Epigenetic Induction of Mitochondrial Fission Is Required for Maintenance of Liver Cancer–Initiating Cells

Mitochondria function as energy stations for adenosine triphosphate and reactive oxygen species (ROS) production, which are required for proliferating cells, including stem cells and cancer cells (1–3). To allow cells to adapt to various stressful circumstances, mitochondria form a dynamic network via the opposing processes of fission and fusion, which divide or fuse the two lipid bilayers that surround mitochondria, resulting in a decrease or increase in energy production, respectively (3, 4). Recently, it has been demonstrated that mitochondrial fission contributes to strengthening the self-renewal capacity in both induced pluripotent stem (iPS) cells and cancer-initiating cells (CIC) via the asymmetric allocation of healthy and oxidatively damaged mitochondria (5, 6). Importantly, inhibition of mitochondrial fission using mdivi-1 suppresses the stem-like traits of CICs, suggesting that mitochondrial fission may be a promising target for cancer therapy (6, 7). However, because of its critical role in normal stem cells and CICs, targeting mitochondrial fission directly would disrupt somatic stem cell activity, resulting in the loss of tissue homeostasis and cause developmental defects. Therefore, it is of great clinical value to unveil the aberrant molecular control of mitochondrial fission in CICs that is distinct from that of normal stem cells, which may provide a new paradigm to develop more effective treatment strategies for cancers.

In response to metabolic demands and stressors, mitochondrial fission aids in allocating mitochondrial contents, enabling asymmetric division, and eliminating dysfunctional mitochondria via mitophagy (3, 8, 9). During the fission process, mitochondrial fission protein dynamin–related protein 1 (DRP1) is recruited from the cytosol to mitochondria by interaction with different adapters located on the outer mitochondrial membrane, such as mitochondrial fission factor (MFF) and FIS1 (10–12), and forms a ring-like structure around mitochondria, which subsequently initiates mitochondrial fission via its GTPase activity (13, 14). Recently, multiple studies have documented that mitochondrial fission is involved in asymmetric cell division, pluripotency, and the fate of stem cells. It is reported that mitochondria become spherical shaped during iPS reprogramming (15), whereas silencing DRP1 impairs both mitochondrial fragmentation and generation of iPS cell colonies (13). Clearance of dysfunctional mitochondria by mitophagy can avoid senescence and prevent stem cell exhaustion (16–18), and inhibition of fission by mdivi-1 disrupts the asymmetric apportioning mitochondrial and caused loss of stem cell properties in the progeny cells (16).

Notably, mitochondrial fission regulators also play vital roles in self-renewal and tumor-initiating capability of CICs (5, 7, 19). For instance, CDK5-mediated DRP1 hyperactivation promotes mitochondrial fission and enhances tumorigenicity in brain tumor–initiating cells (BTIC; ref. 7). AMPK/FIS1-mediated mitochondrial fission/mitophagy is crucial for self-renewal of human AML stem cells, and BRD4/MFF axis–induced mitochondrial fission enables asymmetric division and the self-renewal ability in prostate CICs (5, 19). These studies indicate that mitochondrial fission may be involved in tumorigenesis.

Transcription factor TBX19, characterized by a highly conserved DNA binding motif (T-box), is reported to be upregulated in colorectal cancer and pituitary corticotroph tumors and KRAS-mutant cells (20, 21). However, the clinical significance and biological role of TBX19 in cancer remain largely unknown. Herein, we revealed that TBX19 formed complex with PRMT1 and epigenetically upregulated MFF in liver cancer–initiating cells (LCIC), resulting in increased mitochondrial fission and mitophagy, and enhanced tumor-initiating capability. Pharmacologic inhibition of PRMT1 impaired mitochondrial fission and inhibited the tumor-forming ability of LCICs. Thus, our results revealed the underlying mechanism of mitochondrial fission in LCICs, which provided a new paradigm to develop promising therapeutic strategies for liver cancer treatment.

The hepatocellular carcinoma (HCC) cell lines, including Hep3B, SNU449 were purchased from the ATCC and cultured in DMEM (GIBCO) supplemented with 10% FBS (GIBCO), 100 U/mL penicillin and 100 μg/mL streptomycin (Invitrogen). EpCAM⁺/CD133⁺ LCICs were cultured in ultra-low attachment plates with RPMI-1640 medium, supplemented with EGF (20 ng/mL), bFGF (20 ng/mL), 1 × B27 and 1× antibiotic/antimycotic, in 4% O₂ and 5% CO₂. All the cell lines were tested for Mycoplasma contamination and were authenticated by short tandem repeat fingerprinting at Medicine Laboratory of Forensic Medicine, Department of Sun Yat-Sen University (Guangzhou, China).

