The ability of adult organs to regenerate has been well documented in animal models, and functional studies combined with lineage-tracing experiments have shown that different injuries induce divergent regenerative processes1,2,3,4. However, it is difficult to demonstrate the existence of such events in human organs for technical and ethical reasons. The liver is a particularly interesting organ in this context. The main functional cell types in the hepatic epithelium are the hepatocytes, which are known for their metabolic roles, and the cholangiocytes, which line the biliary tree and transport bile acids. The process by which these cells are replaced after injury depends on the insult encountered. Cell proliferation occurs during acute liver injury5,6,7, but this capacity to proliferate is abolished in chronic diseases8,9. Animal studies have revealed three alternative mechanisms10,11: stem cells or progenitors can be activated and then differentiate into epithelial cells12,13,14,15; cholangiocytes may transdifferentiate into hepatocytes, or vice versa1,16,17,18,19,20,21,22,23; or hepatocytes and cholangiocytes could reverse to a developmental progenitor to restore the corresponding cell compartment24,25,26. Signs of these mechanisms have been observed in humans, but the nature of the regenerative processes that occur during chronic liver disease remain to be fully understood27,28,29. To address this question, we combined single-nucleus analyses, 3D imaging and functional experiments to study both cell behaviour and the regenerative processes that occur during the progression of metabolic dysfunction-associated steatotic liver disease (MASLD), a chronic liver disease that affects a growing population of people worldwide30.

snRNA-seq captures liver cells across MASLD

MASLD is a progressive disease that starts with the accumulation of fat in hepatocytes. Over time, this accumulation can result in cell death, leading to inflammation, fibrosis, cirrhosis and liver failure or liver cancer31. We first assessed whether livers affected by progressive MASLD display evidence of regenerative processes. For that, we performed immunostaining to compare tissue sections of healthy liver with those of biopsies from people at different stages of disease progression (Fig. 1a,b and Extended Data Fig. 1a). Major changes were evident, especially in livers from people with end-stage disease, with the expected appearance of regenerative nodules containing hepatocytes (indicated by the hepatocyte marker ALB) surrounded by large collagen depositions32 (Extended Data Fig. 1b). Immunostaining for the cholangiocyte markers keratin 7 (K7, also known as KRT7) and keratin 19 (K19, also known as KRT19) showed a strong increase in ductal structures around these nodules (Fig. 1a), a process known as the ductular reaction33, which is commonly seen in acute and chronic liver disease34,35. These experiments also revealed cells co-expressing K7 and the hepatocyte markers ALB or HepPar1 (Fig. 1b and Extended Data Fig. 1c). These may represent intermediate hepatocytes, which have been observed histologically in human MASLD28. However, we also observed cells co-expressing ALB, K7 and K19 that seem to be present specifically in end-stage liver (Fig. 1b and Extended Data Fig. 1d), indicating the presence of cells combining hepatocyte and cholangiocyte phenotypes. Importantly, such biphenotypic cells have been associated with the regenerative process32,36,37, so their appearance could be indicative of epithelial regeneration in end-stage MASLD.

Fig. 1: Using snRNA-seq of MASLD progression to analyse cholangiocyte and hepatocyte plasticity.
figure 1

a, Immunofluorescence staining for K7 and ALB in liver sections from healthy donors and those with end stage disease, with high magnification of the areas in the dashed boxes underneath. Scale bars: 1,000 μm for low magnifications and 100 μm for high magnifications; n = 3 healthy and n = 3 end-stage-disease tissue samples. b, Immunofluorescence staining of end-stage MASLD tissue sections. High magnification of the dashed boxes shows examples of cells that are double-positive for K19 or K7 and for ALB in the hepatocyte nodule and in the surrounding ductal structures. An example is indicated by the yellow arrows; n = 3 tissue samples. Scale bars: left, 100 μm; middle, 10 μm; right, 15 μm. c, Schematic of the snRNA-seq experimental workflow; n shows the number of samples at each stage. d, Overall UMAP showing cell annotation from all disease stages after quality control. e, Bubble plot of the expression of cell-type markers. f, Overall UMAP shown by disease stage.

To further examine the events leading to the emergence of biphenotypic cells and their role in disease, we decided to study MASLD progression at the single-cell level. To this end, we collected liver biopsies from 47 people across the different stages of MASLD progression defined by histology as healthy, MASLD, metabolic dysfunction-associated steatohepatitis (MASH), cirrhosis and end-stage disease (Fig. 1c and Supplementary Tables 1 and 2). Half of the biopsy was allocated for diagnostic and staging work, and the other half was rapidly frozen to be processed at a later stage (Fig. 1c). However, we quickly abandoned using cells isolated from fresh biopsy because many hepatocytes and cholangiocytes were lost using this method, as shown by previous studies38,39. To bypass this limitation, we developed a protocol for nucleus isolation involving tissue lysis and fluorescence-activated cell sorting (FACS), which allowed the purification of high-quality nuclei even from fibrotic tissues. Using this protocol, just under 100,000 nuclei were isolated after quality control, which excluded cells expressing stress markers such as mitochondrial and ribosomal proteins (Fig. 1d and Extended Data Fig. 2a,b). Further analyses confirmed that our method captured all the expected liver cell types from all disease stages at similar proportions to the native tissue (Fig. 1d–f, Extended Data Fig. 3a–c and Supplementary Table 3). Accordingly, our collection was enriched in hepatocytes (n = 69,426) and cholangiocytes (n = 5,412). Notably, cell-type-specific clusters mostly overlap independently of the disease stage, except for hepatocytes, which display clear transcriptional changes following disease progression, even after a batch correction of technical effects using Harmony40 (Fig. 1f and Extended Data Fig. 3d,e). Thus, hepatocytes seem to be the cell type most affected by the disease. Finally, our single-nucleus analyses also revealed the presence of cells co-expressing hepatocyte and cholangiocyte markers. We observed cells that appeared to bridge hepatocyte and cholangiocyte clusters and that co-expressed specific markers for both cell types (Extended Data Fig. 3f). Quality control was performed to confirm that these cells were not the result of doublets or RNA contamination. Together, these experiments show that our single-nucleus isolation protocol is compatible with single-cell-level transcriptomic analyses of liver biopsies and confirm the presence of biphenotypic cells, which have previously been associated with regenerative processes in the liver of people with progressive MASLD.

