Periductal fibroblasts participate in liver homeostasis, fibrosis, and tumorigenesis

Hepatic fibroblasts represent a type of stromal cells in the liver that is distinct from the vitamin A–storing hepatic stellate cells (HSCs). They were originally identified as connective tissue cells surrounding the portal triad (bile duct, portal vein, and small artery) and observed to be characterized by long processes and usually surrounded by fibrils, including elastic fibers (Carruthers et al., 1962). Hepatic fibroblasts were first described as mesenchymal cells not related to sinusoids (Popper and Uenfriend, 1970; Schaffner et al., 1963). Later, fibroblast-like cells were also observed surrounding the central veins (second layer cells) and in the liver capsule beneath the mesothelium (submesothelial cells) (Bhunchet and Wake, 1992; Chapman and Eagles, 2007). Therefore, hepatic fibroblasts are a heterogeneous population that includes several anatomically different subpopulations. Although many markers, such as Col1a1, Col15a1, elastin, ectonucleoside triphosphate diphosphohydrolase 2, and Thy1, have been used to identify hepatic fibroblasts (Dranoff et al., 2002; Fausther et al., 2015; Iwaisako et al., 2014; Katsumata et al., 2017; Lemoinne et al., 2015; Li et al., 2007; Lua et al., 2016), their expression is not restricted to this population in the liver (Karin et al., 2016; Wells, 2014). Moreover, the expression patterns of these markers in different fibroblast subpopulations are also unclear. Thus, overall, our current knowledge about the cellular identity of different hepatic fibroblast subpopulations is limited.

Genetic lineage-tracing studies using HSC-specific Cre or CreER drivers have shown HSCs to be the major contributor to myofibroblast formation in liver fibrosis (Henderson et al., 2013; Mederacke et al., 2013; Wang et al., 2021). Although hepatic fibroblasts have been proposed as precursors of myofibroblasts during liver fibrosis, especially biliary fibrosis, their precise contribution remains controversial (Dranoff and Wells, 2010; Kisseleva, 2017; Wells, 2014). Recent studies have investigated the potential contributions of portal fibroblasts (identified according to the expression of a Col1-GFP reporter) to the myofibroblast pool during periductal fibrosis, as well as mapping their transcriptome (Iwaisako et al., 2014; Lua et al., 2016). A major challenge in understanding the contribution of hepatic fibroblasts to fibrosis lies in the similarity of the molecular signatures of fibroblasts, HSCs, and myofibroblasts. To date, Cre alleles able to distinguish the fate of hepatic fibroblasts from other non-parenchymal cells of the liver, especially HSCs, have not been available.

Single-cell RNA sequencing (scRNA-seq) has now been widely used to identify cell (sub)populations within tissues, including the liver (Dobie et al., 2019; Krenkel et al., 2019; Wang et al., 2021; Yang et al., 2021). scRNA-seq also facilitates the characterization of cellular identity, allowing fine discrimination between cell types that have similar characteristics (Dobie et al., 2019). In this study, we aimed to identify fibroblast-specific genes that would enable the fate mapping of these cells in vivo. From scRNA-seq data, we found that Cd34, Dpt, and Acta2 were expressed in hepatic fibroblasts, but not in the majority of other liver cell types, including HSCs. Furthermore, a combination of Cd34-CreER and Dpt-CreER discriminated between different subpopulations of hepatic fibroblasts. By quantifying and comparing the lineage outputs of Cd34-CreER⁺ and Dpt-CreER⁺ cells, we determined the respective contributions of periductal, periportal-venous, and pericentral-venous fibroblasts to ductal homeostasis, liver fibrosis, and tumorigenesis.

We aimed to identify markers that would allow the specific and inducible fate mapping of hepatic fibroblasts. To this end, we compared the single-cell transcriptomes of all liver cell types. We chose scRNA-seq datasets from Wang et al. (2021) that included most liver cell types except for vascular smooth muscle cells (VSMCs) and Dobie et al. (2019) that included HSCs, fibroblasts, and VSMCs (Fig. S1 A). These datasets were combined to ensure a complete representation of all liver cell types for our analyses, and all samples originated from normal liver conditions. t-Distributed stochastic neighbor embedding (t-SNE) analysis identified eight main cell clusters (Fig. 1 A). Based on the expression patterns of known cell type–specific marker genes (Fig. 1 B) and the top 100 differentially expressed genes (Fig. S1 B), we defined these cell types as HSCs (Lrat⁺), fibroblasts (Lrat^-Pdgfra⁺Col1a1^high), VSMCs (Myh11⁺), mesothelial cells (Msln⁺), endothelial cells (Pecam1⁺), immune cells (Ptprc⁺), hepatocytes (Gnmt⁺), and cholangiocytes (Sox9⁺).

