Article Text

Original research
Lineage tracing and single-cell analysis reveal proliferative Prom1+ tumour-propagating cells and their dynamic cellular transition during liver cancer progression
  1. Lei Zhou1,2,
  2. Ken HO Yu1,3,
  3. Tin Lok Wong1,4,
  4. Zhao Zhang5,
  5. Chun Ho Chan1,
  6. Jane HC Loong1,
  7. Noelia Che1,
  8. Hua Jian Yu1,
  9. Kel Vin Tan6,
  10. Man Tong1,2,4,
  11. Elly S Ngan7,
  12. Joshua WK Ho1,3,
  13. Stephanie Ma1,2,4
  1. 1 School of Biomedical Sciences, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong SAR, China
  2. 2 Department of Clinical Oncology, The University of Hong Kong-Shenzhen Hospital, Shenzhen, China
  3. 3 Laboratory of Data Discovery for Health Limited (D24H), Hong Kong Science Park, Hong Kong SAR, China
  4. 4 State Key Laboratory of Liver Research, The University of Hong Kong, Hong Kong SAR, China
  5. 5 Department of Medicine, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong SAR, China
  6. 6 Department of Diagnostic Radiology, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong SAR, China
  7. 7 Department of Surgery, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong SAR, China
  1. Correspondence to Dr Stephanie Ma, School of Biomedical Sciences, The University of Hong Kong, Hong Kong, Hong Kong; stefma{at}; Dr Joshua WK Ho, School of Biomedical Sciences, The University of Hong Kong, Hong Kong, Hong Kong; jwkho{at}


Objective Hepatocellular carcinoma (HCC) has high intratumoral heterogeneity, which contributes to therapeutic resistance and tumour recurrence. We previously identified Prominin-1 (PROM1)/CD133 as an important liver cancer stem cell (CSC) marker in human HCC. The aim of this study was to investigate the heterogeneity and properties of Prom1+ cells in HCC in intact mouse models.

Design We established two mouse models representing chronic fibrotic HCC and rapid steatosis-related HCC. We performed lineage tracing post-HCC induction using Prom1C-L/+; Rosa26tdTomato/+ mice, and targeted depletion using Prom1C-L/+; Rosa26DTA/+ mice. Single-cell RNA sequencing (scRNA-seq) was carried out to analyse the transcriptomic profile of traced Prom1+ cells.

Results Prom1 in HCC tumours marks proliferative tumour-propagating cells with CSC-like properties. Lineage tracing demonstrated that these cells display clonal expansion in situ in primary tumours. Labelled Prom1+ cells exhibit increasing tumourigenicity in 3D culture and allotransplantation, as well as potential to form cancers of differential lineages on transplantation. Depletion of Prom1+ cells impedes tumour growth and reduces malignant cancer hallmarks in both HCC models. scRNA-seq analysis highlighted the heterogeneity of Prom1+ HCC cells, which follow a trajectory to the dedifferentiated status with high proliferation and stem cells traits. Conserved gene signature of Prom1 linage predicts poor prognosis in human HCC. The activated oxidant detoxification underlies the protective mechanism of dedifferentiated transition and lineage propagation.

Conclusion Our study combines in vivo lineage tracing and scRNA-seq to reveal the heterogeneity and dynamics of Prom1+ HCC cells, providing insights into the mechanistic role of malignant CSC-like cells in HCC progression.

  • hepatocellular carcinoma
  • stem cells

Data availability statement

Data is available on GEO (GSE181515). We uploaded our raw data to GEO (GSE181515), which will be open to public access upon paper acceptance. The following secure token has been created to allow review of record GSE181515 while it remains in private status: mxwnsyuwjnexbaj. We also uploaded the scripts for all the analysis performed in the paper onto Github (

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Significance of this study

What is already known on this subject?

  • Liver cancer stem cells are a unique subset of hepatocellular carcinoma (HCC) cells with stem cell features and dictate hierarchical organisation within the tumour.

  • Early studies using cell sorting and xenotransplantation techniques found that Prominin-1 (PROM1)/CD133 indicated an important functional liver cancer stem cell subpopulation in human HCCs.

What are the new findings?

  • Lineage tracing identified a population of Prom1-derived proliferative tumour-propagating cells HCC cells with liver cancer stem cell (CSC) features in vivo.

  • In vivo targeted depletion of the PROM1+ lineage in HCC tumours impedes tumour growth and malignant progression.

  • scRNA-seq combined with lineage tracing revealed the heterogeneity and dedifferentiation trajectory of Prom1+ cancer cells.

  • Prom1-lineage gene signature predicts poor prognosis in HCC and the enriched reactive oxygen species detoxification genes are key for lineage propagating.

How might it impact on clinical practice in the foreseeable future?

  • This study establishes a preclinical model that shares similar transcriptomic spectra with the human good to poor HCC differentiation grades.

  • The gene signature of Prom1-lineage predicts a worse prognosis in patients with HCC.

  • The findings of this study provide novel insight into the underlying molecular mechanism for CSC-like HCC cells with unexpected degree of cellular heterogeneity and potential targeted drug screening.


The key concept underlying the cancer stem cell (CSC) or tumour-initiating cell (TIC) theory is that tumours are highly heterogeneous and maintained in a hierarchical manner, in which different cell populations have different functionalities in pathophysiology. CSCs are characterised by a strong capacity for self-renewal, a potential for differentiation into different subtypes that constitute the tumour and resistance to conventional treatment. Accordingly, CSCs represent the specific subset of cells that are critical for tumour initiation and growth. Thus, the eradication of CSCs, their signalling pathways and their niche is believed to be critical to achieving stable remission of, and even curing, aggressive malignancies.1

Extensive evidence has shown that hepatocellular carcinoma (HCC), the most common form of liver cancer, is also promoted by CSCs.2 CD133/PROM1 (Prominin-1) is one of the best-studied functional and phenotypic markers of liver CSCs,3 4 but most studies have primarily relied on flow cytometric sorting followed by limiting dilution xenotransplantation into immunodeficient mice. The assay, while still regarded as the gold standard to determine TIC frequency in a tumour, has disadvantages, as reviewed in Rycaj and Tang.5 To circumvent the limitations of this assay, researchers have lineage traced genetically marked cells in more recent years to identify CSCs in tractable mouse tumour models. Lineage tracing studies have shown that mature hepatocytes but not cholangiocytes are the origin of HCC6 7 and that EpCAM+ ductal cells in the inflamed liver or Lgr5+ pericentral hepatocytes are source of HCC development.8 9 Specifically, Prom1+ biliary epithelial cells contributed to biliary structure reconstruction on thioacetamide-induced chronic liver damage.10 Prom1+ cells from neonatal liver and 3,5-diethoxycarbonyl-1,4-dihydrocollidine (DDC)-treated biliary injured liver had a high regenerative capacity and the potential to develop into cancer.11 However, HCC mainly occurs in adult, and bile duct injury is not a common etiological factor of HCC. Therefore, the role of Prom1 cells in HCC initiation and progression has not been fully explored, while their complex cell heterogeneity and molecular cascade at the single-cell level remain unclear. A deeper understanding of the heterogeneity and potential fate of CSCs in an intact HCC tumour bulk will offer new insights into novel therapeutic opportunities. The aim of this study was to investigate the role, heterogeneity and properties of Prom1+ cells in HCC by lineage tracing, lineage ablation and scRNA-seq strategies.



