Differential stiffness between brain vasculature and parenchyma promotes metastatic infiltration through vessel co-option

Abstract

In brain metastasis, cancer cells remain in close contact with the existing vasculature and can use vessels as migratory paths-a process known as vessel co-option. However, the mechanisms regulating this form of migration are poorly understood. Here we use ex vivo brain slices and an organotypic in vitro model for vessel co-option to show that cancer cell invasion along brain vasculature is driven by the difference in stiffness between vessels and the brain parenchyma. Imaging analysis indicated that cells move along the basal surface of vessels by adhering to the basement membrane extracellular matrix. We further show that vessel co-option is enhanced by both the stiffness of brain vasculature, which reinforces focal adhesions through a talin-dependent mechanism, and the softness of the surrounding environment that permits cellular movement. Our work reveals a mechanosensing mechanism that guides cell migration in response to the tissue's intrinsic mechanical heterogeneity, with implications in cancer invasion and metastasis.

Extended Data Fig. 1 |. Characterization of cell shape and invasion depth in ex vivo brain cultures.

(a) Percentage of 1205Lu cells adhered to vessels after 24h of ex vivo culture (n=7 slices, 4 experiments). Graph shows mean ± standard deviation. (b) Brain slice immunostaining of laminin (magenta), lifeact-GFP (green) and nuclei (hoechst, blue) of a 1205Lu cell spreading over the outer surface of a vessel. 3D view,11μm-thick image stack. Scale bar 10μm. (c) Left: single image plane of the cell in (b), including isolectin-IB4 labelling of endothelial cells (grey). Scale bar 10μm. Right: laminin (magenta), lifeact-GFP (green) and isolectin-IB4 (grey) fluorescence profile along the dashed yellow line. (d) Example of cell invasion depth quantification. Left: immunostaining of laminin (magenta) and lifeact-GFP (green). Yellow dashed lines outline a cell in the bulk (cell 1) and a cell spreading along a vessel (cell 2). Scale bar 50μm. Right: region 2 (centred on the cell of interest) is used to average laminin intensity across a 60μm stack and define the reference point for invasion depth (0μm). Because of the opacity of brain tissue, averaged laminin intensity shows a peak at the top of the slice and decays as imaging deeper into the tissue. The origin of invasion depths is defined as z_{peak_region2} −6μm. Region1 (that follows the cell 2D shape), is used to average both cell and laminin fluorescence. The intensity of the GFP signal is used to determine the cell depth within the slice, and local laminin intensity is used to determine if a cell is in the bulk or on a vessel (if observing a local peak of laminin intensity). (e) Violin plots of 1205Lu cell shape circularities when adhered to vessels or to the bulk brain tissue 4 and 24h after seeding. (f) Violin plots of 1205Lu invasion depths within the slice when adhered to vessels or to the bulk brain tissue 4 and 24h after seeding. For e-f, n=562 for Bulk, n=621 for Vessel at 4h, n=337 for Bulk, n=706 for Vessel at 24h, from 6 slices, 3 experiments. Graphs show median ± interquartile range. Dots represent the median values for each slice analysed. Statistical analysis: Kruskal–Wallis test followed by a post hoc Dunn’s test.

Extended Data Fig. 2 |. Vessel co-option is a common pattern of invasion across many different cell lines.