Among known fibroblast markers, Entpd2, but not Thy1, showed robust and specific expression in fibroblasts (Fig. 1, C and D); Col1a1 was highly expressed by fibroblasts and lowly expressed by HSCs (Fig. 1 E). Therefore, Col1a1 expression is not sufficient to distinguish fibroblasts from HSCs. Cd34 was highly expressed by fibroblasts and lowly expressed by mesothelial cells, but not by HSCs (Fig. 1 F). Dpt was specifically expressed by fibroblasts (Fig. 1 G). Acta2 was highly expressed by VSMCs and moderately expressed by fibroblasts (Fig. 1 H). The mesothelial cell cluster expressed Gpm6a, Wt1, and Msln (Fig. 1, I–K), consistent with previous reports (Balog et al., 2020; Li et al., 2013; Lua et al., 2016). We chose Cd34, Dpt, and Acta2 for the generation of fibroblast-specific lineage-tracing mice.

We performed immunostaining to examine the protein expression patterns of CD34 in the liver of 2-mo-old wild-type mice. Confocal imaging was performed on 15-μm-thick liver sections co-stained with anti-CD31, anti-Epcam, and anti-CD34 antibodies (Fig. 2 A). Anti-CD31 staining depicted the whole vasculature of the liver, in which the portal and central veins were CD31⁺ and relatively large, whereas sinusoids were CD31⁺ and relatively small (Fig. 2 A). Anti-Epcam antibody stained the bile ducts and thereby distinguished portal from central zones (Fig. 2 A). Anti-CD34 staining was robustly detected around the CD31⁺ portal veins and Epcam⁺ bile ducts, but was not detected in cells surrounding CD31⁺ sinusoids or CD31⁺ central veins (Fig. 2 A). These data suggested that CD34⁺ fibroblasts are mostly restricted to the periportal-venous and periductal regions.

In order to fate-map CD34⁺ fibroblasts, we generated Cd34^creER knock-in mice, in which a creER^T2 coding sequence was inserted into the Cd34 genomic loci in front of the initiation codon ATG (Fig. S2 A). These mice were crossed with Rosa26^{CAG-loxp-STOP-loxp-tdTomato} (R26^tdTomato) mice (Madisen et al., 2010) to constitutively label Cre-expressing cells and their descendants with Tomato fluorescent protein. Cre expression was induced at 6 wk of age, and the mice were analyzed at 8 wk of age. Co-staining of liver sections with anti-CD31 and anti-Epcam antibodies detected robust Tomato expression around CD31⁺ portal veins and Epcam⁺CD31⁻ bile ducts, but not around CD31⁺ central veins or CD31⁺ sinusoids (Fig. 2 B). Confocal imaging of anti-CD34 antibody–stained liver sections confirmed the specific and efficient (∼67%) recombination of Cd34-CreER in CD34⁺ cells in the liver (Fig. 2 C and Fig. S2 C). Most of the Col1⁺ and Thy-1⁺ fibroblasts were labeled by Tomato around the portal triad (Fig. 2, D and E). In contrast, few VSMCs surrounding the arteries and portal veins expressed Tomato in the liver (Fig. 2 F, Fig. S1 C, and Fig. S2 D). On-slide staining of isolated non-parenchymal liver cells from Cd34^creER;R26^tdTomato mice detected Tomato expression in a subset of Col1⁺ cells, but not Reln⁺ or Myh11⁺ cells (Fig. S2, E–G). Whole-lobe tissue clearing and 3D reconstruction of the Tomato⁺ cells from Cd34^creER;R26^tdTomato liver outlined the portal venous/bile ductal network in the liver (Fig. S2 B). Thus, Cd34-CreER labeled periductal- and periportal-venous fibroblasts, but not pericentral-venous fibroblasts.

Besides the Tomato⁺ fibroblasts around the portal triad, a few sparsely distributed Tomato⁺ cells were also observed in the liver from Cd34^creER;R26^tdTomato mice (Fig. 2 B). Confocal imaging showed that these sparsely distributed Tomato⁺ cells were not Reln⁺ HSCs (Fig. 2 G and Fig. S1 D), Gpm6a⁺ mesothelial cells (Fig. 2 H), Hnf4a⁺ hepatocytes (Fig. S2 H), or F4/80⁺ Kupffer cells (Fig. 2 I). Most of these cells expressed the pan-immune cell marker CD45 (Fig. 2 J). Both intravascular and extravascular Tomato⁺ cells were frequently detected on liver sections (Fig. 2 L). By flow cytometry, Tomato⁺ cells represented ∼21% of all CD45/Ter119⁺ immune cells in the liver (Fig. 2 K). Consistent with this, ∼25% of all hematopoietic stem cells in the bone marrow were Tomato⁺ (Fig. S2 I). Flow cytometric analysis of enzymatically dissociated liver cells from Cd34^creER;R26^tdTomato mice showed that Tomato expression was not detected in VE-cadherin⁺ endothelial cells (Fig. 2 M).

