MDA-MB-231 parental (MDA-P), brain-seeking (MDA-BR), and bone-seeking (MDA-BO) clones were provided by Dr. Toshiyuki Yoneda in Osaka, Japan. Organ-seeking clones were developed in his lab using a protocol described previously (Yoneda et al., 2001). Briefly, MDA-MB-231 cells were injected into mice and allowed to metastasize to the brain or bone. This process was repeated until a population that solely metastasizes to either the brain or the bone was established. Upon arriving in our lab, the cells were checked for authenticity using STR testing (Laragen, Inc.), and all three cell lines best matched the MDA-MB-231 line (see Supplementary Material for full results provided by Laragen, Inc.). The cells were cultured in medium consisting of Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with high glucose (ThermoFisher Scientific, Waltham, MA, United States), 10% Fetal Bovine Serum (FBS; HyClone Characterized GE Healthcare, Pittsburgh, PA, United States or ThermoFisher Scientific), and 1% Penicillin-Streptomycin 10,000 U/ml. Cells were washed with Phosphate-Buffered Saline (PBS) (VWR, Radnor, PA, United States), and detached with 0.25% Trypsin-EDTA (ThermoFisher Scientific). Cells were cultured at 37°C, 50% humidity, and 5% CO₂:95% air. Cells were passaged every 2 days (up to passage 12) when about 80% confluency was reached. Unless otherwise noted, trials from different passages were pooled, while Figure 2 specifically assessed passage-dependence.

To assess the morphology of suspended (non-adherent) cells, a total of 4 × 10⁵ cells were plated into 6-well non-tissue culture treated plates (VWR) and imaged immediately. Ten-micron diameter (9.94 ± 1.01 µm according to manufacturer for specific batch) Envy Green fluorescent beads (Bangs Laboratories, Inc., Fishers, IN, United States) were suspended in blank wells of the same plates to validate imaging and analysis based on the known diameter. For cells in suspension (and beads), images were modified in ImageJ for maximum brightness and contrast to achieve the sharpest contours. The image was then converted to binary black and white after thresholding. The ImageJ function to fill holes was applied and the ImageJ built in particle analyzer was used to analyze particles 300–1,250 pixels² in area with circularity values of 0.65–1 to eliminate any aggregates from analysis. To assess the morphology of adherent cells, 1 × 10⁵ cells were plated into 6-well tissue culture treated plates (VWR) and allowed to attach overnight. Phase contrast imaging was performed using a 20x objective. ImageJ was used to process images and quantify area, inverse aspect ratio, circularity, and solidity. Inverse aspect ratio is defined as the ratio of the minor axis to the major axis. Solidity is defined as the ratio of the area to the convex area. Circularity is the ratio of 4 π A to P ² , where A is the projected area of the cell in the image, and p is the perimeter of the cell.

For 2D cell migration experiments, cells were plated onto substrates coated with collagen I from rat tail (Sigma Aldrich, St. Louis), fibronectin from human plasma (Sigma Aldrich), or poly-d-lysine hydrobromide (PDL) (Sigma Aldrich). Collagen I and fibronectin were dissolved in PBS, and PDL was dissolved in sterile milliQ water. Twenty-four well glass bottom plates (MatTek, Ashland, MA) were coated with 300 µl (per well) of 20 μg/ml of the varying protein solutions and incubated for 1 h at 37°C. Following incubation, wells with collagen and fibronectin were washed 3 times with PBS, while wells with PDL were washed 3 times with sterile milliQ water. After washing the substrates, 1 × 10⁴ cells were plated into each well. Cells were imaged and analyzed as described below.

