Introduction

Liver cancer, which is the fourth-leading cause of cancer-related death1, remains a global health challenge. Its incidence is growing worldwide, with an estimated incidence of more than 1 million cases by 20252. HCC is the most common form of liver cancer and accounts for 75% to 85% of cases3. Due to the complex etiology of liver cancer, early diagnosis is rather challenging. By the time most patients receive a definitive diagnosis, the disease has already progressed to an advanced stage4. Therefore, it is essential to find more effective early diagnostic and prognostic markers to help HCC patients improve their quality of life.

The methods used to diagnose HCC consist of clinical manifestations, imaging techniques5, molecular markers6, and “Omics” diagnostics7. In the imaging workup, dynamic contrast-enhanced computed tomography and magnetic resonance imaging are deemed the gold standard diagnostic techniques for HCC, while ultrasonography has been extensively utilized as the preliminary screening examination for HCC, demonstrating sensitivities ranging from 51% to 87% and specificities from 80% to 100%8. Currently, “omics” data, such as genomics, proteomics, and metabolomics, have been utilized to detect biomarkers of HCC7. These diverse biotechnologies have significantly contributed to the detection of HCC biomarkers. Many molecular biomarkers of HCC have been identified, such as alpha-fetoprotein (AFP)9, Osteopontin (OPN)10, and glypican-3 (GPC3)11.

The use of high-throughput sequencing was first proposed in 200512. RNA sequencing (RNA-seq), a key component of high-throughput sequencing, has significantly advanced over the past decade and has become the standard technique for transcriptome research. The range of oncological biomarkers residing in tumor specimens encompasses a vast array of DNA, RNA, enzymes, metabolites, transcription factors, and cell surface receptors13. Recent high-throughput RNA sequencing has demonstrated the potential of mRNAs, microRNA (miRNA), long non-coding RNA (lncRNA), and circular RNA (circRNA) as biomarkers for various cancer types14,15,16,17. eRNAs are a class of lncRNAs transcribed from enhancer regions of DNA that help regulate the expression of neighboring protein-coding genes. For instance, PSA eRNA is transcribed from an enhancer region of the PSA gene. It can activate AR expression in cancer cells, thereby promoting tumor progression in CRPC18. The eRNAs are expected to serve as exceptionally useful biomarkers.

Recent studies have systematically documented eRNA expression in humans. The landscape of eRNA expression in cancer tissues was studied using the Cancer Genome Atlas (TCGA) data19. The analysis of eRNA expression atlases in normal human tissues was performed using the Genotype-Tissue Expression (GTEX) data20. Several researchers have identified eRNAs as potential targets for therapeutic intervention and valuable biomarkers21,22,23,24,25. Regarding HCC, Cai et al. utilized the “PreSTIGE” algorithm to identify eRNA associated with immunological genes from TCGA-LIHC (liver hepatocellular carcinoma) data and construct a prognostic model26. Wu et al. identified DCP1A as the most significant survival-associated eRNA from TCGA-LIHC data27. Their tissue-specific and cancer-type-specific expression patterns make them attractive candidates for precision oncology approaches. Additionally, recent experimental evidence has shown that targeting MYC enhancer-derived eRNAs through genetic deletion or antisense oligonucleotide inhibition effectively suppresses HCC cell proliferation and tumorigenicity28. However, comprehensive studies systematically exploring the differential expression patterns of eRNAs between normal and cancerous tissues in HCC, and their functional validation across diverse patient cohorts, remain critically needed.

In this study, we identified three eRNA biomarkers for HCC using RNA-seq data, and we investigated their functions, prognostic implications, and connections to immunological infiltration and drugs. Immunohistochemistry (IHC) and RT-qPCR was then used to confirm the results.

Materials and methods

Data collection and preprocessing

Three RNA-seq datasets were utilized in this study. The high-throughput sequencing data GSE144269 and GSE124535 were downloaded from the GEO database (Available from: https://www.ncbi.nlm.nih.gov/geo)29. The GSE144269 dataset included 140 samples from cancerous tissues and adjacent normal tissues of 70 HCC patients, and the GSE124535 dataset included 70 samples from cancerous tissues and adjacent normal tissues of 35 HCC patients. The third dataset, E-MTAB-5905, was obtained from the EMBL-EBI database (Available from: https://www.ebi.ac.uk)30. It included 62 tissues from 11 HCC patients, comprising cancerous and normal tissues. For each patient, we merged the multiple cancerous tissues belonging to them as well as the normal tissues. In the subsequent analysis, we removed a patient with only one cancerous tissue from this dataset. We ultimately obtained combined eRNA expression level data for 10 patients.

