Acquisition of epithelial plasticity in human chronic liver disease

For many adult human organs, tissue regeneration during chronic disease remains a controversial subject. Regenerative processes are easily observed in animal models, and their underlying mechanisms are becoming well characterized1–4, but technical challenges and ethical aspects are limiting the validation of these results in humans. We decided to address this difficulty with respect to the liver. This organ displays the remarkable ability to regenerate after acute injury, although liver regeneration in the context of recurring injury remains to be fully demonstrated. Here we performed single-nucleus RNA sequencing (snRNA-seq) on 47 liver biopsies from patients with different stages of metabolic dysfunction-associated steatotic liver disease to establish a cellular map of the liver during disease progression. We then combined these single-cell-level data with advanced 3D imaging to reveal profound changes in the liver architecture. Hepatocytes lose their zonation and considerable reorganization of the biliary tree takes place. More importantly, our study uncovers transdifferentiation events that occur between hepatocytes and cholangiocytes without the presence of adult stem cells or developmental progenitor activation. Detailed analyses and functional validations using cholangiocyte organoids confirm the importance of the PI3K–AKT–mTOR pathway in this process, thereby connecting this acquisition of plasticity to insulin signalling. Together, our data indicate that chronic injury creates an environment that induces cellular plasticity in human organs, and understanding the underlying mechanisms of this process could open new therapeutic avenues in the management of chronic diseases.

since they are difficult to study for technical and ethical reasons.The liver is a particularly interesting organ in this context.The main functional cell types of the hepatic epithelium are (i) the hepatocytes which are known for their metabolic roles, and (ii) the cholangiocytes which line the biliary tree transporting bile acids.The process by which these cells are replaced after injury depends on the insult encountered.Cell proliferation occurs during acute liver injury [5][6][7] , however, this capacity of proliferation is abolished in chronic diseases 8,9 .Animal studies have revealed three alternative mechanisms 10,11 (i) Stem cells or progenitors can be activated and then differentiate into epithelial cells [12][13][14][15] (ii) cholangiocytes may transdifferentiate into hepatocyte or vice versa 1,16-23 and (iii) hepatocytes and cholangiocytes could reverse to a developmental progenitor to restore the corresponding cell compartment [24][25][26] .While signs of these mechanisms have been observed in human, the nature of the regenerative processes occurring during chronic liver disease remain to be fully understood [27][28][29] .To address this major question, we combined single-nuclei analyses, 3D imaging and functional experiments to study cell behaviour and regenerative processes during progression of MASLD, a chronic liver disease which is affecting a growing population of patients worldwide 30 .

