Study design and participants

Our prospective study (ID: 201,903,078) was approved by the Institutional Review Board, and all participants provided written informed consent. We collected clinical samples from CRC patients who underwent comprehensive evaluation at Xiangya Hospital, Central South University, China. Treatment plans were determined through discussions with the multidisciplinary team (MDT). This study did not influence the treatment choice; patients received the best treatment based on MDT evaluations. Recruitment took place from December 2018 to December 2019, with follow-up continuing until December 2023, resulting in an average follow-up period of 58 months. Samples were collected during surgery or biopsy, following standard laboratory practice guidelines. The hospital's internal review board and ethics committee reviewed and approved all sample collection and preservation procedures.

MRI image acquisition and MR elastography

MRI and MRE were conducted using a sequence and setup similar to our previous research [19, 20]. All patients began a liquid diet one day before their MRI scan and adhered to a strict 4-h fasting protocol before imaging. The MRI scans were performed using an 18-channel phased-array body coil on a 3 T scanner (Magnetom Prisma, Siemens Healthcare, Germany). Conventional T2-weighted (T2w) images of the rectum were acquired using 2D fast spin-echo (FSE) sequences in oblique, sagittal, and coronal planes, with an in-plane resolution of 3 × 3 mm. Additionally, 3D FSE (SPACE) T2w images were obtained with an in-plane resolution of 0.8 × 0.8 mm2. The total time to acquire the anatomical images was 12 min.

As for MRE, Mechanical waves at 40, 50, 60, and 70 Hz vibration frequencies were delivered to the pelvic region using three surface-based pneumatic actuators. Two actuators were positioned at the posterior region with a static pressure of 0.8 bar, while one actuator was placed at the anterior pelvic region, specifically at the top of the pubic symphysis, with a static pressure of 0.7 bar. A single-shot spin-echo echo-planar imaging (SE-EPI) sequence with motion-encoded gradients (MEG) for flow compensation was used to capture a complete 3D wave field. The entire vibration cycle was sampled at eight phase offsets. Continuous acquisition of fifteen 5 mm thick sagittal slices with a resolution of 3 × 3 mm2 was performed during free-breathing. MRE frequencies were optimised for each vibration frequency (40, 50, 60, and 70 Hz) and set at 47.89, 47.89, 47.89, and 52.41 Hz, respectively. Additional imaging parameters included an echo time of 56 ms, a repetition time of 1670 ms, a GRAPPA factor of 2 for parallel imaging, and a MEG amplitude of 50 mT/m. The total acquisition time was 3.5 min.

Image analysis

After imaging, a radiologist with 5 years of experience in gastrointestinal imaging assessed the tumour location, TNM staging, and other relevant details on T2w images using the DISTANCE method. This method provides a systematic approach for evaluating all clinically significant features on MR images, which is crucial for treatment planning. In the DISTANCE method: DIS stands for the distance from the inferior part of the tumour to the transitional skin; T stands for T staging; A stands for the anal complex; N stands for nodal staging; C stands for circumferential resection margin; and E stands for extramural vascular invasion. MRI-based T and N staging were assigned to each patient using this method.

The analysis method for MRE datasets was detailed in previous studies. In brief, the MRE data was processed using wave-number multifrequency-inversion (k-MDEV) to create parameter maps of shear wave speed \(\backslash(c\backslash)\) (measured in m/s), which serves as a proxy for tissue stiffness. The data processing followed the k-MDEV pipeline, which is available at www.bioqic-apps.com. To characterise the tumour, 9 to 18 circular regions of interest (ROIs), each measuring 0.3 ± 0.02 cm2, were placed in the anterior and posterior rectal wall across three consecutive slices covering the most significant solid tumour cross-section, based on anatomical T2w images and avoiding areas of necrosis, cyanosis, and blood vessels. The stiffness values were averaged within these manually defined ROIs. Distal tumour-adjacent tissue (DTT), located 2 cm from the tumour, was analysed in 6 circular ROIs measuring 0.1 ± 0.02 cm2 as a reference. In order to avoid the influence of individual differences in tissue stiffness on tumour stiffness, tumour stiffness was calculated as the ratio of the average shear wave velocity in tumour tissue to the average shear wave velocity in normal tissue. A radiologist, blinded to the clinical outcomes, placed all ROIs using MRE magnitude images and the corresponding elastograms.