MASLD remodels the liver microenvironment

Before investigating the origin of biphenotypic cells in more detail, we decided to probe the transcriptomic changes occurring in each cell type. All cell types exhibited differentially expressed genes across disease progression with strong separation of cells in end-stage disease for cholangiocytes, stellate cells and endothelial cells observed in the UMAP (uniform manifold approximation and projection) space, indicating that disease progression affects all the liver cells (Extended Data Fig. 4a–f). However, hepatocyte populations displayed the strongest transcriptional change in end-stage disease (Fig. 2a), and gene-set enrichment analyses (GSEA) showed a diversity of pathways upregulated during disease progression. Of particular interest, we observed major adjustments in pathways related to the microenvironment, such as hypoxia-inducible factor I signalling and gluconeogenesis, indicative of changes in liver zonation (Extended Data Fig. 5a). In the healthy liver, hepatocytes located in different zones of the liver lobules diverged in their expression of functional markers. For example, hepatocytes closer to the central vein (pericentral) express the WNT signalling genes LGR5 and AXIN2, whereas hepatocytes closer to the portal triad (periportal) express the metabolic enzymes HAL and ASS1 (ref. 41). Accordingly, healthy hepatocytes can be clearly separated using correlation analyses for known zonation markers (Fig. 2b). However, this distinction breaks down during disease progression, with end-stage hepatocytes co-expressing pericentral and periportal markers (Fig 2b and Extended Data Fig. 5b). These observations were validated by immunostaining and 3D fast light-microscopic analysis of antibody-stained whole organs (FLASH) imaging for pericentral marker GLUL, periportal marker ASS1 and the pan-hepatocyte marker ALB in optically cleared tissue. Cells aberrantly co-expressing these markers were observed across regenerative nodules in end-stage livers (Fig. 2c, Extended Data Fig. 5c and Supplementary Videos 1 and 2), indicating a loss of zonation at the transcriptional and protein level in hepatocytes. These results reinforce previous studies42,43, but by showing that hepatocytes acquire progressively the capacity to co-express zonation markers that are mutually exclusive in healthy liver, they also indicate that disease progression strongly modifies the liver microenvironment, resulting in the loss of functional zonation.

Fig. 2: Major changes in hepatocyte zonation and biliary-tree remodelling in end-stage MASLD.
figure 2

a, UMAP of hepatocytes annotated by disease stage. b, Correlation analysis examining expression of pericentral and periportal hepatocyte markers across disease progression. c, Immunofluorescence staining for pericentral marker GLUL, periportal marker ASS1 and pan-hepatocyte marker ALB in healthy and end-stage MASLD tissue sections. The yellow dashed arrow indicates the central vein; n = 3 healthy and n = 3 end-stage tissue samples. Scale bars: 50 μm (healthy) and 10 μm (end stage). d, 3D projections of cleared healthy and end-stage MASLD liver samples. Staining with K7 for cholangiocytes and MRP2 for hepatocytes; n = 3 healthy and n = 3 end-stage tissue samples. Scale bars: 100 μm (healthy), 50 μm (end stage). e, The area in the yellow box (left) is shown in higher magnification in the other three boxes, highlighting an example of a MRP2–K19 co-positive cell at the end of a duct; n = 3 healthy and n = 3 end-stage tissue samples. Scale bars: 30 μm (left), 10 μm (the other panels). See also Supplementary Videos 3 and 4.

The organization of the biliary epithelium can also be strongly affected during disease progression by the ductular reaction44,45. Accordingly, cholangiocytes also display a strong disease signature (Extended Data Fig. 4d) characterized by an increase in the ductular-reaction markers NCAM1 and TNFRSF12A. In parallel, we observed increased numbers of bile ducts during MASLD progression (Fig. 1a) to an extent indicating that this process has a major effect on liver architecture. To confirm this hypothesis, we imaged the biliary tree in 3D using FLASH technology on healthy and end-stage tissue. As expected, K7 staining of healthy tissue revealed a network of ducts that formed a branching tree-like structure (Fig. 2d). By contrast, end-stage samples exhibited complex basket-like structures surrounding the hepatocyte nodules. (Fig. 2d and Supplementary Videos 3 and 4). Such structures indicate a profound remodelling of the biliary tree to an extent not previously suspected. Furthermore, FLASH imaging was also performed to define the location of the biphenotypic cells in the diseased biliary tree. This revealed that cells co-expressing K7 and hepatocyte marker MRP2 tended to be located towards the ends of the small ducts (Fig. 2e and Supplementary Video 5), which undergo major transformation during disease progression by becoming bulkier and containing multiple cells in the end stage. Interestingly, the same region has previously been associated with hepatic stem cells46 (Extended Data Fig. 5d). Taken together, these results indicate that the appearance of biphenotypic cells could be associated with a major reorganization of the biliary tree and the liver microenvironment during disease progression.