For lineage tracing, Prom1C-L mice11 were crossed with reporter mice, and the heterozygous offspring (Prom1C-L/+; Rosa26tdTomato/+) were used. For lineage depletion, the Prom1C-L/+; Rosa26DTA/+ and control littermates were used.

A detailed description of all methods used in this study can be found in the online supplemental information.

Supplemental material


Prom1 expression is upregulated with HCC tumour progression in immunocompetent mouse models and HCC clinical samples

We first established two clinically relevant HCC models with immunocompetent mice to elucidate the role of Prom1-derived cells in tumour initiation and progression. Since HCC mostly develops in the presence of advanced fibrosis or cirrhosis, we mimicked this aetiology with a well-established chemical diethylnitrosamine (DEN) plus carbon tetrachloride (CCL4) administration protocol (figure 1A and online supplemental figure S1A–C).12 Concomitant with Afp, Prom1 mRNA expression showed a drastic increase following tumour initiation compared with that in normal (vehicle treated or before CCL4 treatment) and fibrotic (16–20 weeks) livers (figure 1B); with findings further confirmed by immunostaining (figure 1C). Abnormal lipid accumulation in hepatocytes, often presented as non-alcoholic fatty liver disease, is increasingly recognised as a critical contributor to hepatocarcinogenesis.13 Likewise, Prom1 expression concomitant with Afp was also upregulated in a lipogenesis-related HCC model induced by hydrodynamic tail vein delivery of the N-RasV12 and Myr-AKT (N-Ras+ AKT) proto-oncogenes (figure 1D–F).14 Enhanced lipid accumulation and increased MAPK/AKT signalling activation were also confirmed by immunostaining with Oil red O and for p-Mapk1/3 and p-Akt, respectively (online supplemental figure S1D). To demonstrate the clinical relevance of these findings, we then analysed PROM1 gene expression in The Cancer Genome Atlas (TCGA)—Liver Hepatocellular Carcinoma (LIHC) dataset (hereafter TCGA-LIHC) and found that in patients with HCC, PROM1 expression was also higher in advanced stage III/IV HCCs and stage II HCCs than in early stage I HCCs (figure 1G).

Figure 1

Prom1 expression is upregulated with hepatocellular carcinoma (HCC) tumour progression in immunocompetent mouse models and human HCC clinical samples. (A) Gross image and H&E staining of HCC tumours from the diethylnitrosamine plus carbon tetrachloride (DEN+CCL4) mouse model. (B) Afp and Prom1 mRNA expression levels at different time points in the DEN+CCL4 HCC model. At the age of 16–20 weeks, the mice showed fibrotic changes in the liver without tumours. At the age of 22–24 weeks, the mice showed HCC tumours in the liver. VEH, vehicle control. n=3 mice per group. (C) Immunofluorescence staining for Prom1 in the tumour and adjacent non-tumour tissues in the DEN+CCL4 model. (D) Gross image and H&E staining of HCC tumours in the N-Ras+ AKT mouse model. HCC nodules developed in lesions consisting of clear-cell hepatocytes. (E) Afp and Prom1 mRNA expression levels at different time points in the N-Ras+ AKT model. EV, empty vector control. Tumour nodules developed ~3 weeks after plasmid injection by hydrodynamic tail vein delivery. n=3 to 6 mice per group. (F) Immunofluorescence staining of Prom1 in the tumour and adjacent non-tumour tissues in the N-Ras+ AKT model. Scale bar=100 µm. T=tumour. (G) Log2 normalised counts of PROM1 in The Cancer Genome Atlas (TCGA)—Liver Hepatocellular Carcinoma (LIHC) cohort, stratified by TNM stage (stages I, II and III/IV). *p<0.05 and **p<0.01 by Kruskal-Wallis one-way analysis of variance test.

Genetic lineage tracing shows competitive amplification and clonal expansion of the Prom1-derived tumour cells during HCC progression

To assess and characterise Prom1+ cells and their progenies in HCC tumours, we performed lineage tracing following visible tumour nodules formed in the DEN+CCL4 HCC model, in which Prom1-CreERT2-LacZ mice crossed with Rosa26-LSL-Tomato reporter mice (hereafter Prom1-tdTomato mice) were used (figure 2A). On confirmation of effective labelling (online supplemental figure S2A,B), tumours were examined over time. The Prom1-derived tdTomato+ cells covered 50.7% (34/67 tumour nodules from 3 mice) of the tumours collected 30 days after tamoxifen treatment (figure 2B). The frequency of the tdTomato+ tumour cells (live tdTomato+ cells/all live resident cells), as determined by flow cytometry, increased progressively from day 3 (9.30±1.05%) to day 10 (13.57±0.13%) and day 30 (21.54±3.08%), representing a 2.3-fold increase over time, indicating that the Prom1-derived tdTomato+ cells exhibited competitive amplification (figure 2C,D). With low-dose tamoxifen (75 µg), which labels only a small proportion of the Prom1+ cells 3 days after treatment, we could observe clustered tdTomato+ cells widely distributed in the tumours at day 30, with 8.1% (21/259 clones) of the clones comprising at least 40 cells (figure 2E,F). Notably, these tdTomato+ cells did not overlap with Krt19 but primarily overlapped with Afp staining (figure 2G), suggesting that they mainly represented HCC cells but not ductular cells. Over the 30-day tracing period, the tdTomato+ cells showed gradual loss of Hnf4a expression but no gain of progenitors or cholangiocytes markers in situ (online supplemental figure S2C), indicating Prom1+ cells to be cellularly dynamic during tumour growth.

