individually addressable piezoelectric actuators in a braille display.^[26,27] Similar to the commercial platforms, cells are typically seeded onto silicone or other synthetic substrates and allowed to form a monolayer that can be stretched to induce a cellular response. Optical and acoustic tweezers present an interesting modality to stretch and probe individual cells but their throughput is limited and require specialized equipment.^[28,29] The advantages of magnetic actuation include the untethered transmission of energy and the reduced risk of contamination for dynamic cell culture studies.^[30–34] However, magnetic actuation systems to date have used synthetic substrates to interface with cells directly or used intracellularly-embedded magnetic nanoparticles to apply localized strains without the extracellular matrices.^[33–34] A few recent technologies allow stretching of cells in collagen matrices.^[35–38] However, due to the importance of FN during early dissemination, utilizing this extracellular matrix protein provides a key model of the premetastatic lung niche.

Here, we describe a simple new platform that can capture the effects of biomimetic cyclic stretching on cancer cells and their surrounding microenvironments. We show that this platform is compatible with standard 24-well plate culture dishes, creating a high-throughput design. Using this system, we report that cyclical tensile strain of the suspended breast cancer cells displayed a reduced rate of metabolic activity and preserved fibronectin integrity when compared to the static counterparts, without widespread cell death after a week of actuation.

2. Results and Discussion

2.1. Actuator Design and Characterization

We designed a soft actuator capable of inducing a range of strains between 5–25%, which are representative of the strains of the alveoli (Equation S1–S2, Supporting Information).^[39] The actuator body consists of an outer frame and a cantilever made out of polydimethylsiloxane (PDMS) for structural integrity and biocompatibility (Figure 1a). Embedded at the end of the cantilever is a 2-mm diameter and 1-mm thick neodymium (NdFeB) permanent magnet. The actuator fits within the 16 mm diameter well of a standard 24-well culture dish. When an external magnetic field is applied, the cantilever will deflect out of plane (Equation S3, Supporting Information).

Figure 1.

Actuator fabrication and characterization. a) Schematic representing the manufacturing of PDMS magnetic actuator with fibronectin coating. Image of an actuator with a fibronectin coating. b) Components of the magnetic actuation platform. c) Magnetic field is strength produced by the platform at the designed distances as a function of time (black) (n=3, mean ± s.d.). Transient strain experienced by fibronectin according to our theoretical strain (blue). d) One cycle of fibronectin stretching at 0.3 Hz. e) Deflection vs. magnetic field strength of the theoretical response of the actuator (dashed red line), and the response of the actuator with (black) or without (red) fibronectin coating (n=3, mean ± s.d.). f) Deflection vs. magnetic field strength of fibronectin-coated actuators depicting region of physiological strain (blue), and lower (5% to 10%) and higher strain (20% to 25%) ranges in yellow (n=3, mean ± s.d.). Scale bar for a. and d. is 2 mm.

To vary the magnetic field strength, we fixed 12 19-mm-wide and 2.4-mm-thick NdFeB permanent magnets onto a 3D-printed actuating platform that can be programmed to travel along the z-axis (Figure 1b). We used numerical analysis via COMSOL Multiphysics software to identify a magnet arrangement that could achieve a uniform magnetic field while allowing for the largest sampling size, resulting in 12 samples across the 24 well culture plate (Figure S1, Equation S4, Supporting Information). This arrangement also minimized the dipole-induced magnetic interaction between the actuator and the surrounding magnets so that the out-of-plane deflection was the dominant mode of actuation with negligible lateral twisting or bending. The programmable magnetic array continuously actuated at 0.3 Hz to mimic the average respiration rate (18 breaths/min) of an older patient, when BC is typically diagnosed (Figure 1c–d).^[40–43]

