Model of fibrolamellar hepatocellular carcinomas reveals striking enrichment in cancer stem cells

The aetiology of human fibrolamellar hepatocellular carcinomas (hFL-HCCs), cancers occurring increasingly in children to young adults, is poorly understood. We present a transplantable tumour line, maintained in immune-compromised mice, and validate it as a bona fide model of hFL-HCCs by multiple methods. RNA-seq analysis confirms the presence of a fusion transcript (DNAJB1-PRKACA) characteristic of hFL-HCC tumours. The hFL-HCC tumour line is highly enriched for cancer stem cells as indicated by limited dilution tumourigenicity assays, spheroid formation and flow cytometry. Immunohistochemistry on the hFL-HCC model, with parallel studies on 27 primary hFL-HCC tumours, provides robust evidence for expression of endodermal stem cell traits. Transcriptomic analyses of the tumour line and of multiple, normal hepatic lineage stages reveal a gene signature for hFL-HCCs closely resembling that of biliary tree stem cells—newly discovered precursors for liver and pancreas. This model offers unprecedented opportunities to investigate mechanisms underlying hFL-HCCs pathogenesis and potential therapies.

Clustering analysis was based on Euclidean distance and complete linkage and performed using the 3,000 most highly expressed genes (a), the 10,000 most variable genes (b), or the 3,000 most variable genes (c). All genes used had an average expected normalized count > 100 across all samples.   . Expression data for representative cell type marker genes. (a) RNAseq normalized expected count data shown across all cell types for genes that have previously been reported as markers of stem cells/progenitors (SOX9, FOXL1, CD44, ITGA6 CD49f and ITGB1 CD29), hepatocytes (DLK1, AFP, HNF4A, CPS1, and APOB), biliary tree (KRT7, ITGB4 and ONECUT2), and pancreas (PDX1 and PCSK1). (b) Expression data for genes encoding histone deacetylases. Log 2 relative expression value shown across all cell types for genes that code for histone deacetylases. Histone deacetylase 9 (HDAC9) is lost entirely in hFL-HCCs. (c) Expression data for genes in the hedgehog signaling pathway. Log 2 relative expression value shown across all cell types for genes in the hedgehog signaling pathway. Error bars represent standard error of the mean. The normal counterparts are likely to be stage 2 BTSCs. However, since all culturing of the hBTSCs is correlated with LGR5 expression, and since we do not know if culturing of the cells activates LGR5, there is the possibility that the tumor line is actually derived from stage 1 BTSCs. Pluripotency genes: SOX2, OCT4, KLF4, NANOG, TROP-2, BMI1 and SALL4. Cells at all stages express CK8 and CK18. NIS= sodium iodide symporter. **Stage 1 hBTSCs have been observed in situ in the biliary tree but have not been successfully cultured under the conditions tested. hBTSCs surviving in culture in serum-free, Kubota's Medium and on plastic or on hyaluronans are either stage 2 or stage 3 hBTSCs.  The values of these ratios for most samples are close to 2, which is the standard for high purity RNA samples. We obtained > 180 million reads for each sample and, on average, >85% of these reads were uniquely mapped. A small percentage of reads were mapped to multiple locations or not mapped at all. The rest of the mapping statistics provides detailed information on the proportion of reads that corresponded to unspliced regions, splice junctions, or fusion junctions. Many of the "candidate fusions" are not high-confidence. As described in the manuscript, only one high-confidence recurrent fusion was detected, and this was DNAJB1-PRKACA. It was unique to and present in all the hFL-HCCs we analyzed.

