Chondroitin sulfate proteoglycan 4 (CSPG4) overexpression as an oncogenic driver and prognostic marker for unfavorable outcomes in hepatocellular carcinoma.

Introduction
Liver cancer, predominantly in the form of hepatocellular carcinoma (HCC), ranks as the principal primary hepatic neoplasm and stands as a significant cause of cancer-related deaths worldwide, with an estimated 905,677 new cases and 830,180 deaths in 2020 [1]. Despite advances in surgical techniques and targeted therapies, the long-term survival rates for HCC patients remain dismal, with only 18% surviving 5 years post-diagnosis [2]. Current treatment options, including surgical resection, liver transplantation, and systemic therapies, are limited in their efficacy, and post-operative recurrence rates remain high [3]. The diverse characteristics of HCC and the scarcity of reliable molecular markers for early diagnosis and prognostic evaluation underscore the urgent need for innovative, potent biological indicators to refine individualized therapeutic approaches.
Notably, existing prognostic tools have critical limitations. For instance, alpha-fetoprotein (AFP), the most widely used biomarker, lacks sensitivity in approximately 40% of cases, particularly in early-stage HCC, and fails to stratify risk in AFP-negative patients [4]. Meanwhile, Tumor Node Metastasis (TNM) staging, although a clinical standard, often underestimates the heterogeneity in intermediate-risk groups, leading to suboptimal treatment stratification [5]. These limitations underscore the pressing need for novel biomarkers to improve prognostic accuracy and enable personalized management.
Chondroitin Sulfate Proteoglycan 4 (CSPG4), alternatively referred to as Neural/Glial Antigen 2, functions as a cell surface protein that plays crucial roles in cellular growth, movement, and blood vessel formation [6]. The structure of CSPG4 encompasses an extensive external segment, a membrane-spanning portion, and a brief intracellular extension, allowing it to interact with various extracellular and intracellular signaling molecules [6]. Increased levels of CSPG4 have been detected in various malignant neoplasms, such as skin melanomas, brain glioblastomas, and hormone-resistant breast tumors. In these contexts, higher CSPG4 presence is associated with unfavorable outcomes and increased risk of metastasis [7–10]. Research on melanoma has revealed CSPG4’s utility as both a tool for diagnosis and prognosis assessment, while also highlighting its promise as a target for novel treatment strategies [11, 12]. Mechanistically, CSPG4 promotes tumor progression by activating integrin-dependent pathways, enhancing growth factor signaling, and modulating the extracellular matrix [9, 13]. In glioblastoma, CSPG4 contributes to therapy resistance by altering the tumor microenvironment and promoting cancer stem cell properties [7, 14]. Despite its significant roles in other cancers, the biological functions and potential clinical significance of CSPG4 in HCC remain largely unexplored.
In response to these issues, we investigated the clinical relevance and mechanistic role of CSPG4 in HCC. Using public databases, two independent patient cohorts, and in vitro experiments, we aimed to clarify CSPG4’s association with clinicopathological features, validate its prognostic value, and explore its potential as a component of a multi-parameter prognostic model. Our findings may position CSPG4 as a promising biomarker and therapeutic target for HCC, and the newly-established prognostic model incorporating CSPG4 may offer a more precise method for forecasting patient outcomes in HCC cases.

Materials and methods

Bioinformatics analysis
The expression of CSPG4 was evaluated in HCC using The Cancer Genome Atlas (TCGA) Liver Hepatocellular Carcinoma (LIHC) dataset. Transcripts per million (TPM) values, which were indicative of gene expression profiles, were obtained for 371 HCC and 50 adjacent non-tumor samples. Clinical data associated with these samples were also retrieved. Differential expression analysis and visualization were performed using R programming language (version 4.2.1) and relevant packages, including DESeq2 for normalization and differential expression testing, and ggplot2 for generating expression plots.

Study population
This study included two independent cohorts of HCC patients who underwent curative hepatectomy. The training cohort consisted of 153 patients from Eastern Hepatobiliary Surgery Hospital (Shanghai, China), treated between March 2013 and September 2016. The validation cohort included 112 patients from Changhai Hospital (Shanghai, China), treated between February 2017 and August 2021. All patients in both cohorts underwent R0 resection, defined as complete tumor removal with microscopically clear margins. HCC diagnosis was confirmed through histopathological evaluation of the resected specimens. Exclusion criteria were: (a) presence of portal vein or hepatic vein tumor thrombus; (b) any pre-operative anti-tumor therapy; (c) death within one month after surgery or incomplete follow-up data. The study followed the REMARK guidelines and obtained ethical approval from the institutional review boards of both hospitals. All patients provided written informed consent for the use of their tissue samples and clinical data, in accordance with the Declaration of Helsinki.