Microchannel devices were fabricated as we previously described in detail (Balzer et al., 2012; Doolin and Stroka, 2018). Once fabricated, the polydimethylsiloxane (PDMS) devices were sonicated in 100% ethanol (Pharmco) for 5 min. The PDMS devices and 35 mm by 75 mm glass coverslips were rinsed with deionized water, followed with 100% ethanol, and then dried thoroughly with pressurized air. Both the PDMS devices and coverslips were plasma-treated using a plasma cleaner (Harrick Plasma, Ithaca, NY, United States) and pressed together for irreversible bonding. 40 μl of type I rat tail collagen (20 μg/ml) solution was pipetted into each inlet and outlet of the device, the device was incubated for 1 h at 37°C, and then the device was washed two times by pipetting PBS into and out of all wells of the device. Meanwhile, cells growing in culture were washed with PBS, detached with 0.25% Trypsin-EDTA (ThermoFisher Scientific), centrifuged, and resuspended to a final concentration of 1 × 10⁵ cells/25 µl. Twenty-five microliters of cell suspension were added to the cell inlet, and the device was incubated for 5 min at 37°C for initial cell seeding. Excess liquid was removed from the cell inlet and outlet, and 50 µl of serum-free media was added to the bottom inlet, along with the 2 lower inlets of the upper channel. Fifty microliters of serum-full media were added to the top-most inlet of the main upper channel.

Polyacrylamide (PA) gels (∼80 μm thick) were formed using a method initially described by Wang and Pelham (Wang and Pelham, 1998) and described in our previous publications (Stroka and Aranda-Espinoza, 2009; Stroka and Aranda-Espinoza, 2011; Stroka et al., 2012; Hamilla et al., 2014; Gray et al., 2019) on 22 × 22 mm coverslips (ThermoFisher Scientific). PA gels were coated with 50 μg/ml collagen I using sulfo-SANPAH activation, also as previously described in our work (Stroka and Aranda-Espinoza, 2009; Stroka and Aranda-Espinoza, 2011; Stroka et al., 2012; Hamilla et al., 2014; Gray et al., 2019). A total of 1 × 10⁵ cells was plated onto each gel. For AFM experiments on gels, coverslips containing gels were secured to 50 mm AFM grade glass bottom dishes (VWR).

The Young’s modulus of tumor cells on PA gels (as described in the section above) or on glass (50 mm AFM grade glass bottom dishes, VWR) was measured using atomic force microscopy (AFM). After 1 day in culture, AFM was performed on live cells with the stage heated to 37°C. AFM was performed using an Asylum MFP-3D-BIO Atomic Force Microscope with TR400PB(L) probes (Asylum Research) as previously described (Gray et al., 2019). Asylum’s “Get Real” approach was used to measure the spring constant and inverse optical lever sensitivity of TR400PB(L) probes via the Sader method and thermal noise method, respectively. The average spring constant of the cantilevers was within a factor of 1.33 to Asylum’s nominal value of 0.02 N/m and all within the nominal range of 0.01–0.05 N/m. One 100-curve force map covering a 5 μm² area was collected for each cell using a 2 μm force distance, a 1 V trigger point (approximately 1.35 nN), and a scan rate of 0.99 Hz. The Hertz model was used to fit the force curves within Asylum’s Igor Pro-based software using the equation F =   4 3 ∗ ( E 1 − υ 2 ) ∗ r ∗ δ 3 / 2 , where δ is the measured indentation of the sample and the Young’s modulus E was the fitting parameter. The Poisson’s ratio υ of the sample was assumed to be 0.45 and the tip radius of curvature r was approximately 30 nm. Three biological replicates were performed in which three cells per condition per trial were measured (n = 9 cells total, or 900 force curves per condition).

Images were acquired on an Olympus IX83 microscope (Olympus, Center Valley, PA, United States) using a 10x objective. To maintain the cell viability during imaging, a chamber calibrated to 37°C, 50% humidity, and 5% CO₂:95% air was used on the microscope stage. Images were taken at 5-min intervals (as specified in figure captions). On the following day, a collection of phase contrast images was taken using a 20x objective. 2D and confined cell migration tracking parameters were determined as described previously (Shumakovich et al., 2017). Cells were tracked over 390–750 min for 2D assays and approximately 720 min for confinement assays. Cells that were migrating out of the frame or dividing were not tracked. Mean squared displacement (MSD) was calculated as previously described (Stroka and Aranda-Espinoza, 2009; Shumakovich et al., 2017). Briefly, the average of the square of the distance traveled was calculated as the MSD between each pair of points. The MSDs were averaged for every time interval and plotted vs. the corresponding time interval. The diffusion coefficient was acquired by fitting MSD vs. time plots to the Langevin-type equation r 2   =   4 D [ t − π ( 1 − e − 1 / τ ) ] , where D is the diffusion coefficient, r 2 is the MSD, t is time, and τ is the persistence time.

