Matrix mechano-sensing at the invasive front induces a cytoskeletal and transcriptional memory supporting metastasis

Abstract

The extracellular matrix (ECM) controls tumour dissemination. We characterise ECM organization in human and mouse tumours, identifying three regions: tumour body, proximal invasive front and distal invasive front. Invasive areas show increased matrix density, fibre thickness, length, and alignment, with unique radial fibre orientation at the distal invasive front correlating with amoeboid invasive features. Using patient samples and murine models, we find that metastases recapitulate ECM features of the primary tumour. Ex vivo culture of murine cancer cells isolated from the different tumour regions reveals a spatial cytoskeletal and transcriptional memory. Several in vitro models recapitulate the in vivo ECM organisation showing that increased matrix induces 3D confinement supporting Rho-ROCK-Myosin II activity, while radial orientation enhances directional invasion. Spatial transcriptomics identifies a mechano-inflammatory program associated with worse prognosis across multiple tumour types. These findings provide mechanistic insights into how ECM organization shapes local invasion and distant metastasis.

Fig. 1. Regional organisation of the extracellular matrix in solid tumours.

a TMA cores from Tumour Body (TB) and Invasive Front (IF) of primary melanoma (n = 53), stained with Gomori trichrome. Insets show regions of interest (ROIs). Yellow lines separate proximal (PIF) and distal invasive fronts (DIF). Scale: 200 µm. b ROIs from TB, PIF, DIF: trichrome staining, QuPath Pixel Count maps, Haralick Entropy alignment maps. Scale: 50 µm. c Quantification of Matrix Percentage (%), Matrix Intensity (OD Mean arb. units), and Matrix Alignment (Haralick Entropy arb. units) in melanoma (n = 53 patients). d ROIs from TB, PIF, DIF: 8-bit matrix images and CurveAlign alignment maps. Scale: 50 µm. e ROI quantification: Median Fibre Length, Width, Straightness, Distance to Fibres, Alignment Score, Fibre number/mm² (n = 53 patients). f Angle deviation histogram comparing PIF and DIF (n = 53 patients). g TMA cores from TB and IF of primary breast cancer (BRCA) (n = 30 patients), Masson trichrome stained. Insets show ROIs. Yellow lines separate PIF from DIF. Scale: 100 µm. h BRCA ROIs: trichrome staining, QuPath Pixel Count maps, Haralick Entropy alignment maps. Scale: 20 µm. i Quantification of Matrix Percentage (%), Matrix Intensity (OD Mean arb. units), and Matrix Alignment (Haralick Entropy arb. units) in BRCA (n = 30 patients). j BRCA ROIs: 8-bit matrix images and CurveAlign alignment maps. Scale: 20 µm. k BRCA ROI quantification: Median Fibre Length, Width, Straightness, Distance to Fibres, Alignment Score, Fibre number/mm² (n = 30 patients). l Angle deviation histogram comparing PIF and DIF in BRCA (n = 30 patients). One-way ANOVA with Tukey post-hoc (c, e, i, k), displayed as truncated violin plots showing medians and quartiles. Histograms (f, l) show mean and standard error of the mean (SEM). Cartoons in Fig. 1(a, g) created in BioRender. Maiques, O. (2025) https://BioRender.com/f02u038https://BioRender.com/n61t687.

Fig. 2. Regional organisation of ECM is associated with cytoskeletal remodelling, local invasion and metastatic spread in human tumours.