Figure 2

Genetic lineage tracing shows competitive amplification and clonal expansion of the Prom1-derived tumour cells during HCC progression. (A) Schematic plot of the lineage tracing setting for the diethylnitrosamine plus carbon tetrachloride (DEN+CCL4) model. After tumours formed, the Prom1-tdTomato mice received a single dose of 4 mg tamoxifen, and tissues were collected and examined at the indicated time points. (B) Gross bright-field and fluorescent images of livers from the HCC group 3 and 30 days after tamoxifen (4 mg) treatment. The livers of the mice treated with olive oil (No TAM control) are also shown and were used as a negative control. (C, D) Quantification of the tdTomato+ cells in HCC tumours by flow cytometry following a 4 mg tamoxifen injection. n=3–8 mice per group. (C) Representative images showing flow cytometric gating. (D) Quantification analysis by one-way analysis of variance. ***p<0.001. ns=not significant. (E–G) Clonal expansion assays with the low-dose tamoxifen treatment (75 µg, n=3–4 mice per group). (E, F) Size distribution of the tdTomato+ clones at the indicated time points. Size was defined as the number of tdTomato+ cells per clone. Cells with an interval larger than 20 µm were counted as two independent clones. Each dot in (E) represented a clone in (F). (G) Representative images showing clonally expanding lineage-labelled Prom1+ cells in HCC tumours. The upper panel shows low-power tile scan microscope images of Krt19 and Prom1 costaining, and the two middle panels show high-power images of the former, indicated by white rectangles. The two lower panels show costaining of Afp and Prom1 in serial sections. Upper, scale bar=1 mm. Middle and bottom, scale bar=50 µm.

The Prom1-derived tdTomato+ HCC cells show higher proliferative and CSC-like properties

We treated the DEN+CCL4 HCC mice with BrdU for short-term labelling (figure 3A). A higher percentage of BrdU incorporation was noted in the lineage-traced tdTomato+ HCC cells than in the tdTomato- HCC cells, indicating a faster turnover of the Prom1+ HCC cells (figure 3B). Consistently, fluorescence activated cell sorting (FACS) analysis showed that the tdTomato+ HCC cells displayed enhanced Ki67 expression compared with their unlabeled counterparts (figure 3C). These data indicated that the Prom1-derived tdTomato+ cells have enhanced proliferative property. To examine this issue from a functional perspective, we isolated viable tdTomato+ and tdTomato- HCC cells by gating parenchymal cells from tumours based on size and granularity (figure 3A and online supplemental figure S3A).15 Single tdTomato+ HCC cells could form compact tumour organoids (figure 3D,E). The plating efficiency was 23.2%±2.0% for the tdTomato+ HCC cells, two folds higher than that of the tdTomato- HCC cells (11.1%±2.4%) (figure 3D). An in vitro limiting dilution assay also confirmed that the tdTomato+ tumour cells possessed an increased repopulating frequency relative to the tdTomato− tumour cells in both HCC models (online supplemental figure S3B–D). To provide an in vivo growing niche for HCC cells, we transplanted the tdTomato+ and tdTomato− HCC cells into mouse livers through splenic injection. The estimated in vivo repopulating frequency was 1/2396 for the tdTomato+ cells, 17.9-folds higher than that for the tdTomato− cells (1/42 962) (figure 3F). These data showed that the tdTomato+ HCC cells had a superior capacity for tumour initiation and propagation. To evaluate the cell fate of transplanted tdTomato+ HCC cells, we examined the recipient livers by immunohistochemistry staining. Around 8% of the tdTomato+ HCC tumours showed positive Krt19 signal, whereas none of the tdTomato− HCC tumours have such phenotype (figure 3G). Further transplantation of organoid cultured tdTomato+ cells yielded cholangiocarcinoma (CCA)-like tumours with glandular morphology and vimentin+ epithelial-mesenchymal transition (EMT)-like tumours (online supplemental figure S3E), suggestive of the potential of Prom1-lineage HCC cells to differentiate or transdifferentiate. Taken together, Prom1-derived tdTomato+ HCC cells exhibited higher tumourigenicity and CSC-like features.

Figure 3

The Prom1-derived tdTomato+ hepatocellular carcinoma (HCC) cells show higher proliferative and cancer stem cell (CSC)-like properties. (A) Schematic plot for labelling Prom1+ HCC cells in the diethylnitrosamine plus carbon tetrachloride (DEN+CCL4) model followed by in vitro and in vivo functional assays. After tumour initiation, the Prom1-tdTomato mice received a single dose of 4 mg tamoxifen on day 0. For the BrdU proliferation assay, BrdU (50 mg/kg) was injected 4 times every 12 hours, and livers were dissected 1 day after the last BrdU injection. For FACS analysis and sorting, the tdTomato-negative (tdT−) and tdTomato-positive (tdT+) HCC cells were isolated on day 3. (B) BrdU, tdTomato and DAPI co-staining in the traced Prom1-tdTomato mice (left) and quantification (right) of BrdU+ percentages in the tdT− and tdT+ hepatocytes/HCC cells. Arrowheads indicate the BrdU+; tdT+ cells. Scale bar=50 µm. **p<0.01 by unpaired t-tests. (C) FACS analysis of Ki67 expression in the tdT− and tdT+ cells after excluding dead cells and non-parenchymal cells. The Ki67-positive population was gated based on fluorescence minus one (FMO) control. Data are presented as paired cells from four individual mice. **p<0.01 by paired t-tests. (D) Numbers of organoids formed per 1000 tdT− or tdT+ HCC cells from three independent HCC mice and their representative bright-field images. *p<0.05 by unpaired t-test. Scale bar=50 µm. (E) Clonal cultures of primary HCC organoids from single tdTomato+ HCC cells. Images taken at the indicated time points. Scale bar=100 µm. (F) Reconstitution efficiency of the tdT+ and tdT− HCC cells transplanted into the livers of recipient mice through splenic injection. Representative gross liver photos from transplantation with 4000 and 40 000 TdT+ and tdT− cells are shown. Arrowheads indicate tumours. P values obtained by Pearson’s χ2 test with 95% CIs. (G) Representative H&E and immunostaining images of tumours derived from tdT+ and tdT− cells in (F). Hepatic/HCC markers (Hnf4a, Afp), ductular marker (Krt19) and EMT marker (Vim) were examined. Quantification was generated from all examined tumours. Asterisks indicated ductular cells of recipient mice (tdTomato-negative) and arrowheads indicate tumour cells (tdTomato-positive). Low magnification (tdTomato antibody staining): scale bar=3 mm. High magnification: scale bar=20 µm.

