Ethical considerations
Use of animals: we confirm that our research complies with all relevant ethical regulations. Mouse intratongue injection was ethically assessed and authorized by the National Animal Experiment Board and in accordance with The Finnish Act on Animal Experimentation (animal license number ESAVI/6253/2024). All efforts were made to minimize animal suffering and to reduce the number of animals used. All experiments respected the maximum tumour diameter (15 mm) permitted by the authorization bodies.
Patient data: patient samples were obtained at the Department of Otorhinolaryngology—Head and Neck Surgery at the Turku University Hospital under the Finnish Biobank Act, with written informed consent from the sample donors (§279, 9/2001). On collection, the samples were given an arbitrary identifier and no patient identifiers, excluding age, gender, prior treatments and histopathological features, were available or recorded. Tissue samples were snap frozen with liquid nitrogen and stored at –80 °C until further processing.
The study and utilization of human tissue samples were approved by the Finnish National Authority for Medicolegal Affairs (V/39706/2019), the Institutional Review Board at the Helsinki University Hospital (HUS/745/2021) and a research permission was granted (HUS/85/2021).
Formalin-fixed and paraffin-embedded tumour samples were obtained from the pathology archives of the Helsinki Biobank. Patient consent was waived due to the retrospective nature of the data in accordance with approval from the Finnish National Supervisory Authority for Welfare and Health and the Regional Ethics Committee of the University of Turku. The authors affirm that the study was conducted following the rules of the Declaration of Helsinki of 1975, revised in 2013.
Cell lines: UT-SCC-11 (T1) and UT-SCC-103 (T3) cell lines generated at the Turku University Hospital have undergone scientific evaluation by Auria Biobank with a positive decision of release (AB22-7195) to be used in the study. Use of these cell lines for other purposes requires ethical approval and permission from the Auria Biobank.
TCGA data acquisition and analysis
TCGA head and neck squamous cell carcinoma dataset was retrieved and filtered for patient IDs with laryngeal cancer as the primary tumour site. Pathology reports were then reviewed to assess the tumour subsite and the involvement of vocal folds. Raw files were downloaded from the Xena browser (https://xenabrowser.net/). Differentially expressed genes (log2[fold change]) were assessed using Bioconductor R package reproducibility-optimized test statistic (ROTS; v1.14.0), defining genes with FDR < 0.05 as differentially expressed87. GO was performed using clusterProfiler (v4.8.3) in R88.
TMA
TMA blocks with duplicate core biopsies were prepared from the formalin-fixed and paraffin-embedded samples using a TMA Grand Master (3DHISTECH) at the Helsinki University Hospital. A total of 198 patients with known TNM staging (a recognised system to describe the extent of the disease, based on tumour characteristics, spreading to nearby lymph nodes and metastasis to other organs) and survival endpoints were included in the study.
Chick embryo CAM model
The shell of fertilized chicken eggs was cleaned with 70% ethanol before starting development, that is, placing the eggs in a humidified incubator (50% moisture + 37 °C). On day 3 of development, a small hole was made with a needle and tweezers in the eggshell to drop the CAM away from the shell. On developmental day 7, the hole was widened with tweezers to place a plastic ring on the CAM. One million UT-SCC-11 or UT-SCC-103 cells were implanted inside the ring in 20 µl of 50% Matrigel (Corning) diluted in phosphate-buffered saline (PBS) supplemented with control or drug treatment (dimethyl sulfoxide (DMSO) or 10 μM of TEAD inhibitor K-975), after which the hole was covered with a parafilm to avoid drying of the CAM. Tumours were harvested 4–5 days post-implantation by placing the eggs on ice for 30 min before dissecting, weighing and fixing the tumours in 10% phosphate-buffered formalin (pH 7; VWR).