a Representative multiplex images of Tumour Body (TB), Proximal Invasive Front (PIF), and Distal Invasive Front (DIF) cores in melanoma patients. Images show merged pan melanoma, pMLC2, and MLC2 markers, with individual stainings. Nuclei are stained blue. QuPath maps display tumor mask, circularity, pMLC2 H-score, and MLC2 H-score. Scale bar: 200 µm. b Quantification of tumour circularity (0–1), Myosin II activity (pMLC2 H-score, 0–300), and MLC2 H-score (0–300). n = 53 patients. c Representative images of melanoma invading cells across PIF and DIF. Scale bar: 30 µm. d Percentage of individual and grouped cells invading PIF and DIF in melanoma. n = 53 patients. e Number of invading melanoma cells/mm² in PIF and DIF. n = 53 patients per group. f Representative images of TB, PIF, and DIF in breast cancer (BRCA). Images show merged CK19, pMLC2, and MLC2 markers, with individual staining. QuPath maps display tumour circularity. Scale bar: 100 µm. g Quantification of tumour circularity, Myosin II activity, and MLC2 H-score in BRCA. n = 30 patients. h Images of BRCA invading cells across PIF and DIF. Scale bar: 20 µm. i Percentage of individual and grouped cells invading PIF and DIF in BRCA. n = 30 patients. j Number of invading BRCA cells/mm² in PIF and DIF. n = 30 patients. k Melanoma invasion by stage (II, III, IV; n = 24 patients) and BRCA invasion by lymph node status (N0, N1; n = 30 patients). l Trichrome staining and fibre extraction in melanoma and BRCA at PIF and DIF. Scale bar: 30 µm. One-way ANOVA with Tukey post-hoc (b, g, k) displayed as truncated violin plots. Two-way ANOVA with Tukey post-hoc (d, i), and Mann-Whitney U test (e, j). Bar graphs (d, i) show mean percentages with standard deviation error bars. Boxplots (e, j) show medians, interquartile ranges, and individual data points. Cartoons in Fig. 2(a, f, l) created in BioRender. Maiques, O. (2025) https://BioRender.com/f02u038https://BioRender.com/n61t687.

Fig. 3. ECM in vivo organisation is associated with a cytoskeletal memory.

a Diagram showing 4599 mouse melanoma orthotopic model. Intradermal cancer cell injection leads to development of primary tumours and spontaneous lung metastasis. b Histological images of primary tumour (TB, PIF, DIF) and metastasis (micro/macro) showing trichrome staining, QuPath intensity map, and CurveAlign fibre alignment. Scale bars: 100 µm (primary), 50 µm (micro), 200 µm (macro). c Quantification of Matrix Percentage (%), Matrix Intensity (OD Mean arb. units), and Matrix Alignment (Haralick Entropy arb. units). n = 11 mice. d Angle deviation line plot comparing PIF (red) and DIF (green) fibre distributions. n = 11 mice. e Quantification of individual versus grouped cell invasion at PIF/DIF, and invading cells/mm² count. n = 11 mice. f Multiplex images of primary tumour and metastases showing merged and individual channels (mCherry, pMLC2, MLC2). Nuclei in blue. Scale bars: 100 µm (primary), 20 µm (micro), 300 µm (macro). g Quantification of tumour circularity (0-1), pMLC2 H-score (0-300), and MLC2 H-score (0-300). n = 11 mice. h Ex vivo analysis of cells isolated from TB, PIF, DIF and metastasis in Collagen I, showing F-actin and pMLC2 staining. Nuclei in blue. i Quantification of cell rounding index and pMLC2/cell area from ex vivo isolates. n = 100, 76, 90 and 90 cells per group, 3 independent experiments. j Spheroid invasion assay comparing TB (green) and DIF (red) cells and TB (green) and Met (red), with quantification at 600 µm from core. Scale bars: 200 µm, inset 100 µm. n = 13 and 14 spheroids per group. k Lung colonisation assay showing TB (green) versus DIF (red) cells in lung colonization. Nuclei in blue. n = 16 mice. One-way ANOVA with Šídák (c, g) and Tukey post-hoc (i) displayed as truncated violin plots showing medians and quartiles. Unpaired t-test (e right, j, k) shown as bar graphs with mean ± SD (e left, j) and boxplots showing median, interquartile range and individual data points (e right). Two-way ANOVA with Sidak post-hoc (e left). Angle deviation data (d) presented as line plot with SEM error bars. Cartoons in Fig. 3 (a, h, j, k) created in BioRender. Maiques, O. (2025) https://BioRender.com/z35l297https://BioRender.com/f02u038.

