BEHAV3D Tumor Profiler pipeline

Schematic representation of the workflow showing data preparation, input and outputs in each of the three modules. a, Time-lapse image processing and preparation, based on segmentation and tracking of tumor (and labelled TME cells). Mounting Google Drive is optional but recommended to work in the Google Colab framework. b, Heterogeneity Module provides distinct behavioral clusters for cells, and relates them to dynamic features and provides a back-projection of the tracked cells. c, d, Optional modules, additional to the Heterogeneity module. c, Large-scale phenotyping module combines TME large-scale information with cell morphodynamic profiling to further depict population frequencies in distinct TME regions. d, Small-scale phenotyping module quantifies TME proximity and interaction features at the single-cell level. These features can be incorporated upstream as part of the clustering process to define behaviorally distinct subpopulations (option 1), or downstream to analyze how spatial relationships to TME components such as TAMMs or vasculature vary across behavioral clusters identified in the heterogeneity module (option 2).

Behavioral profiling of DMG cells with BEHAV3D-TP heterogeneity module.

a, Representative intravitally imaged position showing DMG cells tracked relative to tumor Edge (left panel). Tracks were classified according to DMG behavior (right, color coded by cluster). Representative of n = 18 independent positions from n=6 mice. b, UMAP plot showing seven color-coded DMG cell behavioral clusters identified by BEHAV3D-TP. Each datapoint represents one DMG cell track. A total of 981 individual cells were tracked in n = 18 independent positions from n=6 mice. c, Representative intravitally imaged positions and enlarged sections for showcasing distinct DMG behavioral clusters Invading (yellow), Retreating (green), Erratic (dark purple) and Static (light purple) through the time series. n = 18 independent positions from n=6 mice. Scale bars 10µm. d, Heatmap depicting relative values of DMG cell features indicated for each cluster, named according to their most distinct characteristics. Arbitrary units in respect to maximal and minimal values for each feature. See Table 1 for features description. Bubble size according to significance levels, all features represented have a significance level of P < 0.001 (***): mean_displ2, mean squared displacement; speed; delta displacement; displacement length; persistence; invasion. P-values represent the significance of differences across clusters tested through ANOVA for each feature. e, f, UMAP representation of DMG cells’ velocity relative to the tumor edge(µm) (e) and speed (µm/h) (f). Each datapoint represents one cell track and is colored following a color gradient. For each feature colors represent the mean DMG track values. Representative of n = 18 independent positions from n=6 mice.

Mapping distinct DMG behavioral patterns to distinct TME regions.

a, Overview of the large-scale TME region phenotyping workflow. After IVM imaging session, the tissue is fixed and stained and then analyzed using mLSR-3D. CytoMAP Spatial Analysis Toolbox is used to identify distinct large-scale regions, that are subsequently matched to IVM imaging dataset (c). b, Representative fixed multispectral image of a mLSR3D imaged DMG tumor and zoomed-in images of TME large-scale regions: Void (blue outline), TAMM/vascularized (red outline) and TAMM/Oligo (yellow outline). DMG nuclei (blue), CD31 (pink), Olig2 (purple), and Iba1 (yellow) are represented on the imaged. Scale bars, 100 µm (overview), 30 µm (zoomed-in regions). c, 3D multispectral image DMG cell rendering color-coded per large TME region: Void (blue), TAMM/Oligo (yellow) and TAMM/vascularized (red) (left). Based on the spatial coordinates of DMG cells classified into specific TME regions by CytoMAP, cells within the mLSR-3D images were subsequently assigned to their respective TME regions. Regions were then visually mapped on intravital imaging data (right). Representative of n=6 mice. White outline indicates one of the tumor regions analyzed with BEHAV3D-TP from the full tumor volume. d Boxplots showing the frequencies of different DMG behavioral clusters in Void, TAMM/Oligo and TAMM/vascularized TME large-scale regions. For each behavioral cluster, the y-axis displays the z-scored percentage of cells, normalized per mouse to account for inter-mouse variability. Each point represents an individual imaged position, with shape indicating the mouse of origin. P-values represent the significance of differences across TME regions tested through ANOVA with Tukey post hoc, for each behavioral cluster. n = 18 independent positions from n=6 mice.