We then investigated the potential contribution of Cd34⁺ cells to liver fibrosis. Pericentral fibrosis was induced in 2-mo-old tamoxifen-treated Cd34^creER;R26^tdTomato mice by administration of CCl₄ twice per week for 4 wk. Successful induction of liver fibrosis was evidenced by Sirius red staining on paraffin sections (Fig. S3, A and B) and anti-Col1 staining on frozen sections (Fig. 3, A and B). Confocal analysis showed that there was no obvious expansion of Tomato⁺ cells around the central veins of CCl₄-treated Cd34^creER;R26^tdTomato mice, compared with untreated controls (Fig. 3, A and B). Furthermore, few of the Col1⁺ myofibroblasts formed after CCl₄ treatment expressed Tomato (Fig. 3, B and C), suggesting that Cd34-CreER⁺ cells are not a source of myofibroblasts during CCl₄-induced pericentral fibrosis.

Biliary fibrosis was induced by 3,5-diethoxycarbonyl-1,4-dihydrocollidine (DDC) treatment. Successful induction of liver fibrosis was evidenced by Sirius red staining 14 days after treatment (Fig. S3 C). On liver sections from Cd34^creER;R26^tdTomato mice, we observed a marked increase of Tomato⁺ cells around the portal triad (Fig. 3 D). By quantification, 25 ± 8.2% of all Col1⁺ myofibroblasts formed 14 days after the treatment expressed Tomato (Fig. 3, D and E). These data suggested that Cd34⁺ cells are a source of myofibroblasts in periportal fibrosis.

To determine the contribution of Cd34⁺ cells to the CAF pool in liver tumors, we used an experimental mouse model of liver tumor formation. After a single injection of notch intracellular domain (NICD)/AKT plasmids at 8 wk of age, 100% of mice developed liver tumors by 5 wk (Fig. S3 D). Characteristics of ICC were observed in photographic comparisons of normal and tumorous livers (Fig. S3 D), H&E-stained sections (Fig. S3, E and F), and liver sections immunostained with anti-Sox9 and anti-Hnf4a antibodies, respectively (Fig. S3, G and H). By anti-Col1 staining, we detected CAFs throughout the tumoral tissue in Cd34^creER;R26^tdTomato mice (Fig. 3 F). Analysis of ICC livers from Cd34^creER;R26^tdTomato mice revealed that ∼40% of Col1⁺ CAFs were Tomato⁺ (Fig. 3, F and G), suggesting that Cd34⁺ cells serve as a source of CAFs in ICC.

In order to fate-map Dpt⁺ fibroblasts, we generated Dpt^creER knock-in mice, in which a 2A ribosome “skipping” sequence fused with creER^T2 was inserted before the 3′UTR of the endogenous Dpt gene (Fig. S4 A). Dpt^creER mice were crossed with R26^tdTomato mice to generate Dpt^creER;R26^tdTomato compound mutant mice. 2 wk after tamoxifen treatment, confocal imaging revealed robust Tomato expression around CD31⁺ portal and central veins, much less around Epcam⁺CD31⁻ bile ducts, but not around CD31⁺ sinusoids (Fig. 4 A). On liver sections, most of the Tomato⁺ cells expressed the fibroblast markers Col1, Thy-1, and CD34, accounting for ∼12% of all CD34⁺ fibroblasts (Fig. 4, B–D and Fig. S4 B), but not the SMC marker Myh11 (Fig. 4 E), the HSC marker Reln (Fig. 4 F), the mesothelial cell marker Gpm6a (Fig. 4 G), the hepatocyte marker Hnf4a (Fig. S4 F), or the Kupffer cell marker F4/80 (Fig. S4 G). On-slide staining of isolated non-parenchymal liver cells from Dpt^creER;R26^tdTomato mice detected Tomato expression in a subset of Col1⁺ cells, but not Reln⁺ or Myh11⁺ cells (Fig. S4, C–E). Thus, Dpt-CreER specifically labels periportal- and pericentral-venous fibroblasts.

We investigated the potential contribution of Dpt-CreER–labeled cells to liver fibrosis. Pericentral fibrosis was induced in 2-mo-old tamoxifen-treated Dpt^creER;R26^tdTomato mice by administration of CCl₄, as described above. As a result, we did not observe a significant expansion of Tomato⁺ cells in Dpt^creER;R26^tdTomato mice after CCl₄ treatment (Fig. 4 H). Furthermore, anti-Col1 immunostaining showed that only 1.8 ± 1.6% of all Col1⁺ myofibroblasts were also Tomato⁺ in the fibrotic livers (Fig. 4, H and I), indicating that Dpt-CreER–labeled cells do not contribute to pericentral fibrosis.

In the DDC-induced biliary liver periportal model, we observed expansion of Tomato⁺ cells in the livers of Dpt^creER;R26^tdTomato mice (Fig. 4 J). Quantification of cell numbers showed that 15 ± 3.4% of all Col1⁺DAPI⁺ myofibroblasts were Tomato⁺ in DDC-induced fibrosis (Fig. 4 K). Thus, Dpt-CreER–labeled cells are a source of myofibroblasts in periportal fibrosis.

