Single-cell transcriptomics reveals anti-inflammatory effects of prednisolone on the hepatic niche in biliary atresia

Abstract

Biliary atresia (BA) is a life-threatening paediatric liver disease, characterized by bile duct obstruction, progressive inflammation and fibrosis. Treatment options are limited, often necessitating surgery or liver transplantation to prevent irreversible liver damage. Prednisolone has shown promise in treating BA; however, there is currently no consensus on its effectiveness, and the mechanism of action remains unknown. We investigated the effects of prednisolone in a BA model induced by Rhesus Rotavirus (RRV). Liver samples were collected from day-21 and day-28 post-RRV infected mice injected with PBS or steroid, and subjected to single-cell RNA sequencing (scRNA-seq) and immunohistochemistry. We identified subpopulations of hepatic stellate cells (HSCs) and macrophages that serve as key effectors of prednisolone-mediated disease amelioration. Prednisolone treatment at 28-days post RRV-infection suppressed pro-inflammatory Ly6c-high macrophages, increased immunosuppressive M2 Kupffer cells, and promoted deactivation of HSCs into the quiescent state. Computational reconstruction of HSC differentiation pathways revealed that the Nr3c1-Klf7 gene regulatory network (GRN) modulates the transition of HSCs between quiescence and activation, and we show that HSCs differentiate into three myofibroblast (MyoFbs) subpopulations, each exhibiting different pathogenic potential. We identified a pro-inflammatory MyoFB subpopulation (MFB-2) characterized by the Fosl1-Hnf1b GRN. MFB-2 secretes elevated levels of macrophage-activating cytokines, exhibits resistance to apoptotic signals, and displayed impaired responsiveness to prednisolone. The HSC and MyoFb subtypes were also identified in BA patient livers with highly conserved gene expression and functional profiles, supporting their importance in BA progression. Prednisolone treatment transforms the liver immune-mesenchymal niche from an inflammatory to regenerative state in BA and modulation of GRNs in HSCs plays a pivotal role.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-48897-2.

Introduction

Biliary atresia (BA) is a severe fibroinflammatory liver disease affecting the hepatic bile ducts resulting in obstructive bile flow. It has a higher incidence in East Asia¹, where it affects around 1 in 5000 newborns. It is the most common cause for paediatric liver transplant, often being fatal without surgical intervention. Despite biliary inflammation and subsequent liver fibrosis being well known features of the disease, the precise causes of BA remain elusive and treatments to manage disease progression are currently suboptimal. Kasai operation is the most commonly used surgical intervention. However, even after surgery, liver fibrosis may continue² and children may develop complications later in life, such as cholangitis³, highlighting the urgent need for research into the disease mechanisms and alternative therapies to improve clinical outcomes. The steroid drug prednisolone has shown promise in mitigating clinical features of the disease⁴, however there is no consensus on optimal adjuvant therapy, and the mechanism of action remains unknown.

Numerous hypotheses exist regarding the underlying causes of BA⁵, with some studies suggesting it to be initiated through injury by an environmental toxin⁶ or viral infection⁷ leading to inflammation, whereas others point towards genetic factors⁸ and/or developmental defects⁹. Regardless of the initial trigger, the blockage of bile flow in the bile ducts invariably leads to immune dysregulation and persistent inflammation, followed by excessive fibrosis and liver damage. Hepatic stellate cells (HSCs) are one of the key players in the progressive fibro-inflammatory process^10,11. They are located in the perisinusoidal space of the liver, where under healthy conditions function to store retinoids, and maintain liver homeostasis through regulation of hepatocyte metabolism, the extracellular matrix (ECM), and the local immune environment. In fibroinflammatory liver disease, HSCs are activated and thought to undergo a transformation into proliferative, fibrogenic myofibroblasts (MyoFbs) resulting in excessive ECM deposition and scaring^10,11. However, there is increasing evidence to suggest that HSCs/MyoFbs are not just passive collagen producers but are also active players in the inflammatory process and immune dysregulation in fibrotic liver diseases¹¹. Indeed, immune cells are known to infiltrate the perisinusoidal space, bringing them in close proximity with HSCs/MyoFbs. The most prominent immune infiltrates being macrophages and CD4 +/CD8 + T-cells, whose interplay with HSCs/MyoFbs can result in the formation of immune-mesenchymal niche¹², which may play a pivotal role in BA by promoting a persistent inflammatory response, inducing epithelial-mesenchymal transition (EMT) and liver fibrosis. Although pro-inflammatory HSC/MyoFb subtypes and macrophages have been previously identified in fibrotic liver disease¹³, the subtypes present in BA have not been characterized, and their involvement in disease mechanisms as well as their association with patient prognosis and treatment response remains to be elucidated.