a-c, Mean invasion depth per slice for cells adhered to vessels or to the bulk brain tissue for 1205Lu (n=7), WM983B (n=4), U2OS (n=4), MCF7 (n=6) and MDA-MB-231 (n=6) (a), A375P (n=8) and A375M2 (n=6) (b), nHDF (n=7), MCF10A (n=5) and HBVPC (n=5) cell lines (c). Full distributions shown in Fig. 1d, j–l. Data corresponds to at least 3 experiments. Statistical analysis: two-sided paired T-test or an ordinary one-way ANOVA test followed by a post hoc Tukey’s test (for b). All graphs show mean ± standard deviation. (d) Violin plots of cell shape circularities for MDA-MB-231 TGL, BrM2831, BoM-1833 and LM2–4175 cells adhered to the vasculature. (e) Violin plots of bulk invasion depths for MDA-MB-231 TGL, BrM2–831, BoM-1833 and LM2–4174 cells. (f) Violin plots of vessel invasion depths for MDA-MB-231 TGL, BrM2–831, BoM-1833 and LM2–4175 cells. (g) Violin plots of cell shape circularities for cells adhered to vessels or to the bulk brain tissue for B16-F10, E077 and NIH-3T3 mouse-derived cell lines. (h) Violin plots of invasion depths for cells adhered to vessels or to the bulk brain tissue for B16-F10, E0771 and NIH-3T3 mouse-derived cell lines. For MDA-MB-231 TGL (parental cell line): n=394 for Bulk, n=441 for Vessel, from 5 slices, 3 experiments. For BrM2–831 (brain tropic): n=652 for Bulk, n=479 for Vessel, from 6 slices, 3 experiments. For BoM-1833 (bone tropic): n=520 for Bulk, n=496 for Vessel, from 6 slices, 3 experiments. For LM2–4175 (lung-tropic): n=564 for Bulk, n=529 for Vessel, from 6 slices, 3 experiments. For B16-F10 (melanoma): n=239 for Bulk, n=212 for Vessel, from 4 slices, 3 experiments. For E0771 (breast cancer): n=269 for Bulk, n=359 for Vessel, from 6 slices, 3 experiments. For NIH-3T3 (embryonic fibroblast): n=165 for Bulk, n=530 for Vessel, from 6 slices, 3 experiments. (i) Brain slice immunostaining of laminin (magenta) and T-cells labelled with CellTracker-488 (green). Left: z-plane 2μm below the top surface. Right: z-plane 50μm below the top surface. White insert labels zoomed area. Scale bars: 100μm, 10μm (inserts). (j) Violin plots of T-cell shape circularities when adhered to vessels or to the bulk brain tissue. (k) Violin plots of T-cell invasion depths within the slice when adhered to vessels or to the bulk brain tissue. For T-cells: n=380 for Bulk, n=208 for Vessel, from 3 slices, 2 experiments. For all violin plots, dots represent the median value for each slice analysed. Statistical analysis: two-sided Mann–Whitney or a Kruskal–Wallis test followed by a post hoc Dunn’s test (for d-f). Graphs show median ± interquartile range.

Extended Data Fig. 3 |. Depletion of CD44, integrin β1, zyxin or paxillin does not significantly impair invasion into brain slices.