We determined the contribution of Dpt-CreER–labeled cells to CAF formation in ICC. ICC was induced as described above. Analysis of liver sections revealed that only 2.5 ± 1.7% of all Col1⁺ CAFs were Tomato⁺ in ICCs of Dpt^creER;R26^tdTomato mice (Fig. 4, L and M), significantly lower than that in ICCs of Cd34^creER;R26^tdTomato mice (Fig. S4 H), suggesting that Dpt-CreER–labeled cells do not contribute to CAF formation in ICC.

Co-staining with anti-CD31, anti-Epcam, and anti-aSma antibodies revealed the robust expression of aSma protein around the CD31⁺ portal and central veins, but rarely around CD31⁺ sinusoids or Epcam⁺CD31⁻ bile ducts (Fig. S4 I). To fate-map aSma⁺ cells, we used Acta2-creER bacterial artificial chromosome transgenic mice, in which the creER^T2 coding sequence was inserted into a 175-kb mouse genomic segment encompassing the Acta2 gene (Wendling et al., 2009). Confocal imaging of anti-aSma antibody–stained liver sections confirmed specific and efficient recombination of Acta2-CreER in aSma⁺ cells from the liver of Acta2-creER;R26^tdTomato mice (Fig. 5 B). Co-staining of liver sections with anti-CD31 and anti-Epcam antibodies showed robust Tomato expression around CD31⁺ portal and central veins, but not around CD31⁺ sinusoids or Epcam⁺CD31⁻ bile ducts (Fig. 5 A). Most of the Tomato⁺ cells in the liver of Acta2-creER;R26^tdTomato mice were positive for Myh11 expression (Fig. 5 C). In contrast, Tomato⁺ liver cells poorly colocalized with anti-CD34 or anti-Col1 staining (Fig. 5, D and E). The Tomato⁺ cells and Col1⁺ fibroblasts appeared to localize at different layers of the portal veins and arteries (Fig. S4, J and K). Tomato expression was not detected in Reln⁺ HSCs, F4/80⁺ Kupffer cells, or Hnf4a⁺ hepatocytes (Fig. 5, F and G; and Fig. S4 L). These data suggested that Acta2-CreER labels VSMCs rather than fibroblasts. Considering the high expression of Acta2 mRNA in VSMCs and moderate expression in fibroblasts (Fig. 1 H), the moderate expression of Acta2 in fibroblasts appeared insufficient to activate Cre expression.

To further investigate the expression patterns of Cd34-CreER, Dpt-CreER, and Acta2-CreER in VSMCs, fibroblasts, and HSCs, we sorted Tomato⁺CD45⁻ cells from the livers of Tcf21^creER;R26^tdTomato mice, Cd34^creER;R26^tdTomato mice, Dpt^creER;R26^tdTomato mice, and Acta2-creER;R26^tdTomato mice, respectively, and examined their expression of Col1a1, Cd34, Myh11, and Reln. We have shown that Tcf21-CreER labels HSCs (Wang et al., 2021). As a result, Tomato⁺ cells from Cd34^creER;R26^tdTomato and Dpt^creER;R26^tdTomato mice expressed robust levels of Col1a1 and Cd34, but not Myh11 or Reln (Fig. S5). In contrast, Tomato⁺ cells from Acta2-creER;R26^tdTomato mice expressed robust levels of Myh11, but not Col1a1, Cd34, or Reln (Fig. S5). These data confirmed the specific expression of Cd34-CreER and Dpt-CreER in fibroblasts and Acta2-CreER in VSMCs.

We investigated the potential contribution of Acta2-CreER–labeled cells to myofibroblast formation in liver fibrosis and CAF formation in ICC. The results showed that Acta2-CreER–labeled cells generated 1.1 ± 0.84% and 2.1 ± 2.1% of all Col1⁺ myofibroblasts formed in CCl₄- and DDC-induced fibrosis, respectively (Fig. 5, H–K). In ICC, they generated 1.7 ± 1.6% of all Col1⁺ CAFs (Fig. 5, L and M). Thus, Acta2-CreER–labeled cells contribute minimally to myofibroblast formation in liver fibrosis and CAF formation in ICC.

To determine the physiological importance of the fibroblast in the liver, we crossed Cd34^creER mice with R26^{CAG-loxp-STOP-lox-DTA} (R26^DTA) mice to generate a strain in which Cre-expressing cells and their descendants would be constitutively depleted by the expression of diphtheria toxin (Voehringer et al., 2008). Cre expression was induced in Cd34^creER;R26^DTA compound mutants at 6 wk of age, as above, and the mice were analyzed at 8 wk. Confocal imaging confirmed the efficient depletion of Tomato⁺ cells from the liver of Cd34^creER;R26^tdTomato/DTA mice upon tamoxifen treatment (Fig. 6, A and B). H&E staining on paraffin sections did not reveal obvious changes to liver morphology in Cd34^creER;R26^DTA mice, compared with R26^DTA mice (Fig. S5 E). However, staining of the ductal epithelium by anti-Sox9 antibody revealed that the bile ducts were enlarged in diameter and the number of cholangiocytes lining the bile ducts was significantly increased. In normal liver, intrahepatic bile ducts were lined by four to eight cholangiocytes (except for the large intrahepatic bile ducts lining the primary portal vein), whereas in livers from Cd34^creER;R26^DTA mice, bile ducts were lined by 6 to 10 cholangiocytes (Fig. 6, C and D). These findings indicated that depletion of Cd34⁺ cells resulted in enlargement of the bile ducts.

