Collective directional migration drives the formation of heteroclonal cancer cell clusters

Abstract

Metastasisation occurs through the acquisition of invasive and survival capabilities that allow tumour cells to colonise distant sites. While the role of multicellular aggregates in cancer dissemination is acknowledged, the mechanisms that drive the formation of multiclonal cell aggregates are not fully elucidated. Here, we show that cancer cells of different tissue of origins can perform collective directional migration and can actively form heteroclonal aggregates in 3D, through a proliferation-independent mechanism. Coalescence of distant cell clusters is mediated by subcellular actin-rich protrusions and multicellular outgrowths that extend towards neighbouring aggregates. Coherently, perturbation of cytoskeletal dynamics impairs collective migration while myosin II activation is necessary for multicellular movements. We put forward the hypothesis that cluster attraction is mediated by secreted soluble factors. Such a hypothesis is consistent with the abrogation of aggregation by inhibition of PI3K/AKT/mTOR and MEK/ERK, the chemoattracting activity of conditioned culture media and with a wide screening of secreted proteins. Our results present a novel collective migration model and shed light on the mechanisms of formation of heteroclonal aggregates in cancer.

Keywords: 3D models; collective migration; directional migration; heterogeneity; imaging; protrusions.

Fig. 1

CCs grow, migrate and aggregate in 3D cultures. (A) Six cancer cell (CC) lines from different tissues of origin were seeded as single‐cell suspension in a 3D growth assay and imaged by means of bright‐field microscopy each day for 2–3 weeks. Representative snapshots at indicated time points are shown. Each image is obtained by creating an extended depth‐of‐field (EDF) projection starting from a wide Z‐stack (around 1 mm thick). Each row represents a cell line, from top to bottom: PC‐3 (initial seeding density calculated a posteriori: 31 cells·mm⁻³); MDA‐MB‐231 (22.2 cells·mm⁻³); ACHN (35.1 cells·mm⁻³); MG‐63 (84.2 cells·mm⁻³); NCI‐H23 (36.2 cells·mm⁻³) and Hs 746T (90.4 cells·mm⁻³). The smaller white squares in the leftmost image of each row are zoomed in the corresponding inserts at the bottom right corner of the images to highlight aggregation events. Scale bar: 200 μm. Each representative snapshot is chosen among at least three biological replicates each performed in double technical replicate. (B) Sketch of a human silhouette to illustrate the point of origin of each aggregating CC line. (C) Confocal microscopy images showing preaggregated three spheroids of PC‐3 (top and bottom‐right panels) and MDA‐MB‐231 (bottom‐left panel) infected with LifeAct Ruby (top) or with H2B‐GFP/LifeAct‐Ruby (bottom) and seeded in Matrigel. Both subcellular protrusions (top) and long multicellular outgrowths (bottom) are well visible. Scale bar top insets: 50 μm; bottom‐left inset: 100 μm; enlarged bottom‐left inset: 20 μm; bottom‐right inset: 200 μm; bottom‐right enlarged inset: 50 μm. (D) Representative snapshots as in panel (A) of nonaggregating cell lines. From top to bottom: MIA PaCa‐2 (initial seeding density calculated a posteriori: 71 cells·mm⁻³); DU145 (62.5 cells·mm⁻³) and A549 (63.7 cells·mm⁻³). White squares are meant to indicate clusters that grow without aggregating. Scale bar: 200 μm. Each representative snapshot is chosen among at least three biological replicates each performed in double technical replicate. (E) Ratio of protrusions to bulk regions of the spheroid for different cell lines obtained by segmentation of EDF‐projected images of aggregation assays at 12 days after seeding. Each colour refers to a cell line, as indicated in panel (I). (F) Fraction of cells seeded at the beginning of the assay that are subsequently involved (blue) or not involved (red) in aggregation events over the course of an aggregation assay. ‘N’ indicates the initial seeding densities (cells·mm⁻³). (G) Sketch depicting the three types of aggregation events included in the measure shown in panel (F). Light blue circles represent cell clusters, with the distance from centroid to centroid indicated with black markings on top. Top row: aggregation by growth. Middle row: aggregation by directional migration. Bottom row: aggregation by protrusive activity. (H) Correlation between the presence of protrusions/outgrowths shown in panel (E) and the percentage of aggregating cells shown in panel (F). Each colour refers to a cell line, as indicated in panel (I). (I) Colour legend for panels (E) and (H).

