Multiple Machine Learning Methods Identified RRAGD as Novel Biomarkers for Hepatocellular Carcinoma and Liver Cirrhosis

doi:10.21203/rs.3.rs-4836745/v1

Download PDF

Article

Multiple Machine Learning Methods Identified RRAGD as Novel Biomarkers for Hepatocellular Carcinoma and Liver Cirrhosis

https://doi.org/10.21203/rs.3.rs-4836745/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background

Hepatocellular carcinoma (HCC) is a common malignant tumor worldwide, usually developing from cirrhosis. Distinguishing biomarkers between HCC and liver cirrhosis is crucial and limited. Disulfidptosis is a recently discovered form of cell death, and it has important prognostic value for various tumors. The mechanism of disulfidptosis in HCC and liver cirrhosis is still unclear

Methods

RNA sequencing data and single-cell sequencing data related to HCC and liver cirrhosis were applied for high dimensional weighted gene co-expression network analysis (hdWGCNA) and Weighted co-expression network analysis (WGCNA) methods. These methods were used for analysis of disulfidptosis related to HCC and liver cirrhosis. A diagnostic model was constructed based on machine learning. Moreover, in vitro assays demonstrated the influence of RRAGD on disulfidptosis of HCC cells.

Results

Applying machine learning methods, we found 7 disulfidptosis-related genes in HCC and liver cirrhosis, including FXN, HSPA1A, AGPAT2, CCND1, RRAGD, SUSD4 and DKK4. These disulfidptosis-related genes in HCC and liver cirrhosis may be used for diagnosis of HCC and liver cirrhosis. RRAGD was significantly up-regulated in both HepG2 and Huh7 cells. RRAGD knockdown induced disulfidptosis of HCC cells under glucose starvation and SLC7A11 overexpression.

Conclusion

Multiplex analysis based on DRGs correlated strongly with HCC and liver cirrhosis, providing new insights for developing clinical diagnosis tools and designing immunotherapy regimens for HCC and liver cirrhosis patients.

Biological sciences/Cancer

Biological sciences/Cell biology

Biological sciences/Molecular biology

disulfidptosis

hepatocellular carcinoma

liver cirrhosis

diagnosis

Hepatocellular carcinoma (HCC) is the most common primary liver cancer, accounting for about 90% of all liver cancers. Its incidence and lethality are among the highest in the world [1]. The pathogenesis of HCC is complex, with approximately 70%-80% of patients with HCC occurring in the setting of liver cirrhosis [2]. Hepatitis B virus infection, hepatitis C virus infection, alcoholic fatty liver disease, as well as non-alcoholic steatohepatitis are the most common risk factors for liver cancer [3, 4]. Under the long-term effect of these risk factors, the development of HCC usually goes through chronic hepatitis, liver fibrosis, and eventually cirrhosis, which is the classic "trilogy" of liver cancer [5]. The annual incidence of HCC in patients with liver cirrhosis is 2–4%. Liver cirrhosis to HCC undergoes a long developmental process. Under the stimulation of long-term persistent inflammation, the liver microenvironment is altered leading to hepatocellular damage, metabolic and biological function alterations, and ultimately carcinoma, which develops into HCC after proliferation of abnormal nodules [6]. However, the specific mechanism of liver cirrhosis in the development of HCC is still unclear.

In 2023, Liu et al. have first proposed a novel cell death mode, which is a rapid death mode caused by disulfide stress caused by excessive accumulation of intracellular cysteine, and named this new cell death disulfidptosis [7]. This research reveals the mechanism of disulfide stress-induced cell death, that is, excessive accumulation of disulfide molecules causes abnormal disulfide bonding between actin cytoskeletal proteins, interferes with their tissues, and ultimately leads to the breakdown of the actin network and cell death [8]. Wang et al. have established a risk score related to disulfidptosis to guide the prognosis prediction, immune infiltration, and immune therapy response in HCC [9]. Chen et al. have utilized LASSO Cox regression method and identified five immune checkpoint genes highly associated with disulfidptosis, which used to assess HCC prognosis [10]. The potential role of disulfidptosis in HCC and liver cirrhosis is still unclear.

