Bipotent transitional liver progenitor cells contribute to liver regeneration

Following severe liver injury, when hepatocyte-mediated regeneration is impaired, biliary epithelial cells (BECs) can transdifferentiate into functional hepatocytes. However, the subset of BECs with such facultative tissue stem cell potential, as well as the mechanisms enabling transdifferentiation, remains elusive. Here we identify a transitional liver progenitor cell (TLPC), which originates from BECs and differentiates into hepatocytes during regeneration from severe liver injury. By applying a dual genetic lineage tracing approach, we specifically labeled TLPCs and found that they are bipotent, as they either differentiate into hepatocytes or re-adopt BEC fate. Mechanistically, Notch and Wnt/β-catenin signaling orchestrate BEC-to-TLPC and TLPC-to-hepatocyte conversions, respectively. Together, our study provides functional and mechanistic insights into transdifferentiation-assisted liver regeneration.

The liver performs critical life-enabling metabolic, endocrine and secretory functions via its two epithelial cell compartments. Hepatocytes metabolize nutrients and xenobiotics, produce and recycle proteins, and generate bile acids. BECs (also termed cholangiocytes) constitute the bile duct network responsible for collecting and transporting bile into the gut, thereby supporting metabolite excretion and digestion 1,2 . Maintaining a functional hepatocyte pool is essential to guarantee liver function during homeostatic cell turnover or in response to injury [3][4][5][6] .
Previous genetic lineage tracing studies demonstrated that the hepatocyte pool is mainly replenished through self-renewal of pre-existing hepatocytes rather than differentiation from liver stem/ progenitor cells during homeostasis and injury conditions leaving hepatocyte proliferation intact [7][8][9][10][11] . BECs are also able to proliferate and generate auxiliary biliary ducts in a regenerative process called ductular reaction 12 . However, when hepatocytes become senescent and hepatocyte-mediated liver regeneration is impaired in mice, BECs can serve as facultative liver progenitor cells (LPCs) and transdifferentiate into functional replication-competent hepatocytes [13][14][15][16][17][18] . In zebrafish, hepatic BECs or LPCs convert into hepatocytes after severe loss of hepatocytes 19,20 , in a process that is tightly modulated by genetic and epigenetic regulators to enable efficient liver regeneration [21][22][23] . Given the widespread hepatocyte senescProtein Data Banence and impaired liver regeneration in patients with chronic liver disease and cirrhosis 24,25 , BEC-to-hepatocyte transdifferentiation could be an important repair mechanism in humans. Therapies promoting this transdifferentiation could open up a new treatment avenue to address this highly Article https://doi.org/10.1038/s41588-023-01335-9 Of note, all tdT + hepatocytes were positive for FAH in TAM-treated mouse livers (Extended Data Fig. 2g), which substantially reduced the severity of liver injury (Extended Data Fig. 2h). These tdT + hepatocytes did not express p21 and showed increased proliferation when compared with tdThepatocytes (Extended Data Fig. 2i,j). Large clones of tdT + hepatocytes re-established metabolic zonation by expressing periportal and pericentral zonation markers in the respective lobular zones (Extended Data Fig. 2k). Of note, tdT + hepatocytes only occurred in mice treated with TAM, and these BEC-derived hepatocytes substantially increased the long-term survival of mice in our injury model (Extended Data Fig. 3a,c-f). While this clearly supports that BEC-to-hepatocyte transdifferentiation promotes liver regeneration, short-term (5 d) NTBC reintroduction was necessary to ensure survival and enable this regenerative process to compensate for the fulminant liver failure in our model (Extended Data Fig. 3a,b). We did not detect any tdT + hepatocytes in TAM-treated CK19-CreER;Fah-LSL/LSL;R26-tdT mice when NTBC was given throughout the experiment (Extended Data Fig. 2l,m), excluding potential ectopic activation of CK19-CreER in hepatocytes, and also suggesting that loss of FAH function and associated liver injury is necessary to induce BEC-to-hepatocyte transdifferentiation.
Furthermore, we crossed CK19-CreER;Fah-LSL/LSL with multicolor fluorescence reporter (R26-Confetti mice 31 ) for clonal analysis of single BEC-derived cells during regeneration (Extended Data Fig. 4a). Due to sparse labeling of BECs with one of the four reporters in CK19-CreER;Fah-LSL/LSL;R26-Confetti mice, a single-color clone detected at the end of the experiment would be regarded as progeny from a single BEC (Extended Data Fig. 4a). While TAM treatment selectively labeled single BECs in livers at 0 weeks, we detected single-color hepatocyte clones, which were located near the portal veins but not central veins at 10 weeks (Extended Data Fig. 4b-e). Of note, a subset of single-color clones contained both BECs and hepatocytes, suggesting single BECs could give rise to both cell lineages over time (Extended Data Fig. 4d,f). Together, our Fah-LSL mice provide a new model for studying BEC-to-hepatocyte transdifferentiation.
Here we generated a mouse model, in which the fumarylacetoacetase (Fah) gene is deleted, causing hepatocyte senescence during liver regeneration, modeling human hereditary tyrosinemia type I caused by a deficiency in FAH 26 and inducing BEC-to-hepatocyte transdifferentiation. Combining single-cell RNA sequencing (scRNA-seq) with dual recombinase-mediated lineage tracing and pathway modulations, we identified a subset of BECs with LPC potential, as well as the mechanisms coordinating stepwise BEC-to-hepatocyte transdifferentiation.