Two clinical human liver cancer tissues were diagnosed and obtained from Sun Yat-sen University Cancer Center according to the Declaration of Helsinki, and each patient signed a written informed consent for all the procedures. The study protocols were approved by the Institutional Research Ethics Committee of Sun Yat-sen University Cancer Center for the use of these clinical materials for research purposes.

The EpCAM⁺/CD133⁺ LCICs were isolated from Hep3B and SNU449 cell lines and two clinical HCC primary tissues as previously described (22, 23). Briefly, liver cancer specimens were immediately digested with collagenase and DNase to obtain cell suspensions, and further washed and adjusted to a concentration of 2 × 10⁷ cells/mL. These cells were incubated with antibodies against EpCAM (1:100; ab237396, Abcam) and CD133 (1:100; ab252126, Abcam) and sorted by FACS to isolate EpCAM⁺/CD133⁺ LCIC and EpCAM⁻CD133⁻ non-LCIC populations.

Streptavidin affinity purification of dCas9-Captured DNA and proteins was performed following the published protocol (24). Briefly, EpCAM⁺/CD133⁺ LCICs and EpCAM⁻/CD133⁻ non-LCICs (5 × 10⁷) transfected with FB-dCas9 plasmid and MFF promoter sgRNAs were harvested and treated with 1% formaldehyde to cross-link the proteins to DNA. The cells were lysed in 1 mL Lysis buffer 1 (1 mol/L HEPES-KOH, 1 mol/L NaCl, 0.5 mol/L EDTA, 50% glycerol, 10% NP-40, 10% Triton X-100, and pH 7.5), and rotated for 15 minutes at 4°C, then centrifuged at 2,300 × g for 5 minutes at 4°C to isolate the nuclei, which were suspended in 1 mL of lysis buffer 2 (1 mol/L Tris-HCl, 1 mol/L NaCl, 0.5 mol/L EDTA, 0.5 mol/L EGTA, and pH 8.0). Cell lysates were centrifuged at 2,300 × g for 5 minutes at 4°C to isolate precipitate and suspended in 1 mL of lysis buffer 3 (1 mol/L Tris-HCl, 1 mol/L NaCl, 0.5 mol/L EDTA, 0.5 mol/L EGTA, 5% Na-Deoxycholate, 5% N-lauroylsarcosine, and pH 8.0). The cells were subjected for sonication to shear chromatin fragments and centrifuged at 12,000 × g for 10 minutes at 4°C. Supernatant was then incubated with 50 μL of Streptavidin T1 Dynabeads (Thermo-Fisher Scientific) at 4°C overnight. After overnight incubation, Dynabeads were washed twice with 1 mL of Wash buffer (1 mol/L HEPES-KOH, 5 mol/L LiCl, 0.5 mol/L EDTA, 5% Na-Deoxycholate, 10% NP-40, and pH 8.0), and twice with 1 mL of TE buffer (1 mol/L Tris-HCl, 0.5 mol/L EDTA, 1 mol/L NaCl, and pH 8.0). The proteins were separated by SDS-PAGE and analyzed by Western blot and IP-MS analysis (Shanghai Applied Protein Technology Co. Ltd.).

Quantification of mitochondrial morphology followed as reported previously (25). Briefly, cells from more than 10 randomly selected fields were imaged and analyzed per sample from three independent experiments, and it was done by 2 researchers who were blinded to genotype and treatment of the samples. Cells were divided into three categories based on mitochondrial morphology stained by MitoTracker probe dye: “Tubulated” is with a majority of mitochondria in a cell forming interconnected and elongated morphology with mitochondrial length >10 μm; “Intermediate” is with mixed short tubular mitochondria with mitochondrial length <10 μm in a cell; “Fragmented” is with a majority of punctiform mitochondria in a cell.

All of the animal procedures were complied with all relevant ethical regulations for animal testing and research and the ethical approval was approved by the Sun Yat-sen University Animal Care Committee. Different numbers (5 × 10⁵, 1 × 10⁴, 5 × 10³, 1 × 10³, 5 × 10², 1 × 10²) of indicated LCICs and non-LCICs were inoculated with Matrigel (final concentration, 25%) subcutaneously into the inguinal folds of NOD/Shi-scid/IL-2Rγ^null (NOG) mice (5–6 weeks old; CREA Japan Inc.). Tumorigenic cell frequency (TIC frequency) was calculated by Extreme Limiting Dilution Analysis (ELDA) using publically available website tool (http://bioinf.wehi.edu.au/software/elda/), and displayed by 95% confidence intervals for the frequency in each population group. The statistical P value was obtained using a χ² test.

