2.1 The cancer genome atlas (TCGA) database

Clinical data of 372 HCC patients and their tumor gene expression profiles were downloaded from TCGA database. Since the expression levels of lysyl oxidase (LOX) and collagen 1 (COL1) can indicate the grade of matrix stiffness [40, 41], the median expression values of LOX and collagen 1A1 (COL1A1) were taken as the threshold to stratify the patients. Of the 372 patients, 119 with COL1A1^High_- LOX^High HCC were classified as the high stiffness group, 120 with COL1A1^Low_- LOX^Low HCC were classified as the low stiffness group, and the rest 133 were excluded due to an inconsistent expression trend between COL1A1 and LOX. The mRNA expression levels of target genes, Piezo1, CD31, CD34 and vascular endothelial growth factor A (VEGFA), were compared between the two groups.

2.2 Cells and cell culture

Two human HCC cell lines, MHCC97H cells and Hep3B cells, were obtained from the Liver Cancer Institute of Fudan University (Shanghai, China) and Cell Bank of Shanghai Institute of Biochemistry and Cell Biology (Shanghai, China), respectively. MHCC97H cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM, Gibco, New York, NY, USA) supplemented with 10% fetal bovine serum (FBS, Biowest, Riverside, MO, USA) and 1% penicillin/streptomycin (Gibco), and Hep3B cells in Minimum Essential Medium (Gibco) with 12.5% FBS and 1% penicillin/streptomycin. HUVECs, purchased from ScienCell Research Laboratories, Inc, were cultured in endothelial cell medium (ScienCell, San Diego, CA, USA) with 5% FBS (ScienCell), 1% penicillin/streptomycin (ScienCell), and 1% endothelial cell growth supplement (ScienCell). Buffalo rat HCC cells McA-RH7777, acquired from the American Type Culture Collection (Manassas, VA, USA), were grown in the same culture medium as MHCC97H cells.

2.3 Establishment of SD rat HCC models with high or normal liver stiffness background

For SD rat HCC models with high liver stiffness background, 5-week-old male SD rats (Shanghai JieSiJie Laboratory Animal Co., Ltd., Shanghai, China) were subcutaneously injected with 100% carbon tetrachloride (CCl₄) (3 mL/kg) in the abdomen, and then injected with 50% CCl₄ olive solution (2 mL/kg) twice a week. Twelve weeks later, SD rat models with high liver stiffness were formed, their liver stiffness was evaluated by liver elasticity ultrasound, and the expression of COL1 and LOX in HCC tissues was evaluated by immunohistochemical staining.

From two days before transplantation to three days after transplantation, the rats were intramuscularly injected with 2.5 mg dexamethasone per day. Approximately 1.6 × 10⁶ McA-RH7777 cells mixed in Matrigel (BD Biosciences, Franklin, NJ, USA) were orthotopically injected under the capsule of rat liver. Penicillin (50,000 U/day) was used from the day of surgery to the second postoperative day to prevent infection. Twelve days later, orthotopic liver cancer SD rat models with high liver stiffness background were established, and their blood samples, tumors, and lung tissues were collected for further analysis.

The experimental method for the establishment of SD rat HCC models with normal liver stiffness background was the same as the method for orthotopic liver cancer SD rat models with high liver stiffness background, except for CCl₄-free pretreatment before transplantation.

All animal care and experiments used in the study followed the guideline for the Care and Use of Laboratory Animals published by the US National Academy of Science (Washington, WA, USA), and the experiment design of the animal model was approved by the Animal Care Ethical Committee of Zhongshan Hospital, Fudan University (Shanghai, China).

2.4 Lentivirus infection

The target fragment of genes (Supplementary Table S1) and packaging recombinant plasmid of lentivirus were designed and constructed by GeneChem Co. Ltd. (Shanghai, China). The fragment targeting human gene integrin β1 was inserted into the plasmid pGCSIL, and the fragment targeting human gene Piezo1 was inserted into the GV112 vector. The target sequence to miR-625-5p was cloned into the GV280 vector, miR-625-5p overexpression sequence was inserted into GV369. The fragments targeting rat genes integrin β1 and Piezo1 were inserted into the GV112 vector, respectively. When HCC cells grew and reached 40% confluence, they were infected with lentivirus and selected by puromycin (2 μg/mL). The efficiency of inhibition or overexpression was evaluated by quantitative real-time polymerase chain reaction (qRT-PCR) and Western blotting.

2.5 Hematoxylin-eosin (H&E) staining and Sirius red staining

H&E staining: Tissue sections were dewaxed with xylene and rehydrated with ethanol at different concentrations, and then they were stained with hematoxylin for 5 min and rinsed with running water. After that, they were stained with eosin for 2 min.

