Transcriptome of PROM1+ cells from AH resembles known DRP and HCC gene signatures
To characterize a gene expression profile of PROM1+ cells in AH, we isolated PROM1+ cells from AH mouse livers14 by FACS using antibodies against PROM1 and CD49f which were used as stem cell markers for mouse liver TICs4,16. In normal mouse liver, a PROM1+CD49f+CD45– population accounts for ~ 0.1% which increases to 1.93% in diethyl nitrosamine (DEN)-liver cancer model4. The yield of PROM1+CD49f+CD45– cells from AH mouse livers ranged from 1 to 4% of sorted cells (Fig. 1a), suggesting PROM1+ cells are increased in the liver of AH mice. To characterize PROM1+ cells, PROM1+CD49f+CD45– (R6) versus PROM1– CD49f+CD45– (R5) cells were sorted and total RNA was sequenced as shown in a flow chart in Fig. 1b. We also sorted PROM1+CD49f+ cells present in CD45+ population as hematopoietic PROM1+ cells (R8).
Total RNAseq analysis revealed a unique transcriptomic landscape of R6 PROM1+ cells as compared to R5 PROM1– or R8 hematopoietic PROM1+ cells (Fig. 1c). To identify differentially regulated genes in R6 (PROM1+) cells, we compared gene expression profiles of R6 versus R5 (PROM1–), R6 versus R8 (hematopoietic PROM1+), and R6 versus all CD45– sorted cells. As shown in the Venn diagrams in Fig. 1d, this analysis identified 613 upregulated and 708 downregulated genes specific for PROM1+ liver cells (Suppl. Table 3 in Excel format). The upregulated gene list included twofold or higher expression of genes considered as markers for progenitors (Prom1, EpCAM, Sox9), biliary epithelial cells (Krt19, Hnf1β, Spp1) and mature hepatocytes (HNF4α, Alb), suggesting hepatic PROM1+ cells in AH are bipotent progenitor cells. Ingenuity Pathway Analysis of the PROM1+ cell transcriptome, also revealed significantly regulated stemness and tumorigenesis pathways such as pluripotency, EMT, tissue factor in cancer, Ephrin receptor, HIPPO pathway, and ILK signaling, suggesting their potential involvement in liver tumorigenesis (Fig. 1e).
Lineage of PROM1+ cells in AH and their fate in liver cancer
To determine how PROM1+ cells are generated in AH, we performed lineage tracing in the AH mouse model. To test the contribution of hepatocytes to the genesis of PROM1+ cells, we utilized two approaches to label hepatocytes in Rosa26mTmG reporter mice before AH induction: the use of the conventional Albumin-Cre (Alb-Cre) and the more mature hepatocyte specific Transthyretin-Cre (Ttr-Cre). Alb-Cre; Rosa26mTmG mice showed strong GFP expression by hepatocytes. Using these cells, we were able to establish the gating for GFP positive cells.
To generate Ttr-Cre mice, we injected AAV8-Ttr-Cre virus to Rosa26mTmG mice via tail vein and allowed 1 month for recombination to occur. The labeling efficiency was 98.8% for Alb-Cre mice and 99.2% for Ttr-Cre mice according to GFP positivity (Suppl. Figure 1b and 1c). To trace GFP+ hepatocytes to PROM1+ DRPs in the AH mice, we isolated CD45– nonparenchymal liver cells by a low speed centrifugation to remove hepatocytes and magnetic depletion of CD45+ cells. The cells from C57/BL6 mice after the AH protocol were used as a negative control because the Western diet regimen contributed to high background fluorescence (Suppl. Figure 1d). The cells were sorted from Alb-Cre;Rosa26mTmG mice (Suppl. Figure 1g) and Rosa26mTmG mice injected with AAV8-Ttr-Cre (Suppl. Figure 1j) for PROM1 and GFP. From the cells from AH liver of Alb-Cre;Rosa26mTmG mice, PROM1+ /GFP+ cells were detected at 0.118% of frequency versus PROM1+ /GFP– cells at 0.172%, suggesting 40.7% of PROM1+ cells are GFP-traced cells from Alb-Cre-based labeling in this representative mouse (Suppl. Figure 1g). Using Ttr-Cre labeling method which is considered more specific for hepatocytes17, the frequency of GFP+ /PROM1+ cells was shown to be 0.195% as compared to 0.691% of total PROM1+ cells with or without GFP expression, indicating 28.2% of the PROM1+ cells are hepatocyte-derived in this particular Ttr-Cre;Rosa26mTmG mouse subjected to AH (Suppl. Figure 1j, right panel). After repeating the same analysis in 4 AH mice, the average percentage of GFP+ /PROM1+ cells was determined to be 24% (Suppl. Figure 1k). These results suggest that PROM1+ cells are generated in part from mature hepatocytes in AH and the origin appears heterogeneous. The latter notion is also consistent with the recent finding of the heterogeneity of PROM1+ cells in HCC18.
