Tumor specimens for tissue microarrays were collected from 236 HCC patients who underwent surgery from April 2005 to September 2008 at Department of Liver Surgery, Zhongshan Hospital of Fudan University. All patients were informed of the aim of the study and gave consent to donate their samples.

Two-tailed t-test was used to compare the differences between two sets, and two-way analysis of variance was used to assess the statistical significance of two-factor interactions with multiple time points. The X-tile software was used to generate the optimum cut-off point for continuous variables,24 and HCC patients was divided into CBS high expression group and CBS low expression group according to the optimum cut-off point. Kaplan-Meier survival analysis with log-rank test was performed to evaluate risk factors for overall survival (OS) of HCC patients, then multivariate Cox analysis was used to analyze predictors for the prognosis. Data were analyzed with SPSS software V.19.0 (SPSS Inc). P<0.05 was considered statistically significant.

Other experimental procedures are described in the online supplemental material.

To explore the role of CBS in the development of HCC, we first examined CBS expressions in different organs. As shown in online supplemental figure 1, both mRNA and protein levels of CBS were considerably higher in liver than in the other seven organs, which implied that CBS might play an essential role in maintaining liver function and homeostasis. Next, we performed statistical analysis on CBS expression in TCGA-LIHC and GEO (GSE14520) databases. Consistent with a previous study,25 we observed that the mRNA levels of CBS in HCC tumor tissues from both databases were remarkably lower than that in peritumor tissues (figure 1A,B, left panel), and the OS of HCC patients with low CBS expression was significantly shorter than those with high CBS expression (figure 1A,B, right panel). We also analyzed CBS expression in fresh HCC samples by western blot and real-time PCR, and results showed that both protein and mRNA levels of CBS were remarkably reduced in HCC tumor tissues compared with matched peritumor tissues (figure 1C–E). As CBS is one of the key enzymes involved in H₂S production, we also examined H₂S concentration. As expected, H₂S contents in HCC tumor tissues were much lower than that in adjacent non-tumor tissues (online supplemental figure 2A), then a tissue microarray was employed to examine CBS expressions in 236 HCC patients with different stages. Representative IHC staining of CBS in tumor and peritumor tissues were shown in figure 1F. It was observed that CBS was mainly localized in the cell cytoplasm. Although CBS was most abundant in normal tissues, it was downregulated in matched HCC tumor tissues (figure 1G). In order to further investigate the role of CBS in HCC development, HCC patients were stratified into different subgroups according to disease stage, and CBS expression was observed to be lower in patients with advanced stage than those with early stage (figure 1H). Kaplan-Meier analysis revealed that patients with low CBS expression had shorter OS than those with high CBS expression (figure 1I). Furthermore, we found that the expression of CBS was positively correlated with the proportion of cleaved Caspase-3 and negatively correlated with the number of Tregs in HCC clinical samples (figure 1J–K). CBS expression was also slightly correlated with the number of CD4⁺ T cells and TAMs but not with other seven kinds of immune cells including pan T cells, CD8⁺ T cells, mononuclear cells, B cells, plasma cells, naive T cells and memory T cells (online supplemental figure 3). These findings indicated that downregulation of CBS might reduce tumor cell apoptosis and increase tumor infiltrating of Tregs in HCC.

Univariate and multivariate analyses were subsequently performed to determine the prognostic factors for the OS of HCC patients. As shown in online supplemental table 5, the levels of alpha fetoprotein, γ-glutamyl transpeptidase (GGT) and alanine aminotransferase (ALT), tumor diameter, tumor differentiation, and CBS expression were identified as risk factors for the OS of HCC patients. Further estimation by multivariate Cox regression analysis demonstrated that GGT level, tumor diameter, tumor differentiation and CBS expression were independent predictors for the OS of HCC patients (online supplemental table 5). Taken together, these results implied that CBS could be a predictor for the OS of patients with HCC.