(a) Percentage of scramble control and CD44 KO cells adhered to vessels after 24h of ex vivo culture (n=8 slices, 4 experiments). (b) Left: representative western blot of CD44 expression in scramble control and CD44 KO cells. Right: quantification of all western blots (one per experiment, 4 experiments). (c) Violin plots of invasion depths for scramble control and CD44 KO cells moving through the bulk brain tissue (Scr: n=571, CD44 KO: n=708, 8 slices, 4 experiments). (d) Violin plots of invasion depths for scramble control and CD44 KO cells moving along the vasculature (Scr: n=652, CD44 KO: n=511, 8 slices, 4 experiments). (e) Percentage of scramble control and integrin β1 KO cells adhered to vessels after 24h of ex vivo culture (n=7 slices, 4 experiments). (f) Left: representative western blot of integrin β1 expression in scramble control and integrin β1 KO cells. Right: quantification of all western blots (one per experiment, 4 experiments). (g) Violin plots of invasion depths for scramble control and integrin β1 KO cells moving through the bulk brain tissue (Scr: n=500, Itgβ1 KO: n=607, 7 slices, 4 experiments). (h) Violin plots of invasion depths for scramble control and integrin β1 KO cells moving along the vasculature (Scr: n=720, Itgβ1 KO: n=354, 7 slices, 4 experiments). i-u, Adaptor proteins. The zyxin CRISPR KO cell lines were established as clonal cell lines by single cell sorting after infection with the CRISPR lentivirus, and two cell lines were analysed (data shown in Fig. 2 corresponds to clone 1). (i) Percentage of scramble control, zyxin clone-1 KO and paxillin KO cells adhered to vessels after 24h of ex vivo culture (Scr: n=7 slices, Zyx KO c1: 8 slices, Pxn KO: 8 slices, 3 experiments). (j) Percentage of scramble control and zyxin clone-2 KO cells adhered to vessels after 24h of ex vivo culture (n=6 slices, 3 experiments). (k) Left: representative western blot of zyxin expression in scramble control and zyxin clone-1 KO cells. Right: quantification of all western blots (one per experiment, 4 experiments). (l) Left: representative western blot of paxillin expression in scramble control and paxillin KO cells. Right: quantification of all western blots (one per experiment, 4 experiments). (m) Left: representative western blot of zyxin expression in scramble control and zyxin clone-2 KO cells. Right: quantification of all western blots (one per experiment, 3 experiments). (n) Violin plots of invasion depths for scramble control, zyxin clone-1 KO and paxillin KO cells moving through the bulk brain tissue (Scr: n=467, 7 slices; Zyx KO c1: n=353, 8 slices; Pxn KO: n=660, 8 slices, 4 experiments). (o) Violin plots of invasion depths for scramble control and zyxin clone-2 KO cells moving through the bulk brain tissue (Scr: n=288 for Scr, Zyx KO c2: n=303, 6 slices, 3 experiments). (p) Violin plots of invasion depths for scramble control, zyxin clone-1 KO and paxillin KO cells moving along the vasculature (Scr: n=593, 7 slices; Zyx KO c1: n=704, 8 slices; Pxn KO: n=597, 8 slices, 4 experiments). (q) Violin plots of invasion depths for scramble control and zyxin clone-2 KO cells moving along the vasculature (Scr: n=275, Zyx KO c2: n=481, 6 slices, 3 experiments). r-u, FAK inhibition with FC11. (r) Percentage of cells adhered to vessels in control conditions or under 1μM FC11 after 24h of ex vivo culture (n=6 slices, 3 experiments). (s) Violin plots of cell shape circularities for cells adhered to the vasculature in control conditions or under 1μM FC11 (Control: n=631, FC11 1μM: n=642, 6 slices, 3 experiments). (t) Violin plots of invasion depths for cells moving through the bulk brain tissue in control conditions or under 1μM FC11 (Control: n=466, FC11 1μM: n=289, 6 slices, 3 experiments). (u) Violin plots of invasion depths for cells moving along the vasculature in control conditions or under 1μM FC11 (Control: n=631, FC11 1μM: n=642, 6 slices, 3 experiments). For all violin plots, dots represent the median values per slice analysed. Statistical analysis: two-sided Mann–Whitney or a Kruskal–Wallis test followed by a post hoc Dunn’s test (for n,p). Graphs show median ± interquartile range. For all other graphs, statistical analysis was performed using a two-sided T-test or an ordinary one-way ANOVA test followed by a post hoc Tukey’s test (for i). Graphs show mean ± standard deviation. For western blot quantification, protein expression was normalized to loading control (β-actin), and to the expression on the scramble control condition.

Extended Data Fig. 4 |. Depletion of talin significantly impairs invasion into brain slices.

a-e, Talin 1 KO. The talin 1 CRISPR KO cell lines were established as clonal cell lines by single cell sorting after infection with the CRISPR lentivirus, and two cell lines were analysed (data shown in Fig. 2 corresponds to clone 1). (a) Percentage of scramble control, talin 1 clone-1 KO and talin 1 clone-2 KO cells adhered to vessels after 24h of ex vivo culture (Scr: n=7, Tln1 KO: n=8, 4 experiments). (b) Left: representative western blot of talin 1 and talin 2 expression in scramble control, talin 1 clone-1 KO and talin 1 clone-2 KO cells. Right: quantification of all western blots (one per experiment, 4 experiments). (c) Violin plots of cell shape circularities for scramble control, talin 1 clone-1 KO and talin 1 clone-2 KO cells adhered to the vasculature (Scr: n=826, 7 slices; Tln1 KO c1: n=725, 8 slices; Tln1 KO c2: n=741, 8 slices; 4 experiments). (d) Violin plots of invasion depths for scramble control, talin 1 clone-1 KO and talin 1 clone-2 KO cells moving through the bulk brain tissue (Scr: n=425, 7 slices; Tln1 KO c1: n=521, 8 slices; Tln1 KO c2: n=721, 8 slices; 4 experiments). (e) Violin plots of invasion depths for scramble control, talin 1 clone-1 KO and talin 1 clone-2 KO cells moving along the vasculature (Scr: n=826, 7 slices; Tln1 KO c1: n=725, 8 slices; Tln1 KO c2: n=741, 8 slices; 4 experiments). f-i, Talin1+2 KO. (f) Percentage of scramble control, talin 2 KO and talin 1 and 2 KO cells adhered to vessels after 24h of ex vivo culture (n=8 slices, 4 experiments). (g) Left: representative western blot of talin 1 and talin 2 expression in scramble control, talin 2 KO and talin 1 and 2 KO cells. Right: quantification of all western blots (one per experiment, 4 experiments). (h) Violin plots of invasion depths for scramble control, talin 2 KO and talin 1 and 2 KO cells moving through the bulk brain tissue (Scr: n=560, Tln2 KO: n=509, Tln1+Tln2 KO: n=515, 8 slices, 4 experiments). (i) Violin plots of invasion depths for scramble control, talin 2 KO and talin 1 and 2 KO cells moving along the vasculature (Scr: n=717, Tln2 KO: n=640, Tln1+Tln2 KO: n=544, 8 slices, 4 experiments). For all violin plots, dots represent the median values per slice analysed. Statistical analysis: Kruskal–Wallis test followed by a post hoc Dunn’s test. Graphs show median ± interquartile range. For all other graphs, statistical analysis was performed using a one-way ANOVA test followed by a post hoc Tukey’s test. Graphs show mean ± standard deviation. For western blot quantification, protein expression was normalized to loading control (β-actin), and to the expression on the scramble control condition.