Generation of a model for BEC-to-hepatocyte conversion
We first generated a Fah-LSL mouse line, which contains a Fah deletion by introducing a LoxP-flanked Stop sequence (LSL) between exon1 and exon2, while allowing for Cre-induced Fah re-expression when needed (Extended Data Fig. 1a). Homozygous Fah-LSL/LSL mice lacked FAH expression and did not survive into adulthood without 2-(2-nitro-4-trifluoromethylbenzoyl)-1,3-cyclohexanedione (NTBC) treatment, a drug preventing injury in hepatocytes with FAH deletion 27 . In contrast, Fah-LSL/+mice and mice with Cre-LoxP-mediated removal of Stop sequence (ACTB-Cre;Fah-LSL/Fah-LSL mice), expressed FAH, were healthy and displayed normal growth (Extended Data Fig. 1b-e). We next analyzed the phenotypes in Fah-LSL/LSL mice after NTBC withdrawal. Fah-LSL/LSL mice were maintained with NTBC-containing water until 8 weeks of age. Compared to littermate Fah-LSL/ + mice, Fah-LSL/LSL mice showed significant body weight loss at 2 weeks after NTBC withdrawal and were moribund within 8 weeks (Extended Data Fig. 1f,g). Fah-LSL/LSL mouse livers exhibited abnormal hepatic architecture, widespread injury throughout liver lobules, disrupted metabolic zonation and hepatocyte senescence (p21 staining) in virtually all hepatocytes (Extended Data Fig. 1h-j). These data demonstrate that our Fah-LSL mice recapitulate the common pathological phenotypes of Fah −/− mice 28,29 , characterized by fulminant liver failure and impaired hepatocyte-mediated regeneration, with the advantage that our Fah-LSL knockout allele allows for Cre-induced restoration of Fah expression.
To assess whether CK19 + HNF4α + cells were also present in patients with severe liver injury, we analyzed biopsies from healthy livers and 11 different liver disease indications (Supplementary Table 3), including NASH and viral hepatitis with and without cirrhosis, primary biliary cirrhosis (PBC), primary sclerosing cholangitis (PSC), acute liver failure, autoimmune hepatitis (AIH) and alcoholic steatohepatitis (ASH) with cirrhosis (ASH cirrhosis). We observed significant numbers of p21 + hepatocytes in patients with ASH cirrhosis (Fig. 1i-k) and in the majority of other liver disease indications (Extended Data Fig. 6a), indicating substantial senescence, similar to what we observed in our animal model. In contrast to healthy livers, ASH cirrhosis patients (Fig. 1j,l) and most other patients with severe liver disease (Extended Data Fig. 6a) showed CK19 + cells with nuclear HNF4α staining. Interestingly, we found a positive correlation between hepatocyte senescence and CK19 + HNF4α + LPCs across 11 liver disease indications, with consistently higher percentages of both p21 + hepatocytes and LPCs in patients with cirrhosis compared to those with a non-cirrhotic milder form of the respective disease ( Fig. 1j-m and Extended Data Fig. 6a). This suggests that LPCs are common in patients with senescent hepatocytes across multiple liver disease indications. Notably, the majority of BECs in cirrhotic livers did not express p21, supporting their potential to proliferate during a ductular reaction and to convert into functional hepatocytes via LPCs (Extended Data Fig. 6b,c). Interestingly, CK19 + HNF4α + BECs were not restricted to ductular reactions, but we also found them in canals of Hering and bile ducts of cirrhotic livers (Extended Data Fig. 6d).

