Development and validation of genomic and epigenomic signatures associated with tumor immune microenvironment in hepatoblastoma

GSE81928 [8], GSE75271 [9] and GSE78732 datasets were retrieved from the Gene Expression Omnibus (GEO; https://​www.​ncbi.​nlm.​nih.​gov/​geo/​) database. Among them, GSE81928 dataset was composed of hepatic RNA-seq profiling of normal (n = 3), background (n = 6), and hepatoblastoma (n = 23) tissue specimens on the Illumina HiSeq2500 platform. GSE75271 dataset contained microarray expression profiles of hepatoblastoma (n = 50) and normal liver (n = 5) tissues on the GPL570 platform. GSE78732 dataset included hepatic methylation profiling of hepatoblastoma (n = 7) and normal differentiated liver (n = 7) samples on the GPL13534 platform.

The original microarray profiles of GSE75271 dataset were read using “affy” package [10]. The raw data were standardized via robust multichip averaging (RMA) method. The, the standardized expression data were re-annotated according to the annotation file of the microarrays. Probes that did not correspond to or correspond to multiple gene symbols were deleted. For one gene symbol corresponding to multiple probes, the maximum value was chosen as the expression value of this gene. The raw data from GSE81928 dataset were dimensionally tested. The matrix was standardized by log2 (TPM + 1) to eliminate the data dimension. Principal component analysis (PCA) was presented to remove outliers. Using “ChAMP” package, the raw data of GSE78732 dataset were read [11]. After quality control, filtering and clustering, outlier samples were deleted. Then, methylation β values were normalized with the BMIQ method.

Differential expression analyses were presented based on the GSE75271 and GSE81928 datasets. Differentially expressed genes between hepatoblastoma and normal hepatic tissues were screened by “limma” package [12]. The screening threshold was p-value < 0.05 and |log2fold change (FC)| > 1.5. Then, common up- and down-regulated genes were intersected between the two datasets.

The jveen tool (http://​jvenn.​toulouse.​inra.​fr/​app/​index.​html) was utilized to comprehensively analyze the differentially expressed genes and differentially methylated genes [13]. The genes obtained by the intersection were regarded as differentially methylated and expressed genes.

Differentially methylated and expressed genes were uploaded to the STRING database (version: 11.0; http://​string-db.​org/​) [14] and Metascape database (http://​metascape.​org/​) [15] for functional enrichment analysis. Then, this study combined the enrichment results from the two databases.

PPI analysis was performed using the STRING database. The parameters were set according to the default. PPI analysis results were visualized using the Cytoscape (https://​cytoscape.​org/​) [16]. The MCODE plug-in was used to mine key sub-networks from the PPI network as follows: degree cutoff = 2, node score cutoff = 0.2, max depth = 100 and k-score = 2. By combining PPI and key sub-networks, hub genes were determined based on degree > 5 in the PPI network and genes in the key sub-networks. The list of hub genes was uploaded to the GENEMANIA (http://​genemania.​org) database [17]. Functionally similar genes of hub genes were analyzed and gene functions were then predicted.

CIBERSORT (http://​cibersort.​stanford.​edu/​) tool was used for inferring the compositions of 22 kinds of tumor-infiltrating immune cells (TIICs) in normal and hepatoblastoma samples through deconvolution algorithm [18]. The LM22 matrix was used for 1000 calculations. Monte Carlo sampling was used to calculate the p-value for the deconvolution of each sample. The differences in immune cell compositions between normal and hepatoblastoma specimens were evaluated via student’s t test. The correlation between immune cells was analyzed via Spearson correlation analysis. Furthermore, Spearson correlation between hub genes and immune cells was analyzed.

We collected 20 paraffin section specimens of hepatoblastoma in the Hunan Children’s Hospital between 2019 and 2020. Adjacent normal tissues were used as controls. Additional file 1 listed detailed clinical information for each patient. All of them did not receive any anti-cancer treatment before surgery, such as oral chemotherapy, radiotherapy, or immunotherapy. All subjects were confirmed by pathological examination. This study was approved by the Ethics committee of Hunan Children’s Hospital (HCHLL-2021-09). The research has been carried out in accordance with the World Medical Association Declaration of Helsinki, and obtained the written informed consent from their parent and/or legal guardian.

