Of the control. Persistence length P is normalized by the Triptorelin average persistence length of the control group. Persistence length along the gradient direction Px is normalized by subtracting that of the control group. Nonparametric t-test compared to the control group (Mann-Whitney test) within Prism (GraphPad Software, Inc., La Jolla, CA) was used for statistical analysis.Results and Discussion Microfluidic Device Setup, 3D Cell Culture and Dynamic Cell TrackingChemical gradients were generated using a recently developed microfluidic device (Fig. 1A ) [8,24]. Briefly, three parallel channels were patterned on a 1 mm thin agarose gel membrane. Medium containing either SDF-1a, EGF, or neither was introduced to each of two side channels, and allowed to form gradients by diffusion across the agarose gel ridges between the channels. Cell-seeded type I collagen was introduced into the center channel. This device has been recently characterized for its ability to generate stable and well-defined gradients both computationally and experimentally using FITC Title Loaded From File conjugated proteins [8,24], which was used to explore dendritic cell chemotaxis previously [8,24]. Various chemical gradients were generated in the center channel by changing the chemical concentrations in the source and sink channels. The chemical gradient is calculated as the difference of the chemical concentrations in sink and source channels divided by the distance between the source and sink channels (i.e. 900 mm). The average chemical concentration is computed as the average of the chemical concentrations in sink and the source channels. A uniformKD Scientific, Holliston, MA) and a syringe (3 ml, BD, Franklin Lakes, NJ).Imaging and Data AnalysisFor live cell imaging, the device was transferred onto an environmentally controlled microscope stage. The device was surrounded by a small Plexiglas chamber (120 mm675 mm645 mm) set at 37uC, 100 humidity and 5 CO2 (CO2-200, In Vivo Scientific, World Precision Instruments, Inc., Sarasota, FL). The microscope stage were surroundedRoles of Two Cytokines in Tumor Cell MigrationFigure 2. Plasticity and heterogeneity of tumor cell morphology and motility behavior. A . Bright field images of MDA-MB-231 cells embedded in 3D collagen matrices within the microfluidic channel at t = 0 and 8 h. Here, t = 0 is defined as the time when buffer and 100 nM SDF-1a solution were introduced into the two side channels. Cells were pre-incubated for 24 hours after seeding before the introduction of the gradients. C. Graphical description of cell speed U’, cell velocity along the gradient direction Vx ‘ , persistence length P’, and gradient-directed persistence length Px ‘ . D. Graphical description of aspect ratio. Distribution of cell aspect ratios at t = 0 and 8 h. E. Distribution of cell speed of elongated cells (aspect ratio larger than 3) and amoeboid-like cells (aspect ratio less than 3). doi:10.1371/journal.pone.0068422.gconcentration in the center channel is generated by supplying the same chemical concentration solution in both side channels. The 3D cell culture consisted of 106 tumor cells/ml embedded in type I collagen extracted from 23977191 rat tail [34]. Different concentrations (0.15, 0.25 and 0.35 ) of collagen were compared, and we measured the average cell speed to be 0.4260.02, 0.3060.02, and 0.2660.02 mm/min, respectively. We therefore used 1.5 mg/ml collagen for all the experiments shown here. In our culture model, cells remained evenly distributed after.Of the control. Persistence length P is normalized by the average persistence length of the control group. Persistence length along the gradient direction Px is normalized by subtracting that of the control group. Nonparametric t-test compared to the control group (Mann-Whitney test) within Prism (GraphPad Software, Inc., La Jolla, CA) was used for statistical analysis.Results and Discussion Microfluidic Device Setup, 3D Cell Culture and Dynamic Cell TrackingChemical gradients were generated using a recently developed microfluidic device (Fig. 1A ) [8,24]. Briefly, three parallel channels were patterned on a 1 mm thin agarose gel membrane. Medium containing either SDF-1a, EGF, or neither was introduced to each of two side channels, and allowed to form gradients by diffusion across the agarose gel ridges between the channels. Cell-seeded type I collagen was introduced into the center channel. This device has been recently characterized for its ability to generate stable and well-defined gradients both computationally and experimentally using FITC conjugated proteins [8,24], which was used to explore dendritic cell chemotaxis previously [8,24]. Various chemical gradients were generated in the center channel by changing the chemical concentrations in the source and sink channels. The chemical gradient is calculated as the difference of the chemical concentrations in sink and source channels divided by the distance between the source and sink channels (i.e. 900 mm). The average chemical concentration is computed as the average of the chemical concentrations in sink and the source channels. A uniformKD Scientific, Holliston, MA) and a syringe (3 ml, BD, Franklin Lakes, NJ).Imaging and Data AnalysisFor live cell imaging, the device was transferred onto an environmentally controlled microscope stage. The device was surrounded by a small Plexiglas chamber (120 mm675 mm645 mm) set at 37uC, 100 humidity and 5 CO2 (CO2-200, In Vivo Scientific, World Precision Instruments, Inc., Sarasota, FL). The microscope stage were surroundedRoles of Two Cytokines in Tumor Cell MigrationFigure 2. Plasticity and heterogeneity of tumor cell morphology and motility behavior. A . Bright field images of MDA-MB-231 cells embedded in 3D collagen matrices within the microfluidic channel at t = 0 and 8 h. Here, t = 0 is defined as the time when buffer and 100 nM SDF-1a solution were introduced into the two side channels. Cells were pre-incubated for 24 hours after seeding before the introduction of the gradients. C. Graphical description of cell speed U’, cell velocity along the gradient direction Vx ‘ , persistence length P’, and gradient-directed persistence length Px ‘ . D. Graphical description of aspect ratio. Distribution of cell aspect ratios at t = 0 and 8 h. E. Distribution of cell speed of elongated cells (aspect ratio larger than 3) and amoeboid-like cells (aspect ratio less than 3). doi:10.1371/journal.pone.0068422.gconcentration in the center channel is generated by supplying the same chemical concentration solution in both side channels. The 3D cell culture consisted of 106 tumor cells/ml embedded in type I collagen extracted from 23977191 rat tail [34]. Different concentrations (0.15, 0.25 and 0.35 ) of collagen were compared, and we measured the average cell speed to be 0.4260.02, 0.3060.02, and 0.2660.02 mm/min, respectively. We therefore used 1.5 mg/ml collagen for all the experiments shown here. In our culture model, cells remained evenly distributed after.