A subpopulation of PROM1+ cells observed in HCC is suggested to possess tumor initiating properties4. Indeed, the DRP markers such as Prom1, EpCAM, Krt19, and Sox9 detected in AH, were also induced in non-tumor liver tissue (NTL) of mice subjected to DEN injection and tumor promotion with WAD as compared to normal liver or tumor tissues, suggesting a more selective increase of DRPs in NTL (Fig. 2a). Indeed, immunofluorescent microscopy detected proliferating DRPs were concentrated in NTL around the peripheral border of the tumor mass as Ki67+ SOX9+ cells (purple arrows, Fig. 2b), revealing a close spatial association of DRPs with tumor cells.
To determine the fate and role of PROM1+ cells in liver tumorigenesis, we traced PROM1+ cells using the DEN-initiated and WAD-promoted liver tumor model in Prom1C-L mice, a strain engineered to have CreERT2 and Lacz genes in the Prom1 locus19. While this mouse serves to express CreERT2 under the Prom1 promoter for labeling of PROM1+ cells, the homozygous strain of Prom1C-L/C-L can also be used as Prom1 knockout because it completely ablates the Prom1 expression due to the biallelic knock-in of the CreERT2-Lacz cassette. We crossed Prom1C-L/C-L mice with Rosa26mTmG mice to generate Prom1 C-L; Rosa26mTmG (Prom1WT) and Prom1C-L/C-L;Rosa26mTmG (Prom1KO) mice, injected DEN at 2-week age and fed WAD from 5-week age for 6 months. Tamoxifen (Tam) was injected to label PROM1+ cells at 3 weeks after DEN injection (Fig. 2c). Liver tumor development in Prom1WT and Prom1KO mice was assessed via gross and histologic examination (Fig. 2d,e). As predicted, Prom1 expression in tumors of Prom1C-L/C-L;Rosa26mTmG (Prom1KO) mice were almost undetectable by qPCR analysis, validating their knockout status (Suppl. Figure 2a).
Surprisingly, total liver tumor volumes calculated in these two genetic groups, were not different (Fig. 2d), indicating PROM1 is not essential in liver tumorigenesis. To determine the fate of PROM1+ cells in tumor development, we stained the liver sections with visible tumors with anti-GFP and anti-KRT19 antibodies (Fig. 2f,g). While DRPs were detected as GFP+ KRT19+ cells in NTL region, tumor cells were also labeled with GFP but not for KRT19. These results suggest that PROM1+ cells give rise to liver tumor cells even though PROM1 is not functionally required for tumor development.
scRNAseq analysis reveals Prom1+ TIC and DRP subpopulations in liver tumor
To understand the molecular characteristics of Prom1+ cells in the single cell resolution, we performed scRNAseq analysis. To enrich the LPCs and TICs, we have developed a unique approach of sorting Col1a1-expressing cells by FACS. The rationale for this approach is based on: (1) LPCs were previously reported to express high levels of Col1a1 expression3; (2) our own results confirmed 23-fold and 90-fold higher expression of Col1a1 in LPC line (PIL4) and TICs as compared to mouse primary hepatocytes, respectively (Suppl. Figure 3a); and (3) our additional data validated higher expression of Col1a1 in PROM1+ versus PROM1– cells isolated via MACS from the DEN-WAD mouse livers (Suppl. Figure 3b). We used Col1a1-driven Cre recombinase expression to label LPCs and TICs in Col1a1Cre;Rosa26mTmG mice subjected to the DEN-WAD protocol. Tumor bearing livers were perfused and digested with collagenase to collect liver cells. Mature hepatocytes were removed by low speed centrifugation, and the remaining liver cells were subjected to FACS sorting based on ultraviolet (UV) and GFP. UV-excited gating was used to exclude vitamin A-storing hepatic stellate cells (HSC), the major Col1a1-expressing cell type in the liver. The UV–GFP+ population was sorted as tumor-associated Col1a1-expressing cells and submitted for scRNAseq analysis (Fig. 3a). Barcode rank plot of the scRNAseq sample showed an excellent separation between the cell-associated barcodes and barcodes associated with empty partitions (Suppl. Figure 4). From a single cell library with the target cell number of 10,000, 8578 cells passed this quality control assessment. By this scRNAseq analysis, we indeed identified two subpopulations which included Prom1+ cells (Cluster 1 and 2) (Fig. 