To conduct our analysis, we obtained the gene expression, mutation data, and clinical information from the TCGA-LIHC project via the UCSC Xena platform (http://xena.ucsc.edu). We included a total of 377 patients, and for the follow-up analysis, we also selected 50 normal individuals. Additionally, we acquired a single-cell RNA sequencing dataset (accession number: GSE149614) consisting of 10 samples and 34,414 cells from the GEO database to validate the immune characteristics of hepatocellular carcinoma.

For immune infiltration analysis, we employed the ESTIMATE algorithm (16) to evaluate the ratio of stromal and immune cells in the tumor microenvironment as well as tumor purity. Furthermore, we utilized EPIC algorithms (17) to compute the proportions of immune cells in each tumor microenvironment. Additionally, we downloaded the activation levels of the seven-step anti-tumor immune cycle from the TIP database (18).

To subtype the liver cancer patients and perform differential gene analysis, we employed a consensus clustering algorithm based on the expression of cuproptosis-related genes. This approach classified the patients into two subtypes: Group A (high expression of DLAT, GLS, and CDKN2A genes) and Group B (high expression of PDHA1, FDX1, MTF1, LIAS, and LIPT1 genes). We repeated the hierarchical clustering 500 times, each time selecting 80% of the samples. Finally, we determined the optimal number of clusters based on the cumulative distribution function and selected stable clustering results.

To screen for differentially expressed genes in both isoforms, we used the R package DESeq2 (19) with a significance threshold of |log2FoldChange| > 1 and an adjusted P-value < 0.05. Subsequently, the differential genes were subjected to gene ontology (GO) functional enrichment analysis using the R package ClusterProfiler (20). We also performed functional analysis based on the KEGG database using the GSEA algorithm. The ssGSEA (21) method was employed to measure the activity of cancer markers in tumor samples, and the enrichment scoring method was set as “singscore.” Furthermore, we analyzed the difference in marker activity between the two subtypes using the Wilcox test.

We conducted co-expression network analysis of the differential genes using the WGCNA package in R. For this analysis, we set 4 as the soft threshold. Through clustering analysis, we identified highly similar modules. We then assessed the correlation between these modules and various factors such as stromal score, immune score, tumor purity, and different immune cell ratios. To further investigate the role of methylation, we examined the methylation patterns in the UALCAN database (22). This allowed us to identify key genes associated with methylation. Specifically, we focused on genes that exhibited higher methylation levels in the tumor group compared to the normal group.

To ascertain the predictive capabilities of our model, we randomly divided the TCGA dataset into a training and test set. In the training set, we utilized the glmnet R package to construct a LASSO Cox regression model, employing the key genes identified earlier. This model was then evaluated in the test set to assess its validity. Next, we employed the median risk score as a threshold to partition the patients into high and low-risk groups. The Kaplan-Meier method was employed to generate survival curves for these groups, and the log-rank test was utilized to determine any significant differences in survival. Additionally, we calculated the time-dependent AUC by utilizing the R package timeROC (23) to evaluate the accuracy of our predictions. To further visualize the prognostic value of our model, we employed the R package rms to create a nomogram that incorporates pooled clinical variables. We employed calibration curves to assess the concordance between the predicted and observed risks indicated by the nomogram. Moreover, decision curve analysis was performed to measure the net benefit (NB) of utilizing the nomogram as a screening tool to determine if interventions are warranted for truly high-risk patients.

The R package maftools was used to analyze the mutation data obtained from patients with liver cancer. By employing waterfall plots, we visualized the mutation profile of the patients. Additionally, we calculated the tumor mutation load (TMB) by determining the total number of mutations present in each patient. To assess the mutation differences between the two groups of patients, we conducted the Fisher test.

The Seurat (24) package was used to perform unsupervised clustering of individual cells, and we performed principal component analysis using the top 2000 high variance features in the dataset and downstream analysis using the top 20 principal components. Cell subclusters were identified using the “FindClusters” function (resolution = 0.6) and visualized using Uniform Manifold Approximation and Projection (UMAP).

The irGSEA R package is used to calculate cuproptosis associated gene scores, obtained using the singscore algorithm. The scores of all cells in the samples were averaged, and all samples were divided into groups with high expression and low expression of cuproptosis related genes in a ratio of 1:1.

The R package Cellchat (25) was used to analyze cell-cell communication in single cell data and to compare the differences in cell-cell communication between the high and low expression groups of hepatocytes.