Intratongue mouse model
Female immunocompromised mice (NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ; Charles River) between 9 and 12 weeks of age were treated with a painkiller and anti-inflammatory mixture (0.07 mg kg−1 of buprenorphine and 5 mg kg−1 of carprofen in 100 ml of PBS injected intraperitoneally) 30 min before tongue injection and twice a day for 3 days after injection. Mice were anaesthetized (75 mg kg−1 of ketamine and 6 mg kg−1 of xylazine in 200 μl of PBS injected subcutaneously) for injections. A single-cell suspension (30,000 cells) was prepared in 30% growth-factor-reduced Matrigel (Corning) diluted in PBS and was injected into the tongue using syringes (0.3 ml, 0.30 mm (30 G) × 8 mm; BD Micro-Fine). Mice were given softened food ad libitum to ensure that they could eat regardless of tumour growth. On day 7 post-injection, 80 mg kg−1 of the K-975 inhibitor (diluted in 10% DMSO (D2650, Sigma), 40% PEG400 (63012, Reidel-de Haen), 5% Tween 80 (P4780, Sigma) and 150 mM of NaCl) was orally administered to mice daily until the animals were killed. Oral gavaging was performed by coating needles in 24% sucrose (diluted in sterile water) to reduce oesophageal irritation89. Animal weight and tumour growth were closely monitored for 2 weeks (endpoint). Mice were killed and tongues were collected in 10% phosphate-buffered formalin (FF-Chemicals) or freshly frozen in an optimal cutting temperature compound (Tissue-Tek). Formalin-fixed tissues were embedded in paraffin and 4–5-µm sections were stained with haematoxylin and eosin and Masson’s trichrome using standard protocols. Optimal-cutting-temperature-embedded tissues were used to assess the tissue stiffness using AFM (see the ‘AFM’ section).
AFM
For all mechanical measurements, tissues were freshly frozen in the optimal cutting temperature compound (OCT) and cut into either 16 or 30 μm cryosections, as indicated, at –20 °C, followed by immediate transfer to poly-L-lysine-coated glass-bottom dishes or slides. Before the measurements, PBS containing complete EDTA-free protease inhibitor (Sigma) was utilized to thaw sections.
Tongue xenograft tumour samples (30 μm cryosections) were incubated with 40 µg ml−1 CNA35-GFP to label collagen and 5 µg ml−1 DAPI in PBS containing protease inhibitors and 0.2% Triton X-100 for at least 1 hour at room temperature. AFM measurements were done on the same day on a JPK NanoWizard II system (Bruker Nano) with a CellHesion module mounted on a ZEISS LSM510 confocal microscope (Carl ZEISS NTS) utilizing JPK SPM Control Software (v4.2). Patient biopsy (16 μm cryosections) measurements were performed within 30 minutes of tissue thaw on a JPK NanoWizard 4 system (Bruker Nano) with a CellHesion module mounted on an Eclipse Ti2 inverted fluorescence microscope (Nikon) and operated via JPK SPM control software (v6). Silicon nitride cantilevers (nominal spring constant, 0.06 N m−1; spherical 4.5-μm-diameter tip; Novascan Technologies) were used to assess xenograft stiffness and MLCT triangular silicon nitride cantilevers (Bruker) were used to measure basement membrane stiffness in patient biopsies. Spring constant and deflection sensitivity were calibrated in fluid via the thermal noise method90. All AFM measurements were performed utilizing a 5 × 5 point grid (25 µm × 25 µm). At least five regions were measured per sample. Forces of up to 3 nN were applied at 20 µm s−1 constant cantilever velocity. All analyses were performed with JPK Data Processing Software (v4.2 for tongue xenograft tissues and v6 for patient biopsies; Bruker Nano) by first removing the offset from the baseline of raw force curves, then identifying the contact point and subtracting cantilever bending before fitting the Hertz model with the correct tip geometry to determine the Young modulus.
Cell lines and culture
Spontaneously immortalized human epidermal keratinocyte-derived cell line HaCaT (ref. 91), obtained from the collection of BioCity Turku (University of Turku, Finland; referred to as NC cells in this manuscript), UT-SCC-11 (T1 human glottic laryngeal cancer, Turku University Hospital) and UT-SCC-103 (T3 human glottic laryngeal cancer, Turku University Hospital) cells were cultured in Dulbecco’s modified Eagle’s medium (Sigma-Aldrich) supplemented with 10% fetal bovine serum (Sigma-Aldrich), 2 mM of L-glutamine (Sigma-Aldrich) and 1% Minimal Essential Medium non-essential amino acid solution (Sigma-Aldrich) at 37 °C, 5% CO2. UT-SCC-11 and UT-SCC-103 cell lines generated at the Turku University Hospital have undergone scientific evaluation by Auria Biobank with a positive decision of release (AB22-7195) to be used in the study. Use of these cell lines for other purposes requires ethical approval and permission from the Auria Biobank. All cell lines were regularly tested for mycoplasma using a MycoAlert Mycoplasma Detection Kit (LT07-418, Lonza) and MycoAlert Assay Control Set (LT07-518, Lonza) to ensure mycoplasma-free culturing. Cells were washed with PBS (Gibco) and detached enzymatically with a 0.25% trypsin-EDTA solution (L0932, Biowest).