Extended Data Fig. 5 |. Roughness ratio calculation.

(a) 1205Lu lifeact-GFP image (grey). From left to right, representative images of a scramble control and a talin1 and 2 KO tumour one week after brain injection. Overlayed, the tumour edge (green) and boundary (magenta) that were used to calculate the roughness ratio as (boundary length)/(edge length). Scalebar 500μm.

Extended Data Fig. 6 |. Modelling vessel co-option in vitro.

(a) Device immunostaining of HMBEC lifeact-Ruby (magenta), 1205Lu lifeact-GFP (green) and nuclei (hoechst, blue). From left to right, representative images of a scramble control, a talin 1 KO, a talin 2 KO and a talin 1 and 2 KO cell adhering to the micro-vessel. Maximum projection, scale bar 20μm. (b) Cell sphericity of scramble control, talin 1 KO, talin 2 KO and talin 1 and 2 KO cells adhered to the micro-vessel (Scr: n=53, 7 devices; Tln1 KO: n=25, 6 devices; Tln2 KO: n=42, 6 devices; Tln1+Tln2 KO: n=29, 8 devices; 3 experiments). Lighter dots represent each cell value and darker dots represent the median value per device. Statistical analysis: Kruskal–Wallis test followed by a post hoc Dunn’s test. Graphs show median ± interquartile range. (c) Cell velocity of scramble control and talin 1 and 2 KO cells moving along the micro-vessel (Scr: n=47, 7 devices; Tln1+Tln2 KO: n=42, 8 devices; 3 experiments). Lighter dots represent each cell value and darker dots represent the mean value per device. Statistical analysis: two-sided T-test. Graph show mean ± standard deviation. d-e, Laminin intensity (d) and cell sphericity (e) of 1205Lu cells spreading on control devices or in devices where the apoptosis of endothelial cells was triggered with the small molecule CID (Control: n=26, 7 devices; CID 5nM: n=118, 18 devices; 4 experiments). Lighter dots represent each cell value and darker dots represent the median value per device. Statistical analysis: two-sided Mann–Whitney test. Graphs show median ± interquartile range.

Extended Data Fig. 7 |. Depletion of vinculin significantly impairs invasion into brain slices.

The vinculin CRISPR KO cell lines were established as clonal cell lines by single cell sorting after infection with the CRISPR lentivirus, and two cell lines were analysed (data shown in Fig. 5 corresponds to clone 2). (a) Percentage of scramble control, vinculin clone-1 KO and vinculin clone-2 KO cells adhered to vessels after 24h of ex vivo culture (n=5, 3 experiments). Statistical analysis: one-way ANOVA test followed by a post hoc Tukey’s test. Graphs show mean ± standard deviation. (b) Left: representative western blot of vinculin expression in scramble control, vinculin clone-1 KO and vinculin clone-2 KO cells. Right: quantification of all western blots (one per experiment, 3 experiments). (c) Violin plots of cell shape circularities for scramble control, vinculin clone-1 KO and vinculin clone-2 KO cells adhered to the vasculature (Scr: n=411, Vcl KO c1: n=453, Vcl KO c2: n=474, 5 slices, 3 experiments). (d) Violin plots of invasion depths for scramble control, vinculin clone-1 KO and vinculin clone-2 KO cells moving through the bulk brain tissue (Scr: n=242, Vcl KO c1: n=385, Vcl KO c2: n=411, 5 slices, 3 experiments). (e) Violin plots of invasion depths for scramble control, vinculin clone-1 KO and vinculin clone-2 KO cells moving along the vasculature (Scr: n=411, Vcl KO c1: n=453, Vcl KO c2: n=474, 5 slices, 3 experiments). For all violin plots, dots represent the median values per slice analysed. Statistical analysis: Kruskal–Wallis test followed by a post hoc Dunn’s test. Graphs show median ± interquartile range. For western blot quantification, protein expression was normalized to loading control (β-actin), and to the expression on the scramble control condition.