RIPA lysate (Beyotime, Beijing, China) was utilized to extract total protein from tissue or cell samples, followed by BCA protein quantification. 30 μg protein was taken for SDS-PAGE gel electrophoresis and transferred to membrane. Membranes were sealed with 5% skimmed milk powder solution at room temperature for 2 h. They were then incubated with monoclonal antibodies against AXIN2 (1/1000; ab109307, abcam, USA), LAMB1 (1/1000; ab256380), NOTUM (1/1000; ab106448), SERPINC1 (1/1000; ab124808), E-cadherin (1/1000; ab238099), Vimentin (1/1000; ab137321), Snail (1/1000; ab78105) and β-actin (1/5000; ab179467) at 4 °C overnight. On the next day, membranes were incubated by HRP-labeled IgG secondary antibody (1/1000; ab7090) at room temperature for 2 h. ECL chemiluminescence detection was presented. The expression of target proteins was quantified.

Paraffin-embedded sections were deparaffinized with xylene and hydrated with gradient ethanol. Then the sections were incubated in 3% methanol-H₂O₂ and transferred to 0.01 mol/L sodium citrate solution for antigen retrieval. The sections were blocked by 5% skimmed milk powder solution at room temperature and then incubated with anti-NOTUM (1/100; ab106448) antibodies and secondary antibodies (1/1000; ab7090), respectively. DAB was used to develop color and neutral gum was utilized for mounting the film.

Human HuH-6 and HepG2 hepatoblastoma cells (ATCC, USA) were cultured in DMEM medium (Gibco, USA) containing 10% fetal bovine serum (FBS). The cells were cultivated in a 37 °C, 5% CO₂ incubator. Cells in the logarithmic growth phase were inoculated on a 6-well plate (1 × 10⁶/well). After incubating for 12 h, Lipofectamine 3000 (BOSTER, Wuhan, China) was utilized for transfection. According to the Lipofectamine 3000 instructions, the experiment was divided into negative control (NC) and sh-NOTUM groups. After transfection, cells were cultured for 6 h, and then replaced with DMEM medium. 48 h after transfection, transfected cells were collected. Western blot was used to detect the expression of NOTUM protein.

Totally, 10 male 5-week-old nude mice (18-22 g) were purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd. (Beijing, China). This experiment was approved by the Animal Care Committee of Hunan Children’s Hospital (HCHLL-2021-09). HuH-6 and HepG2 (1 × 10⁷) stably transfected with NC or sh-NOTUM were separately subcutaneously injected into the left gluteal region of nide mice (n = 5). The tumor volume was measured every 4 days, and tumor size was calculated based on the following formula: volume (mm³) = width2 (mm²) · length (mm)/2. All nude mice were euthanized after 21 days, and tumors were weighted following ethical dissection and photography.

HuH-6 and HepG2 cells were inoculated in a 6-well class. Cells were transfected with shRNA overnight. A 10 μl pipette tip was used to make a uniform scratch on the cell culture plate. Exfoliated cells were washed away by PBS. After culturing the cells in the incubator for 0 h, 24 h and 48 h, 3 fields of view were randomly selected to calculate the scratch width.

The matrigel was diluted with pre-chilled serum-free DMEM and spread in the upper chamber of transwell (Corning, USA) overnight at 37 °C. 200 μL transfected cells were added to the upper chamber (2 × 10⁵ cells / L). 500 μL DMEM medium containing 10% FBS was added to the lower chamber. Cells were incubated for 48 h. Then, cells were fixed in pre-cooled formaldehyde for 20 min and were stained with crystal violet for 30 min in the dark. The cells in the upper chamber were gently wiped with a cotton swab. The number of invaded cells was counted under a microscope.