Immunohistochemistry (IHC) and masson staining

Immunohistochemistry was performed on human colorectal cancer tissues and tumour and liver metastasis tissues from nude mice. The detection of Lysyl Oxidase-Like 1 (LOXL1), HSF4, and Alpha-Smooth Muscle Actin (αSMA) followed a standardised protocol. Paraffin-embedded sections were first deparaffinised and rehydrated using an eco-friendly deparaffinisation solution (Servicebio, G1128) and anhydrous ethanol (SCRC, 100,092,683), followed by thorough rinsing with distilled water to ensure complete removal of residual chemicals. Antigen retrieval was subsequently performed using a 20 × Citrate Antigen Retrieval Solution (pH 6.0) (Servicebio, G1202). The specific retrieval conditions for each antigen were optimised based on tissue type and antigen characteristics, with careful attention to prevent buffer evaporation during the retrieval process. For all targets (LOXL1, HSF4, and α-SMA), antigen retrieval involved microwave heating at medium heat for 8 min, followed by an 8-min pause and then continued with low-medium heat for 7 min. After retrieval, the slides were washed with PBS (Servicebio, G0002), and endogenous peroxidase was exhausted by incubating the sections in a 3% hydrogen peroxide solution (Angergech, 04008978274207), followed by another PBS wash. Serum blocking was performed by covering the tissue with 3% BSA (Servicebio, GC305010) at room temperature. The primary antibody, prepared in the appropriate dilution, was applied to the sections and incubated overnight at 4 °C. The secondary antibody, HRP-labeled and corresponding to the species of the primary antibody, was then applied and incubated at room temperature. DAB (Servicebio, G1212) staining was performed next, with controlled development time, followed by a rinse with tap water to stop the reaction. Nuclei were counterstained, and the sections were dehydrated and mounted (Servicebio, G1404-100 mL). The results were examined under a bright-field microscope for LOXL1 and α-SMA; nuclei stained blue with hematoxylin (Servicebio, G1039), while positive expression indicated by DAB appeared brownish-yellow. HSF4 immunohistochemical staining intensity was classified into four levels: 0 (negative), 1 (weak), 2 (moderate), and 3 (strong). Staining frequency was quantified using ImageJ software through pixel analysis, with manual threshold adjustments based on each image's background noise and staining characteristics. To minimise bias, we performed multiple repeated observations and image comparison analyses. In the frequency quantification, the denominator included only tumour region pixels, excluding background noise and non-tumour tissues. The final HSF4 score was calculated by multiplying staining intensity by staining frequency, where a score > 1 was classified as high expression and a score of 1 or less as low expression. The HSF4 scoring criteria were referenced from Zhang et al.'s study [21]. These methods aim to enhance the quantitative analysis's transparency and reliability, ensuring the results' accuracy and reproducibility.

Masson staining is a widely used histochemical technique highlighting structures such as collagen and muscle fibres within tissues. For this study, sections were processed according to the manufacturer's instructions (Servicebio) for Masson staining. The procedure involved deparaffinisation and rehydration of paraffin-embedded sections or warming and fixing frozen sections, followed by sequential staining with the provided solutions. After staining, the sections were differentiated in 1% acetic acid, dehydrated, cleared, and mounted with neutral gum. Under the microscope, collagen fibres appeared blue, while muscle fibres, fibrin, and red blood cells were stained red. Details of the specific antibodies and staining kit used are provided in Table S1.

