Histone demethylase JMJD2D promotes the self-renewal of liver cancer stem-like cells by enhancing EpCAM and Sox9 expression

Cancer stem-like cells (CSCs) contribute to the high rate of tumor heterogeneity, metastasis, therapeutic resistance, and recurrence. Histone lysine demethylase 4D (KDM4D or JMJD2D) is highly expressed in colon and liver tumors, where it promotes cancer progression; however, the role of JMJD2D in CSCs remains unclear. Here, we show that JMJD2D expression was increased in liver cancer stem-like cells (LCSCs); downregulation of JMJD2D inhibited the self-renewal of LCSCs in vitro and in vivo and inhibited the lung metastasis of LCSCs by reducing the survival and the early lung seeding of circulating LCSCs. Mechanistically, JMJD2D promoted LCSC self-renewal by enhancing the expression of CSC markers EpCAM and Sox9; JMJD2D reduced H3K9me3 levels on the promoters of EpCAM and Sox9 to enhance their transcription via interaction with β-catenin/TCF4 and Notch1 intracellular domain, respectively. Restoration of EpCAM and Sox9 expression in JMJD2D-knockdown liver cancer cells rescued the self-renewal of LCSCs. Pharmacological inhibition of JMJD2D using 5-c-8HQ reduced the self-renewal of LCSCs and liver cancer progression. Collectively, our findings suggest that JMJD2D promotes LCSC self-renewal by enhancing EpCAM and Sox9 expression via Wnt/β-catenin and Notch signaling pathways and is a potential therapeutic target for liver cancer.

In the present study, we reported that JMJD2D was upregulated in LCSCs and downregulation of JMJD2D markedly inhibited the self-renewal and proliferation of LCSCs in vitro and in vivo. Mechanistically, JMJD2D promoted the selfrenewal of LCSC through enhancement of EpCAM and Sox9 expression Wnt/β-catenin and Notch signaling pathways. The JMJD2D inhibitor 5-c-8HQ could attenuate the self-renewal of LCSCs in vitro and in vivo. Our findings indicate that JMJD2D may be a potential therapeutic target against LCSCs.

Downregulation of JMJD2D reduces the self-renewal of LCSCs in vitro
We previously reported that JMJD2D was overexpressed in colorectal and liver cancers and promoted cancer progression (25)(26)(27); however，the role of JMJD2D in CSCs remains This article contains supporting information. ‡ These authors contributed equally to this work. * For correspondence: Wengang Li, lwg11861@163.com; Chundong Yu, cdyu@xmu.edu.cn.
unclear. Formation of tumorsphere in CSC culture media represents the principal characteristic of CSC self-renewal. To investigate the role of JMJD2D in liver LCSCs, we enriched the LCSCs by inducing hepatoma spheroid formation from human liver cancer cell lines HepG2 and Huh-7 as well as mouse liver cancer cell line Hepa1-6 and then measured the protein and mRNA expression of JMJD2D. As shown in Fig. 1A and Fig. S1A, both the protein and mRNA levels of JMJD2D were upregulated in LCSCs (tumorsphere) compared with non-CSCs (attached cells), suggesting that JMJD2D may promote the self-renewal of LCSCs. To test this hypothesis, we knocked down JMJD2D in HepG2 and Huh-7 using two different JMJD2D shRNAs (sh2D-1 and sh2D-2) and knocked out JMJD2D in Hepa1-6 using CRISPR-Cas9 system (2D-KO) (Fig. 1B) and then performed MTT assay to measure the cell proliferation and tumorsphere formation assay to measure the tumorsphere formation ability. Downregulation of JMJD2D significantly inhibited liver cancer cell proliferation (Fig. S1, B-D) and tumorsphere formation ability as demonstrated by reduced tumorsphere number and size (Fig. 1C), indicating that JMJD2D can promote the proliferation and tumorsphere formation of liver cancer cells. Furthermore, we employed the limiting dilution assay to determine the CSC self-renewal frequency. At 200, 400, 600, and 800 cell levels, downregulation of JMJD2D resulted in a significant decrease in tumorsphere number (Fig. 1D). Serial sphere formation assays validated that the self-renewal capacities of JMJD2Ddownregulated cells were decreased as compared with control cells (Fig. 1E). Collectively, these results suggest that downregulation of JMJD2D reduces the self-renewal of LCSCs in vitro.