In BA, steroids may be used as an adjuvant to reduce fibrosis, increase bile flow, and improve the rate of jaundice clearance⁴. Despite the mechanism of action being unclear, they are thought to suppress the immune response, reduce inflammation, leading ultimately to the resolution of liver fibrosis. However, despite improvements in clinical features reported after steroid treatment, not all BA patients respond well to steroid treatment¹⁴. Importantly, the effects of steroid at the molecular and cellular level remain a mystery. In this study, we performed scRNA-seq analysis of the RRV BA model (Fig. 1A), to investigate the effects of steroid treatment on the immune-mesenchymal niche within the liver.

Fig. 1.

Single cell RNA-sequencing (scRNA-seq), histological and immunohistochemical (IHC) analysis of the effects of steroid treatment. (A) Experimental design showing time points for prednisolone treatment and sample collection. (B) IHC staining of liver sections comparing the localization of markers for bile duct (CK19), endothelial cell (CD105), and (C) T-cell (CD4, CD8) and macrophages (F4/80) between prednisolone-treated (Pred) and PBS-RRV (PBS) groups. Sirius Red staining indicates fibrotic liver regions. Number of immune cells and % of fibrosis of liver sections in Pred and PBS groups were quantified and shown (mean ± SD; n = 6) in the graphs next to the immuno-stained or Sirius Red stained photos of liver sections. ****p < 0.0001 (Mann–Whitney test). (D) Uniform manifold approximation and projection (UMAP) of scRNA-seq data showing cell subpopulation clusters in Pred and PBS groups. (E) Proportion of immune versus non-immune cells. (F) Proportion of cells in each treatment group belonging to distinct subpopulation clusters for total immune cells and (G) total liver cells.

Materials and methods

Generation of RRV‑BA mouse model

RRV (rhesus rotavirus; 20 µl of 1 × 106 pfu/ml RRV, MMU 18006, ATCC^® VR-1739™) was injected to postnatal day 2 BALB/c mice (< 3 g) via peritoneal route. All the animals were bred and provided by Centre for Comparative Medicine Research, Faculty of Medicine, The University of Hong Kong, Hong Kong. All animals were monitored daily after inoculation of RRV. The development of acholic stools, and jaundice on day 5 to 6 post-injection indicated a successful induction of BA. At the day of sacrifice, mice were euthanized with an overdose of (150 mg/kg) Pharmaceutical Grade Pentobarbitone via intra-peritoneal injection.