Fig. 2

Aggregation kinetics is independent of initial seeding density and coupled to cell proliferation rates. (A) The number of separate objects in each extended depth‐of‐field (EDF)‐projected image over time is fitted with a sigmoidal curve for different initial seeding densities. Each set of lines and points represents a single aggregation assay and points are experimental data, while lines represent the fit. Each time series represents one field of view. (B) Same as (A) but for nonaggregating cell lines. The legend indicates the colours associated with cell lines. Each time series represents one field of view. At least three biological replicates performed in duplicate were performed for each cell line/density. (C) The plot of the sigmoid function used to fit the data of aggregation kinetics with indicated parameters used in the quantification. The operational definitions of each parameter are as follows: halving time is the time at which the number of objects is half the initial fitted number of objects; lag time t_l is the time corresponding to the first change of convexity of the function; n_0‐b is the number of objects at the end of the assay. (D, E) The density fraction (difference between the number of objects at the beginning and the end of the assay as resulting from the fit parameters divided by the number of objects at the beginning) is plotted for each cell line and initial starting density. (F) The plot reports the halving times obtained by the fit parameters of curves shown in panel (A) and plotted as a function of the seeding density. The horizontal line represents the median. At least three biological replicates performed in duplicate were performed for each cell line/density. (G) Scatter plots reporting fitted starting density vs. the total area of the growing spheroids, i.e. the total projected area in an EDF‐projected field of view. Each dot represents a single time point from the aggregation assays. Lines parallel to the y‐axis would indicate pure proliferation with no aggregation, while lines parallel to the x‐axis would indicate aggregation with no proliferation. Each time series represents one field of view. (H, I) The plots report the approximated doubling times (inferred from time series of total projected area), and the scatterplot halving time vs. doubling time obtained by the fit parameters of curves shown in panel (A) and plotted as a function of the seeding density. Doubling times were obtained by a separate fit (see details in Section 2, and growth curves reported in Fig. S2a). Each dot represents an aggregation assay performed with a cell line at a given initial seeding density with colour codes reported in the legend. The horizontal line represents the median.

Fig. 3

Preformed cell spheroids perform collective directional migration. (A) Each row shows representative snapshots of a time‐lapse experiment with an aggregation assay performed starting from preaggregate spheroids (the same six cell lines shown in Fig. 1A are shown). Seeding density: 2.5 spheroids·mm⁻³. Scale bar: 500 μm. Each representative snapshot is chosen among at least three biological replicates each performed in duplicate technical replicates. (B) The preformed spheroids aggregation assay performed with the MIA PaCa‐2 cell line (among the nonaggregating cells). A single merging event is observed when two spheroids come into contact due to growth. Seeding density: 2.5 spheroids·mm⁻³. Scale bar: 500 μm. Each representative snapshot is chosen among at least three biological replicates each performed in duplicate technical replicate. (C) Top: representative snapshots of an aggregation assay with preformed spheroids where five clusters were manually segmented and tracked over the course of the assay. Each colour defines the outline of an object, and one of the two colours of two merging objects is retained in an aggregation event. Bottom: the outlines of the spheroids (sampled every 3 h) as in the top row were superposed with a time‐dependent colour code (red: earliest time point; green: latest time point) in order to highlight the protrusions mediating the aggregation event. Each panel ends when two spheroids merge.