Tumors were initiated by implantation of 5 × 10⁵ or 1 × 10² EpCAM⁺/CD133⁺ LCIC and EpCAM⁻/CD133⁻ non-LCIC populations derived from patients, coated with Matrigel and media in a 1:1 ratio, into liver of NOG mice (5–6 weeks old; CREA Japan Inc.). Orthotopic liver injection was performed using a 25‐μL syringe with a cemented 22‐gauge needle. The NOG mice were anesthetized with 50 mg/kg of pentobarbital and a small incision was made in the abdomen. The left lobe of the liver was gently exposed and the re-suspended cells were carefully injected into the lobe. After injection, the abdominal wound was sutured. For drug treatment, vehicle or furamidine (20 mg/kg) were injected intraperitoneally into mice. The mice were sacrificed when the mice become moribund and tumors were excised and paraffin embedded.

All values are expressed as the mean ± standard deviation of individual samples. n represents the number of independent experiments performed on different mice or different batches of cells or different clinical tissues. Statistical analysis was performed using the Student two-tailed t test and one or two-way ANOVA. Survival curves were plotted by the Kaplan–Meier method and compared by the log-rank test. The P values of <0.05 were considered as statistically significant. Statistical analysis was performed using GraphPad Prism 8.

Additional information is provided in Supplementary Materials and Methods.

To examine the impact of mitochondrial dynamics on LCIC function, we isolated EpCAM⁺/CD133⁺ LCICs and matched EpCAM⁻/CD133⁻ non-LCIC populations from two patient-derived HCC tissues as previously reported (26–28). An in vitro tumorsphere formation assay further confirmed that the EpCAM⁺/CD133⁺ LCICs exhibited higher sphere-forming ability than the EpCAM⁻/CD133⁻ non-LCICs (Supplementary Fig. S1A). However, we did not observe significant growth differences between 2D-cultured LCICs and non-LCICs (Supplementary Fig. S1B), suggesting that the differences of tumorsphere formation were caused by stemness. We further validated the mitochondrial morphology of LCICs and non-LCICs using three different methods, including mitochondrial red fluorescent probe (MitoTracker probe), mitochondrial marker (TOM20), and transmission electron microscopy (TEM). As shown in Fig. 1A and Supplementary Fig. S1C and S1D, mitochondria in matched LCICs displayed more a fragmented and less tubular morphology, whereas non-LCICs exhibited elongated tubular mitochondria with longer cristae and higher cristae density. Meanwhile, expression of total DRP1 and phosphorylated DRP1 (p-DRP1, S616) was found increased on mitochondria and colocalized with TOM20 in LCICs compared with non-LCICs (Fig. 1B–D; Supplementary Fig. S1E). Consistently, the LCICs isolated from liver cancer cell lines Hep3B and SNU449 also exhibited shorter, rounded mitochondria in LCICs as compared with elongated, tubular structures in non-LCICs (Fig. 1A and B; Supplementary Fig. S1A and S1C–S1E). Taken together, these results suggest that LCICs display increased mitochondrial fission.

To determine the mechanism of increased mitochondrial fission in LCICs, we examined 58 mitochondrial dynamics–related mediators, including 39 genes involved in mitochondrial fission and 19 genes involved in mitochondrial fusion. Real-time PCR analyses revealed that expression of MFF, but not DRP1, FIS1, MID49 and MID51, was highly and consistently upregulated in four LCICs isolated from patient-derived HCC tissues and HCC cell lines compared with that in four matched non-LCICs (Fig. 1C; Supplementary Fig. S1F). Immunoblotting (IB) analysis further confirmed that mitochondrial MFF was also markedly increased in LCICs (Fig. 1D). Importantly, MFF-downregulated LCICs not only displayed elongated, tubular mitochondrial structures and reduced total and phosphorylated DRP1 on mitochondria, but also showed decreased the tumor-forming potential (Fig. 1E–G; Supplementary Fig. S2A and S2B). Correspondingly, MFF-overexpressing non-LCICs exhibited more a fragmented mitochondrial morphology and increased total/phosphorylated DRP1 on mitochondrial (Supplementary Fig. S2C and S2D). Overexpression of MFF significantly enhanced the tumorsphere formation ability of non-LCICs, and obviously increased the percentage of the side-population (SP) cells (SP⁺ population), which was a subpopulation of cells based on SP assay that may represent a universal TIC-like phenotype (Supplementary Fig. S2E and S2F). Therefore, these results indicate that MFF-mediated mitochondrial fission enhances LCIC traits.