Glass-bottom dishes were coated with 20 μg/ml collagen I or a 4:1 ratio of 20 μg/ml collagen IV and 20 μg/ml fibronectin and incubated at 37°C for 1 hour, and then washed at least two times with PBS. A total of 1 × 10⁴ cells were plated into the dishes and incubated overnight. The next day, cells were fixed with 3.7% formaldehyde (Millipore Sigma) for 10 min at room temperature. Following three 5-min PBS washes, the cells were permeabilized with 1% Triton-X 100 (Millipore Sigma) at room temperature. The cells were washed 2 times with PBS for 5 min each. The samples were blocked in 2.5% bovine serum albumin (Millipore Sigma) for 1 h at room temperature. Phospho-Paxillin (Tyr118) antibody (Rabbit, monoclonal antibody, Cell Signaling Technology #2541) or Phospho-FAK (Tyr 397) (Rabbit, monoclonal antibody, and Invitrogen #700255) was diluted with 1% bovine serum albumin at a 1:100 ratio and added to the cells. The primary antibodies were incubated with cells overnight at 4°C and samples were washed twice the next day with PBS for 5 min per wash. The samples were then blocked with 2.5% BSA for 1 h at room temperature. Cells were then incubated with 1:500 phalloidin (Alexa Fluor 488, Life Technologies #A12379), 1:2,500 Hoechst (Hoechst 33,342, trihydrochloride, trihydrate, Life Technologies #H3570, and 1:200 secondary antibody (Alexa Fluor 568 goat anti-rabbit (H + L) #A11011) for 1 h at room temperature. Finally, samples were washed 3 times with PBS for 5 min each and stored at 4°C until imaging.

Fluorescent images were acquired using a FV 3000 RS Olympus Laser Scanning Confocal Microscope (Olympus, Center Valley, PA, United States) using a 60x oil objective. Vertical Z-stack images were taken with the appropriate filters using the FluoView FV31S-SW software. Image brightness was adjusted separately for each channel to optimize visibility, but adjustments were done consistently across image sets. For PY-paxillin and phospho-FAK analysis, ImageJ was used to analyze focal adhesion size and count using images at the cell-substrate interface. Images were converted to 8–bit and background was subtracted at 50 pixels. Auto threshold was determined by the software and the “analyze particles” plug-in was used to determine focal adhesion area and count. A minimum focal adhesion size of 4.14 μm was considered to prevent counting background staining. CellProfiler, a cell image analysis software, and was used to calculate the total cell areas from the actin images. Focal adhesion density was calculated by dividing the number of focal adhesions in a cell image by the total cell area in the corresponding image.

GraphPad (GraphPad, San Diego, CA) was used for all statistical analyses. Normality was tested using the D’Agostino-Pearson normality test (Supplementary Table S1). If the data were found to follow or partially follow (e.g., where some groups in the set were normally distributed, while some were not) a normal distribution, a one-way ANOVA or two-way ANOVA was performed to determine statistical significance. If the data did not follow a normal distribution, a Kruskal Wallis test followed by a Dunn’s multiple comparison was used. Statistical significance was determined by using a cutoff of p value less than 0.05. At least three independent trials were conducted for each experiment. Cells from all three experiments were pooled between passages 6–12 for all experiments. A ROUT test was used to identify outliers in the focal adhesion expression data set. The data with outliers excluded were used to perform the statistical analysis as previously described.

