CCN2–MAPK–Id-1 loop feedback amplification is involved in maintaining stemness in oxaliplatin-resistant hepatocellular carcinoma

A total of 268 patients who underwent curative liver resection for HCC between January 2004 and December 2006 at the Zhongshan Hospital Fudan University were enrolled in this study. Among which, the training set contained 64 cases (including HCC tissues and the paired non-tumor liver tissues from 48 patients were used for real-time PCR and that from 16 patients for Western blot), another set of 184 patients was used for validation (the validation set). Curative resection was defined as the complete resection of tumor nodules, leaving the tumor margins free of cancer upon histologic examination. The histopathologic diagnosis was confirmed by two independent experienced pathologists. Patients were followed after surgical treatment until December 2013, and the median follow-up was 63 months (range 0–110 months). Details of the follow-up procedures were previously described [13]. The clinicopathologic characteristics of HCC patients in the validation set were provided in Supplementary Tables 1 and 2.

cDNA microarrays were performed using the Human OneArray^® (Phalanx Biotech Group, San Diego, CA) to evaluate the alterations of expression profiling. Total RNA was extracted from Hep3B-CCN2 and Hep3B-vector cells and the isolations and microarray analyses were performed in triplicate according to the manufacturer’s instructions. All data were uploaded to the Gene Expression Omnibus (GEO accession number 124529).

The human full-length Id-1 (NM_002165) and CCN2 (NM_001901.2) cDNA were obtained from GeneCards (Shanghai, China) and cloned into the pCDH lentiviral expression vector (System Biosciences, CA, USA). Using the In-Fusion^® HD Cloning Kit (Takara, Tokyo, Japan), the amplified fragment was inserted into the pCDH plasmid (between XbaI and EcoRI sites). Lentiviral shRNA expression plasmids PLVT, PLKO.1, and the target sequences were listed in Supplementary Table 7.

The increased expression levels of CCN2 and Id-1 in the subcutaneous oxaliplatin-resistant tumors were confirmed by immunohistochemical staining (Fig. 1a). Significantly increased expression of vimentin, CD44, aldehyde dehydrogenase 1 (ALDH1), and epithelial cell adhesion molecule (EpCAM), which are related to epithelial–mesenchymal transition (EMT) and stemness, were also upregulated in subcutaneous oxaliplatin-resistant tumors (Supplementary Fig. 1). The expression levels of Id-1 and CCN2 in the oxaliplatin-resistant HCC cell lines, MHCC97H-OXA and Hep3B-OXA, were demonstrated to be significantly increased by Western blot and real-time PCR (Fig. 1b). The expression levels of Id-1 and CCN2 were investigated in seven HCC cell lines with different malignancy phenotypes of oxaliplatin resistance (IC₅₀: HCC-M3, 32.10 ± 3.28; HCC-97H, 28.98 ± 2.37; PLC, 7.10 ± 1.49; Hep3B, 4.52 ± 0.88; HepG2, 7.01 ± 0.75; Huh7, 3.86 ± 0.58; L02, 5.78 ± 0.32). Both the Id-1 and CCN2 levels were significantly increased in HCC cell lines with high metastatic potentials and strong resistance to oxaliplatin (HCC-LM3 and HCC-97H), whereas relatively lower CCN2 and Id-1 levels were detected in HCC cell lines with low metastatic potential and weak oxaliplatin resistance (PLC, HepG₂, Huh7, and Hep3B) and in the human liver cell line (L-02; Fig. 1c). Both Id-1 and CCN2 levels were also obviously increased in HCC patients who were resistant to transcatheter arterial chemoembolization (TACE) compared with those TACE-susceptible patients (Fig. 1d; Supplementary Fig. 2).

We detected the mRNA expression levels of Id-1 and CCN2 in 48 HCCs and their adjacent non-tumor liver tissues, and found increased Id-1 expression in 70.83% (34/48), and increased CCN2 expression in 64.58 (31/48) of HCC tissues compared with the corresponding non-tumor liver tissues (Fig. 2a, b). These results were validated in proteins level of 16 randomly selected HCC samples, and a positive correlation between the expression levels of Id-1 and CCN2 was demonstrated (Fig. 2c).

To illustrate the clinical role of Id-1 and CCN2 in HCC, 184 patients in the validation cohort were sorted according to the Id-1 and CCN2 expression levels (Supplementary Fig. 3A, B). The patients in the Id-1^high group had significantly lower overall survival (OS) and higher cumulative recurrence rates (CCR) compared with those in the Id-1^low group (Fig. 2d). The patients in the CCN2^high group also had significantly lower OS and higher CCR compared with those in the CCN2^low group (Fig. 2e). We then classified the patients into three subgroups according to CCN2 and Id-1 expression levels. Group I had low expression levels of both CCN2 and Id-1, group II had high expression of either CCN2 or Id-1, and Group III had high expression of both CCN2 and Id-1. The patients in group I had the best prognosis, their OS rate was significantly higher than that of the patients in groups II and III, and their CCR was significantly lower (Fig. 2f).

