Human tissue specimens and immunohistochemical analysis
From February 1, 2014 to December 31, 2015, 123 human tissues were authorized by the Hepatic Surgery Center, Tongji Hospital, Huazhong University of Science and Technology (Wuhan, China). All the participants provided written informed consent. Patients who had received radiotherapy or chemotherapy before surgery were excluded. Survival time was calculated from the date of surgery to the date of death or last follow-up. Tumor staging was performed according to the Seventh edition of the Tumor-Node-Metastasis (TNM) Classification of the International Union Against Cancer . HCC samples (n=123) were used to generate a tissue microarray (Shanghai Biochip Co., Ltd. Shanghai, China). The fundamental processes of the Immunohistochemistry assay have been previously mentioned . The pathological types of paraffin-embedded slides were rechecked by HE staining before IHC analysis. A DAB substrate kit (Zsbio Commerce Store) was used according to the manufacturer’s instructions. Scores for staining frequency (0 <10%, 1 = 10–25%, 2 = 26–50%, 3 = 51–75%, 4 = >75%) and intensity (0 = negative, 1 = weak, 2 = moderate, 3 = intense staining) were used. A DAB substrate kit (Zsbio Commerce Store) was used according to the manufacturer’s instructions. The overall staining score (OSS) was calculated by multiplying the staining area percentage score by the intensity score. 0–6 were considered low, and 9–12 were considered high. The results were scored by two pathologists who were blinded to clinicopathological data. All procedures were approved by the Ethics Committee of Tongji Hospital and conducted according to the Declaration of Helsinki Principles. Written and informed consent was obtained from each patient.
HLF, Hep3B, SK-Hep1, and HEK293T cells were cultured in high-glucose Dulbecco’s modified Eagle’s medium (DMEM, HyClone, Illinois, USA) supplemented with 10% fetal bovine serum (FBS, Gibco, New York, USA), 100 U/ml penicillin and 100 mg/ml streptomycin (HyClone, Illinois, USA) at 37 °C with 5% CO2. Collagen1 treatment was administered at a final concentration of 20 μg/ml. All the cell lines were obtained from the Hepatic Surgery Center, Tongji Hospital, Huazhong University of Science and Technology, China. 293 T cells were purchased from the China Center for Type Culture Collection (Wuhan, China).
To calculate sphere-forming efficiency, cells were single-cell sorted into 96-well plates coated with an ultra-low attachment surface (Corning). The cells were grown under anchoring-independent conditions in selective serum-free DMEM/F12 medium supplemented with 1X B27 supplement minus vitamin A (LifeTechnologies), human recombinant EGF (hEGF) (R&D System) (20 ng/mL), and bFGF (R&D system) (20 ng/mL). After seven days, the spheres formed were counted.
Colony formation assay.
Cells were suspended and seeded at a density of 1000 cells per well. After three weeks of cultivation, the cells were fixed and stained with 10% formalin and 0.1% crystal violet. The relative number of colonies was counted.
Flow cytometric analysis.
Cells were stained with a PE-conjugated CD133 (BD Biosciences) antibody in PBS with 2% FBS at 4 °C for 30-60 mins. Isotype-matched mouse immunoglobulins were used as the controls. The samples were analyzed using the CytoFLEX flow cytometer and CytExpert software (Beckman Coulter).
Drug resistance and IC50
Cells in the logarithmic growth phase were uniformly inoculated into 96-well plates (1000/well) and different diluted drug concentrations were added to the medium. Six replicate wells were used for each concentration gradient assay. The cells were cultured in an incubator for 48 h. Subsequently, 10 μL of CCK-8 solution (Vazyme, Nanjing, China) and 90 μL of culture medium were added, followed by incubation in the dark at 37 °C for 1 h. Absorbance at 450 nm was detected using GraphPad Prism (Version 8, GraphPad Software, San Diego, CA).
Small interfering RNA, plasmids and lentivirus
Transient small interfering RNA (siRNA) assays were described before . Sequences of siRNA are listed in Supplementary Table S4. Synthesizing siRNA duplexes were produced and validated by Ribobio (Guangzhou, China). Human DDR1 (NM_001954.4) cDNA and pcDNA3.1, plasmids were gifts from the Hepatic Surgery Center, Tongji Hospital, Huazhong University of Science and Technology, China. pBABE-puro (plasmid #1764), gag/pol (plasmid #14887), pMD2.G(plasmid #12259), pLKO.1 - TRC cloning vector (plasmid #10878), and psPAX2 (plasmid #12260) were purchased from Addgene (Cambridge, MA, USA). To establish pBABE- FLAG- DDR1, human cDNA was cloned into the BamHI/EcoRI site of the pBABE-puro retroviral vector and identified via sequencing (TSINGKE, Wuhan, China). To construct pLKO.1-scramble, pLKO.1-shDDR1, and pLKO.1- shCD44 plasmid, the target double-stranded oligonucleotides (shRNA) sequences, and one non-targeting sequence (negative control, scramble) were annealed and cloned into the AgeI/EcoRI site of the pLKO.1 vector. The shRNA oligo-pair sequences are listed in Supplementary Table S4. Viral production, infection, and establishment of stable cell clones have been described previously . The pcDNA3.1 plasmid inserted by FlAG- or HA-tagged DDR1 and its mutants, FlAG- or HA-tagged CD44, FlAG-tagged PP2AA were constructed according to the ClonExpress II One Step Cloning Kit and Mut Express II Fast Mutagenesis Kit V2 (Vazyme, Nanjing, China) protocol and were identified by sequencing (TSINGKE, Wuhan, China). The CD44-overexpressing lentivirus was purchased from DesignGene Biotechnology (Shanghai, China)
Coomassie blue staining and mass spectrometry
293 T cells transiently transfected with FLAG-DDR1 or FLAG-vector were lysed in IP lysis buffer (25mM Tris-HCl (pH 7.4), 150mM NaCl, 1% NP-40, 1mM EDTA, 10% glycerol, and protease inhibitor cocktail), and IP assays were performed as described previously . The eluted proteins were separated by SDS-PAGE followed by coomassie blue staining. There was a significant difference in the gel bands between the FLAG-DDR1 and FLAG-vector groups among the molecular weight 70kd-125kd regions. Mass spectrometry was performed and analyzed using the ptm-bio lab (PTM BIO, Hangzhou, China).