In vivo depletion of the Prom1+ lineage after tumour initiation impedes HCC growth and malignant progression

The importance of Prom1+ cells in HCC progression was further evaluated by genetic depletion using conditional CreER-induced diphtheria toxin (DTA) expression and selective cell death (hereafter Prom1-DTA mice) (figure 4A). The knock-in of the CreERT2-nLacZ cassette resulted in the disruption of one Prom1 allele.11 In both HCC models, no significant differences in tumour size, tumour number (online supplemental figure S4A) or overall survival (online supplemental figure s4B) were found between the control and Prom1-DTA mice without interventiononline supplemental figure S4A online supplemental figure S4B, indicating that one wild-type allele of Prom1 is sufficient for its function in HCC.

Figure 4

In vivo depletion of the Prom1+ lineage after tumour formation impedes hepatocellular carcinoma (HCC) growth and malignant progression. (A) Schematic plot showing Prom1-lineage depletion in the diethylnitrosamine plus carbon tetrachloride (DEN+CCL4) model. Five doses of 4 mg tamoxifen or oil vehicle were administered to the Prom1-DTA mice with tumours. MRI scanning was performed to capture the tumour size at baseline 3 days prior to tamoxifen treatment (pre-TAM). Tissues were collected and examined at the indicated time points after the last dose. (B) Representative transverse MRI images of the control (Ctrl) and Prom1-DTA mice pre- and post-tamoxifen treatment. Red-dotted lines indicate tumour areas in the livers. Scale bar=1 cm. Right panel: quantitative analysis of tumour diameter normalised to the pretreatment baseline. ****p<0.0001 by unpaired t-tests. n=23 and 21 tumours from 3 mice in the Ctrl and Prom1-DTA groups, respectively. (C, D) Representative gross appearance of the livers (C) and quantitative data of the liver weight, number of tumour nodules and maximum tumour volume (D) from the Ctrl and Prom1-DTA mice 16 days post tamoxifen treatment. *p<0.05, **p<0.01 by unpaired t-tests. n=6 and 8 mice in each group. (E) Immunofluorescence staining showing colocalisation of Prom1 and cleaved caspase-3 12 hours after tamoxifen treatment in the Prom1-DTA mice. Left, scale bar=100 µm. Right, higher magnification, scale bar=20 µm. (F) X-gal staining (blue) of tumour sections with vehicle or tamoxifen treatment at the indicated time points. Scale bar=500 µm. (G) Representative images (upper panel) and quantitative data (lower panel) of Ki67 staining in the liver tumour tissues. Upper left, scale bar=500 µm. Upper right, high magnification, scale bar=100 µm. *p<0.05 by unpaired t-tests. (H) Gene set enrichment analysis (GSEA: hallmark gene sets) within the differentially expressed genes (adjusted p<0.05; fold change >2) between the Ctrl and Prom1-depleted DEN+CCL4 HCC tumours, 16 days post tamoxifen treatment (n=3 in each group).

In the DEN+CCL4 HCC model, MRI scanning allowed confirmation of tumour formation prior to depletion (baseline). Sixteen days after 5 doses of 4 mg tamoxifen treatment, the mice were scanned by MRI again. The tumour size increased 1–3 folds over baseline in the control mice, compared with less than one fold in the Prom1-DTA mice (figure 4B). The Prom1-DTA mice exhibited significantly fewer tumour nodules per liver and smaller tumour sizes than the control mice (figure 4C,D). The apoptotic marker expression in Prom1+ cells and the X-gal staining indicated that DTA diminished the majority of Prom1+ tumour cells with minimal effect on other tumour cells (figure 4E,F). The same treatment scheme was applied to the N-Ras+AKT HCC model in the control and Prom1-DTA mice. Consistently, a similar phenomenon on Prom1 depletion was noted with reduced tumour weight, reduced tumour multiplicity and prolonged survival compared with those of the control or vehicle treated mice (online supplemental figure S4C–I). Ki67 staining indicated a significant reduction of hepatocellular proliferation in the tumours of the Prom1-DTA mice treated with tamoxifen (figure 4G and online supplemental figure S4J), which is a trait of tumour malignancy.16 Bulk RNA-seq further showed downregulated ‘EMT’ and ‘angiogenesis’ hallmarks, reduced Afp expression, as well as recovery of liver functional genes in Prom1 depleted tumours of the two HCC models (figure 4H and online supplemental figure S4K,L). Taken together, these data indicated that the Prom1-derived lineage plays a key role in tumour growth and malignant progression of both fibrosis-driven and lipogenesis-driven hepatocarcinogenesis.

Single-cell transcriptome analysis reveals the cell lineage properties of the Prom1-derived cells in HCC

To further elucidate the lineage characteristics and molecular signatures of the Prom1-derived HCC cells, we performed scRNA-seq. All viable tdTomato+ and tdTomato- resident cells in the traced DEN+CCL4 tumours were sorted and sequenced by 10X Genomics to reconstruct a single-cell landscape of HCC tumours (figure 5A). A total of 33 473 cells from six tumour samples (the tdTomato+ and tdTomato- cells from day 3, 10 and 30 after labelling) were integrated, partitioned into 36 clusters and visualised by t-distributed stochastic neighbor embedding (t-SNE) (online supplemental figure S5A). Based on the expression of known lineage marker genes of liver cells (online supplemental figure S5B), these clusters were classified into 13 cell populations (figure 5B). Clusters 1, 2 and 8 showed higher copy number variation (CNV)17 levels than the other populations; therefore, we defined them as HCC cells (figure 5C,D, online supplemental figure S5C). The non-HCC clusters were merged according to their shared lineage markers. Ultimately, eight major cell lineages were annotated (figure 5E). These liver tumour resident cells included HCC cells, cholangiocytes, macrophages/monocytes, endothelial cells, hepatic stellate cells/fibroblasts, B cells, neutrophils and a cluster representing hybrid hepatocytes expressing both hepatocyte and cholangiocyte markers15 (figure 5F). Cells from the tdTomato+ samples overlapped with those from the tdTomato- samples mainly within epithelial lineages (figure 5G). The tdTomato+ cells on day 3 comprised 94.6% HCC cells, 1.8% hybrid hepatocytes and 3.1% cholangiocytes. The proportion of the tdTomato+ HCC cells was maintained at 97.2% on day 10% and 96.3% on day 30, confirming that the tdTomato+ HCC cells did not transdifferentiate into cells of other lineages in this chronic HCC model in situ (figure 5H). The scRNA-seq data confirmed the Prom1-derived cells as expanding epithelial tumour cells during HCC progression.