To obtain hepatoblastoma-related genes, this study retrieved GSE75271 and GSE81928 datasets. Firstly, we normalized microarray profiles of GSE75271 dataset by RMA method (Fig. 1A). Figure 1B showed the results of dimension analysis. By PCA, we removed three hepatoblastoma samples “GSM1948566”, “GSM1948577”, “GSM1948562” that were close to normal liver tissues. Figure 1C displayed the PCA results of GSE75271 expression matrix after removing outliers. Totally, 1364 differentially expressed genes were screened for hepatoblastoma than normal samples, including 756 up- and 608 down-regulated genes (Fig. 1D). Their expression in each sample was visualized into a heat map (Fig. 1E). For RNA-seq data of GSE81928 dataset, we standardized the expression matrix by log2 (TPM + 1) in Fig. 2A, B. Figure 2C depicted the dimension analysis results following normalization. Background profiles were then removed (Fig. 2D). According to differential expression analysis, 528 up- and 274 down-regulated genes were screened for hepatoblastoma (Fig. 2E). These genes may distinguish hepatoblastoma from normal samples (Fig. 2F). After intersection of differentially expressed genes between GSE75271 and GSE81928 datasets, 134 up-regulated genes were identified for hepatoblastoma (Fig. 3A). Meanwhile, 191 down-regulated genes were intersected for hepatoblastoma (Fig. 3B). These common genes were further analyzed.

Methylation profiles were obtained from GSE78732 dataset. Firstly, all samples were clustered before normalization. An outlier sample “GSM2074844” was removed (Fig. 4A). Afterwards, we analyzed differentially methylated sites between hepatoblastoma and normal samples. As a result, 1125 hypermethylated and 2337 hypomethylated sites were screened in hepatoblastoma (Fig. 4B). There was a distinct difference in methylation between hepatoblastoma and normal specimens (Fig. 4C). Here, genes corresponding to differentially methylated sites were considered as differentially methylated genes. Then, we identified 83 differentially methylated and expressed genes in hepatoblastoma (Fig. 4D), as follows: TNFRSF19, SP5, NOTUM, NKD1, BMP4, ODAM, ROBO1, BAMBI, AXIN2, TRH, TRIB2, ASPSCR1, MSX1, HOXA3, ARHGEF3, PTK7, LAMB1, TBX3, C6orf48, IGDCC3, NT5DC2, GSTP1, RERE, COL4A1, LZTS2, RPS12, NPM1, PLCG1, CYP2C8, SLC22A1, HAL, SLC10A1, XDH, HAO2, C3P1, TDO2, PGLYRP2, CYP1A2, GLS2, TTC36, HABP2, ADRA1A, HPD, HPX, FBP1, FMO3, LBP, THRSP, CDA, ALDOB, HFE2, SRD5A2, GOT1, NAMPT, MAT1A, CLU, TAT, CD14, TRPM8, GREM2, ATF3, AKR7A3, CBS, G0S2, MT1G, CSRNP1, ACSM5, DPYS, FNDC5, IGFALS, DAO, STEAP3, DUSP1, MARCO, APBB1IP, SLC43A3, SERPINC1, NRG1, AGXT2, ACSL1, ABCB11, UGP2 and MGST1.

To investigate the biological functions of differentially methylated and expressed genes, we presented functional enrichment analysis by the STRING database. Our data showed that these genes were mainly enriched in metabolic or catabolic processes such as carboxylic acid metabolic process, alpha-amino acid catabolic process, small molecule catabolic process, small molecule metabolic process, carboxylic acid catabolic process, small molecule biosynthetic process, xenobiotic metabolic process, cellular amino acid metabolic process, catabolic process, and organic substance catabolic process (Fig. 5A). Cellular components enriched by these genes were then analyzed. The results showed that only peroxisome and peroxisomal membrane were enriched. Furthermore, they had various molecular functions such as cofactor binding, oxidoreductase activity, protein binding, identical protein binding, oxidoreductase activity, acting on CH or CH2 groups, coenzyme binding, binding, pattern recognition receptor activity, pyridoxal phosphate binding and transaminase activity, indicating that they were mainly involved in catabolic processes. KEGG pathway enrichment analysis also confirmed that that these genes primarily participated in metabolic pathways (such as drug metabolism - cytochrome P450 and other enzymes, cysteine and methionine metabolism, phenylalanine metabolism, metabolism of xenobiotics by cytochrome P450, biosynthesis of amino acids, caffeine metabolism, phenylalanine, tyrosine and tryptophan biosynthesis, peroxisome, chemical carcinogenesis, alanine, aspartate and glutamate metabolism, tyrosine metabolism, ubiquinone and other terpenoid-quinone biosynthesis, carbon metabolism, pentose phosphate pathway, linoleic acid metabolism, arginine biosynthesis and bile secretion) and cancer-related pathways (such as Wnt signaling pathway, hepatocellular carcinoma, pathways in cancer, NF-κB signaling pathway) in Fig. 5B.