The BEHAV3D-TP small-scale phenotyping correlates DMG behavioral profiles with microenvironmental composition at a single-cell resolution.

a, Overview of the small-scale TME phenotyping module. During IVM imaging, information about spatial distribution of SR101+, CD20r+, and CD31+ cells is collected and processed by the BEHAV3D-TP small-scale phenotyping module. Single DMG cell spatial TME features like distance to its neighbors, number of cells in a radius or minimal distance to a cell type were measured. b, Heatmap depicting DMG cell small-scale niches’ features across distinct DMG behavioral clusters. Arbitrary units in respect to maximal and minimal values for each feature. Bubble size according to significance levels, from smallest (P < 0.05 (**)) to medium (P < 0.01 (**)) to biggest (P < 0.001 (***)): dist_10_cell, Distance to nearest 10 cells, P < 0.01 (**); n_SR101, number of SR101 cells, P < 0.05 (*); min_SR101, minimal distance to a SR101 cell, P < 0.05 (*); n_CD20r, number of CD20r cells, P < 0.05 (*); min_Cd20r, minimal distance to a CD20r cell, P < 0.01 (**); min_BV_dist, minimal distance to a Blood vessel, P < 0.001 (***). P-values represent the significance of differences across clusters tested through ANOVA for each feature. c, e, Boxplots depicting the percentage of DMG cells across behavioral clusters that are close to CD20r+ TAMM (c) or to blood vessels (e). Each point represents an individual imaged position, color depicts the TME large-scale region type and shape denotes the mouse. Statistical differences across clusters were assessed using a linear mixed-effects model including mouse and TME class as random effects. P < 0.05 (*), P < 0.01 (**), P < 0.001 (***) indicate significant differences between clusters. c, n = 10 independent positions from n=3 mice. e, n = 8 independent positions from n=3 mice. d, Multispectral image showing the movement of DMG cells (grey) relative to CD20r+ TAMM (blue) and the Tumor core. The trajectories show a path from the initial (blue) to the final position (red). Scale bar, 10 µm. f, Time lapse multispectral images (left to right) showing the DMG cells (blue) displacement along CD31 blood vessel (purple) away from the tumor edge. Scale bars, 30 µm.

Example application of BEHAV3D-TP for 2D movement pattern analysis of breast cancer cells tracked with MTrackJ Fiji plugin

a) Representative intravitally imaged position showing breast tumor cells tracked in 2D with the MTrackJ plugin. Red: MDA-MB-231 cells; Blue: Second Harmonic Generation; Multicolored tracks: individually tracked cells. b) UMAP plot showing six color-coded breast tumor cell behavioral clusters identified by BEHAV3D-TP. Each datapoint represents one cell track. A total of 935 individual cells were tracked. c) Heatmap depicting relative values of breast tumor cell dynamic and spatial features indicated for each cluster, named according to their most distinct characteristics. Arbitrary units in respect to maximal and minimal values for each feature. See Table 1 for features description. d) Boxplots showing the percentages of different breast cancer behavioral clusters for 4T1 and MDA-MB-231 cells. Each point represents a different imaging position. P-values represent the significance between both conditions measured with a Student’s T test. e) Breast tumor cell tracks classified according to behavior (color coded by cluster in c). In all panels: representative of n = 13 independent positions from n=4 mice.

Example application of BEHAV3D-TP for 2D morphodynamic analysis of cells segmented with CellPose2.0 and tracked using TrackMate

a) Schematic overview of the segmentation and tracking pipeline using CellPose2.0 and TrackMate (see Methods). b) UMAP plot showing four color-coded morphodynamic clusters of healthy breast epithelial cells from TEBs identified by BEHAV3D-TP. Each datapoint represents one cell track. A total of 291 individual cells were tracked. c) Heatmap depicting relative values of epithelial cell dynamic and morphological features indicated for each cluster, named according to their most distinct characteristics. Arbitrary units in respect to maximal and minimal values for each feature. d) Stacked bar plot showing the distribution of morphodynamic cell types across the two indicated TEB populations: HR+ luminal progenitor cells and HR− luminal progenitor cells. Combined data from 19 mice and 72 independent positions. Chi-square test indicates p=0.02 for cluster 2.

Example application of BEHAV3D-TP for 3D morphodynamic analysis of cells segmented with μSAM and tracked using TrackMate

a) Schematic overview of the segmentation and tracking pipeline using μSAM and TrackMate. An online app was developed to integrate morphological and dynamic features (see Methods). b) UMAP plot showing four color-coded morphodynamic clusters of GBM cells identified by BEHAV3D-TP. Each datapoint represents one cell track. Total of 460 individually tracked cells. c) Heatmap depicting relative values of GBM cell dynamic and morphological features indicated for each cluster, named according to their most distinct characteristics. Arbitrary units in respect to maximal and minimal values for each feature. See Table 1 for features description. d) Boxplot showing Average distance to the 3 closest GBM neighbors (pixels) across distinct GBM morphodynamic clusters. Each data point represents the per-mouse average, calculated from a total of six independent imaging positions. Statistical significance across behavioral clusters was assessed using a linear mixed model with mouse as a random effect. P < 0.05 (*).