We administered 5-ethynyl-2′-deoxyuridine (EdU) to Cd34^creER;R26^DTA and R26^DTA control mice immediately after tamoxifen treatment to directly assess cholangiocyte proliferation following depletion of Cd34⁺ cells. After 4 wk, we found that ∼38% of all Epcam⁺ cholangiocytes were EdU-positive in Cd34^creER;R26^DTA mice, which was significantly higher compared with about 16% in R26^DTA control mice (Fig. 6, E and F). This indicates a marked increase in cholangiocyte proliferation following the depletion of Cd34⁺ cells. In contrast, there was no significant difference in the overall percentage of EdU⁺ cells across the liver tissue between the two groups (Fig. 6, E and F), suggesting that the observed effects were specific to cholangiocytes.

Next, we subjected tamoxifen-treated Cd34^creER;R26^DTA mice and littermate R26^DTA mice to DDC-induced liver fibrosis. Livers from all mice exhibited clear liver fibrotic phenotypes, as revealed by Sirius red staining (Fig. 6 G). We did not detect a significant difference between Cd34^creER;R26^DTA and R26^DTA mice with respect to the percentage of liver area stained by Sirius red (Fig. 6 H). Consistent with this, the mRNA levels of Col1 and Acta2 in fibrotic livers from Cd34^creER;R26^DTA and R26^DTA mice did not significantly differ from each other (Fig. 6, I and J). These data suggested that conditional depletion of Cd34⁺ cells does not affect biliary fibrosis.

We conditionally deleted Tgfbr2 from Cd34⁺ cells to determine the requirement of Tgfβ signaling in myofibroblast formation by portal fibroblasts. We performed DDC treatment on both control and mutant mice. After 2 wk, we detected significantly less Tomato⁺Col1⁺ myofibroblasts in Cd34^creER;R26^tdTomato;Tgfbr2^fl/fl mice than in control mice, despite that the overall Col1⁺ myofibroblasts did not differ between control and mutant mice (Fig. 6, K and L). Thus, the formation of myofibroblasts by Cd34⁺ fibroblasts requires Tgfβ signaling.

We also subjected tamoxifen-treated Cd34^creER;R26^DTA mice and R26^DTA littermates to ICC induction. At 3 wk after plasmid injection, the percentage of tumor area was significantly greater in Cd34^creER;R26^DTA liver than in R26^DTA liver (Fig. 6, M and N). In contrast, at 4 wk after plasmid injection, there was no significant difference between Cd34^creER;R26^DTA and R26^DTA mice with respect to the percentage of tumor area within the liver (Fig. 6, O and P). Consistent with this, the mRNA levels of Col1 and Acta2 in ICC livers from Cd34^creER;R26^DTA mice were significantly higher than those from R26^DTA mice at 3 wk, but not at 4 wk, after plasmid injection (Fig. 6, Q and R). These data suggested that conditional depletion of Cd34⁺ cells accelerated the progression of ICC.

We transplanted bone marrow cells from Cd34^creER;R26^tdTomato mice into lethally irradiated wild-type mice. 8 wk after transplantation, we detected many Tomato⁺ immune cells, but no Tomato⁺ fibroblasts in the liver of the recipients (Fig. 7, A and B). After DDC treatment for another 2 wk, we did not detect any Tomato⁺ myofibroblasts in the fibrotic liver (Fig. 7 C). At 4 wk after ICC induction, we did not detect any Tomato⁺ CAFs in the tumor (Fig. 7 D). These data excluded the possibility that Cd34-CreER–labeled immune cells contribute to liver fibrosis or ICC.

We also transplanted bone marrow cells from wild-type mice into lethally irradiated Cd34^creER;R26^tdTomato, Cd34^creER;R26^DTA, and R26^DTA control mice. 8 wk after transplant, we induced Tomato or DTA expression using tamoxifen. Confocal imaging of liver sections from Cd34^creER;R26^tdTomato recipients showed that CD45⁺ immune cells no longer expressed Tomato, confirming that hematopoietic cells were effectively replaced and did not contribute to the observed phenotypes (Fig. 7, E and F). Further analysis of bile duct morphology through anti-Sox9 staining revealed that bile ducts were significantly enlarged, and the number of cholangiocytes lining the ducts was increased in Cd34^creER;R26^DTA recipients compared with R26^DTA controls (Fig. 7, G and H), replicating phenotypes observed in non-transplanted settings. When inducing ICCs in Cd34^creER;R26^DTA and R26^DTA recipients, the tumor area was significantly greater in Cd34^creER;R26^DTA recipients than in R26^DTA controls at 3 wk after induction (Fig. 7, I and J). These findings demonstrated that the effects seen are attributable to the depletion of Cd34⁺ fibroblasts, not hematopoietic cells.

