We obtained 371 Liver hepatocellular carcinoma (LIHC) samples and 50 normal samples from TCGA. Additionally, we downloaded two publicly available datasets GSE45267 and GSE49515 from GEO (https://www.ncbi.nlm.nih.gov/geo/). Raw data from TCGA and GEO were preprocessed using the “affy” and “limma” packages in R software. Patients’ Clinical information, including age, sex, stage, and survival data, was extracted from the TCGA database (https://portal.gdc.cancer.gov/).

Differentially expressed genes (DEGs) identification was screened on three datasets: GSE45267, GSE49515 and TCGA. Venn diagrams were used to identify overlapping up-and-down-regulated genes. WGCNA is a powerful bioinformatics tool that can identify groups of highly related genes and their relationship to phenotypic traits. We utilized WGCNA to analyze the overlapping genes to determine key module for the following analysis.

Next, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis were conducted. p 0.05 was used to determine the significance of the enriched results. The co-expression network of essential genes also had major modules that we could locate using the Molecular Complex Detection (MCODE) algorithm. MCODE analysis was carried out using the Cytoscape (version 3.8.0, Institute for Systems Biology, Seattle, WA, USA) program by k-core (2), degree cutoff (2), maximum depth (100), and node score cutoff (0.2). These helped us identify key biological pathways and subnetworks relevant to HCC progression.

Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis was conducted on the 75 node genes screened out by the MCODE algorithm, and the minimum lambda value (lambda. min = 0.0166) was selected in this study. A risk score was determined for each TCGA-HCC tumor sample to create a prognostic risk model. We created scatterplots and heatmaps of gene expression for the risk model and identified 26 significant genes. The TCGA samples were split into high- (n = 185) and low-risk (n = 185) groups according to the average risk score, and the Kaplan-Meier database performed the overall survival (OS) analyses. The risk score formula was as follows: Riskscore = (–0.0199) $\times$ RPA1 + (0.2099) $\times$ WDR12 + (0.0206) $\times$ CDK4 + (0.0147) $\times$ DCAF13 + (0.1308) $\times$ FTSJ3 + (0.3767) $\times$ CPSF3 + (–0.1757) $\times$ RFC1 + (–0.1788) $\times$ RRP1B + (0.1366) $\times$ RMI1 + (–0.0077) $\times$ POLD3 + (0.1838) $\times$ ASF1A + (0.1581) $\times$ CPSF2 + (0.0078) $\times$ NUP43 + (0.1364) $\times$ XPOT + (0.0332) $\times$ CCNF + (0.1683) $\times$ ESF1 + (0.2319) $\times$ PAK1IP1 + (–0.0111) $\times$ NOL12 + (–0.1896) $\times$ DDX59 + (–0.4887) $\times$ RFC3 + (0.487) $\times$ TTK + (0.0187) $\times$ RRM2 + (0.0052) $\times$ CEP55 + (0.0855) $\times$ ECT2 + (0.0763) $\times$ FBXO5 + (–0.6687) $\times$ ZWILCH. Then, using receiver operating characteristic (ROC) analysis, we reported 1-, 3-, and 5-year OS and compared OS curves between subgroups. The capability of the model was evaluated by area under the area under curve (AUC).

First, we obtained gene data from TCGA, GSE45267 and GSE49515 datasets, respectively. Then, univariate Cox regression analysis was employed to determine the most critical prognostic genes. Following the identification of potential prognostic genes, we employed a batch survival analysis to evaluate their collective impact on patient survival. For this purpose, we utilized the Kaplan-Meier estimator, a non-parametric statistic, to estimate the survival function from lifetime data. The patient cohort was divided into high and low-expression groups based on the median expression level of each gene.

The Cox proportional hazards model is a widely applied statistical method for survival analysis, which considers the influence of multiple variables on the survival time of patients. Herein, we applied univariate/multivariate Cox analyses to investigate the prognostic value of 19 genes with significant expression in survival analysis. Multivariate Cox analysis was used to assess the individual prognostic potential of each gene after adjusting for other covariates. Finally, we selected genes substantially related to survival outcomes in univariate and multivariate Cox analyses to construct prognostic nomograms. The nomogram was further validated using a calibration curve.