Human liver cancer cells HepG2 and SMMC7721 were procured from the Chinese Cell Bank (Shanghai, China). They were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco, NY, USA) supplemented with 10% fetal bovine serum and 100 U/ml penicillin-streptomycin, maintained at 37°C with 5% CO₂. siRNAs targeting DLAT and CDKN2A, along with their respective negative control (NC) sequences, were obtained from RiboBio Co., Ltd. (Guangzhou, China). Cell transfections were carried out using Lipofectamine 3000 reagents (Invitrogen, Grand Island, NY, USA) in accordance with the manufacturer’s guidelines. After a 48-hour incubation period with Lipofectamine 3000, the HepG2 and SMMC7721 cells were harvested and utilized for the pertinent experiments. The sequences of siRNAs targeting DLAT were as follows: 5’-CAGTGAATTGTCTTTTAGACAAC-3’; siCDKN2A: 5’-GAUGCUUCGUCUACGAGAATT-3’; si-NC: Ribobio, # siN0000001-1.

The HepG2 and SMMC7721 cells were harvested and rinsed with ice-cold PBS solution for the Western blot assay. Subsequently, cell lysis was carried out in RIPA cell lysis buffer (Solaibao, Beijing, China), supplemented with complete™ Protease Inhibitor Cocktail. Protein concentrations in the resulting supernatant were determined using the bicinchoninic acid (BCA) Assay (Solaibao, Beijing, China). The protein extracts were resolved on 10% SDS-PAGE gels and subsequently transferred onto PVDF membranes (Immobilon-P, CA, USA). After blocking with 5% skimmed milk, these membranes were subjected to overnight incubation at 4°C with DLAT (Cat#: 12362, Cell Signaling Technology, USA) or CDKN2A (Cat#: 18769, Cell Signaling Technology, USA) antibodies. Following three washes in TBST, the membranes were exposed to the secondary antibody at room temperature for 1 hour. Signal detection was accomplished using an enhanced chemiluminescence (ECL) kit (Immobilon-P, CA, USA), and band visualization was facilitated by the ChemiDoc™ Touch Imaging System (Bio-Rad). Relative quantification was performed by utilizing β-actin as an internal reference for normalization.

Cell proliferation capacity was assessed using the Cell Counting Kit‐8 purchased from Dojindo Laboratories (Japan). Initially, cells were plated at a density of 3×10³ cells/well in 100 μL of medium within 96‐well microplates (Corning, NY, USA). Following 24 hours of incubation, 10 μL of CCK‐8 reagent was added into each well and allowed to incubate for 1.5 h. Subsequently, the absorbance was measured at 450 nm using a microplate reader (Molecular Devices, Sunnyvale, USA). Cellular proliferation was quantified based on the absorbance readings.

The HepG2 and SMMC7721 cells were collected, resuspended in complete medium, and then inoculated into the wells at a density of 2000 cells per dish. The culture medium was changed periodically until visible colonies formed. Finally, the colonies were fixed with methanol and stained with a crystal violet solution (Beyotime Biotechnology, China). This experiment was repeated three times.

Transwell assays were conducted utilizing 8-μm pore transwell compartments (Cat. No. PTEP24H48, Millipore, Billerica, USA). In the migration assay, 200 µL of HepG2 and SMMC7721 cells (1×10⁵) were placed in the upper chamber with serum-free medium, while the lower chamber held 800 µL of DMEM supplemented with 10% FBS. For invasion assay, Matrigel (BD Biosciences, USA) was diluted in serum-free medium at a ratio of 1:19, then added to each well, and 1×10⁵ cells were suspended in the upper compartment. The subsequent steps mirrored those of the migration assay. Following incubation at 37°C for 24 h, the medium was discarded, and cells that had invaded the lower surface of the membrane were fixed in methanol for 30 minutes. Staining was performed using crystal violet (Beyotime Biotechnology, China) for 30 minutes at room temperature. The migrated and invaded cells were counted in 5 separate visual fields at ×20 magnification using an Olympus microscope.