Figure 5

Single-cell transcriptome analysis reveals the cell lineage properties of Prom1-derived cells in hepatocellular carcinoma (HCC). (A) An illustration of our experimental plan, including lineage tracing, cell isolation and scRNA-seq analysis of the tdTomato- and Prom1-derived tdTomato+ cells after tumour initiation (day 3 post-tamoxifen treatment) and progression (day 10 and day 30 post-tamoxifen treatments), in the diethylnitrosamine plus carbon tetrachloride (DEN+CCL4) model. Note that all viable cells isolated from tumours were included in the analysis. (B) Based on lineage marker expression, cells were grouped into 13 clusters and illustrated as a t-SNE plot. (C) Violin plots showing distributions of absolute copy number variation (CNV) gain/loss scores among the numbered clusters from all tumour samples. (D) Violin plots showing distributions of absolute CNV gain/loss scores among the lineages from all tumour samples. Same colour codes as in (E). (E) t-SNE visualisation of 33 473 cells from all tumour samples with annotated cell lineages. (F) Heatmap showing marker gene expression of the eight cell lineages. Endo=endothelial cell; MP=macrophage/monocyte; HSC=hepatic stellate cell/fibroblast; Chol=cholangiocyte; HHP,=hybrid hepatocyte; NP=neutrophil. (G) t-SNE visualisation of the tdTomato− (upper in grey) and tdTomato+ (lower in orange) cells. (H) Percentage distribution of lineage types in the six samples.

Heterogeneous HCC cells of the Prom1-lineage display a trajectory of dedifferentiation

Next, we reintegrated the HCC cells from the 6 samples (17 431 tdTomato+ and 4593 tdTomato- cells in total), in which 4 major HCC subclusters (C0 to C3) were subsequently identified (figure 6A,B), indicating a high degree of intratumoral heterogeneity in the HCC cells. The C0 and C1 clusters accounted for the major difference in proportion between the tdTomato+ and tdTomato− cells, where the C1 subpopulation showed an expansion in the tdTomato+ cells during tumour progression (day 3 vs day 10, p=3.01E−28; day 3 vs day 30, p=2.13E−13) but not in the tdTomato- cells (day 3 vs day 10, p=0.06; day 3 vs day 30, p=0.54) (online supplemental figure S6A). Besides, the tdTomato+ HCC cells showed significantly higher CNV levels than the tdTomato− HCC cells (online supplemental figure S6B). Considering that HCC cells at various time points showed marginal changes in copy number alteration and no single subcluster could represent cells from a particular time point, we combined cells of the three time points and focused on the Prom1+ lineage for further analysis.

Figure 6

Heterogeneous hepatocellular carcinoma (HCC) cells of the Prom1-lineage display a trajectory of dedifferentiation. (A) Uniform manifold approximation and projection (UMAP) clustering of 22 024 tdTomato− and Prom1-derived tdTomato+ HCC cells at three time points combined. (B) UMAP visualisation of the tdTomato+ (tdT+, orange, 17 431 cells) and tdTomato− (tdT−, grey, 4593 cells) HCC cells. (C) UMAP plot showing clustering of 17 431 Prom1-derived tdTomato+ HCC cells into 4 subclusters (C3, C0, C2 and C1). Heatmap of representative liver function-related Gene Ontology and pathway terms enriched in each subcluster of tdTomato+ cells. Colour key from blue to red indicates the adjusted p value. (D) Violin plots showing the SCENIC regulon activities (AUCell) of Hnf4a, Cebpa and Foxa3 in the four subclusters of tdTomato+ HCC cells, with boxplots indicating the upper, median and lower quartiles. ***p<2.22e−16 by Wilcoxon rank-sum tests. Transcription factor expression levels in the tdTomato+ HCC cells are shown at the bottom. (E) UMAP visualisation of the tdTomato+ HCC cells in the same dimensions as in figure 6A–C, with visualisation of the RNA velocity field (black arrows). (F) Left panel presents loess-smoothed curves fitted to the z-scored averaged expression of all genes in the modules 1 to 3 of coregulated genes along the pseudotemporal trajectory (related to online supplemental figure S6E). Right panel shows representative Gene Ontology annotations, with the number of hits shown in brackets. (G) Gene set activities of EMT, proliferation and embryonic-like tumour signatures in the HCC cells of four subclusters shown by AUCell score. Random_500 g gene set contained 500 genes randomly picked, as a internal control. (H) Correlation matrix between mouse HCC subclusters and human HCC histological grading (Edmondson-Steiner 4-tier of The Cancer Genome Atlas (TCGA)–Liver Hepatocellular Carcinoma (TCGA-LIHC) cohort). Normalised average expression values of differentially expressed genes for each cluster or grade were used to calculate the correlation coefficient. NT=non-tumour. G1=well differentiated (low grade); G2=moderately differentiated (intermediate grade); G3=poorly differentiated (high grade); G4=undifferentiated (high grade). Colour keys from blue to red indicate low to high Pearson correlations.

The well-known CSC markers were found to be expressed in distinct patterns in the tdTomato+ cells (online supplemental figure S6C), revealing a previously unappreciated heterogeneity of the Prom1-derived HCC subpopulation. The C3 tdTomato+ cells represented pericentral lesions, with high expression of Lgr518 and Cyp2e1 (online supplemental table S3). Functional annotation analysis showed that C3 and C0 were mainly related to carbohydrate, fat, protein and drug metabolism. C2 showed decreased abilities in processing macromolecules but enhanced xenobiotic and oxidant detoxification. C1 lost most of the normal liver functions but had enhanced oxidative phosphorylation, tricarboxylic acid cycle (TCA) cycle activity, and purine nucleoside processing levels (figure 6C). Consistently, key hepatic differentiation-related transcriptional regulons targeted by Hnf4a, Cebpb and Foxa319 were decreased in C2 and C1 compared with C3 and C0 (figure 6D, online supplemental table S4). These results indicated that the C2 and C1 clusters may represent dedifferentiation statuses. RNA velocity analysis20 suggested directionality from C3 (pericentral lesions)/C0 to C2 to C1 in the tdTomato+ HCC cells (figure 6E), whereas the tdTomato− cells did not show a similar pattern (Supplementary Figure S6D). Along the trajectory of tdTomato+ cells, five modules of coregulated genes were identified (figure 6F, online supplemental figure S6E,F, online supplemental table S5). Genes of modules 1 and 5 were downregulated along the pseudotime, which included epithelial differentiation and Wnt signalling genes. Module 2 consisted of genes with highest expression levels in the middle of transition, with annotations of cellular respiration, energy derivation and redox homeostasis. Module 3 represented a small proportion of genes upregulated at the end of pseudotime, associated with cell mitosis process. Similarly, genes of module 4 rebounded towards the end of trajectory, regulating cell fates such as proliferation, programmed cell death and EMT. In accordance with pseudotime analysis, enriched positive EMT regulation, proliferation and embryonic-like21 signatures were found in the C1 subcluster of Prom1-lineage HCC cells (figure 6G, online supplemental figure S6G). These data not only supported the loss of differentiation phenotype, but also suggested a transition to a proliferative and mesenchymal-like status along the trajectory. Furthermore, the dynamic transition of Prom1-lineage HCC cells shared similar transcriptomic spectra with the human HCC differentiation grades. C1 had the highest correlation with poorly differentiated HCC (G4); C3 and C0 had combined signatures with a higher correlation with normal (non-tumour) to low-grade HCC (G1), while the C2 signature correlated with intermediate grade HCC (G1 and G2) (figure 6H). In sum, our data suggested that highly heterogeneous Prom1-derived HCC cells follow a trajectory of dedifferentiation during tumour progression. These characteristics potentially contribute to the proliferation and tumourigenicity of Prom1+ HCC progenies.