Morphological and histological examination of RRV‑BA mouse liver

Mouse neonates were sacrificed on day 7, 14, 21 and 28 post-RRV inoculation. Morphological examination of the liver and the gall bladder was performed to look for BA phenotypes. Liver tissues were fixed in 4% paraformaldehyde (w/v) in PBS for 48 h at 4 ℃, dehydrated in graded series of alcohol, and cleared in xylene before being embedded in paraffin. Sections (6 µm in thickness) were prepared and mounted onto TESPA-coated microscope glass. Haematoxylin and eosin (H&E) staining of sections were performed for histology. Picro Sirius Red staining was performed to detect collagen fibre on liver sections according to manufacturer’s protocol (ab150681, Abcam). To quantify the degree of fibrosis, we captured images of Sirius Red-stained liver sections at 100 × magnification from five randomly selected fields. For each mouse, two liver sections were examined. The fibrotic areas in each image were outlined using the lasso tool, and the total area of all fibrotic regions was measured using the measurement log function in Photoshop. Additionally, the liver borders in each image were delineated with the lasso tool, and the total liver area was also measured. The percentage of fibrosis was calculated as the ratio of the fibrotic area to the total liver area, multiplied by 100%. Data are reported as mean ± standard deviation (SD).

Steroid treatment of RRV‑BA mice

Post-RRV-injected day 21 BA mice (n = 24) were equally divided into two groups. Mice that were allocated to steroid group received daily intraperitoneal injection of prednisolone (Redipred, Aspen, Australia; 4 mg/kg body weight) from post-RRV day 21 to day 28. The other group of BA mice received daily intraperitoneal injection of same volume of PBS from post-RRV day 21 to day 28. Details of viral strain, drug, and experiment reagents are shown in Table S1.

Immunohistochemistry

Dewax and hydration of paraffin Sects. (6 µm in thickness) was performed following a standard protocol. Antigen retrieval was performed by incubation in 10 mM sodium citrate buffer (pH 6.0) or Tris–EDTA (pH 9.0) at 95 °C for 10 min. Endogenous peroxidase activity was blocked by 10-min incubation of slides in hydrogen peroxide methanol solution (3% H₂O₂ in methanol) followed by PBS wash (3 times, 5 min each). Sections were then incubated in blocking buffer (PBS with 0.1% Triton supplemented with 1% (w/v) Bovine Serum Albumin (USB Corporation, Cleveland, OH USA)) for 1 h at room temperature. After blocking, sections were incubated with diluted antibodies diluted in blocking buffer for overnight at 4 °C. The sections were washed in PBS-T and secondary body incubation and signal development were performed using EnVision Detection Systems Peroxidase/DAB, Rabbit/Mouse system according to manufacturer’s protocol (Dako). After washing in water, sections were counterstained with haematoxylin, washed in water, incubated in 2% (W/V) NaHCO₃ for 30 s before being washed in water. Sections were dehydrated in a graded series of alcohol, cleared in xylene and then mounted in DPX mountant (BDH). Images were taken with Nikon Eclipse E600 microscope mounted with Nikon Digital Camera DXM1200F. Primary antibodies and their antigen retrieval methods were shown in Table S2. To quantify the number of immune cells on sections, we counted the immuno-positive immune cells of liver sections at 200 × magnification from five randomly selected fields. For each mouse, two liver sections were examined. The number of immune cells on each section are reported as mean ± standard deviation (SD).

Statistical analysis

Continuous variables were presented as mean ± standard derivation (SD). Data from quantitative analysis was analysed between groups on GraphPad Prism using Mann–Whitney test (nonparametric distribution and without assumption of equal SD). Exact P < 0.05 was regarded as statistical significance. Data normality was assessed using Shapiro–Wilk test.