Hepatocyte and cholangiocyte plasticity

Having established the presence of biphenotypic cells and their association in part with the abnormal organization of the biliary tree during disease, we next focused on defining their origin by performing detailed subclustering of hepatocytes and cholangiocytes. These analyses revealed two cholangiocyte subpopulations, namely MUC1-expressing cholangiocytes from the larger ducts and small cholangiocytes that expressed BCL2 (Extended Data Fig. 6a–d). The population of small cholangiocytes was also associated with the ductular-reaction markers NCAM1 and TNFRSF12A (ref. 29; Extended Data Fig. 6e–g), indicating that our sampling did capture ductular-reaction structures. These ductular-reaction cells were more common in end-stage disease, in line with our observations of the tissue sections (Extended Data Figs. 1a,d and 6j). Furthermore, subclustering of cholangiocyte populations identified biphenotypic cells expressing multiple hepatocyte markers (Fig. 3a,b) in clusters 5, 9 and 1 (Fig. 3c,d). Interestingly, the same analyses performed on hepatocytes also revealed that cluster 9 includes cells expressing multiple cholangiocyte markers. Thus, both cell types could be able to generate biphenotypic cells. Hepatocyte cluster 9 contains cells from different disease stages, so we decided to further subcluster this population to identify a more-biphenotypic phenotype. Notably, hepatocytes expressing the highest level of cholangiocyte markers were more common in end-stage disease (Fig. 3e), indicating that cell plasticity could occur with disease progression (Fig. 1b). These cells tended to express cholangiocyte markers for small but not large cholangiocytes (Extended Data Fig. 7a) and may indicate that they are more likely to be found in small ducts, in line with our observations from 3D staining (Fig. 2d). Similar analyses performed on cholangiocytes showed that cholangiocyte cluster 1 cells were more prominent in end-stage disease, whereas clusters 5 and 9 were also found in earlier stages (Extended Data Fig. 7b–d). Biphenotypic cells were found to express comparable levels of cholangiocyte and hepatocyte markers to the main cholangiocyte and hepatocyte populations (Extended Data Fig. 7e). These results indicate that biphenotypic cells could appear earlier in disease than was initially suggested by our immunostaining analyses (Extended Data Fig. 1a). These early cells could represent intermediate cells described previously28 that display limited plasticity. The full phenotype, and thus the capacity to generate cells with a biphenotypic transcriptome, seems to be acquired only towards the end stage of progression.

Fig. 3: Using snRNA-seq identifies cholangiocyte and hepatocyte plasticity.
figure 3

a, Subclustering of hepatocytes. b, Relative expression of hepatocyte and cholangiocyte markers across hepatocyte subclusters. c, Subclustering of cholangiocytes. d, Relative expression of hepatocyte and cholangiocyte markers across cholangiocyte subclusters. e, Proportion of hepatocytes classified as cholangiocyte-like hepatocytes (identified by subclustering hepatocyte cluster 9 in b) that are expressing cholangiocyte markers, by disease stage. Statistical significance was calculated using two-sided Welch’s t-test (n = 47 biologically independent donors: healthy, 4; NAFLD, 7; NASH, 27; cirrhosis, 4; end stage, 5). The P value was 0.03058 (significant under a 0.05 threshold). The mid-point, minimum and maximum of the boxplot summary correspond to the median, first and third quartiles. The extent of the whiskers corresponds to the largest and smallest values no further than 1.5 IQR from the inter-quartile range. f, UMAP of cholangiocytes and hepatocytes from end-stage MASLD disease only. g, Cholangiocyte-like hepatocytes and hepatocyte-like cholangiocytes (identified by subclustering hepatocyte cluster 9 in b and subclustering cholangiocyte cluster 1 in d, respectively) that express hepatocyte markers are plotted to show their location on the UMAP. h, RNA velocity using cholangiocyte-like hepatocytes. i, RNA velocity using hepatocyte-like cholangiocytes. j, Pseudotime trajectory across the connected region of the two cell types. k, Heat map of DEGs (differentially expressed genes) across the trajectory. l, Immunofluorescence staining of the cholangiocyte-like hepatocyte markers SOX4 and K23 alongside ALB and K19 in healthy and end-stage sections; n = 3 healthy and n = 3 end-stage tissue samples. Scale bars: 30 μm in the SOX4 images (top); 10 μm in the K23 images (bottom).

We then decided to define the origin of these biphenotypic cells. In the biphenotypic population, no cells were found to co-express the adult stem-cell markers LGR5 and TROP2 (also known as TACSTD2) (Extended Data Fig. 7f–i). We also hypothesized that stem cells, by definition, should be able to self-renew. However, proliferative markers such as MKI67 were rarely co-expressed with LGR5 (n = 1 of 46 LGR5-positive cells), and no proliferative cells expressed TROP2 (Extended Data Fig. 7f–i). Notably, quiescence-marker expression increased during disease progression in hepatocytes (Extended Data Fig. 7j), confirming that regeneration by proliferation is limited in chronic injury. Together, these results indicate that biphenotypic cells are unlikely to originate from a stem-cell population. We next tried to determine whether a dedifferentiation or redifferentiation process could be occurring in the biphenotypic cells. For that, we examined the expression of the fetal liver markers AFP and SPINK1 (ref. 47). No cells co-expressing AFP and SPINK1 were found, and although rare AFP+ cells were observed, none were proliferative (Extended Data Fig. 7k,l). Thus, biphenotypic cells do not seem to originate from a stem-cell population or from a dedifferentiated or developmental progenitor. Together, these results indicate that biphenotypic cells appear during disease progression, whereas transdifferentiation is prominent in end-stage disease. These data do not rule out a role for the ductular reaction and/or intermediate hepatocytes in this acquisition of plasticity, and that these cells may act as precursors to biphenotypic cells, which increase over time during chronic injury.