3b) with no or minimal expression of the HSC marker Lrat, portal fibroblast marker Thy-1, hematopoietic marker Cd45, and endothelial cell markers Cd31 and Cdh5 (Suppl. Figure 5a). As predicted, Cluster 1 and 2 cells expressed the biliary or/and progenitor markers such as Sox9, Spp1, Hnf1b, Krt19, Krt23, Epcam (Fig. 3c). To further validate these results, we assessed co-expression of these genes with Prom1 at the single cell level. As shown in Suppl Fig. 5b, Cluster 1 and 2 are enriched with cells which co-expressed Prom1 and Sox9, Spp1, Hnf1b, or Krt23 and most Prom1+ cells in both clusters did not co-express the HSC, portal fibroblast, hematopoietic, and endothelial cell markers, validating these Prom1+ Cluster 1 and 2 cells are LPCs. The cells co-expressing Prom1 and Afp, the known marker of liver tumor cells, were detected primarily in Cluster 1 (Suppl Fig. 5a and Fig. 3a). In fact, Cluster 1 contained both Prom1+ Afp+ and Prom1–Afp+ cells, suggesting TICs and tumor cells were enriched in this Cluster. Prom1+ Afp– cells enriched in Cluster 2 are DRPs as the biliary/DRP markers such as Hnf1b, Krt19, Epcam, Mmp7 and Cftr were more selectively expressed in Cluster 2 cells (Fig. 3c).
We next assessed whether the known TIC/HCC transcripts are expressed in the scRNAseq profiles of the two clusters. For this, we assessed a set of genes reported to characterize cancer progenitor cells with tumor-initiating properties which were isolated from DEN-induced tumor bearing livers20. Many of the genes upregulated in these tumor initiating cells were expressed in our Cluster 1 cells including Tff3, Tspan8, Gpc3, Ly6d, Cldn2 (Fig. 3c). These genes are involved in malignancies and cancer stemness21,22,23, further supporting the notion that Cluster 1 cells are TICs and tumor cells. We then acquired gene expression profiles unique to each of the three populations of Prom1+ Afp+ , Prom1+ Afp–, and Prom1–Afp+ cells by comparing among them by the locally distinguishing comparison feature of the Loupe Browser as depicted by heatmaps and volcano plots (Fig. 4a–d). As shown in the heatmaps, Prom1+ Afp+ and Prom1–Afp+ cells share coordinately upregulated genes most likely reflecting their common tumorigenic function yet with some discordant gene expression which may reflect genetic differences between TICs with stemness property and tumor cells which have lost the stemness. In contrast, Prom1+ Afp– cells which we consider as DRPs have a clearly distinct gene expression pattern from other two Afp+ cell populations. Expression of 4 top genes uniquely expressed in each population are shown in Fig. 4e–g, and complete lists of the genes are provided in Suppl. Tables 4–6 in Excel format.
Discoidin domain receptor 1 is a potential oncogenic driver of PROM1+ cells
To address the relevance of the transcriptomic data from PROM1+ cells to AH livers in mice and patients, we compared our PROM1+ cell RNAseq profile with a RNAseq profile of AH mouse liver and a previously reported proteomic profile of human AH liver24. This integrated analysis identified 7 upregulated and 7 downregulated genes that are common in PROM1+ cells, mouse and human AH livers (Fig. 5a). Since our results showed PROM1+ cells gave rise to liver tumor cells, we looked for potential oncogenic drivers unique to PROM1+ cells by first screening the expression of the 7 upregulated genes in NTL tissues of DEN/WAD mice compared to normal liver. This analysis revealed that Prom1, Ddr1, Ddr2, Lamc2, Lamc3 and Pkhd1 were induced significantly in DEN-WAD livers (Fig. 5b). Next, we examined the expression of these genes in cell clusters revealed by our scRNAseq analysis. Of these genes examined, Ddr1 and Pkhd1 expressions were detected in Cluster 1 and 2 (Fig. 5c). DDR1 is known as a receptor for collagen25 and implicated in oncogenesis, metastasis and chemoresistance of cancers of pancreas, lung, prostate, ovary, and colon26,27,28,29. Ddr2, in contrast, is expressed in clusters of activate Lrat+ HSC and Thy1+ portal fibroblasts (Fig. 5c and Suppl. Figure 5). Spint1 and Ehf were expressed more selectively in DRPs in Cluster 2 (Fig. 5c), suggesting their importance in the biology of Prom1+ DRPs.