Reverse transcription PCR and Real-time quantitative PCR
Total cell RNA was extracted using TRIzol (Invitrogen, Carlsbad, CA, USA). Reverse transcription was carried out using HiScript II Q Select RT SuperMix (+gDNA wiper) (Vazyme, Nanjing, China) according to the manufacturer’s instructions. Real-time fluorescence quantitative PCR was performed using the ChamQ Universal SYBR qPCR Master Mix (Vazyme, Nanjing, China). Gene expression levels were normalized to those of glyceraldehyde-3-phosphate dehydrogenase (GAPDH) in the same samples. Each sample was analyzed independently in triplicate. The primers used are listed in Supplementary Table S5.
Immunoblotting, co-immunoprecipitation (co-IP)
Immunoblotting and co-immunoprecipitation assays were performed as previously described . Briefly, cells were collected and lysed on ice using IP lysis buffer. Lysates were incubated with protein G agarose for 2 h and immunoprecipitated with indicated antibodies overnight at 4 °C. The lysates were incubated with protein G agarose beads for 1 h followed by 1wash using IP lysis buffer and three washes with washing buffer (300mM NaCl, 1.0mM EDTA, 25mM Tris-HCl, pH7.4, 1.0% NP-40). The beads were eluted with 2×SDS-PAGE loading buffer and subjected to immunoblotting.
Glutathione S transferase (GST) pull-down
The human DDR1 gene encoding was subcloned into a pET-42b vector. Following transformation and amplification in BL21(DE3) E. coli, recombinant GST-DDR1 fusion proteins were purified by GST Purification MagBeads (Absin Bioscience, Shanghai, China). GST (5 μg) or GST-DDR1 (5 μg) was incubated with recombinant human His-CD44 protein (5 μg) (ABclonal Technology, Wuhan, China) in PBS at 4 °C for 4 h under constant mixing. The bound proteins were incubated and immunoprecipitated with GST antibody-protein A beads (Absin Bioscience, Shanghai, China). After washing away the unbound proteins three times, the bound proteins were analyzed by SDS-PAGE and immunoblotting.
Immunofluorescence and Confocal microscopy imaging.
Immunofluorescence assays were performed as described previously . Briefly, after the indicated treatments, cells were cultured on coverslips for 12 h, fixed in 4% paraformaldehyde for 15 min at room temperature, and permeabilized with 0.5% Triton X-100 for 10 min. After blocking, slides were incubated with primary antibody overnight at 4 °C in a humidified box. The slides were then washed thrice and incubated with secondary antibody for 4 h at room temperature in a humidified box. Finally, the cell nuclei were stained with 40, 60-diamidino-2-phenylindole (DAPI, Sigma-Aldrich) for 5 min. The resulting signals were visualized using a confocal laser scanning microscope (Olympus FV1000, Tokyo, Japan).
Dual-Luciferase Reporter Assay
Cells were seeded in a 24-well plate at a density of 10000 cells per well. The next day, cells were co-transfected with 200ng pGL4.17 and 4ng pRL-TK plasmids, and transfections were performed using Lipofectamine 3000 (Thermo Fisher Scientific, Waltham, MA, United States) according to the manufacturer's instructions. 12 h after transfection, cells were replaced with fresh medium and allowed to grow for 48 h. Luciferase activity was detected with the Dual-Luciferase Reporter Assay System (Promega, Madison, WI, United States) using a GloMax 20/20 Luminometer (Promega, Madison, WI, United States). Firefly luciferase activity was normalized to the Renilla luciferase activity.
Reagents and antibodies
Rattail collagen I (BD Bioscience, MA, USA), Puromycin, trypsin-EDTA, Opti-MEM, and polybrene were obtained as previously described . DDR1 inhibitor 7rh, and YAP inhibitor verteporfin were purchased from MedChemExpress (Shanghai, China). The Lipofectamine 3000 reagent was purchased from Invitrogen (Life Technologies, Carlsbad, CA, USA). All the antibodies used in this study are listed in Supplementary Table S6.