Identification of plasticity factors

We next investigated the mechanisms that increase plasticity by focusing on end-stage cells because they display the highest level of marker co-expression. We generated a UMAP including only hepatocytes and cholangiocytes from end-stage disease (Fig. 3f) and then localized the biphenotypic cells from a subcluster of hepatocyte cluster 9 and a subcluster of cholangiocyte cluster 1. As expected, the selected cells bridge cholangiocytes and hepatocytes, confirming their transdifferentiating state (Fig. 3g). To address the directionality of this transdifferentiation, we calculated the RNA velocity for these cells. Cholangiocyte-like hepatocytes indicated bidirectionality, whereas hepatocyte-like cholangiocytes showed a predominant direction from cholangiocytes to hepatocytes (Fig. 3h,i). Thus, transdifferentiation seems to occur in both directions. We then inferred the pseudotime to identify genes expressed specifically during transdifferentiation (Fig. 3j). This analysis revealed numerous genes that were upregulated in the biphenotypic population (Fig. 3k), and we selected SOX4, KRT23, KLF4 and NCAM1 for further validation. Immunostaining revealed SOX4+ nuclei in end-stage cholangiocytes and hepatocytes, whereas SOX4 was not observed in healthy liver (Fig. 3l and Extended Data Fig. 8a). Similarly, K23+ cholangiocytes were observed in end-stage disease, with some cells co-expressing K19, ALB and HepPar1 (Extended Data Fig. 8b). By contrast, K23 was not found in healthy cholangiocytes. Cells co-expressing SOX4 and K23 were also observed (Extended Data Fig. 8a). Similar staining patterns were found for KLF6 and NCAM1, with clear increases in end-stage disease (Extended Data Fig. 8c–e). Notably, analysis of proliferation in the biphenotypic population identified some cells co-positive for MKI67 and SOX4 (n = 4 of 16 proliferative cells) or KRT23 (n = 2 of 16 proliferative cells), indicating that transdifferentiation may be associated with cell division (Extended Data Fig. 8f). Together, these observations demonstrate that our single-nucleus analysis has identified factors that mark transdifferentiating cells in end-stage MASLD and could be relevant for monitoring disease progression.

PI3K–AKT signalling regulates plasticity

Interestingly, GSEA of the genes specific to biphenotypic cells indicated an enrichment in Gene Ontology terms for processes such as cell differentiation, and in KEGG terms including tight junction and PI3K–AKT signalling (Extended Data Fig. 9a). This pathway has been associated with obesity and metabolic syndrome48, both of which are tightly linked to MASLD49. To further investigate the functional importance of the PI3K–AKT pathway in the molecular mechanisms regulating cholangiocyte and hepatocyte plasticity, we decided to take advantage of intra-hepatic cholangiocyte organoids (ICOs). These cells can be grown for an extended period of time in vitro and maintain their biliary identity50 and their capacity to differentiate into cells expressing hepatocyte markers26. We first generated ICOs from end-stage MASLD livers (Supplementary Table 4). These cells expressed K19, confirming their identity (Fig. 4a, left). MASLD ICOs were then differentiated towards cells expressing hepatocyte markers, as previously described26. As expected, the resulting organoids contained cells positive for ALB (Fig. 4a, right), and quantitative PCR (qPCR) showed that they display increased expression of the genes CYP3A4, HNF4A and ALB (Fig. 4b). However, only some of the cells in an organoid become ALB+ (Fig. 4a), confirming previous observations that this process is heterogenous26. Notably, ALB+ cells were also K19+, indicating a biphenotypic identity. Expression of cholangiocyte markers K7 and K19 was also found to increase, but expression of the cholangiocyte transcription factor gene SOX9 decreased (Fig. 4b), indicating that the biliary nature of the cells was mainly maintained. Notably, we also performed differentiation using ICOs derived from healthy and end-stage MASLD livers in parallel and found no difference in their capacity for differentiation (Extended Data Fig. 9b), indicating that our culture conditions can induce cellular plasticity without disease environment. More importantly, qPCR analyses showed that several genes associated with biphenotypic cells in vivo also increased during ICO differentiation, including SOX4 and KRT23 (Extended Data Fig. 9c). Thus, ICOs differentiated in vitro provide a model for the transdifferentiation events observed in vivo (Fig. 3).

We next used ICOs to validate the importance of PI3K–AKT signalling. ICOs differentiated in the presence of the mTOR inhibitor rapamycin, the PI3K inhibitors LY294002 and copanlisib, and the AKT inhibitor MK2206 displayed a strong reduction in hepatocyte marker expression (Fig. 4c,d). Furthermore, differentiation of ICOs in the presence of the mTOR activator MHY1485 enhanced differentiation (Fig. 4e,f), indicating that this pathway can increase the expression of hepatocyte markers in cholangiocytes. Finally, inhibition of mTOR, PI3K or AKT blocked differentiation when applied at the start of the differentiation (10 days of treatment) but had less or no effect when applied from the half-way point (5 days of treatment) or just for the final 24 h, respectively (Fig. 4e and Extended Data Fig. 9d). Together these data indicate that the mTOR–PI3K–AKT pathway could be necessary for cholangiocytes to differentiate into biphenotypic cells, but not for the survival of these cells. Importantly, the PI3K–AKT–mTOR pathway is activated by insulin51, and insulin resistance is commonly associated with MASLD progression30. We measured the serum insulin levels of patients across the disease stages and observed a sharp increase in all stages compared with controls, with levels highest at the stage of cirrhosis (Extended Data Fig. 9e). Taken together, these findings indicate that increased circulating insulin during disease progression could have a key role through the PI3K–AKT–mTOR pathway in inducing plasticity in the hepatic epithelium. However, our single-cell analyses also indicated that the acquisition of plasticity is progressive and occurs only after large changes in the liver microenvironment. Thus, we hypothesized that the PI3K–AKT–mTOR pathway may be one of various pathways involved and decided to test for other pathways in vitro. We first identified FGF13 as being upregulated in biphenotypic cells (Fig. 3k), and we found that differentiation of ICOs in the presence of FGF13 caused a limited increase in hepatocyte marker expression (Extended Data Fig. 10a). We also performed differentiation in the presence of the proinflammatory cytokine TWEAK and fatty acids, because both play a role in MASLD progression52,53, and found no change in hepatocyte marker expression (Extended Data Fig. 10b,c). Finally, we observed increased expression in YAP-signalling genes in cholangiocytes and hepatocytes from end-stage livers. We therefore performed differentiation in the presence of a YAP activator and observed a strong decrease in the expression of hepatocyte marker genes (Extended Data Fig. 10d,e), indicating that the YAP–TAZ pathway could limit cholangiocyte plasticity but promote the ductular reaction, as shown in mouse studies54,55. Finally, we performed differentiation of ICOs in a matrix containing an increased amount of collagen to mimic more closely the cirrhotic liver environment. Strikingly, this change in the composition of the extracellular matrix caused organoid branching and the appearance of ALB+ cells in tubular K19+ structures (Extended Data Fig. 10f). Thus, changes in the composition of the extracellular matrix may instruct tubulogenesis, which resembles the ductular reaction, but without substantially improving transdifferentiation. Taken together, these data suggest that the acquisition of plasticity could involve complex interplays between different signalling pathways, including the YAP and PI3K–AKT–mTOR pathways.