The annotation of enhancers was acquired from Ensembl (Available from: https://useast.ensembl.org/)31, FANTOM (Available from: http://fantom.gsc.riken.jp/index.html)32, and Roadmap Epigenomics (Available from: http://www.roadmapepigenomics.org/)33. These three datasets were merged, and only enhancers annotated in a minimum of two datasets were utilized. The eRNA region was defined as the ± 3 kb region surrounding the central locus of the enhancer19. The human reference genome was collected from the UCSC Genome Browser Home (Available from: http://www.genome.ucsc.edu/, hg38)34. Alignments of the raw RNA-seq data to the human genome were performed using this reference. Annotation of protein-coding genes was obtained from GENCODE (Available from: https://www.gencodegenes.org/, v39)35 and used to obtain the expression levels of protein-coding genes. We downloaded the raw sequence data of LIHC from the TCGA data portal36 and used it for the validation of eRNA biomarkers. Clinical features forTCGA-LIHC were downloaded from the University of California Santa Cruz (UCSC) Xena Browser built on TCGA data (Available from: https://xenabrowser.net ) and used for the prognostic analysis of eRNA biomarkers.

During the processing of the raw RNA-seq data, fastp software37 was used to trim the adapter, fastqc software38 was used to visualize sequence quality, and Hisat2 software39 was used to map the human reference genome (hg38) to clean sequence data. The SAMtools40 toolkit was used to convert the file format and merge data from organizations belonging to the same sample in the E-MTAB-5905 dataset. The eRNA annotation file and processed RNA-seq data were implemented as inputs in Bedtools41. The output included the raw read counts for eRNA, which were then converted to Reads Per Million mapped reads (RPM) values in R. For each sample, the total number of mapped reads was extracted from the alignment statistics files. RPM values were calculated using the formula: RPM = (raw read count × 10^6)/total mapped reads per sample. This normalization method accounts for differences in sequencing depth between samples, enabling direct comparison of eRNA expression levels across different samples. Subsequently, for each sample, the presence of eRNA was defined if its RPM value was ≥ 1. StringTie was used to derive raw read counts for protein-coding genes from the processed RNA-seq files. The Ballgown package in R was then used to convert the raw read counts into FPKM values42. A total of 19 982 protein-coding genes were screened in these datasets.

Research process

After we obtained the RNA-Seq datasets, eRNA annotation, and gene annotation, we obtained the eRNA and gene expression in three datasets. Subsequently, we identified eRNA biomarkers for HCC patients from the eRNA expression matrices and conducted survival analysis on these eRNA biomarkers using LIHC data from TCGA as validation. Correlation analysis between these eRNA biomarkers and protein-coding genes was conducted to identify protein-coding genes associated with eRNA biomarkers (PCGAeRs). Function analysis was then performed for PCGAeRs. The host genes of these eRNA biomarkers were uncovered and subjected to pan-cancer and immune infiltration analysis. Drug-target information for common chemotherapy drugs in HCC was collected from the HCDT database and integrated with the eRNA-host gene relationships. The flowchart of the entire study is shown in Fig. 1.

Fig. 1
figure 1

The flowchart of this study.

Full size image

Screening for eRNA biomarkers and the host genes

For each eRNA in each dataset, we used McNemar’s Test (formula 1) for tumor and normal tissues to identify potential biomarkers. P value < 0.05 was considered statistically significant. The categorization criteria and the corresponding formula are presented in Table 1.

Table 1 Categorization of mcnemar’s test for paired samples.
Full size table
$$\:{\chi}^{2}=\frac{{\left(B-C\right)}^{2}}{B+C}$$
(1)

Due to the limitations of McNemar’s test, it cannot cover all the data in the table as A and D were not used. B or C should be large for a useful biomarker, but A and D should be small. Therefore, we determined an eRNA to be a biomarker using three criteria: It had to be present in all three datasets, have a sample count in group B or C which was at least 60% of the overall sample count, and have a significant McNemar’s test p-value. As a result, seven eRNAs were screened as potential biomarkers (one present in tumor tissue and six in normal tissue). Several recent investigations have shown that eRNAs can function as cis-acting elements24,43,44,45,46. We predicted the host gene of the eRNA biomarkers based on the closest distance using BEDTools and ultimately discovered three host genes. The specific patterns of eRNA biomarkers were validated within the TCGA-LIHC dataset.