Extended Data Fig. 8 |. Tuning the stiffness of the vasculature by filling it with a stiff gel or by inhibiting contractility.

a-b, Brain stiffness characterization. Young’s modulus (a) and force-indentation curves (b) of fresh 700μm-thick brain slices and 20μm-thick cryosections (Fresh: n=120, Cryo: n=120, 6 slices, 3 experiments). Statistical analysis: two-sided Mann–Whitney test. Lighter dots represent each indentation value and darker dots represent the median value per experiment. Graphs show median ± interquartile range. (c) Force-indentation AFM curves for vessel and bulk locations in 20μm-thick cryosections (n=40 pairs of bulk and vessel measurements from 4 brains). Graph shows median ± interquartile range. (d) Left: representative western blot of myosin-IIa expression in scramble control and myosin-IIa KO endothelial cells. Right: quantification of all western blots (one per experiment, 3 experiments). Graph shows mean ± standard deviation. (e) Force-indentation curves for scramble control and myosin-IIa KO endothelial cells (Scr: n=73, Myh9: n=60, 3 experiments). Graph shows median ± interquartile range. (f) Young’s modulus of the collagen-HA gel mixture used in the microfluidic devices (n=81, 9 gels, 3 experiments). Lighter dots represent each indentation value and darker dots represent the median value per experiment. Graph shows median ± interquartile range. g-l, Dextran perfusions. (g) Proton NMR of dextran methacrylate, 70% degree of functionalization (calculated based on a previously published procedure, see methods). (h) Mean Young’s modulus of perfused dextran gels for both soft and stiff gel formulations (n=3 perfusions). For each perfusion, a gel sample was kept for nanoindentation. Graph shows mean ± standard error of the mean. (i) Young’s modulus of vessels in brain cryosections from perfusions in (h) measured with AFM (Soft: n=25, Stiff: n=21, 3 brains). Statistical analysis: two-sided T-test. Graph shows mean ± standard deviation. (j) Percentage of cells adhered to vessels after 24h of ex vivo culture in control, soft and stiff dextran brain slices (Control: n=11, Soft: n=11, Stiff: n=10, 3 experiments). Statistical analysis: one-way ANOVA test followed by a post hoc Tukey’s test. Graphs show mean ± standard deviation. (k) Violin plots of invasion depths for cells moving through the bulk brain tissue after 24h of ex vivo culture in control, soft and stiff dextran brain slices (Control: n=517, 11 slices; Soft: n=727, 11 slices; Stiff: n=741, 10 slices; 3 experiments). (l) Violin plots of invasion depths for cells moving along the vasculature after 24h of ex vivo culture in control, soft and stiff dextran brain slices (Control: n=1349, 11 slices; Soft: n=1278, 11 slices; Stiff: n=1050, 10 slices, 3 experiments). (m) Mean Young’s modulus of the stiff alginate gel formulation in (n) (n=3 experiments). Graph shows mean ± standard error of the mean. (n) Violin plots of cell velocities for cells migrating between alginate gels (Soft-Stiff: n=450, Stiff–Stiff: n=428, 4 experiments). The soft gel formulation was the same as in Fig. 5. For all violin plots, dots represent the median values per slice (k,l) or experiment (n). Statistical analysis: two-sided Mann–Whitney test or a Kruskal–Wallis test followed by a post hoc Dunn’s test for multiple comparisons in k,l. Graphs show median ± interquartile range. For western blot quantification, protein expression was normalized to loading control (β-actin), and to the expression on the scramble control condition.

Extended Data Fig. 9 |. Proposed model of vessel co-option.