Cox regression analysis revealed that increased Id-1 expression was only significantly correlated with tumor differentiation. No significant association was found between Id-1 expression and the other clinico-pathological characteristics including age, gender, hepatitis B virus (HBV) surface antigen (HBsAg) status, cirrhosis, serum alpha-fetoprotein (AFP) and alanine aminotransferase (ALT) levels, tumor dimension, tumor number, vascular invasion, or tumor encapsulation (Supplementary Table 1). Cox regression analysis also revealed that increased CCN2 expression was significantly correlated with tumor number and tumor differentiation. However, no significant association was found between CCN2 expression level and the other clinical and pathological characteristics (Supplementary Table 2).

A univariate analysis revealed that tumor size, tumor number, vascular invasion, Id-1 and CCN2 expression were significantly associated with the post-operative OS and CCR of HCC patients. However, no prognostic significance was found for the other characteristics including age, gender, HBsAg, HCV-Ab, liver cirrhosis, AFP, ALT, tumor encapsulation or differentiation (Supplementary Table 3). The multivariate Cox proportional hazards model revealed that tumor size, number of tumor nodules, and Id-1 and CCN2 expression were also independent prognostic indicators for OS and CCR of HCC patients; however, vascular invasion was an independent prognostic indicator for OS in HCC patients (Supplementary Table 4).

To confirm the prognostic value of Id-1 and CCN2 for HCC, we detected its protein levels in frozen tissue samples from HCC patients by immunoblotting, and found a significantly increased Id-1 and CCN2 protein levels in HCC tissues with early recurrence compared with the others. In addition, increased Id-1 and CCN2 expression was found to be associated with poor differentiation of HCC (Supplementary Fig. 4).

To evaluate the exact function of Id-1, we stably overexpressed and silenced Id-1 expression in HCC cells. The upregulation of Id-1 significantly enhanced oxaliplatin resistance which could be reversed after Id-1 expression was silenced (Fig. 3a). The migration, invasion, and sphere formation abilities of MHCC-97H cells were also significantly enhanced following overexpression of Id-1, and were inhibited after Id-1 silencing (Supplementary Fig. 5). The in vivo subcutaneous tumor growth capacity of MHCC-97H cells transfected with Id-1 shRNA in nude mouse models was significantly diminished, whereas it is significantly enhanced when the Id-1 was upregulated through lentiviral transfection in MHCC-97H cells (Fig. 3b). Moreover, all of the mice with subcutaneous implantation of MHCC-97H cells transfected with Id-1 expression vector exhibited pulmonary metastasis, with an average of more than five metastatic nodules per lung, whereas none of the mice subcutaneous implantation of MHCC-97H cells transfected with control vector was found to have lung metastases (Supplementary Fig. 6). Alterations of the key stemness-related molecules in subcutaneous tumor tissues were investigated by immunohistochemistry. In the Id-1-sh group, the expression levels of CD44, EpCAM, Osteopontin (OPN) and ALDH were significantly downregulated compared with the Mock group, and were significantly upregulated after Id-1 expression was overexpressed (Supplementary Fig. 7). These indicate that Id-1 is significantly involved in oxaliplatin resistance and stemness in HCC.

To evaluate the role of CCN2 in HCC, we stably silenced and restored CCN2 expression in HCC cells. Among of the three CCN2-shRNA, CCN2-sh1 was found to exert the most efficient interference of CCN2 compared with MHCC-97H-mock cells, and restoration of CCN2 expression could rescue the altered expression of CCN2 (Fig. 3c; Supplementary Fig. 8). The downregulation of CCN2 significantly inhibited the oxaliplatin resistance which could be reversed when CCN2 expression was restored. The downregulation of CCN2 expression significantly impaired the invasiveness, migration, proliferation and sphere formation abilities compared with the controls. Furthermore, after rescuing CCN2 expression, the impaired abilities were significantly restored (Supplementary Fig. 9). The in vivo subcutaneous tumor growth capacity of MHCC-97H cells transfected with CCN2 shRNA in nude mouse models was significantly diminished, whereas it is significantly enhanced when the CCN2 was rescued (Fig. 3d).

To investigate the potential contribution of CCN2 to the stemness of HCC cells, we compared the gene expression profiles (32,679 genes) of Hep3B-CCN2 and Hep3B-vector using cDNA microarrays. The > twofold differences in expression levels were found in 768 genes, including 581 upregulated and 167 downregulated genes, between Hep3B-CCN2 and Hep3B-vector cells (Fig. 4a). Differentially expressed genes in HCC with CCN2 overexpression were evaluated by Gene Ontology (GO) enrichment in the Gene Set Enrichment Analysis. The top ten significantly enriched GO terms were selected (Fig. 4b), and 14 genes related to cell cycle including CDK1, ROCK2, CCNE2, CCNT2, TTK, PTEN, MAP3K2, MAPK6, SOS1, and RASA1 were determined to be significantly changed. Further evaluation of signaling pathway using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database revealed that the MAPK signaling pathway was significantly enriched, which indicates a common regulatory network between MAPK signaling and cell cycle (Fig. 4c).

