Antibody reagents, growth factors, and inhibitors
Antibodies against EGFR, IGF1R, C-Met, and glyceraldehyde 3-phosphate dehydrogenase (GAPDH) were purchased from Cell Signaling Technology. EGF, PDGF-BB, and TGF-α were purchased from Invitrogen. IGF1, hepatocyte growth factor, heparin-binding EGF-like growth factor, and Heregulin β1 were purchased from PeproTech. EGF and IGF1 were used at 100 ng/ml, Heregulin β1 was used at 80 ng/ml, and all others were used at 50 ng/ml for all experiments. (S)-(-)-Blebbistatin, cytochalasin D, and nocodazole were purchased from Santa Cruz Biotechnology, Inc. and used at 50 µM, 25 nM, and 10 µM, respectively.
MDA-MB-231, BT-549, and MDA-MB-157 cells were cultured in high-glucose DME supplemented with 10% FBS and 1% penicillin–streptomycin. SUM-159 cells were cultured in Ham’s F12 media supplemented with 5 µg/ml insulin (Lonza), 1 µg/ml hydrocortisone (BD), 5% FBS, and 1% penicillin–streptomycin.
For 3D migration assessment, cells were labeled with CMPTX (Invitrogen) for 20 min and mixed with 2.2 mg/ml pH-neutralized, acid-extracted, nonpepsin digested collagen I (BD) with DME at 500,000 cells/ml. The matrix–cell solution was placed in a glass-bottom multiwell plate (MatTek), polymerized for 30 min at 37°C, and then overlaid with full serum media overnight. Cells were stimulated 4 h before imaging on an environment-controlled microscope (TE2000; Nikon) with a camera (C4742-95-12ERG; Hamamatsu Photonics) and a Plan Apochromat 10× 0.45 NA differential interference contrast (DIC) L air objective. Image stacks of 70 3-µm slices were obtained every 60 min for 16 h using MetaMorph (Molecular Devices). Where indicated, inhibitors were added simultaneously to stimulation. To avoid artifacts caused by potential gradients in stiffness near the edges of the gel, analysis fields were selected >200 µm from the glass surface.
For soft 2D migration assays, 100 µl pH-neutralized, acid-extracted 2.2 mg/ml collagen I or 100% matrigel was spread across wells of a 48-well plate and allowed to polymerize. For stiff 2D migration assays, either 100 µg/ml collagen I in 20 mM acetic acid or 0.2% matrigel in serum-free medium was used to coat uncoated glass multiwall plates for 30 min (MatTek). Cells were then labeled with CMFDA (Invitrogen) for 20 min and seeded sparsely on wells with matrix or directly on tissue-culture plastic. The next day, cells were stimulated 4 h before imaging every 10 min for 16 h. Where indicated, inhibitors were added simultaneously to stimulation.
Cells were tracked using Imaris (Bitplane). From each track, the RMS cell speed was calculated from position intervals between time points as well as the standard deviation of the mean. Total and net speeds were calculated by dividing the total path length and net displacement by the duration of the experiment. Each track was then fit to a random walk model using the method of nonoverlapping intervals as previously described to calculate the random motility coefficient (Kim et al., 2008
). In brief, individual cell speeds were calculated from each track using the mean of each time interval. Mean squared displacements were then calculated as the mean squared distance of all nonoverlapping intervals within a track. Using the mean squared displacement (MSD) for varying time intervals, the following equation was fit using least squares to yield a fit persistence:
From the fit, the random motility coefficient (RMC) was calculated as RMC = S2P, analogously to the diffusion coefficient (S, speed; P, persistence; t, time).
Glass-bottomed dishes (MatTek) were coated with 0.2% matrigel in serum-free media for 30 min. Cells were seeded sparsely overnight and then serum starved for 4 h in L15 media with 0.35% bovine serum albumin. Inhibitors, when indicated, were added at the beginning of serum starvation. DIC images were acquired every 10 s for 1 min before stimulation and 9 min after stimulation. Cell areas were traced immediately before stimulation and 9 min after stimulation using ImageJ (National Institutes of Health). Single-cell information was aggregated from at least three independent experiments.
Receptor expression and activation measurement
Cells were plated sparsely on 15-cm plates overnight, washed with PBS, and lysed with 500 µl of radioimmunoprecipitation assay buffer containing protease inhibitor (Roche) and phosphatase inhibitor cocktail (Boston BioProducts). Equal protein was loaded for SDS-PAGE analysis using a bicinchoninic acid assay and blotted using standard techniques with antibodies against GAPDH, EGFR, IGF1R, and C-Met. Densitometry was performed on an imager (Odyssey; LI-COR Biosciences) and normalized to GAPDH as a loading control.
For activation measurement, cells were seeded sparsely overnight and starved for 4 h the next day followed by stimulation with either 100 ng/ml EGF or 50 ng/ml TGF-α for 5 min. Cells were lysed using lysis buffer (Bio-Rad Laboratories) containing protease inhibitor (Roche) and phosphatase inhibitor cocktail (Boston BioProducts). EGFR pan-pY was measured using a bead-based ELISA assay (Bio-Rad Laboratories) loaded with equal protein using a bicinchoninic acid assay. Linearity of the assay with respect to protein concentration was verified.
All analysis was performed in MATLAB (MathWorks). Single-cell data from each experiment were imported, and each quantile of interest (50th, 90th, or 95th) was calculated for each motility metric and condition. Each set of growth factor conditions within a given quantile was then normalized to the condition absent of growth factor stimulation (or mean centered if indicated). The mean and standard error were then calculated from independent experiments. Comparisons of motility metrics between single growth factor conditions were performed using the Student’s t test or the Mann–Whitney test for single-cell data.
For single cell–based migration parameter clustering, Spearman correlation (ρ) of the parameters for individual cells from all experiments for a single cell line in 3D (n
≥ 3) were calculated and used to calculate pairwise distances, in which perfect correlation corresponds to a distance of zero, and anticorrelation corresponds to a distance of two (1 − ρ). Multidimensional scaling in two dimensions captured >99% of the distance quantities for all cell lines. The first principle component captured 90–95% of the distance quantities, whereas the second principle component captured 5–10%. Standard error was calculated by jackknife, during which each cell was removed separately, and each time, both the distances and scaling repeated (Efron and Gong, 1983
). For clustering of quantile-level migration metrics, each growth factor profile was mean centered and averaged across experimental replicates. Clustering of profiles and growth factors was performed by rank correlation and mean linkage.
Receptor expression and motility enhancement comparison were tested for significance using the binomial distribution. Where indicated, motility enhancement was variance normalized between cell lines by log transformation (so as to center fold-change values around zero) and by division by the standard deviation across growth factor conditions.
Growth factor profiles were compared by calculating the Spearman correlation, and significance was calculated by permutation. Where many comparisons were performed, multiple hypothesis testing was performed as indicated. When comparing migration in 2D and 3D contexts, multiple hypothesis correction was not performed, as type I error is not a concern. When comparing protrusion and 3D migration results, identical quantiles of single-cell data were always used.
Online supplemental material
Fig. S1 shows an example of fluorescent stacks from a 3D migration assay and preliminary analysis of the migration phenotypes. Fig. S2 shows the RMS speed measurements for the experiments shown in . Table S1 shows the subtype characteristics of each cell line examined and whether it was observed to migrate in 3D. Online supplemental material is available at http://www.jcb.org/cgi/content/full/jcb.201201003/DC1