Furthermore, analysis of The Cancer Genome Atlas (TCGA) revealed that increased MFF expression correlated significantly with HCC grade and associated reversely with relapse-free survival of patients with HCC (n = 345; P = 0.009), even in HCC grade I group (n = 54; P = 0.012; Supplementary Fig. S2G; Fig. 1H), providing further evidence that MFF overexpression contributed to the initiation and progression of liver cancer.

To investigate the molecular mechanism of MFF upregulation in LCICs, the CRISPR affinity purification in situ of regulatory elements (CAPTURE) approach (24) and quantitative mass spectrometry were used to identify the trans-regulatory factors targeting MFF in LCICs (Fig. 2A). Compared with those in the non-LCICs, we found that the levels of 12 potential trans-regulatory factors were increased on the MFF promoter in LCICs (Fig. 2A). However, qPCR analysis revealed that MFF mRNA expression was only significantly decreased in TBX19-, or PRMT1-, or RNA polymerase II (RNAP II)–, or TFIID-silenced LCICs (Fig. 2B). Furthermore, chromatin immunoprecipitation (ChIP) assays showed that the enrichment of TBX19 and PRMT1 on the MFF promoter was increased in LCICs compared with that in non-LCICs (Fig. 2C; Supplementary Fig. S3A). Moreover, IB analysis showed that expression levels of TBX19, PRMT1, and MFF were elevated overtly in all four LCICs derived from two clinical HCC tissues and in Hep3B, SNU449 cell lines compared with that in non-LCICs and liver progenitor cells (LPCs#1, LPCs#2; Fig. 2D). All these results suggested that TBX19 and PRMT1 might regulate MFF expression in LCICs. Indeed, overexpressing TBX19 in non-LCICs dramatically increased expression of mitochondrial MFF and total/phosphorylated DRP1. However, we did not observe the inducing effect of TBX19 on MFF upregulation in LPCs (Fig. 2E). Taken together, our data indicate that TBX19 transcriptionally upregulates MFF in LCICs.

IF staining using the MitoTracker probe showed that the mitochondria became more fragmented and less tubular in TBX19-overexpressing non-LCICs (Fig. 2F; Supplementary Fig. S3B), suggesting that TBX19 contributed to mitochondrial fission. Consistently, the mitochondrial/nuclear ratio (Mito./Nuc. area ratio) and mitochondrial mtDNA, accompanied by mitochondrial ROS, were decreased in TBX19-overexpressing non-LCICs (Fig. 2G and H; Supplementary Fig. S3C). These phenomena were also found in LCICs compared with that in matched non-LCICs (Supplementary Fig. S3D and S3E). These results suggested that TBX19 overexpression might promote mitochondrial clearance.

Previously, it was reported that mitophagy, representing selective mitochondrial autophagy, plays a critical role in mitochondrial clearance and maintenance of stem cells and cancer stem cells (CSC; refs. 29, 30). We therefore examined whether TBX19 overexpression induced mitophagy. TEM analysis showed that mitochondria in TBX19-overexpressing non-LCICs displayed increased numbers of mitochondrial engulfed in autophagosomes (Fig. 2I). Correspondingly, IF staining showed that upregulation of TBX19 significantly increased LC3B-positive signals and the colocalization of LC3B with the MitoTracker probe in LCICs (Fig. 2J). Therefore, these results indicate that TBX19 overexpression promotes mitophagy. However, the inducing effect of TBX19 overexpression on mitochondrial fission was significantly abrogated by MFF ablation (Fig. 2K; Supplementary Fig. S3F), demonstrating that TBX19-mediated mitochondrial fission functions through upregulation of MFF.

Consistently, we found that silencing TBX19 in LCICs significantly decreased the expression of mitochondrial MFF and total/phosphorylated DRP1 (Fig. 3A), and also converted the LCIC mitochondrial morphology into elongated, tubular structures in LCICs (Fig. 3B). Importantly, the TBX19 downregulation induced increasing of mitochondrial content in LCICs was significantly rescued by exogenous MFF transduction, as indicated by reduced ratio of Mito./Nuc. area and mtDNA/nDNA (Fig. 3C,–E). Therefore, these results provided further evidence that TBX19/MFF axis induced mitochondrial fission.