Correlation analysis of eRNA biomarkers with protein-coding genes

A Spearman correlation analysis was performed between the RPM values of these three eRNA biomarkers (CAP2e, COLEC10e, and MARCOe) and the FPKM values of protein-coding genes. For statistical significance, p < 0.05 was chosen. Protein-coding genes were classified into positively and negatively correlated genes based on the Spearman correlation coefficient with the eRNA. The top 1% of correlated protein-coding genes were defined as PCGAeRs. To construct the protein-protein interaction network, the identified PCGAeRs were submitted to the STRING47 database, the organisms were set as Homo sapiens, and the default parameters were used. Export in TSV format from STRING and then import it into Cytoscape (version 3.10.0) for visualization and analysis.

Gene ontology and KEGG pathway enrichment analysis

To identify the biological functions of PCGAeRs in HCC, we performed functional enrichment analysis for gene ontology and KEGG using DAVID48. The significance threshold was set at a p-value < 0.05. The results were visualized utilizing the “ggplot2” (v 3.3.6) R package.

Prognostic analysis of eRNA biomarkers in HCC patients

Survival analysis was performed using the GSE144269 and TCGA-LIHC datasets. The samples in the GSE144269 and TCGA-LIHC datasets that lacked clinical information were filtered, and then the samples were split into two groups based on the best cut-off value for eRNA biomarker expression. To determine the cut-off value, the receiver operating characteristic (ROC) was constructed, and the cut-off value was chosen based on the highest sensitivity and smallest distance from the cut-off value to the top left corner of the ROC curve. Kaplan-Meier survival curves were created using the “survival” (v 3.3.1) R package to discern potential associations between eRNA biomarkers and the prognosis of HCC. The statistically significant threshold was set at a log-rank p-value < 0.05.

Association of host gene with therapeutic drugs

We collected therapeutic drugs for the treatment of HCC, which were obtained from several papers available on PubMed. The Highly Confident Drug-Target Resource (HCDT, available from: http://hainmu-biobigdata.com/hcdt/) was a combined database for drug-target interactions49. We used HCDT to identify the target genes for these therapeutic drugs. KEGG pathway enrichment in DAVID was used to analyze these target genes, and then we speculated on the eRNA-gene-pathway-drug relationships.

Pan-cancer and immune infiltration analysis

The correlation between eRNA biomarkers and the expression levels of immune-infiltrating cells was explored in HCC, and the expression levels of eRNA biomarkers were examined in other cancers in the TCGA dataset. TIMER 2.0 (Available from: http://timer.cistrome.org) was utilized to investigate the expression landscape of host genes for eRNA biomarkers across various cancers in the TCGA dataset and their infiltration level in various types of immune cells50.

IHC staining evaluation

Tissue microarrays for liver cancer tissues and normal adjacent tissues to the tumor (NAT) were procured from Wuhan Shuangxuan Biotechnology Co., Ltd., with product codes IWLT-N-60LV61 and IWLT-N-73LV31. These arrays included 63 cases of human HCC and their paired NATs. All specimens were fixed in 10% neutral buffered formalin and embedded in paraffin. Preoperative samples were not subjected to any neoadjuvant therapy. The thickness of each microarray section was 4 μm, with a core diameter of 2 mm. This study was approved by the Ethics Committee of Hainan Medical College.

IHC staining was performed using the ultra-sensitive S-P kit (mouse/rabbit general, KIT-9710, Fuzhou MaXim Biotechnology Co., Ltd., China). The primary antibody used was a MARCO rabbit anti-human/rat polyclonal antibody (ab231046; Abcam, Cambridge, MA, USA) at a working concentration of 1:100. Paraffin-embedded tissue sections were routinely deparaffinized and rehydrated, with subsequent steps conducted according to the kit’s instructions. It included incubation with the primary antibody at 37 ℃ for 1 h and antigen retrieval using citrate buffer solution (pH 6.0). After DAB staining, the cell nuclei were counterstained with hematoxylin.

The IHC staining results were independently assessed by two observers who were blind to the clinical information of the patients. The observers were unaware of the anticipated experimental outcomes and reached a consensus in case of any disagreement. In this study, MARCO expression was localized in the cytoplasm and exhibited relatively uniform staining intensity within the same core. The positive intensity was graded into four categories: 0 = no staining, 1 = light yellow, 2 = yellow, and 3 = yellow-brown. MARCO expression ≤ 1 was defined as absent, and > 1 as present. During the assessment, it was noted that two liver cancer tissue cores and one adjacent tissue core were missing and were excluded from the statistical analysis.