In this study, we utilized the University of Alabama at Birmingham Cancer (UALCAN) (http://ualcan.path.uab.edu/index.html) database to study the levels and trends of key genes in the clinical characteristics of liver cancer, including age, gender, individual cancer, TP53 mutation and tumor grade. UALCAN is an interactive portal that provides easy access to TCGA data. The immune scores were then further evaluated using immuneDeconv (version 2.0.2; https://github.com/grst/immunedeconv), which uses gene expression data to evaluate the comparative abundance of immune cell types in tumor samples.

American type culture collection (ATCC) provided HCC cell lines (Huh7, Hep3B, HepG2, MHCC97H) and normal liver cells (LO2). HepG2 and MHCC97H were put in RPMI-1640 media with 10% FBS, whereas Huh7 and Hep3B were in Dulbecco’s Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F12) conditions. 10% FBS was added to DMEM to boost the culture of LO2 cells. Experiments were conducted on cells between passages 3 and 8, and cells were subcultured every three to four days. All cell lines were validated by short tandem repeat (STR) profiling and tested negative for mycoplasma.

We conducted siRNA transfection using Lipofectamine RNAiMAX Transfection Reagent (13778150; Thermo Fisher Scientific, Waltham, MA, USA). Total RNA was collected from the cells 48 hours after transfection by TRIzol reagent (15596018; Invitrogen, Carlsbad, CA, USA). Utilizing the StepOnePlus Real-Time PCR System (4376357; Thermo Fisher Scientific, Waltham, MA, USA) and SYBR Green PCR Master Mix (4309155; Thermo Fisher Scientific, Waltham, MA, USA), qRT-PCR analysis was carried out. The primers for PAK1IP1 were as follows: Forward 5^′-AGTTATGCTCAGTTCCAATCCAGT-3^′ and Reverse 5^′-CAAGGAGGCAGTGTGAGCAT-3^′; for GAPDH: Forward 5^′-ACAGTCAGCCGCATCTTCTT-3^′ and Reverse 5^′-GTTAAAAGCAGCCCTGGTGA-3^′.

Using RIPA buffer enhanced with protease and phosphatase inhibitors (78430; Thermo Fisher Scientific, Waltham, MA, USA), total protein was recovered from cells. The BCA protein assay kit (23225; Thermo Fisher Scientific, Waltham, MA, USA) was applied to calculate the protein concentration. Electrophoresis was used to separate equal quantities of protein (20–40 µg) put onto 10–12% SDS-PAGE gels. A wet transfer technique was then employed to transfer the isolated proteins onto nitrocellulose or PVDF membranes. Membranes were kept with primary antibodies at 4 °C overnight after being blocked with 3% BSA in TBST for an hour. Primary antibodies include: PAK1IP1 (Cat No. 16071-1-AP, 1:2000; Proteintech, Wuhan, China), gasdermin E (GSDME) (ab215191, 1:1000; Abcam, Shanghai, China), Pro-CASP3 (ab32150, 1:1000, Abcam, Shanghai, China), Cleaved-CASP3 (Cat no. 82707-13-RR, 1:5000; Proteintech, Wuhan, China), GSDMD (Cat No. 20770-1-AP, 1:2000; Proteintech, Wuhan, China), GSDMD-N (FNab10690, 1:1000; Wuhan Fine Biotech Co., Ltd., Wuhan, China), Cleaved-caspase-1 (Cat no. 4199, 1:1000; Cell signaling Technology, Danvers, MA, USA), GAPDH (ab8245, 1:10000; Abcam, Shanghai, China) The membranes were TBST-washed before incubation for an hour with secondary antibodies (ab9482, 1:5000; Abcam, Shanghai, China) that were HRP-conjugated. An excellent chemiluminescent substrate (ECL) substrate was used to see the protein bands, and imaging equipment (Model Gel Doc XR+, Hercules, CA, USA) was used to take pictures of them. To balance the amounts of protein expression, we employed either $\beta$-actin or GAPDH as an internal control. With ImageJ (version 1.53; National Institutes of Health, Bethesda, MD, USA), the bands were quantified.

5 $\times$ 10³ cells were planted in each well in 96-well plates, and the cells were then left to attach for 24 hours. After treatment with the experimental reagents, the CCK-8 solution (CK04; Dojindo Molecular Technologies, Kumamoto, Japan) was diluted to 10 µL in each well, and the plates were incubated at 37 °C for 1–4 hours. The absorbance at 450 nm was measured with a microplate reader (Multiskan™ GO Microplate Spectrophotometer Thermo Scientific, Waltham, MA, USA). By contrasting the absorbance of treated and control cells, the vitality of the cells was determined. The experiments were carried out three times, and the mean standard deviation was used to show the findings.