Clinical characteristics and follow-up
Comprehensive patient data, including demographic and tumor characteristics such as age, gender, hepatitis B envelope antigen (HBeAg) positivity, preoperative AFP levels, maximum tumor diameter, tumor number, tumor differentiation, microvascular invasion (MVI) status, Child-Pugh grade, and TNM stage were collected and summarized in Supplementary Table 1. Although several variables differed between the two cohorts, these differences reflected real-world clinical heterogeneity rather than systematic bias, supporting their suitability and overall comparability for subsequent validation analyses. The primary endpoints were overall survival (OS) and progression-free survival (PFS). OS was defined as the time from surgery date to death or last follow-up, while PFS was the interval between surgery and first recorded recurrence or disease progression. Follow-up occurred every 2–3 months, including serum AFP levels, hepatic function tests, abdominal ultrasound, and enhanced CT or MRI. Other examinations like chest CT or PET-CT were performed if metastasis was suspected.

Cellular experiments and RNA interference
Huh-7 and HCCLM3, two human HCC cell lines, were acquired from the Chinese Academy of Sciences’ Cell Bank (Shanghai, China). These cells were maintained in DMEM (Gibco) with 10% fetal bovine serum (FBS) at 37 °C in a 5% CO2 incubator. CSPG4-specific small interfering RNAs (si-CSPG4) and non-targeting controls (si-NC) were purchased from Sangon Biotech (Shanghai, China) with sequences: si-CSPG4−1: 5’-GGAAGACCUCAGUGUCAAUTT-3’; si-CSPG4−2: 5’-CCGAGACCACAGAAGAUGATT-3’; si-CSPG4−3: 5’-ACCUGUCGGUGGA.
GACCAATT-3’; si-NC: 5’-UUCUCCGAACGUGUCACGUdTdT-3’. Transfection of HCC cells with siRNAs was performed using Lipofectamine 2000 (Invitrogen, CA, USA) as per the manufacturer’s protocol. Cells were collected 48 h post-transfection for subsequent analyses. Recombined CSPG4 (rCSPG4) protein were purchased from UA Bioscience (UA010060).

Cell viability, migration and invasion assays
Cell viability was assessed using the CCK-8 assay. HCC cells, either transfected with siRNA or treated with recombinant CSPG4 (0, 20, 50, 100 ng/ml), were seeded in 96-well plates. Optical density at 450 nm was measured at 0, 24, 48, 72, 96, and 120 h after incubation with CCK-8 solution for 2 h at 37 °C. For wound healing assays, HCC cells were grown to 90% confluence in 6-well plates, and a scratch was created using a 200 µl pipette tip. After washing with PBS, serum-free medium was added, and wound closure was photographed at 0 and 48 h to calculate the closure rate. Cell migration and invasion were evaluated using 8.0 μm transwell inserts. Transfected cells in serum-free medium were added to the upper compartment, while 10% FBS-supplemented medium was added to the lower compartment. After incubation, non-migrating cells were removed, and invasive/migrating cells were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet, and counted in five random fields.

Construction of tissue microarray (TMA) and immunohistochemistry (IHC) staining
HCC tissue microarrays (TMAs) were constructed using paraffin-embedded samples from 265 HCC patients across both cohorts (153 from cohort 1 and 112 from cohort 2). Section (4 µm thickness) underwent IHC staining following a standardized protocol, including antigen retrieval and blocking steps [15]. Anti-CSPG4 primary antibody (1:200 dilution; rabbit polyclonal, 55027-1-AP; Proteintech) was applied overnight at 4 °C, followed by incubation with HRP-conjugated secondary antibody (ZSGB-BIO, China) for 30 minutes at room temperature. Visualization used 3,3’-diaminobenzidine with hematoxylin counterstaining.
Two pathologists independently evaluated the staining while blinded to clinical data. Interobserver agreement was quantified using the kappa coefficient based on 50 randomly selected slides, and any discrepancies were resolved by consensus. For each specimen, analysis was conducted on three randomly identified representative regions. The proportion of positively stained tumor cells was assessed on a scale of 0–100%. Staining intensity was categorized using a four-tier system: 0 (negative), 1 (weakly positive), 2 (moderately positive), and 3 (strongly positive). CSPG4 expression was quantified using the H-score method [16], calculated as ∑ [intensity (1, 2, 3) × percentage of positive cells (0–100%)], resulting in a range of 0–300.