Cell culture and transfection

HepG2 cells were cultured in high-glucose DMEM medium (containing L-glutamine, without sodium pyruvate and HEPES; C11965500BT) supplemented with low-endotoxin fetal bovine serum (catalog no. 11011 − 8611) and penicillin-streptomycin (100×; BL505A). Cells were seeded in 6-well plates and cultured for 12 h, followed by serum starvation for 12 h.

For Marco gene overexpression, cells were divided into two groups: control and overexpression groups. The Marco overexpression plasmid (GenePharma, Jiangsu, China; 339 ng/µL) was prepared for transfection using Lipofectamine 3000 (L3000-001). In a 15 mL tube, 24 µL Lipofectamine 3000 was mixed with 3 mL serum-free DMEM. In a separate tube, 90 µL plasmid was diluted in 3 mL serum-free DMEM. After 5 min incubation at room temperature, the two solutions were combined and incubated for an additional 20 min at room temperature. The culture medium was removed and replaced with 1 mL of the transfection mixture per well. Six hours post-transfection, the medium was replaced with complete culture medium. Total RNA was extracted 24 h after transfection for subsequent experiments.

RNA extraction and reverse transcription

Total RNA was extracted using Trizol reagent. Culture medium was removed from 6-well plates and cells were washed with PBS. Five hundred microliters of Trizol was added to each well, and cells were scraped and transferred to RNase-free 1.5 mL microcentrifuge tubes. After 20 min incubation at room temperature, 100 µL chloroform was added, and the mixture was vigorously shaken and incubated for 15 min at room temperature. Samples were centrifuged at 12,000 × g for 15 min at 4 °C. The upper aqueous phase was collected and mixed with 250 µL isopropanol by inversion. Following 10 min incubation at room temperature, samples were centrifuged at 12,000 × g for 15 min at 4 °C. The supernatant was discarded, and the RNA pellet was washed twice with 1 mL 75% ethanol, with vortexing and centrifugation at 7,500 × g for 5 min at 4 °C for each wash. After removing the ethanol, the RNA pellet was air-dried and dissolved in 20 µL DEPC-treated water. RNA concentration was determined using a NanoDrop One spectrophotometer.

RT-qPCR

We identified the intersection genes between chemotherapy drug target genes and MAPK signal pathway genes, calculated Spearman correlation coefficients between these genes and MARCOe, and selected the top 10 genes with the highest correlation for subsequent qPCR validation experiments.

For reverse transcription, a 20µL reaction system was prepared for each sample containing 5µL RT Master Mix for qPCR (MCE, HY-K051A), 1µL total RNA, and 14µL DEPC-treated water. The reverse transcription was performed with the following thermal cycling conditions: 25 °C for 5 min, 55 °C for 15 min, and 85 °C for 2 min.

Quantitative real-time PCR was performed after dissolving primers according to the manufacturer’s instructions. The 20µL PCR reaction mixture consisted of 10µL SYBR Green Master Mix, 1µL forward primer (10µmol/L), 1µL reverse primer (10µmol/L), 1µL cDNA template, and 7µL DEPC-treated water. The PCR amplification was carried out with the following protocol: initial denaturation at 95 °C for 5–10 min, followed by 40 cycles of denaturation at 95 °C for 15 s, annealing at 50–60 °C for 30 s, and extension at 72 °C for 30 s. Melt curve analysis was performed with one cycle at 95 °C for 15 s, 60 °C for 60 s, and 95 °C for 15 s. The reaction volume ranged from 10µL to 50µL. The 2-ΔΔCq method was used for relative quantification.

Ethics declaration

This study was conducted in accordance with the Declaration of Helsinki and relevant ethical guidelines and regulations, with approval from The Ethics Committee of Hainan Medical University (approval number: HYLL-2023-473). In this study, 63 samples of human HCC and their paired NATs were utilized. The samples were obtained from commercially available tissue microarrays (Wuhan Shuangxuan Biotechnology Co., Ltd., product codes IWLT-N-60LV61 and IWLT-N-73LV31). The informed consent had been obtained from all participants by the tissue supplier.