To create a mimic of the early metastatic niche, we deposited a 76.7 ± 9.0 μm thick network of fibrillar FN that spanned between the top edge of the magnet reservoir in the cantilever and the edge of the frame at the base of the actuator. These points of attachment have a horizontal distance of 1.45 mm, but are offset by 1.1 mm in height, giving the FN a total length of 1.8 mm. FN fibrillogenesis across this gap was achieved following our previously described rotational coating technique. Utilizing fibrillar FN creates a thicker, 3D environment than traditional adsorption techniques and exposes cryptic binding domains of the FN which make the environment more biologically active than while in its globular conformation.^{[44, 45]} The FN architecture that formed on the actuators consisted of bundled fibers interlaced into an aligned mesh-like network. The orientation of the aligned FN fibrils occurred along the principle direction of the applied fluid flow that develops during the rotational coating process (Figure S2, Supporting Information). Furthermore, during cyclic stretching, strain was applied in the direction of the aligned FN fibrils. The fiber alignment was determined by fitting the directionality histogram to a gaussian distribution in order to measure the dispersion. The directionality histogram identifies the amount of a structure that occurs in a given direction. Therefore, an aligned sample would have a peak that occurs at a specific orientation, but an isotropic sample would have a flat distribution with no discernable peak. A representative histogram can be seen in Figure S2, where a single peak was observed for the aligned FN network. The dispersion is the standard deviation of the gaussian fit, and the FN matrix that formed on the actuators had an average dispersion of 6.77 ± 1.16° (Figure S2, Supporting Information).

Next, by calculating the change in length of FN as a function of the measured cantilever deflection, we determined that a deflection angle between 2–12º would apply the desired 5–25% strain on the FN band (Figure 1e–f, Equation S3, Supporting Information). To determine if the FN network was damaged by cyclic stretching, confocal images of the FN network were obtained before and after 4360 duty cycles were applied to the FN. No significant changes were observed in the FN network after more than 4,000 duty cycles of stretching (Figure S2, Supporting Information).

We then developed a finite element (FE) model to quantify the local strains across the FN region of interest. We first quantified the mechanical properties of FN fibrils using a custom FN-coated polyethylene terephthalate (PET) frame and tensile testing system (Figure 2a–d). Specifically, FN bands were fluorescently labeled using AF488-conjugated wheat germ agglutinin, marked with photobleached fiducial lines, and uniaxially loaded on a custom spring mount connected to a micromanipulator with a microforce sensing probe (Figure 2b–c).

Figure 2.

Fibronectin tensile testing and finite element modeling a) Schematic of PET frame (purple) with PDMS blocks (blue) and fibronectin coating (green). Scale bar is 1 mm. b) Zen Blue 3D rendering of fibronectin coating, stained with AF488-conjugated wheat germ agglutinin, with photobleached fiducial lines imaged using confocal microscopy. Scale bar is 500 μm. c) Representative images of fibronectin coating before (top) and during (bottom) tensile testing. PDMS blocks outlined in white. Photobleached fiducial lines marked by orange arrows. The bottom image is shown at 29% strain. Scale bar is 1 mm. d) Stress-strain curve fit showing the empirical data from a representative FN sample. e) Stress-strain curve fits from 10 FN samples (black) overlaid with the average stress-strain curve that was chosen for the finite element simulation (green). f) Computer-aided design (CAD) of the actuating device (left) with finite element representation (right) of the region of mechanical interest (red dashed). Orange indicates the fixed boundary conditions of x, y, and z translations at nodes. The average fibronectin band shape was used for all simulations. g) Side view of the finite element model with force value of 0.03 N approximated to resemble the experimental deformation associated with a 12° cantilever angle. h) The nonuniform shape of the fibronectin band results in uneven strain during deflection, with the bottom corners experiencing far less strain than the center axis, and areas of largest strain at the top corners.