Western blotting
Protein extracts from siRNA-transfected HCC cells were quantified via BCA assay. Equal amounts of protein underwent SDS-PAGE separation and transfer to PVDF membranes (Millipore, USA). Blots were probed with anti-CSPG4 (1:1000 dilution, 55027-1-AP, Proteintech, China) and anti-Hsp90 (loading control; 1:2000 dilution, rabbit monoclonal, HA750090, HUABIO, China) antibodies overnight at 4 °C, followed by appropriate secondary antibodies for 1 h at room temperature. Protein signals were detected using enhanced chemiluminescence and visualized with an Odyssey infrared imaging system (LI-COR Biosciences, Lincoln, NE, USA).

Statistical analysis
Data are presented as mean ± SD for continuous variables and frequencies for categorical variables. Group comparisons utilized Student’s t-test, Wilcoxon test, χ2 test, or Fisher’s exact test as appropriate. Kaplan-Meier method with log-rank test assessed survival outcomes. Cox proportional hazards models identified independent prognostic factors. The optimal CSPG4 H-score cut-off value was determined via time-dependent receiver operating characteristic (ROC) analysis (R package timeROC v0.4) in the training cohort (5-year OS as endpoint, accounting for censored data), using the Youden index (Sensitivity + Specificity-1) to maximize discriminatory power. Predictive performance was evaluated via Harrell’s C-index and area under the ROC curve (AUC), while the prognostic model weighted CSPG4, tumor size, and MVI by their multivariate Cox coefficients. All analyses were conducted using SPSS v21.0 and R v3.5.3 (http://www.r-project.org/), with P < 0.05 considered significant.

Results

CSPG4 is up-regulated in HCC
Firstly, the expression pattern and clinical significance of CSPG4 in HCC was determined using public dataset and clinical HCC cohorts. Analysis revealed markedly elevated CSPG4 transcript levels in HCC samples (n = 371) relative to peritumoral tissues (n = 50) (Fig. 1A). This differential expression pattern was further corroborated through examination of 50 matched tumor-adjacent tissue pairs (Fig. 1B). IHC staining for CSPG4 on TMAs from two independent HCC cohorts validated these findings at the protein level. CSPG4 protein levels were significantly elevated in most HCC cases compared to corresponding adjacent liver tissues (Fig. 1C, D). These results indicate that CSPG4 is commonly elevated in human HCC cases.

CSPG4 knockdown inhibits malignant biological phenotypes of HCC cells
To further elucidate the potential biological functions of CSPG4 in HCC, CSPG4 knockdown was achieved in HCC cell lines using siRNA technology. Western blot analysis confirmed the efficacy of CSPG4 silencing in Huh-7 and HCCLM3 hepatocellular carcinoma cell lines following siRNA transfection (Fig. 2A). Subsequent proliferation assays utilizing CCK8 methodology revealed that CSPG4 suppression significantly impaired the growth rates of both cell lines compared to their counterparts treated with non-targeting control siRNA (si-NC) (Fig. 2B). Conversely, treatment with rCSPG4 protein promoted cell proliferation in a dose-dependent manner, with maximal effects observed at 100 ng/ml concentration (Fig. 2C), further supporting CSPG4’s role in promoting HCC cell proliferation. To evaluate the impact of CSPG4 on cell migration, wound healing assays were performed. The results demonstrated that CSPG4 knockdown significantly reduced the wound closure ability of both Huh-7 and HCCLM3 cells compared to control groups at 48 h post-wounding (Fig. 2D). Additionally, migration and invasion assays revealed that CSPG4 interference suppressed the migration and invasion abilities of HCC cells (Fig. 2E, F). Collectively, these in vitro findings suggest a pivotal role for CSPG4 in enhancing cellular proliferation and metastatic potential, warranting further validation through in vivo studies to fully elucidate its significance.