Interestingly, spp1, which encodes OPN, a secreted phosphoprotein that plays a crucial role in HCC metastasis [14‐17], was also significantly upregulated in HCC with CCN2 overexpression, whereas CDH1, which encodes E-cadherin and mediates the epithelial phenotype in tumor cells [18], was significantly downregulated. Many other significantly altered genes that play important roles in tumor progression were also identified to be closely related to CCN2 (all data were uploaded into GEO).

To investigate the effect of CCN2 on Id-1 expression and the associated MAPK/Erk signaling, we overexpressed CCN2 in Hep3B cells, and found that CCN2 activated the MAPK/Erk signaling cascade with concomitant upregulation and phosphorylation of p-C-RAF, p-MEK, and p-Erk1/2, and ultimately upregulated Id-1 levels. On the other hand, when the CCN2 expression in HCC-97H cells was silenced using specific shRNA, the MAPK/Erk signaling and Id-1 expression were obviously inhibited. Moreover, after rescuing CCN2 expression, the impaired MAPK/Erk signaling and Id-1 expression were reversed (Fig. 5a). In the HCC tissues from subcutaneous implantation model of mice, the knock-down of CCN2 was significantly associated with a decreased protein level of Id-1 detected by immunohistochemical staining, and the impaired Id-1 expression were reversed after rescuing CCN2 expression (Fig. 5b). Taking together, these suggest that CCN2 plays important roles in the regulation of MAPK/Erk signaling cascade and Id-1 expression. To validate this mechanism, we treated HCC cells with recombinant CCN2 in a dose-dependent manner (0–2000 ng/m) and found that the p-Erk and Id-1 expression levels were significantly increased (Fig. 5c). And, sorafenib (2 μmol/L) or the MEK1/2 inhibitor U0126 (10 μmol/L) treatment induced a significant inhibition on MAPK/Erk signaling and the downregulation of Id-1 (Fig. 5d).

In the case of construction of oxaliplatin-resistant HCC cell lines, firstly, we treated the oxaliplatin-resistant MHCC-97H cells with different concentrations of oxaliplatin, and found that the expression levels of CCN2, Id-1, LRP6, p-Erk, and E-cadherin varied in a dose-dependent manner (Supplementary Fig. 10A). And, the variations in CCN2, Id-1, and p-Erk expression levels also occurred in a time-dependent pattern with a significant difference in the maximum time point during the oxaliplatin treatment (Supplementary Fig. 10B).

We re-analyzed the difference between oxaliplatin-resistant and wild-type HCC, and the activation of the MAPK/Erk signaling cascade was demonstrated to be concomitant with the upregulation and phosphorylation of p-C-RAF, p-MEK, and p-Erk1/2, which eventually led to the upregulation of Id-1 (Fig. 5e). When sorafenib or U0126 was added to the oxaliplatin-resistant HCC cells, the MAPK/Erk signaling pathway was impaired and Id-1 expression was downregulated (Fig. 5f). On the other hand, when the CCN2 expression in oxaliplatin-resistant HCC cells was silenced using specific shRNA, the MAPK/Erk signaling and Id-1 expression were obviously inhibited (Fig. 5g). Furthermore, Id-1 overexpression or interference induced an obvious variation of CCN2 protein level in the same trend in HCC cells (Fig. 3a).

As mentioned above, sorafenib and the MEK1/2 inhibitor U0126 could significantly inhibit MAPK/Erk signaling and downregulate the expression of Id-1. We treated MHCC-97H and Hep3B cells with the combination of sorafenib or U0126 with oxaliplatin to investigate their possibilities in reversing oxaliplatin resistance of HCC cells, and found that the combination of sorafenib or U0126 significantly decreases the IC₅₀ of HCC cells to oxaliplatin (Fig. 6a). The synergistic effect of sorafenib, U0126, or CCN2 silencing in oxaliplatin-resistant HCC cells also resulted in inducing a significantly decreased IC₅₀ of HCC cells to oxaliplatin (Fig. 6b).

Furthermore, in the subcutaneous xenograft models established in nude mice using MHCC-97H cells, the combination of sorafenib with oxaliplatin induced a more significant inhibitory effect on in vivo tumor growth compared with the oxaliplatin alone (Fig. 6c). In the subcutaneous xenograft models of C57BL/6 mice established with Hep1-6 cells, oxaliplatin did not exhibit an inhibitory effect on tumor growth, but when in combination with sorafenib, oxaliplatin exhibited a significant inhibitory effect on in vivo growth of HCC (Fig. 6d).