Disulfidptosis play a crucial role in HCC and liver cirrhosis. However, their correlation and specific role in HCC and liver cirrhosis have hardly been studied. We compared disulfidptosis-related genes between HCC and liver fibrosis, and investigated the prognostic significance of these genes. We created a prognostic model to observe the interaction between these genes and HCC and liver fibrosis, which may guide the prognosis prediction, immune infiltration, and immune therapy response in HCC and liver fibrosis.

Data collection

RNA sequencing data (GSE54236) and single-cell sequencing data (GSE212046) were of obtained from Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/). GSE54236 dataset contained 80 HCC tumor tissues and 81 cirrhotic non-malignant tissue samples. GSE212046 dataset included 2 cases of on-tumor tissues with background of liver cirrhosis with steatohepatitis and 2 cases of tumor tissues from HCC patients. Eighty HCC tumor tissue samples in GSE54236 dataset were randomly divided into 3 and 1 portions. Similarly, 81 cirrhotic non-malignant tissue samples in GSE54236 dataset were randomly divided into 3 and 1 portions. Three portions of HCC tumor tissue samples and 3 portions of cirrhotic non-malignant tissue samples were served as training dataset, one portions of HCC tumor tissue samples and one portions of cirrhotic non-malignant tissue samples were served as test dataset.

mRNA sequencing data

Nine tumor tissue samples from HCC patients and 5 cases of liver cirrhosis were collected for mRNA sequencing. The total RNA was extracted from tissues by Trizol method. The mRNA with polyA tail was enriched by Oligo (dT) magnetic beads to construct the sequencing library. Based on the Illumina NovaSeq6000 sequencing platform, these libraries were sequenced by double terminal (Paired-end, PE).

scRNA data analysis

Applying R packages "Seurat" and "NormalizeData" function, the scRNA data was analyzed. In order to ensure the data quality, the batch effects of the non-biotechnological bias were corrected. The number of highly variable genes was set to 3000 and identified using "FindVariableFeatures". Principal component analysis (PCA) was used to reduce the order of 3000 high variable genes by Seurat RunPCA function. The principal components were displayed by Seurat ELbowPlot function, and the first 15 principal components (PC) were selected for follow-up analysis. All samples were integrated by seurat's rpca method correction. Then, the "DIMS" parameter was set as 30, and the " k-nearest neighbor (KNN)" method was used to cluster the cells. Cell clusters were annotated with known cell types based on the cell type-specific markers obtained from the cellmarker 2.0 database [11].

Disulfidptosis‑related genes (DRGs) scoring

Twenty-three DRGs, SLC7A11, GYS1, NDUFS1, NDUFA11, NUBPL, NCKAP1, LRPPRC, SLC3A2, RPN1, ACTN4, ACTB, CD2AP, CAPZB, DSTN, FLNA, FLNB, INF2, IQGAP1, MYH10, MYL6, MYH9, PDLIM1 and TLN1, were colletced for DRGs scoring [7]. Applying “seurat” with addmoudlescore function, DRGs scoring was calculated. Cells were categorized into two distinct groups following the DRG-AUC (DRGs Area Under the Curve) values: those with high DRG-AUC and those with low DRG-AUC, with the median value being utilized as the threshold for classification.

Single-cell inferred chromosomal copy number variation (CNV)

To distinguish between benign and malignant cells in the tumor microenvironment, the copykat program (version 1.1.0) was applied to determine the genomic copy number distribution of individual cells. By integrating Bayesian techniques and hierarchical clustering, cells were classified as diploid normal cells or aneuploid tumor cells applying copykat program. The copykat program defines a model using a Gaussian mixture model (GMM), which assumes that a cell's gene expression is a mixture of three Gaussian models: amplification, deletion, and neutral states. Cells in which neutral genes account for at least 99% of expressed genes are categorized as high-confidence diploid cells. Therefore, we classify cells as diploid (benign) and aneuploid (tumor).