Preparation of liver cells for single cell RNA sequencing

At the day of sacrifice, liver tissue (0.5 cm³) was dissected for single cell isolation. Liver tissues were minced in cold wash medium (Advanced DMEM/F12; 1% GlutaMAX; 1% FBS; 1% Penicillin/Streptomycin (P/S)) and transferred to a gentleMACS-C Tube (Miltenyi Biotec Inc. CA, USA). Wash medium was removed when the tissue pieces settled to the bottom of the tube. Samples were again washed with 5 ml of wash medium by pipetting up and down 10 times to remove RBCs and fat, and the medium aspired when tissue pieces settled down. 5 ml of digestion medium (Multi Tissue Dissociation Kit 1; Miltenyi Biotec Inc. CA, USA) was added and samples were incubated on gentleMACS™ Octo Dissociator (Miltenyi Biotec Inc. CA, USA) using program 37 °C-Multi-A-01. After digestion, 5 ml of cold wash medium was added to each tube, the digested sample was passed through a 70 µm strainer and then through a 30 µm strainer before centrifugation (300 g; 10 min) to pellet the cells. The supernatant was removed, and the cells were washed once with 5 ml of Advanced DMEM/F12. Red blood cells were removed from the cell pellet using Red Blood Cell Lysis Solution following manufacturer’s protocol (miltenyi 130–094–183). In brief, cell pellet was resuspended in 10 ml of red blood cells lysis buffer and incubated for 5 min at room temperature with shaking, cells were then pelleted by centrifugation (300 g; 10 min) after incubation. Dead cells were removed using Dead Cell Removal Kit (Miltenyi #130-090-101). In brief, 200 µl of microbeads (100 µl/10⁷ cells) was used to resuspend the cell pellet and cell suspension was incubated in dark for 20 min at room temperature. After incubation, 800 ul of binding buffer supplemented with 10% fetal bovine serum (FBS) was added to the cells and kept on ice. LS column (Miltenyi #130–042–401) was placed in a MACS Separator (Miltenyi #130-042-302) and then washed with 3 ml of binding buffer. Cell microbeads mixture was added to the columns on the Separator, collected the flow-through containing live cells. The column was washed with binding buffer (3 ml × 3 times), and collected the flow-through. Live cells in effluent were pelleted by centrifugation (400 g; 5 min). The cells were resuspended in 10% FBS/DPBS and counted using Invitrogen Countess II FL Automated Cell Counter.

Single-cell RNA-sequencing

Samples with more than 1 × 10⁶ cells/ml along with > 80% viability is then resuspended in 50–100 ul, 10% FBS and were submitted for 10X genomics single-cell sequencing at Centre for PanorOmic Sciences (CPOS)-The University of Hong Kong. A total of 10⁵ cells were submitted for each sample. Sequencing libraries were prepared from RNA isolated from single cell suspension using 10X Genomics Chromium Single Cell Controller platform as outlined by the 10 × Genomics Single Cell 3’v2 Reagent Kit user guide. Sequencing was performed by pair ends of 100 base pairs (PE100) on an illumina HiSeq 2500 System for those sequencing libraries passing the QC with good cDNA yield.

Further details regarding computational analysis of scRNA-seq data are provided in the supplementary information.

Results

Prednisolone induces a major shift in the transcriptional and cellular landscape

We investigated the major changes in liver and immune cells after RRV infection and steroid treatment of RRV. In PBS-treated 21-day and 28-day post-RRV, immune cell (F4/80; CD4, CD8) infiltration and accumulation was observed in the fibrotic and ductal regions of the liver, and F4/80 immune cells were observed in necrotic regions (Fig. 1B and Fig. S1). Prednisolone treatment reduced the ductular reaction, and promoted clearance of fibrosis, as evidenced by the reduction of F4/80, CD4 and CD8 immune cells, as well as reduced Sirius red staining (Fig. 1C and Fig. S1). Additionally, prednisolone resulted in an increase in the number of CD105 + cells, representing elevated endothelial cell proliferation and angiogenesis.

