ABSTRACT
-
Background/Aims
Hepatocellular carcinoma (HCC) is characterized by high recurrence and mortality, necessitating the identification of reliable biomarkers. In this study, we aimed to identify the predictive gene signatures for HCC recurrence and evaluate the efficiency of GULP PTB domain-containing engulfment adaptor 1 (GULP1) as a predictive and diagnostic marker and therapeutic target for HCC.
-
Methods
We analyzed genomic datasets from The Cancer Genome Atlas and Gene Expression Omnibus databases via least absolute shrinkage and selection operator Cox regression and 10-fold cross-validation, leading to the development of a 15-gene risk score model, which was validated using three independent datasets. Serum GULP1 and α-fetoprotein levels were assessed to determine the diagnostic accuracy of the model. Using clinical cohorts and patient sera, GULP1 roles were examined, and functional assays in vitro and in vivo were used to evaluate its effects on cell growth, epithelial–mesenchymal transition (EMT), ADP-ribosylation factor 6 (ARF6) activation, and β-catenin signaling.
-
Results
Our newly developed risk-score model accurately predicted recurrent HCC in all datasets. Among the 15 genes in the risk score model, GULP1 was overexpressed in patients with HCC and independently predicted HCC recurrence. Its expression modulation influenced cell growth and EMT, with observed effects on ARF6 activation and β-catenin signaling pathways.
-
Conclusions
GULP1 is a crucial biomarker for HCC, serving as a non-invasive diagnostic and predictive tool. It also plays key roles in HCC progression. Our findings highlight the potential use of GULP1 in treatment strategies targeting EMT and HCC recurrence to improve the personalized care and patient outcomes.
-
Keywords: Liver cancer; GULP PTB domain-containing engulfment adaptor 1; Recurrence; Metastasis; Diagnosis
Study Highlights
• GULP1 exhibited predictive accuracy comparable to that of a 15-gene risk score model for HCC recurrence with high clinical application potential and robustness.
• GULP1 was specifically overexpressed in HCC, distinguishing it from other liver conditions and showing significant prognostic and diagnostic value.
• GULP1 promoted tumor growth, EMT, and invasiveness by modulating β-catenin signaling, playing key roles in HCC progression.
• Our findings suggest GULP1 as a promising non-invasive biomarker and therapeutic target for HCC recurrence and progression.
Graphical Abstract
INTRODUCTION
Hepatocellular carcinoma (HCC) is a prevalent aggressive cancer with high recurrence and mortality rates and low patient quality of life [
1]. Key risk factors for HCC include chronic hepatitis (CH) B and C, excessive alcohol intake, and metabolic disorders [
2,
3]. Due to subtle early symptoms, HCC is often diagnosed at advanced stages, complicating its timely treatment and contributing to recurrence rates as high as 80% [
1,
4]. Although α-fetoprotein (AFP) is widely used in clinical practice as a serum biomarker for HCC, multiple studies showed limited sensitivity and specificity [
1]. This shortcoming highlights the need for more reliable biomarkers for the early diagnosis, risk stratification, and personalized treatment of HCC.
In this study, we applied advanced machine learning techniques, including least absolute shrinkage and selection operator (LASSO) Cox regression with 10-fold crossvalidation, to identify the gene signatures associated with HCC recurrence. Based on the results, we developed a 15-gene risk score (RS) model and identified the GULP PTB domain-containing engulfment adaptor 1 (GULP1) as a potential prognostic marker for HCC recurrence.
GULP1 is the human counterpart of CED6 from Caenorhabditis elegans, with both playing a conserved role in cellular engulfment across species [
5]. GULP1 facilitates EphB/ephrinB trogocytosis—a process of cell surface material transfer—by collaborating with Tiam2 and is essential for recruiting dynamin, a crucial protein for cellular internalization [
6]. Interestingly, although GULP1 is typically recognized as a tumor suppressor, its expression was elevated in HCC, where it promoted β-catenin activity and epithelial– mesenchymal transition (EMT), which are associated with metastasis [
7,
8]. Importantly, our data indicate that GULP1 outperforms AFP in detecting early-stage HCC and predicting its recurrence, highlighting it as a more sensitive and specific biomarker than AFP. Furthermore, our investigation highlighted the potential of GULP1 as a biomarker in liquid biopsies, serving as a minimally invasive tool to monitor HCC recurrence and progression, thereby advancing personalized medicine for HCC.
In summary, our study emphasized GULP1’s role in recurrence and metastasis, providing insights into the mechanisms behind HCC progression and suggesting promising strategies for early diagnosis and targeted therapy.
MATERIALS AND METHODS
Patients and specimens
HCC and adjacent non-cancerous tissues and blood samples obtained from the Ajou University Hospital (Suwon, Korea) were used in this study. Tissue samples were collected from 81 patients with HCC who underwent hepatectomy, and 256 blood samples were collected from healthy individuals and patients with CH, cirrhosis, and HCC. To validate the diagnostic performance of GULP1 across various liver disease etiologies, additional patient samples, including hepatitis C virus–induced liver cirrhosis (HCV-LC; n=30), alcoholic LC (n=30), alcoholic HCC (n=30), metabolic dysfunction-associated steatohepatitis-LC (MASH-LC; n=30), MASH-HCC (n=30; 5 with confirmed MASH and 25 with unknown origins), cholangiocarcinoma (n=8), and combined HCC and cholangiocarcinoma (n=22) samples, were obtained from the Human Biobank. The patient demographics and clinical characteristics are presented in
Supplementary Tables 1–
3.
All experiments were performed in accordance with the Declaration of Helsinki, and the study was approved by the Institutional Review Board of Ajou University Hospital (AJIRB-BMR-KSP-16-365, AJIRB-BMR-SMP-17-189, AJOUIRB-KSP-2019-417, AJOUIRB-EX-2022-389 and AJOUIRB-EX-2024-332). Anonymous serum samples and clinical data were provided by the Ajou Human Bio-Resource Bank; the requirement for informed consent was waived.
All animals were cared for in accordance with the Guide for the Care and Use of Laboratory Animals and experiments were approved by the Ethics Committee for Laboratory Animal Research Center of Ajou University Medical Center (IACUC_2022-0049).
Collection and analysis of gene expression data for HCC recurrence
We analyzed the gene expression data from three publicly available HCC datasets (GSE14520, GSE114564, and The Cancer Genome Atlas [TCGA] liver hepatocellular carcinoma [LIHC]) to identify the genes associated with HCC recurrence. The patients were divided into non-recurrent and recurrent groups based on their recurrence status within two years post-surgery. Initial differential expression analysis identified the candidate genes across all datasets. Using Elastic Net Cox regression and LASSO Cox regression with cross-validation, we refined these candidates to construct a robust 15-gene recurrence RS model. Detailed patient baseline characteristics are presented in
Supplementary Table 4, and additional methodological details, including dataset processing and statistical parameters, are provided in the
Supplementary Methods section.
RESULTS
Development of a 15-gene RS model to predict HCC recurrence
In this study, patient selection criteria included patients who underwent surgical liver resection and remained recurrence-free for over two years (non-recurrence [NR] group), as well as patients who experienced recurrence (recurrence [R] group). To identify the differentially expressed genes (DEGs) associated with HCC recurrence, we analyzed three datasets (GSE14520, GSE114564, and TCGA LIHC) with clinical information on patients with HCC who underwent surgical liver resection and were either recurrence-free for over two years or experienced recurrence within two years (
Supplementary Table 4). Venn diagrambased analysis identified 50 overlapping DEGs between the R and NR groups (
Fig. 1A). To identify the genes potentially associated with recurrence, we conducted Elastic Net Cox regression analysis on the 50 selected genes within TCGA LIHC dataset.
Lambda values, a hyperparameter for model optimization, were examined for each gene (
Fig. 1B, left;
Supplementary Table 5). Through 10-fold cross-validation, optimal lambda value was determined as 0.1829059, as indicated by the blue line in the graph (
Fig. 1B, right panel). From this analysis, 15 genes were identified as independent predictors within the model, and their respective regression coefficients are listed in
Supplementary Table 6.
Next, to further refine the RS model, we calculated RSs for each gene by applying univariate Cox regression to the expression values of the 15 genes using TCGA dataset of HCC with recurrence information. Regression coefficient for the RS model was obtained as follows:
RS=(0.03208×GULP1)+(−0.04351×LCAT)+(0.29381×PPAT)+(−0.05648×LPXN)+(0.09575×NOP56)+(−0.03938×CD4)+(0.12171×ZC2HC1A)+(0.42938×PPIA)+(0.08035×CST7)+(−0.04075×PRKCQ)+(0.12525×PHF20)+(0.03416×RAB23)+(0.07323×PCDHB6)+(−0.17598×CXCR6)+(−0.09678×SLC4A10).