To investigate the effect of CBS on HCC progression, we transduced lentivirus to achieve stable knockdown of CBS in Hep3B cells and overexpression of CBS in MHCC97H cells, and we also transfected Cbs-plasmid into mouse Hepa1-6 cells. Real-time PCR and western blot were performed to confirm the efficiency of CBS knockdown and overexpression (figure 2A–C). CCK-8 assays showed that CBS knockdown promoted the viability of HCC cells, and flow cytometry analysis showed that CBS knockdown inhibited cellular apoptosis (figure 2D,G,H). In contrast, CBS overexpression suppressed the viability and facilitated the apoptosis of HCC cells (figure 2E,F,I–L). Subcutaneous xenograft tumor model was used to verify the antitumor role of CBS in HCC. Results showed that CBS knockdown or inhibition by aminooxyacetic acid (AOAA; a CBS inhibitor)26 significantly enhanced tumor growth in Hep3B-xenografted mice model, while CBS overexpression or activation by S-adenosyl methionine (SAM; a CBS agonist)27 exhibited opposite effects in MHCC97H-xenografted mice model (figure 3A,B,E,F). Moreover, H₂S contents in subcutaneous tumor tissues were declined following CBS depletion and were increased following CBS activation (online supplemental figure 2B,C).

We further detected the influence of CBS on the cell apoptosis marker cleaved Caspase-3 and the cell proliferation marker Ki-67 in these subcutaneous HCC tumor tissues. As shown in figure 3C,G, CBS depletion remarkably diminished cleaved Caspase-3 positive rate, while CBS activation improved the proportion of cleaved Caspase-3. Nevertheless, CBS depletion or activation showed little effect on the expression of Ki-67 (figure 3D,H). A previous study reported the regulation of CBS in glycolysis or angiogenesis,28 which play crucial roles in tumor progression. Extracellular acidification rate calculated from proton production rate was used to evaluate glycolytic capacity,29 30 and CD31 staining was used to identify tumor angiogenesis.31 However, unlike reported results in colorectal and ovarian cancers,28 our results showed that CBS knockdown or overexpression exhibited little effect on glycolysis or angiogenesis in HCC (online supplemental figure 4). Together, these data suggested that CBS suppressed HCC tumor growth through inducing cellular apoptosis but had no effects on cell proliferation, glycolysis or angiogenesis.

Next we used the CBS agonist SAM and a slow H₂S donor GYY4137 to further verify the effects of CBS/H₂S axis activation on Hepa1-6-homografted tumor inhibition in different mice strains.27 32 As shown in figure 4A, both SAM and GYY4137 administration impaired tumor growth in immunocompetent C57BL/6 mice as well as in non-obese diabetic severe combined immunodeficiency (NOD SCID) mice. Interestingly, tumor inhibition rates under GYY4137 or SAM treatment were significantly higher in immunocompetent mice than that in immunodeficient mice, which might be due to a stronger apoptosis-induced capacity in immunocompetent mice (figure 4B–D). These data suggested a possible immunoregulatory role of the CBS/H₂S axis in tumor growth repression. Considering that CBS expression was negatively correlated with tumor-infiltrating Tregs and TAMs in HCC patients (figure 1K and online supplemental figure 3B), we next examined the influence of CBS/H₂S activation on the populations of Tregs and TAMs as well as another immunosuppressive MDSCs in tumor microenvironment. Flow cytometry analysis indicated that the abundance of tumor-infiltrating Tregs in subcutaneous tumors derived from immunocompetent mice treated with GYY4137 or SAM was significantly reduced compared with control mice (figure 4E and online supplemental figure 5A). However, MDSCs and TAMs were almost unchanged after GYY4137 or SAM administration (online supplemental figure 5B,C). To confirm the effect of CBS on the immunosuppressive function of Tregs, we detected the abundance of tumor-infiltrating IFNγ⁺CD8⁺ T cells, and results showed that both GYY4137 and SAM treatment significantly increased the proportion of IFNγ⁺CD8⁺ T cells (figure 4F and online supplemental figure 5D). In GYY4137 and SAM-treated tumor tissues, the inhibition of CD8⁺ T cell proliferation by Tregs was impaired compared with control group (figure 4G). Furthermore, the tumor killing capacity of CD8⁺ T cells was higher in GYY4137 and SAM-treated group compared with control group (figure 4H). The mRNA levels of IL-10 in Tregs and IFNγ in CD8⁺ T cells were also examined, and results showed that GYY4137 and SAM treatment significantly diminished IL-10 mRNA level in Tregs but accelerated IFNγ expression in CD8⁺ T cells (online supplemental figure 5E,F).

