HepG2 cells were obtained from the Chinese Academy of Sciences Cell Bank (Shanghai, China), cultured in RPMI-1640 medium (Invitrogen, United States) supplemented with 10% fetal bovine serum (FBS, ExCell, China), and were incubated under humidified atmospheric conditions with 5% CO₂ at 37°C [12]. HCC patients’ tumor (n = 10) and tumor-adjacent normal tissues (n = 10) were collected in the Third Xiangya Hospital and used for immunohistochemistry (IHC) staining of BAG2. The clinicopathological characteristics of these patients were summarized in Supplementary Table S1. Use of human tissue was approved by the institutional review board (IRB) of the Third Xiangya Hospital, Central South University (No: 2020-S374). Written informed consent form (ICF) had been obtained from patients whose tissues were used in this study.

Total RNA was extracted by using Trizol (Pufei Biotech Co., Ltd., Shanghai, China), and then quantified with the Nanodrop 1,000 (Thermo Fisher Scientific, United States), followed by cDNA synthesis (Promega, United States). 6 ul SYBR Premix Ex Taq (TAKARA, Japan), 0.3 ul primers, 0.6 ul cDNA, and 5.1 ul RNase-free H₂O were added into 96-well plates for Real-time PCR (RT-PCR) on the LightCycler 480 II (Roche, Switzerland). The running parameters were set as: 95°C for 30 s, 40 cycles of PCR (95°C for 5 s and 60°C for 30 s), and dissociation (95°C for 15 s, 60°C for 30 s and 95°C for 15 s) [13]. The primers used in this work were obtained from Ruan Tuo Biotechnologies (Shanghai, China) and their sequences were included in Supplementary Table S2. Results were first normalized against the housekeeping gene GAPDH, which is stable across all samples, and then against the experimental controls.

Three small interfering RNAs targeting BAG2 (siBAG2), named from si-BAG2-1 to si-BAG2-3, and a negative control (NC) siRNA (si-NC) were designed by and purchased from Ruan Tuo Biotechnologies (Shanghai, China). The sequences were si-BAG2-1 (sense : 5′-GUACUAGGAUCUAGCAUAUUUTT-3′, Anti-sense : 5′-AAAUAUGCUAGAUCCUAGUACTT-3′), si-BAG2-2 (sense : 5′-GAUC-AGAAGUUUCAAUCCAUATT-3′, Anti-sense : 5′- UAU​GGA​UUG​AAA​CUU​CUG​AUC​TT-3′), and si-BAG2-3 (sense : 5′- AGC​AUG​CCA​CAA​GGA​UUA​UUG​TT-3′, Anti-sense : 5′- CAA​UAA​UCC​UUG​UGG​CAU​GCU​TT-3′), si-NC (sense : 5′- UUCUCCGAACGUGUCACGUdTdT-3′, Anti-sense : 5′- ACGUGACACGUUCGGAGAAdTdT-3′). HepG2 cells were seeded into 12-well plates, and transfection of siRNA was conducted by using Lipofectamine™ 2000 transfection reagent (Invitrogen, Shanghai, China) according to the manufacturer’s instruction.

CCK-8 assay was used to evaluate cell proliferation. Briefly, HepG2 cells were seeded at a density of 4,000 cells per well in 96-well plates and incubated for 1, 2, 3, 4, or 5 days respectively. 10 μl CCK-8 (Dojindo Molecular Technologies, Japan) was added to each well, incubated for 4 h, and mixed gently on an orbital shaker for 2 min before the absorbance value (OD) of each well was measured at 450 nm. Experiments were carried out in triplicate.

1 * 10⁵ transfected cells were seeded on 6 cm-diameter plates with RPMI-1640 medium containing 10% FBS for 48 h. Then the cells were harvested, washed twice with ice-cold phosphate-buffered saline (PBS), and re-suspended with binding buffer at a concentration of 1 * 10⁶ cells/ml. The cells were then labeled by using an annexin V-FITC/PI staining kit (eBioscience, United States) according to manufacturer’s instructions. The DNA content of labeled cells was analyzed with fluorescence activated cell sorting (FACS) cytometry (Millipore, United States). Experiments were performed in triplicate.