Using TCGA model as a reference, a cut-off value was used in each dataset to classify the patients into high- and low-risk groups. Correlation analysis of all datasets consistently demonstrated a negative correlation between patient RS scores and survival (
Fig. 1C). Notably, our 15-gene RS model outperformed the existing 7-gene models, showing superior predictive performance and higher net benefits across all datasets (
Supplementary Fig. 1A) [
9]. Receiver operating characteristics (ROC) analysis further confirmed the predictive capacity of the 15-gene RS model, with area under the curve (AUC) ≥0.7, indicating relatively high accuracy in predicting HCC recurrence (
Fig. 1D). Additionally, GSEA revealed the significant enrichment of gene sets from Hallmark Collection in the high-risk group, indicating their involvement in specific signaling pathways, such as E2F targets, G2/M checkpoint, mitotic spindle, and MYC targets v1 (
Supplementary Fig. 1B). This connection is crucial because MYC influences EMT and recurrence in various cancers, including HCC [
10-
12]. Notably, Kaplan–Meier curves showed that the low-risk group (RS_Low) exhibited a significantly longer recurrence-free survival (RFS) than the high-risk group (RS_High) in all three datasets (
Fig. 1E). Through univariate analysis of the clinical information in each dataset, RS model exhibited a consistent and independent association with the diagnostic outcome in all analyzed datasets (
Fig. 1F). Cox regression analysis of RFS in TCGA cohort revealed that the RS model exhibited significant associations in both univariate (hazard ratio [HR] 2.72; 95% confidence interval [CI] 2.146–3.443) and multivariate (HR 2.08; 95% CI 1.311–3.313) analyses, showing the highest HR compared to other variables (
Supplementary Table 7).
Within the 15-gene RS model,
GULP1 emerged as a potent biomarker, showing predictive accuracy comparable to that of the entire model for HCC recurrence. Although the RS model robustness was enhanced by integrating multiple genes,
GULP1 alone was sufficient to stratify patients according to the recurrence risk with similar precision (
Supplementary Fig. 2). These findings suggest
GULP1 as a simple and cost-effective biomarker for clinical use that reliably predicts the recurrence risk without the need for complex multigene profiling. Additionally, analysis of
GULP1 in 33 different TCGA cancer types revealed that
GULP1 was generally downregulated in other cancer types (
Supplementary Fig. 3A). This finding was contrary to our expectations and prompted us to specifically investigate its role in HCC. We investigated the expression levels of
GULP1 in HCC at various stages and conditions using GepLiver DB datasets. This analysis revealed differences in
GULP1 expression levels among the normal, adjacent nontumor (ADJ_HCC), and HCC liver tissues, with HCC tissues exhibiting markedly elevated
GULP1 levels (
Fig. 2A). Quantitative analysis showed that
GULP1 levels were significantly higher in HCC tissues than in normal, viral hepatitis, nonalcoholic fatty liver disease, and cirrhosis tissues, underscoring its potential in distinguishing HCC from other liver conditions (
Fig. 2B).
High-throughput analysis of spatial transcriptomics showed a pronounced increase in
GULP1 levels in the malignant hepatocytes, suggesting a distinct spatial distribution of the gene in the tumor microenvironment (
Fig. 2C, 2D;
Supplementary Fig. 3B).
Single-cell RNA sequencing data from GepLiver DB revealed diverse cellular landscapes in various liver cell types. The datasets included various liver-specific and immune cells, providing a comprehensive view of the cell populations present in both normal and diseased liver tissues (
Fig. 2E, left;
Supplementary Fig. 3C). Within this diverse cellular environment, we specifically examined the
GULP1 expression levels in hepatocytes. The proportion of
GULP1-positive hepatocytes increased progressively with the progression of liver malignancy. In normal tissues only 0.82% of hepatocytes were
GULP1-positive; however, this number increased to 1.91% in ADJ_HCC tissues, 2.9% in non-malignant HCC tissues, and 4.39% in malignant HCC tissues (
Fig. 2E, right). Similarly, analysis of the scAtlasLC database and subsequent quantification of
GULP1-positive hepatocytes confirmed the increased
GULP1 expression in HCC, emphasizing its significance in liver cancer pathology (
Fig. 2F). Mean expression levels of
GULP1 in different cell types in the GSE151530 dataset further confirmed that the hepatocytes were the primary source of
GULP1 expression, highlighting the hepatocyte-specific roles of
GULP1 in HCC (
Fig. 2G;
Supplementary Fig. 3D). Subsequently, hepatocytes were classified into
GULP1-positive and -negative populations to explore the functional implications of
GULP1 expression (
Fig. 2H;
Supplementary Fig. 3E).
GULP1-positive hepatocytes were transcriptionally enriched in the key oncogenic pathways, including the EMT, hypoxia, and KRAS pathways critical for HCC progression (
Fig. 2I;
Supplementary Fig. 3F,
3G). These findings suggest that hepatocyte clusters with high
GULP1 expression are transcriptionally aligned with key oncogenic processes.
Considering the predictive strength of GULP1, Kaplan– Meier analysis was conducted to assess the survival differences based on GULP1 expression levels in HCC. Patients in the three datasets, GSE14520, GSE114564, and TCGA LIHC, with high
GULP1 expression levels (GULP1_High) showed worse prognosis than those with low
GULP1 levels (GULP1_Low;
Supplementary Fig. 4A–4C). To explore the diagnostic and prognostic value of GULP1 in HCC, we examined an independent cohort of 81 patients with HCC who underwent hepatic resection (
Supplementary Table 1). Quantitative reverse transcription-polymerase chain reaction (qRT-PCR) analysis revealed significantly increased
GULP1 transcript levels in the tumor (T) samples compared to those in the non-tumor (NT) samples, confirming its elevation in HCC tissues (
Supplementary Fig. 4D,
4E;
Fig. 3A, left box plot in the tissue sample panel). In a separate validation cohort, we assessed the protein levels of GULP1 as a diagnostic marker in serum samples (
Supplementary Table 2). Serum GULP1 levels were significantly elevated in patients with HCC compared to those in the NT group, confirming its upregulation in HCC (
Fig. 3A, left box plot in the blood sample panel). GULP1 concentrations were significantly higher in patients at different HCC stages (mUICC I, I/II, and III/IV) than in those with normal liver (NL), CH, and LC (
Fig. 3A, middle scattered dot plot in the blood sample panel). Although the diagnostic power of GULP1 at the transcriptomic level in tissue samples showed a modest AUC of 0.67 for detecting liver cancer; its diagnostic performance as a serum-based marker was more robust with an AUC of 0.85, demonstrating superior accuracy in distinguishing HCC from the NL and early liver diseases (
Fig. 3A, ROC curves for tissue and blood sample panels).
Compared to AFP, GULP1 showed significantly higher AUC values in the high-risk group (CH/LC) for all patients with HCC (0.827 for GULP1 and 0.595 for AFP) (
Fig. 3B, left panel). In the mUICC III/IV group, GULP1 exhibited a higher AUC (0.877) than AFP (0.766), although the difference was not statistically significant (
Fig. 3B, second panel). In the mUICC I/II group, GULP1 demonstrated a relatively strong diagnostic ability (AUC=0.816) compared to that of AFP (AUC=0.556;
Fig. 3B, third panel). In the earliest stage of HCC (mUICC I, tumor size ≤2 cm), GULP1 showed a high AUC (0.749), whereas the diagnostic accuracy of AFP was significantly lower (AUC=0.516;
Fig. 3B, the fourth panel).
Serum GULP1 and AFP levels showed distinct positivity rates across healthy subjects and CH, LC, and HCC groups. Although GULP1 demonstrated a lower incidence than AFP in the non-HCC groups (i.e., NL, CH, and LC), it exhibited a higher positivity rate in the HCC group (
Fig. 3C, left panel). Detection rates of AFP and GULP1 in all 145 patients with liver cancer were 44% and 70%, respectively. When both markers were used together, HCC positivity rate increased to 81% (
Fig. 3C, right panel).