These mechanical properties informed the FE model, in which the average dimensions and shape of the FN band observed in experimental results were simulated (Figure 2e–f). Specifically, the band was 1.73 mm wide at the magnetic reservoir, 3 mm wide across the middle, 4.27 mm wide at the base, and 80 μm thick. An upward force was applied in line with the middle of the magnet of the cantilever (Figure 2g). The strain was recorded at three locations on the FN band as the applied force increased, corresponding to a 0–12° deflection angle to match the experimental results (Figure 2h, Figure 1e). Due to the non-uniform shape of the FN band, various regions on the band experience different local strains. On average, the strain is relatively uniform along the central axis of the band, as seen in the strain map and by the strains plotted at the “center” and “bottom middle” band locations (Figure 2h). However, at the bottom corners of the band, much less local strain is experienced, becoming more divergent with increasing deflection angle. Similarly, at the top corners where the FN band is the smallest, there is a concentration of high local strain (Figure 2h). Based on the FE model, during peak longitudinal strain, corresponding with a 12° deflection (Figure 2h), the width of the center of FN band will compress slightly to 2.78 mm (7.3% transverse strain) (Figure S3, Supporting Information).

2.2. Cellular Response to Actuation

We tested the effects of cyclic stretching on three BC cell lines and human lung fibroblasts (HLF) at a low strain amplitude (5–10%) or a high strain amplitude (20–25%), marked on the graph as LS and HS, respectively (Figure 1f). The BC cell lines were MDA-MB-231 (231), HME2, and MCF10CA1a (Ca1a). The 231 cells are a triple negative, highly invasive, metastatic human cell line isolated from a pleural effusion in the lung.^[46] The HME2 cells were constructed via directed transformation of the human mammary epithelial cell line (HMLE) by overexpression of the oncogene HER2.^[17] These cells are capable of primary tumor formation but have little metastatic potential.^[17] Lastly, the Ca1a cells are an H-RAS transformed cell line derived via serial in vivo passage of MCF10A-T1K cells They are triple negative and have slight metastatic potential, but retain their epithelial-like phenotype in vitro, providing a comparison to the 231 cells, which are more mesenchymal-like ^[47] (Table 1).

Table 1.

Relative comparison of breast cancer lines used.

	MDA-MB-231	HME2	CA1A
HORMONE STATUS	Triple-Negative	HER2+	Triple-Negative
PHENOTYPE	Mesenchymal-like	Epithelial-like	Epithelial-like
MIGRATORY/INVASIVE	Very (+++)	Little (+)	Little (+)
METASTATIC POTENTIAL	Some (++)	None (−)	Little (+)

The average seeding efficiency on the actuators was 31.85% ± 5.4%, which is typical for a suspended porous structure.^[48] Within 48 h, cells infiltrated the FN network, utilizing the 3D volume created by the fibril bundles (Figure S4, Supporting Information). By counting the cells in the supernatant after 24 h of actuation, we determined that actuation does not dislodge cells. We calculated a maximum daily knock-off percentage of 4.68% ± 0.91% based on the number of cells in the supernatant and the calculated total population. This is an overestimate, as it assumes no cells in the supernatant in static cultures and that all floating cells in actuating groups are due to mechanical dislodging, which does not account for typical cell death. The seeding efficiency and the knock-off percentage were not noticeably variable between the cell types.

In all three BC lines, cyclic stretching significantly reduced the global metabolic activity of the sample, detected via resazurin (Figure 3a,c–e). 231 cells undergoing the high strain regime experienced similar results. We saw no statistical difference in the metabolic rate between low and high strain, indicating that, at a global level, 5–10% strain is sufficient to inhibit cell vitality or population growth (Figure 3a,e). Importantly, we still cultured all static samples on the FN-coated actuators, but they were not actuated. This removed bias between the groups, including those that may have resulted from differences in 3D and 2D proliferation rates, seeding efficiencies, and biological stimuli induced by FN. The mechanically-induced reduction in metabolic activity is not seen in HLF samples, indicating that the tensile force is only detrimental to a non-pulmonary cell type (Figure 3b,e). The knock-off percentage was subtracted from the static culture values as a daily compounded rate to further demonstrate that this reduction in mitochondrial activity of the sample is not a function of the mechanical force dislodging the BC cells (Figure 3a–d). In addition, live/dead staining of a sample after 6 d of high-strain actuation shows that the mechanical force did not induce widespread apoptosis (Figure 3f).