Sirius red staining: Slides were stained with Sirius red reagent (Polysciences Inc, Warrington, PA, USA) for 1 h, and then washed in acetic acid, dehydrated in ethanol, and cleared in xylene. Quantifications of collagen proportional area (CPA) were measured by ImageJ software (National Institutes of Health, Bethesda, MD, USA).

2.6 Immunohistochemistry (IHC)

Immunohistochemical staining was the same as the method described previously [31]. The primary antibodies were diluted as follows: LOX (1:100, Abcam, Cambs, Cambridge, UK), COL1 (1:100, Affinity, Cincinnati, OH, USA), Piezo1 (1:50, Abcam), CD31 (1:100, Abcam), hypoxia inducible factor-1α (HIF-1α) (1:100, Abcam), VEGFA (1:100, Proteintech, Wuhan, Hubei, China), insulin-like growth factor binding protein 2 (IGFBP2) (1:200, BOSTER, Wuhan, Hubei, China), CXC chemokine ligand 16 (CXCL16) (1:200, Proteintech). Photos of representative sites were captured with an Olympus microscope (Tokyo, Japan). The density of positive staining was measured by ImageJ software. Microvascular density (MVD) was assessed based on IHC staining of CD31. The slides were examined under 100× magnification to identify the highest vascular density area (hot spot) within the tumor, and five areas of highest MVD under 200× magnification were selected to calculate the average MVD. The average MVD of the five areas was recorded as the MVD level.

2.7 FN-coated substrate gels with different stiffness

FN-coated substrate gels with the stiffness of 6, 10, and 16 kPa were prepared as described previously [30]. The suspended HCC cells (1 × 10⁶) in 0.6 mL culture medium were spread onto an FN-coated gel in a dish and cultured for 2-3 h at 37°C with 5% CO₂. Subsequently, 10 mL culture medium was added to the dish, and the attached cells were further cultured for 36-48 h at 37°C with 5% CO₂. Cells were collected from the gel surface with a cell scraper for the following analysis.

2.8 Western blotting

Western blot was performed as in our previous work [18] with a little modification. Briefly, 100 μg protein extracted from HCC cells was loaded and resolved on sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) for Piezo1 detection, and 20 μg protein extract for other target protein detection. The conditions of membrane transfer were optimized as 350 mA, 210 min for Piezo1, 300 mA, 45 min for von Hippel-Lindau (VHL), 300 mA, and 90 min for other target proteins. The diluted primary antibodies were as follows: integrin β1 (1:1000, Cell Signal Technology, Boston, MA, USA), Piezo1 (1:500, Abcam), COL1 (1:1000, Affinity), α-tubulin (1:5000, Proteintech), HIF-1α (1:1000, Abcam), VHL (1:1000, Abcam), IGFBP2 (1:1000, Abcam), CXCL16 (1:1000, Abcam), L-vascular endothelial growth factor A (L-VEGFA) (1:1000, Proteintech), ubiquitin (UB) (1:1000, Cell Signal Technology). The secondary horseradish peroxidase (HRP)-conjugated antibody (Proteintech) was diluted at 1:5000.

2.9 An HCC tissue microarray from buffalo rat HCC models with different liver stiffness backgrounds

An HCC tissue microarray was constructed previously [18] from buffalo rat HCC models with different liver stiffness backgrounds.

2.10 HCC patients

Clinical data of 88 HCC patients who underwent curative resection at the Department of Liver Surgery, Zhongshan Hospital of Fudan University between January 2008 and December 2008 were analyzed. Patients were followed up till December 2013. HCC patients were diagnosed according to the diagnostic criteria of the American Association for the Study of Liver Diseases (2018) [42], and their clinical stage was determined according to the Barcelona Clinic Liver Cancer staging system (2004) [43] and the 8^th edition of Tumor Node Metastasis (TNM) staging system [44], respectively. Tumor differentiation grade was evaluated by the Edmondson grading system [45]. Vascular invasion, tumor number, tumor size, and other clinical parameters such as age, gender, hepatitis B surface antigen (HBsAg), alpha-fetoprotein (AFP), and liver function indicators were collected for univariate and multivariate analyses.

2.11 The miRTarBase database

Target microRNA (miRNA) to Piezo1 and COL1A1 genes were identified seperately from the miRTarBase database (https://miRTarBase.cuhk.edu.cn/). The common miRNAs targeting both Piezo1 and COL1A1 genes were identified.