Cell culture and cell transfection

Two human colorectal cancer cell lines, SW480 and HCT116, and a colon fibroblast cell line, CCD-18Co, were obtained from the American Type Culture Collection. SW480 cells were maintained in Leibovitz's L-15 medium (Gibco) with 10% fetal bovine serum (FBS) (Hyclone, Logan, UT, USA), 2 mM L-glutamine, 0.1 mg/mL streptomycin, and 100U/mL penicillin at 37 °C in a standard humidity incubator. HCT116 cells were cultured in RPMI 1640 medium (Invitrogen) with 10% FBS and maintained in a humidified incubator at 37 °C with 5% CO2. CCD-18Co cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) with high glucose (Life Technology, NY, USA), supplemented with 100 mg/ml streptomycin, 100 units/ml penicillin, and 10% fetal bovine serum. The cells were maintained at 37 °C with 5% CO2 and ≥ 90% humidity.

Genchem Biotechnology Co. Ltd. (Shanghai) provided lentiviral-based small hairpin RNA (shRNA) targeting HSF4 and a control lentivirus with scrambled shRNA. The sequences used were si-HSF4 #1: 5’-GCAAGCUGAUCCAGUGUCUTT-3', si-HSF4 #2: 5’-CGCCAACUCAACAUGUACGTT-3', and scrambled: 5’-UUCUCCGAACGUGUCACGU-3'. Experiments found that si-HSF4 #2 was more effective than si-HSF4 #1; therefore, only si-HSF4 #2 was used for shRNA constructs. SW480 and HCT116 cells were infected with these lentiviral particles and maintained in L-15 or RPMI 1640 medium containing 2 μg/mL or 1 μg/mL puromycin, respectively. After two weeks of selection, western blot analysis was performed to measure HSF4 expression, confirming the creation of stable cell lines.

Construction of different matrix stiffness models

The construction of different matrix stiffness models followed established protocols reported in the literature [22]. Soft and stiff matrix media were obtained from MATRIGEN, USA (CAT SW6-COL-2-EA). The parameters for the stiff matrix media were 25 kPa and 50 hydrogels coated in different numbers of healthy plates. The parameters for the soft matrix media were 2 kPa hydrogels coated in different numbers of well plates.

RNA extraction and deep sequencing

For RNA extraction and deep sequencing, SW480 cells were cultured on 6-well plates coated with 2 kPa or 25 kPa hydrogels (MATRIGEN, USA). Plates were washed with PBS, and 2 ml of cell suspension at 1 × 106 cells/ml was added to each well. Each stiffness condition was tested in triplicate, and cells were incubated at 37 °C in a CO2-free incubator for 72 h. Cells were harvested by scraping in a complete medium. The cell suspension was centrifuged, washed with PBS, and treated with TRIzol reagent (Life Technologies) for RNA extraction, which was then sent to BGI Genomics for processing. BGI extracted and reverse-transcribed it into cDNA and constructed libraries for deep sequencing with a depth of 200 × . RNA was quantified using a NanoDrop and Agilent 2100 bioanalyser (Thermo Fisher Scientific). Libraries were prepared from 1 μg of total RNA per sample, and small RNA (18–30 nt) was purified via PAGE gel electrophoresis. Adapter-ligated small RNA was reverse-transcribed into cDNA, amplified by PCR, and enriched cDNA fragments were selected and purified. Library quality and quantification were assessed using an Agilent 2100 bioanalyser and real-time qPCR with TaqMan probes. Sequencing was performed on the Illumina HiSeq platform (BGI, Shenzhen). All sequencing results were uploaded to the Sequence Read Archive (SRA) database in FASTQ format. A total of six files were submitted, with each specimen generating an average of 12.64 GB of data. The dataset can be accessed at (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA522032).