snRNAseq captures liver cells across MASLD
MASLD is a progressive disease which 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 cancer 31 .We first assessed if livers affected by progressive MASLD display evidence of regenerative processes.For that, we performed immunostaining to compare healthy liver tissue sections to that of biopsies from patients at different stages of disease progression (Extended Data Fig. 1a and Fig. 1a-b).Major changes were evident especially in end-stage livers, with the expected appearance of regenerative nodules containing hepatocytes (ALB+) surrounded by large collagen deposition 32 (Extended Data Fig. 1b).Immunostaining for cholangiocyte markers keratin 7 (K7) and keratin 19 (K19) showed a strong increase in ductal structures around these nodules (Fig. 1a), termed ductal reaction 33 , commonly seen in acute and chronic liver disease 34,35 .In addition, these experiments revealed cells co-expressing K7 and hepatocyte markers ALB or HepPar1 (Fig. 1b and Extended Data Fig. 1c).These may represent intermediate hepatocytes, which have been observed histologically in human MASLD 28 .However, we also observed cells co-expressing ALB, K7, and K19 which seem to be present specifically in end stage liver (Fig. 1b and Extended Data Fig. 1d) and suggesting the presence of cells combining hepatocyte and cholangiocyte phenotypes.Importantly, such biphenotypic cells have been associated with regenerative process 32,36,37 , and thus, their appearance could be suggestive of epithelial regeneration in end stage MASLD.
To further examine the events leading to the emergence of bi-phenotypic cells and their significance in disease, we decided to study MASLD progression at the single-cell level.To this end, we collected biopsies from 47 patients 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 Table 1-2).Half of the biopsy was allocated for diagnostic/staging while the other half was rapidly frozen to be processed at a later stage (Fig. 1c).Of note, we quickly abandoned using cells isolated from fresh biopsy since many hepatocytes and cholangiocytes were lost using this method as shown by previous studies 38,39 .To bypass this limitation, we developed a protocol for nuclei isolation involving tissue lysis and FACS sorting, which allowed for the purification of high-quality nuclei even from fibrotic tissues.Using this protocol, just under 100,000 nuclei were isolated post-QC which excluded cells expressing stress markers such as mitochondria and ribosomal proteins (Fig. 1d Extended Data Fig. 2a and b).Further analyses confirmed that our method captured all 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).Of note, cell-type specific clusters mostly overlap independently of the disease stage, except for hepatocytes, which display clear transcriptional changes upon disease progression even after a batch correction of technical effects, using Harmony 40 (Fig. 1f and Extended Data Fig 3d and e).Thus, hepatocytes seem to be the cell type most affected by the disease.Finally, our single-nuclei analyses also revealed the presence of cells co-expressing hepatocyte and cholangiocyte markers.We observed the existence of cells "bridging" hepatocyte and cholangiocyte clusters and co-expressing specific markers for both cell types (Extended Data Fig. 3f).Importantly, QCs were performed to confirm that these cells were not due to doublets or RNA contamination.Together, these experiments show that our single nuclei isolation protocol is compatible with single-cell level transcriptomic analyses of liver biopsies and confirm the presence of biphenotypic cells previously associated with regenerative processes in the liver of patients 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 types.All cell types exhibited differentially expressed genes across disease progression with strong separation of cells in end-stage disease for cholangiocytes, stellate, endothelial cells observed in the UMAP space thereby indicating that disease progression impacts 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) shows a diversity of pathways upregulated during disease progression.Of particular interest, we observed major adjustments in pathways related to microenvironment such as hypoxiainducible factor I signalling and gluconeogenesis suggestive of changes in liver zonation (Extended Data Fig. 5a).In the healthy liver, hepatocytes located in different zones of the liver lobules diverge in their expression of functional markers.For example, hepatocytes closer to the central vein (pericentral) express WNT signalling genes LGR5 and AXIN2, while hepatocytes closer to the portal triad (periportal) express metabolic enzymes HAL and ASS1 41 .Accordingly, healthy hepatocytes can be clearly separated using correlation analyses for known zonation markers (Fig. 2b).However, this distinction breaks down with 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 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 video 1 and 2), suggesting a loss of zonation at the transcriptional and protein level in hepatocytes.These results reinforce previous studies 42,43 but by showing that hepatocytes acquire progressively the capacity to co-express zonation markers which are mutually exclusive in healthy liver.These observations also indicate that disease progression strongly modifies the liver microenvironment, resulting in the loss of functional zonation.
The organisation of the biliary epithelium can also be strongly affected by disease progression through a process known as ductular reaction 4445 .Accordingly, Cholangiocytes also display a strong disease signature (Extended Data Figure 4d) characterised by the increase of ductular reaction markers NCAM1 and TNFRS12A.In parallel, we observed increased numbers of bile ducts during MASLD progression (Fig. 1a) to an extent suggesting a major impact of this process 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, Keratin 7 staining of healthy tissue revealed a network of ducts which forms a branching tree-like-structure (Fig. 2d).In contrast, end-stage samples exhibited complex basket-like structures surrounding hepatocyte nodules.(Fig. 2d and Supplementary video 3 and 4).
Such structures suggest a profound remodelling of the biliary tree to an extent not suspected before.Furthermore, FLASH imaging was also performed to define the location of the bi-phenotypic cells in the diseased biliary tree.This revealed that cells co-expressing Keratin 7 and hepatocyte marker MRP2 tended to be located towards the end 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 end stage.Interestingly, the same region has been associated with hepatic stem cells by previous reports 46 (Extended Data Fig. 5d).Taken together, these results suggest that the appearance of bi-phenotypic cells could be associated with major reorganisation of biliary tree and liver microenvironment during disease progression.

Hepatocyte and cholangiocyte plasticity
Having established the presence of bi-phenotypic cells and their association in part with the abnormal organisation of the biliary tree in disease, we next focused on defining their origin by performing detailed subclustering of hepatocytes and cholangiocytes.These analyses reveal two cholangiocyte subpopulation corresponding to MUC1 expressing cholangiocytes from the larger ducts and to small cholangiocyte expressing BCL2 (Extended Data Fig. 6a-d).The small cholangiocyte population was also associated with ductal reaction markers (NCAM1 and TNFRS12A) 29 (Extended Data Fig. 6e-g), suggesting that our sampling did capture ductal reaction structures.These ductal reaction cells were more common in end stage disease, in line with what we observed on the tissue sections (Extended data Fig. 1a, d and 6j).Further, subclustering of cholangiocytes populations identified bi-phenotypic cells expressing multiple hepatocytes markers (Fig. 3a and b) in clusters 5, 9 and 1 (Fig. 3c and d).
Interestingly, the same analyses performed on hepatocytes also revealed that cluster 9 includes cells expressing multiple cholangiocytes markers.Thus, both cell types could be able to generate biphenotypic cells.Hepatocyte cluster 9 contains cells from different disease stage and thus, we decided to further subcluster this population to identify a more biphenotypic phenotype.Interestingly, hepatocytes expressing the highest level of cholangiocyte markers were more common in end-stage disease (Fig. 3e) suggesting that cell plasticity could occur with disease progression (Fig. 1b).These cells tended to express cholangiocyte markers for small and not large cholangiocytes (Extended Data Fig. 7a) and may indicate 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 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 suggest that biphenotypic cells could appear earlier in disease than initially suggested by our immunostaining analyses (Extended Data Fig. 1a).These early cells could represent intermediate cells described previously 28 which display very limited plasticity.The full phenotype and thus the capacity to generate cells with biphenotypic transcriptome seems to be acquired only towards the end stage of progression.
We then decided to define the origin of these bi-phenotypic cells.In the biphenotypic population no cells were found to co-express adult stem cell markers LGR5 and TROP2 (Extended Data Fig. 7f-i).
We also hypothesised that stem cells, by definition, should be able to self-renew.However, expression of 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).Of note, quiescence marker expression increased during disease progression in hepatocytes (Extended Data Fig. 7j), confirming regeneration by proliferation is limited in chronic injury.Together, these results suggest that bi-phenotypic cells are unlikely to originate from a stem cell population.We next interrogated our data to determine if a dedifferentiation/ re-differentiation process could be occurring in the biphenotypic cells.For that, we examined the expression of foetal liver markers AFP and SPINK1 47 .
No cells co-expressing AFP and SPINK1 were found and while 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/ developmental progenitor.Together, these results suggest that bi-phenotypic cells appear during disease progression while transdifferentiation is prominent in end-stage disease.These data do not rule out a role for ductal reaction and/or intermediate hepatocytes in this acquisition of plasticity, and that these cells may act as precursors to the biphenotypic cells which increase over time during chronic injury.