Cells were put into the upper chamber of a Transwell insert with an 8 µm hole size for the migration test after being suspended in serum-free DMEM. 10% FBS-containing media was put into the bottom chamber. The upper chamber of the Transwell insert had Matrigel (BD Biosciences, San Jose, CA, USA) precoated for use in the invasion experiment. Cells were suspended in serum-free media to seed the upper chamber. 10% FBS-containing media was put into the bottom chamber. The migrating and invading cells were fixed and stained with DAPI. The amount of invading or migrating cells was counted after microscopically (BX51; Olympus, Tokyo, Japan) captured images.

To assess the amounts of IL-1$\beta$ protein expression in cell culture supernatants, the enzyme-linked immunosorbent assay (ELISA) was used. At 4 °C, 96-well plates were coated with anti-IL-1$\beta$ capture antibody (ab9722, 0.5 µg/mL; Abcam, Shanghai, China) overnight. The wells were filled with the cell culture supernatants and cultured for 2 hours indoor after blocking with 1% bovine serum albumin (A7030; Sigma-Aldrich, St. Louis, MO, USA) in Phosphate buffered saline (PBS) for 2 hours. Anti-IL-1$\beta$ detection antibody that had been biotinylated was then added to the wells and incubated for an additional hour indoor in a solution containing 1% bovine serum albumin and PBS. Streptavidin-horseradish peroxidase (S911, 1:5000 dilution; Thermo Fisher Scientific, Waltham, MA, USA) was added to each well after three PBS washes with 0.05% Tween 20 (P1379; Sigma-Aldrich, St. Louis, MO, USA) were completed. Each well was then kept for 30 minutes. Finally, the wells were rinsed with PBS with 0.05% Tween 20, and the reaction was developed using a tetramethylbenzidine substrate solution (N301; Thermo Fisher Scientific, Waltham, MA, USA). A microplate reader (Multiskan™ GO Microplate Spectrophotometer Thermo Scientific, Waltham, MA, USA) detected the optical density at 450 nm. The content of IL-1$\beta$ in the cell culture supernatants was computed based on the standard curve. All samples were run in duplicate.

Various bioinformatics and statistical analysis packages of R language (version 4.1.3; R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org/) were employed, such as edgeR (version 3.9; http://bioconductor.org/packages/edgeR/), limma (version 4.1; http://bioconductor.org/packages/limma/), survival (version 3.2.11; http://CRAN.R-project.org/package=survival), ggplot2 (version 3.5.1; https://ggplot2.tidyverse.org/), etc. We preprocessed each dataset, including normalization of gene expression levels and correction for batch effects. For the screening of DEGs, we applied the edgeR and limma packages to screen for differentially expressed genes, and set the significance level as FDR 0.05 and log₂ fold change $>$1 or –1. The data were analyzed using a one-way analysis of variance (ANOVA) by the R function “aov” for multiple comparisons and a two-tailed Student’s t test for group comparisons. The findings were expressed as mean $\pm$ standard deviation representing the data. We conducted the aforementioned experiments at least three times, and used p 0.05 for the purpose of determining statistical significance.

We performed a comprehensive analysis of liver cancer using a multi-database integration. We found 8977 up- and 1001 down-regulated DEGs from the TCGA dataset (Fig. 1A), 4717 up- and 2343 down-regulated DEGs from the GSE45267 dataset (Fig. 1B), and 3442 up and 1682 down-regulated DEGs (Fig. 1C). The Venn diagram revealed 1163 upregulated and 76 downregulated genes at the intersection (Fig. 1D,E). By using WGCNA analysis, we determined a soft threshold power of 1 (Fig. 1F). Two modules were identified based on the clustering dendrogram (Fig. 1G) and measured the relation between module eigengenes (ME) and clinical characteristics; the turquoise module was the key module for subsequent analysis (Fig. 1H).

Fig. 1.

Identification and WGCNA analysis of DEGs in liver cancer. (A–C) Volcano plots showing identified DEGs from the TCGA dataset (A), GSE45267 dataset (B) and GSE49515 dataset (C). Orange and green dots represent up-regulated and down-regulated genes, respectively. (D) Venn diagram showing the intersection of upregulated DEGs from the three datasets. (E) Venn diagram showing the intersection of downregulated DEGs from the three datasets. (F) Determination of soft threshold power using WGCNA analysis. (G) Dendrogram of gene clustering in WGCNA analysis, with modules indicated by color. (H) Correlation heatmap showing correlations between modular eigengenes (MEs) and clinical features. The Turquoise module was identified as a key module related to clinical features. WGCNA, weighted gene co-expression network analysis; DEGs, differentially expressed genes.