The differential stiffness between vessels and parenchyma promotes vessel co-option though a talin-dependent mechanism.

Fig. 1 |. Adherent cells invade brain tissue by migrating along the vasculature.

a, An llustration of the experimental model. b, A brain slice immunostaining of laminin (magenta) and 1205Lu LifeAct–GFP (green) at 2 μm (left) and 22 μm (right) below the top surface. The white arrows indicate zoomed-in cells. Scale bars, 50 μm and 20 μm (insets). The brain slice was created with BioRender.com. c, Violin plots of cell shape circularities for cells adhered to vessels or the bulk brain tissue 24 h after seeding. For reference, the contour and circularity of cells in b, top, are shown in grey. (d) Violin plots of invasion depths for cells adhered to vessels or the bulk brain tissue 24 h after seeding. e,f, A brain slice immunostaining of laminin (magenta) and LifeAct–GFP (green) (maximum projection, 24-μm-thick stack) of cancerous cell lines (WM983B, U2OS, MCF7, MDA-MB-231, A375P and A375M2) (e) and non-cancerous cell lines (nHDF, MCF10A and HBVPC) (f). Scale bar, 20 μm. g–i, Violin plots of cell shape circularities for cells on vessels or in the bulk for WM983B, U2OS, MCF7 and MDA-MB-231 (g); A375P and A375M2 (h); and nHDF, MCF10A and HBVPC (i). j–l, Violin plots of invasion depths for cells on vessels or in the bulk for WM983B, U2OS, MCF7 and MDA-MB-231 (j); A375P and A375M2 (k); and nHDF, MCF10A and HBVPC (l). For 1205Lu: n = 463 for bulk and n = 584 for vessel from seven slices and four experiments. For WM983B: n = 295 for bulk and n = 254 for vessel from four slices and four experiments. For U2OS: n = 176 for bulk and n = 268 for vessel from four slices and three experiments. For MCF7: n = 146 for bulk and n = 223 for vessel from six slices and three experiments. For MDA-MB-231: n = 326 for bulk and n = 523 for vessel from six slices and three experiments. For A375P: n = 540 for bulk and n = 532 for vessel from eight slices and four experiments. For A375M2: n = 500 for bulk and n = 356 for vessel from six slices and four experiments. For nHDF: n = 382 for bulk and n = 651 for vessel from seven slices and four experiments. For MCF10A: n = 168 for bulk and n = 459 for vessel from five slices and four experiments. For HBVPC: n = 130 for bulk and n = 473 for vessel from five slices and three experiments. The graphs show median ± interquartile range. The dots represent the medians per slice. For the statistical analysis, a two-sided Mann–Whitney or a Kruskal–Wallis test followed by a post hoc Dunn’s test (for h and k) was performed.

Fig. 2 |. Talin-driven migration regulates vessel co-option.

a, A brain slice immunostaining of paxillin (grey), LifeAct–GFP (green) and nuclei (Hoechst, blue) (maximum projection of 24- and 3-μm-thick stacks (inset)). Scale bars, 10 μm and 5 μm (inset). Yellow dashed lines label the zoomed areas. The images are representative of three experiments. b–l, KO experiments. Graphs show data for cells adhered to the vasculature. Violin plots of cell shape circularities for scramble control (Scr) and CD44-KO cells (Scr: n = 652 and CD44 KO: n = 511 for eight slices and four experiments). c, The mean invasion depth per slice for scramble control and CD44-KO cells (n = 8 for four experiments). d, Violin plots of cell shape circularities for scramble control and integrin β1-KO cells (Scr: n = 720 and Itgβ1 KO: n = 354 for seven slices and four experiments). e, The mean invasion depth per slice for scramble control and integrin β1-KO cells (n = 7 for four experiments). f, Violin plots of cell shape circularities for scramble control, zyxin-KO and paxillin-KO cells (Scr: n = 593 for seven slices; Zyx KO: n = 704 for eight slices; and Pxn KO: n = 597 for eight slices (four experiments)). g, Violin plots of cell shape circularities for scramble control and talin 1-KO cells (Scr: n = 826 for seven slices; Tln1 KO: n = 725 for eight slices (four experiments)). h, Immunostaining of laminin (magenta) and LifeAct–GFP (green) showing scramble control, talin 2-KO and talin 1- and talin 2-KO cells (maximum projection; scale bar, 10 μm). i, Violin plots of cell shape circularities for scramble control, talin 2-KO and talin 1- and talin 2-KO cells (Scr: n = 717, Tln2 KO: n = 640 and Tln1 + Tln2 KO: n = 544 for eight slices and four experiments). j, The mean invasion depth per slice for scramble control, zyxin-KO and paxillin-KO cells (n = 7 for Scr and n = 8 for Zyx KO and Pxn KO for four experiments). k, The mean invasion depth per slice for scramble control and talin 1-KO cells (n = 7 for Scr and n = 8 for Tln1 KO for four experiments). l, The mean invasion depth per slice for scramble control, talin 2-KO and talin 1- and talin 2-KO cells (n = 8 for four experiments). The violin plots show the median ± interquartile range, and the dots represent median values per slice. n.s., not significant. For the statistical analysis, a two-sided Mann–Whitney or a Kruskal–Wallis test followed by a post hoc Dunn’s test (for f and i) was performed. The mean invasion depth graphs show the mean ± standard deviation. For the statistical analysis, a two-sided t-test or an ordinary one-way analysis of variance test followed by a post hoc Tukey’s test (for j and l) was performed.

