Glycolysis-related metabolites and genes were highly expressed in HCC cases with low miR-192-5p expression
We first investigated the metabolic features in HCCs with low miR-192-5p levels using HCC Cohort 1 with available metabolome and transcriptome data (Fig. 1a). We performed an integration analysis of miR-192-5p with the global metabolome in tumor tissues among 22 HCC patients, and found that 17 metabolites were significantly correlated with miR-192-5p (|r-value| >0.4, Fig. 1b, Supplementary Table S3). Among them, 7 metabolites presented |r-value| >0.5 and three of them were glycolysis-related metabolites, i.e., Glucose-6-Phosphate (G6P), Fructose-6-Phosphate (F6P), and Nicotinamide Adenine Dinucleotide Phosphate (NADPH). Meanwhile, 652 genes were significantly correlated with miR-192-5p with |r-value| >0.4, revealed by an integration analysis of miR-192-5p with mRNA transcriptome in tumor tissues among 176 HCC patients (Fig. 1c). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis using these genes revealed 13 enriched metabolic features (p<0.001), three of which were associated with glycolysis and glycolysis-related pathways. These results suggest an altered glycolytic feature in HCC cases with low miR-192-5p expression.
Available glycolysis-related metabolites and genes in our profiling data (Fig. 1d) were then compared between HCCs with high miR-192-5p levels (termed HCC192High) and HCCs with low miR-192-5p levels (termed HCC192Low), based on a miR-192-5p median cut-off in HCC tumors. Levels of G6P, F6P, and NADPH were significantly higher in tumors from HCC192Low patients than HCC192High patients, while no difference was found in their non-tumor tissues (Fig. 1e, Supplementary Fig. S1a). Consistently, many genes coding for key glycolytic enzymes such as GLUT1, HK2, PFKFB3, PFKP, and PKM2 were significantly upregulated in HCC192Low tumors compared to HCC192High tumors (Fig. 1e) but showed negligible alteration in their non-tumor tissues (Supplementary Fig. S1b). MCT1 was used as a negative control due to its main role in lactate import, but not in glycolysis[34, 35]. These results demonstrate a hyperglycolytic metabolic feature in HCC cases with low miR-192-5p level.
We further investigated the hyperglycolytic feature in CSC+ HCC cases, i.e., cases with the top quartile expression of CSC markers as previously defined [23]. Hierarchical clustering analysis with glycolytic genes in Cohort 1 revealed two HCC subgroups with different expression levels of glycolytic genes. Consistently, in HCC subgroup with high expression levels of glycolysis-related genes, miR-192-5p level was low while various groups of CSC+ HCC cases were enriched (Fig. 1f). Statistical analysis also showed that glycolytic genes were expressed at significantly higher levels in various groups of CSC+ HCCs than in CSC- HCCs, but no difference was observed in the comparisons of their non-tumor tissues (Supplementary Fig. S1c-d). Comparable data were observed in Cohort 2 with 372 HCC patients (Supplementary Fig. S1e). Together, these data indicated that the hyperglycolytic feature was present in various groups of CSC+HCCs with low level of miR-192-5p.
HCC cells with miR-192-5p loss were hyperglycolytic.
We next investigated the role of miR-192-5p in regulating glycolysis. In Huh7, glycolytic genes with significant differential expression between HCC192High and HCC192Low patients were examined and eight out of nine genes showed a significant up-regulation after suppressing miR-192-5p (Supplementary Fig. S2a). Meanwhile, Huh7 cells with suppressed miR-192-5p exhibited a distinctly increased ECAR and a reduced OCR (Fig. 1g). The extracellular acid produced by cells is derived from lactate produced by glycolysis and CO2 produced during respiration. OCR mainly represents mitochondrial respiration. Therefore, the increased ECAR in Huh7 cells with suppressed miR-192-5p was mainly due to lactate produced from glycolysis but not CO2 from mitochondrial respiration. Consistently, over-expressed miR-192-5p in HLF and HLE cells lowered the extracellular acid production from glycolysis as shown by a reduced ECAR but an increased OCR (Fig. 1h). Most examined glycolytic genes were significantly reduced by miR-192-5p overexpression in both HLF and HLE cells (Supplementary Fig. S2b). These results demonstrate an important role of miR-192-5p in modulating a Warburg-like effect in HCC cells.