Results

Identification of eRNA biomarkers in HCC

By applying McNemar’s Test, seven eRNAs were identified as biomarkers (Fig. 2A and B), and their host genes were then identified. Due to the existence of multiple eRNAs corresponding to one host gene, a filtration was performed to select the most important eRNA (tag eRNA) in a gene. In this procedure, the classification accuracy of the eRNA and eRNA combination was used to measure the significance of eRNAs (Doc.S1). As a result, three tag eRNAs were identified. ENSR00000194159 exhibited exclusive expression in tumor tissue, while ENSR00000229373 and ENSR00000122322 expressed in normal tissue. The corresponding host genes were CAP2, COLEC10, and MARCO (Table 2).

These findings were confirmed in the TCGA-LIHC dataset (Fig. 2C-E). The expression levels of ENSR00000194159 were significantly upregulated in tumor tissues compared to normal tissues (p < 0.05, Fig. 2C). In contrast, the expression levels of ENSR00000229373 and ENSR00000122322 were significantly reduced in tumor tissues (p < 0.05, Fig. 2D and E).

Table 2 Tag eRNA biomarkers and the host genes.
Full size table
Fig. 2
figure 2

Distribution of eRNA biomarkers. (A) Tumor-specific eRNA in the multi-datasets. (B) Normal-specific eRNA in the multi-datasets; the graph values indicate the number of eRNAs that satisfy the screening criteria. (C) The expression levels of ENSR00000194159 in tumor tissues in the LIHC dataset were significantly higher than in normal tissues (P < 0.05). (D) The expression level of ENSR00000229373 in tumor tissues is significantly lower than that in normal tissues (P < 0.05). (E) The expression level of ENSR00000122322 in tumor tissues is significantly lower than that in normal tissues (P < 0.05).

Full size image

Strong correlation between the eRNA biomarker and the host gene

In these three datasets, we utilized Spearman correlation analysis to investigate the relationship between the eRNA biomarkers and their host genes. The results demonstrated a strong positive correlation, with Spearman’s rank correlation coefficient (rho) exceeding 0.8 (p < 0.05, Fig. 3A-I), between the expression levels of ENSR00000194159 (CAP2e) and CAP2, ENSR00000229373 (COLEC10e) and COLEC10, along with ENSR00000122322 (MARCOe) and MARCO. A similar correlation is observed between eRNA and host gene in the TCGA-LIHC dataset (Fig. S1). To further validate the differential expression patterns of these eRNA biomarkers, we generated Manhattan plots displaying the genome-wide distribution of eRNA differential expression between tumor and non-tumor adjacent tissues across the three datasets (Fig. S2 A-C). The Manhattan plots revealed distinct peaks corresponding to the specific eRNAs, demonstrating their significant differential expression in HCC tissues compared to adjacent non-tumor tissues. These results emphasize significant consistency between the transcriptional activities of the eRNAs and their host genes, indicating potential interconnected functional roles within the context of HCC.

Fig. 3
figure 3

Correlations of expression levels between eRNA biomarkers and host genes across multiple datasets. (A-C) Correlation between eRNA biomarkers and host genes in the E-MTAB-5905 dataset; the red line is the fitted curve and “R” is the correlation coefficient. (D-F) Correlation between eRNA biomarkers and host genes in the GSE124535 dataset. (G-I) Correlation between eRNA biomarkers and host genes in the GSE144269 dataset.

Full size image

Multiple cancer-related signaling pathways are linked to the eRNA biomarkers

To further explore the potential functions of eRNA biomarkers, gene ontology and KEGG pathway enrichment analyses of PCGAeRs were conducted using DAVID, followed by visualization using R Studio. Summary statistics for PCGAeRs are presented in Table 3 (The list of genes shown in the Supplemental Table S1, S2 and S3). The results of pathway enrichment analysis indicated significant enrichment in the cell cycle and ribosome biogenesis across all three eRNA biomarkers. CAP2e exhibited significant enrichment in several cancer-related pathways, including the Hippo signaling pathway, glycosylphosphatidylinositol (GPI)-anchor biosynthesis, and ubiquitin-mediated proteolysis (Fig. 4A). Similarly, COLEC10e exhibited significant enrichment in several pathways, including the spliceosome, base excision repair, and N-Glycan biosynthesis (Fig. 4B). Finally, MARCOe exhibited significant enrichment in the basal transcription factors, Base excision repair, and RNA polymerase pathways (Fig. 4C). Additional details regarding the functional analysis are discussed in Fig. S3. To further elucidate the molecular interactions of PCGAeRs, we constructed protein-protein interaction (PPI) networks for each eRNA biomarker using the STRING database (Fig. S4).Previous studies have shown that hepatitis viral infection, which is the primary promoter of HCC, interacts with ribosomes and plays a major role in the development and advancement of HCC51.