Identification of plasticity factors
We next investigated the mechanisms increasing plasticity by focusing on end stage cells since they display the highest level of markers co-expression.We generated a UMAP including only hepatocytes and cholangiocytes from end-stage disease (Fig. 3F) and then localised the biphenotypic cells from sub cluster of hepatocyte cluster 9 and sub cluster 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 suggested a bi-directionality, whereas hepatocyte-like-cholangiocytes showed a predominant direction from cholangiocytes to hepatocytes (Fig. 3h and i).Thus, transdifferentiation appears to occur in both directions.We then inferred the pseudotime to identify genes expressed specifically during transdifferentiation (Fig. 3j).This analysis revealed numerous genes up-regulated in the bi-phenotypic population (Fig. 3k) and we selected SOX4, KRT23, KLF6 and NCAM1 for further validations.Immunostaining revealed SOX4+ nuclei in end stage cholangiocytes and hepatocytes, whereas SOX4 was not observed in healthy liver (Figure 3L and (Extended data Figure 8a).Similarly, K23+ cholangiocytes were observed in end stage disease, with some cells co-expressing K19, ALB and HepPar1 (Extended data Figure 8b).Contrastingly, K23 was absent from healthy cholangiocytes.Cells co-expressing SOX4 and K23 were also observed (Extended data Figure 8a).Similar staining patterns were found for KLF6 and NCAM1, with clear increases in end stage disease found (Extended data Fig. 8c-e).Of note, analysis of proliferation in the biphenotypic population showed that some cells co-positive for MKI67 and SOX4 (n= 4 of 16 proliferative cells) or KRT23 (n= 2 of 16 proliferative cells) were observed, suggesting the transdifferentiation may be associated with cell division (Extended data Fig. 8f).Together these observations demonstrate that our single nuclei analysis has identified factors which mark transdifferentiating cells in end stage MASLD which could be relevant to monitor disease progression.