Fig. 4. Collagen I accumulation induces cytoskeletal remodelling.

a Experimental setup showing protease inhibitor (PI) treatment of cells in a collagen I matrix. Images show collagen reflectance and CurveAlign matrix alignment maps at 4 h and 24 h. Scale bar: 5 µm. b Matrix analysis quantifying fibre length, width, straightness, distance, alignment score, and fibre number/mm² with/without PI at 4 h and 24 h. n = 27, 25, 26 and 27 samples analysed from three independent experiments. Sidak one way. c Collagen reflectance images with gap analysis map and quantification comparing control versus PI treatment. n = 1066 and 2052 gaps analysed, three independent experiments. Scale bar: 20 µm. d Immunofluorescence of F-actin (magenta) and pMLC2 (green) in 4599 cells ±PI. Nuclei in blue. Scale bar: 10 µm. e Quantification of cell rounding index and pMLC2/cell area. n = 134, 155, 122 and 155 cells from three independent experiments. f Collagen images at increasing concentrations (1.7, 3.5, 6.6 mg/mL) by confocal reflectance microscopy. Scale bar: 20 μm. g Matrix analysis quantifying fibre length, width, straightness, distance, alignment score, and fibre number/mm² at increasing concentrations (1.7, 3.5, 6.6 mg/mL). n = 90, 78 and 55 images, three independent experiments. h Collagen reflectance images with gap analysis comparing 1.7 and 6.6 mg/mL collagen. n = 3584 and 10493 gaps analysed, three independent experiments. Scale bar: 20 µm. i Immunofluorescence showing F-actin and pMLC2 in 4599 cells at different collagen concentrations. Nuclei in blue. j Quantification of rounding index and pMLC2/cell area across collagen concentrations. n = 110, 95, 151 cells, three independent experiments. One-way ANOVA with Šídák (b, e) and Tukey post-hoc (g, j) displayed as violin plots showing medians and quartiles. Mann-Whitney U test between (c, h) shown as cumulative plots. Cartoons in Fig. 4(a, d, f, i) created with BioRender Maiques, O. (2025) https://BioRender.com/z57v040https://BioRender.com/f02u038.

Fig. 5. Collagen I accumulation and radial alignment support Myosin II driven invasive and metastatic behaviour.

a Schematic showing how MgCl₂ affects collagen matrix alignment, illustrated in side and top views. b Collagen reflectance images comparing H₂O and MgCl₂ treatment, showing reflectance and CurveAlign matrix alignment maps. Scale bar: 20 µm. c Matrix analysis quantifying fibre parameters (length, width, straightness, distance, alignment score, fibre number/mm²). n = 150 and 132 images, three independent experiments. d Angle deviation histogram comparing control versus MgCl₂-treated collagen fibres (n = 30 samples per group). e AFM elasticity (Pa) measurements of control and MgCl₂-treated collagen. n = 15 per group, two independent experiments. f Reflectance images with gap analysis maps. Scale bar: 20 µm. g Gap analysis quantification comparing control and MgCl₂ collagen fibres. n = 6941 and 6125 gaps analysed, three independent experiments. h Immunofluorescence of F-actin (magenta) and pMLC2 (green) in 4599 cells ±MgCl₂. Nuclei in blue. Scale bar: 20 µm. i Cell rounding index and pMLC2/cell area quantification in 4599 cells ±MgCl₂. n = 140 and 160 cells, three independent experiments. j Spheroid assay showing 4599 cell invasion length in non-aligned (H₂O) /aligned (MgCl₂) matrices. Scale bar: 200 µm. k Area coverage quantification from 4599 tumour spheroids at 900 µm from spheroid core comparing non-aligned (H₂O) /aligned (MgCl₂) matrices. n = 4 independent experiments. l Diagram of 3D invasion assay schematic using educated 4599 cells in non-aligned (H₂O) /aligned (MgCl₂) matrices. m Western blot of ppMLC2 and MLC2 in primed cells. n = 3 independent experiments. The samples derive from the same experiment but different gels for pp-MLC2^T19/S19 and GAPDH, and another for MLC2 were processed in parallel. n 3D invasion images at 0 µm and 50 µm with quantification. Scale bar: 100 µm. n = 3 independent experiments. o Diagram of in vivo lung invasion assay schematic using pre-educated 4599 cells. p Lung parenchyma representative image at 48 h post-injection. Scale bar: 100 µm. ROI quantification at 30 min and 48 h. n = 14 mice, 2 independent experiments. Unpaired t-test (c, i, k, n) and Mann Whitney test (e, f) between -/+ conditions. One-way ANOVA with Šídák post-hoc test (m). Data is displayed as violin plots showing individual data points and medians (c, e, i), boxplots showing median, interquartile range and individual data points (k, n), Bar plot with Mean and SD (m, p), and histogram in (d). Cartoons in Fig. 5(a, h, l, o) created in BioRender. Maiques, O. (2025) https://BioRender.com/z35l297.