Fig. 3 |. Talin-KO tumours have smoother boundaries and lose the association with the vasculature at the invasive front.

a, Immunostaining of laminin (magenta) and LifeAct–GFP (green) showing scramble control (Scr) and talin 1- and talin 2-KO tumours 1 week after injection. Yellow dashed lines label zoomed areas. Scale bars, 500 μm and 10 μm (inset). b, Tumour volumes for scramble control and talin 1- and talin 2-KO tumours (n = 4 animals). The graph shows the median ± interquartile range. For the statistical analysis, a two-sided Mann–Whitney test was performed. c, The edge roughness ratio for scramble control and talin 1- and talin 2-KO tumours (Scr: n = 22 and Tln1 + Tln2 KO: n = 21 for four animals). n.s., not significant. The lighter dots represent single plane values and the darker dots the mean value per tumour. The graph shows the mean ± standard deviation. For the statistical analysis, a two-sided t-test was performed.

Fig. 4 |. The ECM composition of the basement membrane is not sufficient to drive spreading and invasion of melanoma cells along the vasculature.

a, Decellularized slices. Brain slice immunostaining of laminin (magenta), LifeAct–GFP (green) and nuclei (Hoechst, grey) (maximum projection; scale bar, 100 μm). b, The percentage of cells adhered to the basement membrane (n = 3 experiments). c, Violin plots of cell shape circularities for cells adhered to the basement membrane (vessels) or the bulk (bulk: n = 191, vessel: n = 743 for 11 slices and three experiments). The dots represent the median values per experiment. For the statistical analysis, a two-sided Mann–Whitney test was performed. The graph shows the median ± interquartile range. d, The mean invasion depth for cells adhered to vessels or the bulk (n = 3 experiments). For the statistical analysis, a two-sided paired t-test was performed. The graph shows the mean ± standard deviation. e, Microfluidic device. Schematic of the device modelling in vitro vessel co-option. f, Immunostaining of laminin (grey), 1205Lu LifeAct–GFP (green), HMBEC LifeAct–Ruby (magenta) and nuclei (Hoechst, blue). Top: maximum projection. Bottom: XZ plane along the dashed line. Scale bar, 20 μm. g, 1205Lu cell sphericity on the micro-vessel or in the bulk gel (bulk: n = 47, vessel: n = 47 for eight devices and three experiments). h, Representative 3D cell shapes from g (sphericities indicated, S). i, Immunostaining of laminin (magenta), 1205Lu LifeAct–GFP (green) and HMBEC LifeAct–Ruby (yellow) in control, BME-coating and BME-gel devices (3D views). Scale bar, 100 μm. j, The sphericity of cells adhered to the micro-vessel in control, BME-coating and BME-gel devices (control: n = 46 for eight devices; BME coating: n = 61 for six devices; BME gel: n = 50 for seven devices and three experiments). k, The velocity of cells along the micro-vessel in control or BME gel devices (control: n = 83, BME gel: n = 65 for eight devices and three experiments). l,m, Immunostaining of laminin (grey), 1205Lu LifeAct–Ruby (magenta), HMBEC iCasp9–GFP (green) and nuclei (Hoechst, blue) in control (l) and 5 nM CID (m) devices. Scale bar, 20 μm. n,o, The cell sphericity (n) and laminin intensity (o) of cells spreading on control or CID-induced endothelial cell apoptosis devices (control: n = 26 for seven devices; CID 5 nM: n = 54 for 11 devices (for four experiments)). Only cells with substantial laminin intensity are shown for CID. The full data are in Extended Data Fig. 6. n.s., not significant. The lighter dots and the darker dots represent individual cells and device medians. For the statistical analysis, a two-sided Mann–Whitney or a Kruskal–Wallis test followed by a post hoc Dunn’s test (for j) was performed. The graphs show the median ± interquartile range.