1 * 10⁵ transfected cells were seeded in 500 μl RPMI-1640 medium on the matrigel in the upper chamber of the Corning® BioCoat™ Matrigel® Invasion Chambers (8 mm pore size; Corning, United States), 750 μl RPMI-1640 medium containing 30% FBS was added in the bottom chamber. The cells were incubated for 24 h at 37°C with 5% CO₂ and then were fixed in 4% paraformaldehyde and stained with crystal violet solution (Sigma, United States). The cells on the bottom of the membrane were visualized under a microscope and quantified by counting the number of cells in three randomly chosen fields at 100-fold magnification. Experiments were performed in triplicate.

IHC staining of BAG2 on 8 μm paraffin sections was performed according to a standard 3,3′-Diaminobenzidine (DAB) staining protocol (Abclone, China) with primary antibody against BAG2 (Abclone, ab115205, 1:200). Two independent pathologists assessed all IHC samples. The extent of cell staining (0–10% positive cells for 0; 11–50% positive cells for 2; 51–80% positive cells for 3; >80% positive cells for 4) and the staining intensity (no staining for 0; slight staining for 1; moderate staining for 2; strong staining for 3) were scored separately and then added to reflect the expression of BAG2 [14].

Western blotting analysis was conducted as reported in our previous studies [13]. Briefly, transfected cells were cultured for 24 h before total protein was extracted by using RIPA buffer (Beyotime, China) containing proteinase inhibitor cocktail (Roche, Switzerland). The BCA Protein Assay Kit (Beyotime, China) was used to measure the protein concentration, and approximately 30 μg of total protein was separated by 10% SDS-PAGE and transferred onto PVDF membranes (Millipore, United States). The membranes were incubated in 5% skim milk for 1 h at room temperature and then overnight at 4°C with primary antibodies. After the membranes were washed three times, they were incubated with HRP-conjugated secondary antibody. The primary antibodies were rabbit polyclonal anti-BAG2 (Abclonal, China) and rabbit polyclonal anti-GAPDH (Abcam, United States). The secondary antibody was HRP-conjugated anti-rabbit (Sigma, United States).

Normalized gene expression data (HTSeq-FPKM), phenotype data, survival information, and miRNA expression quantification data of liver hepatocellular carcinoma (LIHC) were downloaded from TCGA through GDC Xena Hub (https://xena.ucsc.edu/public) on March 03, 2020. Progression free survival (PFS) and disease specific survival (DSS) data were downloaded from Kaplan Meier-plotter (http://www.kmplot.com), with the gene expression data of BAG2. The Oncomine database (https://www.oncomine.org/resource/login.html) was used to validate the expression of BAG2 in HCC, with default thresholds: p-value < 1E-4, fold change >2 and the gene ranks in the top 10%. GSE64041 (60 pairs of tumor and tumor-adjacent normal samples), GSE25097 (6 liver samples from healthy people, 40 cirrhosis samples, 243 tumor-adjacent normal samples, and 268 tumor samples) and GSE121248 (37 tumor-adjacent normal samples and 70 tumor samples) datasets were downloaded from the Gene Expression Omnibus (GEO) database through R (version 3.6.2), by the GEOquery package. All the data in this part were processed, analyzed and plotted by R (version 3.6.2), and the following packages were used: BiocManager, clusterProfiler, dplyr, ggplot2, ggpubr, limma, org. Hs.eg.db, plyr, survival, survminer, and tidyverse.

The co-expressed genes of BAG2 were constructed into protein-protein interaction (PPI) networks in the STRING database (http://string-db.org). Cytoscape (version 3.7.2) was then used for the visualization of these networks. The hub genes were then identified as the top 10 nodes based on the score generated by the cytoHubba plug-in in Cytoscape (ranked by degree). Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis were conducted by the clusterProfiler package in R (version 3.6.2) [15].

Overall survival (OS), PFS, and DSS of high- and low-BAG2 subgroups of HCC patients were compared using the Kaplan-Meier method with the log-rank test. Cox regression analysis were conducted with SPSS (version 21), to evaluate prognostic factors. The cut-off value of BAG2 was determined by X-tile (version 3.6.1) [16]. Levene’s test was used for homogeneity of variance test. When the variance was assumed even, unpaired t test was used to compare the data from two groups and one-factor ANOVA test for multiple groups without post-hoc analysis. Wilcoxon signed rank test was used for the comparison between two groups when the variance was assumed uneven. Correlation analysis between the gene expression of BAG2 and that of the rest genes was conducted by R software (version 3.6.2) with spearman method. The data from in vitro experiments were analyzed and plotted by Graphpad Prism 8 (GraphPad Software, La Jolla, CA). All statistical tests were two tailed and p value <0.05 was considered significant.