Breast tumor cells circulate in a suspended state in the blood stream during the metastatic cascade, prior to adhesion to the vascular endothelium (Mentis et al., 2020). Circulating tumor cells are associated with increased risk of metastasis and lower progression-free and overall survival in breast cancer patients (Bidard et al., 2014; Riebensahm et al., 2019; Cristofanilli et al., 2019; Park et al., 2020). We first questioned whether the three breast tumor cell clones had different sizes in suspension. First, to validate morphological analysis of cells in suspension, we added a suspension of 10 µm diameter beads to a glass-bottom dish, imaged the beads using phase contrast microscopy, and used the analyze particles plugin in ImageJ to characterize size of the beads (Figure 1A). Bead diameters using our method were measured to be 11.8 ± 1.25 µm, while bead area was 108.4 ± 16.1 µm². Hence, we moved on to evaluate the size of breast tumor cells in suspension by plating parental and organ-seeking clones into non-treated tissue culture plates and imaging the cells using phase contrast microscopy (Figure 1A). MDA-BO cells were found to have a smaller median diameter (Figure 1B) and area (Figure 1C) when compared to the MDA-BR and MDA-P cells. It is important to note that the cell populations are highly heterogenous in size, even in suspension, as evidenced by the relatively large interquartile ranges of the data (Figures 1B,C). It is possible that differences in cell cycle stage may have contributed to the heterogeneity of suspended cell size, and/or that cells, even in the same cell cycle phase, are innately heterogeneous in size.

To evaluate the effects of cell passage on the morphology of adherent organ-seeking clones, we plated cells from different passages onto tissue culture treated 6-well plates. Cells were imaged using phase contrast microscopy (Figure 2A), and cell morphological features, including area (Figure 2B), inverse aspect ratio (Figure 2C), circularity (Figure 2D), and solidity (Figure 2E) were analyzed. We found that, for all passages the area, circularity, and solidity of the MDA-BO cells were significantly smaller than the MDA-P cells (Figure 2B). Further, the MDA-BO cells significantly decreased in area, circularity, and solidity (Figures 2B,D,E) between passages 6 and 12. This same trend was not seen in the MDA-P and MDA-BR cells. Inverse aspect ratio (a proxy for cell elongation) was mostly consistent across passages and cell clones (Figure 2C). Overall, these qualitative and quantitative results indicate that the MDA-BO cells are smaller, less circular, and more protrusive than their MDA-P or MDA-BR cell counterparts, and that these morphological features are modestly altered over passage.

Morphological alterations in attached and suspended breast tumor cell clones motivated us to evaluate random migration (i.e., chemokinesis) on three different substrate coatings–collagen I, fibronectin, and poly-d-lysine (PDL). Collagen I is an abundant protein in the bone extracellular matrix (source), while fibronectin is found in the brain basement membrane and extracellular matrix (source); both promote cell integrin-based adhesions. PDL promotes electrostatic interactions between cells and the substrate, thus presumably blocking integrin-based binding. Qualitative inspection of phase contrast images suggested that MDA-BO cells were more protrusive (less solid) on collagen I and on fibronectin (Figure 3A). Meanwhile, during chemokinesis, there were no statistically significant differences in migration speed between cell types on collagen I or fibronectin. On PDL, all cell types were migrating significantly slower compared to their corresponding cell types on collagen I substrate, while only the MDA-P and MDA-BO cells were moving slower compared to a fibronectin substrate (Figure 3B). Meanwhile, on collagen I-coated substrates (but not fibronectin), MDA-BO cells had a higher mean squared displacement (MSD) over time, and diffusion coefficient compared to MDA-BR and MDP-P cells (Figures 3C–F). These results suggest that even though the MDA-BO speed was not different from the other clones, the MDA-BO cells covered a larger area over time while migrating, and hence were more persistent. Because we observed the largest differences in 2D chemokinetic migration between cell clones on collagen I, the subsequent confined migration, atomic force microscopy, substrate stiffness, and focal adhesion experiments used collagen I as the substrate.