Mitochondrial fission is reported to be involved in self-renewal and tumor-initiating capability of CICs (5–7), which prompted us to examine the effect of TBX19 on the properties of LCICs. Overexpression of TBX19 significantly enhanced the tumorsphere formation ability of non-LCICs, whereas silencing TBX19 significantly decreased the tumor-initiating cell (TIC) traits of LCICs, as indicated by fewer and smaller tumorsphere formation (Fig. 4A; Supplementary Fig. S4A–S4C). Importantly, an in vivo limiting dilution assay showed that TBX19-overexpressing non-LCICs exhibited a significantly higher tumor incidence and tumor-forming capacity, whereas TBX19-silenced LCICs displayed a markedly lower capacity to form tumors (Fig. 4B). SP assay, a flow cytometry method based on the dye efflux properties of ABC transporters was further performed to examine the effect of TBX19 on SP⁺ population, a subpopulation of cancer cells that have been identified in cancers and display increased capacity of self-renewal and tumorigenicity (31, 32). As shown in Supplementary Fig. S4D, silencing TBX19 significantly reduced the percentage of SP⁺ population. In line with the notion that mitochondrial fission–mediated asymmetric stem cell division leads to functional inheritance of stem cells and CSCs (16). We also observed that the frequency of asymmetric cell division was significantly increased in TBX19-overexpressing non-LCICs but reduced in TBX19-silenced LCICs (Supplementary Fig. S4E and S4F). These results demonstrate that TBX19 contributes to the tumor-forming and self-renewal capabilities of HCC cells.

However, TBX19 overexpression–induced tumorsphere formation in non-LCICs was significantly impaired by MFF knockdown (Fig. 4A; Supplementary Fig. S4A). Meanwhile, the inhibitory effects of TBX19 silencing on LCIC traits and asymmetric cell division were rescued by ectopic expression of MFF (Fig. 4C; Supplementary Fig. S4F and S4G). These results indicate that MFF is a prerequisite for TBX19-induced TIC traits in liver cancer cells. Consistent with the effect of silencing MFF, inhibition of DRP1 using siRNAs or mdivi-1 or treated with the mitophagy inhibitor chloroquine or a pharmacologic inhibitor of DRP1 significantly abolished the stimulating effect of TBX19 on TIC traits, as indicated by the decreased SP and number of tumorspheres (Fig. 4D and E; Supplementary Fig. S4H–S4J). Therefore, these data provided further evidence that the TBX19/MFF axis–induced mitochondrial fission contributed to the LCIC traits.

Mitochondrial fission is necessary for both pluripotential and metabolic reprogramming (10), in which a metabolic shift from oxidative phosphorylation (OXPHOS) to glycolysis has been found to be beneficial for stem cells, including CSCs, for self-renew and maintenance of stem cell states (33). Consistently, overexpression of TBX19 significantly suppressed OXPHOS in non-LCICs, as indicated by a reduced oxygen consumption rate (OCR), but increased the glycolytic activity, as demonstrated by the elevated extracellular acidification rate (ECAR; Fig. 4F; Supplementary Fig. S5A), indicating that the metabolic profile of TBX19-overexpressing non-LCICs shifted from OXPHOS to glycolysis. The mitochondrial ROS level also decreased significantly in TBX19-overexpressing non-LCICs, but increased in TBX19-silenced LCICs (Fig. 4G; Supplementary Fig. S5B). Importantly, silencing MFF dramatically abolished the effects of TBX19 on the OXPHOS/glycolysis switch and reversed the change in the mitochondrial ROS level (Fig. 4F,–G; Supplementary Fig. S5A and S5B), suggesting that TBX19/MFF-induced mitochondrial fission promoted metabolic reprogramming from OXPHOS to glycolysis.

Strikingly, the levels of the pluripotency factors OCT4 and NANOG, which are key regulatory genes that initiate stem cell characteristics in liver cancer (23, 34–36), increased significantly in the TBX19-overexpressing non-LCICs, but were decreased in the TBX19-silenced LCICs. However, silencing MFF abrogated the inductive effect of TBX19 on OCT4 and NANOG levels (Fig. 4H; Supplementary Fig. S5C and S5D). Interestingly, TBX19 and MFF had no impact on the OCT4 mRNA level (Supplementary Fig. S5D), suggesting that the TBX19/MFF axis–mediated OCT4 upregulation might act via posttranscriptional regulation. We found that expression of oxidized OCT4 decreased in TBX19-overexpressing non-LCICs and increased in TBX19-silenced LCICs (Fig. 4H; Supplementary Fig. S5C). Meanwhile, treatment with cell-permeable glutathione (GSH), which phenocopied MFF overexpression, also resulted in a reduction in oxidized OCT4 levels (Fig. 4H; Supplementary Fig. S5C). Therefore, these results demonstrate that the TBX19/MFF axis–mediated metabolic switch from OXPHOS to glycolysis results in mitochondrial ROS reduction and prevents ROS-mediated OCT4 degradation.