To investigate the variation in GULP1 expression levels across different etiologies of HCC, we analyzed the serum GULP1 levels in cohorts stratified by the hepatitis B virus (HBV), HCV, alcohol, and MASH (
Supplementary Table 3). GULP1 levels were significantly higher in HCC than in LC across all tested etiologies (
Supplementary Fig. 5A–
5E). In HBV-induced liver disease, GULP1 levels were significantly higher in HCC than in LC (
P<0.001), with an AUC of 0.871, surpassing the diagnostic performance of AFP (AUC=0.707;
P=0.006;
Supplementary Fig. 5A). In HCV-, alcoholic-, and MASH-related liver diseases, GULP1 levels were consistently and significantly elevated in HCC compared to those in LC. Although ROC analysis indicated that GULP1 demonstrated superior diagnostic performance over AFP for these etiologies, the observed differences were not statistically significant (
Supplementary Fig. 5B–
5D). We further evaluated the diagnostic potential of GULP1 in non-HCC liver tumors, including cholangiocarcinoma and combined hepatocellular-cholangiocarcinoma. GULP1 levels were significantly higher in non-HCC liver tumors than in NT samples (NL and LC;
P<0.001). ROC analysis revealed that GULP1 exhibited an AUC of 0.833 in distinguishing the NT samples (n=145) from the non-HCC liver tumors (n=30), markedly outperforming AFP (AUC=0.540;
P<0.0001;
Supplementary Fig. 5E). Comparing LC to non-HCC liver tumors, GULP1 showed an AUC of 0.790, consistently surpassing AFP per formance (AUC=0.524;
P=0.002;
Supplementary Fig. 5F).
Considering the strong association between GULP1 and the 15-gene RS model for predicting recurrent HCC, we further analyzed its potential as a recurrence biomarker. qRT-PCR analysis of tissue samples revealed significantly increased
GULP1 expression levels in R versus NR cases (
Fig. 3D, left box plot in the tissue sample panel). Time-dependent AUROC curves were plotted to evaluate the ability of GULP1 to predict recurrence over time. Time‐dependent AUC‐based C‐index for GULP1 in tissue samples was 0.745 (95% CI 0.611–0.859), surpassing that of AFP 0.663 (95% CI 0.489–0.815). Statistical comparison of mean time‐dependent AUCs confirmed the superior predictive performance of GULP1 (
P=8.82×10
-18;
Fig. 3D, time-dependent AUC curve in the tissue samples panel). High serum GULP1 levels were significantly associated with R, further supporting its potential as a biomarker for liver cancer recurrence (
Fig. 3D, left box plot in the blood sample panel). Time-dependent AUROC analysis revealed that GULP1 exhibited a slightly higher predictive capacity than AFP, with a C-index of 0.726 (95% CI 0.623–0.826) vs. 0.703 (95% CI 0.576–0.814), although the difference was not statistically significant (
P=0.188;
Fig. 3D, time-dependent AUC curve in the blood samples panel). Western blotting analysis confirmed this upregulation, showing elevated GULP1 protein levels in RT vs. PT tissues, underscoring its involvement in recurrence mechanisms (
Fig. 3E). Furthermore, tissue (left) and serum (right)-based RFS evaluations revealed that high GULP1 expression was significantly associated with a poor patient prognosis (
Fig. 3F).
To investigate the roles of GULP1 in tumor growth, proliferation, and invasion in HCC, we analyzed its expression levels in various liver cancer cell lines. Endogenous GULP1 levels were the highest in PLC/PRF/5 and Huh-7 cells, leading to their selection for subsequent experiments (
Supplementary Fig. 6A).
GULP1 knockdown (siGULP1 group) significantly reduced the cell growth, proliferation, and clonogenic capacity. These effects were partially rescued by GULP1 overexpression (GULP1_OE;
Fig. 4A). To further validate the oncogenic potential of GULP1
in vivo,
GULP1-suppressed Huh-7 cells were subcutaneously injected into female BALB/c nude mice.
GULP1-depleted cells exhibited a significantly lower growth rate than the negative control cells (
Fig. 4B).
Furthermore, immunohistochemistry (IHC) analysis of xenograft tumor tissue sections revealed reduced Ki-67 and PCNA expression levels in the
GULP1-depleted group (
Fig. 4C). These findings highlight the roles of GULP1 in promoting tumor growth and proliferation. However, as GULP1 was upregulated in recurrent HCC and affected RFS, we examined its impact on metastasis and invasion, which are critical phenotypes influencing HCC recurrence, using a wound-healing assay to observe the effect of GULP1 on the migration of liver cancer cells.
GULP1 knockdown significantly reduced the wound-healing capacity of these cells, whereas
GULP1 re-expression effectively restored their migratory potential (
Fig. 4D). Transwell invasion assay showed a similar effect of
GULP1 expression on cell invasiveness (
Fig. 4E, left panel). Three-dimensional sphere cultures showed that
GULP1 modulation significantly affected sphere formation and cell outgrowth (
Fig. 4E, right panel), suggesting that GULP1 contributes to cell invasion.
In a subcutaneous xenograft model,
GULP1 knockdown significantly affected the expression levels of various markers. Specifically, expression levels of epithelial markers (Ecadherin and zonula occludens-1) were upregulated, whereas those of mesenchymal markers (vimentin, fibro-nectin, and slug) and angioinvasion markers (CD31 and vascular endothelial growth factor [VEGF]) were downregulated upon
GULP1 suppression (
Fig. 4F).
To validate the effect of GULP1 on HCC recurrence
in vivo, GULP1-suppressed Hepa1-6 cells were orthotopically injected into the mouse liver and recurrent tumors were resected for evaluation (
Fig. 5A). IHC analysis revealed a significant increase in GULP1 expression levels, particularly in RT tissues, compared to those in NL and PT tissues, highlighting its role in tumor recurrence and progression (
Fig. 5B;
Supplementary Fig. 6B). However, GULP1 suppression significantly reduced the number, size, and weight of recurrent tumors (
Fig. 5C).
IHC analysis also demonstrated changes in the expression levels of key markers in the
GULP1-suppressed group (
Fig. 5D). Specifically, decreased expression of the mesen-chymal marker (vimentin) indicated reduced tumor invasion and metastasis. Increase in the levels of epithelial markers, such as E-cadherin, indicated the reversal of EMT, a critical step in cancer metastasis. Significant changes were observed in PT, with more pronounced alterations in RT than in the NT, indicating the substantial impacts of GULP1 suppression on recurrent tumors (
Fig. 5D).
The above-mentioned findings were further validated using a lung metastasis model of ras-transformed NIH-3T3 cells. GULP1 expression was regulated followed by tail vein injection, further supporting the metastatic role of GULP1 in promoting tumor recurrence (
Supplementary Fig. 6C). Upon resection, metastatic nodules were significantly reduced in the GULP1-suppressed group (
Fig. 5E, left panel). Hematoxylin and eosin (H&E) staining showed small metastatic lesions in the lungs of the GULP1-suppressed group, reinforcing the reduction in metastasis (
Fig. 5E, right panel). Additionally, analysis of RNA-seq data of the biopsies of patients with liver cancer (GSE164359) revealed higher GULP1 expression levels in the RT than in the adjacent liver tissue (AL) and PT (
Fig. 5F).
Notably, no significant differences in body weight were observed in the three
in vivo models after GULP1 knockdown (
Supplementary Fig. 6D), indicating the potential of GULP1-targeting therapeutics for recurrent and metastatic HCC.
DEAD-box helicase 5 directly interacts with β-catenin, facilitating its nuclear translocation and transactivation. This interaction is associated with alterations in the expression of GULP1, a protein associated with neuroblastoma progression [
13]. In our initial assessment of liver cancer cell lines, β-catenin suppression considerably decreased the GULP1 protein levels (
Fig. 6A,
6B). Interestingly, modulation of GULP1 expression significantly influenced β-catenin localization, as observed via fluorescence microscopy and supported by quantitative analysis, indicating a strong correlation between GULP1 levels and β-catenin subcellular distribution (
Fig. 6C,
6D).