Figure 3.

Actuating reduces proliferation. Metabolic activity is analyzed for cells in static conditions (red) or under cyclic actuation (black) in a) 231 cells, b) human lung fibroblasts, c) HME2 cells, and d) Ca1a cells. The knockoff percentage is subtracted from the static values as a daily compounding rate (red x) e) Normalized growth rates from the metabolic data indicate significant growth reduction in 231, Ca1a, and HME2 cells, with no difference seen between the low and high strain actuation in 231 cells (* = p<0.05, n=3, mean ± s.d.). f) Staining of 231 cells after 6 d under high strain conditions shows live (green) and dead (red) cells with the quantified viability (n=3, mean ± s.d.). Scale bar is 100 μm. The translucent regions represent the approximate location of the PDMS cantilever and body edges.

Immunofluorescent images of the 231 cells after 3 d in culture indicate the presence of more cells on the static devices (Figure S5, Supporting Information). From this and the reduction in global metabolic activity, actuation likely inhibited population outgrowth on the devices. However, we saw no statistical difference in the percentage of EdU positive cells due to actuation for 3d in 231 or HME2 samples, with the majority of cells in both static and actuating conditions entering S phase within 11 h (Figure S5, Supporting Information). Because EdU is incorporated when cells enter S phase, this may indicate that mechanical stimulation does not slow the length of time it takes a cell to enter the S phase, but prolongs the length of the S, G2, or M phases. However, further cell cycle analyses are required to fully understand the mechanisms of reduced cell viability due to mechanical stretching.^{[49, 50]}

YAP and its paralog, TAZ are well-known regulators of the Hippo pathway and relay mechanical information into biochemical response in a process known as mechanotransduction. To date, the majority of studies exploring mechanotransduction and YAP/TAZ activation have solely investigated nontumorigenic cells. Wang et al. showed that cyclic stretching of smooth muscle cells on Matrigel-coated silicone chambers (0.5 Hz, 13% strain) for 24 h markedly increased YAP/TAZ activity and proliferation rates.^[51] Cui et al. also showed that proliferation rates were significantly reduced in primary fibroblasts cultured on a soft substrate as compared to PDMS pillars, but the growth rate was recovered under dynamic stretching (0.1 Hz, 5% strain).^[52] We analyzed YAP colocalization to the nucleus or cytoplasm after 24 h in culture, where a larger ratio of YAP in the nucleus indicates more YAP activation in the cell.^[3] We found higher YAP colocalization into the nucleus in static cultures for all cell types except for the 231 cells (Figure 4a–c). Therefore, cyclic stretching of these cancerous cells does not seem to induce the same effect as stretching nontumorigenic cells, with significant inhibition of YAP and cell viability established after 2 d in culture (~1.5 d actuation) for some cell types. This may be due to cancer cells’ already enhanced ability to resist apoptosis, such that they are instead pushed into a state of quiescence or dormancy.^[3,53–55] In addition, the PDMS actuator body and the FN fibrils are both stiffer than the hydrogels typically utilized in mechanical loading studies. Our FE model estimates the FN as a nonlinear strain stiffening material with an equivalent modulus of >30 KPa, which is above the canonical threshold for classifying a stiff material.^[56,57] This difference may explain why static HLF cells also displayed high levels of YAP nuclearization and why mechanical stimulation did not cause an increase in YAP activation typically reported in the field.^{[51, 52]} Yadav et al. also noted apoptosis of BC cells after 4 h of cyclic stretching.^{[58, 59]} However, as this was performed on a 2D synthetic substrate, it is possible that the 3D FN environment partially protected the cells, reducing metabolic activity and/or proliferation, but not inducing apoptosis.