To further investigate the shift in cellular composition upon prednisolone treatment at high resolution, we performed scRNA-seq analysis of post-RRV livers, which after quality control filtering, resulted in 21,803 cells in PBS-treated 21-day post-RRV group (RRV-21D), 19,544 cells in PBS-treated 28-day post RRV group (RRV-28D), and 29,443 cells in prednisolone-treated 28-day post RRV group (St-28D). Uniform manifold approximation and projection (UMAP), together with unsupervised clustering, shows that the cells largely cluster together according to cell type (Fig. 1D). We calculated the differential abundance of each cell type as a proportion of total cells in each group (Fig. 1E–G). This shows that whilst immune cells comprise the majority of total cells in RRV-21D (73%) and RRV-28D (60%), they comprise only 13% of total cells in the St-28D (Fig. 1E). In terms of immune cell subsets, both RRV-21D and RRV-28D were largely dominated by macrophage, Cd8 + cytotoxic T-cell, Cd4 + T-cells, and neutrophils (Fig. 1F). In contrast, the St-28D was largely dominated by B-cells, and naïve/central memory T-cells, as well as an increase in M2 Kupffer cells (KCs) compared with RRV-21D and RRV-28D. For non-immune cells of the liver, RRV-21D has a higher proportion of HSCs/MyoFbs. Both RRV-21D and RRV-28D have a higher proportion of cholangiocytes and specific subsets of endothelial cells, including activated venous endothelial cells and interferon-activated endothelial cells (Fig. 1G). St-28D on the other hand, have a higher proportion of non-activated endothelial cells.

Prednisolone promotes KC polarization from M1-like to M2 phenotype

The shift in macrophage populations between PBS-treated RRV and prednisolone-treated RRV groups may be a key mechanism underlying the anti-fibrotic effects of prednisolone, therefore we sought to further investigate these cells. UMAP and clustering of the macrophage populations revealed six distinct subpopulations of macrophages (Fig. 2A). Notably, the Ly6c-high subpopulation was abundant in RRV-21D and RRV-28D, but absent from St-28D (Fig. 2B and Fig. S2). Differential expression analysis revealed that the Ly6c-high subpopulation was characterized by overexpression of pro-inflammatory factors linked with M1 macrophage polarization such as Ly6c2, Trem1, Mmp8, Thbs1, Slpi, and C3 (Fig. 2C). Additionally, Thbs1 and Fn1 were overexpressed Ly6c-high macrophages, and are strongly linked to liver fibrosis through their activation of mesenchymal cells¹⁵. Therefore, suppression of this subpopulation in the prednisolone-treated group may partly underly the anti-fibrotic and anti-inflammatory effects of steroid. We also identified a Cd163-expressing M2 KC subpopulation which accounted for the majority of macrophage/KCs in St-28D, and was less abundant in RRV-28D (Fig. 2B,C and Figs. S2,S3). Immunohistochemistry confirmed that Cd163 protein marker for M2 KCs is more abundant in the prednisolone-treated group (Fig. 2D). The M2 KC subpopulation strongly expressed genes associated with M2 polarization and suppression of inflammation such as Cd163, Cd5l, Timd4, Vsig4, Slc40a, and Clec4f (Fig. 2C). Several of these such as Clec4f, Cd5l, Vsigt4, and Timd4 are also known to be immunosuppressive and tolerogenic. Taken together, this indicates that the M2 KCs play an important role in mediating the effects of prednisolone in the RRV model, and suggests they are key players in the resolution of liver fibrosis in BA.

Fig. 2.

Steroid treatment induces Kupffer cell (KC) M2 polarization. (A) Uniform manifold approximation and projection (UMAP) of batch-corrected scRNA-seq gene expression data showing transcriptionally distinct KC/macrophage subtypes present in prednisolone-treated (Pred) and PBS-RRV (PBS) groups. (B) Proportion of cells from each macrophage subtype from steroid-RRV and PBS-RRV groups. (C) Top 90 differential expressed genes between KC/macrophage subtypes. (D) Immunohistochemistry staining of the Cd163 protein marker for M2 KC in the Pred and PBS groups. Number of Cd163 + cells in Pred and PBS groups were quantified and shown (mean ± SD; n = 6) in the graphs next to the immuno-stained photos of liver sections. ****p < 0.0001 (Mann–Whitney test).