CSPG4 overexpression associates with aggressive HCC phenotype and unfavorable patient outcomes
An investigation into the association between CSPG4 expression and HCC clinicopathological profiles was conducted. Utilizing time-dependent ROC analysis on the H-score derived from IHC staining in the primary cohort, an optimal CSPG4 expression threshold for predicting 5-year OS was established at 100 (H-score), yielding an AUC of 0.745 (Fig. 3A). Based on this cut-off, patients across both cohorts were stratified into CSPG4-high and CSPG4-low groups. As detailed in Table 1 and Supplementary Table 2, the CSPG4-high group exhibited more aggressive disease characteristics, including advanced tumor stage, increased tumor dimensions, elevated serum AFP levels, and higher MVI incidence. Survival analyses using Kaplan-Meier methodology demonstrated that patients with high CSPG4 expression had significantly poorer OS (P < 0.001) and PFS (P < 0.001) compared to the CSPG4-low group. This trend was consistent across both the primary (Fig. 3B) and validation cohorts (Fig. 3C). To further validate the prognostic value of CSPG4 in HCC, data from the public database TCGA were analyzed. Analysis revealed a strong correlation between heightened CSPG4 levels and diminished patient outcomes, including OS, DSS, and PFS (Fig. 3D-F). These findings suggest that CSPG4 may function as a valuable indicator for clinicopathological characteristics and prognosis of HCC patients, highlighting its potential as a prognostic biomarker in HCC.

Integrating CSPG4 with clinical indicators improves prognostic accuracy for postoperative HCC patients
Further assessment of CSPG4’s prognostic significance in HCC involved comprehensive statistical analyses in the primary study group. Univariate and multivariate examinations revealed CSPG4 expression, tumor size, and MVI as independent indicators of unfavorable post-surgical OS and PFS (Table 2). The validation cohort corroborated these findings (Supplementary Table 3). Subsequently, the predictive accuracy of these factors was evaluated both individually and in combination through C-index analysis within the primary cohort. As shown in Table 3, the combination of these three factors demonstrated the highest predictive efficiency for both OS and PFS. The C-index values for this panel were significantly higher than those of any individual factor or other combinations. These findings were further confirmed in the validation cohort (Supplementary Table 4). Collectively, these data indicate that integrating CSPG4 expression with established clinical indicators provides superior prognostic value for predicting postoperative survival in HCC patients.

Evaluation of the CSPG4-based prognostic model against established staging systems
The novel CSPG4-based model’s predictive capabilities were assessed in comparison to widely-adopted staging methodologies, specifically the TNM and Child-Pugh systems. Supplementary Fig. 1 illustrates the Kaplan-Meier analyses for long-term outcomes across different stages in both cohorts. The model’s discriminative power for survival prediction was evaluated using ROC curves, with AUCs demonstrating superior performance in both sets compared to conventional classifications for OS and PFS (P < 0.001, Supplementary Fig. 2). Given that approximately 30% of HCC cases are AFP-negative in clinical settings, a focused analysis was conducted on this subgroup across both cohorts. Within the 64 AFP-negative (≤ 20 ng/mL) HCC cases, elevated CSPG4 expression correlated with notably poorer OS and PFS outcomes (Supplementary Fig. 3). Thus, this model shows particular promise in stratifying prognosis for AFP-negative HCC patients, a subgroup where traditional biomarkers may be less informative. In summary, these findings highlight the enhanced prognostic accuracy achieved by integrating the tumor-specific biomarker CSPG4 with established clinical parameters. This approach offers a more refined and reliable prognostic tool compared to current single-factor models used in clinical practice for survival estimation.