PI3K-AKT signalling regulates plasticity
Interestingly, GSEA on the genes specific to biphenotypic cells suggested an enrichment in GO terms for processes such as cell differentiation, and KEGG terms including tight junction and PI3K-AKT signalling (Extended data Fig. 9a).This pathway has been associated with obesity and metabolic syndrome 48 both of which are tightly linked to MASLD 49 .To further investigate the functional importance of the PI3K-AKT pathway in the molecular mechanisms regulating cholangiocyte/hepatocyte plasticity, we then decided to take advantage of intra-hepatic cholangiocyte organoids (ICOs).These cells can be grown for extended period of times in vitro while maintaining their biliary identity 50 and their capacity to differentiate into cells expressing hepatocytes markers 26 .
We first generated ICOs from end stage MASLD livers (Supplementary Table 4).These cells expressed cholangiocyte marker K19, confirming their identity (Fig. 4a left panels).MASLD-ICOs were then differentiated towards cells expressing hepatocytes markers as described 26 .As expected, the resulting organoids contained cells positive for ALB (Fig. 4a right panels) and display an increased expression of CYP3A4, HNF4A and ALB by qPCR (Fig. 4b).However, only some of the cells within an organoid become ALB+ (Fig. 4a) confirming previous observations that this process is heterogenous 26 .Notably, ALB+ cells were also K19+, suggesting a biphenotypic identity.Expression of cholangiocyte markers KRT7 and KRT19 were also found to increase while cholangiocyte transcription factor SOX9 decreased (Fig. 4b), indicating that the biliary nature of the cells was mainly maintained.Of note, we also performed differentiation using ICOs derived from healthy and end stage MASLD livers in parallel and found no differences in their capacity of differentiation (Extended Data Fig. 9b) suggesting that our culture conditions can induce cellular plasticity without disease environment.More importantly, qPCR analyses showed that several genes associated with bi-phenotypic cells in vivo also increased during ICOs 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 mTOR inhibitor rapamycin, PI3K inhibitors LY294002, copanlisib and AKT inhibitor MK2206 displayed a strong reduction in hepatocyte marker expression (Fig. 4c and d).Furthermore, differentiation of ICOs in the presence of mTOR activator MHY1485 enhanced differentiation (Fig. 4e and f) suggesting that this pathway can increase the expression of hepatocytes markers in cholangiocytes.Finally, inhibition of the mTOR/PI3K/AKT blocked differentiation when applied at the start of the differentiation (10d treatment), had less or no effect when applied from the half-way point (5d treatment) or just for the final 24h respectively (Fig. 4e and Extended Data Fig. 9d).Together these data suggest the mTOR/PI3K/AKT pathway could be necessary for cholangiocytes to differentiate into bi-phenotypic cells, but not for the survival of these cells.Importantly, the PI3K-AKT-mTOR pathway is activated by insulin 51 ,while insulin resistance is commonly associated with MASLD progression 30 .We measured the serum insulin levels of patients across disease stages and observed a sharp increase in all stages compared to control patients, with levels highest at the stage of cirrhosis (Extended Data Fig. 9e).Taken together this suggests that increased circulating insulin during disease progression could play a key role through the PI3K-AKT-mTOR pathway in inducing plasticity in the hepatic epithelium.
However, our single cell analyses also suggested that acquisition of plasticity is progressive and only occurs after major changes in the liver micro-environment.Thus, we hypothesised that the PI3K-AKT-mTOR pathway may be one of various pathways involved and thus we decided to test additional pathways in vitro.We first identified FGF13 as being upregulated in biphenotypic cells (Fig. 3k) and we found 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 proinflammatory cytokine TWEAK and fatty acids as both play a role in MASLD progression 52,53 and found no change in hepatocyte marker expression (Extended Data Fig. 10b and c).Finally, we observed increased expression in YAP signalling genes in cholangiocytes and hepatocytes from end stage livers.Thus, we performed differentiation in the presence of a YAP activator, but interestingly observed a strong decrease in the expression of hepatocyte marker genes (Extended Data Fig. 10d and e) thereby suggesting that YAP/TAZ pathway could limit cholangiocyte plasticity while promoting ductal reaction as shown in mouse studies 54,55 .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 ECM composition caused organoid branching and the appearance of ALB+ cells in tubular K19+ structures (Extended Data Fig. 10f).Thus, changes in ECM composition may instruct tubulogenesis resembling ductular reaction without significantly improving transdifferentiation. Taken together, these data suggest the acquisition of plasticity could involve complex interplays between different signalling including the YAP and PI3K-AKT-mTOR pathways.

Discussion
Our single cell analyses provide an advanced resource to study factors driving disease progression.
The use of snRNAseq as opposed to scRNAseq allows the unbiased capture of hepatocytes and cholangiocytes without overrepresentation of immune cells as reported in previous scRNAseq studies.
Recent comparison between the two approaches for human liver suggests snRNAseq also enhances the detection rate of rare populations 56 .This, plus the suitability of snRNAseq for processing frozen biopsies made this approach more suitable for our study and aims.The information contained in this data set certainly goes beyond mechanisms of regeneration, and subsequent analyses will likely reveal new cellular activity involving additional cell types.However, we decided to focus on regenerative process since this aspect remains challenging to investigate in human and could have profound implications for organs targeted by progressive disorders.By combining snRNAseq and advanced imaging of patient tissue, we showed that cellular plasticity between cholangiocytes/ hepatocytes increases with disease progression to culminate during end stage disease.This supports findings from animal studies which have reported cholangiocyte-to-hepatocyte plasticity 1,[16][17][18][19] .Furthermore, our analysis builds on histological observations of intermediate hepatocytes (K7 expressing hepatocytelike cells) in MASLD 28 by showing that transdifferentiating cells and thus truly biphenotypic cells are mainly found in end stage liver.These biphenotypic cells are different from the hepatobiliary hybrid progenitors recently identified 57 which are only present in healthy tissue.In addition, we could not find evidence of liver stem cells or dedifferentiation processes.However, single cell data resolution can be a limitation and we can't totally exclude the existence of a rare population of adult stem cells in the liver.Such cells could be hypothetically activated by other types of injury.On the other hand, the resolution of our data set was sufficient to capture cells representing ductal reaction and intermediate hepatocytes during the early stage of the disease.While lacking the expression of plasticity factors, these cells share a transcriptional signature with the biphenotypic cells identified in our study since they express both hepatocytes and cholangiocytes markers.Thus, our data does not exclude that ductal 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 represent a true event of regeneration.Indeed, transdifferentiating cells were mainly observed in end stage livers which display little function, represent a very damaged environment, and have a very high incidence of liver cancer.Thus, while we cannot rule out a regenerative effort, this 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/cholangiocytes evidenced by the abnormal zonation, loss of cellular identity and aberrant remodelling of the biliary tree, all which are difficult to associate with a healthy regenerative process.Interestingly, SOX4, KLF6 and KRT23 expression has been associated not only with liver steatosis 58,59 , biliary remodelling 60,61 , but also with hepatocellular carcinoma 62,63 .Finally, our data indicates a role of the PI3K-AKT-mTOR pathway in regulating cholangiocyte-to-hepatocyte transdifferentiation. Interestingly, this pathway was recently implicated in in the conversion of biliary epithelial cells to hepatocytes in zebrafish 64 , which may suggest that this mechanism is conserved between species.The involvement of the insulin signalling pathway in the regulation of plasticity also highlights the potential role for insulin resistance, which is commonly associated with an increased risk for cancer.Thus, the plasticity observed in the liver could reflect a broader mechanism occurring in several organs of MASLD patients with type II diabetes.Future work investigating the interplay of insulin resistance and cellular plasticity may address this important question.Our data also suggests the PI3K-AKT-mTOR pathway is likely one of various pathways involved in plasticity.Thus, acquisition of cellular plasticity in human epithelium is likely a disease mechanism driving by multiple signals and modifications of the microenvironment over prolonged period of time.Consequently, a deeper understanding of the signals controlling the appearance of plasticity will pave the way for the development of efficient and safe therapeutic strategies against chronic liver diseases.

