Open Access

Machine learning model reveals roles of interferon‑stimulated genes in sorafenib‑resistant liver cancer

  • Authors:
    • Deok Hwa Seo
    • Ji Woo Park
    • Hee Won Jung
    • Min Woo Kang
    • Byung Yoon Kang
    • Dong Yeup Lee
    • Jae Jun Lee
    • Seung Kew Yoon
    • Jeong Won Jang
    • Jae Gyoon Ahn
    • Pil Soo Sung
  • View Affiliations

  • Published online on: July 15, 2024     https://doi.org/10.3892/ol.2024.14571
  • Article Number: 438
  • Copyright: © Seo et al. This is an open access article distributed under the terms of Creative Commons Attribution License.

Metrics: Total Views: 0 (Spandidos Publications: | PMC Statistics: )
Total PDF Downloads: 0 (Spandidos Publications: | PMC Statistics: )


Abstract

HCC (Hepatocellular carcinoma) is the most common malignant tumor; however, the molecular pathogenesis of these tumors is not well understood. Sorafenib, an approved treatment for HCC, inhibits angiogenesis and tumor cell proliferation. However, only ~30% of patients are sensitive to sorafenib and most show disease progression, indicating resistance to sorafenib. The present study used machine learning to investigate several mechanisms related to sorafenib resistance in liver cancer cells. This revealed that unphosphorylated interferon‑stimulated genes (U‑ISGs) were upregulated in sorafenib‑resistant liver cancer cells, and the unphosphorylated ISGF3 (U‑ISGF3; unphosphorylated STAT1, unphosphorylated STAT2 and IRF9) complex was increased in sorafenib‑resistant liver cancer cells. Further study revealed that the knockdown of the U‑ISGF3 complex downregulated U‑ISGs. In addition, inhibition of the U‑ISGF3 complex downregulated cell viability in sorafenib‑resistant liver cancer cells. These results suggest that U‑ISGF3 induced sorafenib resistance in liver cancer cells. Also, this mechanism may also be relevant to patients with sorafenib resistance.

Introduction

HCC is the most common form of liver cancer and the fourth leading cause of mortality worldwide (1,2). It predominates as the most common form of liver cancer, representing approximately 90% of all cases (3). Infection with Hepatitis B virus (HBV) or Hepatitis C virus (HCV) is a major risk factor for HCC development (4). Metabolic-associated steatohepatitis (MASH) and metabolic-associated fatty liver disease (MAFLD) associated with metabolic syndrome have emerged as contributing factors for HCC (57). Surgical treatments result in high patient survival rates. However, this approach has only been applied to patients with early stage HCC (8,9). Recently, atezolizumab plus bevacizumab has emerged as the therapy for unresectable HCC (10), sorafenib remains a treatment option for these patients.

Sorafenib, a multi-targeted tyrosine kinase inhibitor, has been approved as a therapeutic agent for advanced-stage HCC (11). Sorafenib inhibits tumor cell proliferation by suppressing the activity of BRAF, RAF1, and kinases in the MEK/ERK signaling pathways (12). In addition, the anti-angiogenesis effect of sorafenib is mediated by platelet-derived growth factor receptor (PDGFR-β), vascular endothelial growth factor receptors (VEGFR-1 and VEGFR-2), and c-KIT (13). Despite sorafenib treatment, drug resistance persists in some patients with advanced HCC (14). Only 30% of the patients with HCC benefit from sorafenib, and acquired resistance commonly occurs within 6 months (15). Therefore, the mechanisms underlying sorafenib resistance must be elucidated.

Type I interferons (IFNs) are cytokines with antiviral, anti-proliferative, and immunomodulatory effects that play crucial roles in suppressing viral infections (16). Type I IFNs bind to IFN receptors, leading to the phosphorylation of JAK1 and TYK2 (17). Subsequently, STAT1 and STAT2 are phosphorylated to form a complex with IRF9 (18). The ISGF3 (phosphorylated STAT1, phosphorylated STAT2 and IRF9) complex translocates into the nucleus and upregulates the expression of genes such as interferon-stimulated genes (ISGs) with antiviral functions (19). However, low concentrations of interferon lead to the dephosphorylation of STAT1 and STAT2. At this stage, unphosphorylated STAT1 and STAT2 bind to high levels of IRF9 to form the U-ISGF3 complex (17,20). The U-ISGF3 complex translocates to the nucleus and regulates the expression of genes such as U-ISGs, OAS1, MDA5, and BST2 (21,22). U-ISGs are associated with resistance to chemotherapy and irradiation, which are correlated with resistance to DNA damage (23).