LPCs are a transitional cellular state between BECs and hepatocytes
To further characterize LPCs and their dynamics in liver regeneration, we first analyzed scRNA-seq data using UMAP embedding of RNA velocity from isolated EPCAM + cells in mice with injury-induced BEC-to-hepatocyte transdifferentiation (Fah-LSL/LSL mice 25 d after NTBC removal). The dynamic trajectories indicated a transition from BECs to LPCs (Fig. 2a). Using CK19-CreER;Fah-LSL/LSL;R26-tdT mouse model, we labeled BECs before injury and collected livers for analysis at 25 and 33 d after NTBC withdrawal (Fig. 2b). At day 25, we observed lineage-labeled (tdT + ) LPCs expressing both BEC markers (CK19, EPCAM, A6 and OPN) and hepatocyte marker HNF4α (Fig. 2c), indicating that LPCs originated from BECs. At day 33, we barely found LPCs anymore but instead detected tdT + hepatocytes (Fig. 2c), suggesting that LPCs were in a transient or transitional state during transdifferentiation. Therefore, we considered this LPC population to be transitional liver progenitor cells (TLPCs). CK19 + HNF4α + TLPCs did not acquire mature hepatocyte markers such as FAH at day 25, whereas tdT + hepatocytes became positive for FAH at day 33 while no longer expressing CK19 (Fig. 2d). TLPCs were quiescent (Ki67-negative) at day 25 ( Fig. 2e), which is consistent with our scRNA-seq data (Fig. 1b,c). In contrast, we found pronounced proliferation of tdT + hepatocytes at day 33 ( Fig. 2e), suggesting that transdifferentiated hepatocytes contributed to liver regeneration. Collectively, our data indicate that TLPCs transitionally emerged during BEC-to-hepatocyte transdifferentiation.

Notch signaling suppresses the differentiation of BECs into TLPCs
GO term analysis of scRNA-seq data and immunostaining data from CK19-CreER;Fah-LSL/LSL mice showed that Notch target genes, such as Onecut1 and Sox9, were substantially reduced in TLPCs compared with BECs ( Fig. 1g and Extended Data Fig. 5g), suggesting reduced Notch activity may promote the activation of TLPCs. To test this hypothesis, we blocked Notch signaling via Rbpj deletion in BECs and simultaneously traced these cells in our liver injury model, using CK19-2A-CreER mice that provide high recombination efficiency in BECs (Extended Data Fig. 7a-d). CK19-2A-CreER;Fah-LSL-LSL;R26-GFP;Rbpj fl/fl (Rbpj fl/fl ) mice enabled TAM-induced Rbpj deletion and GFP reporter expression in BECs, whereas CK19-2A-CreER;Fah-LSL-LSL;R26-GFP;Rbpj fl/+ (Rbpj fl/+ ) mice (enabling GFP expression in BECs while leaving one Rbpj allele intact) were used as controls (Fig. 4a). Rbpj expression was substantially reduced in BECs of Rbpj fl/fl mice when compared with littermate Rbpj fl/+ control mice (Fig. 4b). We only observed an insignificant decrease in ductular reaction and comparable serum total bilirubin in BEC-specific Rbpj knockout mice compared with littermate controls (Extended Data Fig. 7e-l). This suggests that the inducible Rbpj deletion, in only the subset of BECs we traced, did not impair bile duct integrity as reported during developmental Rbpj deletion 42 . A comparable percentage of GFP + BECs between the Rbpj fl/fl and Rbpj fl/+ groups (Fig. 4g) indicates functional ductular reaction in mice with liver-specific Rbpj deletion 43 . TLPC numbers were significantly increased at day 25 ( Fig. 4c,d), followed by increased numbers of GFP + BEC-derived hepatocytes at week 7 post-NTBC removal in Rbpj fl/fl mice compared with Rbpj fl/+ mice ( Fig. 4e-g). Proliferation of GFP + hepatocytes was comparable between Rbpj fl/fl and Rbpj fl/+ mice (Fig. 4h). Similarly, mice treated with Notch inhibitor DBZ showed increased numbers of TLPCs and hepatocyte clones in CK19-CreER;Fah-LSL/LSL;R26-tdT mice (Extended Data Fig. 8). These data suggest that loss of Notch signaling promotes BEC-to-TLPC activation, therefore increasing BEC-to-hepatocyte conversion.
To assess whether activation of Notch signaling would inhibit BEC-to-TLPC activation, we generated R26-NICD-GFP mice, in which Cre recombinase leads to the co-expression of the dominant active Notch intracellular domain (NICD) and GFP (Extended Data Fig. 9a,b). After crossing with CK19-2A-CreER mice, TAM treatment induced simultaneous GFP expression and Notch pathway activation in BECs (Extended Data Fig. 9c-f). Next, we generated CK19-2A-CreER;Fah-LSL-LSL;R26-NICD-GFP (NICD overexpression, NICD-OE) mice and CK19-2A-CreER;Fah-LSL-LSL;R26-tdT control mice, and collected livers at day 25 after NTBC removal for analysis (Fig. 5a). The number of TLPCs was significantly reduced following NICD overexpression, whereas BEC density and proliferation was dramatically increased (Fig. 5b-h and Supplementary Table 1). EPCAM + cells sorted from NICD-OE livers revealed substantial enrichment for genes representative of Notch signaling and gene signatures indicating cell proliferation, and substantially decreased expression of hepatocyte genes, compared with EPCAM + cells sorted from control mice ( Fig. 5f and Supplementary Table 4). We did not detect any GFP + hepatocytes at 7 weeks after NTBC removal in NICD-OE mice, compared with a considerable number of tdT + hepatocytes in the control mice (Fig. 5g). Taken together, activation of Notch signaling suppresses BEC-to-TLPC induction and promotes BEC proliferation in injured livers (Fig. 5i).