Prednisolone mediates HSC quiescence and supresses pathogenic differentiation states

Given the differential abundance of the HSC/MyoFb population observed between treatment groups, we investigated the subpopulations of HSCs and MyoFbs in greater detail. Trajectory graph analysis of the scRNA-seq data was performed using Monocle3¹⁶ on the HSC/MyoFb cell population and showed a clear cell differentiation path from quiescent HSCs towards the activated HSC and MyoFb states (Fig. 3A). The analysis revealed two MyoFb subpopulations (MFB-1 and MFB-2) and one portal MyoFb population (pMFB) were present in the RRV model, each defined by distinct sets of differentially expressed genes (Fig. S4). Trajectory graph analysis showed that the MyoFb subpopulations each form a distinct branch (Fig. 3A), with each representing a separate lineage. All three lineage branches appear to be derived from activated HSCs (aHSCs), which are differentiated from a pool of quiescent HSCs (qHSCs). Pseudotime analysis showed each of the three branches having distinct gene activation signatures, with all signatures having an inverse temporal relationship with the quiescent HSC gene signature (Fig. 3C–E) indicating that the gradual repression of quiescent HSC marker is a core feature of HSC activation.

Fig. 3.

Distinct differentiation states and signalling pathway activities of hepatic stellate cells (HSCs). (A) Monocle3 trajectory graph showing HSC differentiation pathways. ‘DIHM’ refers to damage-associated intermediates between HSCs and mesothelial cells. (B) Bar chart representing the proportion of each HSC cell type derived from the prednisolone-treated (Pred) and PBS-RRV (PBS) groups. Expression dynamics of gene signatures used to define (C) MFB-1 (D) pMFB, and (E) MFB-2. Average expression was calculated along the pseudotime for each trajectory. Immunohistochemistry staining showing localization of MyoFb markers (F) S100a6, and (G) Spp1. Number of cells in Pred and PBS groups were quantified and shown (mean ± SD; n = 6) in the graphs next to the immuno-stained photos of liver sections. ****p < 0.0001 (Mann–Whitney test).

We observed that the subpopulations representing distinct differentiation states were differentially enriched between PBS-treated versus steroid-treated groups indicating that prednisolone modulates HSC differentiation (Fig. 3B). The qHSCs were comprised of cells almost exclusively from the steroid-treated RRV group demonstrating the effectiveness of prednisolone in promoting the non-fibrotic quiescent state, whereas the aHSCs and MFB-1 states were comprised of cells from both the steroid and PBS groups. pMFB and MFB-2 on the other hand, were comprised of cells almost exclusively from the PBS-RRV groups (Fig. 3B). The different lineage branches express distinct branch-specific activation markers, with the MFB-1 branch activation signature (Fig. 3C), comprising canonical pro-fibrotic HSC marker genes such as Acta2, Tagln, and the pMFB branch activation signature comprising portal fibroblast markers Fbln2 and Eln (Fig. 3D), as well as the S100a6 marker for activation (Fig. 3F), which are involved in extracellular matrix (ECM) remodelling. The MFB-2 branch comprised of activation signatures which are less typical of HSCs/MyoFbs (Fig. 3E), consisting of genes such as Spp1 and Lcn2 which encode pro-inflammatory cytokines with known profibrogenic activity. Spp1 protein was more abundant in the RRV-28D group compared with the St-28D group (Fig. 3G), confirming the enrichment of MFB-2 in RRV-28D.

Prednisolone mediates HSC deactivation through modulation of GRNs

Next, we sought to investigate the potential regulatory mechanisms responsible for the shift in HSC differentiation that we observed upon steroid treatment, as well as the role of MyoFb subtypes identified in the RRV model. Bioinformatics analysis of the scRNA-seq data identified gene regulatory networks (GRNs) with differential activity between the MFB-1, MFB-2 and pMFB branches (Fig. 4A,C and Fig. S5). We identified a network of transcription factors (TFs), termed the Nr3c1-Klf7 GRN (Fig. S5) which was active in all of the HSCs, MFB-1, and pMFB cells, but largely absent in the MFB-2 branch (Fig. 4A,C). The activity of this GRN across the trajectory graph is correlated with the expression and regulatory activity of Nr3c1 (Fig. 4A and Fig. S6), which encodes the steroid-binding glucocorticoid receptor, suggesting that Nr3c1-Klf7 GRN plays a central role in mediating the effects of prednisolone. The MFB-2 branch on the other hand, does not express Nr3c1, instead showing activation of a number of regulons constituting the Fosl1-Hnf1b GRN (Fig. 4B,C).