Machine learning has become a powerful tool for the identification of diagnostic genes. We organized the data for input into the machine learning model by generating a vector of gene features that represented the influence of genes on other genes and the influence they received (24). The gene feature vector was extracted from the impact matrix generated by applying a modified PageRank algorithm to a patient-specific gene network constructed using the integrated gene network, gene variants, and gene expression data. A machine learning model was constructed by combining two autoencoders and a deep neural network. The model was trained using gene feature vectors generated from data obtained from patients with liver cancer in The Cancer Genome Atlas (TCGA) (25). These gene feature vectors are labeled based on the presence of genetic mutations and a list of known cancer driver genes. We utilized this trained machine learning model to input gene feature vectors generated from Huh-7 cells to investigate genes associated with sorafenib resistance. The model learns the patterns of influence that known cancer driver genes have on other genes; the more the input gene feature vector has feature patterns similar to the cancer driver genes, the closer the output value will be to 1. Among the genes with high model outputs in samples with sorafenib resistance, particularly those with low model outputs in samples without sorafenib resistance, we selected candidate genes involved in the mechanism of sorafenib resistance and identified the role of STAT1 in this process.

In the current study, machine learning revealed that U-ISGs were highly expressed in sorafenib-resistant liver cancer cells. We further found that the U-ISGF3 complex upregulated U-ISGs in sorafenib-resistant liver cancer cells. Our findings suggest that U-ISGs play a significant role in sorafenib resistance in liver cancer cells, and U-ISGF3 induces sorafenib resistance in liver cancer cells.

Materials and methods

RNA-sequencing analysis

RNA-sequencing analysis was performed using Human liver cancer cell lines (Huh-7, sorafenib resistant Huh-7, HepG2 and sorafenib resistant HepG2). The generated RNA-sequencing libraries were sequenced using an Illumina sequencing system (Macrogen). RNA-sequencing analysis was conducted on Huh-7 (SAMN41561228), sorafenib resistant Huh-7 (SAMN41561229) HepG2 (SAMN41561230) and sorafenib resistant HepG2 (SAMN41561231), obtained from the NCBI Sequenced Read Archive (SRA) database (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1117191, last accessed on May 27, 2024).

Overall process to identify sorafenib resistance-associated genes

The first step in identifying genes associated with sorafenib resistance using machine learning involves generating gene feature vectors from LIHC (Liver Hepatocellular Carcinoma) tumor patient samples obtained from TCGA, as well as Huh-7 samples with sorafenib resistance, along with Huh-7 samples with sorafenib resistance. These gene feature vectors represent the rows and columns of the impact matrix, calculated by applying a modified PageRank algorithm to the patient-specific gene network. This network is constructed using integrated gene network data, the patient's genetic variants, and gene expression data. Once the gene feature vector generation is complete, we construct the training data using the TCGA LIHC samples' gene feature vectors and proceed to train the model. To identify candidate genes related to sorafenib resistance, the gene feature vectors from Huh-7 samples with and without sorafenib resistance are input into the machine learning model. By comparing the sorafenib-resistant and non-sorafenib-resistant sample groups, genes exhibiting particularly high scores in the sorafenib-resistant group are considered candidate genes associated with sorafenib resistance. A visual representation of the overall process is presented in Figure 1.

Construction of patient-specific gene networks

The construction of a patient-specific gene network involves selecting relevant edges from the integrated gene network, considering the specific characteristics of each patient. The integrated gene network was created by combining directed edges from functional interaction networks obtained from Reactome (26). Additionally, gene regulation networks were incorporated from the RegNetwork and TRRUST (27).

For an edge to be included in the patient-specific gene network, it had to satisfy at least one of two conditions: (1) at least one gene from a gene pair connected by an edge exhibited a mutation, and the mutational status utilized information specific to individual patients; (2) the expression of two connected genes aligned with the overall expression pattern observed in the entire cancer sample. To determine this, the RANSAC algorithm was executed 10 times to generate 10 regression models. RANSAC uses the expression values of genes corresponding to the departure node of an edge as input and estimates the regression model parameters to predict the expression of genes corresponding to the arrival node. Regression models with regression coefficients below 0.1 were excluded from consideration. If at least one regression model identified a patient as an inlier, that particular edge was included in the patient's gene network. After acquiring a set of edges that met one or more of the previously mentioned conditions, the construction of the patient-specific gene network was accomplished using Equation (1):

W = (IΨ)Α + Ψ

where A is an adjacent matrix of individual patients, and each component has one value among 0, 1, and 2. Specifically, if the patient has an edge connecting genes i and j, the value of Aij becomes 2 if there is a mutation in gene i, and 1 if there is no mutation. If there are no edges connecting genes i and j, the value of Aij becomes zero. I is an identity matrix, and Ψ is a diagonal matrix meaning the weight of a self-loop in the gene network for each patient. The self-loop weight was calculated using a one-sample t-test on the gene expression data of the patient and control groups. When the self-loop weight was 0, the gene was incapable of affecting itself; conversely, when it was set to 1, the gene remained unaffected by the gene network.