WNT/β-catenin signaling promotes TLPC-to-hepatocyte conversion
scRNA-seq profiling of tdT + BECs, TLPCs and TLPC-derived hepatocytes from CK19-CreER;Fah-LSL/LSL;R26-tdT mice at 28 days after NTBC withdrawal showed WNT/β-catenin signaling target genes, such as Lect2, Cyp2e1 and Cyp1a2, highly enriched in TLPC-derived hepatocytes as well as in a subset of TLPCs ( Fig. 6a-d). Immunostaining confirmed the expression of WNT/β-catenin-regulated CYP2E1 in newly formed hepatocytes from TLPCs at day 28 (Fig. 6e). Considering the higher expression of WNT/β-catenin-regulated genes in TLPC-derived hepatocytes (Fig. 6d,e), Axin2 expression in BECs from mice with impaired hepatocyte regeneration 44 , and the role of WNT/β-catenin signaling in promoting hepatocyte fate 36 , we hypothesized that WNT/β-catenin signaling may promote transdifferentiation into hepatocytes Middle panel shows violin plot of cell cycle score of genes related to G2M and S phases. Lower panel shows UMAP plot of gene set score of S, G2M and cell cycle. g, Immunostaining for tdT or GFP, HNF4α, and CK19 on liver sections collected at week 7 after NTBC removal. Quantification of the ductular reaction per ×10 field and the percentage of tdT + or GFP + hepatocytes is shown on the right panel. Data represent mean ± s.d.; n = 5 mice; CK19 density (10×): *P < 0.0001; Reporter + Heps: *P < 0.0001. Scale bars, 100 µm. h, Immunostaining for tdT or GFP, Ki67 and CK19 on liver sections collected at week 7 after NTBC removal. Percentage of Ki67 + BECs is shown on the right panel . Data represent mean ± s.d.; n = 5 mice; *P < 0.0001. Scale bars, 100 µm. i, Schematic showing that Notch signaling inhibited BEC-to-TLPCS activation. Statistical analysis was performed by twotailed unpaired Student's t test in c, d, g and h. w, weeks.
Article https://doi.org/10.1038/s41588-023-01335-9 during liver regeneration. To test this hypothesis, we first generated CK19-2A-CreER;Fah-LSL/LSL;Ctnnb1 fl/fl ;R26-Confetti (mutant) mice, in which TAM induced deletion of β-catenin (resulting in loss of WNT/β-catenin signaling) and the simultaneous expression of a Confetti reporter 37 in BECs (Fig. 6f). Littermate CK19-2A-CreER;Fah-LSL/ LSL;Ctnnb1 fl/+ ;R26-Confetti mice were used as controls (Fig. 6f). We collected livers from the mutant and control mice at day 25 and ~7 weeks after NTBC removal for analysis of TLPCs and hepatocytes ( Fig. 6f) and confirmed deletion of β-catenin in a subset of BECs (Fig. 6g). Comparable numbers of CK19 + HNF4α + TLPCs between the mutant and control groups (Fig. 6h) suggest that WNT/β-catenin signaling is not required for BEC-to-TLPC induction. However, it is likely that Ctnnb1 deletion was not complete in all BECs, which may underestimate the role of WNT/β-catenin signaling in BEC-to-TLPC conversion. The number of BEC-derived hepatocyte clones was significantly reduced in mutant compared with control mice (Fig. 6i)      CK19 + cells remained similar between the mutant and control groups (Fig. 6j), suggesting that WNT/β-catenin signaling is dispensable for a ductular reaction as previously reported 45,46 . These data suggest that WNT/β-catenin signaling is required for efficient hepatocyte transdifferentiation during liver regeneration (Fig. 6k).