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

Tissue collection and freezing
Liver biopsies were carried out under ultrasound guidance using a 16g end cut needle (Biopince).Two ultrasound guided needle core liver biopsies of approximately 2cm were obtained.Half of the second biopsy (1cm) was placed in a cryo-vial and frozen immediately using liquid nitrogen.For healthy donor and explant tissue, a cube of approximately 1cm 3 was cut and was frozen as above.For 2 healthy (Hl1 and HL3) and all end stage patients, samples were taken from each of the 3 liver lobes (left, right and caudate) and so each of those patients has 3 samples contributing to the dataset.Samples were then kept at -80.Details of patient demographics and disease staging are included in Table 1 and 2.

Nuclei isolation
Frozen samples were transferred to a dounce homogeniser and lysed in 1ml of lysis buffer (IGEPAL 0.1%, NaCl 10mM, Tris-HCL pH 7.5 10mM, MgCl2 3mM in nuclease free water supplemented with RNasin plus (0.2U/ul)).Lysis was carried out by performing 5 strokes with part 'A' and 10-15 strokes with part 'B' on ice, with a 2 minute incubation on ice in between use of 'A' and 'B'.Following a further 2 minutes on ice, the sample was mixed using a p1000 by pipetting up and down 10 times, before waiting a further 1 minute on ice.
Sample was then passed through a pre-wet 40um cell strainer, transferred to a 1.5ml low bind microfuge tube, and centrifuged at 500G for 5 min at 4C. Pellet was resuspended in 1ml of wash buffer (Ultrapure BSA 1% in tissue culture grade supplemented with RNasin plus (0.2U/ul)), and centrifuged at 500G for 5 min at 4C. Pellet was resuspended in 400ul wash buffer and transferred for tube for FACS and kept on ice and the sample was treated with 3uM DAPI.FACS sorting was performed on an Influx or Aria Fusion cell sorter.Nuclei were defined by strict FCS and SCA 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 pre-cooled.20,000 DAPI positive nuclei were sorted into a 1.5ml microfuge tube containing 500ul of wash buffer and the tube was topped up and centrifuged at 500G for 5min at 4C. Pellet was resuspended in 43ul wash buffer and kept on ice, until loading on the 10x chromium.As part of protocol optimisation, 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 over 95% lysed cells pre-sorting.Following sorting, nuclei were examined to ensure a single nuclei suspension of intact nuclei (nuclear membrane intact with minimal blebbing).

Single Nuclei RNAseq
Single-nuclei RNA-seq libraries were prepared using the following: Chromium Single Cell 3′ Library & Gel Bead Kit v3.1, Chromium Chip G Kit and Chromium Single Cell 3' Reagent Kits v3.1 User Guide (Manual Part CG000316 Rev A; 10X Genomics).One sample was run per lane of the 10X chip.For each sample 16k nuclei were loaded on the Chromium instrument with the expectation of collecting gel-beads emulsions containing nuclei cells.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 performed according to the manufacturer's protocol with 19 PCR cycles in cDNA amplification step.cDNA quality and quantity was measured with Agilent TapeStation 4200 (High Sensitivity 5000 ScreenTape) after which 25% of material was used for gene expression library preparation.Library quality was confirmed with Agilent TapeStation 4200 (High Sensitivity D1000 ScreenTape to evaluate library sizes) and Qubit 4.0 Flourometer (ThermoFisher Qubit™ dsDNA HS Assay Kit to evaluate dsDNA quantity).Each sample was normalized and pooled in equal molar concentration.To confirm concentration pool was qPCRed using KAPA Library Quantification Kit on QuantStudio 6 Flex before sequencing.Pool was sequenced on Illumina NovaSeq6000 sequencer with following parameters: 28 bp, read 1; 10 bp, i5 index; 10bp, i7 index; 90 bp, read 2.'