Downregulation of JMJD2D inhibits LCSC-derived tumor initiation and progression in vivo
CSCs have a strong ability to form tumors (29). To characterize the role of JMJD2D in LCSC-derived tumor initiation and progression, in vivo limiting dilution assay was performed using LCSCs disassociated from cultured spheroids. LCSCs derived from JMJD2D-downregulated spheroids displayed a lower tumorigenicity compared with the cells from control spheroids (Fig. 2, A-B). LCSCs from JMJD2D-downregulated spheroids exhibited decreased subcutaneous graft tumor growth and tumor weight (Fig. 2, C-D). Furthermore, we performed Ki67 staining to determine the effects of JMJD2D downregulation on cell proliferation in vivo. As shown in Fig. 2, E-F, downregulation of JMJD2D dramatically reduced Ki67-positive cell number in subcutaneous graft tumors. Although subcutaneous graft tumor is the most common graft tumor model, orthotopic graft tumor model is representative of natural progression of liver cancer. Therefore, we performed orthotopic graft tumor model to determine the effect of JMJD2D downregulation on the initiation and progression of LCSC-derived tumors in vivo. We found that orthotopic graft tumors derived from JMJD2D-downregulated spheroids grew much slower than control tumors (Fig. 2, G-H). These results indicate that downregulation of JMJD2D inhibits the initiation and progression of LCSCderived tumor in vivo.
Downregulation of JMJD2D inhibits the lung metastasis of LCSCs by reducing the survival and the early lung seeding of circulating LCSCs Tumor metastasis is the major cause of cancer-associated mortality. Tumor cells invade the surrounding tissue of the primary tumor, intravasate into blood to become circulating tumor cells (CTCs), translocate to distant tissues, and eventually seed, proliferate, and colonize to form metastatic tumors (30). CSCs are regarded to be the source of CTCs and responsible for tumor metastasis. To evaluate the function of JMJD2D in CTCs, we injected green fluorescent protein (GFP)-labeled LCSCs into mouse tail vein to establish a mouse model of CTCs and then detected GFP-labeled CTCs in mouse peripheral blood by flow cytometry as well as counted the early lung seeding GFP-labeled CTCs under the fluorescence microscope at 36 h after LCSCs injection. Downregulation of JMJD2D significantly reduced the number of CTCs in mouse peripheral blood (Fig. 3, A-B) and the number of the early lung seeding CTCs (Fig. 3, C-D), suggesting that JMJD2D promotes the survival of CTCs in blood and the early seeding of CTCs in the lung. Consequently, the number of metastasizing tumors in the lungs was significantly decreased in mice injected with JMJD2D-downregulated LCSCs as compared with control cells (Fig. 3, E and F). These results suggest that downregulation of JMJD2D inhibits the lung metastasis of LCSCs by reducing the survival and the early lung seeding of circulating LCSCs.
JMJD2D demethylates the H3K9me3 on the promoters of EpCAM and Sox9 to facilitate the recruitment and transactivation of β-catenin/TCF4 and NICD1, respectively To investigate the molecular mechanisms by which JMJD2D promotes the self-renewal of LCSCs, we examined the effects of JMJD2D knockdown on the expression of several LCSCsrelated CSC markers, including CD13, CD44, CD133, OCT4, Sox2, KLF4, Nanog, EpCAM, CD90, and Sox9. As shown in Fig. S2A, the mRNA levels of EpCAM and Sox9, but not other CSC markers, were significantly reduced in two JMJD2Dknockdown HepG2 cell lines. Knockdown of either EpCAM or Sox9 significantly inhibited the proliferation and tumorsphere formation ability of HepG2 cells (Fig. S2, B-C), suggesting that downregulation of EpCAM and Sox9 may, at least in part, be responsible for the inhibitory effects of JMJD2D knockdown on the self-renewal of LCSCs.
To show that the inhibitory effects of JMJD2D knockdown on EpCAM and Sox9 expression is not just limited to HepG2 cells, we also measured the mRNA levels of EpCAM and Sox9 in JMJD2D-downregulated Huh-7 and Hepa1-6 cells. As shown in Fig. 4A, downregulation of JMJD2D Figure 1. Downregulation of JMJD2D inhibits the self-renewal of LCSC in vitro. A, JMJD2D protein levels were upregulated in LCSCs compared with non-CSCs (attached cells). B, downregulation of JMJD2D protein level by JMJD2D shRNAs and CRISPR-Cas9 system was confirmed by western blot analysis in HepG2, Huh-7, and Hepa1-6. C, downregulation of JMJD2D in liver cancer cells reduced tumorsphere number and size. D, the in vitro limiting dilution assay showed that downregulation of JMJD2D decreased CSC self-renewal frequency. E, The serial sphere formation assay showed that downregulation of JMJD2D decreased self-renewal capacities of LCSCs. These experiments were performed at least three times with similar results. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001. JMJD2D promotes the self-renewal of LCSCs significantly decreased EpCAM and Sox9 mRNA levels in HepG2, Huh-7, and Hepa1-6 cells. Consistent with the mRNA levels, the protein levels of EpCAM and Sox9 were decreased in JMJD2D-downregulated HepG2, Huh7, and Hepa1-6 cells compared with control cells (Fig. 4B). Furthermore, we found that the expression of JMJD2D, EpCAM, and Sox9 was upregulated in human liver cancer specimens in GEO database (Fig. S3A), and the mRNA levels of JMJD2D were  To determine whether JMJD2D regulates EpCAM and Sox9 expression at the transcriptional level, we transfected EpCAM and Sox9 promoter reporters into JMJD2Dknockdown and control cells, respectively. The results showed that knockdown of JMJD2D decreased the promoter activities of EpCAM and Sox9 (Fig. 4C), suggesting that JMJD2D regulates EpCAM and Sox9 expression at the transcriptional level. It has been reported that EpCAM is a transcriptional target gene of Wnt/β-catenin signaling pathway with two TCF4 binding elements (TBE) on the EpCAM promoter (TBE1 and TBE2) (31), and Sox9 is a transcriptional target gene of Notch signaling pathway with a NICD1 binding site on the Sox9 promoter (32). Therefore, we performed promoter reporter assays to determine whether JMJD2D can cooperate with β-catenin/TCF4 and NICD1 to enhance the promoter activities of EpCAM and Sox9, respectively. As shown in Fig. 4D, ectopic expression of JMJD2D and β-catenin/TCF4 as well as JMJD2D and NICD1 synergistically increased the promoter activities of EpCAM and Sox9, respectively, indicating that JMJD2D cooperates with β-catenin/TCF4 and NICD1 to enhance the transcription of EpCAM and Sox9, respectively.
Next, we wondered whether JMJD2D and β-catenin/TCF4 can be recruited to the endogenous EpCAM promoter, and JMJD2D and NICD1 can be recruited to the endogenous Sox9 promoter. To this end, chromatin immunoprecipitation (ChIP) assays were conducted. As shown in Fig. 4E, JMJD2D could be recruited to the promoters of EpCAM and Sox9, but JMJD2D knockdown reduced its recruitment as expected. As a histone demethylase, JMJD2D promotes gene transcription through demethylating H3K9me3 on the promoter. The results of ChIP assays showed that the H3K9me3 levels on the promoters of EpCAM and Sox9 were increased in JMJD2Dknockdown cells compared with control cells (Fig. 4F), suggesting that JMJD2D is responsible for demethylating H3K9me3 on the promoters of EpCAM and Sox9. The results of ChIP assays showed that β-catenin/TCF4 could be recruited to the EpCAM promoter and NICD1 could be recruited to the Sox9 promoter as expected, but their recruitment was reduced in JMJD2D-knockdown cells (Fig. 4G), indicating that H3K9me3 demethylation facilitates the recruitment of β-catenin/TCF4 and NICD1 to the promoters of EpCAM and Sox9, respectively. Collectively, these results suggest that JMJD2D demethylates H3K9me3 on the promoters of EpCAM and Sox9 and facilitates the recruitment and transactivation of TCF4/β-catenin and NICD1, respectively.
To determine whether the demethylase activity of JMJD2D is required for the transcription of EpCAM and Sox9 to promote the self-renewal of LCSCs, we transfected wild-type JMJD2D and demethylase-defective JMJD2D S200M mutant plasmids into HepG2 cells, respectively. Compared with wildtype JMJD2D, JMJD2D S200M mutant failed to cooperate with β-catenin/TCF4 and NICD1 to enhance the promoter activities of EpCAM and Sox9 (Fig. S4A), failed to induce the expression of EpCAM and Sox9 in HepG2 cells (Fig. S4B), and failed to promote tumorsphere formation of LCSCs (Fig. S4C). These results suggest that the demethylase activity is required for JMJD2D to enhance EpCAM and Sox9 expression to promote the self-renewal of LCSCs.