Table 3 Number of protein-coding genes associated with eRNA biomarkers in the datasets.
Full size table
Fig. 4
figure 4

KEGG pathway enrichment analysis of PCGAeR. (A) KEGG pathway enrichment analysis of protein-coding genes associated with CAP2e. (B) KEGG pathway enrichment analysis of protein-coding genes associated with COLEC10e. (C) KEGG pathway enrichment analysis of protein-coding genes associated with MARCOe.

Full size image

eRNAs are potential independent prognostic biomarkers

To evaluate the potential prognostic prediction of eRNA biomarkers, a survival analysis was performed using the clinical data from the TCGA and GSE144269 datasets. Kaplan-Meier analysis revealed that heightened expression of CAP2e exhibited a significant correlation with unfavorable prognosis in both datasets (p < 0.05, Fig. 5A and D), while decreased expression of COLEC10e was associated with shorter overall survival (p < 0.01, Fig. 5B and E). Reduced expression of MARCOe was linked to a poor prognosis (p < 0.05, Fig. 5C) in the GEO dataset, but no significant association was observed in the TCGA dataset (Fig. 5F). These findings suggest that eRNA biomarkers hold promise as potential prognostic indicators for HCC.

Fig. 5
figure 5

Kaplan-Meier survival analysis was performed on eRNA biomarkers in both the GSE144269 and TCGA-LIHC datasets. (A) The high expression of eRNA CAP2e is associated with a poor prognosis in dataset GSE144269. (B) The low expression of eRNA COLEC10e is associated with a poor prognosis in dataset GSE144269. (C) The low expression of eRNA MARCOe is associated with a poor prognosis in dataset GSE144269. (D) The high expression of eRNA CAP2e is associated with a poor prognosis in TCGA. (E) The low expression of eRNA COLEC10e is associated with a poor prognosis in TCGA. (F) The expression status of eRNA COLEC10e in relation to prognosis in TCGA.

Full size image

MARCOe is a potential HCC therapeutic drug target

The current therapeutic drugs used in the treatment of HCC include Lenvatinib, Sorafenib, Nivolumab, Cabozantinib, Atezolizumab, Bevacizumab, Imatinib, and others52,53,54,55,56. We obtained their target genes from the HCDT database (Table 4). Then, we conducted pathway enrichment analysis for the top four medications in terms of the amounts of targets (see more details in Fig. S5). The results revealed a significant enrichment of all four drugs in the MAPK pathway (FDR < 0.0001). Meanwhile, existing evidence has demonstrated a positive correlation between MARCO expression and the MAPK pathway57. Therefore, MARCOe may represent a potential target for therapeutic drugs in HCC.

Table 4 Statistics of drug-target number in the HCDT database.
Full size table

Relationship between MARCO expression and immune infiltration

MARCO is an immune-associated gene available in the ImmPort Portal database58 (Available from: https://www.immport.org/home). We utilized Timer 2.0 to examine the association between MARCO expression and immune cell infiltration. MARCO expression was found to have a negative correlation with tumor purity (r = −0.448, p = 1.90e-18, Fig. 6). Additionally, MARCO expression was positively associated with the infiltration level of several immune cells, including CD8 + T cells (r = 0.286, p = 6.73e-08), neutrophils (r = 0.245, p = 4.27e-06), macrophages (r = 0.262, p = 8.23e-07), and myeloid dendritic cells (r = 0.317, p = 1.63e-09, Fig. 6). Immune infiltration analysis for other eRNA biomarkers is described in the Fig. S6.

Fig. 6
figure 6

Relationship between the expression level of MARCO and immune cell infiltration.

Full size image

Pan-cancer analysis of eRNA biomarkers

To investigate the expression patterns of the eRNA biomarker host genes in other cancer types, we performed a pan-cancer expression analysis for CAP2, COLEC10, and MARCO. The findings revealed that these genes exhibited the same pattern in some other cancer types, e.g., CAP2e exhibited a similar expression pattern in CHOL (cholangiocarcinoma) and KIRC (kidney renal clear cell carcinoma) patients as was observed in HCC patients, displaying an upregulated expression in tumor tissues (Fig. 7A). Similarly, both COLEC10 and MARCO displayed a comparable expression pattern in CHOL, LUSC (lung squamous cell carcinoma), and LUAD (lung adenocarcinoma) patients as that observed in HCC patients, with downregulated expression in tumor tissues (Fig. 7B and C). The shared expression pattern of eRNA biomarkers in CHOL and HCC may potentially be attributed to the differentiation of liver stem cells, which give rise to both hepatocytes and cholangiocytes. Additionally, the phenotypic overlap between LIHC and CHOL has been acknowledged as a continuous spectrum within liver cancer59.