2.12 miRNA qRT-PCR

Total RNA was extracted from HCC cells using TRIzol reagent (Invitrogen). The cDNA of miRNA was reversely transcribed using All-in One^TM miRNA qRT-PCR Detection Kit 2.0 (GeneCopoeia, Rockville, MD, USA) according to the manufacturer's protocol. Specific primers for U6 and miR-625-5p were all synthesized by GeneCopoeia. Reaction conditions were performed following the manufacturer's protocol of GeneCopoeia. The target gene was amplified by a QuantStudio^Ⓡ 5 Real-Time PCR instrument (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer's protocol. Heterogeneous nuclear RNA (hsnRNA) U6 was selected as an endogenous reference. Relative gene expression was normalized to U6 and reported by 2^{− ΔΔ Ct} method.

2.13 Recombinant plasmid construction and transient transfection

miR-625-5p interference sequence (5'-GGACUAUAGAACUUUCCCCCU-3') and its scramble sequence (5'- CAGUACUUUUGUGUAGUACAA-3') were constructed and synthesized by Sangon Biotech (Shanghai, China). When they grew and reached 80% confluence, the cells were transfected with a negative control siRNA or si-miR-625-5p in Opti-Medium at the concentration of 50 nmol/L via lipofectamine 2000 (Invitrogen).

2.14 Dual-luciferase reporter assay

The wild-type Piezo1 3’-untranslated region (UTR) or COL1A1 3’-UTR containing the binding site of miR-625-5p and its mutant type were all designed and synthesized by OBiO Technology Co. Ltd. (Shanghai, China). 293T cells were seeded into a 24-well culture plate on the day before plasmid transfection. The cells were transfected with the designed plasmids. After 48 h culture, the cells were harvested, and their luciferase activity was analyzed using the Dual-luciferase Reporter Assay System (Promega, Madison, WI, USA).

2.15 Scanning electron microscopy (SEM)

Rat rail COL1 (3.58 mg/mL, Corning, New York, NY, USA) and serum-free DMEM were mixed in volume ratios of 1:20 and 1:4 to prepare two types of substrate gels for SEM analysis. The mixtures containing different amounts of COL1 on a slide were solidified into the gels at 4°C overnight. Then the gels were sampled and fixed in 2% glutaraldehyde at 4°C. Samples were stained with 1% osmium tetroxide and observed under a scanning electron microscope (JEOL, Tokyo, Japan).

2.16 Collection of conditioned medium (CM) from shPiezo1-transfected MHCC97H cells

Wild-type, shCtrl-transfected, and shPiezo1-transfected MHCC97H cells were cultured on the high-stiffness substrate for 48 h, and their culture supernatants were collected and concentrated by a centrifugal filter (Millipore, Schwalbach, SL, Germany) at 4000 × g for 30 min at 4°C, and then the concentrated supernatant was filtered through 0.22 μm filters and stored at -80°C for subsequent use. The protein concentration of the concentrated supernatant was measured by bicinchoninic acid assay (Beyotime, Shanghai, China).

2.17 Tube formation assay

HUVECs (5 × 10⁵) suspended in extracellular matrix (ECM) containing CM were seeded onto a Matrigel-coated dish. After 8 h culture, tube-like structures on Matrigel were photographed by an inverted microscope (Olympus). The state of tube forming in the picture was analyzed using ImageJ software.

2.18 Wound healing assay

Wound healing assay was performed to assess the effect of CM intervention on cell migration. After HUVECs grew to reach a tight cell monolayer in a 6-well plate, the cell monolayer was scratched with a plastic pipette tip. The remaining cells were washed twice with PBS and then treated with CM for 48 h. The migrated cells at the wound front were photographed using an Olympus microscope and analyzed by ImageJ software.

2.19 Analysis of angiogenesis-related factors in CM using a human angiogenesis array

Angiogenesis-related factors in CM were analyzed by Human Angiogenesis Array Kit (Proteome Profiler^TM, R&D Systems, Minneapolis, MN, USA). Samples were mixed with a detection antibody cocktail for 1 h. The sample/antibody mixture was then incubated with an angiogenesis array overnight at 4°C. After unbound materials were removed from the array, streptavidin-HRP and chemiluminescent detection reagents were sequentially added. Array signals were recorded by the Bio-Rad Chemi Doc XRS imaging system (Hercules, CA, USA) and analyzed using ImageJ software.

2.20 Intracellular Ca²⁺ measurement

HCC cells were cultured on different stiffness substrates for 24 h, and then their culture solution was replaced with a fresh culture medium containing Ca²⁺ probe (Ca²⁺-GPCR analysis-calcium ion indicator probe Fluo-4, AM, KeyGEN BioTECH, Nanjing, China, 1 μmol/L). The cells were further cultured for 30 min, and the culture medium was discarded. The cells were washed once with PBS and cultured in a fresh culture medium for 1 h. The fluorescence intensity of intracellular Ca²⁺ in HCC cells was observed and recorded by a fluorescence microscope (Olympus) with cellSens software (Olympus).