Fig. 4: The PI3K–AKT–mTOR pathway is a key regulator of cholangiocyte-to-hepatocyte plasticity.
figure 4

a, Bright-field images and immunofluorescence staining of organoids treated with cholangiocyte organoid medium (uICO) or with differentiation medium (dICO) for ALB and K19; n = 6 patient-derived organoid lines. Scale bars: 500 μm, bright-field; 20 μm, immunofluorescence. b, mRNA expression of hepatocyte markers (ALB, CYP3A4 and HNF4A) and cholangiocyte markers (KRT19, KRT7 and SOX9) in uICOs and dICOs; n = 14 biologically independent experiments (unpaired two-tailed t-test; errors bars indicate s.e.m.). c, mRNA expression of hepatocyte markers in organoids differentiated in the presence of DMSO, copanlisib (a PI3K inhibitor), LY294002 (a PI3K inhibitor), MK2206 (an AKT inhibitor) or rapamycin (an mTOR inhibitor); n = 3 biologically independent experiments (P values indicated, ordinary one-way ANOVA, adjusted for multiple comparisons; errors bars show mean ± s.d.). d, Immunofluorescence staining for K19 and ALB in organoids differentiated in the presence of DMSO or the mTOR inhibitor rapamycin (top); bottom image shows high magnification of the area in the white box; n = 3 patient-derived organoid lines. Scale bars: 100 μm, top; 50 μm, bottom. e, Immunofluorescence staining for K19 and ALB in uICOs and dICOs treated with DMSO, an mTOR activator (MHY1485), an AKT inhibitor (MK2206) or an mTOR inhibitor (rapamycin) for the number of days indicated; n = 3 patient organoid lines. Scale bars: top, left to right: 100 μm, 40 μm, 100 μm, 70 μm, 70 μm; bottom, left to right: 40 μm, 80 μm, 100 μm, 100 μm, 100 μm. f, mRNA expression of the hepatocyte markers ALB and CYP3A4 in dICOs treated for 10 days with DMSO or an mTOR activator (MHY); n = 8 biologically independent experiments (P values indicated, two-tailed t-test; error bars indicate s.e.m.).

Source Data


Our single-cell analyses provide an advanced resource to study the factors driving disease progression. The use of snRNA-seq, as opposed to scRNA-seq, allows the unbiased capture of hepatocytes and cholangiocytes without the over-representation of immune cells, which has been reported in scRNA-seq studies. Comparison of the two approaches for human liver indicates that snRNA-seq also enhances the detection rate of rare populations56. These benefits, plus the suitability of snRNA-seq for processing frozen biopsies, made this approach suitable for our study and aims. The information contained in this dataset certainly goes beyond mechanisms of regeneration, and subsequent analyses will probably reveal new cellular activity involving more cell types. However, we decided to focus on the regenerative process because this aspect is difficult to investigate in human tissue and could have profound implications for organs targeted by progressive disorders. By combining snRNA-seq with advanced imaging of tissue, we showed that cellular plasticity between cholangiocytes and hepatocytes increases with disease progression to culminate during end-stage disease. This finding supports the results from animal studies, which have reported cholangiocyte-to-hepatocyte plasticity1,16,17,18,19. Furthermore, our analysis builds on histological observations of intermediate hepatocytes (K7-expressing hepatocyte-like cells) in MASLD28 by showing that transdifferentiating cells, and thus truly biphenotypic cells, are found mainly in end-stage liver. These biphenotypic cells are different from the hepatobiliary hybrid progenitors previously identified57, which are present only in healthy tissue. Furthermore, we could not find evidence of liver stem cells or dedifferentiation processes. However, single-cell data resolution can be a limitation and we are unable to totally exclude the existence of a rare population of adult stem cells in the liver. Such cells could hypothetically be activated by other types of injury. The resolution of our dataset was sufficient to capture cells representing the ductular reaction and intermediate hepatocytes during the early stage of the disease. Although lacking the expression of plasticity factors, these cells share a transcriptional signature with the biphenotypic cells identified in our study because they express markers of both hepatocytes and cholangiocytes. Thus, our data do not exclude the possibility that the ductular reaction or intermediate hepatocytes could represent early precursors necessary for the production of biphenotypic cells in end-stage liver. More importantly, our analyses revealed that transdifferentiation might not be a real event in regeneration. Indeed, transdifferentiating cells were observed mainly in end-stage livers, which display little function, represent a damaged environment and have a high incidence of liver cancer. Thus, although we cannot rule out a regenerative effort, the acquisition of plasticity represents a disease process, rather than a repair mechanism. This hypothesis is reinforced by the major changes occurring in the niche surrounding hepatocytes and cholangiocytes, as evidenced by the abnormal zonation, the loss of cellular identity and the aberrant remodelling of the biliary tree, all of which are difficult to associate with a healthy regenerative process. Interestingly, expression of SOX4, KLF6 and KRT23 has been associated not only with liver steatosis58,59 and biliary remodelling60,61, but also with hepatocellular carcinoma62,63. Finally, our data indicate a role for the PI3K–AKT–mTOR pathway in regulating cholangiocyte-to-hepatocyte transdifferentiation. This pathway has previously been implicated in the conversion of biliary epithelial cells to hepatocytes in zebrafish64, which may suggest that this mechanism is conserved between species. The involvement of the insulin signalling pathway in the regulation of plasticity also highlights a potential role for insulin resistance, which is commonly associated with an increased risk of cancer. Thus, the plasticity observed in the liver could reflect a broader mechanism occurring in several organs of MASLD patients with type 2 diabetes. Future work investigating the interplay of insulin resistance and cellular plasticity may address this important question. Our data also indicate that the PI3K–AKT–mTOR pathway is probably one of various pathways involved in plasticity. Thus, the acquisition of cellular plasticity in human epithelium is likely to be a disease mechanism involving multiple signals and modifications of the microenvironment over a prolonged period of time. Consequently, a deeper understanding of the signals controlling the appearance of plasticity could pave the way for the development of efficient and safe therapeutic strategies against chronic liver diseases.