Analysis of two public HCC datasets (TCGA-LIHC and GSE25097) revealed that TBX19 was significantly upregulated in HCC tissues (Supplementary Fig. S6A and S6B), and patients with higher TBX19-expressing HCC had poorer overall survival and a higher rate of HCC recurrence than those with lower TBX19-expressed HCC (P = 0.001, P < 0.001; P < 0.001, P < 0.001; Supplementary Fig. S6C and S6D). Notably, increased TBX19 levels also correlated with shorter progression-free and disease-specific survival in HCC patient, as well as shorter progression and relapse-free survival in patients treated with sorafenib (P < 0.001, P < 0.001; P < 0.001, P = 0.010; Supplementary Fig. S6E and S6F). Taken together, these results indicate that TBX19 upregulation is associated with worse prognosis and leads to therapeutic failure in HCC.

Our previous dCas9 CAPTURE/MS analysis showed that PRMT1 was also enriched on the MFF promoter and silencing PRMT1 resulted in a reduction in MFF expression (Fig. 2A–C; Supplementary Fig. S3A). Although silencing PRMT1 significantly abrogated the inductive effect of TBX19 on the mRNA and protein levels of MFF in non-LCICs, overexpressing PRMT1 did not increase MFF expression at mRNA and protein levels in TBX19-silenced LCICs (Fig. 5A and B). These results suggested that PRMT1-mediated MFF upregulation was dependent on TBX19. Consistent with this hypothesis, ChIP assays showed that overexpressing TBX19 in non-LCICs significantly increased, but silencing TBX19 in LCICs decreased, enrichment of PRMT1 on MFF promoter (Fig. 5C; Supplementary Fig. S7A). However, dysregulation of PRMT1 has no impact on the association of TBX19 with MFF promoter. Furthermore, reciprocal coimmunoprecipitation (co-IP) analysis and far-Western blotting analyses showed that TBX19 directly interacted with PRMT1 (Fig. 5D and E), and co-IP assays using a series of truncated forms of TBX19 indicated that the N-terminal domain of TBX19 mediated the protein–protein interaction with PRMT1 (Fig. 5F). Taken together, these results demonstrate that TBX19 recruits PRMT1 to the MFF promoter.

The arginine methyltransferase PRMT1 plays roles in transcriptional regulation via diverse histone methylation modifications, such as histone H2AR3, H4R17, and H4R3 (37–39). ChIP assays revealed that overexpressing TBX19 significantly increased level of H4R3me2a, but not H4R17me1a and H2AR3me2a, on the MFF promoter. Meanwhile, either silencing PRMT1 or treatment with the PRMT1 special inhibitor furamidine drastically abolished the inductive effect of TBX19 on H4R3me2a expression (Fig. 5G). The histone H4R3me2a modification could induce histone H3 acetylation on Lys9/Lys14 or histone H4 acetylation on Lys8/Lys12, resulting in the general transcriptional machinery binding to promoters and initiating transcription (40, 41). In line with these reports, we observed that TBX19 overexpression significantly increased the enrichment of H3K9ac, RNAPII, and TFIID on the MFF promoter, which was abrogated by PRMT1 knockdown or furamidine treatment (Fig. 5H). Consistent with the effect of PRMT1 silencing, furamidine treatment also resulted in a reduction of MFF in LCICs (Supplementary Fig. S7B and S7C). Therefore, these results suggest that PRMT1 contributes epigenetically to TBX19-mediated MFF upregulation.

Consistent with the results that PRMT1 enhanced TBX19-mediated MFF upregulation, overexpressing PRMT1 in non-LCICs significantly promoted, but silencing PRMT1 reduced, mitochondrial fragmentation and colocalization of LC3B with mitochondria (Fig. 6A and B; Supplementary Fig. S8A–S8C). These results indicated that PRMT1 was also involved in mitochondrial fission and mitophagy. Correspondingly, PRMT1-overexpressing non-LCICs displayed increased asymmetric stem cell division, but reduced mitochondrial ROS stress (Fig. 6C and D). However, overexpressing PRMT1 in TBX19-silenced non-LCICs or overexpressing TBX19 in PRMT1-silenced non-LCICs failed to promote mitochondrial fission and mitophagy (Fig. 6A and B; Supplementary Fig. S8A–S8E). These results further demonstrate that TBX19 and PRMT1 form a complex that has a collaborative effect on mitochondrial fission.