HdWGCNA analysis

The malignant cells and non-malignant cells in GSE212046 dataset were analyzed by R packages hdWgcna. The hdWGCNA package with KNN algorithm was employed to identify similar cells that could be aggregated. The average or sum expression of these cells was then calculated, resulting in a low sparse metacell gene expression matrix. The SetDatExpr function was utilized to specify Treg cells for constructing the expression matrix. Next, we performed parameter scans using the TestSoftPowers function to determine the optimal soft power threshold for constructing the co-expression network. The ConstructNetwork function was employed to establish the co-expression network based on the optimal soft threshold. Subsequently, the ModuleEigengenes (ME) function was utilized to calculate the module feature genes by performing principal component analysis (PCA) on a subset of the gene expression matrix specific to each module. Additionally, the ModuleExprScore function with either Seurat or UCell algorithm was used to compute the central gene feature score for each module. To visualize the correlation between modules, the ModuleCorrelogram function was applied, considering the hME, ME, or hub gene scores. Important modules were filtered based on their correlation coefficients and p-values with specific traits (e.g., tissue type, malignant cells, disulfide apoptosis score, and disulfide apoptosis group) using the GetModuleTraitCorrelation function.

Transcriptome data analysis

The R package ssGSEA was used to calculate the disulfidptosis gene score in the transcriptome data. The immune infiltration score for 28 immune cells was calculated based on the immune cell gene set [12]. The GSE54236 dataset was analyzed using the R package WGCNA, and the Pearson correlation coefficient was used to construct a Weight co-expression network. The cutreeDynamic function was used to divide the different modules with the "minClusterSize" parameter was set to 30. The correlation of each module with tissue type and disulfidptosis score was calculated to select significant modules. KEGG and GO analyses were performed using the R package ClusterProfiler based on the Entrez IDs of the genes in the important modules.

Construction of machine learning models

Based on the results of hdWgcna and Wgcna analyses with both the single-cell level and the transcriptome level, clusters of genes that correlate with both tissue type and disulfidptosis scores were selected for machine learning to select features and train models. Feature selection was performed using SelectFromModel in python scikit_learn library on RidgeCV, LassoCV, LDA, RandomForest, LinearSVC, LogisticRegression respectively. LogisticRegression, KNN, RandomForest, GradientBoosting, Linear Discriminant Analysis (LDA) were utilized to train the model. GridSearchCV function was used to optimize the model parameters. Models with fewer number of features, higher AUC values without overfitting and underfitting were selected to diagnose liver cancer cirrhosis. The models were validated on nine samples of liver cancer and five samples of cirrhosis.

Cell culture

Human hepatic stellate cells (LX-2) and liver cancer cells (HepG2 and Huh7) were obtained from ATCC (Manassas, VA, USA). All cells were maintained in RPMI-1640 medium (Gibco, Grand Island, NY, USA) in the presence of 10% fetal bovine serum (FBS; Gibco) and 1% penicillin-streptomycin (Sangon Biotech, Shanghai, China) at 37°C and 5% CO₂ atmosphere. Cells were treated with different cell death inhibitor, ferroptosis inhibitor ferrostatin-1 (Ferr-1; 3 µg/mL; MedChemExpress, Monmouth Junction, NJ, USA), Caspase inhibitor Z-VAD-FMK (Z-VAD; 50 µM; MedChemExpress), disulfide bond reductant tris − (2-carboxyethyl) - phosphine (TCEP 2 mM; MedChemExpress). Cells were treated with 1% DMSO (MedChemExpress) as control.

Cell transfection

The SLC7A11 overexpression vector (SLC7A11) and small interfering RNA specially targeting RRAGD (si-RRAGD-1/2/3) were obtained from Sangon Biotech (Shanghai). Empty vector (EV) and scrambled siRNA (si-NC) were served as control. Cells were transfected with SLC7A11/EV or si-NC/si-SLC7A11 applying lipofectamine 2000 reagent (Thermo Fisher Scientific, San Jose, CA, USA).