2.21 Intervention of Yoda1, GsMTx4 and MG132

Yoda1 (an agonist for Piezo1, MedChemExpress, Shanghai, China, 25 μmol/L for MHCC97H and 5 μmol/L for Hep3B) and GsMTx4 (an antagonist for Piezo1, Abcam, 2.5 μmol/L for both MHCC97H and Hep3B) were respectively applied to treat HCC cells grown on different stiffness substrates for exploring the effect of Piezo1 activation on pro-angiogenic factor expression. MG132 (a proteasome inhibitor, Cell Signal Technology, 10 μmol/L) was applied to treat HCC cells grown on different stiffness substrates for exploring the effects of Piezo1 and matrix stiffness on HIF-1α ubiquitination.

2.22 Co-immunoprecipitation (Co-IP)

Total protein was extracted from the cells using immunoprecipitation lysis buffer (20 mmol/L Tris-HCl, pH 7.6; 150 mmol/L NaCl; 1 mmol/L ethylene diamine tetraacetic acid (EDTA); 0.5% NP-40; 10% glycerol; 1 mmol/L phenylmethanesulfonyl fluoride). The extracted protein was pretreated with protein A/G plus-agarose beads (Millipore) overnight at 4°C. Subsequently, these mixtures were incubated with IgG or HIF-1α antibody at 4°C for 24 h, respectively. Afterwards, the collected beads were washed three times using washing buffer (50 mmol/L Tris-HCl, pH 7.6; 300 mmol/L NaCl; 1 mmol/L EDTA; 0.5% NP-40; 10% glycerol), and the post-elution beads were boiled in the loading buffer, resolved on SDS-PAGE. Immunoblotting was finally performed to detect VHL and HIF-1α.

2.23 Statistical analysis

GraphPad Prism 8.0 (San Diego, CA, USA) and SPSS 22.0 statistical software (SPSS, Chicago, IL, USA) were used for all statistical analyses. Data are presented as mean ± standard deviation (SD). Quantitative variables were analyzed by the analysis of variance (ANOVA) test among three groups and Student's t-test between two groups. Qualitative variables were analyzed by the chi-squared test and rank sum test. Kaplan-Meier analysis and log-rank test were performed to evaluate the association between Piezo1 expression and patients’ overall survival (OS). The interval time of OS was calculated from surgery to death of any reason or the most recent follow-up. COX regression model was performed to clarify the prognostic value of Piezo1 expression. P < 0.05 was considered statistically significant.

3.1 Piezo1 participated in matrix stiffness-induced angiogenesis and metastasis

On the basis of our previous findings that increased matrix stiffness potentiated HCC angiogenesis [31, 32], we further explored the contribution of Piezo1 to matrix stiffness-induced angiogenesis in HCC. We first downloaded clinical data of 372 HCC patients and their tumor gene expression profiles from the TCGA database to clarify Piezo1 expression and its association with HCC angiogenesis. Considering that the expression levels of LOX and COL1 can indicate the grade of matrix stiffness [40, 41], we used the median expression values of LOX and COL1A1 as the threshold to stratify HCC patients into COL1A1^High-LOX^High group (119 patients) and COL1A1^Low-LOX^Low group (120 patients). Compared with that in the COL1A1^Low-LOX^Low TCGA-HCC tissues, Piezo1 presented an obvious upregulation in COL1A1^High-LOX^High TCGA-HCC tissues (Figure 1A), and its high expression was associated with an unfavorable prognosis (Figure 1B). Additionally, CD31, CD34, and VEGFA also exhibited higher expression in COL1A1^High-LOX^High TCGA-HCC tissues than in COL1A1^Low-LOX^Low TCGA-HCC tissues (Figure 1A), in agreement with our previous results in rat HCC tissues [31, 32]. Importantly, Piezo1 was positively associated with CD31 in expression level (Figure 1C). These results described above suggest a potential linkage between Piezo1 and matrix stiffness-induced HCC angiogenesis.