Fig. 4

Collective directional migration proceeds independently from proliferation. (A) The plots show the growth rate fitted by measuring the number of cells by image segmentation in cell lines cultured in 2D and treated with mitomycin at the indicated concentration. Positive growth rates indicate the presence of cell proliferation while negative growth rates indicate toxicity. The duration of the time lapse is 19 h. Note that growth rates in control conditions are consistent with doubling times measured in aggregation assays. Each value of concentration is assessed with a technical quadruplicate relative to at least 100 cells. Horizontal lines represent medians. (B) Snapshots of spheroid‐based aggregation assays performed with mitomycin‐treated cells (0.75 μg·mL⁻¹). Scale bar 200 μm. Each representative snapshot is chosen among two biological replicates each performed in double technical replicate. (C) Fraction of EdU incorporating cells in deconvolved images of spheroids treated with mitomycin for MG‐63 spheroids as those shown in panel (D). Each point is the mean of the fractions obtained by image segmentation in four different fields of view. (D) Snapshots of aggregating spheroids stained with EdU and DAPI for MG‐63. Scale bar 200 μm. Images are equalised with the same limits. (E) Quantification of two time points corresponding to time‐lapse experiment shown in panel (B). The number of independent multicellular aggregates (blue) and single cells (red) was quantified at the beginning of the experiment (0 h), and at the end (90 h), normalised with respect to the total number of objects at 0 h.

Fig. 5

The role of protrusions in collective directional migration. (A) Representative images showing preformed spheroids of MG‐63 cells transduced with LifeAct‐GFP (green) or LifeAct‐Ruby (red) expressing lentiviral vectors, embedded in Matrigel. White triangles indicate protrusions between the two spheroids. The region of the images corresponding to protrusions was enlarged and shown in the top‐right inserts. Scale bar: 100 μm. (B) Representative images of preformed spheroids of LifeAct‐Ruby expressing MDA‐MB‐231 embedded in Matrigel. White triangles point at protrusions directed towards the neighbouring object. Green triangles point at smaller protrusions that develop all around the spheroids. Scale bar: 100 μm. (C) Preformed spheroids of three selected aggregating cell lines were either left untreated or treated with 1 μm latrunculin A or 20 μm Y‐27632. (From left to right: MDA‐MB‐231, PC‐3 and MG‐63; from top to bottom: untreated, latrunculin A, Y‐27632). Snapshots are extracted from time‐lapse experiments at 0, 24, 48 and 80 h after seeding. Seeding density: 2.5 spheroids·mm⁻³. Scale bar: 200 μm. Each representative snapshot is chosen among at least two biological replicates each performed in double technical replicate. (D) The images corresponding to experiments as those shown in panel (C) were segmented in order to extract the ratio of protrusion to bulk regions for each treatment, as detailed in the legend. Each curve represents a time series picked among at least two biological replicates each performed in double technical replicate. (E) Quantification of two time points corresponding to time‐lapse experiment shown in panel (c). The number of independent multicellular aggregates (blue) and single cells (red) was quantified at the beginning of the experiment (0 h), and at the end (80 h), normalised with respect to the total number of objects at 0 h for control, latrunculin and rock inhibitor. (F) Area velocity extracted from images as detailed in Section 2. Each colour represents a different treatment as detailed in the legend. Each curve represents a time series picked among at least two biological replicates each performed in double technical replicate.

Fig. 6

Inhibition of downstream effectors MEK, AKT, PI3K and mTOR affects collective directional migration to different extents. (a; left column) An aggregation index (equivalent to the density fraction shown in Fig. 2D,E) is shown for each inhibitor and each cell line to indicate the impact of each inhibitor on the aggregation process in each cell line. Aggregation assays are started from single cells. (a; right column) A proliferation index (obtained as the ratio between the total area of the spheroids in the images at 15 and 3 days after seeding) is shown for each inhibitor and each cell line in order to indicate the impact of each inhibitor on proliferation. Top to bottom: MDA‐MB‐231, PC‐3 and MG‐63. Each bar is coming from at least three biological replicates each performed in double technical replicate. (b; left column) The ratio of protrusion to bulk regions in the images for each of the treatments and the three cell lines is shown. (b; right column) Area velocity extracted from images as detailed in Section 2. Each colour represents a different treatment as detailed in the legend. Each line is a time series selected from at least three biological replicates each performed in double technical replicate. (C) Preformed spheroids seeded in Matrigel were imaged for 4 days. Representative pictures at three time points (0, 40 and 96 h) for control and each inhibitor are shown. Each of the three panels of images corresponds to a cell line; from left to right: MDA‐MB‐231, PC‐3 and MG‐63. Each row corresponds to one experimental condition, as reported on the left of each panel; from top to bottom: untreated, MEK inhibitor AZD‐6244 0.5 μm, PI3K inhibitor BYL719 1 μm (MG‐63) or 3 μm (MDA‐MB‐231, PC‐3), PI3K‐mTOR inhibitor BEZ235 0.1 μm and AKT inhibitor MK‐2206 5 μm. Scale bar: 200 μm. Each representative snapshot is selected from at least three biological replicates each performed in double technical replicate.