Cbs-deficient (Cbs^+/−) immunocompetent mice were used to further certify the suppressive roles of CBS in HCC. As shown in figure 4I–L, Cbs deficiency notably facilitated Hepa1-6-homografted tumor growth and abnormally elevated the tumorous infiltration of Tregs while reduced tumor-infiltrating IFNγ⁺CD8⁺ T cells. Similar to GYY4137 and SAM treatment, Cbs deficiency exhibited little effect on the abundance of tumor-infiltrating MDSCs and TAMs (online supplemental figure 5G,H). In addition, Tregs separated from Cbs^+/− mice showed stronger suppression on the proliferation of CD8⁺ T cells than those from wild type group (figure 4M), and the tumor-killing capacity of CD8⁺ T cells derived from Cbs^+/− mice was weakened compared with those from wild type group (figure 4N). Cbs deficiency enhanced IL-10 expression in Tregs while repressed IFNγ expression in CD8⁺ T cells (online supplemental figure 5I,J). Furthermore, we constructed a primary liver carcinogenesis mouse model with intraperitoneal injection of the chemical carcinogen diethyl-nitrosamine (DEN). As shown in online supplemental figure 5K, after a short-induced period (4 months), no tumor lesions were observed in wild type mice, while 33.3% of the Cbs-deficient mice developed microtumor lesions. These data implied that Cbs deficiency improved DEN-induced HCC incidence.

To explore the underlying mechanisms of how CBS restrain HCC progression, we conducted RNA sequencing and human phospho-kinase array in MHCC97H-CBS OE cells in comparison with MHCC97H-NC cells to screen out potential CBS-regulated signal pathways. Gene set enrichment analysis based on RNA sequencing demonstrated that CBS overexpression was significantly negatively related to STAT3 signaling pathway (figure 5A). Consistently, the phospho-kinase array also manifested an obviously attenuation of STAT3 (Y705) phosphorylation when CBS was overexpressed (figure 5B). Western blot results confirmed that CBS overexpression inhibited STAT3 phosphorylation, while knockdown of CBS facilitated STAT3 phosphorylation (figure 5C). Meanwhile, the expression of the apoptosis marker cleaved Caspase-3 was significantly reduced, and the antiapoptosis factor Bcl-3 was increased when CBS was knockdown in Hep3B cells, while opposite trends were observed when CBS was overexpressed in MHCC97H and Hepa1-6 cells (figure 5C), then we aimed to confirm whether CBS depletion-mediated tumor growth acceleration was STAT3 dependent. As shown in figure 5D, the STAT3 inhibitor Stattic completely reversed the promotion of cell viability induced by CBS knockdown. The inhibition of cellular apoptosis and cleaved Caspase-3 expression as well as the improvement of Bcl-3 expression induced by CBS depletion were also blocked by Stattic (figure 5E,F). Furthermore, intraperitoneal injection of Stattic reversed the facilitation of tumor growth in mice when CBS was knockdown or partly deficient (figure 5G,H). Cbs deficiency-mediated tumor infiltration of Tregs was also disrupted by Stattic (figure 5I). Interestingly, we found that the expression of Foxp3 in CD4⁺CD25⁺ Tregs was increased by Cbs depletion in a STAT3-dependent manner (figure 5J).

Recent studies indicated that the development and function of Tregs are governed by particularly DNA demethylation of Foxp3 promoter,33 and ten-eleven translocation (Tet) family is able to convert 5-methylcytosine to 5-hydroxymethylcytosine (5hmC) to remove existing methylation marks.34 H₂S was also shown to promote methylcytosine dioxygenases Tet1-mediated and Tet2-mediated Foxp3 demethylation in spleen and lymph nodes-derived CD4⁺ T cells.35 To clarify whether the upregulation of Foxp3 in tumor-infiltrating CD4⁺CD25⁺ Tregs mediated by Cbs depletion was also related to DNA demethylation, we examined the levels of Tet1, Tet2 and 5hmC in tumor infiltrating CD4⁺ T cells. However, the expression of all these factors was almost unchanged following Cbs depletion (online supplemental figure 6A,B). Instead, the mRNA level of IL-6 was increased in Cbs deficient tumor-infiltrating Tregs (online supplemental figure 6C). These results suggested that Cbs deficiency-mediated IL-6/STAT3 activation promoted the expression of Foxp3 in tumor-infiltrating Tregs. Additionally, we investigated the role of STAT3 in the influence of Cbs deficiency on IFNγ⁺CD8⁺ T cells. As shown in figure 5K,L, the reduction of tumor-infiltrating IFNγ⁺CD8⁺ T cells following Cbs depletion was reversed by Stattic administration, while tumor killing capacity of IFNγ⁺CD8⁺ T cells was remarkably enhanced by STAT3 inhibition.