Biopsy collection and processing of human samples were carried out under ethics approval by Addenbrookes Hospital REC 18/WM/0397. The study met all the UK criteria for the responsible use of human tissue. Every donor whose samples were used was offered the patient information sheet and provided informed consent. Healthy deceased transplant organ tissue and explants were taken under ethics approval by the National Research Ethics Service Committee East of England - Cambridge South (REC number REC 15/EE/152).

Tissue collection and freezing

Liver biopsies were done with ultrasound guidance using a 16 g end cut needle (Biopince). Two ultrasound-guided needle core liver biopsies of approximately 2 cm were obtained. Half of the second biopsy (1 cm) was placed in a cryo-vial and frozen immediately using liquid nitrogen. For healthy donors and explant tissue, a cube of approximately 1 cm3 was cut and frozen as above. For two healthy donors (Hl1 and HL3) and all end-stage patients, samples were taken from each of the three liver lobes (left, right and caudate), so these individuals contributed three samples to the dataset. Samples were then stored at −80 °C. Details of patient demographics and disease staging are included in Supplementary Tables 1 and 2.

Nucleus isolation

Frozen samples were transferred to a Dounce homogenizer and lysed in 1 ml lysis buffer (IGEPAL 0.1%, NaCl 10 mM, Tris-HCL pH 7.5 10 mM, MgCl2 3 mM in nuclease-free water supplemented with 0.2 U μl−1 RNasin plus). Lysis was done by performing five strokes with part A and 10–15 strokes with part B on ice, with 2 min incubation on ice between using parts A and B. After a further 2 min on ice, the sample was mixed using a P1000 by pipetting up and down ten times before a further 1 min on ice. The sample was then passed through a pre-wet 40 μm cell strainer, transferred to a 1.5 ml low-bind microfuge tube and centrifuged at 500g for 5 min at 4 °C. The pellet was resuspended in 1 ml wash buffer (Ultrapure BSA 1% in tissue-culture grade supplemented with 0.2  U  μl−1 RNasin plus) and centrifuged at 500g for 5 min at 4 °C. The pellet was resuspended in 400 μl wash buffer and transferred to a tube for FACS and kept on ice, and the sample was treated with 3 μM DAPI. FACS sorting was performed on an Influx or Aria Fusion cell sorter. Nuclei were defined by strict FSC (forward scatter) and SSC (side scatter) gating to remove debris and intact cells (larger events on the FSC). A strict singlet gate was applied and nuclei were sorted in high-purity mode with the sorter precooled. Then 20,000 DAPI-positive nuclei were sorted into a 1.5 ml microfuge tube containing 500 μl wash buffer and the tube was topped up and centrifuged at 500g for 5 min at 4 °C. The pellet was resuspended in 43 μl wash buffer and kept on ice until loading on the 10x chromium. As part of the protocol optimization, a series of lysis buffers and incubation times were tested and lysis was examined using Trypan blue and a cell counter, with efficient lysis showing more than 95% lysed cells before sorting. After sorting, nuclei were examined to ensure a single nuclei suspension of intact nuclei (nuclear membrane intact with minimal blebbing).

Single-nucleus RNA-seq

Single-nucleus RNA-seq libraries were prepared using the following: Chromium Single Cell 3′ Library and Gel Bead Kit v.3.1, Chromium Chip G Kit and Chromium Single Cell 3′ Reagent Kits v.3.1 User Guide (manual part CG000316 Rev A; 10x Genomics). One sample was run per lane of the 10x chip. For each sample, 16,000 nuclei were loaded on the Chromium instrument with the expectation of collecting gel–bead emulsions containing cell nuclei. RNA from the barcoded nuclei for each sample was subsequently reverse-transcribed in a C1000 Touch Thermal cycler (Bio-Rad) and all subsequent steps to generate single-nuclei libraries were done according to the manufacturer’s protocol with 19 PCR cycles in the cDNA amplification step. cDNA quality and quantity were measured using Agilent TapeStation 4200 (High Sensitivity 5000 ScreenTape) after which 25% of the material was used to prepare the gene-expression library. Library quality was confirmed with Agilent TapeStation 4200 (High Sensitivity D1000 ScreenTape to evaluate library sizes) and Qubit 4.0 Fluorometer (ThermoFisher Qubit dsDNA HS Assay Kit to evaluate the double-stranded DNA quantity). Each sample was normalized and pooled in equal molar concentrations. To confirm the concentration of the pool we performed qPCR using a KAPA Library Quantification Kit on QuantStudio 6 Flex before sequencing. The pool was sequenced on an Illumina NovaSeq6000 sequencer with the following parameters: 28 base pairs (bp), read 1; 10 bp, i5 index; 10 bp, i7 index; 90 bp, read 2.