Quantitative real-time PCR (qRT-PCR)

RNA extraction was carried out applying TRIzol regent (Thermo Fisher Scientific). PrimeScript™ RT reagent Kit (Takara, Dalian, China) and TB Green® Premix Ex Taq™ (Takara) were applied to synthesize cDNA and PCR reaction. GAPDH served as loading control. The relative expression of mRNA was analyzed by 2^−ΔΔCT method.

Western blotting

Utilizing RIPA Lysis Buffer (Thermo Fisher Scientific), total proteins were extracted from cells. The proteins were separated by 10% SDS-PAGE, and then transferred on a nitrocellulose membrane. The membranes were incubated with 5% skimmed milk, and then incubated with primary antibodies, SLC7A11 (1:1000 dilution; Cat#ab307601; Abcam, Cambridge, MA, USA), RRAGD (1:1000 dilution; Cat# ab187679; Abcam) or anti-GAPDH (1:5000 dilution; Cat#10494-1-AP; Proteintech, Wuhan, China). Goat anti-rabbit HRP-IgG (1:2000 dilution; Cat#SA00001-2; Proteintech) was utilized to visualize protein bands. Finally, WB bands were developed by Enhanced chemiluminescence reagent (Beyotime, Shanghai, China) and analyzed by Image J software.

Detection of ATP

ATP Assay Kit (Beyotime) was used to detect the levels of ATP in cells following the protocol of manufacturer. Chemiluminescence of each sample was detected on a Luminoskan Ascent (Labsystems, Franklin, MA, USA).

Cell death

Annexin V-FITC Apoptosis Detection Kit (Beyotime, Shanghai, China) was applied to assess cell death. Cell suspension (10⁴ cells) was incubated with 5 µL Annexin V-FITC and 10 µL PI in darkness at 25°C for 20 min. Finally, cell death was analyzed on a FACSCalibur flow cytometer (BD Biosciences, San Diego, CA, USA) in 1 h.

Statistical methods

All statistical analyses and plots were realized through R language or python package. The spearman correlation coefficient was used to assess the correlation between variables. Two-tailed Student’s t test and one-way ANOVA were used to analyze the statistical difference. P values less than 0.05 were considered statistically significant.

Differential analysis of disulfidptosis status in HCC and liver cirrhosis at single-cell level

Based on GSE212046 dataset, the sifferent intercellular batch effects were processed utilizing the rpca method of the seurat package. Following batch correction, all cells were clustered into 16 clusters (Fig. 1A). Based on Cellmarker 2.0 database, we annotated these clusters and classified them into hepattocytes, endothelial, fibroblasts, cholangiocytes, mesenchymal cells, macrophages, plasma cells and T cells (Fig. 1B-C). Compared with liver cirrhosis samples, hepatocytes dominated HCC tumor tissues. Other cell types were decreased in HCC tumor tissues (Fig. 1D). Moreover, the disulfidptosis score of each cell type was calculated. Based on the median disulfidptosis score, these cell types were categorized into two groups with high or low disulfidptosis score (Fig. 1E-F). Compared with liver cirrhosis samples, the disulfidptosis scores of hepattocytes, endothelial cells, fibroblasts, T cells, and macrophages was increased in HCC tumor tissue samples (Fig. 1G). Prediction of malignant cells using copykat reveals that malignant hepatocytes were predominantly derived from HCC samples and were accompanied by lower of disulfidptosis scores (Fig. 1H).

HdWGCNA identified the hub genes of tumor related to disulfidptosis

A total of 12 gene modules with similar expression patterns were identified based on the optimal threshold for hdwgcna analysis (Fig. 2A-B). The genes in the blue, brown, green, black, greenyellow, tan, pink modules were significantly associated with both tissue type and malignant cells as well as disulfidptosis with high correlation (Fig. 2C-D). The core gene scores of these modules were calculated and presented in the UMAP clustering diagram.