To further elucidate the mechanisms underlying GULP1-mediated modulation of β-catenin translocation, we investigated its interaction with ADP-ribosylation factor 6 (ARF6)-GTP based on the proposed role of GULP1 in stabilizing ARF6 activity [
14]. Consistently, our results revealed that changes in ARF6-GTP were barely affected by the reduction in GULP1 levels (
Fig. 6E). However, in the presence of cycloheximide, a protein synthesis inhibitor, GULP1 depletion significantly decreased the ARF6-GTP levels. Remarkably, GULP1 re-expression under these conditions successfully restored the ARF6-GTP levels, confirming the role of GULP1 in stabilizing ARF6 activity (
Fig. 6E). Direct interaction between GULP1 and ARF6 was further validated via fluorescence resonance energy transfer-based assay. Sm-Bit-tagged ARF6 and Lg-Bit-tagged GULP1 were used to measure the luminescence signals indicative of their interactions (
Supplementary Fig. 7A). Indeed, luminescence signals were significantly reduced when GULP1 was depleted, confirming the dependence of the signals on GULP1 presence. In contrast, luminescence signals were successfully rescued when GULP1 was overexpressed in the depleted state (
Supplementary Fig. 7B). When ARF6 expression was depleted, luminescence signals decreased, confirming the direct interaction between GULP1 and ARF6 (
Supplementary Fig. 7C). Luminescence signals were significantly reduced upon β-catenin depletion (siCTNNB1 group), indicating that β-catenin activity is essential for ARF6–GULP1 interaction (
Supplementary Fig. 7D). These findings demonstrate the dependency of GULP1-mediated β-catenin translocation on ARF6 activation, as knockdown of ARF6 suppressed β-catenin localization (
Fig. 6F). Furthermore, ARF6 downregulation was accompanied by a decrease in GULP1 expression, reinforcing their relation-ship (
Supplementary Fig. 7E).
Bioinformatics analyses identified a transcription factor 3 (TCF3)-binding motif within the
GULP1 promoter region (chr 2:189,158,687-189,158,701), suggesting direct regulation by TCF3 (
Supplementary Fig. 7F). Luciferase promoter mutation reporter assay was performed to confirm the binding of TCF3 to the
GULP1 promoter. Mutations in the TCF3 binding site significantly reduced the luciferase activity, suggesting that TCF3 directly binds to and regulates the
GULP1 promoter (
Fig. 6G).
ARF6 knockdown decreased the binding affinity of β-catenin and TCF3 to the GULP1 promoter, as demonstrated by the chromatin immunoprecipitation assays (
Fig. 6H;
Supplementary Fig. 7G). Modulation of GULP1 expression further altered the TCF3 binding affinity to the
GULP1 promoter region (
Fig. 6I), indicating that these proteins form a complex involving direct binding between β-catenin and TCF3 at the promoter region of GULP1.
GULP1 regulated β-catenin interactions with key adhesion molecules, including N-cadherin and E-cadherin (
Fig. 6J). This regulatory capacity of GULP1 extended to the expression of downstream β-catenin targets, such as SRY-box transcription factor 9, c-Myc, and fibronectin, indicating a positive feedback mechanism in the β-catenin signaling pathway that significantly influences EMT and HCC progression (
Fig. 6K). GULP1 suppression reduced β-catenin expression in the
in vivo experiment (
Supplementary Fig. 7H). Further analysis of the GSE164359 dataset revealed that several key genes, including
MKI67, CTNNB1, VIM, VEGFA, FN1, and
SOX9, exhibited elevated expression levels in RT, similar to the
GULP1 expression levels (
Supplementary Fig. 7I). Correlation analysis revealed a moderately positive relationship between GULP1 and these gene expression levels, suggesting that GULP1 influences these oncogenic pathways in recurrent HCC (
Supplementary Fig. 7J).
In summary, our study identified GULP1 as a β-catenin signaling modulator in HCC. By stabilizing ARF6-GTP, GULP1 facilitated β-catenin localization and transcriptional activation via TCF3 binding at its promoter, thereby driving oncogenic processes, such as EMT and tumor progression (
Fig. 7).
DISCUSSION
Various guidelines have been established for primary liver cancer; however, management of recurrent liver cancer remains challenging despite various treatment options, such as hepatectomy, radiotherapy, transplantation, and systemic therapy [
15,
16]. Despite surgical resection being the preferred treatment, three-year recurrence rate of HCC remains high [
17]. For small recurrent HCC cases following surgical resection, radiofrequency ablation results in overall survival and RFS rates comparable to those observed with repeated surgical resection [
18]. Despite advancements in medical techniques, prediction of recurrent HCC remains challenging. Although such predictions can improve the patient survival rates by enabling proactive measures, accurate HCC recurrence prediction models are lacking. Conventional HCC markers, such as AFP, often exhibit low sensitivity and specificity in diagnosing various liver cancer types; therefore, many studies are exploring novel serum biomarkers, such as glypican-3, and multigene biomarker panels for liver cancer [
19,
20]. The near absence of reliable recurrence predictors underscores the importance of our study in this field.
Thorough understanding of EMT, a hallmark of recurrence, is pivotal to predict HCC recurrence and its detrimental effect on the patient overall survival [
21]. EMT involves a transition, in which epithelial cells gain mesenchymal features and exhibit enhanced migration and invasion [
22-
24]. This transition is accompanied by changes in cell morphology, reduced cell–cell adhesion, and increased migration and invasion. Several markers implicated in EMT are recognized as key players in HCC recurrence, including transcription factors Snail, Slug, and Twist [
25]. Downregulation of the levels of epithelial markers, such as E-cadherin and cytokeratin, and upregulation of the levels of mesenchymal markers, including N-cadherin, vimentin, and fibronectin, are commonly observed during EMT [
26]. Specifically, Wnt/β-catenin pathway is also implicated in EMT regulation, as aberrant activation of β-catenin drives EMT and promotes HCC progression and recurrence [
27-
29].
Building on these, this study explored the oncogenic role of GULP1 in HCC metastasis. Interestingly, GULP1 predominantly acts as a tumor suppressor in many cancer types. For instance, its reduced expression in ovarian cancer, attributed to epigenetic silencing via genomic methylation, is associated with advanced disease stages and unfavorable prognosis [
7].
In urothelial carcinoma, downregulation of GULP1 levels induces cell growth, predominantly via activation of the NRF2–KEAP1 signaling pathway [
8]. However, our findings revealed its different roles in HCC, where it functioned as an oncogene. Specifically, GULP1 inhibition reduced β-catenin activity, while its overexpression enhanced it. This modulation was achieved via ARF6 activation. The role of GULP1 in oncogenesis is thus context-dependent, akin to other cancer-associated genes with dual functions, such as ARID1A and HDAC6 [
30,
31]. While GULP1 promotes β-catenin activation and EMT in HCC—facilitating tumor progression and recurrence—it suppresses tumorigenesis in ovarian and urothelial. These divergent roles underscore the importance of the tumor microenvironment and pathway-specific interactions in defining GULP1’s function. This not only broadens our understanding of the complex associations among GULP1, β-catenin signaling, and EMT but also highlights the pivotal role of EMT in cancer metastasis [
32]. Additionally, experimental results further suggest that GULP1 functions as an oncogene. First, we demonstrated its elevation in both tissue and serum of HCC patients, as well as its association with recurrence through time-dependent AUROC analyses. Second, functional assays
in vitro revealed that GULP1 knockdown dampens tumor cell proliferation, migratory capacity, and invasive behavior, while rescue experiments reversed these effects. Third,
in vivo models established GULP1’s involvement in tumorigenesis and metastasis—particularly in recurrent settings—suggesting its pivotal role in both the initiation and progression of HCC.
We anticipate that GULP1 serves as an oncogenic factor in HCC through diverse mechanisms. Notably, our GSEA findings indicate that GULP1 is linked not only to the Wnt pathway but also to other key signaling cascades such as NOTCH and HEDGEHOG (
Supplementary Fig. 8A–
8D). These pathways are all well-documented to play crucial roles in HCC onset and progression, suggesting that GULP1 may integrate multiple oncogenic signals. Future research efforts should focus on elucidating these interactions to provide a more comprehensive understanding of GULP1-driven hepatocarcinogenesis.
Nevertheless, we acknowledge that GULP1 expression may also fluctuate due to liver inflammation, cirrhosis, and coexisting treatments. These factors highlight the need for additional large-scale, multicenter validation studies to refine assay cutoffs and adjust for confounding clinical variables. While our mechanistic data pinpoint GULP1’s role in ARF6-mediated β-catenin activation, further targeted approaches could more definitively validate these interactions in HCC recurrence. In addition, its modest predictive power in serum underscores the need for further studies to refine detection thresholds, validate assay methods to enhance overall diagnostic accuracy and clinical utility.
Looking ahead, we propose future studies to (1) explore combinational biomarker panels—including GULP1 with AFP or other emergent markers—to enhance predictive accuracy, (2) systematically investigate how GULP1 cooperates with Wnt, Notch, and Hedgehog pathways, and (3) develop therapeutic interventions targeting GULP1-driven ARF6–β-catenin signaling in recurrent HCC. By addressing these gaps, we believe GULP1 may evolve into not only a valuable clinical biomarker but also a therapeutic target for controlling HCC progression and recurrence.