Through the Metascape database, we performed DisGeNET annotation analysis on 83 methylated and expressed genes. We found that these genes were closely related to liver diseases, such as drug-induced liver disease, chemical and drug-induced liver injury, hepatitis, toxic, chemically-induced liver toxicity, drug-induced acute liver injury, hepatitis, drug-induced, fatty liver disease, cholestasis, diabetes mellitus, experimental malnutrition, juvenile arthritis, congenital defects, endotoxemia, thrombosis of cerebral veins, prostatic hypertrophy, homocystinuria, and hypertyrosinemia (Fig. 5C). PaGenBase annotation analysis results revealed that these genes were liver tissue-specific and liver cell-specific (Fig. 5D). Figure 5E indicated the complex interactions between pathways enriched by these genes.

This study explored the interactions between proteins from differentially methylated and expressed genes in depth. A PPI network was established, composed of 60 nodes (Fig. 6A). Among all nodes, 16 genes were up-regulated, while 44 genes were down-regulated in hepatoblastoma. BMP4 (degree = 10) had the highest degree in the network, which had the closest connection with other nodes, indicating that their status in the network was the most important. Using the MCODE plugin, we established two subnetworks from the PPI network. In Fig. 6B, there were CYP1A2, GSTP1, CYP2C8, ABCB11 and MGST1 in the cluster 1. Furthermore, there were BMP4, SERPINC1, LAMB1 and NOTUM in the cluster 2 (Fig. 6C). In this study, 13 hub genes were determined by integration of genes with degree > 5 in the PPI network and genes in the two sub-networks, including AGXT2, HPD, TAT, HAO2, GSTP1, ABCB11, MGST1, LAMB1, CYP1A2, CYP2C8, BMP4, SERPINC1 and NOTUM.

Using the GENEMANIA database, we identified functionally similar genes for hub genes, including genes with co-expression, co-localization, and shared protein domains (Fig. 7A). Furthermore, we predicted the gene functions of these hub genes. The data showed that these hub genes were mainly enriched in various metabolic processes and physiological functions of blood (Fig. 7B). Especially, demethylation was distinctly enriched by them. These data highlighted the pivotal role of these hub genes in hepatoblastoma.

We employed CIBERSORT to assess compositions of TIICs in normal and hepatoblastoma specimens. Figure 8A showed the compositions of each TIIC for each hepatoblastoma sample. Observably, T cells CD4 memory resting and macrophages M2 were main immune cell types in hepatoblastoma tissues (Fig. 8B). Furthermore, we compared the differences in immune cell compositions between normal and hepatoblastoma specimens. Our data showed that there were significantly higher compositions of B cells naïve, T cells CD8, T cells CD4 memory resting, T cells follicular helper, T cells regulatory (Tregs), T cells gamma delta, NK cells resting, NK cells activated, macrophages M0, macrophages M2, dendritic cells resting, dendritic cells activated and mast cells resting in hepatoblastoma compared to normal tissue specimens (Fig. 8C). Meanwhile, B cells memory, plasma cells, monocytes, mast cells activated, eosinophils and neutrophils exhibited distinctly lowered compositions in hepatoblastoma than normal tissues. These data were indicative that abnormal immune cells could be in relation to hepatoblastoma progression.

To explore the interactions between tumor immune cells, we analyzed the correlations between immune cells in hepatoblastoma. Figure 9A, B showed the complex interactions between immune cells in tumor microenvironment of hepatoblastoma. Specially, B cells naïve were negatively correlated to B cells memory (cor = − 0.633 and p-value = 2.171e-07; Fig. 9C). In Fig. 9D, there was a negative correlation between macrophages M1 and dendritic cells activated (cor = − 0.548 and p-value = 1.509e-05). Furthermore, a negative correlation between macrophages M1 and macrophages M2 was found in hepatoblastoma (cor = − 0.436 and p-value = 0.0008716; Fig. 9E). In Fig. 9F, mast cells resting had a negative association with mast cells activated (cor = − 0.618 and p-value = 4.922e-07). Also, there was a negative association between NK cells resting and NK cells activated (cor = − 0.612 and p-value = 6.798e-07; Fig. 9G). In Fig. 9H, plasma cells were negatively associated with mast cell resting (cor = − 0.435 and p-value = 0.0009076). There was a positive correlation between plasma cells and Tregs (cor = 0.403 and p-value = 0.002263; Fig. 9I).