WGCNA identified the hub genes of tumor related to disulfidptosis

A total of 49 gene modules with similar expression patterns were classified at the transcriptional level based on wgcna analysis, as shown in Fig. 3A-B. Among them, the genes in white lightpink4, darkolivegreen, darkred, darkgrey, lightgreen, black, cyan, ivory modules were associated with both tissue type and disulfidptosis score with high correlation (Fig. 3C).

Selection of DRGs using Machine learning models

A total of 299 genes were screened based on the WGCNA and Validation set (Fig. 4A). Then, we conducted GO and KEGG analysis on these 299 DEGs. Cytoplasmic translaton, ribonucleoprotein complex biogensis and ribosomal small subunit were main enriched in biological process (BP) terms. Cytosolic ribosome, ribosomal subunit and rbosome were main enriched in cell component (CC) terms. Structural constituent of ribosome, rRNA binding and cadherin binding were concentrated in molecular function (MF) terms (Fig. 4B). KEGG annotations showed that these genes were associated with ribosome, COVID-19 and non-alcoholic fatty liver disease (Fig. 4C). Afterwards, we carried out standard feature engineering on these genes, and the AUC value was used to evaluate the performance of model in training and test sets. By comparing the results of the training set and test set, we found that there may be overfitting in some models. For KNN and GradiendBoosting, the AUC values were typically as high as 1 in the training set but less than 0.9 in the test set, which may produce overfitting in the training models. When modeling on LogisticRegression using the features selected by LassoCV, the AUC value of 0.9 in the training set was 0.94 in the test set, which may be insufficient in training the model. Considering the results of the training and test sets and the number of features, we finally chose the RandomForest model constructed from the marker genes identified by LassoCV to distinguish HCC from liver cirrhosis (Fig. 4D-E). RandomForest model contained 7 marker genes, FXN, HSPA1A, AGPAT2, CCND1, RRAGD, SUSD4 and DKK4. We demonstrated these seven marker genes and their 11 importance in RandomForest (Fig. 4F).

Identification and verification of potential biomarker combinations for the classification of HCC and liver cirrhosis Patients

We further used ROC curves, confusion_matrix and PCA to analyze the effect of these seven genes on the training, test and Validation sets. On the validation set, we harvested an AUC value of 0.89, confusion_matrix demonstrated the diagnostic results for each sample. PCA analysis revealed the differences of these seven genes between hepatocellular carcinoma and cirrhosis samples (Fig. 5A-I).

Correlation between model genes, disulfidptosis and immune cell infiltration

We calculated immune infiltration scores and disulfidptosis scores using the ssGSEA method in the validation set (Fig. 6A). We found that immune cell scores were higher in HCC as compared with liver cirrhosis, and positively correlated with disulfidptosis scores. The correlation of 7 model genes with immune infiltration was calculated, showning that RRAGD was significantly correlated with most of the immune cell scores (Fig. 6B).

The different expression of model genes and disulfidptosis scores in different cells

The expression of 7 marker genes in single cell level was shown in Fig. 7A-D. These genes were highly expressed in hepatocytes. However, these genes were down-regulated in liver malignant cells, which was consistent with the expression trend of disulfidptosis scores. The expression of some genes differed in liver malignant and non-malignant cells, such as AGPAT2, DKK4, CCND1, RRAGD, SUSD4. We further analyzed the differences of these genes in HCC and liver cirrhosis as well as in the high and low score of disulfidptosis. Most of the genes were more highly expressed in the high disulfidptosis group and the HCC group. Among them, the expression of RRAGD and DKK4 in hepatocytes highly overlapped with the expression trend of disulfidptosis scores and malignant cell distribution. Meanwhile, RRAGD had the highest correlation with the disulfidptosis score in the bulk data with a p-value of 0.004 (Fig. 7E).