Figure 4.

Morphological and mechanotransduction assessment. Immunofluorescence of cells after 24 h in a) static or b) actuating conditions (n=3). Scale bar is 50 μm. c) Arrows indicate an example of loss of nuclear YAP due to actuation. Scale bar is 50 μm. d) The ratio of YAP in the nucleus or cytoplasm, averaged for every cell imaged per condition. e) The weighted average cluster size in each condition. f) The circularity of cells that were not in clusters per condition. 1 denotes a perfect circle. (mean ± s.d., * = p<0.05, ** = p<0.0005, *** = p<0.0001)

We next investigated the morphology of the cells by analyzing the weighted average cluster size and the circularity of single cells. Morphological and cluster size changes are well established indicators of epithelial-mesenchymal transition (EMT) in BC, where cells with an elongated spindle-like body and higher migration are considered more mesenchymal-like.^[2] Ca1a cells had a higher average cluster size in actuating conditions compared to static conditions. Conversely, the 231, HME2, and HLF cells had smaller cell clusters on average in actuating conditions compared to their static counterparts (Figure 4d). Actuation increased the maximum cluster size seen from 8 to 14 cells in Ca1a samples and decreased the maximum cluster size seen from 23 to 20 cells in 231 samples, 90 to 30 cells in HME2 samples, and 96 to 86 cells in HLF samples (Figure S7, Supporting Information). Similarly, the number of single cells across the FN region of interest decreased in Ca1a cells under actuation, but increased 1.9–2.6 fold for 231, HME2, and HLF cells in actuating compared to static cultures (Figure S7, Supporting Information). Single cells were then segmented, and their circularity was measured on a scale from zero to one with one being perfectly circular. Only the morphology of the mesenchymal cells was significantly affected by the mechanical stimulation. 231 and HLF cells both became more circular in actuating conditions (Figure 4e).

2.3. Matrix Degradation

Tumor cells actively degrade the extracellular matrix during metastasis.^[2] We measured the effect of actuation on the integrity of the suspended FN fibrils. After a week in culture under static or actuating conditions, we measured the deflection angle of the cantilever in response to increasing magnetic field strengths. The measurement of cell-free conditions assured that a week of cyclic actuation did not inherently alter the mechanical properties of the FN band. The actuators from all four cell lines exhibited higher deflection angles in static conditions compared to cell-free controls, suggesting a breakdown of the matrix by static cells. (Figure S7a, Supporting Information) This breakdown also caused the FN to completely tear in some of the actuators during the deflection test (Figure 5a).

Figure 5.

Matrix degradation under cyclic actuation. a) Deflection vs applied magnetic field of 231, Ca1a, HLF, and HME2 cells as well as no-cell controls after a week of static (red) or actuating (black) culture. Fibronectin rupture is pictorially indicated. Scale bar is 1 mm (n=3, mean ± s.d.). b) Pliability of the fibronectin band for each actuating cell line. HME2 culturing significantly weakened the fibronectin band compared to every other condition (* = p<0.05, n=3). c) Deflection angle at the maximum applied magnetic field in measurement setup (29 kA/m) (n=3, mean ± s.d., * = p<0.05). d) Representative strain contour for two conditions simulated comparing the effect of cluster size. Logarithmic strain is calculated with respect to a local coordinate along the fibronectin scaffold length. e) Average strain of the fibronectin scaffolds as the mechanical properties are weakened by the presence of cell clusters of different size and occupying overall different area fractions of the scaffolds.