Fig. 7
figure 7

The landscape of host gene expression levels in pan-cancer. (A-C) The landscape of CAP2, COLEC10, and MARCO expression levels in pan-cancer. Red, blue, and purple represent tumor, normal, and metastatic tissues, respectively.

Full size image

Literature validation of eRNA biomarkers

Our study revealed the significant upregulation of CAP2e and its host gene CAP2 in HCC tissues compared to normal tissues. It aligns with Shibata et al.’s research, where they observed elevated levels of CAP2 mRNA and protein in HCC tissues60. Yoon et al. demonstrated that endoplasmic reticulum stress induces CAP2 overexpression, leading to the activation of Rac1 and ERK, which promote epithelial-mesenchymal transition (EMT) and enhance the migration and invasion of HCC cells61.

In this study, COLEC10e exhibited significant downregulation in HCC patients and demonstrated significant prognostic value in survival analysis. Ju et al. identified COLEC10 as an immune-related gene (IRG) associated with the tumor microenvironment (TME), and its expression correlated with overall survival (OS) in HCC patients62. Similarly, Zhang et al. discovered that low expression of the COLEC10 encoded protein was significantly associated with vascular invasion and peripheral invasion in HCC patients, establishing COLEC10 as a potential predictive marker for HCC prognosis63.

MARCOe exhibited similar biological characteristics to COLEC10e in our study. Previous research has indicated the crucial role of MARCO in the uptake and clearance of tumor cells by macrophages64. Dong et al. analyzed the correlation between MARCO, immune infiltration, and prognosis in pan-cancers using the data from TCGA and GTEX65. However, studies focusing on MARCO in HCC patients remain limited.

Biological experiment validation of MARCO

To validate the expression pattern of MARCO, we analyzed the expression levels of MARCO using IHC in an external cohort of HCC paired samples. The IHC staining scores showed that MARCO protein expression is typically absent in tumor tissues (observed in 51 out of 60 samples). However, it is commonly present in NAT (observed in 39 out of 60 samples, chi-square test p-value = 6.523e-08, McNemar’s test p-value = 2.951e-07, see Tables 5 and 6). An example was shown in Fig. 8. MARCO protein is present in NAT but absent in tumor tissue (Fig. 8A). In an IHC staining image, a clear difference in MARCO protein expression between NAT and tumor tissue is evident (Fig. 8B).

To further investigate the functional significance of MARCO in HCC, we performed MARCO overexpression experiments in HCC cells. Upon MARCO overexpression, we examined the expression changes of chemotherapy drug target genes within the MAPK pathway. The results revealed that key genes such as PDGFB and BRAF exhibited significant differences between the MARCO overexpression group and the control group (Fig. 8C, Fig. S7). These findings suggest that MARCO agonists may activate the MAPK pathway through upregulating MARCO expression, thereby suppressing cancer cell proliferation and differentiation (Fig. 8D).

Table 5 Distribution of IHC staining status in tumors and NATs.
Full size table
Table 6 McNemar’s test for MARCO expression in tumors and NATs.
Full size table
Fig. 8
figure 8

Representative IHC Staining Images of MARCO. (A). IHC images of MARCO protein in HCC tissue and its paired NAT. The scale bars were 500 μm and 50 μm. (B). IHC image of MARCO protein in a liver tissue including both HCC tissue and NAT. The scale bar was 100 μm. (C) Alterations in the expression of chemotherapy drug target genes following MARCO overexpression relative to the control group (n = 6 biological replicates per condition). (D) Hierarchical regulatory network illustrating MARCO-mediated regulation of the MAPK signaling pathway, constructed based on KEGG database (https://www.kegg.jp/entry/map04010)66,67,68.