RRAGD knockdown induced disulfidptosis of HCC cells

We further verified the 7 disulfidptosis-related genes in HCC cells. qRT-PCR results showed that FXN, HSPA1A, CCND1 were severely down-regulated in HepG2 and Huh7 cells. Compared with LX-2 cells, the expression of AGPAT2, SUSD4 and DKK4 was notably elevated in HepG2 or Huh7 cells. RRAGD was highly expressed in both HepG2 and Huh7 cells (Fig. 8A). RRAGD protein was also up-regulated in HepG2 and Huh7 cells (Fig. 8B). Then, SLC7A11 was up-regulated in HepG2 and Huh7 cells, as determined by WB assay (Fig. 8C). Moreover, RRAGD was silenced in both HepG2 and Huh7 cells. RRAGD was severely down-regulated in HepG2 and Huh7 cells in the presence of si-RRAGD-1/2/3, especially si-RRAGD-1 (Supplementary Fig. 1A). RRAGD silencing notably inhibited the expression of RRAGD, while had no influence on SLC7A11 expression in HepG2 and Huh7 cells (Fig. 8D). We then test the influence of RRAGD knockdown on cell death of HCC cells. Under glucose starvation and SLC7A11 overexpression conditions, RRAGD silencing notably elevated cell death of HepG2 and Huh7 cells (Fig. 8E, Supplementary Fig. 1B). Glucose starvation depletes ATP. In HepG2 and Huh7 cells, glucose starvation does decrease the intracellular levels of ATP, while RRAGD knockdown had no influence on ATP levels (Fig. 8F). Thus, RRAGD knockdown induced cell death was not caused by ATP depletion under glucose starvation and overexpression of SLC3A7. Additionally, HCC cells were treated with cell death inhibitors, including the ferroptosis inhibitor Ferr-1, apoptosis inhibitor Z-VAD and disulfidptosis inhibitor TCEP. TCEP treatment notably reduced cell death of HepG2 and Huh7 cells. Ferr-1 and Z-VAD-FMK had no influence on cell death of HepG2 and Huh7 cells (Fig. 8G and Supplementary Fig. 1C). All these data indicated that RRAGD knockdown induced disulfidptosis of HCC cells under glucose starvation and SLC7A11 overexpression.

Disulfidptosis-induced cell death is a novel form of cell death caused by disulfide stress, and its mechanism is different from known ferroptosis, apoptosis, autophagy, and necroptosis. Under SLC7A11 overexpression and glucose starvation, a large amount of disulfide molecules are produced inside the cell, inducing severe disulfide stress, ultimately leading to the breakdown of the actin network and cell death [7]. Disulfidptosis death plays an important role in tumor occurrence and development, and can serve as a novel molecular weight target [13–15]. Yu et al. also have pointed out that disulfidptosis is related to tumors, and proposed that disulfidptosis may become an effective treatment strategy for tumors [14]. However, the role of disulfide death in HCC remains unclear. The present work applied RNA sequencing and single-cell sequencing to analyze the differential expression of DRGs between HCC and liver cirrhosis. We developed a diagnostic model, consisting of 7 genes, based on machine learning. These disulfidptosis-related genes in HCC and liver cirrhosis may be used for diagnosis of HCC and liver cirrhosis.

Applying machine learning methods, we found 7 disulfidptosis-related genes in HCC and liver cirrhosis, including FXN, HSPA1A, AGPAT2, CCND1, RRAGD, SUSD4 and DKK4. Among them, RRAGD was significantly up-regulated in both HepG2 and Huh7 cells. Thus, we verified the mechanism of RRAGD in HCC. RRAGD is a member of the Rag GTP-binding protein family. It plays a key role in mediating the amino acid-stimulated mTOR signaling pathway, which is a critical pathway that determines the rate of cell growth and proliferation [16]. Previous studies have confirmed that RRAGD participates in the progression of various cancers, such as cervical cancer and ovarian cancer [17, 18]. Ding et al. have analyzed the correlation between RRAGD expression and prognosis of HCC patients by Kaplan-Meier's analysis, and validated the role of RRAGD in HCC in vitro [19]. The results have showed that high RRAGD expression is closely associated with poor prognosis of HCC. RAGD knockdown inhibits aerobic glycolysis in HCC cells. LncTUG1 elevates RRAGD expression by sponging miR-144-3p, and then aggravates the malignant progression of HCC [20]. Guo et al. have applied machine learning algorithms and identified seven differentially expressed genes, GPC3, ACSM3, SPINK1, COL15A1, TP53I3, RRAGD, and CLDN10, as potential biomarkers associated with HCC immune infiltration [21]. In the present work, we identified 7 disulfidptosis-related genes, including FXN, HSPA1A, AGPAT2, CCND1, RRAGD, SUSD4 and DKK4. RRAGD knockdown induced disulfidptosis of HCC cells under glucose starvation and SLC7A11 overexpression.