Fig. 4.

Functional characterisation of hepatic stellate cell (HSC) differentiation states by gene regulatory network (GRN) and pathway analysis. (A) AUCell scores of the four main regulons of the Nr3c1-Klf7 GRN across the HSC trajectory graph. Higher AUCell scores indicate increased regulon activity. (B) AUCell scores of the four main regulons of the Fosl1-Hnf1b GRN across the HSC trajectory graph. (C) Violin plots comparing AUCell scores between different HSC/MyoFb subpopulations. (D) Heatmap of average UCell pathway signature scores for subpopulations. Arrows indicate pathways regulated by the Nr3c1-Klf7 and Fosl1-Hnf1b GRNs. (E) Circos plot showing total numbers of receptor-ligand interactions between each MyoFb subtype and immune cells. Darker coloured arrows indicate more interactions. (F) Top scoring ligand-receptor interactions between MFB-2 and macrophages. Darker coloured arrows indicate higher LR score.

To investigate the possible functional roles of the branch-specific Nr3c1-Klf7, and Fosl1-Hnf1b GRNs, we performed pathway enrichment analysis (Fig. 4D) on the network TFs and their target genes. The target genes of the Nr3c1-Klf7 GRN were enriched in gene ontology signatures such as Hippo signalling, negative regulation of TGF-beta signalling, nuclear receptor (NR)-steroid signalling and chromatin remodelling. The activity of these gene signatures was high in the quiescent and pre-activated HSCs, and maintained, albeit at lower levels, in aHSCs and MFB-1. Conversely, the activity of ECM organization and integrin signalling was increased in aHSCs and MFB-1. The Hippo signalling pathway and negative regulation of TGF-beta signalling are known to play a critical role in maintaining the HSC quiescence, whereas ECM organization and integrin signalling are hallmarks of HSC activation. Taken together, this indicates that the Nr3c1-Klf7 GRN controls the transition between activated and quiescent HSC states, and this a major effector of steroid-induced deactivation of HSCs/MyoFbs.

MFB-2 is characterized by the Fosl1-Hnf1b GRN, which confers increased pathogenicity

The lack of Nr3c1 expression and inactivity of the Nr3c1-Klf7 GRN in the MFB-2 branch suggests that a loss of control over the quiescent-activation state transition in MFB-2. Instead, we observe that the Fosl1-Hnf1b GRN is the major GRN uniquely active in MFB-2 and is involved in the activation of genes involved in the negative regulation of apoptosis, inflammatory response and immune cell recruitment (Fig. 4D). We also observed that the fibrin clot formation pathway is activated in MFB-2 (Fig. 4D), which may worsen BA progression.

To investigate the pro-inflammatory MFB-2 proliferation further, we performed ligand-receptor analysis on the scRNA-seq data to functionally characterize the interactions between MyoFbs and the major immune cell types (Fig. 4E,F). This revealed MFB-2 has a substantially higher number of interactions with immune cells compared with the MFB-1 and pMFB subpopulations, and that macrophage receptors are the main interaction partners for MFB-2 ligands (Fig. 4E). The top-most significant ligand-receptor pairs between MFB-2 and macrophages (Fig. 4F) consists of pro-inflammatory cytokines such as Spp1, Lcn2, Fgb, C3, Hspa8, and Lgals1, which are secreted from MFB-2 and predicted to interact with macrophage cell surface receptors. For example, Spp1 which directly binds to multiple integrin receptors on macrophages and has been demonstrated by others to promote inflammation and fibrosis, and is linked with poor BA prognosis. Another notable example is the interaction between the amyloid precursor protein (App) expressed in MFB-2, and macrophage Lrp1 surface receptor (Fig. 4F).