In this study, we used Cd34-CreER, Dpt-CreER, and Acta2-CreER mice to trace fibroblasts and VSMCs in the liver under various conditions. An ontological combination of Cd34-CreER and Dpt-CreER distinguished the fates of periductal, periportal-, and pericentral-venous fibroblasts. Hepatic fibroblasts are a heterogeneous population and have been implicated in liver homeostasis and diseases (Dranoff and Wells, 2010; Lemoinne et al., 2015; Wells, 2014). Most previous studies have focused on the fibroblasts surrounding the portal triad, although fibroblasts are also present around the central veins and in the liver capsule beneath the mesothelium (Bhunchet and Wake, 1992; Chapman and Eagles, 2007). scRNA-seq studies have identified several groups of fibroblast subpopulations derived from the biliary–vascular tree, among which the Pdgfra⁺Cd34⁺CD9⁺CD200^low subpopulation conferred mesenchymal stem cell features in vitro (Lei et al., 2022). Our work has revealed that Cd34⁺Dpt^- periductal fibroblasts contributed to ductal homeostasis, biliary fibrosis, and tumorigenesis. Cd34⁺Dpt⁺ periportal-venous fibroblasts contribute to biliary fibrosis. Cd34⁻Dpt⁺ pericentral-venous fibroblasts do not contribute to pericentral fibrosis.

HSCs are a major source of myofibroblasts in multiple models of biliary liver fibrosis (Mederacke et al., 2013; Wang et al., 2021) and are also likely to represent the main source of stromal cell–derived CAFs in liver tumors (Brivio et al., 2017; Wang et al., 2021). Our fate-mapping results showed that fibroblasts marked by Cd34-CreER contributed to ∼25% of all myofibroblasts in biliary fibrosis (Fig. 3, D and E). However, it was notable that conditional depletion of these cells did not impair liver fibrosis (Fig. 6, G–J). We speculated that HSCs compensate for the loss of Cd34⁺ cells by producing additional myofibroblasts. Mesothelial cells have been shown to be involved in capsular fibrosis (Balog et al., 2020), suggesting a division-of-labor among different stromal cells in myofibroblast formation in different types of liver fibrosis. We also showed that conditional depletion of Cd34⁺ fibroblasts accelerated the progression of ICC (Fig. 6, M–R). One possible mechanism is that Cd34⁺ periductal fibroblasts restrict the proliferation of cholangiocytes (Fig. 6, C and D), the cell type from which ICCs arise. This finding is in line with the negative role of fibroblasts in ductal epithelial cell proliferation in a mesenchymal-ductal organoid co-culture system (Cordero-Espinoza et al., 2021). Another mechanism could be a direct regulation of tumor cell proliferation by Cd34⁺ fibroblasts in ICC, potentially through the secretion of antitumorigenic factors.

Increasing evidence has demonstrated the heterogeneity of the CAF compartment in liver tumors. Different CAF subpopulations may play distinct or even opposing roles in the formation and/or progression of liver tumors (Affo et al., 2021; Filliol et al., 2022). We have shown that Cd34⁺ fibroblasts form a significant proportion (∼40%) of all CAFs in a mouse model of ICC (Fig. 3, F and G). While our current study design did not allow for a detailed investigation into whether CD34⁺ fibroblasts contribute to specific subtypes of CAFs, this remains an intriguing possibility for future research.

All mice strains, including R26^tdTomato (Madisen et al., 2010), Acta2-creER (Wendling et al., 2009), Cd34^creER, Dpt^creER, R26^DTA (Voehringer et al., 2008), and Tgfbr2^fl mice (Chytil et al., 2002), were housed in the animal facility at Shanghai Institute of Biochemistry and Cell Biology (SIBCB). Construction of Cd34^creER and Dpt^creER mice was performed by Shanghai Biomodel Organism Co., Ltd. Cd34^creER was generated by knocking creER into the ATG site of the endogenous Cd34 gene. Dpt^creER was generated by knocking 2A-creER before the 3′UTR of the endogenous Dpt gene. To induce CreER activity, intraperitoneal injections of 2 mg tamoxifen (Sigma-Aldrich) dissolved in corn oil were administered daily for three consecutive days. All procedures were approved by the Institutional Animal Care and Use Committee of SIBCB.

To induce pericentral liver fibrosis, CCl₄ (Sigma-Aldrich) dissolved in corn oil (1:4) was injected intraperitoneally at a dose of 1 μl per gram body weight biweekly for 4 wk. To induce biliary fibrosis, 8–12-wk-old mice were fed on a diet supplemented with 0.1% DDC (D17142193; Future Biotech) for 2–4 wk.

ICC was induced according to the literature (Fan et al., 2012). Briefly, 20 μg pT3-EF1a-NICD1, 4 μg pT3-EF1a-HA-myr-Akt, and 1 μg transposase plasmids were diluted in 2 ml 0.9% NaCl, sterile-filtered, and administered to 6–8-wk-old mice by hydrodynamic tail vein injection.