Interestingly, actuators tested with HLF cells under cyclic stretching were stiffer than cell-free actuating controls, indicating that they formed a more robust matrix than the initial FN coating (Figure 5b, and Figure S7a, Supporting Information). Fibroblasts are known to reorient the FN matrix under mechanical stimulation, creating an organized and aligned fibrillar structure.^[20] This anisotropic organization likely strengthened the FN in the direction of the strain. For 231 and Ca1a BC cells, cyclic stretching inhibited matrix breakdown (Figure 5a). Specifically, the deflection angle at the highest magnetic field strength was significantly smaller for actuated 231 and HLF samples compared to their static counterparts (Figure 5c). However, the HME2 cells were unaffected by the mechanical stimulation, achieving almost complete degradation of the FN in both static and actuating conditions (Figure 5a–c). To further compare FN integrity between actuated cell lines, we calculated the pliability of the FN by measuring the slope of each of the deflection tests. The HME2 group had a significantly higher deflection angle per unit of applied magnetic field strength compared to all the other cell lines, and the responses of the other three cell lines were not significantly different than the cell-free control (Figure 5b). Of note, the matrix breakdown was also only halted in the 231, Ca1a, and HLF cells in the high strain (20–25%) cases (Figure S7a–c, Supporting Information). In the current actuator design, we observed deviations in deflection, and subsequently the strain profile, for a given magnetic field strength (Figure 1f, Figure 5a). Therefore, direct comparison between groups undergoing high- or low-strain regimes was limited. Nevertheless, the statistically significant results when comparing the actuating and static groups highlights a clear effect of cyclical mechanical force on cultured breast cancer cells.

We then utilized the FE model to evaluate the different mechanical weakening caused by the three BC cell lines as a function of cell cluster size and total occupied area. Although the number of cells seeded onto the actuator was held constant, each cell type has a unique cell size and propensity to cluster into fewer larger clusters or many smaller clusters, as described in Figure 4. Specifically, HME2 cells had significantly larger cell clusters than the other two BC cell lines (Figure 4e). Therefore, we simulated the strain in the FN band under 27 environments, combining three cluster sizes, three total area fractions, and three matrix weakening percentages (Figure 5d–e).

Clusters of 2500 μm², 5000 μm², and 7500 μm² were evaluated. Based on the size of a single cell and the average number of cells per cluster quantified from immunofluorescent imaging, the Ca1a cells had clusters of ~900 μm², the 231 cells had clusters of ~1300 μm², and the E2 cells had clusters of ~6000 μm². However, it is important to note that these experimental values were obtained after only 24 h of actuation. The total area fraction is the total percent of the area that would be occupied by cells. The evaluated total area fractions were 10%, 25%, and 50%. The mechanical properties at locations with cells were weakened by 10%, 25%, or 50% of the original FN mechanical properties, obtained through the previous characterization (Figure 2). Our simulations indicate that the effect of cluster size on the average strain is minimal when the total area fraction is small, but may play an important role in the differential response seen as the area fraction and the weakening per cell increases (Figure 5 d–e). During the deflection analysis, the larger defects caused by the HME2 cells would likely result in a structurally weaker FN band than the multiple, smaller defects caused by the 231 and Ca1a cells. Further analysis is currently ongoing to investigate additional cellular effects of cyclic stretching in metastatic breast cancer.

3. Conclusion

In conclusion, we present a novel platform that can incorporate cyclic stretching on a biologically-relevant 3D FN culture to create a biomimetic lung model. We demonstrate a dramatic reduction in global metabolic activity in three BC cell lines due to actuation that is not recapitulated in human lung fibroblasts. Moreover, we demonstrate the role of mechanical stimulation on YAP downregulation and matrix degradation rates in tumorigenic cells. This system enables high-throughput experimentation in standard culturing equipment while utilizing a 3D physiologically-relevant environment. Importantly, the culturing of cells on FN fibrils supported by the PDMS actuator body allows for all traditional assays such as fluorometric/colorimetric metabolic activity measurements, removal of cells for flow cytometry and immunoblotting, and immunofluorescent staining and imaging. These assays often require substantial optimization in other 3D culturing platforms because the cells are fully embedded, which interferes with cell removal techniques and/or reagent penetration and removal.