Discussion
While it is established that hepatocytes can re-enter the cell cycle to proliferate and restore a functional hepatocyte pool in response to various injuries, the contribution of facultative LPCs to this process has been controversial 7,[9][10][11]13,[51][52][53][54] . The discovery of BEC-to-hepatocyte transdifferentiation in conditions where hepatocyte-mediated regeneration is impaired provided an important new concept enabling liver regeneration [13][14][15][16][17][18] . Unfortunately, the cellular identity of the BECs with such facultative LPC potential, as well as the molecular mechanisms enabling their transdifferentiation, remained unclear.
We now identified quiescent TLPCs, which are characterized by a hybrid BEC/hepatocyte gene expression signature and represent a transitional LPC state that situates in-between BECs and hepatocytes. Whether all BECs or just a subset of them have the potential to become TLPCs remains unclear. Additional lineage tracing and profiling studies in rats will be necessary to clarify whether oval cells, previously identified in pioneering studies showing hepatocyte transdifferentiation [55][56][57] , are similar to the TLPCs we identified in mice. It is also possible that additional HNF4α-negative LPCs, identified by other markers, contribute to liver regeneration by restoring hepatocytes. Several other markers have been used to identify putative BEC-associated LPCs, including CD24/CD133 (ref. 58 ), Foxl1 (refs. 14,53,54 ) and Tweak/Fn14 (ref. 59 ). However, validation of TLPCs and uncovering their full differentiation potential requires lineage tracing studies with improved genetic approaches. Using dual genetic lineage tracing to specifically label HNF4α + CK19 + TLPCs, we now prove that these are a source of newly generated hepatocytes in conditions where hepatocyte regeneration is impaired.
Co-expression of BEC and hepatocyte markers has also been reported in human livers. While classical immunostaining did not identify such mixed populations in healthy adult human livers 60 , tissue-tethered cytometric analyses found CK19 + BECs with faint HNF4α expression 61 , possibly representing primed BECs that can upregulate HNF4α to acquire LPC potential when needed. Notably, patients with acute liver failure 41 or cirrhotic livers in patients with viral hepatitis or AIH 16 showed strong HNF4α expression in BECs. We now identified HNF4α + CK19 + BECs in 11 different liver disease indications, whereas we only found negligible amounts in healthy patients. We further found that the induction of these TLPCs correlates with b c Article https://doi.org/10.1038/s41588-023-01335-9 the amounts of senescent hepatocytes and severity of the disease. In cirrhotic livers, generation of hepatocytes by BECs has been proposed to be a major mechanism for parenchymal regeneration 62,63 . Together, this suggests that TLPC induction may be a common mechanism in human liver disease. However, lack of lineage tracing possibilities in patients does not allow for proving their transdifferentiation potential.
Mechanistically, we show that WNT/β-catenin and Notch signaling pathways orchestrate the stepwise BEC-TLPC-hepatocyte transdifferentiation process (Fig. 8). Notch signaling is crucial for determining cell lineages in the liver, regulating the differentiation of hepatoblasts to cholangiocytes 38,39,64-66 . We found that inhibition of Notch signaling enhanced BEC-to-TLPC conversion, while increased Notch signaling blocked this process. We further show that activated WNT/β-catenin signaling promoted TLPC-to-hepatocyte conversion and also enhanced the conversion of BECs into TLPCs, consistent with WNT/β-catenin-induced BEC-to-hepatocyte conversion in hepatic organoids 44 . Conversely, abrogation of WNT/β-catenin signaling blocked TLPC-to-hepatocyte conversion and newly generated hepatocytes express WNT-regulated metabolic enzymes, suggesting that the pathway is key to the transdifferentiation process. Because our genetic tool only allows us to activate WNT/β-catenin signaling in BECs, but not in TLPCs specifically, we could not directly distinguish whether WNT/β-catenin pathway activation affects TLPC-to-hepatocyte or TLPC-to-BEC conversion. Notably, extended survival in injured mice following BEC-to-hepatocyte transdifferentiation, and the potential of the Notch inhibitor DBZ and WNT/β-catenin pathway agonist RSPO1 in enhancing transdifferentiation, suggest possibilities to therapeutically exploit this regenerative mechanism. However, Notch blockade in the liver may impair the biliary system 42 and RSPO1 treatment impaired metabolic zonation 49 , suggesting a more targeted therapy would be required to benefit patients with liver disease. Together, our data provide the cellular identity and mechanistic cues for transdifferentiation-mediated liver regeneration, establishing a rational basis for potentially therapeutic concepts leveraging this fundamental regenerative process.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41588-023-01335-9.