Mouse livers were fixed in 4% paraformaldehyde (PFA) overnight, dehydrated through an ethanol series (70%, 80%, 95%, 100%), embedded in paraffin, cut into 5-µm sections using the Leica manual rotary microtome (Leica Biosystems), and placed on adhesion microscope slides (Xinyu Biotech). After dewaxing and hydration, liver sections were stained with Sirius red or hematoxylin and eosin (H&E) and mounted on slides for imaging by a BX51 (Olympus) microscope.

Liver samples were fixed overnight in 4% PFA at 4°C, dehydrated in 30% sucrose for 24 h, embedded in OCT (NEG-50-6502; Thermo Fisher Scientific), and then cut into 15-μm sections using a cryostat microtome (Leica Biosystems). Sections were incubated with primary antibodies overnight at 4°C and then washed three times in PBS. After 2 h of incubation with secondary antibodies at room temperature, sections were washed in PBS and mounted on slides using ProLong Gold Antifade Reagent (Invitrogen). Images were captured by a Leica TCS SP8 WLL or a Leica TCS SP8 STED confocal microscope. Antibodies were diluted in PBS containing 10% donkey serum and 0.1% Triton X-100. The following primary antibodies were used: anti-aSma (1:250, ab5694; Abcam), anti-CD31 (1:250, AF3628; R&D), anti-F4/80 (1:250, MCA497rt; Bio-Rad), anti-CD34 (1:100, ab81289; Abcam), anti-Reln (1:100, AF3820; R&D), anti-Col1 (1:200, A1352; ABclonal), anti-Myh11 (1:100, HPA015310; Atlas Antibodies), and anti-Hnf4a (1:500, ab181604; Abcam). The following secondary antibodies were used: donkey anti-rabbit Alexa Fluor 647 (1:500, A31573; Invitrogen), donkey anti-goat Alexa Fluor 488 (1:500, A21432; Invitrogen), donkey anti-rat Alexa Fluor 488 (1:500, A21208; Invitrogen), donkey anti-rabbit Alexa Fluor 488 (1:500, A21206; Invitrogen), donkey anti-goat Alexa Fluor 647 (1:500, A21447; Invitrogen), and donkey anti-goat BV421 (1:200, 705-675-147; Jackson ImmunoResearch).

To prepare liver cells for flow cytometry, mouse liver was digested according to a previously published protocol (Mederacke et al., 2015). Briefly, after cannulation of the inferior vena cava, the portal vein was cut to allow retrograde stepwise perfusion of Liver Perfusion Medium (17701038; Gibco) for 5 min, followed by pronase (0.4 mg/ml, 5 min, 10165921001; Sigma-Aldrich)- and collagenase type IV (1 mg/ml, 15 min, 17104019; Gibco)– containing HBSS. Livers were then excised and minced, before ex vivo digestion in HBSS containing 0.5 mg/ml pronase, 1 mg/ml collagenase types I and IV, and 1 mg/ml DNase I (11284932001; Sigma-Aldrich). Cells were stained with CD144 (VE-cadherin)-BV421 (1:100, 747749; BD), CD45-APC (1:200, 103112; BioLegend), and Ter119-APC (1:200, 116212; BioLegend) antibodies for 30 min. To prepare bone marrow cells, bone cavities were flushed using a syringe containing 1 ml buffer (HBSS with 2% FBS). Cells were stained with lineage mix on ice for 30 min: Ter119-APC780 (1:200, 47-5921-82; eBioscience), CD3-APC780 (1:200, 47-0031-82; eBioscience), CD5-APC780 (1:200, 47-0051-82; eBioscience), CD8a-APC780 (1:200, 47-0081-82; eBioscience), B220-APC780 (1:200, 47-0452-82; eBioscience), Gr-1-APC780 (1:400, 47-5931-82; eBioscience), Sca1(Ly6a)-PerCPcy5.5 (1:100, 122524; BioLegend), CD117(c-Kit)-BV421 (1:200, 105828; BioLegend), CD48-APC (1:200, 103412; BioLegend), and CD150-PEcy7 (1:100, 115914; BioLegend). Flow cytometry was performed on a BD LSRFortessa flow cytometry system, and data were analyzed using FlowJo software.

After enzymatic digestion of the liver from Cd34^creER;R26^tdTomato mice, non-parenchymal cells were isolated from the interface between the two density cushions of 25% and 50% Percoll (P4937-500 Ml; Sigma-Aldrich) by gradient centrifugation (1,500 g for 20 min). Isolated cells were fixed with 4% PFA for 20 min, permeabilized with 0.5% Triton X-100 for another 20 min, and blocked with 5% BSA for an hour. They were then mounted on poly-D-lysine–coated slides and incubated with antibodies as indicated.