The findings presented here indicate a clear cellular response due to mechanical strain and present a platform that may help the field further investigate the dynamic transition from micro- to macrometastases, including the dormancy and reactivation of latent disseminated BC cells in metastatic niches. Taken together, our work suggests that mechanical force in the lungs is a potential suppressor of BC metastatic outgrowth and demonstrates the need for robust, physiologically-relevant studies that incorporate cyclic stretching. However, the main limitation of the current design is the out-of-plane motion which makes real-time imaging during deflection extremely difficult. Future device design iterations will benefit from establishing a suspended FN network that is stretched in plane and biaxially.

In addition, the current device design has a limited strain resolution due to variation in the fabrication technique, the FN coating process, and the deflection measurement technique. It may be possible to better control the amount of applied strain in the future by characterizing the deflection of each FN-coated actuator prior to experiment. Having an array of microscale actuators with different compliances may also help improve the range of applied strain as well. Furthermore, embedding a strain sensing element, such as a piezoresistive conductive trace, would be more accurate than current deflection measurements and would enable real-time adjustment of the magnetic field strength to maintain a certain strain amplitude even during matrix remodeling by the cells.^[60] We further acknowledge that when designing actuator arrays of varying sizes, a higher density of magnets in the array would induce a larger magnetic gradient that may alter the actuator motion. As we further miniaturize the platform to increase the throughput of this system, it will be crucial to carefully consider the interaction between neighboring magnets, which may fundamentally limit the overall scalability of this approach.

Lastly, future experiments in quantifying the cell biology will likely be expediated by the creation of a larger FN region of interest and an increased throughput to use the maximum capacity of commercial well plates. There is not currently a consensus on the effect of static or alternating magnetic fields on cells.^[61–64] Here, cells on non-actuated control devices were exposed to a static magnetic field from the embedded permanent magnet, but were not exposed to the time-varying magnetic field. Although we do not anticipate that the static or dynamic magnetic fields will affect breast cancer cell behavior, this is worth investigating in future studies which aim to more deeply explore the metabolic activity, matrix deposition, and mechanotransduction of metastatic cancer cells. This can be accomplished by including an additional control group without a magnet in the actuator, but that is placed on the actuating platform to quantify any changes in cellular behavior that occur due to the application of time-varying magnetic fields.

4. Experimental Section

Actuator Preparation:

3D printed molds (Autodesk Ember, San Rafael, CA) were used to cure 10:1 PDMS formulation (Dow Corning Sylgard 184, Midland, MI). The PDMS was cured in an oven for 2 h at 100°C. After curing the PDMS, the device was removed from the mold, and a 2-mm-wide, 1-mm-thick N42 NdFeB permanent magnet (KJ Magnetics, Pipersville, PA) was placed in the reservoir and then coated with PDMS so that the magnet imposed no cytotoxicity. The devices were left in the oven at 70 °C for 24 h to fully cure. The actuators were sterilized by soaking in 70% ethanol, rinsed three times with phosphate buffered saline (PBS), and left under UV in a laminar flow cabinet (30W) for 1–2 h. After sterilization, to suspend the physiological substrate on the region of interest, the actuators were placed in a 100 μg/mL solution of FN suspended at the air-water interface. The actuators were rotated for 2 h on a rotisserie shaker (Barnstead Labquake, Lake Balboa, CA) and maintained at 30°C at 8 RPM.^{[44, 45]} A FN network formed following the device geometry from the magnet reservoir at the end of the cantilever to the edge of the frame, 1.8 mm in total length and 76.7 ± 9.0 μm thick. The geometry of the FN network that forms is dictated by several factors, including the shape and material of the device, the concentration of FN solution used, and the length of time rotating. Using the parameters above, the average FN network formed was 1.73 mm wide at the magnetic reservoir, 3 mm wide across the middle, and 4.27 mm wide at the base where it attached to the inner edge of the device frame. Devices were checked using brightfield microscopy to ensure a robust coating before experimentation.