To delineate the function of BAG2 in HCC, we first examined its expression pattern in the disease. By analyzing the data from the TCGA_LIHC data set, we found that the transcriptional level of BAG2 was considerably increased in tumor tissues when compared with tumor-adjacent normal tissues (Figure 1A, p < 0.001). Such an expression pattern was supported by three independent GEO data sets, namely GSE64041, GSE25097 and GSE121248 (Figures 1B–D, p < 0.0001). We further validated that BAG2 was significantly up-regulated in HCC by comparing its expression between HCC samples and normal samples across five independent cohorts in the Oncomine database (Figure 1E). In addition, we conducted IHC staining of BAG2 on 10 HCC tissues and 10 tumor-adjacent normal tissues and scored the staining based on the intensity and extent [14]. The results confirmed that the expression of BAG2 was significantly elevated in tumor tissues when compared with normal tissues (Figures 1F–H, p < 0.05).

We then examined the relationship between clinicopathological features of HCC patients and the gene expression of BAG2 by manipulating data from the TCGA_LIHC data set. The clinical characteristics of HCC patients were summarized in Supplementary Table S3. A total of 368 cases had gene expression data, survival data and phenotype data. Unavailable or unknown clinical features of the 368 patients were regarded as missing values. Briefly, 44.8% of HCC patients were younger than 60 years old, and more patients were male than female (67.7 vs. 32.3%). Most patients were at relatively good status, because only a small fraction of patients were at T4 stage (3.5%), with lymph node metastasis (N1, 1.1%), with distant metastasis (M1, 0.8%), diagnosed as stage IV (1.1%), or with bad liver function (Child–Pugh classification C, 0.3%). Besides, 56.3% of patients had no vascular invasion, 20.4% had no fibrosis, and 32.1% had no adjacent hepatic tissue inflammation. With respect to cancer status, 33.7% of HCC patients had tumor while 44.0% of them were tumor free. As shown in Figures 2A–I, high level of BAG2 was significantly correlated with advanced T stage (p < 0.01) and disease stage (p < 0.05). Due to the small number of patients with lymph node or distant metastasis, no correlation was found between N or M stage and BAG2 expression. In addition, patients with poor cancer status (with tumor, p < 0.05) or with vascular invasion (p < 0.05) was associated with higher expression of BAG2. No correlation was found between the expression of BAG2 and level of AFP, fibrosis or adjacent hepatic tissue inflammation.

As shown in Figures 2J–L, HCC patients with high expression of BAG2 had significantly shorter overall survival (OS, p = 0.00012), progression free survival (PFS, p = 0.0029), and disease specific survival (DSS, p = 0.007). A univariate analysis revealed that low level of BAG2 [hazard ratio (HR): 0.508; 95% confidence interval (CI): 0.357–0.722; p < 0.001], good cancer status (HR: 0.372, CI: 0.234–0.590; p < 0.001), early T stage (HR: 0.479, CI: 0.334–0.685; p < 0.001), without lymph node metastasis (HR: 0.676, CI: 0.469–0.973; p = 0.035), and without distant metastasis (HR: 0.613, CI: 0.426–0.883; p = 0.009) were all associated with longer OS (Table 1a). We further put these factors into a multivariate Cox regression analysis and found that low BAG2 expression was still an independent factor that correlated to good prognosis (HR: 0.421, CI: 0.247–0.719; p = 0.002) (Table 1b).

To investigate how BAG2 affected biological activities of HCC, we transfected HepG2 cells with three si-BAG2s or si-NC. All the three si-BAG2s dramatically down-regulated the expression of BAG2 on the mRNA and protein level, and si-BAG2-2 had the strongest inhibition (Figure 3A). We then used si-BAG2-2 for further experiments. As shown in Figure 3B, silencing the expression of BAG2 in HCC cells significantly impaired cell proliferation on day 4 (p < 0.01) and day 5 (p < 0.01). Besides, HCC cells with down-regulated expression of BAG2 showed a significantly increased percentage of apoptotic cells (Figure 3C, p < 0.01). Because Lisha Sun et al. reported that BAG2 promoted metastasis of gastric cancer [7], we further conducted invasion assay and the result suggested that BAG2 knockdown dramatically decreased the amount of cells invaded though the matrigel (Figure 3D).