The genes in the turquoise module were related to the DNA metabolic process, Mitotic spindle organization, and DNA repair (Fig. 2A). Furthermore, enriched pathways included RNA transport, Herpes simplex virus 1 infection, Cell cycle, and DNA replication (Fig. 2B). Additionally, we utilized the MCODE algorithm to analyze the protein-protein interaction (PPI) network of the turquoise module genes. Fig. 2C,D present the PPI networks of the turquoise module genes, highlighting two significant clusters, MCODE Cluster 1 and MCODE Cluster 2, with 30 and 45 nodes and 379 and 247 edges, respectively. These clusters indicate densely interconnected groups of genes within the turquoise module, suggesting their involvement in significant biological pathways or processes crucial to HCC tumorigenesis. The nodes symbolize individual genes, and the edges denote their interactions, with larger nodes reflecting higher connectivity. The distinct structures of these clusters may imply different sub-pathways or mechanisms within HCC-related biological processes, potentially linked to tumor progression or treatment resistance. These figures collectively underscore the complex interplay among the genes in the turquoise module and their probable role in HCC development.

Fig. 2.

GO, KEGG pathway enrichment analysis and MCODE networks of turquoise module genes. (A) GO analysis of turquoise module gene enrichment. The abscissa is the Gene ratio, and the ordinate is the enriched term. (B) KEGG pathway enrichment analysis of turquoise module genes. The abscissa is the Gene ratio, and the ordinate is the enriched pathway. (C) PPI network of turquoise module genes. (D) PPI network of MCODE 1 cluster (orange nodes) and MCODE 2 cluster (green nodes) in turquoise module genes. Nodes in the network represent genes and edges represent protein-protein interactions. The size of the nodes reflects the degree of connectivity of the corresponding genes. GO, gene ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; MCODE, molecular complex detection; PPI, protein-protein interaction.

Fig. 3A–C collectively illustrate the process of developing a prognostic risk model and the identification of key genes that can predict survival outcomes in liver cancer. Fig. 3A presents the LASSO regression analysis on nodal genes, showing the distribution of coefficients across a range of lambda values, which is a tuning parameter that controls the regularization strength (Fig. 3A). Fig. 3B identifies the optimal lambda.min value through cross-validation, which is 0.0166, minimizing the mean cross-validated prediction error (Fig. 3B). Fig. 3C depicts the risk score distribution and gene expression levels for the 26 prognostic genes, indicating a correlation between gene expression (z-scores) and patient survival status, with higher risk scores being associated with higher mortality rates. The selection of these 26 genes is based on their statistical significance and strong association with patient survival, as determined by the LASSO regression analysis, aiming to provide a more accurate risk assessment for liver cancer patients. In addition, the OS analysis revealed the high-risk group’s survival prognosis was poorer (Fig. 3D). Besides, ROC analysis showed AUC values of 0.858, 0.79, and 0.749 for the first, third, and fifth years, respectively (Fig. 3E). These findings provided valuable insights into HCC prognosis and highlighted the potential clinical utility of the identified prognostic genes and risk models.

Fig. 3.

LASSO regression analysis and construction of liver cancer risk model. (A) Distribution plot of LASSO coefficients for nodal genes. (B) Optimal lambda. min value selected by 10-fold cross-validation. (C) Patient characteristics ordered by their risk score. From top to bottom are the risk scores of 26 genes, the distribution of patient survival status, and the heat map of patients in the low-risk group and high-risk group. (D) Kaplan-Meier curves of overall survival in high-risk and low-risk groups. (E) ROC curves of the liver cancer risk model at 1, 3, and 5 years. ROC, receiver operating characteristic; CI, confidence interval; AUC, area under the curve; LASSO, least absolute shrinkage and selection operator.

We investigated the expressions of 26 prognostic genes in TCGA, GSE45267, and GSE49515 datasets (Fig. 4A–C). Our results demonstrated these genes were upregulated in tumor samples. Kaplan-Meier survival analysis indicated that 19 genes were significantly linked to poor prognosis, and high expressions demonstrated lower survival (Fig. 4D–V). Together, these 19 genes have potential as prognostic biomarkers for cancer patients and can be used to develop personalized treatment strategies.