Subsequently, we developed new orthotopic liver cancer SD rat models with high and normal liver stiffness backgrounds (Figure 1D and Supplementary Figure S1A) to validate the role of Piezo1 in matrix stiffness-induced angiogenesis and metastasis. We first compared liver stiffness levels of SD rat models with high or normal liver stiffness background (healthy liver) using LOX/COL1 expression (Supplementary Figure S1B) and sirius red staining (Supplementary Figure S1C). SD rats with high liver stiffness background had higher expression of LOX and COL1 and more collagen fibers in tumor tissues than those with normal liver stiffness background (Supplementary Figure S1B-C). Then we detected the stiffness of SD rats with high liver stiffness background using liver elasticity ultrasound and found that the mean stiffness value of SD rats with high liver stiffness background (17.04 ± 1.73 kPa) could represent the tissue stiffness of advanced cirrhosis [46]. These results suggested that liver stiffness levels of the established SD rat models with high liver stiffness background were able to mirror the stiffness ranges of advanced cirrhosis, which were in accordance with reported data [40, 41, 46]. Afterwards, by hepatic subcapsular injection of HCC cells combined with short-term usage of dexamethasone, we successfully obtained orthotopic liver cancer SD rat models with high liver stiffness background (Figure 1D, Supplementary Figure S1A) and ascertained that their serum phenotype (Supplementary Figure S1D) was almost consistent with that of HCC patients with advanced cirrhosis [47]. So, the established animal models can simulate the pathological properties of HCC with advanced cirrhosis. Using shPiezo1-transfected McA-RH7777 cells or shITGB1-transfected McA-RH7777 cells (Figure 1E), we also assessed the roles of Piezo1 or integrin β1 in matrix stiffness-regulated HCC angiogenesis and discovered that tumor volume of the shPiezo1 #1 group was significantly smaller than those of the two control groups (Figure 1F). Simultaneously, the positive expression area of CD31 and the microvascular density (MVD) in HCC tissues of the shPiezo1 #1 group dropped considerably (Figure 1G-H), and Piezo1 suppression evidently hindered the incidence of lung metastasis (Figure 1I), revealing that Piezo1 mediates matrix stiffness-induced HCC angiogenesis and influences the incidence of lung metastasis. On the other hand, knockdown of integrin β1, a stiffness-sensor molecule identified previously [18, 31], also resulted in an apparent decline in tumor growth, angiogenesis, and lung metastasis (Figure 1F-I), which verified that increased matrix stiffness promoted angiogenesis and metastasis via integrin β1. In orthotopic liver cancer SD rat models with normal liver stiffness background (Supplementary Figure S1A), knockdown of Piezo1 or integrin β1 also significantly inhibited tumor growth and angiogenesis (Supplementary Figure S2A-F), but they had little effect on the occurrence of lung metastasis (Supplementary Figure S2G), which were significantly different from the results of the SD rat model with high liver stiffness background. These results indirectly support that higher liver stiffness facilitated the occurrence of lung metastasis in HCC, which were consistent with our previous findings in buffalo rat HCC models [18]. Taken together, in addition to integrin β1, Piezo1 may also play a regulatory role in matrix stiffness-driven angiogenesis and metastasis in HCC.

3.2 Piezo1 expression in HCC cells under different stiffness stimulation and its clinical significance

We applied 6, 10, and 16 kPa stiffness substrates, which represented the stiffness levels of normal, fibrotic, and cirrhotic liver tissues [46, 48, 49], to analyze matrix stiffness-mediated effects on Piezo1 expression in HCC cells. The expression levels of Piezo1 and COL1 in HCC cells all showed an increasing trend with the increase of matrix stiffness (Figure 2A). Integrin β1 suppression remarkably reduced the expression of Piezo1 and COL1 in HCC cells grown on a high-stiffness substrate (Figure 2B). Similarly, in a rat HCC tissue microarray analysis, Piezo1 also presented a significant upregulation in HCC tissues with high liver stiffness background compared with that in HCC tissues with normal and medium liver stiffness background (Figure 2C). Because high expression of COL1 and LOX in rat HCC tissues with high liver stiffness background has been validated in our previous study [18], we easily inferred that Piezo1 expression level was associated with matrix stiffness level or COL1/LOX expression level. Additionally, high Piezo1 expression in COL1^High-LOX^High human HCC tissues (44 cases) (Figure 2D-E) also supported an association between Piezo1 expression and COL1 expression. Thereby, at cellular and tissue levels, increased matrix stiffness indeed promoted Piezo1 and COL1 expression, and both of them showed the same trend in the expression levels.

To evaluate the clinical significance of Piezo1 in HCC patients, we used the median expression value of Piezo1 in HCC tissues as the cut-off value to divide HCC patients into the Piezo1^High group (44 cases) and the Piezo1^Low group (44 cases). The results demonstrated that high expression of Piezo1 was associated with high LOX/COL1 expression, cirrhosis-associated indexes (total bilirubin, albumin), AFP, and tumor size (Table 1), in accordance with the above-mentioned results in orthotopic liver cancer SD rat models. Additionally, Piezo1^High HCC patients had an unfavorable prognosis (Figure 2F), which was identical to the result of the TCGA analysis. Multivariate analysis suggested that apart from TNM stage and vascular invasion, Piezo1 overexpression also served as an independent risk factor for OS of HCC patients (Supplementary Table S2).