Next we aimed to elaborate how CBS represses STAT3 activation. Numerous studies have confirmed the regulatory effect of CBS on IL-617 36 and the activation of STAT3 by IL-6.19 37 As shown in online supplemental figure 7, the mRNA level of IL-6 was reduced when CBS was overexpressed in HCC cells and was increased when CBS was knockdown. Real-time PCR and ELISA analysis also showed that CBS knockdown or inhibition significantly stimulated IL-6 production in subcutaneous tumor tissues, while CBS overexpression or activation inhibited IL-6 expression (figure 6A). Further exploration using IL-6 recombinant protein indicated that CBS-mediated STAT3 inhibition was IL-6 dependent (figure 6B). However, the specific mechanism of CBS regulation on IL-6 was unknown. To elucidate the upstream transcription factor of IL-6, we searched the Jaspar database and found that 93 transcription factors had sequences complementary to IL-6 promotor (online supplemental table 6). We then conducted a transcription factors array on these 93 factors in CBS overexpression MHCC97H cells in comparison with control cells, in order to screen out potential transcription factors. The results showed that 14 transcription factors were significantly reduced along with CBS overexpression (figure 6C). Further real-time PCR analysis verified that only Jun proto-oncogene (JUN), JunB proto-oncogene (JUNB), NK2 homeobox 5 (NKX2-5) and PRRX2 were significantly decreased after CBS overexpression (figure 6D). Moreover, among these four transcription factors, only PRRX2 transfection completely reversed the inhibition of IL-6 by CBS overexpression (figure 6E,F). We also performed a rescue experiment in CBS knockdown Hep3B cells. As shown in figure 6G,H, CBS knockdown significantly facilitated IL-6 expression, but PRRX2 depletion blocked this effect. When PRRX2 was regained by plasmid transfection, the mRNA level of IL-6 was rescued subsequently. In addition, the expression of PRRX2 exhibited significant negative correlation with the expression of CBS in tumor tissues from 236 HCC patients (figure 6I,J). Taken together, these data implied that CBS inhibited the expression of PRRX2. As a potential transcription factor, PRRX2 modulated IL-6 transcription and activated downstream STAT3 in HCC cells.

miRNAs have been demonstrated as important upstream regulators of many tumorigenesis-related genes. To clarify why CBS is downregulated in HCC and elucidate the upstream regulators of CBS, we searched six public databases, including TargetScan, miRanda, miRMap, PITA, RNA22 and RNAhybrid. Among all these six databases, we found that miR-942-5p, miR-139-3p, miR-24-3p and miR-1303 had sequences complementary to CBS 3′-UTR (figure 7A and online supplemental figure 8A), with score values greater than 80 in the TargetScan database. We then examined the expression levels of these four miRNAs in 24 HCC tissues. Further analysis showed that the level of miR-24-3p was inversely associated with CBS mRNA level, whereas the correlation between CBS mRNA level and the other three miRNAs were not significant (figure 7B and online supplemental figure 8B). Also, the level of miR-24-3p was increased in HCC samples compared with peritumor tissues (figure 7C). To verify the hypothesis that miR-24-3p might be an upstream regulator of CBS, we transfected MHCC97H and Hep3B cells with miR-24-3p mimics, and results showed that miR-24-3p overexpression led to a decrease of CBS protein level in both MHCC97H and Hep3B cells (figure 7D,E). Next a luciferase reporter assay was performed to further validate whether miR-24-3p was a direct regulator of CBS. When cotransfected with Wt-CBS 3′-UTR luciferase reporter plasmid, miR-24-3p mimics led to a significant decrease in the luciferase activity of CBS, whereas the luciferase activity of Mut-CBS 3′UTR was not affected by miR-24-3p mimics (figure 7F). In addition, cotransfection with miR-24-3p inhibitor led to a significant increase in the luciferase activity of Wt-CBS 3′-UTR, while this effect was not observed in Hep3B cells transfected with Mut-CBS 3′UTR (figure 7F). Taken together, miR-24-3p directly targeted CBS and negatively regulated the expression of CBS, suggesting that miR-24-3p was an upstream repressor of CBS in HCC.