Generation of gene feature vectors using modified PageRank

The modified PageRank algorithm was applied to the gene network of individual patients to produce an impact matrix. The resulting impact matrix rows and columns served as gene feature vectors, and their calculation involved iterating Equation (2):

τ+1 = W̃ × IMτ

where W̃ is a stochastic matrix, which is a matrix with a column sum of 1, calculated by dividing the components of each column of the matrix W by the sum of the corresponding columns. W can be interpreted as the probability that patient gene i affects gene j. IM (impact matrix) is a square matrix of n × n, where n is the number of genes in the patient-specific gene network. The initial value matrix IM0 is a diagonal matrix in which the values of the diagonal components are all 10,000, and the impact matrix at τ + 1 is calculated as the product of the stochastic matrix W and the impact matrix at τ. If Equation (2) is repeated, each column of IM0 is a one-hot vector whose value exists only in the component of the corresponding column index; therefore, the initial value of each column spreads to other components along the patient-specific gene network. Iteration of Equation (2) ends when the impact matrix converges.

The components of the converged impact matrix, denoted as IMij, represent the influence of gene j on gene i for a given patient. Therefore, the ith column of IM represents the impact that gene i has on all genes in the network, denoted as GIV (give impact vector). The ith row represents the impact that gene i receives from all genes in the network, denoted as the RIV (received impact vector). The gene feature vector encompasses both GIV and RIV. Each patient had a different gene network composition. Therefore, for gene i, each patient had a different gene feature vector.

Training a cancer driver gene identification model

We employed a composite model consisting of two autoencoders and one deep neural network to develop an approach for identifying cancer driver genes. In the initial phase of the model, two autoencoders were used to manage GIV and RIV separately. The encoder compresses and represents high-dimensional gene feature vectors as low-dimensional latent vectors. Subsequently, the decoder reconstructs these latent vectors back into their original input data formats. The latent vectors of the GIV and RIV were concatenated and fed into the deep neural network, where the model predicted the probability that the input gene feature vector represented a cancer driver gene. The parameters used in the machine learning model are shown in Table I.

Table I.

Hyper-parameters used for the deep feed-forward network.

Table I.

Hyper-parameters used for the deep feed-forward network.

GroupsParametersValue
CommonEpoch100
Batch size200
OptimizerStochastic gradient descent
Learning rate0.005
Momentum0.9
AutoEncoderSize of input layerNumber of genes
Number of hidden layers3
Size of hidden layers5,000, 1,000, 5,000
Activation functionReLU
Loss functionMean Squared Error
Deep neural networkSize of input layer2,000
Number of hidden layers2
Size of hidden layers500, 100
Size of output layer1
Activation function of hidden layersReLU
Activation function of output layersSigmoid
Loss functionBinary Cross Entropy

To generate the training data, we gathered 360 tumor patient samples and 50 normal samples from TCGA using search term ‘LIHC’. During the preprocessing phase, gene expression level data excluded genes with an FPKM of zero in more than 80% of samples. Genetic mutation data were constructed by integrating somatic and gene copy number mutation data. Each gene was assigned a value of 1 if one or more mutations occurred and 0 otherwise.

After generating gene feature vectors for TCGA LIHC tumor samples, we labeled them using information from known cancer driver genes and genetic mutation data of the samples. Known cancer driver genes were sourced from the IntOGen and CGC databases (28,29). Only Tier 1 genes with substantial evidence of cancer occurrence in the CGC database were used. The list of known cancer driver genes comprised 30 genes from IntOGen and 28 genes from CGC, with eight genes common to both databases. Gene feature vectors corresponding to known cancer driver genes with mutations in individual samples were labeled true, whereas the remaining vectors were labeled false. To construct the training dataset, a falsely labeled gene feature vector was randomly selected for each truly labeled gene feature vector.

Cell lines

Human liver cancer cell lines (Huh-7 and HepG2) were obtained from American Type Culture Collection (Rockville, MD, USA). Huh-7 and Huh-7/sorafenib resistant (Huh-7-SR) cells were maintained in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (WelGENE, Daegu, Korea), 4.5 g/l glucose, L-glutamine, and 100 U/ml penicillin/streptomycin (Invitrogen, Carlsbad, CA) at 37°C with 5% CO2. HepG2, HepG2/sorafenib resistant (HepG2-SR) cells were maintained in minimum essential medium (MEM) with 10% fetal bovine serum and 100 U/ml penicillin/streptomycin. Huh-7 and HepG2 cells were obtained from American Type Culture Collection (ATCC). In order to establish sorafenib resistant cell lines, Huh-7, HepG2 cells were exposed to 1 µM sorafenib at first, and the concentration was gradually increased by 1 µM per month until reaching 6 µM.

RNA extraction, cDNA synthesis, and RT-qPCR

Total RNA was isolated using the TRIzol reagent (Invitrogen). cDNA was amplified using GoScriptTM Reverse Transcriptase (Promega). RT-qPCR was performed using specific primer sequences and SYBR based was conducted with a Light Cycler 480 (Roche Applied Science) in a total volume of 20 µl. The relative expression was analyzed using the 2-ΔΔCq method. The primer sequences for the gene were provided in Table II.

Table II.

Primers sequence used in SYBR-based reverse transcription-quantitative PCR.

Table II.

Primers sequence used in SYBR-based reverse transcription-quantitative PCR.