The above analysis and experiments indicated that BAG2 was up-regulated in HCC and might favor the progression of the disease; we then intended to understand the underlying regulatory mechanisms. Correlation analysis was conducted independently between the expression of BAG2 and that of other genes in tumor samples of the TCGA_LIHC and GSE64041 data sets. 10,918 genes and 3,837 genes were identified as co-expressed genes in the TCGA_LIHC and GSE64041 data sets respectively (p value cutoff = 0.05). Based on the correlation coefficient, the top 500 co-expressed genes from the two data sets were intersected and 155 genes were found to be the common co-expressed genes for BAG2 (Figure 4A; Supplementary Table S4). These co-expressed genes were then undergoing GO and KEGG analysis. The significantly affected biological processes (BP) were “RNA biogenesis and processing like ribosome biogenesis,” “ribosome RNA (rRNA) metabolic process,” “non-coding RNA (ncRNA) processing,” “rRNA processing,” “maturation of SSU-rRNA from tricistronic rRNA transcript (SSU-rRNA, 5.8S rRNA, LSU-rRNA),” and “positive regulation of telomerase RNA localization to Cajal body” (Figure 4B). The significantly enriched molecular functions (MF) of these co-expressed genes were “catalytic activity,” “unfolded protein binding,” “RNA polymerase activity,” and “aminoacyl-tRNA ligase activity” (Figure 4B). According to the KEGG analysis, “RNA transport,” “RNA polymerase,” “aminoacyl-tRNA biosynthesis,” “Alanine, aspartate and glutamate metabolism,” and “ribosome biogenesis in eukaryotes” were significantly enriched terms (Figure 4C; Table 2).

We then uploaded BAG2 and its 155 co-expressed genes into the STRING database and constructed a BAG2-contained PPI network, which was visualized by the Cytoscape software (Figure 5A). By using the cytoHubba plug-in, the 10 hub genes (ranked by degree) were identified to be WD repeat domain 12 (WDR12), bystin like (BYSL), DDB1 and CUL4 associated factor 13 (DCAF13), WD repeat domain 3 (WDR3), chaperonin containing TCP1 subunit 2 (CCT2), ribosome biogenesis regulator 1 homolog (RRS1), nucleolar pre-RRNA processing protein NIP7 (NIP7), RNA polymerase I and III subunit C (POLR1C), nucleolar protein interacting with the FHA domain of MKI67 (NIFK), chaperonin containing TCP1 subunit 7 (CCT7) (Figure 5A, yellow blocks; Supplementary Table S4). The biological process analysis of these hub genes was conducted by using the BINGO plug-in the Cytoscape software. “Ribosome biogenesis,” “ribosomal large subunit biogenesis,” “ribonucleoprotein complex biogenesis,” “rRNA processing,” “ncRNA processing,” “rRNA metabolic process,” “RNA processing,” “arginyl-tRNA aminoacylation” and “ncRNA metabolic process” were significantly altered, consistent with the GO and KEGG enrichment analysis of the co-expressed genes of BAG2 (yellow modules, Figure 5B). Except WDR3, all the rest nine hub genes were significantly up-regulated in tumors when compared to tumor-adjacent normal tissues in both the TCGA_LIHC and GE64041 data sets (Figures 5C,D). We then analyzed the prognostic significance of the nine hub genes in the TCGA_LIHC data set, and the result suggested that WDR12, BYSL, DCAF13, CCT2, RRS1, NIP7, NIFK, and CCT7 at high expression levels were all associated with significantly unfavorable prognosis in HCC (Figures 6A–H), while no significant correlation was observed between the expression of POLR1C and OS of HCC patients (Figure 6I). In addition, previous studies revealed that most of these hub genes were required or inducible for ribosome biogenesis [17–23]. Further, we detected the transcriptional level of the 10 hub genes in HepG2 cells with BAG2 silencing, and found that all of them showed a significant down-regulation after BAG2 knockdown, suggesting BAG2 might affect ribosome biogenesis through these genes (Supplementary Figure S1).