Proliferation assay
Plastic (Corning) or Softwell Easy Coat (Matrigen; stiffness values, 0.5 kPa, 25 kPa and 50 kPa) 24-well plates were coated with 10 μg ml−1 of collagen I (C8919, Sigma) and 10 μg ml−1 of fibronectin (341631, Sigma) diluted in PBS or 10 μg ml−1 of growth-factor-reduced Matrigel (354230, Corning) diluted in PBS, at 37 °C for 1 h. Coated plates were washed three times with PBS before seeding 10,000 cells in a culture medium (approximately 5,000 cells cm−2). Time-lapse live imaging was performed using Incucyte S3 or ZOOM Live-Cell Analysis System for 96 h with 2-h imaging intervals (×10 objective). The medium was changed every second day.
Migration assay
Softwell Easy Coat (Matrigen) 50-kPa 24-well plates were coated with 10 μg ml−1 of collagen I (C8919, Sigma) and 10 μg ml−1 of fibronectin (341631, Sigma) diluted in PBS, at 37 °C for 1 h. The coated plates were washed three times with PBS before seeding 1,000 cells in the culture medium (approximately 500 cells cm−2). Time-lapse live imaging was performed using Nikon Eclipse Ti2-E (×10/0.3 objective) for 24 h with 10-min imaging intervals. Single-cell tracking was performed using TrackMate plugin in ImageJ (National Institutes of Health).
Invasion assay
Here 200,000 cells were seeded in a serum-free medium on Matrigel transwell inserts (354480, Corning) and placed in the culture medium (approximately 666,000 cells cm−2). After 45 h of invasion, uninvaded cells in the inner well were removed with cotton buds and invaded cells were fixed with 4% paraformaldehyde diluted in PBS for 10 min at room temperature (RT). Inserts were washed three times with PBS and stained overnight with 4′,6-diamidino-2-phenylindole (DAPI). Invaded cells were assessed by confocal imaging (3i Marianas CSU-W1; ×20/×0.8 objective), quantifying the number of invaded cells per FoV in ImageJ (National Institutes of Health).
Viability assay
Here 5,000 cells were seeded in a 96-well plate in the culture medium (approximately 15,000 cells cm−2). DMSO (D265, Sigma) or YAP-TAZ-TEAD inhibitors K-975 (HY-138565, MedChemExpress) or IK-930 (HY-153585, MedChemExpress) and Wnt/β-catenin inhibitor iCRT3 (HY-103705, MedChemEpress) were added at concentrations of 10 nM, 30 nM, 100 nM, 300 nM, 1 µM, 3 µM, 10 µM, 30 µM and 100 µM the following day. Relative cell viability was measured as absorbance at 450 nm after 2-h incubation with a cell counting kit at 37 °C as per the manufacturer’s instructions (Cell Counting Kit-8, ab228554) 48 h after the addition of inhibitor treatment.