Supplemental material

Supplemental material

Supplemental material

The gene signature of Prom1-lineage predicts a poor prognosis in human HCC and its malignancy depends on the activated oxidant detoxification system

To decipher a unique mechanism facilitating the propagation of Prom1+ progenies, we identified conserved markers upregulated in the tdTomato+ cells across four subclusters (figure 7A). This analysis found 836 genes (Prom1-lineage signature, online supplemental table S6). In the TCGA-LIHC (n=340) and cross-validated International Cancer Genome Consortium (ICGC-JP, n=162, no treatment) HCC cohorts, unsupervised consensus clustering revealed two distinct groups of patients with HCC, with high and low levels of human homologs of the signature gene expression in the tumours (online supplemental figure S7A,B). The signature high group comprised 27% and 15% of patients, in TCGA-LIHC and ICGC-JP respectively, and had a significantly worse prognosis than those in the signature low group (figure 7B), revealing a profound clinical impact of the signature of Prom1-lineage. These genes were involved in cellular respiration, purine nucleotide metabolism, and regulation of apoptotic signalling, which are typical hallmarks of rapidly dividing HCC cells.22 Importantly, multiple gene ontology (GO) annotations were associated with oxidant detoxification (figure 7C), which were also a dominant signature of C2 cells during the transition from well to poorly differentiated HCC (figure 6C,F,H). Prom1+ HCC cells, especially those of C2 subcluster, were enriched of the key detoxification genes (Gpx, Gsta, Gstp, Gstm, Nqo1), as well as mTORC1 and NFE2L2 gene sets activities (online supplemental figure S7C,D) that protected HCC-initiating cells from oxidative stress-induced death.23 Functionally, the Prom1-derived tdTomato+ organoids showed a significant increase in cell viability after treatment with the reactive oxygen species (ROS)-inducing reagents H2O2 and glutathione reductase inhibitor carmustine (BCNU),24 compared with the tdTomato− organoids (figure 7D, online supplemental figure S7E). In vivo, the frequency of the tdTomato+ tumour cells increased four folds after inhibition of glutathione reductase for 3 weeks, together with significantly decreased cell death (figure 7E–G), suggesting that the HCC cells of the Prom1-lineage were resistant to oxidative stress and that this property may be a protective mechanism for rapidly expanding tumour cells to survive in a stressed tumour microenvironment.

Supplemental material

Figure 7

The gene signature of Prom1-lineage predicts a poor prognosis in human hepatocellular carcinoma (HCC) and its malignancy depends on the activated oxidant detoxification system. (A) Identification of 836 conserved upregulated genes in the tdTomato+ HCC cells compared with the tdTomato- HCC cells across four subclusters. The full gene list is provided in online supplemental table S6. (B) Consensus clustering of patients with HCC from Cancer Genome Atlas (TCGA)–Liver Hepatocellular Carcinoma (TCGA–LIHC) and International Cancer Genome Consortium (ICGC-JP) cohorts, respectively, based on the Prom1-lineage signature yields two distinct groups (related to online supplemental figure S7A,B). Kaplan-Meier curves showing the overall survival of patients with high and low expression of the signature. Kaplan-Meier curves: **p<0.01 and *p<0.05 by log-rank tests; Expression of signature: **p<0.01 and ***p<0.001 by unpaired t-tests. (C) Top Gene Ontology terms identified for the tdTomato+ HCC cells based on the 836 conserved upregulated genes, with –log10(p value) and number of hits shown. Antioxidant-related pathways highlighted in orange. (D) Organoid cultures of the tdTomato+ and tdTomato- HCC cells treated with vehicle or ROS-inducing agents, for example, hydrogen peroxide (H2O2, 0.1–100 mM) and glutathione reductase inhibitor (BCNU, 0.1–200 µg/mL), for 48 hours, after which cell viability was measured by CellTiter-Glo. Data are presented as percentages compared with the vehicle control. p<0.0001 by sum-of-square F-test. (E) Experimental scheme of the Prom1-tdTomato mice with tumours induced by diethylnitrosamine plus carbon tetrachloride (DEN+CCL4) administered tamoxifen 3 days prior to drug treatment. The glutathione reductase inhibitor BCNU (15 mg/kg) or vehicle was administered three times weekly i.p. for 3 weeks, with mice sacrificed for analysis after the last dose. (F) Representative flow cytometric images and quantification analysis of the tdTomato+ cells in HCC tumours following BCNU treatment compared with the vehicle control. n=3 mice per group, ****p<0.0001 by unpaired t-tests. (G) Representative pictures and quantitative data of terminal deoxynucleotidyl transferase biotin-dUTP nick end labeling (TUNEL) staining in liver tumour tissues from the mice in (E, F). Scale bar=50 µm *p<0.05 by unpaired t-tests.