Genomic PCR
Genomic DNA was extracted from the mouse tail. Tissues were lysed with Proteinase K overnight at 55 °C, followed by centrifugation at maximum speed for 8 min. DNA was deposited with isopropanol and washed in 70% ethanol. All mice were genotyped with specific PCR primers that distinguish knock-in alleles from wild-type alleles. Sequences of all primers were included in Supplementary Table 5.

Immunostaining
Immunostaining was performed as previously described 74 . In detail, livers were fixed in 4% PFA at 4 °C for 1 h, then washed in PBS and dehydrated in 30% sucrose overnight at 4 °C and embedded in OCT (Sakura). For staining, the cryosections were washed in PBS, incubated in blocking buffer (5% normal donkey serum ( Jackson Immunoresearch), 0.1% Triton X-100 in PBS) for 30 min at room temperature then stained with the primary antibodies overnight at 4 °C. Signals were developed with Alexa fluorescence antibodies (Invitrogen). HRP-conjugated antibodies with tyramide signal amplification kit (PerkinElmer) were used to amplify weak signals. HRP-conjugated antibodies with ImmPACT DAB kit (Vector lab, SK-4105) were used to show CK19 density. Nuclei were counterstained with 4'6-diamidino-2-phenylindole (DAPI, Vector lab). The following antibodies were used: tdT (tdTomato,  (20 µm) were used for staining. For TLPC number quantification, we collected ten tissue sections (20 tissue sections in Ctnnb1 gene knockout experiment) from each mouse liver and took five random fields from each liver section for quantification. Immunostainings for CK19 and HNF4α were performed on both mutant and control liver sections at the same time to avoid potential batch differences during staining. Imaging of all immunostained slides was performed under the same exposure and contrast conditions using the same confocal microscope.

DBZ and RSPO1 treatment
To study the effect of dibenzazepine ( H&E staining. Cryosections were washed in PBS for 5 min to remove OCT, then incubated in hematoxylin A for 10 min, followed by washing in water. Then, cryosections were incubated in 1% concentrated hydrochloric acid diluted in 70% ethanol for 1 min and washed in water. Afterward, the sections were incubated in 1% ammonia water for 1 min, followed by washing in water. The sections were stained with Eosin-Y solution for 10 s and dehydrated in ethanol and xylene. Finally, sections were mounted with resinous medium. Images were acquired using an Olympus microscope (Olympus, BX53).
Cell isolation and fluorescence-activated cell sorting. Liver cells were isolated by standard two-step collagenase perfusion as described previously 33 . Briefly, mice were anesthetized and the liver was exposed through an incision in the lower abdomen. A needle was inserted into the inferior vena cava and secured with a hemostatic clamp around the needle. Portal vein was cut immediately when the mouse liver was perfused with perfusion medium using a peristaltic pump. Then, the liver was next perfused with medium containing collagenase type I (150 U ml −1 ; Invitrogen) for 10 min to adequately digest the liver. After removing the gallbladder, the liver was dissected with cold resuspension buffer (0.5% BSA and 2 mM EDTA in PBS) to free the hepatic cells. Then the cell suspension was passed through a 70-µm cell strainer (BD Biosciences, 352350) and centrifuged at 50g for 3 min at 4 °C.
Article https://doi.org/10.1038/s41588-023-01335-9 The non-parenchymal cells that remained in supernatant were collected and passed through a 40 µm cell strainer (BD Biosciences, 352340), then centrifuged at 400g for 5 min at 4 °C. The cell pellet was resuspended in red blood cell lysis buffer (eBioscience, 00-4333-57) for 5 min at room temperature and washed with cold resuspension buffer and centrifuged at 400g for 5 min at 4 °C. The washing step was repeated once again. Subsequently, cells were stained with the positive selection antibody (anti-mouse EPCAM-APC; eBioscience, 17-5791-82) diluted in resuspension buffer for 30 min in the dark at 4 °C. After staining, cells were washed with resuspension buffer and centrifuged at 400g for 5 min. EPCAM + cells were enriched by using APC microbeads (130-090-855, Miltnyi Biotec) according to the manufacturer's protocols before sorting with Sony MA900 equipped with a 100 µm nozzle in purity mode. Cell viability was assessed with DAPI staining. EPCAM + cells were isolated for further scRNA-seq of BECs or bulk RNA-seq or qRT-PCR. For scRNA-seq of tdT + cells (Fig. 6a-c), cells were not stained with anti-mouse EPCAM-APC antibody. Cells were sorted based on the expression of tdT.