PrimerForward sequence (5′-3′)Reverse sequence (5′-3′)
Mx1 GGCTGTTTACCAGACTCCGACA CACAAAGCCTGGCAGCTCTCTA
ADAR TCCGTCTCCTGTCCAAAGG TTCTTGCTGGGAGCACTCACAC
MyD88 GAGGCTGAGAAGCCTTTACAGG GCAGATGAAGGCATCGAAACGC
PKR GAAGTGGACCTCTACGCTTTGG TGATGCCATCCCGTAGGTCTGT
IRF1 GAGGTGAAAGACCAGAGCA TAGCATCTCGGCTGGACTTCGA
ACTB CACCATTGGCAATGAGCGGTTC AGGTCTTTGCGGATGTCCACGT
STAT1 ATGGCAGTCTGGCGGCTGAATT CCAAACCAGGCTGGCACAATTG
STAT2 CAGGTCACAGAGTTGCTACAGC CGGTGAACTTGCCAGTCTT
OAS1 AGGAAAGGTGCTTCCGAGGTAG GGACTGAGGAAGACAACCAGGT
OAS2 GCTTCCGACAATCAACAGCCAAG CTTGACGATTTTGTGCCGCTCG
MAP3K TGGCAAGCACTACCTGGATCAG GCAGAGACTGTAGGTAGTTTCGG
BST2 TCTCCTGCAACAAGAGCTGACC TCTCTGCATCCAGGGAAGCCAT
IFI27 CGTCCTCCATAGCAGCCAAGAT ACCCAATGGAGCCCAGGATGAA
MDA5 CCCAAGACACAGAATGAACAAAA CGAGACCATAACGGATAACAATGT

[i] IRF, interferon regulatory factor; Mx1, Myxovirus resistance protein 1; ACTB, Actin Beta; OAS, oligoadenylate synthetase; MAP3K, Mitogen-activated Protein Kinases; IFI27, Interferon Alpha Inducible Protein 27; BST2, Bone Marrow Stromal Cell Antigen 2; MDA5, melanoma differentiation-associated protein 5; ADAR, Adenosine deaminase Acting on RNA; PKR, Protein kinase R; MyD88, Myeloid differentiation primary response 88.

Immunoblotting

Huh-7 cells, Huh-7-SR cells, HepG2, and HepG2-SR cells were collected and lysed with RIPA buffer (20 mM Tris-HCl, 150 mM NaCl, 1% sodium deoxycholate, 1% Triton-X-100 and 0.1% SDS) containing protease inhibitors and phosphate inhibitors. SDS-PAGE (8%) was performed to separate the protein extracts. The proteins were transferred onto nitrocellulose membranes. After blocking the membrane in TBS containing 5% skim milk for 1 h. The antibodies used for immunoblotting were as follows: rabbit monoclonal anti-STAT1 (Cell signaling Technology, Cat#9176S), rabbit monoclonal anti-PY STAT1 (Cell signaling Technology, Cat#9167S), rabbit polyclonal anti-STAT2 (Cell signaling Technology, Cat#4594S), rabbit polyclonal anti-PY STAT2 (Cell signaling Technology, Cat#4441S), rabbit monoclonal IRF9 (Cell signaling Technology, Cat#28492), and horseradish peroxidase-conjugated secondary antibody (1:5,000).

siRNA transfection

Huh-7-SR cells were seeded at 2.5×105 cells per wells into a 6-well plate in DMEM. The following day, Huh-7-SR cells were transfected with siControl (Santa Cruz, Cat#sc-37007), siSTAT1 (Santa Cruz, Cat#sc-44123), siSTAT2 (Santa Cruz, Cat#sc-29492), or siIRF9 (OriGene, Cat#SR323091) at a concentration of 10 nM in 2 ml of serum-free medium containing RNAiMAX (Invitrogen, Cat#13778075). The medium was changed 4 h after transfection, and the cells were harvested after 48 h. Next, the transfected cells were treated with sorafenib for 24 h, followed by MTT assay.

Cell viability assay

For cell viability assays, liver cancer cells (Huh-7, Huh-7-SR, HepG2, HepG2-SR) were seeded into 96 well plates at 1×104 cells per well and incubated with sorafenib at a concentration of 0 to 32 µM for 24 h. After the addition of 10 µl of MTT solution (Abcam, Cat#ab211091), the samples were incubated for 4 h. Subsequently, the medium was removed, and 100 µl of DMSO was added to each well. The optical density was recorded at 590 nm using an enzyme-linked immunosorbent assay reader (MDS Analytical Technologies).

Statistical analysis

Statistical analyses were performed using GraphPad Prism 8 software (GraphPad Software Inc., San Diego, CA, USA). Human liver cancer cells are presented as the mean ± SEM. Unpaired t tests were used for statistical analyses. The significance was set at P<0.05.