In conclusion, multiplex analysis based on DRGs correlated strongly with HCC and liver cirrhosis, providing new insights for developing clinical diagnosis tools and designing immunotherapy regimens for HCC and liver cirrhosis patients.

Consent for publication

The participant has consented to the submission of the research to the journal.

Availability of data and materials

The datasets presented in this study can be found in online repositories. The datasets generated and/or analysed during the current study are available in the (https://www.ncbi.nlm.nih.gov/geo/) repository.RNA sequencing data (GSE54236) and single-cell sequencing data (GSE212046) were of obtained from Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/).

Competing interests

The authors declare no competing interests.

Funding

Weilong ZOU was supported by Doctoral Research Foundation Project of the Affiliated Hospital of Guizhou Medical University, gyfybsky-2022-1.

Authors' contributions

W-l. Z, B.L planned the experiments and revised the manuscript. H.M ,Z-q. L performed the experiments and prepared a draft of the manuscript. Y.S, B-y. S and T.S performed the statistical analysis. W-l. Z, Z-q. L and J. Z conceived the project and edited the manuscript. W-l. Z, B.L and Z-q. L discussed the results. All the authors read and approved the final manuscript.

Acknowledgements

None.

Xia, C., et al., Cancer statistics in China and United States, 2022: profiles, trends, and determinants. Chin Med J (Engl), 2022. 135(5): p. 584–590.
Moon, A.M., A.G. Singal, and E.B. Tapper, Contemporary Epidemiology of Chronic Liver Disease and Cirrhosis. Clin Gastroenterol Hepatol, 2020. 18(12): p. 2650–2666.
Ganesan, P. and L.M. Kulik, Hepatocellular Carcinoma: New Developments. Clin Liver Dis, 2023. 27(1): p. 85–102.
Huang, D.Q., et al., Changing global epidemiology of liver cancer from 2010 to 2019: NASH is the fastest growing cause of liver cancer. Cell Metab, 2022. 34(7): p. 969–977.e2.
Paradis, V. and J. Zucman-Rossi, Pathogenesis of primary liver carcinomas. J Hepatol, 2023. 78(2): p. 448–449.
Huang, D.Q., et al., Global epidemiology of alcohol-associated cirrhosis and HCC: trends, projections and risk factors. Nat Rev Gastroenterol Hepatol, 2023. 20(1): p. 37–49.
Liu, X., et al., Actin cytoskeleton vulnerability to disulfide stress mediates disulfidptosis. Nat Cell Biol, 2023. 25(3): p. 404–414.
Liu, X., et al., Cystine transporter regulation of pentose phosphate pathway dependency and disulfide stress exposes a targetable metabolic vulnerability in cancer. Nat Cell Biol, 2020. 22(4): p. 476–486.
Wang, T., et al., Disulfidptosis classification of hepatocellular carcinoma reveals correlation with clinical prognosis and immune profile. Int Immunopharmacol, 2023. 120: p. 110368.
Chen, Y., et al., A novel disulfidptosis-related immune checkpoint genes signature: forecasting the prognosis of hepatocellular carcinoma. J Cancer Res Clin Oncol, 2023. 149(14): p. 12843–12854.
Filliol, A., et al., Opposing roles of hepatic stellate cell subpopulations in hepatocarcinogenesis. Nature, 2022. 610(7931): p. 356–365.
Charoentong, P., et al., Pan-cancer Immunogenomic Analyses Reveal Genotype-Immunophenotype Relationships and Predictors of Response to Checkpoint Blockade. Cell Rep, 2017. 18(1): p. 248–262.
Wang, Z., et al., A novel disulfidptosis-associated expression pattern in breast cancer based on machine learning. Front Genet, 2023. 14: p. 1193944.
Meng, Y., X. Chen, and G. Deng, Disulfidptosis: a new form of regulated cell death for cancer treatment. Mol Biomed, 2023. 4(1): p. 18.
Feng, Z., et al., Identification a unique disulfidptosis classification regarding prognosis and immune landscapes in thyroid carcinoma and providing therapeutic strategies. J Cancer Res Clin Oncol, 2023. 149(13): p. 11157–11170.
Di Malta, C., et al., Transcriptional activation of RagD GTPase controls mTORC1 and promotes cancer growth. Science, 2017. 356(6343): p. 1188–1192.
Wang, G., et al., miR-99a-5p inhibits glycolysis and induces cell apoptosis in cervical cancer by targeting RRAGD. Oncol Lett, 2022. 24(1): p. 228.
Wu, M., et al., Integrated analysis of lymphocyte infiltration-associated lncRNA for ovarian cancer via TCGA, GTEx and GEO datasets. PeerJ, 2020. 8: p. e8961.
Ding, L. and X. Liang, Ras related GTP binding D promotes aerobic glycolysis of hepatocellular carcinoma. Ann Hepatol, 2021. 23: p. 100307.
Chen, W., et al., LncTUG1 contributes to the progression of hepatocellular carcinoma via the miR-144-3p/RRAGD axis and mTOR/S6K pathway. Sci Rep, 2023. 13(1): p. 7500.
Guo, X., et al., Identification and Validation of a Novel Immune Infiltration-Based Diagnostic Score for Early Detection of Hepatocellular Carcinoma by Machine-Learning Strategies. Gastroenterol Res Pract, 2022. 2022: p. 5403423.