Human samples and IHC analysis
Glass slides with formalin-fixed and paraffin-embedded (FFPE) sections from patient livers (healthy n = 6, ASH cirrhosis n = 5, acute liver failure n = 9, nonalcoholic steatohepatitis (NASH) noncirrhosis n = 6, NASH cirrhosis n = 6, hepatitis B (HepB) noncirrhosis n = 6, HepB cirrhosis n = 6, Hepatitis C (HepC) noncirrhosis n = 6, HepC cirrhosis n = 5, AIH n = 5, PSC n = 5 and PBC n = 5) were obtained from the University Hospital Basel Tissue Bank. Healthy livers were classified by normal morphology during histopathological assessment. The biopsies were originally acquired for routine diagnostic and patients signed a general informed consent for the use of remaining tissue for research purposes in accordance with the Swiss Federal Human Research Act (HRA

scRNA-seq and bioinformatics analysis
scRNA-seq. Isolated cell suspension was loaded to the 10X Chromium and ~8,000 cells were expected to be captured when Gel Beads-in-emulsions were generated. The library was prepared followed by the instruction manual of Single Cell 3′ Gene Expression kit (v3.1) or Single Cell 5′ Gene Expression kit (v2). Briefly, the Gel Beads-in-emulsions were first incubated and reverse transcripted to first-strand cDNA. The single-strand cDNA was purified by Dynabeads and amplified using 12 cycles to generate the double strands cDNA. Next, dsDNA was fragmented, end-repaired and further ligated with adaptor. Lastly, index PCR was performed before sequencing. The library was sequenced on the Illumina Hiseq X ten PE150 platform.
Single-cell transcriptomic analysis. Sequencing reads were aligned, annotated and demultiplexed using CellRanger (v4.0.0) with the mm10-2020-A reference provided by 10X Genomics. Further downstream analyses were carried out using the Seurat R package (v4.0.5) 75 . Quality control was performed using the subset function using the threshold of nFeature_RNA larger than 2,000 and less than 8,000, nCount_RNA larger than 8,000 and less than 50,000, as well as percentage of expressed mitochondrial gene less than 10% to filter out low-quality cells and potential doublets. PCA was calculated using the scaled expression data of 3,000 most variable genes, which were selected by 'vst' method using FindVariableFeatures function. Dimension reduction and clustering were further performed. Different dims of PCA and different values of resolution parameters were tested 76 . We set the final resolution to 0.2 (testing a range from 0.1 to 0.5) and dims to 15 (testing a range from 10 to 20) first in the sample of CK19-CreER;Fah-LSL/ LSL mice. Given that the obtained clustering sensitivity for a given resolution is dependent on the number of cells of that subpopulation in each respective sample, we swept over the same range of resolutions for the other samples to assure the proportion of TLPCs is comparable with the statistics result.
DEGs and pathway enrichment. Two-sided Wilcoxon rank-sum test was used to define marker genes for clusters and samples using the FindMarkers function in Seurat and P values were Benjamin-Hochberg FDR correction for the total number of comparisons. The GO BP pathway enrichment analyses of DEGs calculated above are performed using Metascape webtool 77 .

Data integration.
To compare the scRNA-seq data from the sample of CK19-CreER;Fah-LSL/LSL mice and two published samples of mice with DDC-induced injury, data integration was performed using the MNN algorithm 78 . In detail, QC filter and preprocessing were performed as described in the original articles 32,33 . The RunFastMNN function in SeuratWrappers package was used to integrate these three datasets. The highest 2,000 variable features were selected to correct the batch effects between samples. The dimensions of the first 19 MNNs and resolution of 0.22 were used to unsupervised cluster all cells. The first 19 MNNs were used to reduce dimensions by RunUMAP function. Other datasets were also integrated by the RunFastMNN function. The same number of MNN dimensions but different resolutions were used for clustering and dimension reduction.