Magnetic Actuation Platform Fabrication & Characterization:

In order to actuate the magnetic PDMS devices and overcome some of the challenges presented by other cell stretching systems, we developed a platform to easily integrate a 24-well culture plate for 3D cell study. The platform consists of a linear actuator (Actuonix, Alberta, BC, Canada) programmed to move an array of permanent magnets of 19 mm in diameter and 2.4 mm in height in a fixture that suspends the multi-well culture plate (Figure 1b). This linear actuator can reach up to 0.6 Hz for the high strain amplitude application and 0.8 Hz for the low strain amplitude group; this range can be expanded by using other linear actuators on the market. Each of the actuators is fixed by the weight of a custom circular 304 stainless steel frame that has 15.4 mm outer diameter, 9.2 mm inner diameter, and 12.9 mm in height to allow free deflection of the cantilever. This prevented reorientation of the actuators when exposed to the magnetic field in the culturing well. The fixture did not affect the magnetic field strength required to achieve the desired deflection angle of the cantilever.

A computational model assessed the distribution of the magnetic field in two different array configurations (Figure S1a,b, Supporting Information). The most equal distribution of the magnetic field was found to be the alternating magnet array allowing for 12 permanent magnets in total. The magnetic field distribution of the 12-magnet permanent magnet array was determined experimentally via a hall effect sensor. Each of the magnets on the platform was measured at 3 mm increments along the z-axis. The transient magnetic field was characterized in the same manner in three regions of the platform.

Cell Seeding & Viability Studies:

Devices were placed on the actuating platform briefly before cell seeding to ensure that each had a robust coating of FN. 70,000 cells were then seeded at high density in media (40–100 μl) onto the suspended FN on the actuators in non-adherent 24-well plates. Media used was DMEM/high glucose with 10% FBS and 1% penicillin-streptomycin. The HME2 parental line additionally had 1% insulin. After 2 h, additional media was added to cover the devices (80–1000 μl). After an additional 5 h, devices were moved to new wells for metabolic testing. Upon completion, half were placed on the platform to begin cyclic stretching at 5–10% strain and 0.3 Hz. Actuating devices remained on the platform for 7 d. Control devices remained static for the same duration. A resazurin-based metabolic assay (Sigma TOX8–1KT) was performed on days 0, 2, 4, and 6. Briefly, media was removed and 10% resazurin stock solution (1 ml, diluted in phenol-red free completed media as described above) was added to each well. After 2 h, 3 supernatant samples from each well were read using fluorescence (excitation 560 nm, emission 590 nm). The remaining solution was removed completely and replaced with completed media. The 231 cells were additionally assessed at 20–25% strain on days 0, 1, 3, and 5 following the same procedure. The entire media volume was collected in actuators undergoing high-strain tensile testing after 24 h of actuation (n=3 per cell type). Cells in the media were counted in triplicate using the LUNA-FL Dual Fluorescence automated cell counter (Logos Biosystems, Annadale, VA) and extrapolated to determine the percentage of cells potentially dislodged by the mechanical force (knock-off percentage).

Additional viability studies were performed using EdU and AOPI in a subset of cells. Actuators were seeded as described above but did not undergo metabolic activity testing. The Click-iT EdU Cell Proliferation Kit for Imaging (Invitrogen) was used to detect cells that entered S phase. EdU (5-ethynyl-2′-deoxyuridine) was incubated for 11 h during day 3 of actuation or static culture. Cells were fixed and the EdU reaction was completed according to manufacturer’s instructions. The cell nuclei were stained using DAPI (1:500 dilution of 0.1 mg ml⁻¹ stock). Separate 231 actuators were stained using Cellometer ViaStain AOPI staining solution (acridine orange/propidium iodide) after 6 d of high strain actuation to differentiate live and dead cells. EdU and live/dead images were taken on a Zeiss LSM 880 confocal microscope.