Fig. 4.

Expression and prognostic analysis of 26 identified genes in liver cancer. (A–C) Expression levels of 26 genes from TCGA, GSE45267 and GSE49515 datasets in normal and tumor tissues. Red and green represent tumor and normal samples, respectively. (D–V) Kaplan-Meier survival analysis of 19 genes in HCC patients in the TCGA dataset. The X-axis represents the survival time, and the Y-axis represents the survival rate. Red and green curves represent high and low expression groups, respectively. p-values and hazard ratios (HR) with 95% confidence intervals (CI) are shown. *p 0.05, **p 0.01, ***p 0.001, ****p 0.0001. HCC, hepatocellular carcinoma.

In this study, we utilized univariate and multivariate Cox regression analysis to discover that PAK1IP1 and pathological Tumor-Node-Metastasis (pTNM) stage were individual predictive variables for HCC (Fig. 5A,B). Subsequently, we developed a nomogram by integrating pAK1IP1 and pTNM stage to enhance the precision of prognosis prediction for HCC patients (Fig. 5C). The calibration plot demonstrated a strong concordance between the projected and observed survival rates (Fig. 5D). These emphasized the value of these two factors in HCC prognosis and the potential of integrating them in a nomogram for clinical use.

Fig. 5.

Construction of the prognostic nomogram. (A) Forest plot of univariate Cox regression analysis of PAK1 interacting protein 1 (PAK1IP1) and clinicopathological variables. (B) Forest plot of multivariate Cox regression analysis of PAK1IP1 and clinicopathological variables. (C) Nomogram for HCC prognosis prediction constructed by integrating PAK1IP1 and pTNM staging. (D) The calibration plot of the nomogram shows the agreement between predicted and observed survival. The x-axis represents predicted survival probabilities and the y-axis represents actual survival probabilities. Diagonal lines represent perfect predictions. HCC, hepatocellular carcinoma; pTNM, pathological Tumor-Node-Metastasis; 95% CI, 95% confidence interval; pT, pathological tumor; Uni, univariate; Muit, multivariate; Pro, probability.

Next, we studied the level and clinical significance of PAK1IP1 in LIHC. According to data from the UALCAN database, PAK1IP1 level was considerably greater in tumor samples (Fig. 6A). Furthermore, PAK1IP1 level was higher in HCC patients with older age, male gender, and TP53 mutation (Fig. 6B,C,E). In addition, as tumor stage and grade increased, PAK1IP1 expression also increased (Fig. 6D,F). Next, we identified most immune cells were downregulated in the low PAK1IP1 expression group, and Myeloid dendritic cells had the highest infiltration percentage in tumor samples (Fig. 6G,H). Our findings suggested that PAK1IP1 may be a predictive biomarker and be connected to immune cell infiltration.

Fig. 6.

Clinical feature and immune analysis of PAK1IP1 in liver cancer. (A) PAK1IP1 expression levels in normal and tumor samples according to the UALCAN database. (B–F) Boxplot displaying PAK1IP1 expression levels in individuals with liver cancer, stratified by (B) age, (C) sex, (D) tumor stage, (E) TP53 mutation status, and (F) tumor grade. (G) Heat map of immune cell scores, where different colors represent expression trends in different samples. (H) The percentage abundance of tumor-infiltrating immune cells in each sample. Different colors represent different immune cell types, the abscissa represents the sample, and the ordinate represents the percentage of immune cell content in a single sample. *p 0.05, **p 0.01, ***p 0.001, ****p 0.0001. Yrs, yeras; LIHC, liver hepatocellular carcinoma.

We discovered through qRT-PCR analysis that HCC cells (especially Hep3B and HepG2) have higher levels of PAK1IP1 expression than normal liver cells (Fig. 7A). Our results demonstrated that si-PAK1IP1#1 and si-PAK1IP1#2 more efficiently reduce PAK1IP1 expression in HCC cells (Fig. 7B). Moreover, western blotting analysis validated the efficacy of si-PAK1IP1#1 and si-PAK1IP1#2 in significantly downregulating PAK1IP1 protein levels in HCC cells (Fig. 7C). We then applied CCK-8 assays to assess the impact of PAK1IP1 knockdown on the proliferation of HCC cells. As expected, PAK1IP1 knockdown inhibited HCC cell growth (Fig. 7D,E). Additionally, Transwell assays demonstrated that PAK1IP1 knockdown markedly impeded the invasive and migratory capabilities of HCC cells (Fig. 7F–I).