The SMMC7721 and HepG2 cells were seeded in 6-well plates at a density of 2 × 10⁵ cells/well and treated using Elesclomol 200 nM for 24 h. The total RNA was extracted using Trizol reagent (Invitrogen, Carlsbad, CA). The cDNA synthesis was performed using RevertAid First Strand cDNA Synthesis kit (Thermo Fisher Scientific). The PowerUp SYBR Green Master Mix (Applied Biosystems) was used for qRT-PCR on an FTC-3000p Real-Time PCR System (Funglyn Biotech, China). The relative gene expression was determined using the 2^-ΔΔCT method with actin as the reference gene. The primer sequences are for BCAT1: forward, 5′-GTGGAGTGGTCCTCAGAGTTT-3′ and reverse, 5′-AGCCAGGGTGCAATGACAG-3; for TLR8: forward, 5′-ATGTTCCTTCAGTCGTCAATGC-3′ and reverse, 5′-TTGCTGCACTCTGCAATAACT-3; for IL4I1: forward, 5′-TGATGTCCGAGGATGGCTTCT-3′ and reverse, 5′-TGTACTGGAGTCTGTCGCTGA-3; for CASP5: forward, 5′-TCACCTGCCTGCAAGGAATG-3′ and reverse, 5′-TCTTTTCGTCAACCACAGTGTAG-3; for FCRL5: forward, 5′-ATGTTCCTTCAGTCGTCAATGC-3′ and reverse, 5′-CCTCAAGGATATTGTCTGGGGTT-3; for CCR8: forward, 5′-CTGTCTGACCTGCTTTTTGTCT-3′ and reverse, 5′-CCACTTTGCACATTACAGTCCC-3; for PDCD1: forward, 5′-CCAGGATGGTTCTTAGACTCCC-3′ and reverse, 5′-TTTAGCACGAAGCTCTCCGAT-3; for IFI30: forward, 5′-CCCCTCTGCAAGCGTTAGAC-3′ and reverse, 5′-CCCGCAGGTATAGATTGCCT-3; for MMP12: forward, 5′-GATCCAAAGGCCGTAATGTTCC-3′ and reverse, 5′-TGAATGCCACGTATGTCATCAG-3; for Actin: forward, 5′-TCCCTGGAGAAGAGCTACGA-3′ and reverse, 5′-TACAGGTCTTTGCGGATGTC-3; The cells without Elesclomol treatment is selected as control.

The above statistics and analysis were performed using R (v 4.1.3) software. Figures were stitched together by Adobe Illustrator software. Comparative analysis of differences in box plots was performed using Wilcoxon rank sum test. Spielman’s coefficient was used for correlation analysis. The chi-square test (Fisher exact test if necessary) was used for comparison of clinical characteristics between the two groups. Multifactorial logistic regression analysis was used to assess the clinical characteristics affecting clustering. Kaplan Meier method was used to plot survival curves. Cox analysis was used to assess the characteristics associated with survival. All hypothesis tests were two-sided, and p-values for multiple tests were corrected by the FDR method, and corrected P-values <0.05 were considered significant.

To investigate the relationship between cuproptosis and hepatocellular carcinoma (HCC), we analyzed the differential profile of 10 cuproptosis-related genes (9) in tumor versus normal tissues. Our findings revealed that these cuproptosis-related genes were expressed at higher levels in tumor tissues compared to normal tissues ( Figure 1A ). Subsequently, we categorized liver cancer patients into two groups based on the median expression of these genes, separating them into high and low expression groups. By analyzing the prognostic differences between these groups, we observed that patients with low expression of DLAT and CDKN2A had significantly better prognosis compared to those with high expression ( Figure 1B ). These results indicated a close relationship between hepatocellular carcinoma and the expression levels of cuproptosis-related genes.

To further explore the connection between cuproptosis-related genes and the tumor immune microenvironment, we employed the ESTIMATE algorithm. The analysis demonstrated significant positive correlations of PDHB and LIAS with tumor purity, accompanied by negative correlations with immune score and interstitial score ( Figure 1C ). Additionally, we employed the EPIC method to assess immune cell infiltration associated with cuproptosis. The results indicated positive correlations between cuproptosis-related genes and various non-immune cells, particularly fibroblasts ( Supplementary Figure S1A ) and CD4⁺ T cells. Among macrophages, only FDX1 displayed a positive correlation ( Figure 1D , Supplementary Figure S1B ), while the remaining nine cuproptosis-related genes exhibited significant negative correlations. As an example, we can consider the gene GLS ( Figure 1D , Supplementary Figure S1C ). Moreover, we performed an analysis of the anti-cancer immune status and tumor-infiltrating immune cells, investigating the seven-step cancer-immunity cycle using the TIP method specifically for cuproptosis-related genes. We observed that the majority of genes were expressed in the initial step of the cancer-immunity cycle, which involves the release of antigens from cancer cells. Specifically, LIPT1 and LIAS displayed negative correlations with the recruitment of various immune cells ( Figure 1E ), whereas GLS was involved in multiple steps of the cycle. These findings indicate a strong association between the regulation of cuproptosis and the tumor immune microenvironment in hepatocellular carcinoma.