To better elucidate the role of miR-192-5p in regulating glycolysis, we established miR-192-5p knockout (termed 192KO) clones from two human HCC cell lines HLF and HLE. A 69bp DNA fragment was deleted in the 192KO mixture clones as well as selected single 192KO clones (Supplementary Fig. S2c-d). One 192KO single clone from each HCC cell lines was used and miR-192-5p expression was undetectable in HLE-192KO and HLF-192KO cells (Fig. 2a). As a control, the expression of miR-194, a nearby miRNA of miR-192-5p, was not affected. As expected, HLF-192KO cells displayed significantly increased CSC features, such as increased populations of CD44+, CD24+ and EpCAM+ CSCs (Fig. 2b, Supplementary Fig. S2e); increased mRNA levels of multiple CSC biomarkers and reduced expression of a differentiation-related gene CYP1A2 (Supplementary Fig. S2f); and enlarged and more spheroid formation (Supplementary Fig. S2g). HLE-192KO cells displayed increased CSC features at a moderate level (Fig. 2b, Supplementary Fig. S2e-f). Consistently, these two 192KO lines also showed the hyperglycolytic features. As shown in Fig 2c, five glycolytic enzymes, i.e., GLUT1, HK2, PFKFB3, ALDOA, and PKM2, as well as c-Myc presented higher levels in 192KO cells than in wild-type cells. 192KO lines also exhibited increased ECARs but decreased OCRs (Fig. 2d), indicating that miR-192-5p loss largely increased the glycolysis-related extracellular acidification. Consistent data were also noticed in other 192KO clones of both HLF and HLE HCC cell lines (Supplementary Fig. S3a-b).
Furthermore, overexpressed miR-192-5p in HLF-192KO cells significantly reduced CSC features and lactate accumulation in the culture medium (Fig. 2e). As a control, the intracellular lactate remained unchanged. Meanwhile, over culturing time, lactate gradually accumulated in the medium and was significantly higher in both HLF-192KO and HLE-192KO cells compared to their corresponding wild-type cells, which could be lowered by overexpressed miR-192-5p (Fig. 2f). A lower pH value was also observed in 192KO cells indicated by the orange/yellow medium vs. the pink medium of wild-type cells at 72 hours after seeding. Comparable data on lactate production were seen in other HLF and HLE 192KO clones (Supplementary Fig. S3c). Consistent data were also obtained through the detection of lactate using non-targeted metabolomics in HLF cells with different expression of miR-192-5p in both internal cells and culture medium (Supplementary Fig S3d). Together, miR-192-5p loss in HCC cells led to a hyperglycolytic phenotype.
HCC cells with miR-192-5p loss had high glucose consumption
We further examined glucose consumption among HCC cells with different levels of miR-192-5p as well as between HCC cells and their co-cultured non-tumor cells. As shown in Fig. 3a, both HLF-192KO and HLE-192KO cells exhibited significantly higher glucose consumption than HCC cells overexpressing miR-192-5p. In Huh7 cells, suppressing miR-192-5p increased their glucose usage (Supplementary Fig. S4a). Consistently, 192KO cells were more sensitive after exposure to 2-DG, a glucose analog, as shown by the significantly reduced cell viability compared to cells with miR-192-5p expression (Fig. 3b). Comparable data were obtained in HuH7 cells (Supplementary Fig. S4b).