Overall, this study revealed GULP1 as a key biomarker for HCC, acting both as a non-invasive diagnostic tool and an oncogene driving HCC progression. Our findings highlight its potential for improving HCC treatment strategies, particularly for developing therapeutics targeting EMT and cancer cell growth. However, further research and clinical validation are crucial to fully harness the potential of GULP1 and substantially advance HCC treatment.
FOOTNOTES
-
Acknowledgements
This work was supported by grants from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute funded by the Ministry of Health and Welfare, Republic of Korea (HR21C1003), Bio and Medical Technology Development Program of the National Research Foundation (NRF-2022R1A2C2092422 and RS-2023-00210847), and Commercialization Promotion Agency for R&D Outcomes (COMPA) funded by the Ministry of Science and ICT (RS-2024-00422549).
Biospecimens and data used in this study were provided by the Biobank of the Ajou University Hospital, a member of the Korea Biobank Network. Additionally, we would like to thank all members of MOAGEN (Daejeon, South Korea) for assistance and guidance with bioinformatics analysis. For this study, a graphical abstract (JZ27K8N3P9; https://biorender.com/w63d148) was created using BioRender (www.biorender.com). ChatGPT (version 4o) was used for error checking and language refinement during the preparation of this manuscript.
-
Authors’ contribution
H.S.K. and J.H.Y. contributed equally to this study. Conceptualization: H.J.C., S.S.K., J.W.E.; Data curation: H.S.K., J.H.Y., M.G.Y., G.O.B., J.E.H.; Methodology: M.G.Y., G.O.B., J.-Y.J., S.W.N.; Investigation: : H.S.K., J.H.Y., W.P., Y.G., J.T.N., S.B.L.; Resources: M.K., S.W.N., J.E.H., H.J.C., S.S.K., J.Y.C.; Visualization: G.O.B., H.J.C., S.H.J., J.E.H.; Funding acquisition: J.Y.C., H.S.K., J.W.E.; Project administration: S.S.K., J.W.E.; Supervision: J.Y.C.; Writing – original draft: : H.S.K., J.H.Y., H.J.C., J.W.E.; Writing – review & editing: S.S.K., J.Y.C., J.W.E. All authors have read and approved the final manuscript.
-
Conflicts of Interest
The authors have no conflicts to disclose.
SUPPLEMENTAL MATERIAL
Supplementary material is available at Clinical and Molecular Hepatology website (
http://www.e-cmh.org).
Supplementary Figure 1.
Evaluating predictive models for hepatocellular carcinoma (HCC) recurrence and their clinical impact. (A) Decision curve analysis (DCA) comparing the net benefits of the 15-gene risk score (RS) model and the existing 7-gene model by Du et al.9 across different threshold probabilities in three datasets, indicating that the 15-gene RS model outperforms the 7-gene model. (B) Biological and functional pathway analysis between the low- and high-risk groups performed using gene set enrichment analysis (GSEA).
Supplementary Figure 2.
Expression levels and diagnostic performance of GULP1 and other genes. Box plots and ROC curves for the 15 genes identified in the recurrence risk score (RS) model, comparing expression levels between non-recurrence (NR) and recurrence (R) groups. Among these genes, GULP1 showed the highest expression in the R group and demonstrates superior sensitivity and specificity for recurrence diagnosis, as indicated by the ROC analysis.
Supplementary Figure 3.
GULP1 as an indicator for HCC aggressiveness. (A) The expression level of GULP1 was analyzed across a wide array of cancer types in the TCGA dataset. GULP1 is significantly overexpressed in HCC compared to its expression in other cancer types, where it appears predominantly downregulated. The data points for individual samples are plotted, with median expression levels indicated by horizontal black lines. TPM, transcripts per million. (B) Spatial transcriptomic analysis. Left: representative spatial transcriptomic spots colored by non-malignant (yellow) and malignant (purple) hepatocytes from tumor tissues. Right: GULP1 expression in spatial sections. (C) UMAP plot displaying cell type clusters across various liver cancer-related datasets, with distinct cell types marked by different colors, including hepatocytes, T cells, B cells, macrophages, endothelial cells, and fibroblasts. (D) Single-cell transcriptomic data (GSE151530) revealing GULP1 expression across various cell types, including T cells, TAMs (tumor-associated macrophages), hepatocytes, TECs (tumor endothelial cells), Tregs (regulatory T cells), B cells, and MDSCs (myeloid-derived suppressor cells). The bar chart on the right confirms that hepatocytes predominantly express GULP1. (E) UMAP-based sub-clustering of 18,539 hepatocytes into nine groups (C1–C9). (F) Hallmark pathway enrichment analysis (MSigDB Hallmark 2020) ranking gene sets enriched in GULP1 (+) samples. (G) Summary table of selected hallmark pathways with corresponding normalized enrichment scores (NES) and P-values, indicating the statistical significance of GULP1-linked oncogenic signaling in HCC.
Supplementary Figure 4.
GULP1 expression correlates with poor recurrence-free survival (RFS) in HCC and is elevated in recurrent tissues. (A–C) Kaplan–Meier curves showing RFS for HCC patients, stratified by high (purple) and low (gray) GULP1 expression levels. Patients with high GULP1 expression demonstrate significantly poorer RFS outcomes. Analysis was conducted across three datasets: (A) GSE14520, (B) GSE114564, and (C) TCGA LIHC, all providing recurrence information. (D) Comparative mRNA expression analysis of GULP1 using qRT-PCR in 81 pairs of normal and HCC liver tissue samples, demonstrating significantly higher expression in HCC tissues. The x-axis numbers represent the unique patient IDs. (E) qRT-PCR analysis in a subset of 24 patients with recurrent HCC, showing elevated levels of GULP1 in recurrent tumor tissues compared to matched adjacent non-tumor (NT) tissues. The x-axis numbers represent the unique patient IDs. *P<0.05, **P<0.01, ***P<0.001.
Supplementary Figure 5.
GULP1 expression and diagnostic performance across liver disease etiologies. (A) Hepatitis B virus (HBV) cohort: Serum GULP1 measurements in liver cirrhosis (LC; n=30) and HCC (n=30). GULP1 levels were notably higher in HCC than LC (P<0.001), yielding an AUC of 0.871, which exceeds that of AFP (AUC=0.707; P=0.006). NL (normal liver) refers to healthy control samples included for baseline comparison. (B) Hepatitis C virus (HCV) cohort: Serum GULP1 levels were assessed in LC (n=30) versus HCC (n=30). GULP1 was substantially upregulated in HCC relative to LC (P<0.001), achieving an AUC of 0.821 compared to AFP’s 0.801 (P=0.729). (C) Alcoholic group: Serum GULP1 levels in alcoholic LC (n=30) and alcoholic HCC (n=30). GULP1 expression was significantly elevated in HCC versus LC (P<0.001), with an AUC of 0.836 surpassing AFP’s 0.810 (P=0.716). (D) Metabolic dysfunction-associated steatohepatitis (MASH) group: Serum GULP1 levels in LC (n=30) compared to HCC (n=30). GULP1 was markedly increased in HCC (P<0.001), generating an AUC of 0.890, higher than AFP’s 0.810 (P=0.120). (E) Comparison of serum GULP1 concentrations among normal liver (NL), liver cirrhosis (LC), hepatocellular carcinoma (HCC), and non-HCC liver tumors (non-HCC LT). Statistical significance was determined via one-way ANOVA followed by multiple comparison tests. (F) Left: Non-tumor vs. non–HCC liver tumor comparison: Serum GULP1 levels for non-tumor samples (NL and LC, n=145) and non–HCC liver tumors (n=30). GULP1 was significantly higher in non–HCC liver tumors compared to non-tumor samples (P<0.001), providing an AUC of 0.833, which exceeds AFP’s 0.540 (P<0.0001). Right: Serum GULP1 levels and ROC analysis for LC (n=115) and non-HCC liver tumors (n=30). GULP1 levels were significantly higher in non- HCC liver tumors compared to LC (P<0.001), with an AUC of 0.790, outperforming AFP’s AUC of 0.524 (P=0.002). Data are represented as the mean±standard deviation. *P<0.05, **P<0.01, ***P<0.001.
Supplementary Figure 6.