Western blotting
Cells were kept on ice and washed with cold PBS and lysed with heated (90 °C) TX-lysis buffer (50 mM of Tris-HCl, pH 7.5, 150 mM of NaCl, 0.5% Triton X, 0.5% glycerol, 1% SDS, complete protease inhibitor (Sigma-Aldrich), and PhosSTOP tablet (Sigma-Aldrich)). Lysed cells were scraped into an Eppendorf tube and boiled for 5 min at 90 °C followed by 10 min of sonication and 10 min of centrifugation at 16,200g at 4 °C in a microcentrifuge. Protein concentrations were determined from the supernatant with DC Protein assay (Bio-Rad) as per the manufacturer’s instructions. Samples were boiled at 90 °C for 5 min before protein separation using precast sodium dodecyl sulfate–polyacrylamide gel electrophoresis gradient gels (4%–20% Mini-PROTEAN TGX, Bio-Rad) and transferred onto nitrocellulose membranes with the semi-dry Trans-Blot Turbo Transfer System (Bio-Rad). Membranes were blocked with AdvanBlock-Fluor blocking solution (AH Diagnostics) diluted 1:1 in PBS for 1 h at RT and incubated overnight at 4 °C, with the primary antibodies diluted in an AdvanBlock-Fluor blocking solution. Membranes were washed for 5 min three times with Tris-buffered saline and 0.1% Tween 20, and incubated 1:2,500 with fluorophore-conjugated Azure secondary antibodies (AH Diagnostics) in the blocking solution for 1 h at RT. Membranes were washed three times with Tris-buffered saline and 0.1% Tween 20 for 5 min at RT. Membranes were scanned using an infrared imaging system (Azure Sapphire RGBNIR Biomolecular Imager), and the band intensities were analysed using Image Studio Lite (Licor) by normalizing the signal to GAPDH or HSP70, which were used as loading controls.
The list of antibodies is provided in Supplementary Table 1.
PIV analysis
A custom PIV algorithm was developed in Python to measure the cell velocities within monolayers and derive different indicators of cellular motility. Velocity fields were first extracted by processing sequences of images. In short, each image is divided into interrogation windows: for each window located at position \(\), the local cell displacement \({\boldsymbol}{{\boldsymbol{r}}}\) is quantified by cross-correlating the intensity of two region-of-interest (ROI) images separated by \({{\Delta }}{{t}}\), which allows estimating the local velocity as \({\boldsymbol{v}}_{t}\left({\boldsymbol{x}}\right)=\frac{\boldsymbol{\Delta }{\boldsymbol{r}}}{{\Delta }{t}}\), where the index \({{t}}\) corresponds to the time of the frame pair used to compute the velocity field. We used ROIs of size \(80\times 80\) px2, which are slightly larger than the typically observed cell size of \(\sim 50\) px, with a spatial overlap factor of 50% between different ROIs. To improve statistics, we also performed a temporal average of the so-obtained velocity fields over chunks of 20 frames (200 min), again with a temporal overlap of 50%. The previous parameters were carefully optimized to find the best trade-off between increasing the spatiotemporal resolution and averaging a sufficient number of data samples to obtain smoother velocity maps, which will be indicated in the following with \({{{\boldsymbol{v}}}}_{{{t}}}\left({{\boldsymbol{x}}}\right)\). We then followed Garcia et al.54 to compute the total r.m.s. velocity as \({\boldsymbol{v}}_{\rm{RMS}}^{\rm{tot}}(t)={\sqrt{\left\langle |{\boldsymbol{v}}_{t}{({\boldsymbol{x}})}^{2}\right\rangle }}_{\boldsymbol{x}}\) and the drift-corrected r.m.s. velocity \({\boldsymbol{v}}_{\rm{RMS}}^{\rm{d.}{\rm{c}}.}\left(t\right)=\sqrt{{\left\langle |{\boldsymbol{v}}_{t}^{\rm{d}.{\rm{c}}.}({\boldsymbol{x}}){|}^{{{2}}}\right\rangle }_{\boldsymbol{x}}}\) as spatial averages of the velocity fields, where we have introduced the drift-collected velocity \({\boldsymbol{v}}_{t}^{{\rm{d}}{{.}}{\rm{c}}.}({\boldsymbol{x}})={\boldsymbol{v}}_{{{t}}}({\boldsymbol{x}})-{\left\langle {{\boldsymbol{v}}}_{{{t}}}({\boldsymbol{x}})\right\rangle }_{\boldsymbol{x}}\). In cell lines with no strong collective motion, \({\boldsymbol{v}}_{t}({\boldsymbol{x}})\) and \({\boldsymbol{v}}_{t}^{\mathrm{d.c.}}({\boldsymbol{x}})\) are similar, but in the presence of collective motion, these two quantities can differ substantially. As suggested by Garcia et al.54, we used the drift-corrected velocity to calculate the radial velocity–velocity correlation function, obtained as
$${C}_{\mathrm{vv}}(\delta {{x}},t)=\frac{\langle {\bf{v}}_{t}^{\rm{d.c.}}({\bf{x}}+{\delta {\bf{x}}, t})\cdot {\bf{v}}_{t}^{\rm{d.c.}}({\bf{x}},t)\rangle\!{\atop\bf{x}}}{\langle {{\bf{v}}_{t}^{\rm{d.c.}}({\bf{x}},t)}^{2}\rangle\!{\atop\bf{x}}}.$$
Furthermore, we fitted this function to a model exponential \({{\rm{e}}}^{\frac{{{\delta }}{\bf{x}}}{{{\xi }}}}\) to extract the spatial correlation length \({{\xi }}\) of the velocity field, quantifying the size of regions with similar velocities once the average monolayer velocity has been removed. Finally, to better visualize spatial correlations in the velocity field, we followed Malinverno et al.55 and calculated the alignment index \({{{a}}}_{{{t}}}\left({{\bf{x}}}\,\right)\) as the cosine of the angle between the average velocity vector of a single velocity field with every other velocity vector.