3.3 A positive feedback loop worked in matrix stiffness-mediated Piezo1 upregulation

Given that Piezo1 and COL1 were all highly expressed in HCC cells under high-stiffness stimulation (Figure 2A), and the alteration of liver matrix stiffness was mainly attributed to extracellular matrix protein COL1 deposition and crosslinking [40, 41], we speculated that there might be a positive feedback regulation loop in matrix stiffness-mediated effect on Piezo1 upregulation. Specifically, matrix stiffness promoted Piezo1 expression, simultaneously enhanced the production of extracellular COL1, and COL1-reinforced tissue stiffening resulted in more expression of Piezo1. Considering that the same miRNA can concurrently regulate different target proteins, we identified 6 common miRNAs from the miRTarBase database that could target both Piezo1 and COL1A1 genes (Figure 3A). Among them, we determined miR-625-5p as the target miRNA by literature review and preliminary experiment analysis and then elucidated its role in matrix stiffness-upregulated Piezo1. Increased matrix stiffness restrained miR-625-5p expression in HCC cells (Figure 3B and Supplementary Figure S3A), and suppression of integrin β1 significantly reversed high-stiffness stimulation-induced miR-625-5p downregulation (Figure 3C and Supplementary Figure S3B). The expression trend of miR-625-5p was opposite to that of Piezo1 and COL1 mentioned above (Figure 2A). These data indicate that there may exist a negative regulation of miR-625-5p on Piezo1 and COL1 expression under different stiffness stimulation.

Based on the predicted binding sites of miR-625-5p to the 3’-UTR of Piezo1 and COL1A1 (Figure 3D-E), we respectively constructed the wild-type recombinant plasmids (WT Piezo1 3’-UTR and WT COL1A1 3’-UTR) and the mutant-type recombinant plasmids (MUT Piezo1 3’-UTR and MUT COL1A1 3’-UTR) to analyze whether there was a binding site between miR-625-5p and the 3’-UTR regions of these two target genes. Dual-luciferase reporter assay for Piezo1 and miR-625-5p showed that the luciferase fluorescence intensity of the miR-625-5p mimics group was decreased by 40.55% compared with that of the control group (P < 0.001), while the luciferase fluorescence intensity of the MUT Piezo1 3’-UTR group was partially restored compared with that of the WT Piezo1 3’-UTR group (P < 0.001) (Figure 3D). Similarly, for miR-625-5p and COL1A1 3’-UTR, the luciferase fluorescence intensity of the miR-625-5p mimics group was decreased by 38.66% (P < 0.001), and the luciferase fluorescence intensity of the MUT COL1A1 3’-UTR group was also partially restored (P < 0.001) (Figure 3E). The above results supported that miR-625-5p is specifically bound to the 3’-UTR sites of Piezo1 or COL1A1. However, a partial reversion but not full reversion in the MUT Piezo1 3’-UTR group or the MUT COL1A1 3’-UTR group implied that other atypical binding sites might exist except the predicted binding sites. We further analyzed the regulatory role of miR-625-5p in matrix stiffness-upregulated Piezo1 and found that miR-625-5p overexpression distinctly decreased the expression of Piezo1 and COL1 in HCC cells grown on high-stiffness substrate (Figure 3F-G and Supplementary Figure S3C-F). Conversely, miR-625-5p inhibition improved Piezo1 and COL1 expression in HCC cells grown on low-stiffness substrate (Figure 3H-I and Supplementary Figure S3G-J). All these results demonstrated that miR-625-5p participated in matrix stiffness-upregulated Piezo1 and COL1.

To assess the effect of COL1 deposition on matrix stiffness level, we prepared two types of substrate gel by mixing COL1 and serum-free DMEM in a volume ratio of 1:20 and 1:4 to simulate different deposition amounts of COL1. The results showed that the stiffness level of the substrate gel (1:4, 1230.190 Pa) was significantly higher than that of the substrate gel (1:20, 341.522 Pa), and collagen I fiber bundles of the substrate gel (1:4) were thicker and more tightly packed than that of the substrate gel (1:20) (Figure 3J-K). It confirmed our speculation that COL1 deposition reinforced tissue stiffening. Together, we proposed that a positive feedback regulation loop as stiff matrix/integrin β1/miR-625-5p/Piezo1 and COL1/stiffer matrix was involved in matrix stiffness-upregulated Piezo1 in HCC (Figure 3L).