GULP1 expression and its validation in vivo HCC models. (A) Relative protein abundance of GULP1 in healthy liver cell lines versus HCC cell lines, determined by Western blot analysis. The results indicate a marked increase in GULP1 protein levels in HCC cell lines, particularly in PLC/PRF/5 and Huh-7 cells. (B) Immunohistochemistry (IHC) analysis of orthotopically injected, GULP1- suppressed Hepa1-6 cells in mouse liver. The stained sections demonstrate a reduction in GULP1 expression in tumor (T) tissues compared to non-tumor (NT) tissues. (C) Validation of GULP1 suppression in ras-transformed NIH-3T3 cells. The bar graph shows the results of qRT-PCR confirming reduction in GULP1 expression (left). The Western blot analysis corroborates the suppression of GULP1 levels (right). (D) Comparative analysis of body weight upon GULP1 suppression in different in vivo models. Data are shown as mean±standard error of the mean. Statistical significance was determined using unpaired t-tests. **P<0.01, ***P<0.001.
Supplementary Figure 7.
Direct modulation of GULP1 by ARF6 and β-catenin in HCC. (A) Schematic diagram of the FRET (Förster resonance energy transfer) assay used to detect ARF6–GULP1 interaction. ARF6 and GULP1 were each tagged with either Sm-Bit (small fragment of luciferase) or Lg-Bit (large fragment of luciferase). When the two proteins come into close proximity, these luciferase fragments reconstitute into an active enzyme that emits a luminescent signal, indicating direct protein–protein binding. (B) FRET-based quantification of luminescence signals under varying GULP1 expression levels in HCC cells. After transfection with GULP1-targeting siRNA (siGULP1) or GULP1-overexpressing vector (GULP_OE), luminescence was measured to assess changes in ARF6–GULP1 interaction. (C) FRET-based evaluation of ARF6 knockdown (siARF6). Luminescence was markedly lower upon ARF6 suppression. (D) Assessment of ARF6–GULP1 interaction following β-catenin (CTNNB1) depletion. Cells were transfected with siCTNNB1, and luminescence signals were measured to determine any impact on the ARF6–GULP1 complex. (E) Western blot analysis confirmed a decrease in GULP1 expression following ARF6 suppression. (F) Sequence logos representing the TCF3 transcription factor consensus DNA binding sites; the y-axis indicates the information amount at each motif position. (G) The binding activity of TCF3 at the GULP1 promoter regions, quantified with the chromatin immunoprecipitation (ChIP) assays, and presented in bar graphs. (H) Immunohistochemistry (IHC) of β-catenin in subcutaneous and orthotopic HCC xenograft models, comparing negative control (siCtrl) and siGULP1 groups. Bar graphs show significantly reduced β-catenin staining with GULP1 suppression. (I) Violin plots from the GSE164359 dataset illustrating elevated GULP1 expression in recurrent tumor (RT) samples relative to primary tumor (PT) and adjacent liver tissue (AL). (J) Pearson correlation analysis from the GSE164359 dataset, revealing significant positive correlations between GULP1 and its downstream targets (CTNNB1, FN1, SOX9, etc.). **P<0.01, ***P<0.001.
Supplementary Figure 8.
GULP1-driven oncogenic pathways in HCC. (A) Bar chart of the top 20 enriched Hallmark gene sets in the GULP1high subgroup (top quartile of GULP1 expression) compared to the GULP1low subgroup (bottom quartile) in the TCGA LIHC dataset (n=371). Rankings are based on normalized enrichment score (NES). (B) GSEA enrichment plot for EPITHELIAL_MESENCHYMAL_ TRANSITION (NES=1.806, NOM P=0.019). (C) GSEA enrichment plot for NOTCH_SIGNALING (NES=1.559, NOM P=0.032). (D) GSEA enrichment plot for HEDGEHOG_SIGNALING (NES=1.541, NOM P=0.043). *P<0.05.
Figure 1.Identification of differentially expressed genes (DEGs) in recurrent hepatocellular carcinoma (HCC), and development of a risk score (RS) model. (A) Flow chart of the selection of genes associated with recurrent HCC. Patient selection criteria included patients who underwent surgical liver resection and remained recurrence-free for over two years (non-recurrence [NR]) and those who experienced recurrence (R). (B) Left: Coefficient profiles of correlated DEGs in least absolute shrinkage and selection operator (LASSO) analysis. Right: LASSO model parameter adjustment with 10-fold cross-validation. Blue dashed line indicates the optimal lambda value cutoff (0.1829059). (C) Correlation analysis of RS and patient survival using three HCC datasets: GSE14520, GSE114564, and The Cancer Genome Atlas (TCGA) liver hepatocellular carcinoma (LIHC). (D) Predictive potency determined via receiver operating characteristic (ROC) analysis. (E) Risk-based analysis of recurrence-free survival (RFS) in the GSE14520, GSE114564, and TCGA LIHC datasets using the Kaplan–Meier approach. (F) Forest plots of univariate Cox regression analyses of clinical parameters affecting RFS. HCC differentiation was defined using the Edmondson grade scale. Statistical significance was determined via Cox proportional hazards regression analysis and log-rank tests for survival data. AFP, α-fetoprotein; AJCC, The American Joint Committee on Cancer; AUC, area under the curve; HR, hazard ratio. *P<0.05, **P<0.01, ***P<0.001.
Figure 2.
GULP1 is a hepatocellular carcinoma (HCC) progression indicator. (A) Heatmap of GULP1 expression levels (Z-score of log2[TPM+1]) in the normal, adjacent non-HCC (ADJ_HCC), and HCC tissues across 22 datasets from GepLiver DB. (B) Comparison of GULP1 expression levels across different liver phenotypes using data from GepLiver DB (normal=362; viral hepatitis=180; nonalcoholic fatty liver disease (NAFLD)=503; cirrhosis=63; HCC=724). (C) Spatial transcriptomic analysis of the tumor tissues. GULP1 expression levels were notably higher in the malignant hepatocytes (purple) of the same tissue, with minimal to no expression observed in the nonmalignant hepatocytes (yellow). P9T and P10T indicate the unique patient IDs. (D) Proportion of GULP1-positive cells (%) across all analyzed tumor tissues from spatial transcriptomics. The labels (P1T, P2T, etc.) indicate patient IDs. (E) Uniform manifold approximation and projection (UMAP) plot of an integrated liver single-cell RNA-sequencing (scRNA-seq) dataset from GepLiver DB. Left: Cells colored by major cell type, indicating the distribution of cell types (e.g., hepatocytes, cholangiocytes, and immune cells) across different phenotypes. Right: Proportion of GULP1-positive cells (%) among hepatocytes in the normal, ADJ_HCC, and HCC tissues. (F) Left: UMAP plot of scRNA-seq data from the scAtlasLC database. Right: Proportion of GULP1-positive cells (%) among non-malignant and malignant hepatocytes from the scAtlasLC database. (G) Mean expression levels of GULP1 in different liver-associated cell types in the GSE151530 dataset. (H) UMAP plot showing the GULP1-positive (+) and GULP1-negative (−) hepatocytes among the analyzed 18,539 hepatocytes. (I) Enrichment score plots of the hallmark pathways with GULP1-positive hepatocytes derived via single-sample gene set enrichment analysis (ssGSEA). EMT, epithelial–mesenchymal transition; MDSCs, myeloid-derived suppressor cells.
Figure 3.Clinical significance of GULP PTB domain-containing engulfment adaptor 1 (GULP1) expression in hepatocellular carcinoma (HCC). (A) GULP1 levels in the paired tissue (left) and blood (right) samples. GULP1 levels were significantly elevated in the HCC tissues and serum samples of patients with HCC compared to those in the non-tumor (NT) groups. Receiver operating characteristics (ROC) analysis showed the high diagnostic accuracy of serum GULP1 in distinguishing HCC from NT and early liver disease cases. (B) ROC analysis results of GULP1 and α-fetoprotein (AFP) in high-risk liver disease groups (CH/LC) and across modified Union for International Cancer Control (mUICC) stages. (C) Left: Comparison of the serum GULP1 and AFP positivity rates among different groups (NL, chronic hepatitis [CH], liver cirrhosis [LC], and HCC). Right: Positivity rates of AFP, GULP1, and their combination in patients with liver cancer. (D) Comparison of GULP1 levels in recurrent (R) vs. non-recurrent (NR) cases using the tissue (left panels) and serum (right panels) samples. GULP1 levels are elevated in R cases, showing superior time-dependent predictive performance in tissues and comparable but slightly higher performance than AFP in the serum samples for recurrence prediction. (E) Left: Western blotting analysis of GULP1 protein levels in the primary tumor tissues (PTs) and recurrent tumor tissues (RTs) of three patients. #P01, #P02, and #P03 indicate the unique IDs of patients with HCC. Right: Densitometry analysis shows the increased GULP1 levels in recurrent tumors. (F) Recurrencefree survival (RFS) analysis based on GULP1 expression in the tissue (left) and serum (right) samples. High GULP1 expression was associated with significantly poor prognosis in both tissue and serum evaluations in the validation cohort. Statistical significance was determined via unpaired t-tests for within-group comparisons and one-way analysis of variance (ANOVA) for multi-group comparisons. ROC curve analysis was performed for diagnostic evaluations. Data are represented as the mean±standard error of the mean. *P<0.05, **P<0.01, ***P<0.001.