Consistent with the inhibitory effect of PRMT1 downregulation on mitochondrial fission, silencing PRMT1 drastically decreased the tumorsphere formation capability of LCICs (Fig. 6E; Supplementary Fig. S8F). Importantly, in vivo limiting dilution and serial transplantation assays revealed a significant reduction in tumor incidence and tumor-forming capacity in PRMT1-silenced LCICs compared with that of the control cells (Fig. 6F), which further supported the notion that PRMT1 contributes to the tumor-forming and self-renewal capabilities of HCC cells.

Finally, we examined whether pharmacologic inhibition of PRMT1 could reduce the tumorigenesis capability of LCICs and prevented tumor progression. As shown in Fig. 7A–D, treatment with PRMT1 inhibitor, furamidine, dramatically converted the fragmented mitochondria in clinical HCC tissue-derived LCICs into tubular and prolonged shape, accompanied by a reduction of asymmetric division and mitochondrial oxidative stress. Furthermore, tumorsphere formation assay revealed that furamidine-treated LCICs formed obviously smaller and fewer tumorspheres than vehicle-treated LCICs (Fig. 7E). Strikingly, furamidine treatment completely inhibited the tumor formation by 1 × 10² LCICs and significantly reduced the tumor incidence by 5 × 10⁵ LCICs (Fig. 7F). In addition, IF staining and IHC showed less MFF expression and TOM20/LC3B colocalization in furamidine-treated LCICs and tumors, suggesting that furamidine could suppress mitochondrial fission and mitophagy in vivo (Fig. 7G). Collectively, our results demonstrate that pharmacologic blocking PRMT1 could decrease the recurrence of liver cancer through inhibition of mitochondrial fission and mitophagy.

CICs are implicated in tumor initiation and progression, which are responsible for high therapeutic failure and recurrence in multiple types of cancers (42). Recently, mitochondrial fission has been reported to enable self-renewal of CICs in multiple human cancers (10, 15), and treatment with mdivi-1, which inhibits the GTPase activity of mitochondrial fission protein DRP1, was sufficient to suppress the propagation of stem-like cells in CICs, such as in brain tumors, prostate cancer, and nasopharyngeal carcinoma (6, 7). However, iPSCs also lost pluripotency when exposed to mdivi-1 treatment, indicating the drawback of mitochondrial fission target therapy when applied in cancer treatment, in which it might also impair normal stem cells (43). Therefore, to identify the pivotal factor that drives CIC-specific mitochondrial fission might be of great clinical value. In the present study, we identified MFF as the major candidate that promoted mitochondrial fission in LCICs, enabling self-renewal and enhanced tumorigenic capability of LCICs. Mechanistically, we identified the TBX19/PRMT1 axis as the central hub for upregulating MFF expression in LCICs, which allowed the application of the PRMT1 inhibitor, furamidine, to restrict the self-renewal and tumorigenesis capacity of LCICs. More importantly, the expression levels of TBX19 and PRMT1 were much higher in LCICs than in non-LCICs and normal liver stem cells, which might eliminate the simultaneous inhibition effect of furamidine for somatic stem cells. Therefore, our study reveals the distinct molecular mechanism of mitochondrial fission in LCICs, suggesting an efficient regimen for liver cancer treatment.

Fission-related proteins required for mitochondrial fission have been discovered in mammals (3, 10). For instance, DRP1 activation–mediated mitochondrial fragmentation is crucial for iPS cell reprogramming (13). Silencing FIS1 inhibited mitochondrial fission–mediated mitochondrial elongation in senescent cells and AML stem cells (19, 44). Downregulation of mitochondrial fission accessory proteins MIEF1/2 leads to mitochondrial elongation via blockage the interaction of DRP1 with mitochondria (45, 46). In our study, we examined 39 genes involved in mitochondrial fission in LCICs compared with non-LCICs, which revealed the indispensable role of MFF in regulating mitochondrial fission, whereas the expression level of other factors, including DRP1, FIS1, and MIEF1/2, showed no significant changes. Moreover, overexpression of MFF increased, whereas suppression of MFF reduced, mitochondrial segregation, asymmetric stem cell division, and mitophagy in LCICs. Importantly, it has been validated using in vivo and in vitro assays that stable MFF-depleted Hep3B and SNU449 CICs had disrupted mitochondrial fission and reduced tumor-initiating capability. Therefore, our study not only demonstrates the crucial role of MFF in mitochondrial fission, but also provides a basis for the discovery of molecular targets for mitochondrial fission–driven LCICs.