Sequencing data processing, differentially expressed genes and functional analysis

FASTQ files were subjected to quality assessment and potential coding predictions. The reads were then normalised to FPKM (Fragments Per Kilobase of exon model per Million mapped reads) to create matrix files. The processed data have been uploaded to the GEO database under accession number GSE273846 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE273846). Differential expression analysis was performed in R using the DESeq2 package, which involved normalising counts, estimating dispersion, and fitting a generalised linear model to identify differentially expressed genes (DEGs) [23]. Gene Ontology (GO) and pathway enrichment analyses were conducted using the clusterProfiler package in R to elucidate the biological significance of the DEGs. Additionally, single-sample gene set enrichment analysis (ssGSEA) was carried out with the GSVA package in R to assess pathway activity at the individual sample level. Statistical analyses and visualisations, including plots of DEGs, enriched GO terms, and pathway activities, were generated using R packages such as ggplot2 [24]. Appropriate corrections for multiple tests were applied to ensure the robustness of the results. DEGs were visualised using volcano plots and heat maps to effectively illustrate the differences and patterns in gene expression. Please refer to our previous publication for detailed analysis methods [24].

Data sources and analysis for HSF4 mRNA expression in CRC patients

HSF4 mRNA expression data and clinical information for CRC patients were retrieved from The Cancer Genome Atlas (TCGA) and the GSE10950 datasets [25]. The TCGA CRC cohort included 41 matched pairs of CRC tumours and adjacent normal tissues and 458 unpaired CRC tumour samples. TCGA-related analyses and visualisations were conducted using the GEPIA2 database (http://gepia2.cancer-pku.cn/) [26].

Cell proliferation assay using CCK-8

In order to evaluate the effect of HSF4 knockdown on tumour cell proliferation, a CCK-8 assay was conducted. The cells were seeded at a density of 2 × 104 cells per well in a 96-well plate containing 100 μl of medium with varying stiffness (2 kPa or 25 kPa). After incubation under standard conditions, the viability of the cells was assessed each day using the CCK-8 reagent. Viable cells were incubated with the CCK-8 solution for 2 h, and the absorbance at 450 nm was measured. This absorbance provided a quantitative measure of cell proliferation, allowing for the assessment of the impact of HSF4 knockdown on the growth of HCT116 and SW480 cells.

Transwell migration and matrigel invasion assays

Transwell migration and Matrigel invasion assays were conducted using 24-well culture plates with inserts containing 8 μm pore membranes (Falcon). These membranes were either uncoated or coated with Matrigel (Corning). The cells were cultured on 2 kPa or 25 kPa matrices for 48 h, followed by starvation for 24 h before being harvested and seeded into the chambers for migration and invasion tests. 4 × 104 cells in 0.2% bovine serum albumin (BSA) were placed in the upper chamber, while the lower chamber was filled with 500 μL of medium containing 10% fetal bovine serum (FBS). After 24 h, the cells that had migrated to the bottom surface of the upper chamber were fixed with 4% paraformaldehyde for 20 min and then stained with 0.1% Crystal Violet. The migrated or invaded cells were counted and photographed using a microscope, with five fields per membrane analysed for each group. These experiments were performed in triplicate. For a more detailed description of this methodology, please refer to our previous publication [27].

Wound healing assay

Cells from each group were trypsinised and seeded into 6-well plates, then incubated at 37 °C until they reached 90% confluence. The cells were subjected to serum starvation in a 0.1% FCS-medium for 24 h. A precise scratch was made across the cell monolayer using a 10 µL pipette tip. After scratching, the cells were gently rinsed with PBS and cultured in 2 mL of 1% FCS-medium. The initial gap width was recorded at 0 h using a Leica microscope. The gap was then re-measured at 24 and 48 h with Image J software, and the percentage of the remaining gap area was calculated.

Western Blot (WB) analysis

The Western blot analysis was conducted according to the methodology described in our previous research [28]. Briefly, cells were lysed in RIPA buffer containing a protease inhibitor cocktail (Selleck). Proteins (30 μg) were denatured by boiling at 95 °C for 5 min and then separated using sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE). The separated proteins were transferred to a PVDF membrane, blocked with 5% non-fat milk in TBS buffer containing 0.1% Tween-20 (TBS-T) at room temperature for 1 h. The membranes were cut according to molecular weights, incubated with primary antibodies at 4 °C overnight, washed with TBS-T, and then treated with HRP-conjugated secondary antibodies for 2 h at room temperature. Signal detection was performed using an enhanced chemiluminescence (ECL) system (Amersham Pharmacia Biotech, Arlington Heights, IL). Each blot was performed in triplicate, with GAPDH as the loading control. The intensity of the bands was measured using ImageJ for relative quantification. The Epithelial-Mesenchymal Transition (EMT) Antibody Sampler Kit #9782 was used to analyse epithelial-mesenchymal transition. The specific antibodies used are detailed in Table S1.