JMJD2D interacts with β-catenin/TCF4 and NICD1
To determine whether JMJD2D could interact with βcatenin/TCF4 and NICD1 in liver cancer cells, we performed Co-IP assays. The results showed that JMJD2D interacted with β-catenin/TCF4 and NICD1 in HepG2, Huh-7, and Hepa1-6, respectively (Fig. 5, A and B). Furthermore, GST pull-down assays showed that JMJD2D interacted with TCF4 at its HMG Box domain and C-terminal domain (Fig. 5C), and TCF4 interacted with JMJD2D at its C-terminal domain (Fig. 5D). We previously reported that the Jmjc domain of JMJD2D interacted with the ARM-3-10 domain of β-catenin (33). Thus, JMJD2D bound to β-catenin and TCF4 via different domains. The N-terminal domain of TCF4 could interact with the ARM-3-10 domain of β-catenin (Fig. S5, A  and B). Thus, JMJD2D, β-catenin, and TCF4 interact with each other to form a ternary complex (Fig. 5E). GST pull-down assays also showed that JMJD2D interacted with NICD1 at its ANK domain (Fig. 5F), and NICD1 interacted with JMJD2D at its C-terminal domain (Fig. 5G). Collectively, these results implicate that β-catenin/TCF4 could interact with JMJD2D to recruit it to the EpCAM promoter and NICD1 could interact with JMJD2D to recruit it to the Sox9 promoter to enhance EpCAM and Sox9 expression, respectively.