Cell-stretching assay
Stretch chambers (STB-CH-4W, STREX cell-stretching systems) were autoclaved and coated with 10 μg ml−1 of collagen I (C8919, Sigma) and 10 μg ml−1 of fibronectin (341631, Sigma) diluted in PBS at 37 °C for 2 h. The coated chambers were washed three times with PBS before seeding 200,000 cells per well in the culture medium (88,888 cells cm−2). Cells were stretched the following day with a STREX cell-stretching system (model number STB-140-10) with 20% stretch (6.40 mm), 1-Hz frequency for 30 min and 1 h.
Cell vibration assay
Flexible-bottomed silicone elastomer plates (BF-3001U, BioFlex) were coated with 10 μg ml−1 of collagen I (C8919, Sigma) and 10 μg ml−1 of fibronectin (341631, Sigma) diluted in PBS for 2 h at 37 °C. The coated chambers were washed three times with PBS before seeding 500,000–900,000 cells in the culture medium (52,083–93,759 cells cm−2). On the following day, stimulation sound files were played for 30 min and 6 h, 1 min off/1 min on in a frequency range of 50–250 Hz with a phonomimetic bioreactor92 connected to a Crown XLS 1502 amplifier.
3D spheroid assay
Spheroid formation in a 3D environment was assessed by embedding cells between two layers of Matrigel (Corning, 354230). First, the bottom of an angiogenesis 96-well µ-plate (89646, ibidi) was coated with 10 µl of 50% Matrigel diluted in the culture medium and centrifuged at 4 °C, 200g for 20 min followed by 1-h incubation at 37 °C. Next, wells were filled with 20 µl of cell suspension in 25% Matrigel diluted in the culture medium (500 cells per well), centrifuged for 10 min at 100g and incubated at 37 °C for 4 h. Wells were filled with the culture medium supplemented with 10 µg ml−1 function blocking antibodies or IgG control; mouse anti-IgG (31903, Invitrogen), mouse anti-human α3 integrin (P1B5, in-house hybridoma), mouse anti-human α6 integrin (P5G10, in-house hybridoma) and rat anti-human β1 integrin (mAb13, in-house hybridoma). Spheroid formation was imaged for 10 days with IncuCyte S3 Live-Cell Analysis system (×10 objective). The culture medium was changed every 2–3 days. Analysis was performed using OrganoSeg software93 and ImageJ (v1.54p).
Wetting assay
Cells were seeded in a low-attachment round-bottom 96-well plate to allow the formation of spheroids. The following day, spheroids were transferred to a multiwell plate previously coated with 10 µg ml−1 of fibronectin (diluted in PBS, incubated overnight at 4 °C and washed twice with PBS). Spheroids were monitored as they wet the substrate by time-lapse imaging for 48 h using IXplore Live Microscope (Olympus Evident; ×4 objective, 10-min time frame). Analysis of spreading area over time was performed using ImageJ. The data were normalized to the area of the spheroid at time 0. To evaluate the impact of integrin perturbations, spheroids were treated with the blocking antibodies described above before starting the wetting experiment.