Results

Genes associated with sorafenib resistance

Raw RNA-seq data were processed using Cutadapt, FastQC and MultiQC (30). Kallisto was used to determine the abundance of transcripts, which were normalized using TPM (31). Genes with a TPM value of zero in >80% of the samples were excluded from the gene expression data. Gene feature vectors were derived from the processed gene expression data of Huh-7 cells. DNA-seq data were processed using the GATK pipeline v4.1.7.0 (32). Subsequently, the gene feature vectors for each sample were input into a model, and the genes were ranked based on the output of the model. For each cell line, a list of genes was obtained that ranked in the top 50 in at least two of the three samples with sorafenib resistance and outside the top 50 in at least two of the three samples without sorafenib resistance. To eliminate genes with insignificant rank differences between the groups with and without sorafenib resistance, the average rank of each group of genes was compared, and genes with an average rank difference of less than twice were excluded. This led to the identification of 21 Huh-7 genes. By intersecting the results from the Huh-7 cell lines, six common genes were identified.

Unphosphorylated ISGF3 is positively associated with sorafenib resistance in liver cancer cells

Sorafenib-resistant liver cancer cell lines were established as follow: liver cancer cells were exposed to gradually increasing concentrations ranging from 1 to 6 µM of sorafenib (increasing 0.25 µM per cycle) for ~4 months (Fig. 2A). To observe the effects of sorafenib resistance, we treated liver cancer cell lines with increasing concentrations of sorafenib for 24 h using the MTT assay. Sorafenib-resistant liver cancer cell lines were resistant to higher concentrations of sorafenib (Fig. 2B). Compared with liver cancer cells, sorafenib-resistant liver cancer cells showed markedly increased levels of IRF9, but no significant difference in STAT1 and STAT2 (Fig. 2C and D).

Sorafenib resistance increases U-ISG levels

U-ISGs, including OAS1, IFI27, BST2, and MDA5, were increased in sorafenib-resistant liver cancer cells (Fig. 3A). In contrast, ISGF3 complex-dependent ISGs did not increase in Huh-7-SR cells (Fig. 3B) or were slightly induced in HepG2-SR cells (Fig. 3B). Moreover, other U-ISGs were robustly upregulated in the sorafenib-resistant liver cancer cells (Fig. 3C). However, ISGs produced by the phosphorylated ISGF3 complex were minimally increased (Fig. 3C).

U-ISGF3 inhibition re-sensitizes sorafenib-resistant liver cancer cells to sorafenib

To explore the role of the U-ISGF3 complex, we used a transfection method to reduce U-ISGF3 levels in Huh-7-SR cells (Fig. 4A). Downregulation of the U-ISGF3 complex in Huh-7-SR cells reduced U-ISG expression (Fig. 4B). We showed that sorafenib resistance was reduced in Huh-7-SR treated with transfection and in the MTT assay. As expected, the U-ISGF3 complex increased the viability of sorafenib-resistant liver cancer cells (Fig. 4C).

Discussion

Several studies have revealed a critical role of the U-ISGF3 complex in cancer (33). Thus, the inhibition of U-ISGF3 is emerging as an attractive therapeutic strategy for cancer. However, the relationship between the U-ISGF3 complex and sorafenib resistance in HCC remains poorly understood. Here, we confirmed that the U-ISGF3 complex promotes sorafenib resistance and that inhibition of the U-ISGF3 complex reduces sorafenib resistance.

Sorafenib is a multityrosine kinase inhibitor used to treat HCC (34). However, its sensitivity appears in only 30% of the patients, and within 6 months, sorafenib resistance is acquired in HCC (35). The first mechanism occurs when there is no initial response to sorafenib treatment and is mainly associated with altered activation of signaling pathways. In contrast, the second mechanism refers to the development of resistance to sorafenib after following an initial response. Sorafenib resistance targets multiple cellular pathways that contribute to tumor survival and proliferation (36,37). The PI3K/Akt/mTOR signaling pathway is strongly activated by prolonged exposure to sorafenib, leading to the development of resistance (38). Machine learning was used to elucidate the mechanisms underlying sorafenib resistance. We employed a deep neural network to identify candidate genes responsible for sorafenib resistance. However, machine learning methods, including deep neural networks, require a substantial number of samples. Therefore, we trained the model using TCGA data and applied it to the data obtained from Huh-7 cells. While attempting to identify driver genes exhibiting distinct patterns in samples with and without resistance, an inherent limitation arises owing to factors such as batch effects between TCGA and Huh-7 cell data, potentially causing a decrease in accuracy.

Using ML, we determined the significance of STAT1 expression in sorafenib resistance. In this study, we found no differences in STAT1 expression in sorafenib-resistant cell lines, whereas only IRF9 was differentially expressed. The U-ISGF3 complex binds to the promoter regions of U-ISGs, particularly IRF9, which contributes to most of the U-ISG promotor region. This suggests that IRF9 is important for U-ISG expression. Fig. 5 shows that in sorafenib-resistant cell lines, the U-ISGF3 complex translocates into the nucleus to regulate the expression of U-ISGs, leading to the acquisition of resistance. Several studies have suggested that U-ISGs are critical regulators of irradiation or chemotherapy. The knockdown of STAT1, STAT2, and IRF9 significantly enhanced the antitumor activity of sorafenib in vitro.