FLASH imaging
FLASH (fast light-microscopic analysis of antibody-stained whole organs) was performed as described (Messal et al., 2021).Samples were fixed overnight in 4% PFA at 4C.The sample was transferred to PBS and sliced using a vibratome to generate 500um thick slices.Depigmentation was performed by incubating samples in DMSO and H2O2 in PBS in a 1:1:4 (vol/vol/vol) ratio overnight.Next day samples were washed briefly in PBS and transferred antigen retrieval solution.To prepare antigen retrieval solution, urea was dissolved in 200mM boric acid to 250 g/L.Zwittergent was then dissolved in the urea-borate solution to (80g/L).Samples were incubated in 1ml of solution in a 2ml microcentrifuge tube at RT for 1h, then overnight at 54C with gentle mixing on a thermo mixer.Next day samples were washed in PBT (0.2% Triton X-100 in PBS) 3x 1h at RT, 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 3 times for 1 hour per wash before fluorophore-conjugated secondary antibodies were applied for 2 nights (dilution 1:200) at room temperature on a nutator.Samples were then washed in PBS 3 times for 30 minutes per wash and passed through a dehydration series of 30%, 50%, 75% and then 2x 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 2x 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.Imaging was performed using an upright LSM 880 microscope, using 10x and 20x water immersion lenses.

Immunofluorescent staining of tissue slides
For all tissue staining experiments, multiple tissue sections from at least 4 different patients of the relevant disease stage were analysed.Slides were dewaxed in HistoClear twice for 5 minutes before being washed in 100% ethanol for 5 minutes.Slides were then passed through a re-hydration series for 5 minutes of 95%, 90%, 80% and 50% ethanol, then dH2O.Heat-mediated antigen retrieval was performed using 10mM citrate (pH 6.2).The buffer was pre-warmed in a microwave until gently bubbling, before slides were submerged and heated for 15 minutes 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 minutes at room temperature in a humidified chamber.Primary antibodies were then diluted in blocking solution (all at 1:100 dilution except for anti-SOX4 used at 1:50) and incubated overnight at 4C in a humidified chamber.Next day, slides were washed for 3x 15 min in PBS before fluorophore-conjugated secondary antibodies (1:500 dilution) plus DAPI were applied for 1h at room temperature in the humidified chamber.Following this, slides were washed 3x 15 min in PBS.Slides were mounted in 1 drop of DAKO fluorescent mounting media.Imaging of slides was performed using a Zeiss inverted 710 confocal microscope.

Organoid derivation
Tissue was stored at 4℃ in basal media (Advanced DMEM/F12, 1% Glutamax, 1% HEPES, 1% P/S) after retrieval and derivation attempted within 24 hrs of tissue storage.Tissue was minced with a scalpel or scissors to small pieces of approximately <1 mm 3 while in basal media.The minced tissue was transferred to a 50 ml conical tube with enough digestion media (collagenase D. 2.5mg/ml and DNAse I 0.1mg/ml in HBSS) to fully cover it and placed in a water-bath at 37 ℃ for 70 minutes with pipetting to mix every 10 minutes.Cold wash media (DMEM + 1% Glutamax +1 % FBS + 1% Pen / Strep) was added to stop the digestion and the sample was centrifuged at 400 g for 4 minutes.The pellet was resuspended in 5ml wash media and centrifuged again as before.The resulting pellet was then resuspended in growth factor reduced Matrigel and plated in 50ul domes in a 24 well plate.The plate was incubated at 37C for 15 minutes before 500ul of isolation media was added (Advanced DMEM/F12, 1% Glutamax, 1% HEPES, 1% P/S, 1% B27 without vitamin A, 1% N2 supplement, 10% conditional RSPO media, 30% WNT conditioned media, 25ng/ml Noggin, 100ng/ml FGF10, 25ng/ml HGF, 50ng/ml EGF, 10mM Nicotinamide 0.4M, 10nM Gastrin, 1mM N-acetyl cysteine (NAC), 10uM FSK, 5uM A8301, Noggin, 10uM Y27632).Details of patient demographics are included in Supplementary Table 4. Extended data Figure 2 -QC showing that snRNAseq protocol generates high quality data A) Violin plots summarising the number of UMIs detected per cell (nCount), number of genes per cell (nFeature), proportions of reads incident to ribosomal genes (rp %) and mitochondrial genes (mt %) per patient.B) gradient of ncount, nfeatures, mt %, rp %, all presented on log10 scale displayed on the expressiondriven UMAP, for all samples.3B.Expression-based correlations for pericentral and periportal genes, compared within groups (pericentral vs pericentral and periportal vs periportal) or between groups (pericentral vs periportal) across disease stages.P-values corresponding to comparisons of distributions for within-group and between-group correlations are indicated.Statistical significance was calculated using two-sided Welch's t-test.Per disease stage n=66 pairwise correlations between unique pairs of genes were compared (Pericentral_Pericentral: 15, Pericentral_Periportal: 36, Periportal_Periportal: 15).Mid-point, minimum and maximum of the boxplot summary correspond to the median, first and third quartiles.The extent of the whiskers correspond to the largest/smallest value no further than 1.5*IQR from the inter-quartile range.Points beyond this range are defined as outliers and are plotted individually.C) FLASH imaging of cleared healthy and end-stage liver tissue, with staining for pan-hepatocyte marker GSTA1 and pericentral hepatocyte marker GLUL.In healthy the high magnification (yellow box and right panel) highlights a region of the central vein with a view displayed through the lumen of the vessel.In end-stage the high magnification examines one side of a hepatocyte nodule.See also supplementary videos