Full size image

Discussion

Abnormal eRNA expression has been closely linked to cancer development in various cancers, as numerous studies have shown69,70. Few researchers have examined eRNAs associated with HCC against normal and cancerous tissues. This study identified three new eRNA biomarkers (CAP2e, COLEC10e, and MARCOe), exhibiting differential expression in HCC tumor tissues relative to normal liver tissues. The results suggest the potential value of these eRNAs as diagnostic and prognostic biomarkers as well as therapeutic targets for HCC.

Our research revealed that tumors tissues showed significantly higher levels of CAP2e than normal tissues, while COLEC10e and MARCOe expression were noticeably lower in tumor tissues compared to normal tissues. The target genes for these biomarkers have been predicted. Subsequently, a correlation analysis was conducted using the expression values of the target genes and eRNA biomarkers; this confirmed the initial prediction as there was a statistically significant positive correlation between them.

Pathway analysis of PCGAeRs showed enrichment in cancer-related pathways for all three eRNAs, emphasizing their broad relevance in HCC pathogenesis. Validation of the prognostic value of CAP2e and COLEC10e supported their clinical usefulness as independent biomarkers.

The pan-cancer analysis also exhibited common expression trends for these eRNAs in cholangiocarcinoma, suggesting their involvement in other types of cancer. As CAP2 and COLEC10 have been validated through biological experiments, we validated MARCO protein via an IHC staining experiment. The results suggest that MARCO could serve as a significant biomarker for HCC. It provides opportunities for further exploration into the role of MARCO in cancer diagnosis and treatment.

This study demonstrated that MARCO exhibited significant prognostic value in our cohort, whereas its prognostic relevance was weaker in the TCGA-LIHC dataset. This discrepancy may be attributed to differences in population characteristics and etiological factors between the two datasets. Our cohort primarily consists of East Asian patients with HBV- and HCV-related HCC, while the TCGA-LIHC dataset is predominantly composed of Caucasian patients with non-alcoholic steatohepatitis (NASH)-related HCC. As MARCO functions as a macrophage scavenger receptor that may be specifically activated during viral infection, the tumor microenvironment associated with viral hepatitis may enhance the prognostic value of MARCO. These population-based and etiological differences may account for the inconsistent prognostic significance of MARCO across different cohorts.

Building upon these observations, the present study further demonstrated that MARCO is highly expressed in normal liver tissue. We investigated the regulatory role of MARCO through overexpression experiments in HCC cell lines. However, this study has not yet explored whether inhibiting MARCO expression using monoclonal antibodies or small-molecule inhibitors might promote HCC cell invasion and metastasis or accelerate disease progression. Therefore, future studies could focus on evaluating the potential risks of MARCO inhibitors to provide a more comprehensive theoretical basis for improving the prognosis of HCC patients.

Despite these promising findings, we acknowledge that this study primarily employed bioinformatics analysis methods, supplemented with limited in vitro experiments, to identify biomarkers, lacking large-scale prospective clinical cohort validation and in-depth in vivo functional mechanistic studies. Additionally, the data were predominantly sourced from public databases, which may introduce limitations such as limited sample size, selection bias, and missing critical clinical information. Moreover, the heterogeneity in sequencing platforms, experimental protocols, and bioinformatics analysis pipelines across different datasets may introduce unexpected batch effects.

More specifically, our biological function analysis was primarily conducted at the protein level. Due to the limitations of technical means, there is a lack of direct experimental evidence at the RNA level. Despite the fact that this method is unable to directly measure the expression of eRNA, it provides valuable evidence for the functional consequences of the identified eRNA biomarkers on their target proteins. In consideration of the established nonlinear relationship between eRNA and the expression of its regulated proteins71, this incomplete validation strategy is considered to represent a significant limitation.

Furthermore, we obtained sequencing data for control samples of tumor tissues and normal tissues in HCC, however, the accompanying dataset is deficient in detailed clinical staging information and etiological data. It is not feasible to systematically assess the progression stage of target eRNA expression and etiological changes in different HCC subtypes and disease stages. In light of the aforementioned population-specific differences in MARCO’s prognostic value and the technical limitations described above, future validation studies should incorporate larger-scale multicenter patient cohorts with comprehensive clinical metadata (including viral infection status, cirrhosis severity, etc.) to further validate the robustness and clinical utility of these eRNA biomarkers across different etiological subtypes of HCC.Overall, this study presents a systematic process for identifying functionally and clinically relevant eRNA biomarkers from RNA-seq data. The utilization of multiple datasets improves the reliability of the presented results. By clarifying eRNA-mediated gene regulation, new targets, and strategies for the treatment of HCC may be identified.