Homeostasis of mitochondrial dynamics is required for cellular reprogramming and differentiation in both stem cells and CSCs (3, 10). However, the mechanism regulating this requirement has not been unraveled. In this study, we found that mitochondrial fission modulated by the TBX19/MFF axis switched the metabolic process from OXPHOS to glycolysis and decreased mitochondrial ROS production. The reduction of ROS resulted in decreased oxidation modification of pluripotency factor OCT4 degradation, which in turn led to upregulation of OCT4 protein levels and its downstream target NANOG, and therefore sustained stemness and augmented the tumorigenicity of LCICs. These findings were consistent with a previous report that the ROS level correlates inversely with OCT4 expression in human embryonic stem cells (47). Therefore, our study shed light on the mechanism of mitochondrial fission and stemness maintenance in LCICs.

Recent studies have provided conflicting evidence regarding effects of mitochondrial fission versus fusion in cancer stem-like cells. For instance, it has been reported that several types of CSCs, such as BTICs, prostate CSCs, and leukemic stem cells, exhibited elevated mitochondrial fission activity and displayed a significantly more fragmented mitochondrial morphology compared with the more differentiated non-CSCs (5, 7, 19). They further demonstrated that inhibition of mitochondrial fission by silencing DRP1 not only decreased tumorsphere formation but also reduced expression of stemness markers, such as NANOG and Oct4 (7). Meanwhile, mitochondrial fission is also found to be associated with pro-survival mitophagy to clear defective mitochondria compromised by the mitochondrial stress generated from oncogenic transformation (19). Consistently, we also found that TBX19/PRMT1/MFF axis–mediated mitochondrial fission sustained the stemness and enhanced tumor-initiating capability of LCICs. However, it was also reported that unbalanced and abnormal activation of mitochondrial fission by aberrant MFF expression drive the impairment of mitochondrial metabolic function, leading to inhibition of breast CSCs (BCSC) propagation (48). Lee and colleagues (49) found that the chemotherapy-resistant BCSCs exhibited increased fusion activity and elongated mitochondrial network, which enhanced mitochondrial oxidative phosphorylation (mtOXPHOS) of BCSCs for survival after neoadjuvant chemotherapy. These results indicated that BCSCs may depend on mitochondrial fusion to adapt the demand of elevated mitochondrial biogenesis and metabolism. Therefore, mitochondria dynamics–associated CSCs might be highly context dependent and might also vary in different tumor environment and various states of the requirement. Moreover, a recent study of drosophila neural stem cell–derived tumor model reported that uncontrolled mitochondrial fusion is the rate-limiting step for irreversible immortalization of neural stem cells during tumor formation (50), which was distinct different from BTICs that depend on fission to maintain the TIC traits (7). Taken together, the opposite effects of mitochondrial fission and fusion on cancer stem-like cells might be caused by the response to distinct environment and meet the specific functional requirement.

In summary, our study provides strong evidence to support the hypothesis that the TBX19/PRMT1/MFF axis–induced mitochondrial fission evokes mitophagy and asymmetric stem cell division, and promotes the metabolic shift from OXPHOS to glycolysis in LCICs. This results in a reduction of mitochondrial ROS and the prevention of ROS-mediated OCT4 degradation. Understanding the precise role of TBX19 in the self-renewal of LCICs and the regulation of mitochondrial fission will not only advance our knowledge of liver cancer pathogenesis, but also will allow the development of new therapeutic strategies against liver cancer.

No disclosures were reported.

M. Tang: Data curation, validation. M. Yang: Data curation, validation. G. Wu: Validation, methodology. S. Mo: Methodology. X. Wu: Software. S. Zhang: Validation, methodology. R. Yu: Methodology. Y. Hu: Methodology. Y. Xu: Methodology. Z. Li: Methodology. X. Liao: Methodology. J. Li: Writing–original draft, project administration, writing–review and editing. L. Song: Writing–original draft, project administration, writing–review and editing.

This work was supported by National Key Research and Development Program of China (2020YFA0509400 to L. Song), Natural Science Foundation of China (81830082 and 82030078 to J. Li, 82003128 to G. Wu, 82072609 to L. Song, and 81773106 to L. Song), Guangzhou Science and Technology Plan Projects (201803010098 to J. Li), and China Postdoctoral Science Foundation (2019M653220 to G. Wu).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.