Immunostaining
Coated (collagen I and fibronectin, as previously mentioned) µ-slide eight-well chambered coverslips (ibidi), standard culture plates (Corning) or Softwell Easy Coat (Matrigen) were fixed at the indicated endpoint with 4% paraformaldehyde in the culture medium for 10 min at RT. Cells were washed with PBS three times for 5 min. Permeabilization and blocking were performed using 0.3% Triton X-100 in 10% normal horse serum diluted in PBS for 20 min at RT. Cells were stained with primary antibodies diluted in 10% normal horse serum overnight at 4 °C. Cells were washed three times for 5 min with PBS and incubated with secondary antibodies diluted in PBS for 1 h at RT, followed by three 5-min washes with PBS. Samples were either imaged right away or stored at 4 °C covered from light until imaging.
The list of antibodies is provided in Supplementary Table 1.
Imaging
Confocal imaging was performed with a 3i spinning-disc confocal (Marianas spinning-disc imaging system with a Yokogawa CSU-W1 scanning unit on an inverted Carl ZEISS Axio Observer Z1 microscope, Intelligent Imaging Innovations) with ×10 ZEISS Plan-Apochromat objective (without immersion, 2-mm working distance, 0.45 numerical aperture), ×40 ZEISS LD C-Apochromat objective (water immersion, 0.62-mm working distance, 1.1 numerical aperture) and ×63 ZEISS Plan-Apochromat objective (oil immersion, 0.19-mm working distance, 1.4 numerical aperture). Wide-field imaging was performed with Nikon Eclipse Ti2-E (Hamamatsu sCMOS Orca Flash4.0, Lumencor Spectra X LED excitation). Live imaging was performed with Incucyte S3 or ZOOM Live-Cell Analysis system.
Confocal microscopy image analysis
FA count and size were assessed by the segmentation of FAs from the maximum intensity projections (MIPs) of a confocal microscopy image z stack (ten bottom slices; vinculin, ITGB1 and ILK staining) using ImageJ software (v1.54p). Junctional intensities were assessed by measuring the integrated density value in MIPs divided by the cell number in each FoV of the confocal microscopy images (β-catenin staining) using ImageJ software. Junction morphology was manually assessed by counting each junction type (linear, reticular and zipper like) per FoV in MIPs of the confocal microscopy images (β-catenin staining) using ImageJ software. Changes in the orientation (coherency) were assessed using the OrientationJ package in ImageJ software from MIPs of the confocal microscopy images (actin and β-catenin staining)94. The total intensities were assessed by measuring the integrated density value divided by total cell number in FoV in MIPs of the confocal microscopy images (β-catenin, E-cadherin, YAP and AMOTL2) using ImageJ software. Total/nuclear intensities were assessed by the segmentation of total cell (cytoplasm + nucleus) and nuclear areas based on actin and DAPI staining, and measuring the integrated density value in MIPs of the confocal microscopy images (β-catenin, YAP and AMOTL2 staining) using ImageJ software.