No competing interests reported.

S1.tif
Supplementary figure 1 Cell death of HepG2 and Huh7 cells. HepG2 cells were transfected with si-RRAGD/si-NC. (E) Flow cytometry examined apoptosis of HepG2 and Huh7 cells. (F) The levels of ATP in HepG2 and Huh7 cells were assessed. (G) The transfected HepG2 and Huh7 cells were treated with DMSO, Z-VAD, Ferr-1 or TCEP. Flow cytometry examined apoptosis of HepG2 and Huh7 cells. *P < 0.05, **P < 0.01, ***P < 0.001 vs si-NC group.

Download PDF

Reviews received at journal
07 Nov, 2024
Reviewers agreed at journal
03 Nov, 2024
Reviewers agreed at journal
31 Oct, 2024
Reviewers agreed at journal
30 Oct, 2024
Reviewers invited by journal
23 Aug, 2024
Editor assigned by journal
23 Aug, 2024
Editor invited by journal
16 Aug, 2024
Submission checks completed at journal
16 Aug, 2024
First submitted to journal
31 Jul, 2024

You are reading this latest preprint version

Multiple Machine Learning Methods Identified RRAGD as Novel Biomarkers for Hepatocellular Carcinoma and Liver Cirrhosis

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

Introduction

Materials and methods

Data collection

mRNA sequencing data

scRNA data analysis

Disulfidptosis‑related genes (DRGs) scoring

Single-cell inferred chromosomal copy number variation (CNV)

HdWGCNA analysis

Transcriptome data analysis

Construction of machine learning models

Cell culture

Cell transfection

Quantitative real-time PCR (qRT-PCR)

Western blotting

Detection of ATP

Cell death

Statistical methods

Results

Differential analysis of disulfidptosis status in HCC and liver cirrhosis at single-cell level

HdWGCNA identified the hub genes of tumor related to disulfidptosis

WGCNA identified the hub genes of tumor related to disulfidptosis

Selection of DRGs using Machine learning models

Correlation between model genes, disulfidptosis and immune cell infiltration

The different expression of model genes and disulfidptosis scores in different cells

RRAGD knockdown induced disulfidptosis of HCC cells

Discussion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1