Subsequently, we conducted an analysis of cuproptosis subtypes in hepatocellular carcinoma based on the expression of cuproptosis-related genes, which exhibited significant correlations with each other ( Supplementary Figure S1D ). Employing consensus clustering, we grouped the hepatocellular carcinoma patients based on the expression patterns of cuproptosis-related genes. With a chosen consistency matrix k value of 2 ( Figure 1F ), the patients were divided into two distinct groups, indicating different cuproptosis-related gene expression profiles. Notably, patients in group B demonstrated a longer overall survival compared to those in group A (P=0.00053, Figure 1G ). Specifically, genes DLAT, GLS, and CDKN2A were significantly overexpressed in group A, while genes PDHA1, FDX1, MTF1, LIAS, and LIPT1 were significantly overexpressed in group B ( Figure 1H ). Furthermore, we identified a significant correlation between the cuproptosis subtypes and the immune subtypes (26) (P=9.12e-07). However, the correlation with the TCGA classification (P=0.1071) was not statistically significant ( Figure 1H , Supplementary Table 1 ), suggesting that the immune response subtype was more prevalent among patients in group B.

We conducted further analysis to examine the variations in the immune microenvironment between the two subtypes. The findings revealed higher levels of proliferation, wound healing, macrophage regulation, and lymphocyte infiltration in group A. Particularly, there was a notable elevation in the transforming growth factor-β response in group A, indicating a strong immunosuppressive effect ( Supplementary Figure S1E ). Additionally, group A exhibited higher levels of macrophage regulation, primarily observed in M0 macrophages. On the other hand, the functions of group B were more related to communication with non-immune cells and cell maintenance ( Supplementary Figures S1F , S2E , S3 ).

These results indicate that the upregulation of cuproptosis genes, such as DLAT, GLS, and CDKN2A, exerted a suppressive effect on the tumor immune microenvironment. Consequently, patients in group A had relatively poorer prognoses, while those in group B demonstrated comparatively better outcomes.

To investigate the variations in biological pathways among the different subtypes, differential expression analysis was initially conducted. At a threshold of ploidy change greater than 2, and after correcting for p-values less than 0.05, a total of 381 genes were up-regulated, while 857 genes were down-regulated in group B compared to group A ( Figure 2A , Supplementary Table 2 ). GSEA analysis revealed that PPAR signaling pathway, citrate cycle (TCA cycle), cytokine-cytokine receptor interaction, JAK-STAT, and cell cycle were associated with regulation in the subtypes. Specifically, genes involved in the PPAR signaling pathway and citrate cycle (TCA cycle) were predominantly up-regulated, whereas genes in other pathways were mainly down-regulated ( Figure 2B ). Gene function enrichment analysis showed that the upregulated genes were primarily enriched in the metabolism of metal ions and fatty acids ( Figure 2C ). On the other hand, the downregulated genes were mainly associated with pathways determining cell fate ( Figure 2D ). Furthermore, a hallmark differential analysis confirmed that subtype A exhibited stronger activity in Wnt, TGF_β, Notch, and other signaling pathways. Conversely, subtype B demonstrated increased levels of metabolic responses, specifically involving KRAS and fatty acids ( Figure 2E ).

To examine the correlation among differentially expressed genes, we utilized the Weighted Gene Co-expression Network Analysis (WGCNA) on TCGA-LIHC dataset, employing a soft threshold (β=4) to construct co-expression modules ( Figure 3A ). The dynamic tree-cut method identified a total of 14 modules ( Figure 3B , Supplementary Table 3 ), and these modules were depicted to reduce dimensionality ( Figure 3C ). By analyzing module properties through a heatmap, Module 10 was selected as the key module based on immunoscore, immune-cell proportion, and CD8⁺ T-cell infiltration ( Figure 3D ). In Module 10, a total of 31 key genes were identified by integrating module membership and gene significance ( Figure 3E , Supplementary Table 3 ). This selection proved beneficial for a more in-depth exploration of the relationship between cuproptosis and the tumor immune microenvironment.