FLASH imaging

FLASH was performed as described65. Samples were fixed overnight in 4% PFA at 4 °C. The sample was transferred to PBS and sliced using a vibratome to generate slices 500 μm thick. Depigmentation was performed by incubating samples in DMSO and H2O2 in PBS in a 1:1:4 (by volume) ratio overnight. The next day, samples were washed briefly in PBS and transferred to an antigen-retrieval solution. To prepare the antigen-retrieval solution, urea was dissolved in 200 mM boric acid to 250 g l−1. Zwittergent was then dissolved in the urea–borate solution to 80 g  l−1. Samples were incubated in 1 ml of the solution in a 2 ml microcentrifuge tube at room temperature for 1 h, then left overnight at 54 °C with gentle mixing on a thermo-mixer. The next day, samples were washed in PBT (0.2% Triton X-100 in PBS) three times for 1 h at room temperature before being moved to blocking buffer (1% BSA, 5% DMSO, 10% FCS and 0.2% Triton X-100) in PBS and incubated overnight at room temperature. Primary antibodies were then incubated in blocking buffer (dilution 1:100) for at least 2 nights at room temperature on a nutator. Samples were washed in PBT three times for 1 h per wash before fluorophore-conjugated secondary antibodies were added for two nights (dilution of 1:200) at room temperature on a nutator. Samples were then washed in PBS three times for 30 min per wash and passed through a dehydration series of 30%, 50%, 75% and then 2 × 100% methanol for at least 30 min in each solution, protected from light. Dehydrated samples were then gradually cleared by submerging in methyl salicylate diluted in methanol at 25%, 50%, 75% and 2 × 100% methyl salicylate for at least 30 min each in a glass dish protected from light. Cleared samples were then mounted on a glass slide in 100% methyl salicylate. Samples were imaged using an upright LSM 880 microscope, using 10× and 20× water immersion lenses.

Immunofluorescence staining of tissue slides

For all tissue-staining experiments, multiple tissue sections from at least four different patients of the relevant disease stage were analysed. Slides were dewaxed in HistoClear twice for 5 min before being washed in 100% ethanol for 5 min. Slides were then passed through a rehydration series for 5 min of 95%, 90%, 80% and 50% ethanol, then distilled water. Heat-mediated antigen retrieval was performed using 10 mM citrate (pH 6.2). The buffer was pre-warmed in a microwave until gently bubbling, before slides were submerged and heated for 15 min in the microwave on 50% power to maintain gentle bubbling. Slides were then cooled and washed twice briefly in PBS. Slides were incubated in blocking solution containing 1% (w/v) BSA, 5% (v/v) donkey serum and 0.1% (v/v) Triton X-100 for 30 min at room temperature in a humidified chamber. Primary antibodies were then diluted in blocking solution (all at 1:100 dilution except for anti-SOX4, which was used at 1:50) and incubated overnight at 4 °C in a humidified chamber. The next day, slides were washed three times for 15 min in PBS before fluorophore-conjugated secondary antibodies (1:500 dilution) plus DAPI were applied for 1 h at room temperature in the humidified chamber. Then, slides were washed three times for 15 min in PBS. Slides were mounted in one drop of DAKO fluorescent mounting medium. Slides were imaged using a Zeiss inverted 710 confocal microscope.

Organoid derivation

Tissue was stored at 4 °C in basal medium (Advanced DMEM/F12, 1% Glutamax, 1% HEPES, 1% penicillin-streptomycin) after retrieval and derivation was attempted within 24 h of tissue storage. Tissue was minced with a scalpel or scissors to small pieces of less than 1 mm3 in basal medium. The minced tissue was transferred to a 50 ml conical tube with enough digestion medium (collagenase D 2.5 mg ml−1 and DNAse I 0.1 mg ml−1 in HBSS) to fully cover it and placed in a water-bath at 37 °C for 70 min with pipetting to mix every 10 min. Cold wash medium (DMEM, 1% Glutamax, 1% FBS, 1% pen-strep) was added to stop the digestion and the sample was centrifuged at 400g for 4 min. The pellet was resuspended in 5 ml wash medium and centrifuged again as before. The resulting pellet was then resuspended in growth-factor-reduced Matrigel and plated in 50-μl domes on a 24-well plate. The plate was incubated at 37 °C for 15 min before 500 μl isolation medium was added (Advanced DMEM/F12, 1% Glutamax, 1% HEPES, 1% pen-strep, 1% B27 without vitamin A, 1% N2 supplement, 10% conditional RSPO medium, 30% WNT-conditioned medium, 25 ng ml−1 Noggin, 100 ng ml−1 FGF10, 25 ng ml−1 HGF, 50 ng ml−1 EGF, 10 mM nicotinamide 0.4 M, 10 nM gastrin, 1 mM N-acetyl cysteine, 10 μM FSK, 5 μM A8301, Noggin, 10 μM Y27632). Details of patient demographics are included in Supplementary Table 4.

Organoid culturing

After organoid derivation, the medium was changed from isolation medium to expansion medium (isolation medium without Y27632, Noggin and WNT-conditioned medium). Organoids were typically passaged every 7–10 days and the medium was changed every 2–3 days. For splitting organoids, the medium was replaced with 500 μl Cell Recovery solution (Corning). The Matrigel dome was scraped and collected using a P1000 and incubated on ice for 20 min. This was then spun at 400g for 4 min and the pellet was resuspended in basal medium using a P1000 to break up the organoids, before being centrifuged as before. The pellet was resuspended in an appropriate volume of Matrigel and plated in 50-μl domes in a 24-well plate. The plate was incubated at 37 °C for 15 min before 500 μl expansion medium was added.

Differentiation of organoids

Cholangiocyte organoids were split into expansion medium with the addition of 25 ng ml−1 BMP7 for 5 days (medium renewed every 2–3 days). Organoids were then passaged as above and plated into differentiation medium for additional 10 days and renewed every 2–3 days (Advanced DMEM/F12, 1% Glutamax, 1% HEPES, 1% pen-strep, 1% B27 without vitamin A, 1% N2 supplement, 25 ng ml−1 HGF, 50 ng ml−1 EGF, 10 nM Gastrin, 1mM N-acetyl cysteine, 0.5 μM A8301, 100 ng ml−1 FGF19, 10 μM DAPT, 3 μM dexamethasone, 25 ng ml−1 BMP7).