In co-culture systems of HCC cells with LX2, HL7702, and THP1, we further compared their glucose uptake via 2-NBDG uptake assay. HLF HCC cells infected with pmiR-ctrl/RFP or pmiR-192/RFP lentiviruses were used, and red fluorescent labeling efficiency was nearly 100% (Supplementary Fig. S4c). In this system, with or without co-culturing with other cells, HLF-192KO cells consistently showed higher 2-NBDG uptake than wild-type cells (Fig. 3c-d). In contrast, LX2 and HL7702 in co-culture with HLF-192KO cells exhibited lower 2-NBDG uptake compared to those in co-culture with HLF-WT cells. Moreover, forced-expression of miR-192-5p in HLF-192KO cells reduced the 2-NBDG uptake in HLF cells but increased 2-NBDG uptake in LX2 and HL7702 cells in the co-culture system. The alteration of 2-NBDG uptake was not observed in THP1 from our co-culture system (Fig. 3d, Supplementary Fig. S4d). Similar data were seen in HLE cells as well as in HLE cells co-cultured with LX2 and HL7702 (Supplementary Fig. S4e-f). These results demonstrate that HCC cells with loss of miR-192-5p actively utilize glucose from their environment to ensure a hyperglycolysis status.
GLUT1, PFKFB3 and c-Myc were miR-192-5p bona fide targets and contributed to glycolytic and stemness features of HCC cells
To investigate the target genes of miR-192-5p in regulating glycolysis flow, we assessed genes negatively correlated with miR-192-5p in 176 HCC cases (r<-0.3) and significantly up-regulated in HCC192Low tumors versus HCC192High tumors (log2fold >0.2, p<0.01). Among these 554 genes, two main groups were observed. One group contained genes related to cell migration as we previously reported [23]. The other group included eight glycolysis-related genes (Fig. 4a) and three of them (GLUT1, HK2, and PKM2) were reported targets of c-Myc, an important regulator of glycolysis[36, 37].
Next, using TargetScan and manual miRNA target prediction, we found that four of these glycolytic genes (PFKFB3, GLUT1, MCT4, and MYC) contained miR-192-5p binding sites in their 3’UTR and/or coding regions (Fig. 4a). In HLF and HLE cells, miR-192-5p overexpression reduced the protein levels of PFKFB3, GLUT1, and c-Myc, but not that of MCT4 (Fig. 4b). Further, the predicted miR-192-5p binding regions of these three genes were cloned into a luciferase reporter and forced expression of miR-192-5p reduced the luciferase activities when the wild-type sequences for PFKFB3 and GLUT1 as well as the #2 binding site of MYC were present (Fig. 4c). These effects were significantly reduced when the corresponding miR-192-5p binding sites were mutated. Moreover, silencing PFKFB3, GLUT1, or c-Myc with 2 siRNAs for each gene reduced ECAR in both HLF-WT and HLF-192KO cells (Fig. 4d, Supplementary Fig. S5a). In HLF-192KO cells, silencing PFKFB3, GLUT1 or c-Myc notably reduced the ECAR rate to a level similar to that of HLF-WT cells with silencing of these genes. Comparable data were also noticed in HCC patients from Cohorts 1 and 2. PFKFB3, GLUT1, and MYC presented higher levels in HCC192Low tumors compared to HCC192High tumors (Supplementary Fig. S5b-c). These results indicate that PFKFB3, GLUT1, and c-Myc are miR-192-5p targets and are involved in the hyperglycolysis caused by miR-192-5p loss.
PFKFB3, GLUT1, and c-Myc were reported to maintain stemness features in cancer at certain levels [38-41]. Consistently, si-PFKFB3, si-GLUT1 or double knockdown led to reduced levels of four CSC biomarkers, i.e., CD44, CD24, EpCAM and CD90, as determined by RT-qPCR (Fig. 4e). Flow cytometry analysis also showed that si-PFKFB3 and GLUT1 reduced the populations of CD44+ and CD24+ CSCs (Fig. 4f). Meanwhile, si-MYC seemed to only reduce CD44+ CSCs moderately, but not the CD24+ CSC population (Supplementary Fig. S5d). Together, these data demonstrate that three glycolytic regulators, PFKFB3, GLUT1 and c-Myc were bona fide targets of miR-192-5p, and they contributed to both hyper-glycolysis and CSC features of HCCs caused by loss of miR-192-5p.