Ectopic expression of EpCAM and Sox9 rescues the selfrenewal of JMJD2D-knockdown LCSCs
To further confirm that EpCAM and Sox9 mediate the promoting effect of JMJD2D on the self-renewal of LCSCs, we ectopically expressed EpCAM, Sox9, and EpCAM plus Sox9 in JMJD2D-knockdown HepG2 cells, respectively, and then measured the proliferation and tumorsphere formation abilities of cells. Restoration of EpCAM or Sox9 expression alone partially rescued the proliferation (Fig. S6, A and B) and the tumorsphere formation abilities of JMJD2D-knockdown cells (Fig. 6, A and B) Restoration of both EpCAM and Sox9 expression more efficiently rescued the proliferation (Fig. S6C) and the tumorsphere formation abilities of JMJD2Dknockdown cells (Fig. 6C). Consistently, restoration of EpCAM and Sox9 expression in JMJD2D-knockdown LCSCs partially rescued the subcutaneous tumor growth (Fig. 6D). Taken together, these results support the notion that JMJD2D promotes the self-renewal of LCSCs through enhancing the expression of EpCAM and Sox9. Since JMJD2D can activate Wnt and Notch signaling pathways, apart from EpCAM and Sox9, JMJD2D may promote LCSC self-renewal by regulating the expression of other CSC-related genes such as c-Myc and Hes1 via Wnt and Notch signaling pathways, respectively (Fig. S7).
JMJD2D promotes the self-renewal of LCSCs