Fig. 6. Matrix organisation changes result in Rho-ROCK engagement and uncoupling of Myosin II from vinculin.

a Confocal SoRa images of HT1080 Fibrosarcoma cells showing nuclei (blue), F-actin (magenta), pMLC2 (cyan), and Vinculin (green) at increasing collagen concentrations (1.7-6.6 mg/ml). Scale bars: 10 µm, insets 2 µm. b Quantification of Vinculin/cell area (%) and percentage of pMLC2/Vinculin levels at the indicated collagen I concentrations. Left, n = 35, 28 and 28 cells and right, n = 23, 25 and 22 cells, three independent experiments. c Western blot analysis of Vinculin levels in HT1080 cells grown in different collagen I concentration gels. n = 3 independent experiments. Vinculin and GAPDH were run in the same gel. d Confocal SoRa images of HT1080 Fibrosarcoma cells showing nuclei (blue), F-actin (magenta), pMLC2 (cyan), and Vinculin (green) ± protease inhibitors (PI) at 4 h and 24 h. Scale bars: 10 µm, insets 2 µm. e Quantification of Vinculin/cell area (%) and percentage of pMLC2/Vinculin levels ± PI. Left, n = 23, 18 and 21 cells and right, n = 30, 27 and 25 cells, three independent experiments. f Western blot analysis of Vinculin levels ±PI. n = 3 independent experiments. Vinculin and GAPDH were run in the same gel. g RhoA and RhoC activation by pull-down assays in HT1080 cells ±PI after 2 h treatment, with quantification of GTP-bound and total protein levels. n = 3 independent experiments. h Immunofluorescence images showing F-actin and pMLC2 in HT1080 cells ±PI and ROCK inhibitor (H1152) at 4 h. Scale bar: 20 µm. i Quantification of cell rounding index and pMLC2/cell area from conditions in (h). n = 61, 66 and 59 cells, three independent experiments. Unpaired t-test for pull-down assays (g) and One-way ANOVA with Tukey post-hoc test for all other comparisons (b left, c, e left, f, i). Two-way ANOVA with Tukey post-hoc (b right and e right). Data displayed as violin plots showing individual data points and medians (b left, e left, i), bar plot with mean and SD (b, right, c, e right, f, g). Cartoons in Fig. 6 (a, d, g, h) created in BioRender. Maiques, O. (2025) https://BioRender.com/c83e894.

Fig. 7. Matrix organisation drives a pro-metastatic transcriptional program.