Extreme limiting dilution xenograft tumor formation
Thirty-six male NOD/SCID mice (4 weeks old) were divided into two groups (18 mice per group): a group receiving SK-Hep1 cells with vector and a group receiving cells overexpressing DDR1.The cells were diluted and transplanted by subcutaneous injection. Tumors were harvested at the end of the experiment for documentation. Tumor-initiating cell frequency was calculated using the extreme limiting dilution analysis (ELDA) software . Animal assays were carried out according to Wuhan Medical Experimental Animal Care Guidelines.
Drug treatments in vivo
Cells (2 × 106) were subcutaneously injected into 4-week-old nude mice. After 2 weeks, the animals were orally administered verteporfin (50 mg/kg) and 7rh (50 mg/kg) twice daily for two weeks. Drugs delivered by oral gavage were dissolved in DMSO and diluted in corn oil. Control mice were treated with the vehicle following an identical procedure. Mice were killed (within 48 h of the last treatment), and samples were obtained for histopathological and immunohistochemical analysis.
Cells were lysed using TRIzol reagent (Invitrogen, Life Technologies, Carlsbad, CA, USA). RNA extraction, library construction, high-throughput sequencing, and data analysis were conducted by Novogene Technology Co., Ltd. (Beijing, China).
Tumor segmentation and feature extraction
65 patients were retrospectively recruited between February 1, 2014 and December 31, 2015. We randomly selected 45 cases as the training data set (25/20=positive/negative) and another 20 cases as the independent testing data set (11/9=positive/negative). Two independent readers manually delineated all MR images as high-resolution T2-weighted images (T2WI) using an open-source software package (ITK-SNAP, version 3.6.0, www.itksnap.org). Tumors were outlined as regions of interest (ROIs). Two types of images, “original images” and “wavelet images,” were used for the analysis in this study. “Original images” were the images without any transformation, and “wavelet images” were used for image denoising and improving image quality. The original images underwent a three-dimensional (i.e., x, y, and z directions) wavelet transformation using the PyWavelet package in Python. Each image was filtered by a high bandpass filter or low band-pass in the three directions, resulting in 8 combinations of different decompositions: LLH, LHL, HLL, LHH, HHL, HLH, HHH, and LLL (H means high, and L means low). The FeAture Explorer Pro (FAE Pro, V 0.4.1) was used to extract radiomic features. The shape features were extracted only from the “original images.” 18 types of first-order statistical features, 14 types of shape features, and 75 types of texture features [24 Gray Level Co-occurrence Matrix (GLCM) + 14 Gray Level Dependence Matrix (GLDM) + 16 Gray Level Run Length Matrix (GLRLM) + 16 Gray Level Size Zone Matrix (GLSZM) + 5 Neighborhood Gray-tone-difference Matrix (NGTDM)], a total of 107 “original images” features were used for analysis in our study. “Wavelet images” contained 144 first-order statistical features [8 wavelet images × 18] and 600 texture features [8 wavelet images × 75]. All 851 characteristics were extracted from the visible primary tumors, recorded, and stored quantitatively (Supplementary Table S7).
Predictive models establishment
To remove the imbalance in the training data set, we used the synthetic minority oversampling technique (SMOTE) to balance the positive/negative samples. Normalization was applied to the feature matrix. Each feature vector was subtracted from the mean value of the vector and divided by its length. Because the dimension of the feature space was high, we compared the similarity of each feature pair. If the PCC value of the feature pair was greater than 0.99, one of them was removed. After this process, the feature space dimension was reduced, and each feature was independent. Before building the model, we used analysis of variance (ANOVA) to select features. The F-value was calculated to evaluate the relationship between the features and labels. We sorted the features according to their corresponding F-values and assigned a specific number of features to build the model. We used a support vector machine (SVM) as the classifier. The kernel function can map the features into a higher dimension to search the hyper-plane to separate cases with different labels. To determine the hyper-parameter of the model (e.g., the number of features), we applied a cross-validation 5-fold on the training data set. The hyper-parameters were selected according to model performance on the validation data-set. The model performance was evaluated using receiver operating characteristic (ROC) curve analysis. The area under the ROC curve (AUC) was calculated for quantification. Acuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were also calculated at a cutoff value that maximized the value of the Yorden index. We also estimated the 95% confidence interval using bootstrapping with 1000 samples. The above processes were implemented using the FeAture Explorer Pro (FAEPro, V 0.4.1) on Python (3.7.6).
Statistical analyses were performed using GraphPad Prism (Version 8, GraphPad Software, San Diego, CA). The levels of statistical significance for comparing the two groups were evaluated using the non-parametric two-sided Mann–Whitney U-test. To compare several groups with a control group, one-way ANOVA followed by Dunnett’s multiple comparison test was applied. Two-way ANOVA followed by Bonferroni post-test was applied to compare several groups with a control group over time. Statistical significance was set at P < 0.05. Survival curves were generated using the Kaplan–Meier procedure. Survival curves and hazard ratios were evaluated using log-rank tests.