In conclusion, our results indicate that the U-ISGF3 complex plays a crucial role in mediating sorafenib resistance in liver cancer cells. These results suggest that this mechanism may have clinical relevance and could potentially be applicable to patients. The current study is limited by the lack of patient samples, and future research should be verified using patient samples.

Acknowledgements

Not applicable.

Funding

This work was supported by the Basic Science Research Program of the National Research Foundation of Korea through the Ministry of Science Information and Communication Technology (grant no. NRF-2019R1A2C3005212) and the Research Fund of Seoul St. Mary's Hospital (grant no. 2022-001). This work was also supported by the Basic Science Research Program of the National Research Foundation of Korea through the Ministry of Science Information and Communication Technology (grant no. RS-2024-00337298).

Availability of data and materials

The datasets have been deposited in the Sequenced Read Archive (SRA) (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1117191) under accession no. PRJNA1117191. The rest of data generated in the present study may be requested from the corresponding author.

Authors' contributions

DHS contributed to collecting raw data, analyzing data interpretation, designing the research and writing the draft of the paper and revision of the manuscript. JWP, HWJ, MWK and BYK contributed to collecting raw data and data analysis. DYL and JJL contributed to the study conception and data interpretation. SKY and JWJ interpreted the data and revised the manuscript. PSS designed the research and supervised the study. JGA contributed to the study design, data interpretation, drafting the paper, and manuscript revision. PSS and JGA confirm the authenticity of the raw data. All authors have reviewed and approved the final manuscript.

Ethics approval and consent to participate

Not applicable.

Patient consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Glossary

Abbreviations

Abbreviations:

HCC

hepatocellular carcinoma

U-ISGs

unphosphorylated interferon-stimulated genes

U-ISGF3

unphosphorylated interferon-stimulated gene factor-3

HBV

hepatitis B virus

HCV

hepatitis C virus

MASH

metabolic-associated steatohepatitis

MAFLD

metabolic-associated fatty liver disease

PDGFR-β

platelet-derived growth factor receptor

VEGFR

vascular endothelial growth factor receptors

IFNs

type I interferons

ISGs

interferon-stimulated genes

LIHC

liver hepatocellular carcinoma

References

1 

Sung PS, Park DJ, Roh PR, Mun KD, Cho SW, Lee GW, Jung ES, Lee SH, Jang JW, Bae SH, et al: Intrahepatic inflammatory IgA+PD-L1high monocytes in hepatocellular carcinoma development and immunotherapy. J Immunother Cancer. 10:e0036182022. View Article : Google Scholar : PubMed/NCBI

2 

Mun K, Han J, Roh P, Park J, Kim G, Hur W, Jang J, Choi J, Yoon S, You Y, et al: Isolation and characterization of cancer-associated fibroblasts in the tumor microenvironment of hepatocellular carcinoma. J Liver Cancer. 23:341–349. 2023. View Article : Google Scholar : PubMed/NCBI

3 

Tumen D, Heumann P, Gulow K, Demirci CN, Cosma LS, Muller M and Kandulski A: Pathogenesis and current treatment strategies of hepatocellular carcinoma. Biomedicines. 10:32022022. View Article : Google Scholar : PubMed/NCBI

4 

Alawyia B and Constantinou C: Hepatocellular carcinoma: A narrative review on current knowledge and future prospects. Curr Treat Options Oncol. 24:711–724. 2023. View Article : Google Scholar : PubMed/NCBI

5 

Thandra KC and Barsouk A, Saginala K, Aluru JS, Rawla P and Barsouk A: Epidemiology of non-alcoholic fatty liver disease and risk of hepatocellular carcinoma progression. Clin Exp Hepatol. 6:289–294. 2020. View Article : Google Scholar : PubMed/NCBI

6 

Kim GA, Moon JH and Kim W: Critical appraisal of metabolic dysfunction-associated steatotic liver disease: Implication of Janus-faced modernity. Clin Mol Hepatol. 29:831–843. 2023. View Article : Google Scholar : PubMed/NCBI

7 

Gallego-Durán R, Albillos A, Ampuero J, Arechederra M, Bañares R, Blas-García A, Berná G, Caparrós E, Delgado TC, Falcón-Pérez JM, et al: Metabolic-associated fatty liver disease: From simple steatosis toward liver cirrhosis and potential complications. Proceedings of the third translational hepatology meeting, organized by the Spanish association for the study of the liver (AEEH). Gastroenterol Hepatol. 45:724–734. 2022.(In English, Spanish). View Article : Google Scholar : PubMed/NCBI

8 

Angeli-Pahim I, Chambers A, Duarte S and Zarrinpar A: Current trends in surgical management of hepatocellular carcinoma. Cancers (Basel). 15:53782023. View Article : Google Scholar : PubMed/NCBI

9 

Yoon JH and Choi SK: Management of early-stage hepatocellular carcinoma: Challenges and strategies for optimal outcomes. J Liver Cancer. 23:300–315. 2023. View Article : Google Scholar : PubMed/NCBI