Mass cytometry
Cells were grown on a 10-cm plate to 90% confluence, washed once with PBS and detached with cell dissociation buffer (number 13150-016, Gibco). The detached cells were dispensed into 15-ml Falcon tubes, centrifuged at 300g for 5 min followed by the removal of supernatant and mixing the pellet by pipetting. Cells were resuspended in 1 ml of serum-free medium. Then, 1 ml of 1-µM cisplatin in a serum-free medium was added to the cells for 5 min, mixed well by pipetting and incubated for 5 min at room temperature. The mixture was quenched with Cell Staining Buffer (Maxpar), 5× volume of the stained cells. Cells were centrifuged at 300g for 5 min, the supernatant was aspirated and the cells were resuspended by pipetting. Cells were washed with 4 ml of Cell Staining Buffer (Maxpar). Cells were counted and three million cells aliquoted into 5 ml of polypropylene tube followed by centrifugation at 300g for 5 min. The supernatant was aspirated and the cells were gently mixed by pipetting. Cells were resuspended in 50 µl of Cell Staining Buffer (Maxpar). Cells were then stained with the antibody panel (Supplementary Table 2), starting with Fc-blocking. Fc Receptor Blocking Solution was added 1:100 to each tube and incubated for 10 min at room temperature. 50 µl of the prepared antibody cocktail was added to each tube and gently mixed by pipetting and incubated at room temperature for 15 min. Samples were gently vortexed and incubated for an additional 15 min at room temperature. After a total of 30-min incubation, samples were washed by adding 2 ml of Cell Staining Buffer (Maxpar) to each tube, centrifuged at 300g for 5 min and the supernatant was removed. Sample wash was repeated three times and the cells were resuspended in residual volume by gently vortexing after final wash and aspiration. Cells were fixed with 1 ml of 1.6% formaldehyde diluted in PBS and gently vortexed before 10 min of incubation at room temperature. Samples were centrifuged at 800g for 5 min and the supernatant was removed. Samples were gently vortexed to resuspend in the residual volume. After cell staining, 1 ml of cell intercalation solution was prepared for each sample by diluting Cell-ID Intercalator-103Rh 1:1,000 into Fix and Perm Buffer (Maxpar) and mixed by vortexing. Then, 1 ml of intercalation solution was added to each tube and gently vortexed. Samples were incubated for 1 h at room temperature or left overnight at 4 °C (up to 48 h). Before acquisition with Helios (WB Injector), cells were centrifuged at 800g for 5 min and washed by adding 2 ml of Cell Staining Buffer (Maxpar), followed by another round of centrifugation. The supernatant was removed and the samples were gently vortexed to resuspend cells in the residual volume. Cells were washed by adding 2 ml of Cell Acquisition Solution (CAS; Maxpar) to each tube and gently vortexed before counting and transferring one million cells into a new tube. Tubes were centrifuged at 800g for 5 min, followed by a careful aspiration of the supernatant. Cells were gently vortexed to resuspend in the residual volume, and finally, one million cells were resuspended in 900-µl CAS. Cells were filtered into cell strainer cap tubes. Sufficient volume of 0.1× EQ beads to resuspend all the samples in the experiment were prepared by diluting one-part beads to nine-parts CAS. Cells were left pelleted until ready to run on Helios. Immediately before data acquisition, the cell concentration was adjusted to 1.0 × 106 cells ml−1 and diluted by the EQ bead solution. Cells were filtered into cap tubes. Samples were run and the data were acquired with Helios CyTOF. Mass cytometry antibodies were either purchased from Fluidigm or self-conjugated.
RNA sequencing
RNA was isolated from three biological replicates of cells (900,000 cells per well in a six-well plate; 93,750 cells cm−2) seeded on coated (collagen I and fibronectin) BioFlex plates. Cells were washed with cold PBS followed by RNA extraction using a NucleoSpin RNA kit (number 740955.250, Macherey-Nagel) as per the manufacturer’s instructions. The total RNA concentration was measured with Nanodrop and the samples were normalized by diluting with RNAse-free water. The sample quality was verified using Agilent Bioanalyser 2100, and the final concentrations were measured using Qubit/Quant-IT Fluorometric Quantitation (Life Technologies). Illumina Stranded Total RNA prep library was prepared using 100 ng of RNA as per the manufacturer’s instructions (Illumina Stranded mRNA Preparation and Ligation kit (Illumina) and sequenced with Novaseq 6000 (S4 instrument, v1.5 (Illumina), 2 × 50 bp, SP flow cell, two lanes (650–800M reads))). The library quality was verified using an Advanced Analytical Fragment Analyser. The sequencing data read quality was ensured using the FastQ (v0.11.14) and MultiQC (v1.5) tools95. Differentially expressed genes were assessed using Bioconductor R package ROTS (v1.14.0), defining genes with FDR < 0.05 as differentially expressed.
Multiplexed fluorescence immunohistochemical staining and imaging
Multiplexed fluorescence immunohistochemical staining and imaging was performed in three cycles, as previously described96, for two sets of seven to eight antibodies and the nuclear marker DAPI (Supplementary Table 1), stained on two serial TMA sections. After the first round of staining and whole-slide imaging of the TMAs, the fluorescence signal was bleached, and the antibodies from the first round of staining were denatured, after which the second round of staining was performed. The process was repeated for the third round of staining. Imaging was performed using a ZEISS Axio Scan.Z1 slide scanner, with each round of staining recorded as an independent .czi image file containing up to five fluorescence channels.