The Prom1-marked cholangiocytes in normal adult liver are not the major origin of HCCs in our mouse models

Prom1 expression was found in cholangiocytes of the normal mouse liver,10 but whether Prom1+ HCC cells are the direct progeny of these cholangiocytes is unclear. We performed lineage tracing before tumour formation to see whether the normal Prom1+ cells contribute to tumour cells (figure 8A). Normal adult Prom1-tdTomato mice treated with multiple doses of tamoxifen showed over 80% labelling efficiency in Prom1+ cholangiocytes, suggesting a minimal possibility of false negative tracing (figure 8B,C). Co-staining for the presence of Sox9high/A6/Krt19 and the absence of the hepatocyte marker Hnf4a further validated that the tdTomato+ Prom1-expressing cells represented the biliary lineage (figure 8D). Prom1-derived cells in the normal liver have a limited ability to differentiate into hepatocytes during liver homeostasis or to replenish injured livers (online supplemental figure S8A,B). This observation is in line with previous reports showing that ductal cells have a minimal contribution to hepatocyte neogenesis in normal adult liver or when hepatocyte proliferation is not completely disrupted by liver injuries.25 26 Additionally, even when HCC was induced after the ductular Prom1+ cells were labelled, very few tdTomato+ cells could be detected in the Afp+ HCC nodules (figure 8E,F). Specifically, the tdTomato+ HCC cells were only detected in 8.1% of the tumours (13/160 tumour nodules from 8 mice) in the DEN+CCL4 HCC model and 4.0% of the tumours (7/176 tumour nodules from 6 mice) in the N-Ras+AKT HCC model. Most tdTomato+ ductal cells remain inactive in the periportal space, suggesting that most tumours (>94% on average of two HCC models) do not originate from them. Consistently, ablation of Prom1+ cholangiocytes with conditional DTA induction before HCC initiation did not interfere with tumour growth (online supplemental figure S8C,D).

Figure 8

The Prom1-marked cholangiocytes in normal adult liver are not the major origin of hepatocellular carcinomas (HCCs) in our mouse models. (A) Experimental scheme for tracing normal Prom1+ cells in liver cancer models. Adult Prom1-tdTomato mice were treated with five doses of 4 mg tamoxifen every other day. Three days after the last treatment carbon tetrachloride (CCL4) treatment or hydrodynamic transfection were administrated. (B) Representative images of the Prom1-expressing cells in adult normal livers (baseline, online supplemental figure S8A). Arrowheads mark clusters of Prom1+ cells. Left and middle, scale bar=100 µm. Right, high magnification, scale bar=20 µm. PV, portal vein. (C) An average baseline labelling efficiency of 83.60%±0.86% was found by merging the tdTomato fluorescent signal with β-gal/Prom1 antibody staining (79 periportal areas from n=4 mice). (D) Liver epithelial markers (Sox9, A6, Krt19 and Hnf4a) were costained with tdTomato. Scale bar=100 µm. Inset, scale bar=20 µm. (E, F) Representative immunofluorescence images (E) and macrophotographs (F) of the HCC tumours in the two liver cancer models. Upper: diethylnitrosamine (DEN)+CCL4 model; lower: N-Ras+AKT model. (E) Afp was costained as an indicator of HCC tumours. High magnification images show typical Prom1-expressing HCC cells and cholangiocytes. Left, scale bar=500 µm. Right, high magnification, scale bar=50 µm.


Lineage tracing has been used to investigate in vivo CSC properties in established tumours.27 In HCC, most lineage tracing studies are devoted to identifying stem cell populations in normal/injured livers and whether they are the potential cell-of-origin of HCC,8 15 while the behaviour of stem-like cancer cells during tumour progression has not been fully elucidated. By reconciling transplantation and lineage tracing approaches, our study indicated that the Prom1-marked HCC cells in early HCC nodules represent a proliferative and actively propagating population. They displayed superior ability of in vivo tumourigenicity, a typical CSC-like feature. In the colon and intestine, differentiated cells showed plasticity of being able to replenish the ablated stem cells/CSC population.27–29 Similarly, though not inherently the same, the mature differentiated hepatocyte can replenish the damaged or lost tissues during hepatic injury and regeneration.30 31 The Prom1+ tumour-propagating cells in the tumours of early stage expressed mature hepatocyte marker Hnf4a rather than progenitor markers, whereas they retained the ability to form progenitor-like HCC21 or CCA-like/EMT-like tumours during transplantations. Regardless of the origin of these tumour-propagating cells, this observation in HCCs was in accordance with the remarkable plasticity of mature hepatocytes in liver regeneration and homeostasis,32 and the ability to form cells of differential lineages represented another key CSC-like trait.

We turned to scRNA-seq for a higher resolution of the lineage and molecular characteristics of Prom1+ cells in HCC tumours. HCC cell subpopulations and their lineage dynamics at the single-cell level remain incompletely elucidated in previous studies.33–36 Two recent papers revealed the heterogeneity of Prom1+/PROM1+ cells in HCC by scRNA-seq, showing the heterogeneity of nonparenchymal Prom1+ cells in vivo37 and differential transcriptomic patterns among marker-defined CSCs in vitro,38 respectively. Our model provided robust recovery of lineage-traced HCC cells by FACS, which circumvented the limitation of tumour cell isolation and took snapshots of the same lineage in vivo at various time points. On this basis, we were able to identify heterogeneous subclusters of Prom1+ HCC cells and further perform trajectory analysis implied by RNA velocity.20 The HCC cells of Prom1-lineage lost the hepatic phenotype and markers along the trajectory, a process we defined as ‘dedifferentiation’. We propose a model of progression from well-differentiated HCC cells, which phenotypically resemble normal hepatocytes, to cells with loss of hepatocyte function and differentiation markers. The former population is enriched in Wnt signalling pathways, in line with the features of CTNNB1 mutant HCC.16 The latter population increased EMT, proliferation and embryonic-like signatures, accumulating at the end of cellular trajectory of the Prom1+ lineage. These signatures are correlated with the stemness21 39 and dedifferentiation events40 in liver CSCs. Importantly, a subset of antioxidant and detoxification genes were upregulated in C2 subcluster during the dedifferentiation transition. Previous reports have shown that the in vitro ROS stress can induce CD133/PROM1 expression,41 and that the activation of Sqstm1/Nfe2l2/mTorc1 signalling pathways maintains survival of stressed HCC-initiating cells.23 How cancer cells of Prom1-lineage are correlated with ROS scavenging in vivo remains uninvestigated. Here we showed at the single-cell level that the Sqstm1, Nfe2l2 target genes,42 as well as mTorc1 and Nfe2l2 regulatory gene set activities were upregulated in Prom1+ HCC cells especially C2 subcluster, highlighting the ability of Prom1+ cells to combat ROS stress in vivo. A recent large-scale drug screening study showed that HCC cells with KEAP1 mutation/NRF2 activation are resistant to sorafenib treatment.43 Nfe2l2 inactivator may prevent the resistance to conventional drug44 and have the potential to impede disease progression for patients with HCC with high Prom1 signature. Notably, selective Prom1+ cell depletion resulted in reduced HCC tumour growth, but not tumour regression, indicating that other subpopulations might also play a role in tumour propagation or replenish Prom1+ cells.28 Besides, Prom1+ cells did not transdifferentiate to other cell types in the in situ tumours; therefore, we did not hypothesise Prom1+ cells to be a canonical CSC/progenitor population, rather, we pointed out that cells with CSC-like features can be heterogeneous and exhibit dynamic cellular traits in vivo. The malignancy of Prom1+ HCC cells may come from their trajectory towards dedifferentiation and proliferation status, as well as the anti-ROS capacity during the trajectory.