MSKCC samples#
Total 4/9 7/9 7/9 7/9 9/9 5/5 Supplementary Table 3. Summary of IHC assays on paraffin sections of original blocks (primary FL-HCCs). The IHC assays on paraffin sections of hFL-HCCs from 9 donors indicated that all are positive for sonic hedgehog (SHH) and, of those assayed, all are positive for HepPar-1. The majority of the tumors (7/9) were positive for SOX9, PDX1, and for NIS, and 4/9 for BMI1. There were two distinct patterns of expression consisting of 1) ones in which most or almost all were positive for a given antigen (e.g. HepPar-1, SHH, NIS and SOX9) but with heterogeneous levels of expression or 2) a pattern in which a percentage of the cells were positive (at least 20%) and the remainder negative (e.g. PDX1 and BMI1).   The values of these ratios for most samples are close to 2, which is the standard for high purity RNA samples. We obtained > 180 million reads for each sample and, on average, >85% of these reads were uniquely mapped. A small percentage of reads were mapped to multiple locations or not mapped at all. The rest of the mapping statistics provides detailed information on the proportion of reads that corresponded to unspliced regions, splice junctions, or fusion junctions. Many of the "candidate fusions" are not high-confidence. As described in the manuscript, only one high-confidence recurrent fusion was detected, and this was DNAJB1-PRKACA. It was unique to the hFL-HCCs.  further tissue for tumor sensitivity studies and debulking. Biopsies were taken, but his disease was too extensive for debulking. Paclitaxel and thalidomide were then started based on sensitivity studies but were poorly tolerated with continued disease progression, so treatment was stopped. After 4 months it was realized that he had widely disseminated disease especially in the ascites fluid. In early February, 2010, a palliative paracentesis was done for massive ascites, and approximately 5 liters of fluid were removed and transferred to several researchers, including those in the UNC research lab, in hopes that studies on the tumor might identify alternate treatments. A week later the patient passed away peacefully.

Background on normal human biliary tree stem cells
More detailed studies and reviews on biliary tree stem cells (hBTSCs) are given elsewhere 1, 2, 3, 4, 5, 6,7,8,9,10,11,12,13,14 . In brief, the biliary tree contains stem cell niches, peribiliary glands (PBGs), mucinous glands scattered as intramural PBGs within the walls of the bile ducts and also found as extramural PBGs tethered to the surface of the bile ducts 15,16,17 . The phenotypes of the cells within the intramural PBGs can be relatively homogeneous in some sites (e.g. hepato-pancreatic common duct and intrahepatic, large bile ducts) and quite heterogeneous in other sites (e.g. cystic duct, common duct, hepatic duct) 6,7,9 . The pattern of phenotypic traits of the PBG cells has been found to indicate maturational lineages in a radial axis from the fibromuscular layer within the duct walls to the lumen of the bile ducts and in a proximal (duodenum)-todistal axis from the duodenum to either liver or pancreas 1, 4 The PBGs deepest within the bile ducts and near the fibromuscular layer contain the most primitive hBTSCs, those that co-express transcription factors for both liver and pancreas (e.g. SOX17, PDX1), that also co-express multiple pluripotency genes (e.g. OCT4, SOX 2, KLF4/KLF5, NANOG) 1, 2, 4 and that have surface markers of CD44, the hyaluronan receptor, and sodium iodide symporter (NIS). These cells do not express epithelial cell adhesion molecule (EpCAM) or even LGR5. We refer to these as primitive hBTSCs or, The reason for the absence of stage 1 hBTSCs in the cultures is unknown but is hypothesized to be either that distinct culture conditions are required for them or that culturing (or proliferative state?) of the cells triggers expression of LGR5. Parallel findings have been shown in intestinal stem cell subpopulations in which LGR5+ with and without EpCAM+ expression identify important subpopulations driven by wnt signaling pathways 18 .
These are questions being addressed in ongoing studies.

Relevance of stem/progenitor cells to liver regeneration
Liver regeneration postnatally has long been known to involve adult parenchymal cells in either hyperplastic (complete cell division) or hypertrophic responses (DNA synthesis with absence of cytokinesis leading to polyploidy) 19 . Hyperplastic responses dominate following injuries to zone 3 parenchymal cells (and even more so if zone 2 cells are also damaged). By contrast, partial hepatectomy elicits a wave of DNA synthesis across liver plates but with minimal, if any, cytokinesis in zones 2 and/or 3 parenchymal cells and so leads to polyploidy that secondarily triggers increased rates of apoptosis 20 .
The extent of liver mass is maintained primarily by adult cells in normal turnover (quiescent livers) and following mild injuries, regulated by effects of the Hippo/Yap signaling pathway 21  can have a role even in normal turnover in quiescent livers or those with mild injuries, since the number of divisions required is small and easily missed 25 . With increasing severity of the injuries, the contributions by stem/progenitor populations are presumed to increase to provide sufficient functional liver tissue for the host to survive 8,19,26 .