Fig. 7

Conditioned medium is chemoattractant for aggregating cell lines. (A) The plots report the migrated cells (fold over control) for different cell lines and conditioned media collected at different times. Each point on the plot represents data coming from a whole membrane. At least three different membranes were considered in each condition. Migration of cells from serum‐free media towards serum‐free media (SF – SF) was used to normalise data as a control. Data are reported as median (horizontal line) with interquartile range. Statistical significance was assessed by performing a parametric one‐tailed t‐test with Welch's correction (unpaired); *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001. (B) Representative images of membranes corresponding to the experimental conditions reported in the graphs are shown. (C) List of gene symbols corresponding to the proteins found in the corresponding conditioned media for 2D vs. 3D, where proteins found in the controls with no cells, in the Matrigel or in the FBS were filtered out. (D) Euler–Venn diagrams of factors that can be found in the corresponding conditioned media for 2D vs. 3D, where proteins found in the controls with no cells, in the Matrigel or the FBS were filtered out. (E) List of gene symbols of proteins found in the 3D conditioned media classified according to their gene ontology features. (F, G) List and Euler–Venn diagram of the three conditioned media and corresponding list with the common proteins annotated. (H) Heatmap of the transcriptional expression (CCLE) of the cognate receptors of all chemokines and growth factors for the three cell lines. Boxed values represent cases in which both the ligand and the receptor are expressed.

Fig. 8

Heterotypic aggregation is observed across different cell lines. (A) Snapshots of time‐lapse experiments performed with preformed spheroids of cell lines couples expressing H2B‐RFP (red) or H2B‐GFP (green) were seeded in Matrigel. Each row represents (from top to bottom): H2B‐RFP MDA‐MB‐231 and H2B‐GFP PC‐3; H2B‐RFP MDA‐MB‐231 and H2B‐GFP MG‐63; H2B‐GFP MG‐63 and H2B‐RFP PC‐3. Scale bar: 200 μm. Each sequence is selected among two biological replicates performed with technical triplicates. (B) All the graphs and images in this panel refer to the aggregation assay between PC‐3 and MDA‐MB‐231 (top‐row images in A). The bar plots show the histogram of the number of composing aggregates of each object for the first (top, 0 h) and the last (bottom, 92 h) timeframe of the movie for the top row of panel (A), top row. Composing aggregates at frame 1 are conventionally considered as single objects as their aggregation happened before the assay started. (C) Images show the segmentation of the first and the last frames of the aggregation assay shown in panel (A), top row. Each object is identified by a numeric label and a unique colour in order to appreciate the aggregation of initially separated objects. (D–F) Chemotaxis assays (Transwell) performed with the cell line couples represented in panel (A). The plots report the migrated cells (fold over control) of different experiments. Each point on the plot represents a whole membrane. At least three membranes were analysed for each condition. Data are reported as median (horizontal line) with interquartile range. Statistical significance was assessed by performing a parametric one‐tailed t‐test with Welch's correction (unpaired); *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001. Representative membranes corresponding to the experimental condition reported in the plots are shown on top of each plot. (G) The matrix reports the migrated cells (fold over control) of different Transwell assay experiments performed to test whether the conditioned medium of each of the six aggregating cell lines acts as a chemoattractant for all the other cell lines. Indicated values correspond to the log₂ of the median fold‐change. Experiments were conducted by seeding onto the Transwell membrane cell lines listed in the column labels and by adding in the lower compartment 48 h conditioned media collected from cell lines listed in the row labels. (H) Representative images of time‐lapse movies showing an aggregation event between preformed spheroids of LifeAct‐Ruby (red) MDA‐MB‐231 and LifeAct‐GFP (green) PC‐3 cells in Matrigel. The section of the images corresponding to the area between the two clusters was enlarged to facilitate the visualisation of the protrusions and it is shown in the top‐right inserts. Time labels indicate hours after seeding. Scale bar: 100 μm.