Pharmacological inhibition of JMJD2D reduces EpCAM and Sox9 expression and inhibits the self-renewal and tumorigenesis of LCSCs
Given that downregulation of JMJD2D could inhibit LCSC self-renewal, we wondered whether a JMJD2D inhibitor 5chloro-8-hydroxyquinoline (5-c-8HQ), which reduces both demythelase activity and protein levels of JMJD2D (26), could inhibit the self-renewal of LCSCs. To this end, HepG2, Huh-7, and Hepa1-6 cells were treated with 5-c-8HQ (Fig. S8A), and then the self-renewal of LCSCs was measured by tumorsphere formation assay. As shown in Fig. 7, A and B, 5-c-8HQ treatment for 48 h significantly reduced the protein and mRNA levels of EpCAM and Sox9 in a dose-dependent manner. Consistently, 5c-8HQ treatment markedly reduced the tumorsphere formation abilities of LCSCs in a dose-dependent manner (Fig. 7C). Furthermore, 5-c-8HQ treatment significantly inhibited LCSC orthotopic graft tumor growth (Fig. 7D) and decreased Ki67positive tumor cells (Fig. 7E). As expected, 5-c-8HQ treatment significantly reduced the protein levels of EpCAM and Sox9 in LCSC orthotopic graft tumors (Fig. 7F). Moreover, JMJD2D inhibitor 5-c-8HQ could also suppress the expression of c-Myc and Hes1 in orthotopic tumors (Fig. S8B). Treatment with 10 mg/kg or 20 mg/kg 5-c-8HQ did not result in obvious adverse effects on mice as demonstrated by no body weight reduction and no toxicity to the major organs after treatment (Fig. S8, C and D). Collectively, these results demonstrate that JMJD2D could be targeted by a chemical inhibitor to reduce the expression of EpCAM, Sox9, c-Myc, and Hes1 and to inhibit the self-renewal and tumorigenesis of LCSCs (Fig. 7G). In addition, transient knockdown of JMJD2D by siRNA reduced the mRNA and protein levels of EpCAM, Sox9, c-Myc, and Hes1 (Fig. S9, A and B), as well as the proliferation and tumorsphere formation of Hepa1-6 cells (Fig. S9, C and D), supporting the notion that inhibition of JMJD2D is an effective way to inhibit the self-renewal of LCSCs for liver cancer therapy.

Discussion
CSCs are believed to be responsible for the initiation, propagation, metastasis, chemoresistance, and heterogeneity of solid tumor (3). Nowadays, increasing evidence shows that epigenetic alterations including DNA methylation and histone modifications participate in CSC self-renewal maintenance (34). Histone-modifying enzymes, such as EZH2, SUV39H1, and LSD1, have been reported to regulate CSC self-renewal (35,36). In the present study, we demonstrate that histone demethylase JMJD2D promotes LCSC self-renewal and could be a potential therapeutic target against LCSCs as follows: (1) JMJD2D is upregulated in LCSCs; (2) downregulation of JMJD2D inhibits the self-renewal of LCSCs in vitro and in vivo; (3) pharmacological inhibition of JMJD2D reduces LCSC selfrenewal and tumorigenesis.
EpCAM and Sox9 could enhance the self-renewal, propagation, and metastasis ability of LCSCs (8,9). In this study, we found that downregulation of JMJD2D reduced the expression of LCSC biomarkers EpCAM and Sox9 and inhibited the selfrenewal of LCSCs, whereas restoration of EpCAM and Sox9 expression in JMJD2D-knockdown LCSCs rescued the tumorsphere formation ability and tumorigenesis, suggesting that JMJD2D enhances LCSC self-renewal through upregulating EpCAM and Sox9 expression. Mechanistically, JMJD2D cooperates with β-catenin/TCF4 and NICD1 to enhance the transcription of EpCAM and Sox9, respectively.
Nowadays, targeting CSC therapy is thought to have great potential for the clinical treatment of cancer patients (37). Many signaling pathways had been reported to be involved in regulating CSC self-renewal, including Wnt/β-catenin, TGFβ, Notch, Hedgehog, NF-κB, and IL6/STAT3 pathways. Among them, Wnt/β-catenin, Notch，and Hedgehog are the main signaling pathways that regulate the maintenance and survival of CSCs and attract much attention. Conventional targeting therapeutics against LCSCs usually target one signaling pathway, which may induce the compensatory activation of other CSC-related signaling pathways and lead to the chemoresistance of tumors. Therefore, it is ideal to find a therapeutic target that can regulate multiple CSCrelated signaling pathways simultaneously. In this study, we found that JMJD2D could promote LCSC self-renewal through Wnt and Notch signaling pathways. Moreover, we recently reported that JMJD2D could severe as a coactivator to enhance Hedgehog signaling to promote colorectal cancer progression (25). Hedgehog signaling pathway is also very important for CSC self-renewal (14). We found that knockdown of JMJD2D in HepG2 could inhibit Hedgehog signaling pathway as demonstrated by Gli2 downregulation (Fig. S7). Taken together, JMJD2D may promote CSC self-renewal through activating Wnt, Notch, and Hedgehog signaling pathways simultaneously. Therefore, targeting JMJD2D could attenuate CSC self-renewal through simultaneously inhibiting Wnt, Notch, and Hedgehog signaling pathways, which have great potential for the clinical treatment of liver cancer patients.
As a demethylase, JMJD2D reduced H3K9me3 levels at the promoter region to increase the recruitment of TCF4/β-catenin and NICD1 to the EpCAM and Sox9 promoters and to facilitate the transcription of EpCAM and Sox9, respectively. Therefore, JMJD2D can be targeted by a demethylase inhibitor to suppress the expression of EpCAM, Sox9, c-Myc, Hes1, and Gli2. Indeed, treatment of a JMJD2D inhibitor 5-c-8HQ suppressed the expression of EpCAM, Sox9, c-Myc, Hes1, and Gli2, as well as LCSC tumorsphere formation ability and LCSC orthotopic tumor growth in vivo, indicating that 5-c-8HQ treatment can simultaneously inhibit Wnt, Notch, and Hedgehog signaling pathways to inhibit the self-renewal and tumorigenesis of LCSCs. Collectively, our study shows the proof-of-concept use of the JMJD2D inhibitor to target LCSCs for liver cancer therapy.