Cd34^creER;R26^DTA and R26^DTA control mice were fed with EdU (0.5 mg/ml in the water supply) for 4 wk right after tamoxifen induction. To detect EdU incorporation, liver sections wer; Carbosynthe incubated with EdU labeling cocktail containing 2.5 μM sulfo-Cyanine5 azide (Abcam) for 20 min. Confocal microscopy was performed as described above.

Liver lobes were fixed in 4% PFA at room temperature for 24 h and then cleared using the PEGASOS method (Jing et al., 2018). Briefly, the liver lobes were treated with 25% Quadrol (122262; Sigma-Aldrich) for 2 days, 5% ammonium for 1 day, a tert-Butanol (360538; Sigma-Aldrich) dilution series of 30%, 50%, and 70% solutions for 4, 6 h, and 1 day, respectively, tert-Butanol/PEG (409529; Sigma-Aldrich) for 2 days, and benzyl benzoate (B6630; Sigma-Aldrich)/PEG for 1 day. 3D imaging data were acquired using a light-sheet microscope Luxendo MuVi SPIM and analyzed using Imaris software.

Mice of the indicated genotype were irradiated using an RS 2000 x-ray irradiator with two doses of 540 rad (total 1,080 rad) delivered at least 2 h apart. One million whole bone marrow cells from donor mice were then injected intravenously into the retro-orbital venous sinus of irradiated recipient mice.

Cells were sorted directly into TRIzol (Life Technologies). RNA was reverse-transcribed using SuperScript III Reverse Transcriptase (Life Technologies). Quantitative real-time PCR was performed using SYBR Green on a LightCycler 96 system (Roche). Primer sequences were as follows: β-actin: 5′CGT​CGA​CAA​CGG​CTC​CGG​CAT​G-3′ and 5′-GGG​CCT​CGT​CAC​CCA​CAT​AGG​AG-3′; Col1a1: 5′-CAC​CCT​CAA​GAG​CCT​GAG​TC-3′ and 5′- GTT​CGG​GCT​GAT​GTA​CCA​GT-3′; Cd34: 5′-TGG​TAG​GAA​CTG​ATG​GGG​AT-3′ and 5′-TGG​GTA​GCT​CTC​TGC​CTG​AT-3′; Reln: 5′-GAA​ACC​GAG​AAG​CAA​AGC​TG-3′ and 5′-CAG​GTG​ATG​CCA​TTG​TTG​AC-3′; Acta2: 5′-CGA​AAC​CAC​CTA​TAA​CAG​CAT​CA-3′ and 5′-GCG​TTC​TGG​AGG​GGC​AAT-3′; Myh11: 5′-ATC​GCC​CAG​AAA​AAC​AAT​GCC​CTA​AA-3′ and 5′-GCG​TAT​CCT​CCA​GCT​CCG​TCT​TGA-3′.

scRNA-seq datasets from Wang et al. (2021) and Dobie et al. (2019) were integrated, and the canonical correlation analysis method implemented in Seurat (version 4.0.1) was used to remove batch effects among different datasets. The top principal components were selected and employed for subsequent dimension reduction and unsupervised clustering. For dimension reduction, we performed t-SNE to visualize the datasets. We used the FindAllMarkers function to detect differentially expressed genes in different clusters.

All data represent multiple independent experiments performed using different mice. Both male and female mice were included. Sample sizes were not based on power calculations. No randomization or blinding was performed. Variation is always indicated using standard deviation (SD). The statistical significance of differences was assessed by two-tailed Student’s t tests.

Fig. S1, related to Fig. 1, shows the scRNA-seq analysis of non-parenchymal liver cells. Fig. S2, related to Fig. 2, shows the expression pattern of Cd34-CreER in the liver. Fig. S3, related to Fig. 3, demonstrates successful induction of liver fibrosis and ICC. Fig. S4, related to Fig. 4 and Fig. 5, shows the expression patterns of Dpt-CreER and Acta2-CreER in the liver. Fig. S5 shows the quantitative real-time PCR analysis of purified Tomato⁺ cells from multiple lineage-tracing mice.

Data supporting the findings of this study are available from the corresponding authors upon reasonable request.

This work was supported by the National Key Program on Stem Cell and Translational Research (2022YFA1104100), the Strategic Priority Research Program of Chinese Academy of Science (XDB0990000), the National Natural Science Foundation of China (32425020, 82273292, 92368101, and 32270785), Shanghai Municipal Science and Technology Major Project Fund, Shanghai Science and Technology Commission (22XD1424000 and 22ZR1469000), and Haihe Laboratory of Cell Ecosystem Innovation Fund (22HHXBSS00016).

Author contributions: S.-S. Wang: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, supervision, validation, visualization, and writing—original draft, review, and editing. J. Yuan: data curation, formal analysis, investigation, validation, and writing—original draft. X.T. Thang: data curation and formal analysis. X. Yin: formal analysis and validation. K. Fang: data curation and investigation. L.V. Chen: formal analysis and investigation. Z. Ren: conceptualization, project administration, supervision, and writing—review and editing. B.O. Zhou: conceptualization, funding acquisition, supervision, and writing—original draft, review, and editing.