10 

Jost-Brinkmann F, Demir M, Wree A, Luedde T, Loosen SH, Müller T, Tacke F, Roderburg C and Mohr R: Atezolizumab plus bevacizumab in unresectable hepatocellular carcinoma: Results from a German real-world cohort. Aliment Pharmacol Ther. 57:1313–1325. 2023. View Article : Google Scholar : PubMed/NCBI

11 

Sung PS: Crosstalk between tumor-associated macrophages and neighboring cells in hepatocellular carcinoma. Clin Mol Hepatol. 28:333–350. 2022. View Article : Google Scholar : PubMed/NCBI

12 

Zhang Y, Li G, Liu X, Song Y, Xie J, Li G, Ren J, Wang H, Mou J, Dai J, et al: Sorafenib inhibited cell growth through the MEK/ERK signaling pathway in acute promyelocytic leukemia cells. Oncol Lett. 15:5620–5626. 2018.PubMed/NCBI

13 

Habiba YH, Omran GA, Helmy MW and Houssen ME: Antitumor effects of rhamnazinon sorafenib-treated human hepatocellular carcinoma cell lines via modulation of VEGF signaling and PI3K/NF-κB p38/caspase-3 axes cross talk. Life Sci. 297:1204432022. View Article : Google Scholar : PubMed/NCBI

14 

Li J, Xuan S, Dong P, Xiang Z, Gao C, Li M, Huang L and Wu J: Immunotherapy of hepatocellular carcinoma: Recent progress and new strategy. Front Immunol. 14:11925062023. View Article : Google Scholar : PubMed/NCBI

15 

Tian X, Yan T, Liu F, Liu Q, Zhao J, Xiong H and Jiang S: Link of sorafenib resistance with the tumor microenvironment in hepatocellular carcinoma: Mechanistic insights. Front Pharmacol. 13:9910522022. View Article : Google Scholar : PubMed/NCBI

16 

Hiebinger F, Kudulyte A, Chi H, Burbano De Lara S, Ilic D, Helm B, Welsch H, Dao Thi VL, Klingmüller U and Binder M: Tumour cells can escape antiproliferative pressure by interferon-β through immunoediting of interferon receptor expression. Cancer Cell Int. 23:3152023. View Article : Google Scholar : PubMed/NCBI

17 

Blaszczyk K, Nowicka H, Kostyrko K, Antonczyk A, Wesoly J and Bluyssen HAR: The unique role of STAT2 in constitutive and IFN-induced transcription and antiviral responses. Cytokine Growth Factor Rev. 29:71–81. 2016. View Article : Google Scholar : PubMed/NCBI

18 

Au-Yeung N, Mandhana R and Horvath CM: Transcriptional regulation by STAT1 and STAT2 in the interferon JAK-STAT pathway. JAKSTAT. 2:e239312013.PubMed/NCBI

19 

Platanitis E, Demiroz D, Schneller A, Fischer K, Capelle C, Hartl M, Gossenreiter T, Muller M, Novatchkova M and Decker T: A molecular switch from STAT2-IRF9 to ISGF3 underlies interferon-induced gene transcription. Nat Commun. 10:29212019. View Article : Google Scholar : PubMed/NCBI

20 

Lee CJ, An HJ, Cho ES, Kang HC, Lee JY, Lee HS and Cho YY: Stat2 stability regulation: An intersection between immunity and carcinogenesis. Exp Mol Med. 52:1526–1536. 2020. View Article : Google Scholar : PubMed/NCBI

21 

Sung PS, Cheon H, Cho CH, Hong SH, Park DY, Seo HI, Park SH, Yoon SK, Stark GR and Shin EC: Roles of unphosphorylated ISGF3 in HCV infection and interferon responsiveness. Proc Natl Acad Sci USA. 112:10443–10448. 2015. View Article : Google Scholar : PubMed/NCBI

22 

Cheon H, Holvey-Bates EG, Schoggins JW, Forster S, Hertzog P, Imanaka N, Rice CM, Jackson MW, Junk DJ and Stark GR: IFNβ-dependent increases in STAT1, STAT2, and IRF9 mediate resistance to viruses and DNA damage. EMBO J. 32:2751–2763. 2013. View Article : Google Scholar : PubMed/NCBI

23 

Cheon H, Wang YX, Wightman SM, Jackson MW and Stark GR: How cancer cells make and respond to interferon-I. Trends Cancer. 9:83–92. 2023. View Article : Google Scholar : PubMed/NCBI

24 

Jung H, Choi J, Park J and Ahn J: A novel machine learning model for identifying patient-specific cancer driver genes. IEEE Access. 10:54245–54253. 2022. View Article : Google Scholar

25 

Tomczak K, Czerwińska P and Wiznerowicz M: The cancer genome atlas (TCGA): An immeasurable source of knowledge. Contemp Oncol (Pozn). 19:A68–A77. 2015.PubMed/NCBI

26 

Croft D, O'Kelly G, Wu G, Haw R, Gillespie M, Matthews L, Caudy M, Garapati P, Gopinath G, Jassal B, et al: Reactome: A database of reactions, pathways and biological processes. Nucleic Acids Res. 39:(Database Issue). D691–D697. 2011. View Article : Google Scholar : PubMed/NCBI