c-Myc suppressed miR-192-5p transcription, ensuring a positive feedback of high c-Myc/low miR-192-5p in hyperglycolytic CSC+HCCs
In our previous miRNA profiles of tumors and non-tumors from a hydrodynamic injection HCC FVB mouse model[31], miR-192-5p expression was significantly reduced in c-Myc-induced HCCs (Fig. 5a). Further, in a hydrodynamic injection HCC ICR mouse model, miR-192-5p level was also reduced >100 times in c-Myc induced HCCs but was not much in Ras-induced HCCs when compared to corresponding non-HCC liver tissues. In four different HCC cell lines, si-MYC led to an increased expression of miR-192-5p, while forced expression of c-Myc reduced the level of miR-192-5p (Fig. 5b). These data indicate that c-Myc might regulate miR-192-5p transcription.
Consistently, among four different lengths of miR-192-5p promoter regions, the -266 nt to +186 nt region showed the strongest promoter activity (Fig. 5c) and the miR-192-5p promoter activity (-266 nt to +186 nt) was reduced by exogenous c-Myc, while enhanced by si-MYC (Fig. 5d). Wild-type p53 could bind to the miR-192-5p promoter region and induce its expression[23, 42]. Consistently, in HepG2 cells with wild-type TP53, the expression of miR-192-5p was induced by p53 via exposure to Nutlin-3a (an MDM2 antagonist to stabilize p53) and reduced by silencing of p53 (Fig. 5e, Supplementary Fig. S6a-b). Over-expressed c-Myc significantly suppressed miR-192-5p expression in HCC cells with either activated p53 or silenced p53, indicating that c-Myc-mediated miR-192-5p down-regulation was independent on p53.
Comparable data were noticed in HCC patients. In both HCC cohorts, hierarchical clustering analysis revealed two subgroups with distinct c-Myc activation status based on 76 c-Myc target genes from the online Human MYC Targets Profiler (Supplementary Fig. S6c-d). In Cohort 1, miR-192-5p expression in the c-Myc activation subgroup was significantly lower than that in c-Myc non-activation subgroup (Fig. 5f). In Cohort 2, miR-192-5p expression was always significantly lower in each c-Myc activation subgroup than in the corresponding non-activation subgroup, which was independent of TP53 mutation and mir-192 promoter methylation (Fig. 5f-g). In addition, different groups of CSC+HCCs consistently presented a low level of miR-192-5p, a high level of c-Myc activation and high frequency of MYC amplification (Supplementary Fig. S7). Together, c-Myc suppressed miR-192-5p transcription, which led to a positive feedback of high c-Myc/low miR-192-5p in CSC+HCC cells with glycolytic feature.
Overproduced lactate from CSC+HCCs activated the ERK pathway in environmental non-tumor cells, and this effect further increased HCC cell stemness and malignancy features
As the end product of glycolysis, the continuously produced lactate in hyperglycolytic miR-192-5p-loss HCC cells might affect their environment and contribute to HCC malignancy. The transport of lactate across the plasma membrane is mainly catalyzed by MCT1 and MCT4, with MCT1 typically involved in the import while MCT4 in export of lactate[34, 35]. In HCC patients, the expression ratio of MCT1 vs. MCT4 showed no difference between tumor and non-tumor tissues of Cohort 1 but was significantly higher in non-tumor tissues than tumor tissues of Cohort 2 (Supplementary Fig. S8a), indicating the possibility of lactate uptake by environmental non-tumor cells. Lactate could also stabilize NDRG3, which in turn activated the ERK pathway to promote cell malignancy[43]. Consistently, lactate treatment stimulated ERK phosphorylation noticeably in non-tumor cells of HCC microenvironment, i.e., LX2, THP1 and HL7702 cells (Fig. 6a).