Immunofluorescence assay

Immunofluorescence assays were performed using the CST immunofluorescence kit (catalogue number 12727S). Cells were plated on substrates with stiffnesses of 2 kPa, 25 kPa, and 50 kPa and cultured overnight until reaching 40–50% confluence. The cells were then fixed with 4% formalin at room temperature for 15 min. After fixation, the cells were washed three times with PBS. Blocking was carried out with the kit's blocking agent for 1 h, followed by incubation with the primary antibody overnight at 4 °C. The next day, cells were washed thrice with PBS and incubated with fluorescence-conjugated secondary antibodies for 1 h at room temperature. After three additional washes, nuclear staining was performed using DAPI (Beyotime, BB-4133) for 5–30 min in the dark. The results were observed using a fluorescence microscope. Specific antibodies used are listed in Table S1.

Tumour formation in nude mice

The Ethics Committee approved the animal experiments for Experimental Animals of Xiangya Hospital, Central South University (Changsha, China; approval number: 202103520), and adhered to internationally accepted principles for the care and use of experimental animals (NRC, 2011). Twenty-four 4-week-old male BALB/C nude mice were sourced from Beijing Vital River Laboratory Animal Technology Co., Ltd. After a one-week acclimation period, the mice were randomly divided into four groups, each consisting of six mice. For the tumour formation experiments, different cell treatments were injected subcutaneously into the left axillary region of the mice as follows: shControl-HCT116 + LVControl-CCD-18Co, shHSF4-HCT116 + LVControl-CCD-18Co, shControl-HCT116 + LVLOXL1-CCD-18Co, and shHSF4-HCT116 + LVLOXL1-CCD-18Co. Each mouse received a 100 µL injection containing a mixture of CCD-18Co cells (4 × 105) and HCT116 cells (2 × 106). Mice were weighed and monitored twice weekly. After 17 days, the experiment was concluded. Tumours were photographed, excised, and fixed in paraformaldehyde.

Liver metastasis model construction

The same groups of 4-week-old male BALB/C nude mice were used for the liver metastasis experiment. Following a one-week acclimation period, the mice were randomly divided into four groups of six. The different treated cells were injected into the spleen of each mouse in the respective groups: shControl-HCT116 + LVControl-CCD-18Co, shHSF4-HCT116 + LVControl-CCD-18Co, shControl-HCT116 + LVLOXL1-CCD-18Co, and shHSF4-HCT116 + LVLOXL1-CCD-18Co. Each mouse received a mixture of CCD-18Co (4 × 105) and HCT116 (2 × 106) cells in 100 µL. Post-injection, the mice were monitored and weighed twice weekly. The experiment was concluded on day 19, and the liver and spleen were collected, weighed, and fixed in 4% paraformaldehyde.

Statistical analysis

This study's statistical analysis and data visualisation were conducted using GraphPad Prism 9, R, IBM SPSS 22.0, ImageJ, Adobe Illustrator CC 2024 software and BioRender.com. Quantitative data are expressed as mean values with standard errors from at least three independent experiments. Comparisons between two data groups were performed using the student's t-test, while differences among multiple groups were evaluated using two-way ANOVA with Tukey's test. A significance level of P < 0.05 was considered statistically significant. Specific statistical methods and results are detailed in the figure legends. For example, in Gene Set Enrichment Analysis (GSEA), a false discovery rate (FDR) of 0.1 was applied to adjust for multiple comparisons. P-values were calculated based on two-sided statistical tests, with significance levels denoted as *P < 0.05, **P < 0.01, and ***P < 0.001. Results with P > 0.05 were considered not significant (ns).