To confirm DDR1 upregulation in AH in patients in our hands, we analyzed DDR1 mRNA levels in liver tissues of AH patients versus healthy subjects received from the explant repository program of John Hopkins University. Indeed, this analysis revealed at least 20-fold higher DDR1 expression in AH livers compared to the normal liver (Fig. 6a). DDR2 level was also statistically higher in AH tissues than normal liver, but the fold induction was much lower than DDR1. To address the role of DDR1 in human liver cancer, we performed the data mining approach. This bioinformatic analysis showed that DDR1 expression was significantly elevated in both HCC30 (Fig. 6b) and cirrhotic livers31 (Fig. 6c) compared to normal livers.
Patients with combined HCC-intrahepatic cholangiocarcinoma (ICC) have poor prognosis33. ICC is characterized by higher expression of liver progenitor markers such as EpCAM and KRT19, suggestive of expanded TICs and/or DRPs. Indeed, DDR1 expression is higher in patients with ICC or combined HCC-ICC compared to those with HCC alone (Fig. 6d)34.
Next we assessed the relationship of DDR1 and PROM1 expression with the HCC patient survival, by analyzing in TCGA HCC cohort data using UCSC Xena database32. We stratified the cohort into DDR1 high and low expression groups. The Kaplan–Meier analysis showed that DDR1high patient group had lower survival than the DDR1low patients (p < 0.039) (Fig. 6e). When PROM1 was used to stratify the cohort, PROM1high patients had a significantly worse outcome in their survival compared to the PROM1low group (p < 0.021) (Fig. 6f). When PROM1 and DDR1 co-expression was analyzed, 23 patients with high expressions of both PROM1 and DDR1 had a worst outcome compared to 69 patients with both genes expressed at the low level (p < 0.014) (Fig. 6g). The difference in survival in these patients was greatest (> 1825 days). In summary, DDR1 is induced in human cirrhosis, HCC, and ICC and inversely associated with the survival of HCC patients. In addition, the expression of DDR1 is correlated with PROM1 expression and predicts the survival of HCC patients in this TCGA cohort. These data suggest that DDR1 expressing PROM1+ DRPs and TICs may determine the clinical course of HCC patients.
Functional significance of DDR1 in TIC-derived tumor development
To determine the functional significance of DDR1 expressed in PROM1+ cells in liver oncogenesis, we first measured Ddr1 expression in TICs, PIL4 liver progenitors and mouse primary hepatocytes. TICs were isolated from a mouse model of alcohol-promoted liver cancer4. PIL4 cells were previously isolated from p53-null mouse fed with choline-deficient, ethionine-supplemented diet35. TICs have self-renewing and tumor-initiating activities while PIL4 cells lack these properties but mimic DRPs. As expected, albumin is expressed exclusively by primary hepatocytes but not by TICs and PIL4 cells while Krt19 (CK19) is expressed by PIL4 cells. Tlr4 which is induced in stem cell-like cells4, is expressed moderately by PIL4 cells and most conspicuously by TICs along with Nanog, Sox2 as previously reported (Suppl. Figure 6). A pattern of Ddr1 expression followed that of Tlr4: undetectable in primary hepatocytes, moderately increased in PIL4 cells, and highest expression (~ 40-fold) in TICs. When Ddr1 but not Ddr2 was selectively knocked down 80% in TICs by using two separate shRNAs (Ddr1-sh1 and Ddr1-sh2) versus scrambled shRNA (Scr-sh) (Fig. 7c), TIC growth was significantly impaired as shown by crystal violet staining and cell count (Fig. 7a,b). Further, the treatment with the DDR1 inhibitor (DDR1in7rh), suppressed the growth of TICs at IC50 of 0.95 µM and that of the liver cancer cell line DIHXY at IC50 of 1.55 µM while PIL4 cells were four–fivefold less sensitive (Fig. 7d). These results suggested that DDR1 has a functional significance in TIC and liver cancer cell growth. To test this potential oncogenic role of DDR1 in vivo, we performed a xenograft experiment in nude mice transplanted with TICs transduced with Ddr1-sh1 versus Scr-sh. Ddr1 knockdown in TICs significantly attenuated TIC-derived tumor growth to one third of that with TICs with Scr-sh, supporting the functional role of DDR1 in tumor development (Fig. 7e).