Figure 4.Targeted inactivation of GULP PTB domain-containing engulfment adaptor 1 (GULP1) suppressed the tumorigenic potential of liver cancer cells. (A) Effects of GULP1 modulation on HCC cell growth, proliferation, and colony formation. Left: 3-(4,5-Dimethylthiazol- 2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay showed the regulated cell growth of PLC/PRF/5 and Huh-7 cells upon the alteration of GULP1 expression. Middle: 5-Bromo-2′-deoxyuridine (BrdU) incorporation assay indicated the regulated proliferation of GULP1-modulated cells. Right: Clonogenic assay revealed the significantly lower colony formation capacity of siGULP1-treated cells compared to that of the controls. (B) Left: Subcutaneous xenograft tumor growth assay revealed that the GULP1-depleted Huh-7 cells exhibited significantly lower growth rates than the control cells. Right: Tumor weight differences in xenografts with GULP1 knockdown. (C) Left: Histopathological examination of tumor sections via hematoxylin and eosin (H&E) staining and immunohistochemistry (IHC) for GULP1, Ki67, and proliferating cell nuclear antigen (PCNA). Right: Inter-group comparative analysis of IHC staining intensity for GULP1, Ki67, and PCNA. (D) Scratch wound-healing assay revealed the effect of GULP1 on cell migration. Left: Representative images of wound closure in PLC/ PRF/5 and Huh-7 cells at 24-hour post-scratch. Right: Quantification of wound closure, showing significantly regulated migration of GULP1-modulated cells. (E) Effects of GULP1 on cell invasion and sphere formation. Left: Invasion assay showed the altered invasiveness of PLC/PRF/5 and Huh-7 cells after GULP1 modulation. Right: Sphere formation assay revealed the significant changes in sphere formation and cell outgrowth upon altered GULP1 expression, with enlarged images highlighting the reduced lamellipodia (arrows). (F) IHC images showing the expression patterns in HCC tissues derived from the subcutaneous xenografts injected with GULP1-depleted Huh7 cells. Statistical significance was determined via unpaired t-tests for comparisons between two groups. siCtrl, scrambled sequence of single interference control RNA; siGULP1, small interfering RNA targeting GULP1; GULP1_OE, GULP1 overexpression. *P<0.05, **P<0.01, ***P<0.001.
Figure 5.Validation of GULP PTB domain-containing engulfment adaptor 1 (GULP1) effects on epithelial–mesenchymal transition (EMT) and hepatocellular carcinoma (HCC) recurrence in vivo. (A) Schematic representation of the experiment illustrating the orthotopic injection of GULP1-suppressed Hepa1-6 cells into the mouse liver. (B) Immunohistochemistry (IHC) analysis showing the significant increase in GULP1 expression levels, particularly in the recurrence tumors compared to those in the normal liver (NL) and primary tumors. Magnifications: 200× and 400×. (C) Left: Representative images of the livers with tumor burden from different groups (siCtrl and siGULP1). Right: Corresponding bar graphs on the right show the calculated nodule volume and weight. (D) Top: IHC analysis of various markers in the NT, primary tumor (PT), and recurrent tumor (RT) tissues. Bottom: Quantification of stained GULP1, Ki-67, E-cadherin, vimentin, and vascular endothelial growth factor (VEGF) expression levels. Proliferation marker: Ki-67. Epithelial marker: E-cadherin. Mesenchymal marker: Vimentin. Angioinvasion marker: VEGF. (E) Lung metastasis model using ras-transformed NIH-3T3 cells with siCtrl or siGULP1 treatment. Left: Lung nodules (arrows) are significantly reduced in the siGULP1-treated tissues. Right: H&E-stained images at 40× and 100× magnification show the dispersed HCC cells (arrows) in siCtrl-treated lungs, whereas siGULP1-treated lungs show improved morphology. Scale bars=50 μm. (F) Analysis of RNA-seq data from the GSE164359 dataset revealed higher GULP1 levels in RT samples than in the adjacent liver tissue (AL) and PT samples. Statistical significance was determined via unpaired t-tests for comparisons between two groups. Data are represented as the mean±standard error of the mean. siCtrl, scrambled sequence of single interference control RNA; siGULP1, small interfering RNA targeting GULP1. *P<0.05, **P<0.01, ***P<0.001.
Figure 6.Mechanistic role of GULP PTB domain-containing engulfment adaptor 1 (GULP1) in β-catenin signaling pathway regulation in hepatocellular carcinoma (HCC). (A) Western blotting analysis revealed that β-catenin suppression considerably decreased the GULP1 protein levels in the HCC cell lines, Huh-7 and PLC/PRF/5. The cells were treated with the CTNNB1-targeting siRNA (siCTNNB1) or siCtrl. (B) Quantification of GULP1 protein levels via enzyme-linked immunosorbent assay (ELISA) after β-catenin suppression in HCC cell lines. Data indicate the significant reduction in GULP1 expression upon β-catenin knockdown. (C) Immunofluorescence (IF) microscopy images illustrating β-catenin nuclear translocation influenced by GULP1 expression in Huh-7 and PLC/PRF/5 cells. siGULP1 or GULP1 overexpressing vector (GULP1_OE) treatment was performed, with nuclei stained with 4′, 6-diamidino-2-phenylindole (DAPI). (D) Quantitative representation of β-catenin translocation normalized to that of histone H3, corresponding to the microscopy findings. Data indicate that GULP1 modulates β-catenin localization. (E) ELISA of the modulation of ARF6-GTP levels in response to changes in GULP1 levels under conditions where protein synthesis was inhibited by cycloheximide (CHX). Expression of ARF6-GTP was normalized to that of total ARF6. siGULP1 and GULP1_OE treatments showed that GULP1 stabilized ARF6-GTP. (F) Visualization of β-catenin nuclear translocation via IF microscopy (left), with accompanying quantitative analysis (right). Suppression of β-catenin localization upon ARF6 knockdown using the ARF6-targeting siRNA (siARF6) supported the critical role of ARF6 in GULP1-mediated β-catenin translocation. (G) Luciferase activity in Huh7 cells transfected with the wild-type (WT) GULP1 promoter construct, first mutant (MT1) with a deletion from +1304 to +2330, and second mutant (MT2) with a sequence alteration at +1006 (CCCGCATCCT). WT construct showed significantly higher activity than MT1 and MT2. (H) Chromatin immunoprecipitation (ChIP) assays showing the decreased binding affinity of β-catenin to the GULP1 promoter region upon ARF6 knockdown in Huh-7 and PLC/PRF/5 cells. (I) Modulation of TCF3 binding affinity to the GULP1 promoter region in response to changes in GULP1 expression. ChIP assay was performed to measure the binding activity. (J) Immunoprecipitation revealed β-catenin nuclear translocation and its enhanced interaction with the key adhesion molecules, N-cadherin and E-cadherin, following GULP1 modulation in Huh-7 and PLC/PRF/5 cells. (K) ELISA showing the expression levels of downstream β-catenin targets, such as SRY-box transcription factor 2 (SOX9), c-Myc, and fibronectin, in response to altered GULP1 expression in Huh-7 and PLC/PRF/5 cells. These results indicate that GULP1 influences the expression levels of these targets via its regulatory effects on β-catenin signaling. Unpaired t-tests were used to determine the statistical significance. Data are represented as the mean±standard error of the mean. ARF6, ADP-ribosylation factor 6; siCtrl, scrambled sequence of single interference control RNA; siGULP1, small interfering RNA targeting GULP1. *P<0.05, **P<0.01, ***P<0.001.
Figure 7.Schematic diagram of the GULP PTB domain-containing engulfment adaptor 1 (GULP1)–β-catenin co-regulatory mechanism in hepatocellular carcinoma (HCC). GULP1 functions as a crucial modulator in the β-catenin signaling axis in HCC that is critical for the stabilization of ARF6-GTP, which further influences the cellular distribution of β-catenin. This stabilization facilitates β-catenin release and subsequent nuclear translocation, where it binds to TCF3 at the GULP1 promoter region, impacting gene transcription. Downstream effects of this binding include enhanced cell proliferation, EMT, and tumor recurrence, which contribute to liver tumorigenesis. ARF6, ADPribosylation factor 6; EMT, epithelial–mesenchymal transition; TCF3, transcription factor 3.