We first showed the mutation profiles of the two subtypes, where the most severely mutated genes included TP53, TTN and MUC16 ( Figure 4A , Supplementary Figure S4 ). Then, we compared the differences in mutation frequencies between the two subtype samples. The results showed that most of the mutation frequencies in group A were significantly higher than those in group B, such as TP53, ADAM18, CABIN1 and RB1, while only the mutation frequency of CTNNB1 gene was significantly higher in group B than in group A ( Figure 4B ). The large number of mutations in oncogenes may be closely related to cuproptosis and one of the reasons for the poor prognosis of patients in group A. We then compared the immunogenicity differences between the two subtypes, and we found that subtype A samples had higher number of segments, homologous recombination defects and TCR abundance, indicating that subtype A patients are more suitable for immunotherapy ( Figure 4C ).

Subsequently, a scoring model was developed to assist in predicting the prognosis of hepatocellular carcinoma patients based on the identified key genes associated with cuproptosis and immune response. The liver cancer samples were randomly divided into training and validation sets. Using the LASSO Cox model, we determined a risk scoring system comprising nine genes (BCAT1, TLR8, IL4I1, CASP5, FCRL5, CCR8, PDCD1, IFI30, MMP12) by constraining the training set ( Figures 5A, B ) with the coefficients presented in Figure 5B . Furthermore, considering the close relationship between tumorigenesis and methylation (27), we validated the methylation patterns in the UALCAN database ( Supplementary Figure S5 ). We then evaluated the prognostic impact of the risk score in the training set by classifying patients into two groups based on the median risk score. The results demonstrated that patients with lower risk scores generally exhibited better survival outcomes compared to those with higher risk scores (P=0.00038, log-rank test; Figures 5C, D ). Moreover, the risk score displayed excellent predictive accuracy at different follow-up times, as indicated by ROC analysis ( Figure 5E ).

To verify that the risk scores retained similar prognostic value across different populations, we applied them to the validation set. Notably, the distribution of risk scores and survival status varied in the validation set ( Figure 5F ). A significant difference in survival time was observed between the high-risk and low-risk groups (P=0.0078, log-rank test; Figure 5G ). The risk score also exhibited outstanding predictive accuracy when assessed in an independently validated cohort ( Figure 5H ).

In comparison with the normal group, the genes expression of BCAT1, IL4I1, CASP5, CCR8, PDCD1, IFI30, and MMP12 were significantly increased in liver cancer patients in the UCSC Xena database ( Figure 6A ). We confirmed the nine cuproptosis-related genes using RT-qPCR in order to verify our results. The HepG2 and SMMC7721 cells were seeded treated using Elesclomol 200 nM for 24 h ( Figures 6B, C ). In comparison with the control group, the expression of BCAT1, TLR8, IL4I1, CASP5, FCRL5, CCR8, PDCD1, IFI30, and MMP12 were significantly down-regulated in Elesclomol treatment group ( Figures 6B, C ).

We downloaded the standardized expression data and clinical information of pancancer from the UCSC Xena database, extracted the information of liver Cancer patients, divided them into Normal and cancer according to the characteristics of ID, and analyzed the differential expression of target genes. According to the median expression of a gene, the patients were divided into high expression group and low expression group, and the survival analysis of the gene in the tumor was performed. According to the median expression of a gene, the patients were divided into high expression group and low expression group, and the survival analysis of the gene in the tumor was performed. The survival analysis curves of these nine genes are shown in Figure 6D .

A prognostic nomogram was developed using a multifactorial Cox regression model ( Figure 7A ). This model integrated the risk score along with two independent predictors, namely age and grading, to better predict the prognosis of patients with hepatocellular carcinoma. The performance of the nomogram was evaluated in both the training set and the independent validation set, demonstrating its effectiveness in predicting patient survival ( Figures 7B, C ). To assess the clinical benefit of the nomogram model, decision curve analysis was conducted on both the training and independent validation sets. The analysis, using 4-year survival as the endpoint, compared the net clinical benefit of the nomogram with several competing intervention strategies, such as intervention for all, intervention for none, and intervention based on different clinical indicators ( Figures 7D, E ). The decision curve analysis revealed that the nomogram provided greater net clinical benefit compared to the other strategies, indicating its potential as a valuable tool for clinical decision-making.