We further analyzed whether hub genes were in relation to immune cells in hepatoblastoma. The data showed that ABCB11 was negatively correlated to mast cells resting (cor = − 0.41 and p-value = 0.001876; Fig. 10A) and positively associated to monocytes (cor = 0.431 and p-value = 0.001021; Fig. 10B) and plasma cells (cor = 0.414 and p-value = 0.001657; Fig. 10C). Moreover, BMP4 exhibited a positive association with mast cell resting (cor = 0.431 and p-value = 0.001017; Fig. 10D) and showed a negative correlation to plasma cells (cor = − 0.482 and p-value = 0.0002354; Fig. 10E). In Fig. 10F, CYP1A2 was positively linked to plasma cells (cor = 0.433 and p-value = 0.001081). CYP2C8 (cor = 0.479 and p-value = 0.0002154) and HAO2 (cor = 0.444 and p-value = 0.0006848) both displayed positive interactions with eosinophils in Fig. 10G, H. LAMB1 possessed a positive association with macrophages M0 (cor = 0.424 and p-value = 0.001242; Fig. 10I). NOTUM was negatively correlated to B cells memory (cor = − 0.487 and p-value = 0.0001653; Fig. 10J) and plasma cells (cor = − 0.405 and p-value = 0.002342; Fig. 10K). SERPINC1 was negatively associated with macrophages M0 (cor = − 0.524 and p-value = 4.093e-05; Fig. 10L) and was positively correlated to plasma cells (cor = 0.444 and p-value = 0.0007725; Fig. 10M). TAT exhibited a positive association with plasma cells (cor = 0.422 and p-value = 0.001473; Fig. 10N). Hence, hub genes could be closely correlated to immune cells in hepatoblastoma.

This study collected 20 pairs of paraffin section specimens from hepatoblastoma and adjacent normal tissues. The expression of AXIN2, LAMB1, NOTUM and SERPINC1 proteins was examined by western blot. Our results demonstrated that AXIN2 (p < 0.01), LAMB1 (p < 0.05) and NOTUM (p < 0.001) were all distinctly up-regulated in hepatoblastoma compared to normal tissues (Fig. 11A-D). SERPINC1 protein (p < 0.01) exhibited lowered expression in hepatoblastoma than normal tissues (Fig. 11E). NOTUM protein was also detected in hepatoblastoma and adjacent normal tissues via immunohistochemistry. There was a higher NOTUM expression in hepatoblastoma than normal specimens (p < 0.05; Fig. 11F, G).

To observe the function of NOTUM hepatoblastoma progression, NOTUM was silenced by transfection with sh-NOTUM in HuH-6 or HepG2 hepatoblastoma cells. Western blot confirmed the decrease in NOTUM expression in HuH-6 cells (p < 0.01; Fig. 11H-J). This study established nude mouse tumor xenograft models injected by HuH-6 or HepG2 cells that were transfected with NC or sh-NOTUM. Tumor volume was measured every 4 days. We found that NOTUM knockdown significantly lessened the tumor volume (Fig. 11K, L). After 21 days, we measured the tumor weight. We observed that tumor weight was distinctly lowered by NOTUM knockdown (Fig. 11M-P).

Migration and invasion of HuH-6 and HepG2 cells were assessed after the deletion of NOTUM. In Fig. 12A-C, the number of invasive HuH-6 and HepG2 cells was reduced under transfection with sh-NOTUM (p < 0.01). Furthermore, NOTUM knockdown distinctly suppressed the migratory ability of HuH-6 and HepG2 cells (p < 0.01; Fig. 12D-F). Epithelial-mesenchymal transition (EMT) is essential for cancer migration and invasion. The expression of EMT markers (E-cadherin, Vimentin and Snail) was examined following the deletion of NOTUM. Our results showed that NOTUM knockdown significantly enhanced the expression of E-cadherin as well as lessened the expression of Vimentin and Snail in HuH-6 and HepG2 cells (Fig. 12G-J).