In a chamber co-culture system, pERK level was increased in LX2 and THP1 cells when co-cultured with HLF-192KO cells compared to when co-cultured with HLF-WT (Fig. 6b). Moreover, pERK was further reduced in LX2 and THP1 when exposed to HLF-192KO cells with miR-192-5p overexpression (Fig. 6b). Similar results were observed from LX2 and THP1 cells co-cultured with HLE cells (Supplementary Fig. S8b). However, pERK was unaltered in HL7702 cells in this co-culture system. Thus, HCC cells with miR-192-5p loss could actively affect certain non-tumor cells via increased production of lactate.
We have also found that lactate-induced pERK in environmental non-tumor cells partially relied on NDRG3 and MCT1 (Fig. 6c-e). LX2 and THP1 cells expressed relatively high levels of MCT1 and NDRG3 (Fig. 6c). Silencing NDRG3 or MCT1 in LX2 and THP1 cells reduced the level of lactate-induced pERK (Fig. 6d-e). Low expression levels of MCT1 and NDRG3 in HL7702 were consistent with its minor response to lactate (data not shown).
We then explored the effects of an altered lactate/MCT1/NDRG3/pERK axis in LX2 or THP1 cells on the malignancy features in HCC cells (Fig. 7a). HLF-192KO/RFP cells were co-cultured with LX2 cells pre-transfected with si-Ctrl, or si-NDRG3, or si-MCT1 (termed LX2si-Ctrl, LX2si-NDRG3, and LX2si-MCT1, respectively). Wound-healing assay of red-fluorescence HCC cells showed that cell migration of HLF-192KO cells was slower upon co-culture with LX2si-NDRG3 or LX2si-MCT1 than with LX2si-Ctrl (Fig. 7b). Consistent data were observed in HLF-192KO cells co-cultured with THP1 (Fig. 7b). Moreover, spheroid assays of HLF-192KO cells were performed in different conditioned medium settings. The number of spheroids of HLF-192KO cells was significantly lower under exposure to conditioned medium from co-culture of HLF-192KO with LX2si-NDRG3 or LX2si-MCT1 than from co-culture of HLF-192KO with LX2si-Ctrl (Fig. 7c). CD44+ and CD24+ HLF-192KO populations were also significantly reduced when they were co-cultured with LX2si-NDRG3 or LX2si-MCT1 (Fig. 7d, Supplementary Fig. S8c). As a control, HLF-192KO with overexpressed miR-192-5p did not exhibit significant alteration of migration, spheroid formation and CSC populations when co-cultured with different LX2 cells (Fig. 7b-d). Consistent data were obtained using a second set of siRNAs for NDRG3 and MCT1 (Supplementary Fig. S9). Therefore, blocking the lactate/ERK pathway in HCC microenvironmental cells suppressed the malignancy and stemness features of HCC cells in co-culture experiments.
In both HCC cohorts, patients were divided into four groups based on miR-192-5p expression in their tumor tissues (HCC192Low and HCC192High, medium cut-off) and levels of NDRG3 and MCT1 in non-tumor tissues (NTHigh_NDRG3 or MCT1 and NT Low_NDRG3 and MCT1, medium cut-offs). There was no expression difference of NDRG3 or MCT1 in non-tumor tissues between HCC192Low and HCC192High patients (Fig. 7e, Supplementary Fig. S10a). In Cohort 1, patients with HCC192High NTLow_NDRG3 and MCT1 had the best prognosis, as shown by a prolonged time to recurrence and overall survival. In HCC192Low subgroup, patients with NTHigh_NDRG3 or MCT1 had worse prognosis compared to patients with NTLow_NDRG3 or MCT1 (Fig. 7f). Similar but less significant data were obtained in Cohort 2, which might be due to the limited number of patients (n=49) with available non-tumor mRNA data (Supplementary Fig. S10b). GSEA analysis in HCC192Low patients revealed that several stem cell related gene-sets were enriched in patients with NTHigh_NDRG3 or MCT1 (Supplementary Fig. S10c-d). Together, HCC cells with miR-192-5p loss exhibited a highly malignant feature when they were surrounded by environmental non-tumors with high MCT1 or NDRG3 expression.