Cell culture
Human liver cancer cell lines HepG2, Huh-7, mouse liver cancer cell line Hepa1-6, and HEK293 T were cultured in high-glucose DMEM (HyClone) supplemented with 10% fetal bovine serum (FBS) and penicillin-streptomycin. All cell lines were maintained in a humidified chamber with 5% CO 2 at 37 C.

Cell proliferation assay
MTT assays were used to measure cell proliferative rate as described previously (38).

Tumorsphere culture and serial passage assay
Tumorsphere culture was performed as described previously (39). For primary tumorsphere culture, cells were seeded at 3000 cells per well. After 5-7 days, primary tumorspheres were centrifuged and dissociated with 0.05 % trypsin-EDTA (Invitrogen) and then sieved (40 μm) to obtain single cell. Single cells derived from primary tumorspheres were seeded at 3000 cells per well for secondary tumorspheres, then single cells derived from secondary tumorspheres were seeded at 3000 cells per well for tertiary tumorspheres.

In vitro limiting dilution assay
In vitro limiting dilution assay was performed as described previously (40).

Reverse transcription and real-time PCR
Total RNA was isolated using trizol reagent (Invitrogen) according to the manufacturer's instructions. cDNA was obtained from 2 μg of total RNA using a reverse transcription kit (Toyobo). Real-time PCR was performed using universal SYBR Green Master (Roche Applied Science), and relative quantification was achieved by normalization to β-actin. Sequences of the primers used for real-time PCR are listed on Table S1.

Luciferase reporter assay
The EpCAM or Sox9 promoter reporter was transfected into cells, the renilla reniformis luciferase reporter serviced as an internal control. The firefly luciferase activity of individual sample was standardized relative to the renilla luciferase activity and expressed in units (-fold change) relative to control sample value assigned a unit of 1. The activities of EpCAM and Sox9 promoter reporters were analyzed by the luciferase reporter assay system (Promega, Madison, US).

Coimmunoprecipitation (Co-IP) and GST pull-down assay
The interactions of JMJD2D with β-catenin, TCF4, and NICD1 in HepG2, Huh-7, and Hepa1-6 were analyzed by the Co-IP assay and GST pull-down assay as described previously (38).