Liver progenitor cells is plausibly one of the origins of HCC.45 During early infancy stages (<P14), Prom1 marks liver progenitor cells because these cells exhibit bipotent differentiation potential.46 However, the Prom1+ cells lose hepatocyte differentiation potential and become cholangiocytes at adolescence and adult stages (online supplemental figure S9A–C).11 Moreover, adult Prom1-marked cholangiocytes rarely differentiated into hepatocytes in liver injury models and mostly stayed in peritumoural ductular area in HCCs. Therefore, we concluded that the Prom1-marked cholangiocytes in normal adult liver are not the major origin of HCCs, in the DEN+CCL4 and N-Ras+AKT mouse models. A few remarks need to be contextualised. First, the major origin of HCC is closely related to the model of HCC induction. DEN, the most commonly used chemical carcinogen for HCC induction, is mainly converted into toxic metabolites by pericentral hepatocytes.47 Consequently, labelling pericentral Lgr5+ hepatocytes traced around 40% of DEN-induced tumours8 while most studies labelling cholangiocytes6 7 25 or periportal hepatocytes15 with similar treatment scheme yielded negative results. Likewise, hydrodynamic transfection predominantly targets pericentral hepatocytes,48 leaving Prom1+ cholangiocytes unaffected to a large extent. Along the same lines, mutations on ductular reaction treatment, such as DDC diets, can induce liver cancers originated from ductular cells.9 11 A recent report showed Prom1+ cells gave rise to DEN-initiated, Western alcohol diet (WAD)-promoted HCC with ductular reaction progenitors (DRPs) activation in peripheral borders.37 It will be interesting to further investigate how WAD activates ductular cell response and the cell-of-origin of this HCC model. Second, the timing of tracing also affects the results. We traced Prom1+ cells in adult livers49 where Prom1 no longer marked progenitors but mature cholangiocytes. Nevertheless, regenerative Lgr5+, Axin2+ or Prom1+ cells in neonatal livers can give rise to traced HCC.8 11 50 Third, we do not exclude the possibility that ductular cells can generate HCCs and yet it remains to be discussed as to what extent. Altogether, our data suggests that HCC cell-of-origin and HCC tumour-propagating cells should be discussed as independent populations.

Regardless of the cell-of-origin, proliferative DRPs are implicated in liver cancers and associated with poor patient outcomes.51 However, in the DEN+CCL4 HCC model, DRPs (Ki67+; Krt7+) and ductular cells (Krt7+ only) predominantly located in peritumour areas, and tdTomato+ DRPs were rarely observed (around 0.9% of all tdTomato+ cells, online supplemental figure S9D–F). Most of the Prom1-derived tdTomato+ cells were intratumoral and nonductular, with an increasing proportion over all tdTomato+ cells during tumour growth, indicating their competitive amplification. Notwithstanding, we do not exclude the possibility that DRPs/ductular cells in peri-tumour area may have paracrine effect for tumourigenesis. Further investigation to specify the role of Prom1+ ductular cells can be performed using Cre-lox/Dre-rox dual recombination system.52 Notably, we identified a hybrid population with both hepatocytes and cholangiocytes markers and pluripotency signature by scRNA-seq (online supplemental figure S9G,H), but this population, as well as cholangiocytes/ductular cells, maintained low CNV levels during HCC progression, suggesting that they are nonmalignant cells, consistent with the previous report.15

In the future, it would be interesting to determine whether the drug-resistant/relapsed/metastatic HCC subclones are derived from dedifferentiated cells or directly branched from well-differentiated HCC cells.53 An equally interesting question is whether Prom1+ cells differ from other tumour cells in regulating or interacting with the tumour microenvironment during HCC progression. In sum, the systematic investigation of Prom1+ cells during HCC progression provides novel insights into HCC cell heterogeneity and opens new avenues for therapeutic and mechanistic investigations in this field.

Supplemental material

Supplemental material

Supplemental material

Data availability statement

Data is available on GEO (GSE181515). We uploaded our raw data to GEO (GSE181515), which will be open to public access upon paper acceptance. The following secure token has been created to allow review of record GSE181515 while it remains in private status: mxwnsyuwjnexbaj. We also uploaded the scripts for all the analysis performed in the paper onto Github (

Ethics statements

Patient consent for publication


The authors would like to thank the Imaging and Flow Cytometry Core as well as the Genomics and Bioinformatics Core in the Centre for PanorOmic Sciences (CPOS) of the University of Hong Kong for providing and maintaining the equipment and technical support needed for single-cell RNA sequencing, flow cytometry and confocal microscopy studies. We also thank the Centre for Comparative Medicine Research (CCMR) of the University of Hong Kong for supporting our animal work studies. We thank Prof. Kathryn Cheah of the University of Hong Kong for providing reporter mouse lines, Prof. Xin Chen of the University of California San Francisco for providing plasmids in establishment of hydrodynamic transfection liver cancer models, and Prof. Hans Clevers of Hubrecht Institute as well as Prof. Suet-Yi Leung and Dr. Helen Yan of the University of Hong Kong for providing Rspo1 cell line and share of know-how for organoid culture. This project is supported by the Research Grants Council of Hong Kong – Collaborative Research Fund (C7026-18G) as well as the Innovative Research Fund of State Key Laboratory of Liver Research (The University of Hong Kong) (SKLLR/IRF/2018/02) to SM, and the National Natural Science Foundation of China (82002605) to LZ.


Supplementary materials


  • LZ, KHY and TLW contributed equally.

  • Correction notice This article has been corrected since it published Online First. A second corresponding author has been added.

  • Contributors LZ and SM conceived the project, designed the experiments and wrote the manuscript. LZ performed the majority of the experiments including mouse model establishment and characterisation, cell culture, sequencing and data analysis. LZ, KHY and TLW analysed the single-cell sequencing data and provided critical scientific input. TLW helped with manuscript editing. ZZ, CHC, JHL, NC, HJY and MT aided in animal experiments. KVT performed MRI scanning. ESN and JWH provided guidance in single cell sequencing analysis and manuscript editing. SM, JWH and LZ provided funding support. SM and JWH supervised the project.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.