To explore the impact of cuproptosis on the microenvironment, we analyzed the GSE149614 scRNA-seq dataset for primary hepatocellular carcinoma. Figure 8A shows the patient origin of each sample, which was further classified into 23 cell types ( Figure 8B , Supplementary Table 5 ). We show the markers of clusters ( Figure 8C , Supplementary Figure S6 ) and the distribution and proportion of immune and non-immune cells in them ( Supplementary Figures S7A, B ). We examined the expression of cuproptosis-related genes within each cell cluster and found that these genes were predominantly expressed at higher levels in non-immune cells ( Supplementary Figure S7C ). To assess the variability in gene expression, we used the singscore method and observed distinct expression patterns among samples and clusters ( Figure 8D , Supplementary Table 6 ). Consequently, we divided the samples into two groups based on high and low expression of cuproptosis-related genes, equally distributed ( Supplementary Figure S7D ). The expression and distribution of cuproptosis-related genes were then analyzed across different cell subpopulations ( Figure 8E ). Furthermore, we compared the proportions of immune cells in the high and low cuproptosis-related gene expression groups. Notably, T_memory and Macrophage_FCN1⁺ were significantly lower in the high-expression group compared to the low-expression group ( Figure 8F , Supplementary Table 7 ). Previous studies have highlighted the role of FCN⁺ macrophages in recognizing and eliminating tumor cells by detecting abnormal glycosylation patterns on their surface (28, 29). Additionally, the infiltration of T memory cells within tumors allows for quicker and stronger immune responses (30). These findings suggest that a low-score cuproptosis environment can activate the immune system, promote inflammation, and facilitate the recruitment of immune cells, thereby enhancing the ability to attack tumor cells.

Based on the expression levels of cuproptosis-related genes, we categorized the T_memory and Macrophage_FCN1⁺ cells into high and low expression groups ( Figures 9A, B ). Our investigation focused on understanding the differences in cell-cell communication and signaling between these groups, with a goal of identifying the key factors contributing to phenotypic variation. Notably, we found no significant differences in the number of interactions and interaction weights/strength between Macrophage_FCN1⁺ cells, regardless of whether they acted as signal senders or receivers ( Supplementary Figures S8A–D ). Similar results were observed when T_memory cells were the signal receivers ( Supplementary Figures S8E, F ). However, we observed distinct patterns when T_memory cells were the signal senders. Specifically, T_memory_low cells emitted a higher number of signals compared to T_memory_high cells. Additionally, in terms of signal intensity, T_memory_low cells generally exhibited stronger signaling towards non-immune cells, while T_memory_high cells displayed stronger signaling towards immune cells ( Figures 9C, D ).

In our further analysis, we investigated the signaling pathways involved in T_memory_high and T_memory_low cells. Notably, T_memory_high cells displayed stronger MIF signals towards almost all other cells, while T_memory_low cells predominantly sent SPP1 signals ( Figure 9E , Supplementary Figure S9A ). Furthermore, we identified receptor-ligand pairs that played a pivotal role in these signaling pathways. T_memory_low cells communicated SPP1/CD44 signals to all other immune cells, indicating a high degree of immune infiltration. Additionally, non-immune cells, such as SPP1-(ITGAV+ITGB5), received SPP1 signals. The IFNG pathway exhibited significant variation, primarily observed in macrophages and a subset of hepatocytes, with the main receptor-ligand pair being INFG-(IFNGR1+IFNGR2). Macrophages also released CCL signals. On the other hand, the MIF signaling pathway was predominantly enriched in T_memory_high cells and relied on MIF-(CD74+CXCR4) and MIF-(CD74+CD44) interactions. This signaling cascade was responsible for communication with T cells, B cells, and macrophages ( Figure 9F , Supplementary Figure S9B ).

We knocked down DLAT and CDKN2A expression by using siRNAs in HepG2 and SMMC7721 cells, which was verified by western blot ( Figures 10A, B ). CCK-8 and colony formation assays were used to evaluate the proliferation of HepG2 and SMMC7721 cells. We found that DLAT and CDKN2A knockdown significantly inhibited cell proliferation ( Figures 10C, D ). Cells of different treatment groups were inoculated into 6-wellplates. The plates were seeded with 2000 cells (HepG2) or 2500 cells (SMMC7721). The cells were fixed after 14 days of culture, stained with crystal violet and then counted. The results showed that the colony formation rate was significantly lower in the DLAT and CDKN2A knockdown group than in the control groups ( Figures 10E, F ). Our studies indicated that inhibition of DLAT and CDKN2A expression significantly reduced the proliferation of liver cancer cells. The effect of knocked down DLAT and CDKN2A on the in vitro tumor cell migration and invasion was investigated using the Transwell migration and Matrigel invasion assay, respectively. Under light microscope, the crystal violet-stained HepG2 and SMMC7721 cells demonstrated reduced motility after knocked down DLAT and CDKN2A ( Figures 10G–J ).