Fig. 5 |. The differential stiffness of the vessel network versus surrounding tissue promotes cell spreading and invasion.

a, The mean invasion depth per slice for scramble control (Scr) and vinculin-KO cells adhered to the vasculature (n = 5 for three experiments). For the statistical analysis, a two-sided t-test was performed. The graph shows the mean ± standard deviation. b, Isolectin-IB4 staining of vasculature in a 20 μm cryosection. Scale bar, 100 μm. c, Left: isolectin-IB4 images showing vessel and bulk locations for AFM indentation (yellow dots). Right: corresponding force-indentation curves. The grey dashed lines show curve fittings. Scale bar, 20 μm. d, The Young’s modulus of vessels and surrounding tissue (n = 40 for four brains). For the statistical analysis, a two-sided Wilcoxon test was performed. e, The Young’s modulus of scramble control and myosin-IIa-KO HMBECs (Scr: n = 73, Myh9 KO: n = 60 for three experiments). For the statistical analysis, a two-sided Mann–Whitney test was performed. The lighter and darker dots represent each indentation and the median value per experiment. f, Immunostaining of HMBEC LifeAct–Ruby (grey), 1205Lu LifeAct–GFP (green) and nuclei (Hoechst, blue) in scramble control and myosin-IIa-KO micro-vessels (maximum projection; scale bar, 20 μm). g,h, The laminin intensity (g) and cell sphericity (h) for cells spreading on scramble control or myosin-IIA-KO devices (control: n = 68 for ten devices; Myh9 KO: n = 70 for 14 devices (three experiments)). For the statistical analysis, a two-sided Mann–Whitney test was performed. The lighter and darker dots represent each cell value and device medians. i, A schematic of the dextran perfusion experiment. The brain slice image shows labelled dextran gel. Scale bar, 100 μm. j, The mean Young’s modulus for each perfused dextran gel (n = 3 experiments). The graph shows the mean ± standard error of the mean. k, A brain slice immunostaining of laminin (blue, inset), isolectin or dextran (magenta) and LifeAct–GFP (green) for brains perfused with isolectin-IB4, a soft or a stiff dextran gel (maximum projection; scale bars, 100 μm and 20 μm (inset)). l, Violin plots of cell shape circularities for cells adhered to the vasculature on control, soft or stiff dextran (control: n = 1,349 for 11 slices; soft: n = 1,278 for 11 slices; stiff: n = 1,050 for 10 slices (3 experiments)). The dots represent the median circularity per slice. For the statistical analysis, a Kruskal–Wallis test, followed by a post hoc Dunn’s test, was performed. m, The mean invasion depth per slice for cells adhered to the vasculature on control, soft or stiff dextran (control: n = 11, soft: n = 11, stiff: n = 10 for three experiments). For the statistical analysis, an ordinary one-way analysis of variance test, followed by a post hoc Tukey’s test, was performed. The graph shows the mean ± standard deviation. n, A schematic of the double-layer gel assay; orthogonal view: fluorescent beads (blue, top gel; magenta, bottom gel) and LifeAct–GFP (green). Scale bar, 20 μm. o, The mean Young’s modulus of each gel condition (n = 3 experiments). The graph shows mean ± standard error of the mean. p, Violin plots of cell velocities for cells migrating between gels (soft–soft: n = 163, soft–stiff: n = 211, stiff–stiff: n = 177 for three experiments). n.s., not significant. The dots represent the median cell velocity for each experiment. For the statistical analysis, a Kruskal–Wallis test, followed by a post hoc Dunn’s test, was performed. The figure graphs show the median ± interquartile range unless stated otherwise. The drawings in c and i were created with BioRender.com.