Abbreviations
adjacent non-tumor tissue
one-way analysis of variance
ADP-ribosylation factor 6
chronic hepatitis B virus
differentially expressed gene
enzyme-linked immunosorbent assay
epithelial–mesenchymal transition
fluorescence resonance energy transfer
gene set enrichment analysis
GULP PTB domain-containing engulfment adaptor 1
liver hepatocellular carcinoma
modified Union for International Cancer Control
metabolic dysfunction-associated steatohepatitis
quantitative reverse transcription polymerase chain reaction
receiver operating characteristics
single-cell RNA sequencing
scrambled sequence of single interference control RNA
small interfering RNA targeting GULP1
singlesample gene set enrichment analysis
uniform manifold approximation and projection
vascular endothelial growth factor
REFERENCES
- 1. Vogel A, Meyer T, Sapisochin G, Salem R, Saborowski A. Hepatocellular carcinoma. Lancet 2022;400:1345-1362.
- 2. Mashiba T, Joko K, Kurosaki M, Ochi H, Hasebe C, Akahane T, et al. Real-world efficacy of elbasvir and grazoprevir for hepatitis C virus (genotype 1): A nationwide, multicenter study by the Japanese Red Cross Hospital Liver Study Group. Hepatol Res 2019;49:1114-1120.
- 3. Huang DQ, Mathurin P, Cortez-Pinto H, Loomba R. Global epidemiology of alcohol-associated cirrhosis and HCC: trends, projections and risk factors. Nat Rev Gastroenterol Hepatol 2023;20:37-49.
- 4. Bruix J, Takayama T, Mazzaferro V, Chau GY, Yang J, Kudo M, et al. Adjuvant sorafenib for hepatocellular carcinoma after resection or ablation (STORM): a phase 3, randomised, doubleblind, placebo-controlled trial. Lancet Oncol 2015;16:1344-1354.
- 5. Beyer AS, von Einem B, Schwanzar D, Keller IE, Hellrung A, Thal DR, et al. Engulfment adapter PTB domain containing 1 interacts with and affects processing of the amyloid-β precursor protein. Neurobiol Aging 2012;33:732-743.
- 6. Gong J, Gaitanos TN, Luu O, Huang Y, Gaitanos L, Lindner J, et al. Gulp1 controls Eph/ephrin trogocytosis and is important for cell rearrangements during development. J Cell Biol 2019;218:3455-3471.
- 7. Maldonado L, Brait M, Izumchenko E, Begum S, Chatterjee A, Sen T, et al. Integrated transcriptomic and epigenomic analysis of ovarian cancer reveals epigenetically silenced GULP1. Cancer Lett 2018;433:242-251.
- 8. Hayashi M, Guida E, Inokawa Y, Goldberg R, Reis LO, Ooki A, et al. GULP1 regulates the NRF2-KEAP1 signaling axis in urothelial carcinoma. Sci Signal 2020;13.
- 9. Du B, Wang F, Jarad B, Wang Z, Zhang Y. A novel signature based on microvascular invasion predicts the recurrence of HCC. J Transl Med 2020;18:272.
- 10. Liu M, Casimiro MC, Wang C, Shirley LA, Jiao X, Katiyar S, et al. p21CIP1 attenuates Ras- and c-Myc-dependent breast tumor epithelial mesenchymal transition and cancer stem cell-like gene expression in vivo. Proc Natl Acad Sci U S A 2009;106:19035-19039.
- 11. Yang L, Wang M, Wang Y, Zhu Y, Wang J, Wu M, et al. LINC00460-FUS-MYC feedback loop drives breast cancer metastasis and doxorubicin resistance. Oncogene 2024;43:1249-1262.
- 12. Liu F, Liao Z, Zhang Z. MYC in liver cancer: mechanisms and targeted therapy opportunities. Oncogene 2023;42:3303-3318.
- 13. Zhao X, Li D, Yang F, Lian H, Wang J, Wang X, et al. Long Noncoding RNA NHEG1 Drives β-catenin transactivation and neuroblastoma progression through interacting with DDX5. Mol Ther 2020;28:946-962.
- 14. Ma Z, Nie Z, Luo R, Casanova JE, Ravichandran KS. Regulation of Arf6 and ACAP1 signaling by the PTB-domain-containing adaptor protein GULP. Curr Biol 2007;17:722-727.
- 15. Saraswat VA, Pandey G, Shetty S. Treatment algorithms for managing hepatocellular carcinoma. J Clin Exp Hepatol 2014;4:S80-89.
- 16. Zhou J, Sun HC, Wang Z, Cong WM, Wang JH, Zeng MS, et al. Guidelines for diagnosis and treatment of primary liver cancer in China (2017 Edition). Liver Cancer 2018;7:235-260.
- 17. Gosalia AJ, Martin P, Jones PD. Advances and future directions in the treatment of hepatocellular carcinoma. Gastroenterol Hepatol (N Y) 2017;13:398-410.
- 18. Li JK, Liu XH, Cui H, Xie XH. Radiofrequency ablation vs. surgical resection for resectable hepatocellular carcinoma: A systematic review and meta-analysis. Mol Clin Oncol 2020;12:15-22.
- 19. Qiao SS, Cui ZQ, Gong L, Han H, Chen PC, Guo LM, et al. Simultaneous measurements of serum AFP, GPC-3 and HCCR for diagnosing hepatocellular carcinoma. Hepatogastroenterology 2011;58:1718-1724.
- 20. Kaur H, Dhall A, Kumar R, Raghava GPS. Identification of platform-independent diagnostic biomarker panel for hepatocellular carcinoma using large-scale transcriptomics data. Front Genet 2019;10:1306.
- 21. van Zijl F, Zulehner G, Petz M, Schneller D, Kornauth C, Hau M, et al. Epithelial-mesenchymal transition in hepatocellular carcinoma. Future Oncol 2009;5:1169-1179.
- 22. Brabletz T, Kalluri R, Nieto MA, Weinberg RA. EMT in cancer. Nat Rev Cancer 2018;18:128-134.
- 23. Ribatti D, Tamma R, Annese T. Epithelial-mesenchymal transition in cancer: a historical overview. Transl Oncol 2020;13:100773.
- 24. Huang Y, Hong W, Wei X. The molecular mechanisms and therapeutic strategies of EMT in tumor progression and metastasis. J Hematol Oncol 2022;15:129.
- 25. Yang MH, Chen CL, Chau GY, Chiou SH, Su CW, Chou TY, et al. Comprehensive analysis of the independent effect of twist and snail in promoting metastasis of hepatocellular carcinoma. Hepatology 2009;50:1464-1474.
- 26. Zeisberg M, Neilson EG. Biomarkers for epithelial-mesenchymal transitions. J Clin Invest 2009;119:1429-1437.
- 27. Deldar Abad Paskeh M, Mirzaei S, Ashrafizadeh M, Zarrabi A, Sethi G. Wnt/β-catenin signaling as a driver of hepatocellular carcinoma progression: an emphasis on molecular pathways. J Hepatocell Carcinoma 2021;8:1415-1444.
- 28. Zhang Q, Bai X, Chen W, Ma T, Hu Q, Liang C, et al. Wnt/β-catenin signaling enhances hypoxia-induced epithelial-mesenchymal transition in hepatocellular carcinoma via crosstalk with hif-1α signaling. Carcinogenesis 2013;34:962-973.
- 29. Yang M, Li SN, Anjum KM, Gui LX, Zhu SS, Liu J, et al. A double-negative feedback loop between Wnt-β-catenin signaling and HNF4α regulates epithelial-mesenchymal transition in hepatocellular carcinoma. J Cell Sci 2013;126:5692-5703.
- 30. Sun X, Wang SC, Wei Y, Luo X, Jia Y, Li L, et al. Arid1a has context-dependent oncogenic and tumor suppressor functions in liver cancer. Cancer Cell 2018;33:151-152.
- 31. Jung KH, Noh JH, Kim JK, Eun JW, Bae HJ, Chang YG, et al. Histone deacetylase 6 functions as a tumor suppressor by activating c-Jun NH2-terminal kinase-mediated beclin 1-dependent autophagic cell death in liver cancer. Hepatology 2012;56:644-657.
- 32. Leung RWH, Lee TKW. Wnt/β-catenin signaling as a driver of stemness and metabolic reprogramming in hepatocellular carcinoma. Cancers (Basel) 2022;14.