a Mouse melanoma model showing intradermal injection of 4599 cells leading to lung metastasis. Cells isolated from different tumour areas (TB, PIF, DIF, Metastasis) for ex vivo experiments and RNA-seq. b Principal component analysis of 4599 cells’ transcriptome from different tumour regions. n = 3 independent experiments. c Gene overrepresentation analysis comparing PIF, DIF, and MET versus TB, grouped by ECM, Mechano-sensing, Cytoskeleton, and Angiogenesis signatures. d Volcano plots of transcription factors in PIF versus TB and DIF versus TB. Threshold: -0.5<Log2FC < 0.5, -Log10 p-adjusted value. e GSEA analysis of inflammatory pathways, comparing PIF, DIF, and Met to TB. Dotplot shows normalized enrichment scores (FDR < 0.1). f–h Immunoblots showing YAP, p-p65, p65, p-STAT3, STAT3, pp-MLC2, and MLC2 protein levels in 4599 cells from different tumour areas (n = 5 independent experiments), collagen concentrations (n = 4), and aligned/non-aligned matrix (n = 3). GAPDH is used as loading control. Quantification in Supplementary Fig. 7g. The samples for each panel derive from the same experiment but different gels for STAT3, MLC2 and GAPDH, another for p-STAT3^Y705, pp-MLC2^T19/S19, another for p65, another for p-p65^s536, and another YAP were processed in parallel. i Top, diagram of experimental setting showing priming in ± MgCl₂ (7 days), and further culture in plastic (6 days). Bottom, corresponding p-STAT3/STAT3 protein levels in 4599 cells and quantification. GAPDH is used as loading control. n = 3 independent experiments. The samples derive from the same experiment but different gels for p-STAT3^Y705 and GAPDH, and another for STAT3 were processed in parallel. j Left, protein levels of p-STAT3/STAT3 in HT1080 cells after YAP or NF-κB knockdown with PI treatment (24 h). GAPDH is used as loading control. Corresponding quantification of p-STAT3/STAT3. n = 3 independent experiments. The samples derive from the same experiment but different gels for p-STAT3^Y705 and GAPDH, another for STAT3, another for YAP1 and another for NF-Ƙβ1 were processed in parallel. k Top, confocal images showing F-actin, pMLC2, and nuclei after PI treatment with siYAP1, siNF-κB1, or siSTAT3. Bottom, corresponding cell rounding index and pMLC2/cell area quantification. Scale bar: 10 µm. n = 91, 97, 91, 91 and 92 cells, three independent experiments. l Multiplex immunohistochemistry of YAP (red), p65 (green), p-STAT3 (magenta) in 4599 primary tumours and lung metastases. Scale bars: 200 µm (primary, macro), 50 µm (micro, insets). m Nuclear H-score quantification (0–300) of transcription factors across tumour regions from (l). n = 11/group. One-way ANOVA with Šídák (I, j, k) and Tukey (m) post-hoc test. Data displayed as violin plots showing individual data points and medians (k, m), bar plot with mean and SD (I, j). Cartoons in Fig. 7 (a, f, h, i, j, l) created with BioRender.com. created with BioRender Maiques, O. (2025) https://BioRender.com/z57v040https://BioRender.com/f02u038.

Fig. 8. Matrix organization drives a pro-metastatic transcriptional program.

a Venn diagram showing Mechano-Inflammatory signature (MIS) derived from 4599 cells extracted from TB and DIF and from HT1080 treated with PI for 24 h. b Overlapping genes from (a) - conserved signature- were scored on the Visium array spots in a representative section of human melanoma lymph node metastasis. Annotations correspond to TB and IF areas. n = 4 slices were analysed. Box plot shows the comparison between TB and IF annotations. n = 12 TB annotations and n = 12 IF annotations were analysed. c Volcano plot showing DEGs upon IF and TB comparison from spatial transcriptomics defined annotations. Highlighted in red DEGs that present average log2 fold change >0.5 and -log₁₀ adjusted p-value. d Heatmap shows median scaled by row, displaying Gene Ontology Biological Pathways (GOBP) and Molecular Functions (GOMF) terms separated by regions (TB and IF) and slices from Visium array analysed in melanoma metastasis samples. Z-score scale. e Representative merged multiplex images of primary and metastatic lesions of melanoma tumours. (bottom) Inset of merged multiplex. Nuclei were visualized with Haematoxylin stain (blue). Scale bar: 100 µm. Scale bar insets: 20 µm. f H-score quantification (0-300) of YAP1, p65 and pSTAT3 in the primary and metastatic lesions. n = 53 primary and 45 metastatic lesions. g Kaplan-Meier curve showing overall survival with the quantification of amoeboid cells YAP1, p65 and pSTAT3 positive in the DIF area (left). Kaplan-Meier curve displays the overall survival comparing high and low fibre alignment (right). One-way ANOVA with Tukey post-hoc test (f). Data displayed as violin plots showing individual data points and medians (f). Cartoons in Fig. 8 (b, e) Created in BioRender. Maiques, O. (2024) https://BioRender.com/h15l089.

Fig. 9. Model.

Matrix organisation in primary tumours and metastasis: matrix heterogenous distribution in human solid tumours coupled to changes in the cytoskeleton and invasive patterns. The observations in primary tumour are mirrored in metastatic lesions. Mechanism at the cellular level (bottom): cancer cells under confinement, imposed by increased matrix, retract their protrusions and contract their cytoskeleton reducing cell-matrix adhesion. This retraction is accompanied by vinculin uncoupling from adhesions and activation of RhoA-ROCK-Myosin II axis leading to long-term transcriptional reprogramming orchestrated by YAP1-NF-κB and STAT3. Figure created in BioRender. Maiques, O. (2025) https://BioRender.com/v18d918.