27 

Liu ZP, Wu C, Miao H and Wu H: RegNetwork: An integrated database of transcriptional and post-transcriptional regulatory networks in human and mouse. Database (Oxford). 2015:bav0952015. View Article : Google Scholar : PubMed/NCBI

28 

Sondka Z, Bamford S, Cole CG, Ward SA, Dunham I and Forbes SA: The COSMIC cancer gene census: Describing genetic dysfunction across all human cancers. Nat Rev Cancer. 18:696–705. 2018. View Article : Google Scholar : PubMed/NCBI

29 

Gundem G, Perez-Llamas C, Jene-Sanz A, Kedzierska A, Islam A, Deu-Pons J, Furney SJ and Lopez-Bigas N: IntOGen: Integration and data mining of multidimensional oncogenomic data. Nat Methods. 7:92–93. 2010. View Article : Google Scholar : PubMed/NCBI

30 

Ewels P, Magnusson M, Lundin S and Käller M: MultiQC: Summarize analysis results for multiple tools and samples in a single report. Bioinformatics. 32:3047–3048. 2016. View Article : Google Scholar : PubMed/NCBI

31 

Bray NL, Pimentel H, Melsted P and Pachter L: Near-optimal probabilistic RNA-seq quantification. Nat Biotechnol. 34:525–527. 2016. View Article : Google Scholar : PubMed/NCBI

32 

Marx V: Genomics in the clouds. Nat Methods. 10:941–945. 2013. View Article : Google Scholar : PubMed/NCBI

33 

Wang W, Yin Y, Xu L, Su J, Huang F, Wang Y, Boor PPC, Chen K, Wang W, Cao W, et al: Unphosphorylated ISGF3 drives constitutive expression of interferon-stimulated genes to protect against viral infections. Sci Signal. 10:eaah42482017. View Article : Google Scholar : PubMed/NCBI

34 

Zhou W, Lou W, Chen J, Ding B, Chen B, Xie H, Zhou L, Zheng S and Jiang D: AG-1024 sensitizes sorafenib-resistant hepatocellular carcinoma cells to sorafenib via enhancing G1/S arrest. Onco Targets Ther. 14:1049–1059. 2021. View Article : Google Scholar : PubMed/NCBI

35 

Guo L, Hu C, Yao M and Han G: Mechanism of sorafenib resistance associated with ferroptosis in HCC. Front Pharmacol. 14:12074962023. View Article : Google Scholar : PubMed/NCBI

36 

Zhai B and Sun XY: Mechanisms of resistance to sorafenib and the corresponding strategies in hepatocellular carcinoma. World J Hepatol. 5:345–352. 2013. View Article : Google Scholar : PubMed/NCBI

37 

Xia S, Pan Y, Liang Y, Xu J and Cai X: The microenvironmental and metabolic aspects of sorafenib resistance in hepatocellular carcinoma. EBioMedicine. 51:1026102020. View Article : Google Scholar : PubMed/NCBI

38 

Sun T, Liu H and Ming L: Multiple roles of autophagy in the sorafenib resistance of hepatocellular carcinoma. Cell Physiol Biochem. 44:716–727. 2017. View Article : Google Scholar : PubMed/NCBI

Related Articles

Journal Cover

September-2024
Volume 28 Issue 3

Print ISSN: 1792-1074
Online ISSN:1792-1082

Sign up for eToc alerts

Recommend to Library

Copy and paste a formatted citation
x
Spandidos Publications style
Seo DH, Park JW, Jung HW, Kang MW, Kang BY, Lee DY, Lee JJ, Yoon SK, Jang JW, Ahn JG, Ahn JG, et al: Machine learning model reveals roles of interferon‑stimulated genes in sorafenib‑resistant liver cancer. Oncol Lett 28: 438, 2024
APA
Seo, D.H., Park, J.W., Jung, H.W., Kang, M.W., Kang, B.Y., Lee, D.Y. ... Sung, P.S. (2024). Machine learning model reveals roles of interferon‑stimulated genes in sorafenib‑resistant liver cancer. Oncology Letters, 28, 438. https://doi.org/10.3892/ol.2024.14571
MLA
Seo, D. H., Park, J. W., Jung, H. W., Kang, M. W., Kang, B. Y., Lee, D. Y., Lee, J. J., Yoon, S. K., Jang, J. W., Ahn, J. G., Sung, P. S."Machine learning model reveals roles of interferon‑stimulated genes in sorafenib‑resistant liver cancer". Oncology Letters 28.3 (2024): 438.
Chicago
Seo, D. H., Park, J. W., Jung, H. W., Kang, M. W., Kang, B. Y., Lee, D. Y., Lee, J. J., Yoon, S. K., Jang, J. W., Ahn, J. G., Sung, P. S."Machine learning model reveals roles of interferon‑stimulated genes in sorafenib‑resistant liver cancer". Oncology Letters 28, no. 3 (2024): 438. https://doi.org/10.3892/ol.2024.14571