Tumor graft
Four to six-week-old male nude mice and C57BL/6 mice were obtained from the Laboratory Animal Center of Xiamen University. JMJD2D-knockdown/knockout cells or control cells derived from tumorspheres were subcutaneously injected into the dorsal flanks of mice. The tumor size was measured along two perpendicular axes every 2 days from day 8 after cell injection using a vernier caliper. The volume of the tumor was calculated using the formula: Volume = Length × Width 2 × 0.52. All experimental procedures involving animals were performed in accordance with animal protocols approved by the Laboratory Animal Center of Xiamen University.

In vivo limiting dilution assay
For in vivo limiting dilution assay, four to six-week-old male BALB/c nude mice or C57BL/6 mice were injected with 1 × 10 2 , 5 × 10 2 , 1 × 10 3 , and 1 × 10 4 cells derived from tumorspheres into the dorsal flanks of mice. After 2 months, the number of tumors was counted.

Detection of circulating liver cancer stem-like cells
Detection of circulating LCSCs was performed as previously described (42). Briefly, GFP-labeled LCSCs (1 × 10 6 ) dissociated from spheroids were suspend in PBS and subsequently injected into the lateral tail veins of mice in a volume of 0.2 ml. Whole blood specimens were collected via heart acupuncture at 36 h after transplantation. Erythrocytes were lysed in RBC Lysis Buffer and peripheral blood mononuclear cells (PBMCs contain circulating tumor cells) were enriched from the blood. Circulating LCSCs were analyzed by detecting GFP-positive cells using flow cytometry on a BD LSRFortessa (BD Biosciences, San Jose, CA).

LCSC GFP-labeling process
Lentiviruses carrying GFP sequence were packed through transfecting lentiviral transfer vectors pLV-GFP and three additional plasmids (pMDL, pREV and pVSVG) into 293T cells. Cancer cells were infected with lentiviruses carrying GFP sequence for 72 h, and then cancer cells were seeded in CSC culture media for 7 days to form tumorsphere. Tumorsphere was disassociated into single cell suspensions by trypsin, and then fluorescence-activated cell sorter (FACS) was used for isolation and enrichment of GFP-labeled (GFP-positive) LCSCs. Almost all of these cells were GFP-positive under the fluorescence microscope (data not shown).

Orthotopic graft tumor model
The orthotopic graft tumor model was performed as previously describe (43). Briefly, LCSCs (1 × 10 6 ), disassociated from cultured HepG2 and Hepa1-6 spheroids and suspended in 30 μl PBS, were slowly injected into the left lobe of the liver using a 28-gauge needle. After cell injection, a small piece of sterile gauze was placed on the injection site, and light pressure was applied for 5 min to prevent bleeding. The abdomen was then closed with a 6-0 silk suture. After surgery, animals were kept in a warm cage, observed for 1-2 h and subsequently returned to the animal room after full recovery from the anesthesia. After 3 days of inoculation of cancer cells, mice were randomly divided into three groups and then gavaged daily for 12 days with 1% sodium carboxymethyl cellulose (Control), 10 mg/kg 5-c-8HQ (low-dose group), and 20 mg/kg 5-c-8HQ (high-dose group) dissolved in 1% sodium carboxymethyl cellulose, respectively. After 12 days of drug administration, mice were sacrificed, and then tumors and organ tissues were collected for section and storage in a -80 C refrigerator for subsequent analysis.

Immunohistochemistry
Immunohistichemistry analysis was performed as previously described (41). The primary antibody used for immunohistochemistry analysis was Ki67 (ab15580; Abcam).

Western blot analysis
Western blot analysis was performed as previously described (41). The antibodies used for western blot analysis are listed in "Antibodies and drugs."

Statistical analysis
All data are shown as the mean ± SD (Standard deviation) of at least three replicates. The statistical analysis of the difference between two groups was determined using two-tailed Student's t-tests, and multiple group comparisons were performed using one-way or two-way ANOVA followed by Tukey's post hoc test. For all figures, ns (p > 0.05); *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Data availability
All data described are contained within the article and the supporting information file. This article contains Table S1 and Figs. S1-S9.  Conflict of interest-The authors declare that they have no conflicts of interest with the contents of this article.