3.4 Differentially expressed angiogenesis-related cytokines between shPiezo1-CM and Scramble-CM

Paracrine is the most frequent regulatory way for tumor cells to induce angiogenesis [50-52]. The above results in animal models and HCC tissues suggested that Piezo1 might participate in matrix stiffness-induced angiogenesis. By comparing the inhibitory effects of three knockdown clones for Piezo1 in HCC cells, the one with the best inhibitory effect was selected for subsequent function analysis (shPiezo1 #2 for MHCC97H cells and shPiezo1 #3 for Hep3B cells, Figure 4A). Subsequently, the CM from wild-type, shCtrl-transfected, and shPiezo1 #2-transfected MHCC97H cells grown on high-stiffness substrate (WT-CM, Scramble-CM, and shPiezo1-CM) were collected to appraise their effects on angiogenesis. Compared with the cells treated with shPiezo1-CM, HUVECs treated with WT-CM and Scramble-CM all presented stronger abilities in tube formation (Figure 4B) and migration (Figure 4C), illustrating that WT-CM and Scramble-CM contained more pro-angiogenic factors than shPiezo1-CM. Thereby, Piezo1 upregulation may influence the expression and secretion of pro-angiogenic factors and promote HCC angiogenesis. Using a human angiogenesis array comprising 55 cytokines, we successfully found out 9 differentially expressed pro-angiogenic and anti-angiogenic factors (fold change >1.2 or <0.5) between shPiezo1-CM and Scramble-CM (Figure 4D and Supplementary Table S3), including 7 downregulated factors (Serpin F1, CXCL16, IGFBP-2, VEGF, PDGF-AA, IGFBP-3, interleukin-8) and 2 upregulated factors [Amphiregulin, urokinase-type plasminogen activator (UPA)] in shPiezo1-CM. Among these differentially expressed angiogenesis-related cytokines, most pro-angiogenic factors in shPiezo1-CM exhibited a decreasing trend in content. Based on the above results, we concluded that the overall effect of shPiezo1-CM was angiogenesis inhibition, and shPiezo1-CM lacked the pro-angiogenic factors required for tube formation and migration of HUVECs. Furthermore, we selected three differentially expressed pro-angiogenic factors, VEGF, CXCL16, and IGFBP2, as the target factors for subsequent validation. The reasons for this selection were as follows: (1) the evidence presented above suggested that Piezo1 was a pro-angiogenic regulator and mediated matrix stiffness-induced angiogenesis; (2) VEGF, IGFBP2, and CXCL16 had a common upstream transcription factor hypoxia-inducible factor (HIF-1α) [15, 16, 53, 54] and affected tumor proliferation, angiogenesis, invasion and metastasis [55-57]; (3) HIF-1α expression was not always dependent on hypoxia [58], and a variety of growth factors and cytokines were able to stabilize HIF-1α under normoxic conditions [58]; (4) increased matrix stiffness also played a regulatory role in HIF-1α expression in the polarized M2 macrophages [59]. In validation analysis, knockdown of Piezo1 remarkably suppressed the expression of VEGF, CXCL16, and IGFBP2 in HCC cells cultured on a high-stiffness substrate (Figure 4E), which was consistent with the results of the human angiogenesis array. Besides that, integrin β1 knockdown also partially attenuated the expression of VEGF, CXCL16, and IGFBP2 in HCC cells grown on a high-stiffness substrate (Figure 4F), meaning that matrix stiffness was also involved in the regulation of angiogenesis-related molecules. Consequently, knockdown of Piezo1 reduced the expression of angiogenesis-related cytokines and restrained matrix stiffness-induced HCC angiogenesis.

3.5 Piezo1 activation and Ca²⁺ influx in HCC cells under high-stiffness stimulation

Under mechanical stimulation, Piezo1 allows the influx of positively charged ions (Ca²⁺, Na⁺) into the cells, and the content of Ca²⁺ influx can reflect the activation level of Piezo1 [34]. As shown in Figure 2A, increased matrix stiffness significantly upregulated Piezo1 expression. Accordingly, the relationship between Piezo1 expression and Piezo1 activation becomes an unavoidable problem for understanding the contribution of Piezo1 to matrix stiffness-induced HCC angiogenesis. We analyzed the levels of Ca²⁺ influx in HCC cells cultured on different stiffness substrates and discovered that the fluorescence intensity of Ca²⁺ in HCC cells was increased significantly with the increase of matrix stiffness (Figure 5A). Instead, suppressing Piezo1 significantly diminished the fluorescence intensity of Ca²⁺ in HCC cells cultured on the high-stiffness substrate (Figure 5B). It suggests that Ca²⁺ influx positively correlate with Piezo1 expression and matrix stiffness. On the other hand, Yoda1 intervention noticeably enhanced the fluorescence intensity of Ca²⁺ in HCC cells grown on low-stiffness substrate (Figure 5C), whereas GsMTx4 intervention clearly reduced it in HCC cells grown on high-stiffness substrate (Figure 5D). Overall, high-stiffness stimulation significantly enhanced Piezo1 expression and activation in HCC cells, and the expression level of Piezo1 was associated with its activation level under high-stiffness stimulation.