In vitro treatments

Cholangiocyte organoids were treated with expansion medium and BMP7 for 5 days. They were then passaged directly into differentiation medium (as above) with the addition of the small molecule of interest per condition (10 μM LY294002, 20 nM copanlisib, 1 μM MK-2206, 100 nM rapamycin, 10 μM MHY1485) for a total of 10 days. The medium was renewed every 2–3 days. For the time-course experiment, inhibitors were applied only at the time point indicated in the figure. For experiments in which organoids were cultured in increased collagen, the cell pellet was resuspended in a 50:50 mix of Matrigel and collagen I with NaOH added to neutralize the collagen before resuspending the cells. Organoids were then cultured as described above.

Immunofluorescence staining of organoids

Organoids that were planned for immunofluorescence staining were plated after splitting in a µ-Slide 8 Well High Glass Bottom (Ibidi) for better imaging quality. For staining on 3D organoid cultures, cells were washed with PBS once and then incubated with 4% PFA–PBS for 20 min at room temperature. After incubation, cells were washed three times with PBS and stored in PBS at 4 °C for up to a month. For intracellular epitopes, organoids were permeabilized using a solution of 10% donkey serum in PBS plus 0.3% Triton X-100 for at least 3 h. Cells were incubated with the primary antibody (1:100 dilution) in 1% donkey serum plus 0.1% Triton X-100 at 4 °C overnight. Cells were washed with PBS three times at room temperature for 1 h per wash. Then, cells were incubated with secondary antibody diluted 1:1000 in 1% donkey serum plus 0.1% Triton X-100 at 4 °C overnight. Cells were washed with PBS three times at room temperature for 1 h per wash. Cells were stained with Hoechst dye at 1:10,000 dilution in PBS for 30 min and washed twice. Cells were stored in PBS at 4 °C for up to a month. A Zeiss LSM 710 confocal microscope was used for imaging.

Collagen and haematoxylin-and-eosin staining

Collagen (Picro Sirius Red) staining and haematoxylin-and-eosin staining was done by the Department of Pathology at Addenbrookes Hospital in Cambridge according to their local protocol.

Statistical analysis of qPCR

Unpaired t-tests were used to perform statistical analysis on the qPCR, comparing uICOs and dICOs. One-way ANOVA adjusted for multiple comparisons was used to analyse in vitro treatments of organoids and patient insulin serum level. P values are indicated in the figure legends.

Computational methods

Sample quantification

The samples were mapped and the expression levels summarized using 10x Genomics CellRanger v.5.0.0 (ref. 66) against version GRCh38.p13 of the H. sapiens genome. To accommodate the characteristics of single-nucleus data, that is, a higher proportion of reads mapped to introns, the option ‘--include-introns’ was enabled.

Quality control

Seurat (v.4.0.3)67 objects were created considering genes expressed in more than three cells, and cells with more than 200 features expressed. Barcodes (nuclei) were excluded that had less than 1,000 or less than 800 features, or for which more than 10% of counts mapped to mitochondrial or ribosomal genes. To remove potential doublets, nuclei with more than 50,000 counts were also removed; the nCount, nFeature, %MT and %RP distributions per patient were visualized. After filtering, mitochondrial and ribosomal protein-coding genes were removed from the dataset, resulting in a dataset of 99,809 cells and 31,257 features across 47 samples.


The preprocessing of raw count matrices was performed using Seurat (v.4.0.3)67. Gene-expression values were normalized for library size using sctransform68. Principal component analysis was carried out using the top 3,000 highly variable genes. Neighbours were identified using the first 50 principal components and clustering was done using the Louvain algorithm with the 20 nearest neighbours per cell. UMAP projections were calculated using ‘RunUMAP(n.neighbors = 20, min.dist = 0.3)’. The clustering parameters used were identified by evaluating the resulting cluster stability using ClustAssess69

Annotation of cells

Expression of cell-type marker genes (Supplementary Table 3 and Extended Data Fig. 3e) was used to assign cell-type labels.

Data integration

For hepatocyte and cholangiocyte cells, some sample-specific segregation was observed within each disease stage. To alleviate potential batch effects, the data were integrated using Harmony40 with default parameters except θ (the diversity clustering penalty parameter), which was minimized such that within each disease stage, all recovered clusters included cells from each patient.

Differential expression analysis

Genes differentially expressed between cell groups were identified using the Seurat FindMarkers function. Differentially expressed genes were called on: abs(log2FC) > 0.5, Benjamini–Hochberg corrected P < 0.05 and a minimum of 25% of cells expressing the gene in the higher-expression group. GSEA of differentially expressed genes was carried out using gprofiler2 (v.0.2.0)70 using all genes detected in the compared cell groups as the background set. Enrichment was tested on the standard Gene Ontology terms, KEGG and Reactome pathway databases, and the microRNA and TF regulatory features. The Benjamini–Hochberg correction for multiple testing was applied to GSEA P values.

Cells labelled as positive for one or multiple genes are those with SCtransform-normalized expression greater than 0, per gene. The proportion of biphenotypic cells across each disease stage was compared using Welch’s t-test. Loss of zonation through disease was assessed by comparing the correlation between pairs of periportal and pericentral markers, which were then contrasted using Welch’s t-test.

RNA velocity

Velocyto (v.0.17.17), and velocyto.R (v.0.6) were used to estimate RNA velocity on the basis of the prevalence of spliced and unspliced mRNA71. Velocyto run10× was run using GRCh38.p13 annotation and repeat mask. The dataset was randomly downsampled to 20,000 cells, and cell distance was calculated as 1 minus correlation in the first 50 principal components. RNA velocity was estimated using ‘gene.relative.velocity.estimates(deltaT = 1, kCells = 20, cell.dist = D, fit.quantile = 0.2)’.

Reporting summary

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