Fig. 7.

Knockdown of PAK1IP1 inhibits growth, invasion and migration of HCC cells. (A) qRT-PCR analysis of PAK1IP1 expression in normal hepatocytes and HCC cells. (B,C) Efficiency of si-PAK1IP1 #1, #2 and #3 in HCC cells detected by qRT-PCR and WB in HCC cells. GAPDH was used as a loading control. (D,E) CCK-8 assay showed the effect of PAK1IP1 knockdown on the proliferation of HCC cells. (F–I) Transwell assay for the effect of PAK1IP1 knockdown on the invasion and migration of HCC cells. Scale: 50 µm. *p 0.05, **p 0.01. HCC, hepatocellular carcinoma; qRT-PCR, quantitative real-time polymerase chain reaction; WB, western blot; CCK-8, cell counting kit-8; si-NC, small interfering RNA negative control.

We constructed a physical interaction network linking PAK1IP1 and 33 pyroptosis-related genes, as identified in a prior study [15] (Fig. 8A). To investigate whether PAK1IP1 was related to the regulation of pyroptosis, we measured the production of IL-1$\beta$, a quintessential marker of pyroptosis, in HCC cells post-LPS stimulation and subsequent PAK1IP1 knockdown by ELISA. According to our findings, IL-1 levels considerably increased following LPS treatment and PAK1IP1 knockdown (Fig. 8B,C).

Fig. 8.

PAK1IP1 regulates gene expression associated with pyroptosis in HCC cells. (A) Physical interaction network between PAK1IP1 and 33 pyroptosis-related genes. Nodes represent genes, and edges represent connectivity between genes. (B,C) ELISA assay to detect the effect of LPS treatment and knockdown of PAK1IP1 hepatocellular carcinoma cells on pyroptosis-related markers (IL-1$\beta$). *p 0.05, **p 0.01. ###p 0.001 vs. si-PAK1IP1#1/2. HCC, hepatocellular carcinoma; ELISA, enzyme-linked immunosorbent assay; LPS, lipopolysaccharide; IL, interleukin; NC, negative control.

Moreover, we found that PAK1IP1 inhibition significantly increased the protein levels of CASP-3, GSDME-N, cleaved caspase-1, and GSDMD-N, as detected by WB (Fig. 9A). We subsequently assessed the effects of si-PAK1IP1 #1 and N-benzyloxycarbonyl-Asp (OMe)-Glu (OMe)-Val-Asp (OMe)-fluoromethyl-ketone (Z-DEVD-FMK) treatment, a CASP-3 inhibitor, on the proliferation of liver cancer cells. The data showed that PAK1IP1 knockdown significantly reduced HCC cell proliferation, while treatment with Z-DEVD-FMK increased cell proliferation (Fig. 9B,C). Transwell assays further revealed that PAK1IP1 knockdown suppressed liver cancer cell invasion and migration after LPS treatment, while Z-DEVD-FMK treatment promoted cellular invasion and migration (Fig. 9D–G). These findings indicate that PAK1IP1 promotes the proliferation, migration, and invasion of HCC cells by inhibiting pyroptosis.

Fig. 9.

PAK1IP1 knockdown suppresses proliferation, invasion and migration of HCC cells through the CASP-3 pathway. (A) WB analysis of caspase 3 (CASP-3), gasdermin E (GSDME)-N, cleaved caspase-1, and gasdermin-D (GSDMD)-N protein expression in HCC cells after PAK1IP1 knockdown. (B,C) CCK-8 assay showing the effect of si-PAK1IP1 knockdown and N-benzyloxycarbonyl-Asp (OMe)-Glu (OMe)-Val-Asp (OMe)-fluoromethyl-ketone (Z-DEVD-FMK) treatment on the proliferation of HCC cells. (D,E) Transwell assay showing the effect of PAK1IP1 knockdown on the invasion and migration of HCC cells after LPS treatment. Scale: 50 µm. (F,G) Transwell analysis showing the effect of Z-DEVD-FMK treatment on the invasion and migration of HCC cells. Scale: 50 µm. *p 0.05. #p 0.05 vs. si-PAK1IP1#1/2. HCC, hepatocellular carcinoma; WB, western blot; CCK-8, cell counting kit-8.