Figure 2 -
Figure 2 -Major changes in hepatocyte zonation and biliary tree remodelling in end stage MASLD

legends Extended data Figure 1 -
Biphenotypic cells are observed in late stages of the disease progression A) Immunofluorescent staining for ALB, K19 and K7 on tissue sections from healthy, MASLD, MASH and Cirrhosis staging.Scale bars = 50 um for Healthy MASLD and MASH panels and 20um for cirrhosis panels B) H&E and collagen staining of healthy and end stage MASLD tissue sections.Scale bars = 500um Healthy H&E, 1000um End stage H&E, 2000 um for Healthy and End stage collagen staining.C) Staining for KRT7 and HepPar1 in end stage liver.An example of a double positive cell is indicated (yellow arrow).Scale bars = 20 um.D) Immunohistochemistry staining for K19 on tissue section of the indicated disease stage.In end stage (bottom panels), a cell with hepatocyte morphology expressing low levels of K19 is indicated in the left panel (yellow arrow) and a cell negative for K19 within a duct with hepatocyte morphology is indicated in the right panel (yellow arrow).Scale bars = 50 um (4 upper panels) and 20 um (4 lower panels).n= 3 patient samples from each indicated disease stage in (a-d).

Figure 3 -Extended data figure 5 -
snRNAseq captures all liver cell types across disease progression.A) Proportions of cells, captured per patient, assigned to each cell type.B) Expression UMAPs of cell type markers corresponding to Figure 1E shown on overall UMAP.C) Heatmap of relative expression of various markers used in the annotation of the cell types.D) Overall UMAP facetted by disease stage.E) Uncorrected overall UMAP show by disease stage.F) Expression UMAPs for cholangiocyte markers KRT7 and BICC1, and hepatocyte marker ABCC2 (MRP2) on overall umap.Extended data Figure 4 -Hepatic cell types display different disease signature.Bubble plot of examples of significantly differential gene expression across disease stages with corresponding UMAPs for A) Stellate cells B) Lymphocytes C) Neutrophils D) Cholangiocytes E) Macrophages F) Endothelial.Hepatocytes and cholangiocytes are strongly affected by disease progression A) GSEA analysis of hepatocytes across disease stages.Examples of significantly enriched terms are shown for each disease stage.Benjamini-Hochberg corrected values shown.B) Statistical analysis corresponding to Figure