Image analysis of multiplexed TMA datasets
Images of individual TMA cores were extracted from the whole-slide images using the TMA dearrayer functionality in QuPath97. Images from the three staining rounds were registered using an affine image registration method operating through the pyStackReg Python dependency98, aligning the DAPI channels of the three staining rounds. Autofluorescence signal from red blood cells and other histology artefacts (for example, wrinkled or folded tissue section areas) were removed using a pixel classifier in ilastik99. Nuclei were segmented from the DAPI channel using a trained StarDist model100. The nuclear ROIs were expanded by 6 px to generate extranuclear ROIs. Pan-epithelial staining was used to threshold cells into epithelial and stromal compartments. A custom Python script101 was then used to calculate the fluorescence intensity in all channels for the relevant nuclear or extranuclear ROI in the relevant tissue compartments. Finally, patient-level average expression values were calculated for all cells and all TMA cores originating from the same patient. Five samples in which fewer than 100 cells could be quantified within the stromal or epithelial tissue compartment were excluded from further analyses to ensure a representative quantification of cellular phenotypes across the tumour tissue. A total of 193 samples were included in the analyses.
Calculation of ECM, YAP and β-catenin scores
For the ECM scores, the median patient-level expressions (intensities) of stromal fibronectin, collagen I, SMA, laminin and vinculin were determined across the full patient dataset. Next, each patient was assigned one point for each instance that the expression of each of the above markers was above the dataset median. The sum of all points was determined as that patient’s ECM score. YAP scores were determined in the same way, with patients being assigned into the YAP-high group if their mean nuclear YAP expression in the tumour epithelium fell above the dataset median. All other patients were assigned into the YAP-low group.
Survival analysis
Kaplan–Meier analysis was used to compare the survival outcomes between patient groups with different phenotypic signatures, with a log-rank test used to measure statistical significance. P ≤ 0.05 was used as a cut-off for statistical significance.
Statistics and reproducibility
GraphPad Prism (v9.3.1) was used for all statistical analyses. Outliers were identified with 0.1% ROTS and indicated in the Source data. Data distribution was determined with the Shapiro–Wilk normality test. Two-sample testing was performed using Student’s t-test (unpaired, two-tailed) with Welch’s correction (normally distributed data) or non-parametric Mann–Whitney U-test (non-normally distributed data). Multiple comparisons were performed using either one-way analysis of variance (ANOVA; normally distributed data) or Kruskal–Wallis (non-normally distributed data) followed by an appropriate post hoc test, as indicated in the figure legends. Data are presented as column graphs, dot plots (mean ± standard deviation (s.d.)) or box plots (defined in the legends). P values less than 0.05 were considered to be statistically significant. Exact P values are provided in the figures where possible. Otherwise, P values are available in the Source data for each figure.
No statistical methods were used to predetermine the sample sizes, but our sample sizes were based on previous reports101,102,103,104. Data were reproduced in three or more biological replicates, unless otherwise indicated in the figure legends. Excluded data have been indicated in the Methods and figure legends. These pertain to the TMA samples in which five patient samples with fewer than 100 cells within the stromal or epithelial tissue compartment were excluded from further analyses to ensure a representative quantification of cellular phenotypes across tumour tissue, bringing the total samples analysed from 198 to 193. In addition, five patient samples with no available tumour staging information were excluded from analyses requiring a defined tumour stage, bringing the total samples analysed in these cases from 193 to 188.
The experiments were not randomized. However, animals were randomly assigned to cages (equal number of animals per cage) by the animal facility staff. Cages were chosen at random for experimentation. Mice assigned to different experimental conditions were run in parallel, and all animals were maintained under the same condition and were at the same developmental stage.
Experiments were not performed in a blinded fashion. Analysis software/statistical packages were used as detailed in the Methods for robust data analysis, removing user bias. In addition, appropriate controls were included in experiments and control versus treated samples were analysed in the same fashion.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