DMG cells heterogeneity captured by intravital imaging.

a, Schematic representation of an intravital imaging experimental design, capturing timelapses up to 5.4 hours of DMG cells expressing H2BmNeonGreen and mScarlet proteins, followed by single-cell tracking. b, Representative intravitally imaged position showing DMG cell tracks colored by behavioral cluster projected over the last timepoint of the original IVM time-lapse. c, Quantification of average DMG cell speed (µm/h) in each identified behavioral cluster per mouse. Significance of differences across behavioral clusters tested through ANOVA followed by Tukey’s post hoc test. All comparisons significant, except cluster 5-2, 4-7, 1-6, 7-3, 3-6, 3-4 and 6-4. d, Behavioral cluster frequency distribution across individual IVM imaged mice. (c,d) Each data point represents the average per mouse, computed from 18 independent imaging positions.

Cell type and behavioral cluster frequency distribution in identified TME large-scale regions.

a, Representative CytoMap output showcasing, 3D spatial distribution of DMG cells in distinct TME large-scale regions: Void (blue), TAMM/Oligo (yellow) and TAMM/vascularized (red). Each datapoint represent a unique DMG cell within a classified large-scale region. b, c, Fold change of cell count of DMG, Iba1+, Olig2+ and CD31+ cell types, identified with mLSR3D fixed imaging, relative to the mean cell count in each TME large scale region (Void, TAMM/Oligo, TAMM/vascularized), both in boxplot (b) and table (c) formats. In (b) each point represents a mouse, n=7. For DMG cell count: TAMM/Oligo versus Void, P < 0.01 (**); TAMM/Oligo versus TAMM/vascularized, P < 0.05 (*). For Iba1 cell count: TAMM/Oligo versus Void, P < 0.001 (***); TAMM/vascularized versus Void, P < 0.001 (***). In Olig2 cell count: TAMM/Oligo versus Void, P < 0.001 (***); TAMM/Oligo versus TAMM/vascularized, P < 0.001 (***). For CD31 cell count: TAMM/Oligo versus Void, P < 0.001 (***); TAMM/vascularized versus Void, P < 0.001 (***); TAMM/Oligo versus TAMM/vascularized, P < 0.001 (***). d, Boxplots showing the percentages of different DMG behavioral clusters in Void, TAMM/Oligo and TAMM/vascularized TME large-scale regions. Each point represents a different imaging position. Color: n = 18 independent positions; Shape: n=6 mice. e, f, g, Boxplots plots showing DMG single cell features (Tumor cell speed (µm/hour) (e), Average speed displacement (f) and Tumor cell velocity (µm/hour) (g)) differences among Void, TAMM/Oligo and TAMM/vascularized per mouse in TME large-scale regions. g, Tumor cell velocity can be either towards brain core (down, negative values) or towards brain parenchyma (up, positive values). For e, f, g: Each data point represents the average per mouse, computed from 18 independent imaging positions. Statistical differences were assessed using one-way ANOVA. No statistically significant differences were found.

Cell distribution and behavioral clusters in small-scale TME.

a, b, Representative 3D IVM rendering of DMG nuclei (cyan), CD31+ cells (grey) and SR101+ (a, red) or CD20r+ cells (b, red). Scale bars, 100 µm. c, d, Boxplots showing Average distance to the 10 closest DMG neighbors (µm) (c) and Number of SR101+ cells in a 30 µm radius (d) across distinct DMG single-cell behavioral clusters. (c) Each point represents an individual mouse. Statistical differences across clusters were assessed using a linear mixed-effects model including mouse as random effect. P < 0.05 (*) indicate significant differences between clusters. n =18 independent positions from n=6 mice. (d) Each point represents an individual imaged position, color depicts the TME large-scale region type and shape denotes the mouse. Statistical differences across clusters were assessed using a linear mixed-effects model including mouse and TME class as random effects. P < 0.05 (*), P < 0.01 (**) indicate significant differences between clusters. n=8 independent positions from n=3 mice.