Discussion
Identifying reliable molecular biomarkers is critical for improving hepatocellular HCC prognosis, but decades of research into candidate markers—from AFP to newer targets like GPC3 and PD-L1—have yielded limited clinical success [17–19]. The core challenge lies in HCC’s extreme heterogeneity with tumors differ widely in etiology, genetic drivers, and microenvironmental features, meaning no single biomarker can fully capture disease complexity or predict outcomes [20]. This limitation has left clinicians reliant on imperfect tools such as AFP, which lacks sensitivity in approximately 40% of cases [4], and TNM staging, which fails to stratify intermediate-risk patients [5]. These issues underscore the urgent need for integrated approaches that combine molecular signatures with clinical parameters. CSPG4, a cell surface proteoglycan with well-established oncogenic roles in melanoma, glioblastoma, and Squamous cell carcinoma [7, 9, 11, 14], emerges as a promising candidate for HCC, though its clinical and biological significance has remained unexplored until now.
CSPG4 may provide a notable improvement over traditional HCC biomarkers by overcoming their shortcomings. AFP, the most commonly used marker, is not sensitive in early-stage or AFP-negative HCC, which makes up about 30% of cases [4], and it doesn’t reliably indicate tumor invasiveness. In contrast, CSPG4 is consistently expressed across all HCC stages, closely linked to MVI [21], and remains a valuable prognostic indicator in AFP-negative patients. GPC3, another marker, is limited by its restricted expression in only about 70% of HCC cases and its absence in well-differentiated tumors [18]. Even PD-L1, which guides immunotherapy decisions, only forecasts treatment response [19], not initial prognosis in resectable HCC. CSPG4, in contrast, serves both as a prognostic tool and a potential therapeutic target, fulfilling a critical need for markers that work across different HCC subtypes and clinical situations, including AFP-negative HCCs.
Although direct studies on CSPG4 in HCC are lacking, research in other tumors provides insights into its potential mechanisms and supports the association between CSPG4 overexpression and poor prognosis observed in this study. CSPG4 modulates integrin and growth factor signaling pathways, promoting tumor cell proliferation, invasion, and resistance to apoptosis. In glioblastoma, CSPG4 activates the Akt–ERK axis and regulates EGR1-mediated transcription, maintaining tumor survival and chemoresistance [22]. In glioma stem cells, CSPG4 binds to integrin αV, activating the integrin–ERK pathway to sustain cell stemness and invasiveness [14]. Also, its deficiency was proven to inhibit tumor growth and affect the IGF pathway in soft tissue sarcomas [23]. Collectively, CSPG4 functions as a multifunctional coreceptor, driving tumor progression through integrin-mediated ERK activation and interactions with growth factor receptors such as PDGFR, VEGFR, and IGF. Given the roles of integrin signaling and growth factor pathways in HCC, CSPG4 likely promotes tumor invasion and angiogenesis, contributing to adverse patient outcomes. Future research should explore these mechanisms in HCC, particularly CSPG4’s interactions with key pathways like FAK/ERK and EGFR/MET.
A key advance of this study is the development of a prognostic model that integrates CSPG4 expression with tumor size and MVI—two well-established predictors of HCC recurrence [24, 25]. This model outperforms traditional TNM staging, Child-Pugh grade, and single biomarkers with a C-index of 0.811 for OS prediction. For instance, TNM staging typically yields C-indexes of 0.62–0.68 for OS in resectable HCC, whereas our model’s higher C-index reflects its ability to capture both clinical and molecular heterogeneity. This supports recent reviews highlighting that multi-parameter models are essential to overcome the limitations of single biomarkers in HCC. It also addresses a clinical need—AFP-negative HCC patients often lack reliable tools for risk stratification, leading to suboptimal treatment decisions.
This study has several limitations that should be addressed in future research. First, our cohorts are limited to resectable HCC patients treated at two Chinese centers, primarily involving HBV-related HCC. This restricts the model’s generalizability to advanced HCC and non-HBV etiologies more prevalent in Western populations [26]. Multi-center prospective studies, including international cohorts with diverse etiologies, are needed to validate the CSPG4-integrated model’s broader utility. Second, while our in vitro data confirm CSPG4 promotes HCC cell proliferation, migration, and invasion, which processes central to tumor growth and spread, we did not employ in vivo models to verify these effects. This prevents us from accounting for discrepancies between in vitro cell behavior and in vivo biology, particularly the tumor microenvironment’s influence on CSPG4 function. Third, our mechanistic investigations were preliminary. We did not explore CSPG4’s interactions with the tumor microenvironment or its cross-talk with other oncogenic pathways, which are central to HCC development [27]. Future research should elucidate these interactions, focusing on pathways already targeted by approved or investigational drugs for HCC.
In conclusion, this study unveils the crucial role of CSPG4 in HCC for the first time. Our findings highlight its significant potential not only as a valuable prognostic biomarker but also as a promising therapeutic target for HCC treatment. The combination of CSPG4 with existing clinical parameters significantly improves the accuracy of prognosis prediction, providing new insights for individualized treatment decisions in HCC. Although further research is needed to elucidate its specific mechanisms and clinical applications, CSPG4 undoubtedly could advance both prognostic assessment and therapeutic strategies for HCC.

Supplementary Information