Trajectory.
To map the differentiation trajectory directions, scVelo was used to calculate the RNA velocity 79 . In brief, the cell filter mentioned above was used to re-UMAP using the spliced assay data. Genes less than 20 counts were filtered out and 1,500 highly variable genes were retained and log-normalized. 30 PCs and 30 neighbors were used to compute moments based on connectivity, and then calculate velocities for each individual cell. The velocity embedding stream plot was drawn and colored by Seurat clusters.
Cell cycle scoring for scRNA-seq. To evaluate the potential ability of proliferation for each cell, we first calculate the S.score and G2M.score for each single-cell data using the CellCycleScoring function in Seurat package. Also, we used enrichIt function in escape 80  Article https://doi.org/10.1038/s41588-023-01335-9 The changes in these scores were inconsistent in the comparison of To perform the KS test, the fgsea package first ranks the genes in the gene set and the genes outside of the gene set by their statistical significance (for example, P values). The cumulative distribution functions of the ranked genes in the gene set and the ranked genes outside of the gene set are then calculated. The maximum difference between these two cumulative distribution functions is then calculated and used as the test statistic.
RNA isolation and quantitative RT-PCR. Total RNA was extracted from the liver of indicated mice or BECs isolated from indicated mice treated with TAM or oil. Cells were lysed with Trizol (Invitrogen, 15596018), and total RNA was extracted according to the manufacturer's instructions. Then, RNA was reverse-transcribed into cDNA using Prime Script RT kit (Takara, RR047A). The SYBR Green qPCR master mix (Thermo Fisher Scientific, 4367659) was used and quantitative RT-PCR was performed on QuantStudio 6 Real-Time PCR System (Thermo Fisher Scientific).
Gapdh was used as internal control. For qPCR of Fah gene, the forward primer for qPCR is located in exon7 and the reverse primer is located in exon8, and their PCR produced is 74 bp overlapping part of exon7 and exon8. Sequences of all primers are included in Supplementary Table 5.
Serum biochemical analysis. The blood was collected from indicated mice and centrifuged at 850g for 15 min at 4 °C. The serum that remained in the supernatant was collected for biochemical analyses. ALT and AST were measured by 7600 clinical analyzer (Hitachi) or 4600 fully automatic biochemical analyzer (VITROS). TBIL was measured by 4600 fully automatic biochemical analyzer (VITROS).
Statistics. For image acquisition, as well as analyses such as quantification by IF and IHC of cell number or CK19 density, the investigators were blinded. Investigators were not blinded to mouse treatment and sacrifice because mouse treatment and sacrifice were performed by the same people. Investigators were not blinded for scRNA-seq analysis studies as there were no separate groups and the samples were annotated. For western and qPCR, the investigators were not blinded to the loading samples. Within an experimental condition, the allocation of mice was random. Data were presented as means ± s.d. Statistical analysis was performed by two-tailed unpaired Student's t test for comparison of differences between two groups, and by ANOVA followed by Tukey's method for multiple comparisons. P < 0.05 was considered to be statistically significant. The P value was added in the figure legend for each comparison, with statistical method included. Each image in Fig. 1e is representative of five individual mice samples. Each image in Fig. 1f is representative of five individual mice samples. Each image in

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
All of the data generated or analyzed during this study are included in Figs

Code availability
This study did not generate any unique code or algorithm. The algorithms used for the analysis during this study are all publicly available.
The code of single cell data processing and analysis in this study have been deposited in Zenodo (https://doi.org/10.5281/zenodo.7576366). Data are the mean ± SD; n = 5 mice. Scale bars, 50 µm. d, Immunostaining for tdT, nGFP/YFP, and CK19 on the liver sections collected from CK19-CreER;Fah-LSL/LSL;R26-confetti mice at week 10. Numbers in the images indicated the sequential order of serial 20 µm-sections of 2 entire clones. White arrowhead, reporter + hepatocytes; yellow arrowhead, reporter + BECs. Scale bars, 100 µm. e, Percentage of reporter + hepatocyte clones located close to the PV or CV region. f, Cartoon image showing that BECs proliferate before conversion into TLPCs, and TLPCs converted into either hepatocytes or BECs. Single BECs could contribute to both hepatocytes and BECs during liver injury. Data are the mean ± SD; n = 5 mice. PV, portal vein; CV, central vein.