Breast tumor cells experience confinement throughout the metastatic cascade, in migrating through ECMs, during intravasation and extravasation, and while migrating along anatomical features (Balzer et al., 2012; Stroka and Konstantopoulos, 2013). To mimic confinement, microchannel devices with widths varying from 3 to 50 µm were used to evaluate cell migration speed and chemotactic index (CI) in response to a chemotactic gradient (FBS) (Figure 4A). MDA-P and MDA-BO cells generally migrated more persistently (Figures 4B–D) and slower (Figures 4E–G) in highly confined (3, 6, and 10 µm) channels. However, interestingly, that same trend was not observed in the MDA-BR cells, which had similar speeds and persistency across all channel widths.

Because the brain and bone-seeking clones target tissues with unique matrix stiffnesses, we hypothesized that one method these organ-seeking cells may use to determine a preferential environment is by sensing the environment’s extracellular matrix stiffness. Further, we wanted to explore whether these clones altered their stiffness in response to the environment, perhaps as a mechanical mechanism that enhances their survival. We compared cells on collagen I-coated glass, which is in the GPa range of stiffness (similar to mature bone tissue), and on collagen I-coated polyacrylamide gels of intermediate stiffness (∼194 kPa) or a stiffness similar to the brain microenvironment (∼1 kPa). The polyacrylamide gels were fabricated, coated with collagen I, and seeded with cells. Atomic force microscopy (AFM) was performed on the cells in force mapping mode and Young’s modulus was calculated by fitting the force vs. distance curves to the Hertz model, as described in the Methods section. MDA-BR and MDA-BO cells’ Young’s moduli on glass were modestly larger in comparison with the respective cell clone on soft 1 kPa gels (Figure 5B). In addition, the MDA-BR cells had a lower stiffness on the 1kPa compared to the 194 kPa (Figure 5B). There were no differences in Young’s modulus between the three cell types for any substrate stiffness.

Since we had previously observed morphological differences between the clones on tissue culture plastic, we quantified whether there were differences in their morphology in response to substrate stiffnesses mimicking either the brain or bone environment. Qualitatively, we observed that all cell types became less spread on soft (1 kPa) substrates (Figure 5A; Supplementary Table S2). We found that the MDA-BO clones had the largest area on both the stiff polyacrylamide gel (∼194 kPa) and soft polyacrylamide gel (∼1 kPa) but the smallest area on glass (Figure 5C) when compared to MDA-P and MDA-BR clones. Further, the MDA-BO clones had the lowest circularity and solidity on all substrates (Figures 5E,F; Supplementary Table S2). Meanwhile, the MDA-BR clones had similar areas to the MDA-P cells on the 1 kPa substrate but had minor differences in the inverse aspect ratio, circularity, and solidity (Figures 5D–F; Supplementary Table S2). Notably, the MDA-BR cells had higher areas compared to the MDA-P cells on both glass and 194 kPa. We note that another study comparing MDA-P and MDA-BR cells has reported a similar trend (Narkhede et al., 2018). Furthermore, both MDA-P and MDA-BR cells had decreasing areas and increasing inverse aspect ratios as stiffness decreased. Finally, all cell types displayed increasing solidity as substrate stiffness decreased.

A previous study has identified a subpopulation of MDA-MB-231s that has a high α5β1 integrin expression, which has been linked to having more prominent focal adhesion expression and higher invasion rates (Mierke et al., 2011). Additional studies have also shown that MDA-BO and MDA-BR cells have higher α5β1 integrin expression (McFarlane et al., 2015; Jahangiri et al., 2017). This motivated us to explore whether there were differences in focal adhesion phenotypes between the MDA-P, MDA-BO, and MDA-PR clones. We imaged PY-paxillin and phospho-FAK, along with F-actin, in cells on collagen I-coated substrates (Figure 6A). We then processed the images using FIJI and Cell Profiler to quantify the area, number of focal adhesions per cell, and density of focal adhesions marked by PY-paxillin or phospho-FAK (Figure 6B). On the collagen I substrate, MDA-BO had a higher PY-paxillin focal adhesion area, count per cell, and density when compared to the other two cell lines (Figure 6C). Furthermore, MDA-BO cells had higher area of phospho-FAK rich focal adhesions compared to the MDA-P & MDA-BR cells, while the focal adhesion count per cell and density were similar (Figure 6C).