3.6 Matrix stiffness-caused Piezo1 activation regulated HIF-1α ubiquitination and the expression of its downstream pro-angiogenic factors in HCC cells

HIF-1α, an important common transcription factor, governs the expression of many pro-angiogenic factors such as VEGF, CXCL16, and IGFBP2 [16, 55, 60]. We first elucidated whether matrix stiffness-caused Piezo1 activation regulated the expression of HIF-1α and its downstream pro-angiogenic factors. As shown on Figure 6A and Supplementary Figure S4A-B, high-stiffness stimulation significantly upregulated the expression of HIF-1α and its downstream target genes (VEGF, IGFBP2 and CXCL16) and significantly attenuated the expression of VHL (VH the E3 ligase of HIF-1α) in HCC cells, suggesting a potential linkage between high matrix stiffness and HIF-1α ubiquitination. Subsequently, we used proteasome inhibitor MG132 to treat HCC cells cultured on low-stiffness substrate and found that MG132 intervention obviously increased the expression of HIF-1α and its downstream target genes VEGF, IGFBP2, and CXCL16 (Figure 6B and Supplementary Figure S4C-D), indirectly indicating that increased matrix stiffness repressed HIF-1α ubiquitination and promoted the expression of its downstream target genes. Afterwards, we evaluated the effects of Piezo1 activation on HIF-1α ubiquitination and the expression of its downstream target genes. The results demonstrated that Yoda1 intervention resulted in a significant upregulation in HIF-1α, VEGF, CXCL16, and IGFBP2 expression, as well as an obvious downregulation in VHL expression in HCC cells cultured on low-stiffness substrate (Figure 6C and Supplementary Figure S4E-F). GsMTx4 intervention achieved an opposite result in HCC cells cultured on high-stiffness substrate (Figure 6D, Supplementary Figure S4G-H). These results together with the data in Figure 5 suggested that Piezo1 activation inhibited HIF-1α ubiquitination and upregulated the expression of its downstream target genes via Ca²⁺ influx. Besides, Piezo1 or integrin β1 knockdown also led to a significant decline in HIF-1α, VEGF, IGFBP2, and CXCL16 expression and an increase in VHL expression in HCC cells cultured on high-stiffness substrate (Figure 6E-F and Supplementary Figure S4I-L), but Yoda1 intervention partially reversed the effects of Piezo1 or integrin β1 knockdown on the expression of HIF-1α, VHL, VEGF, IGFBP2, and CXCL16. Similarly, MG132 intervention also reversed Piezo1 downregulation-caused changes in the expression of HIF-1 and its downstream target proteins (Figure 6E-F and Supplementary Figure S4I-L). Thereby, matrix stiffness-caused Piezo1 activation participates in HIF-1α ubiquitination and the expression of its downstream pro-angiogenic factors.

Taking HIF-1α as a bait protein to capture VHL protein, we observed an endogenous interaction between VHL protein and HIF-1α protein in HCC cells (Figure 6G, Supplementary Figure S5A). Additionally, knockdown of Piezo1 or integrin β1 elevated the HIF-1α ubiquitination level (Figure 6H and Supplementary Figure S5B-C), showing direct evidence that Piezo1 participates in high matrix stiffness-downregulated HIF-1α ubiquitination. In SD rat model with high-stiffness background, the positive expression areas of HIF-1α, VEGF, CXCL16 and IGFBP2 were significantly decreased in HCC tissues of the shPiezo1 group and the shITGB1 group (Figure 6I and Supplementary Figure S5D). Moreover, the positive expression areas of HIF-1α and VEGF were associated with Piezo1-positive expression areas in human HCC tissues (Supplementary Figure S5E), validating that Piezo1 contributes to matrix stiffness-mediated effect on HIF-1α expression and angiogenesis in HCC.

Above all, we concluded that the matrix stiffness/integrin β1/Piezo1 activation/Ca²⁺ influx/HIF-1α ubiquitination/VEGF, CXCL16 and IGFBP2 pathway participates in matrix stiffness-regulated HCC angiogenesis (Figure 7).