1 and 2 .
Scale bars = 400 um low magnifications and 200 um high magnifications (right panels).D) FLASH imaging of highlighting ductal endings in healthy and end stage samples.Yellow arrows indicate single cell endings in healthy and bulkier endings in end stage samples.Scale bars = 50 um.n= 3 healthy and 3 end stage patient tissue samples (c-d).Extended data figure 6 -snRNAseq confirms cholangiocyte diversity and ductal reaction in late-stage disease.A-B) Cholangiocyte UMAP with overlaid gradient of expression of large cholangiocyte markers MUC5B and MUC1 C-D) Cholangiocyte UMAP with overlaid gradient of expression cholangiocyte marker CFTR and small cholangiocyte marker BCL2.E) UMAP indicating disease stage of cells.F-G) Cholangiocyte UMAP with overlaid gradient of expression of ductal reaction markers TNFRS12A and NCAM1.H-I) Quantification of the proportion of cholangiocytes expressing ductal reaction markers TNFRS12A and NCAM1 across disease stages.p-values indicated (two-sided Fisher exact test).J) Immunostaining of K7 and NCAM1 in end stage tissue sections.Scale bar = 15um.Extended data figure 7 -Characterisation of biphenotypic cells suggests an absence of an adult stem or foetal progenitor population.A) Heatmap of relative expression of large cholangiocyte markers MUC1 and MUC5b and small cholangiocyte marker BCL2 across the indicated cell types.B) Cluster 9 cholangiocytes identified in figure 3d plotted as a proportion of cholangiocytes from each disease stage C) Cluster 5 cholangiocytes identified in figure 3d plotted as a proportion of cholangiocytes from each disease stage.D) Cluster 1 cholangiocytes identified in figure 3d plotted as a proportion of cholangiocytes from each disease stage.P-values indicated.(Binomial Generalized Linear Mixed-Effects Model (BOBYQA optimiser, maxfun=2e5) with patient ID as a random effect).E) Violin plots of expression of indicated hepatocyte and cholangiocyte markers comparing the hepatocyte, cholangiocyte and biphenotypic populations F) Upset plot displaying the number of biphenotypic hepatocytes co-expressing the indicated stem/ progenitor cell genes.G-I) End stage hepatocyte and cholangiocyte UMAP with overlaid gradient of expression for indicated stem cell markers.J) Heatmap of relative expression of senescence markers CDKN2A and CDKN1A in bi-phenotypic hepatocytes across disease progression.K-I) End stage hepatocyte and cholangiocyte UMAP with overlaid gradient of expression for indicated liver progenitor cell markers.Extended data figure 8 -Plasticity markers are expressed in end stage liver.A) Immunofluorescence staining for HepPar1, K7 and SOX4 (upper panel) K23, K7 and SOX4 (lower panel) in end stage tissue sections.Scale bars = 15um (lower panel) and 10um (upper panel).B) Immunofluorescence staining for HepPar1 and K23 in end stage tissue sections.Scale bar = 15um.C) Immunofluorescence staining for ALB, K19 and NCAM1 in healthy and end stage tissue sections.Yellow box indicates the region shown in higher magnification.Scale bars = 50 um upper 2 panels and 10um lower 2 panels D) Immunofluorescence staining for HepPar1 and K23 in end stage tissue sections.Scale bar = 20um.E) Immunofluorescence staining for ALB, K19 and KLF6 in healthy and end stage tissue sections.Scale bars = 30um upper 2 panels and 10um lower panel.n= 3 healthy and 3 end stage patient tissue samples (a-e).F) Upset plot displaying the number of biphenotypic hepatocytes co-expressing the indicated plasticity and proliferative genes.Extended data figure 9 -Intrahepatic cholangiocyte organoids (ICOs) differentiation provides a model to study cellular plasticity.A) GSEA analysis of cholangiocyte-like-hepatocytes and hepatocyte-like-cholangiocytes from in vivo data (Figure3) combined.Top significantly enriched terms are shown.B) qPCR of hepatocyte marker expression in uICOs and dICOs derived from cirrhotic (end stage) livers or healthy donor livers.n=17 biologically independent experiments for cirrhosis and n=6 for healthy.Errors bars indicate mean with SD.C) qPCR of KLF6, SOX4 and SERPINE1, which were identified in Figure3kas markers of biphenotypic cells, in uICOs and dICOs.n=10 biologically independent experiments.P-values indicated, two-tailed unpaired t-test.Errors bars indicate SEM) D) qPCR for hepatocyte markers dICOs treated with either DMSO, MK2206 (AKT inhibitor) or rapamycin (mTOR inhibitor) for indicated time.Untreated uICOs included as a control.For CYP3A4 expression n= 6 biologically independent experiments for DMSO, 5 for MK TP1, 4 for MK TP2 and TP3, RAPA TP1 and TP2, 3 for RAPA TP3, 5 for EM control.For ALB expression n= 6 biologically independent experiments for DMSO, 4 for MK TP1 and TP3, 3 for MK TP3, 5 for RAPA TP1, 4 for RAPA TP2, 3 for RAPA TP3, 6 for EM control.P-values are indicated, ordinary one-way ANOVA adjusted for multiple comparisons.Error bars indicate mean with SE.E) Serum insulin levels of patients diagnosed from different MASLD stages.n= 7 biologically independent patients (control), 19 (MASLD), 63 (MASH), 9 (cirrhosis), 3 (end stage).P-values indicated, ordinary one-way ANOVA adjusted for multiple comparisons.Errors bars indicate mean with SD.Extended data figure 10 -Extracellular matrix composition alters ICOs branching and differentiation.A-D) qPCR for hepatocyte and cholangiocyte marker expression in dICOs treated with wither vehicle or indicated treatment for the duration of the differentiation.n= 3 biologically independent experiments, P-values are indicated unpaired t-test.Errors bars indicate mean with SD.E) Heatmap showing relative expression of YAP signalling genes across disease stages in hepatocytes and cholangiocytes combined.F) Examples of dICOs differentiated in a mixture of 50:50 collagen I : Matrigel (lower 2 panels).Immunofluorescence staining for ALB and K19.White box indicates region shown in high magnification (lower panel).uICOs grown in 100 % Matrigel included as a control (upper panel).n= 3 patient organoid lines.Scale bars =100um upper, 200um middle and 150um lower panel.