Validation of HSC and MyoFb subtypes in BA patients

We next examined whether the MyoFb subtypes identified in the RRV model could also be detected in BA patients, thereby ensuring that these subtypes are not unique to the RRV mouse model and are also relevant to human BA. scRNA-seq¹⁷ data were obtained from the GEO database under accession number GSE176189. We applied the same bioinformatics approach used for identifying the MFB-2 subpopulation in our RRV model, to detect MFB-2 in the human scRNA-seq dataset. The UMAP plot (Fig. 5A) illustrates the different HSC and MyoFb subtype clusters present in this BA dataset. Consistent with our results, the BA patient data also clustered into three distinct MyoFb subtypes: MFB-1, MFB-2, and pMFB. Furthermore, the subtype-specific gene signature profiles (Fig. 5B–E) showed substantial overlap with the signature genes used to define the MyoFb subtypes present in our RRV model data (Fig. 5B–E), strongly suggesting that the MyoFb subtypes are also present and involved in human BA.

Fig. 5.

Detection of hepatic stellate cell (HSC) and myofibroblast (MyoFb) subpopulations in BA patient livers. (A) Uniform manifold approximation and projection (UMAP) of scRNA-seq data obtained from patient livers. Violin plots showing the expression profiles of the top signature genes for (B) HSCs, (C) MFB-1, (D) MFB-2, and (E) Portal MFB subpopulation clusters identified in human BA. Genes with expression profiles highly enriched in the corresponding RRV subpopulations are highlighted with stars.

Discussion

This study reveals the effects of prednisolone in BA and identifies novel pathogenic mechanisms involving specific HSC, MyoFb, and macrophage subtypes as key mediators of disease amelioration. Using scRNA-seq, we observed an expansion of Ly6c-high macrophages in mouse livers after RRV infection. Remarkably, treatment with steroid resulted in the complete ablation of this inflammatory macrophage population. Ly6c-high macrophages originate from bone marrow-derived Ly6c-high monocytes, and are known to be recruited by injured bile-duct cells in BA¹⁸. They are known to exhibit M1-like properties, secrete pro-inflammatory cytokines, promote tissue damage in response to liver injury, exacerbate chronic inflammation and have been identified in BA models previously¹⁹. The human equivalent of murine Ly6c-high macrophages, CD14 + + CD16 + monocytes, may be affected in the same manner because their functional characteristics and gene expression profiles are conserved, and both are recruited by the same chemokine receptor-2 mechanism^20,21. Therefore, the elimination of CD14 + + CD16 + monocytes in BA after steroid treatment could explain the anti-fibrotic and anti-inflammatory effects of steroid on BA patients.

Although steroid treatment drastically decreased the abundance of most monocyte-derived macrophage and Kupffer cells (KCs) in post-RRV livers, the M2 KC subpopulation showed the opposite trend, with an increase observed after treatment. M2 KCs are less well studied compared with M2 macrophages, but like M2 macrophages they are generally associated with anti-inflammatory function²², protection against fibrotic-related injury²², and liver regeneration²³, and their gene expression profile is in agreement with this. We identified genes such as Vsig4, Timd4, Clec4f, and Cd5l overexpressed in M2 KCs, which are involved in the negative regulation of T-cell activation and contribute to immune tolerance, which may limit bile duct damage, inflammation and fibrosis. M2-polarized cells may also suppress autoreactive T-cell activation²⁴, which mitigate the autoimmune-mediated bile duct destruction and fibrosis observed in some BA cases²⁵. M2 polarization also plays a crucial role in the resolution of liver fibrosis and liver regeneration, as evidenced by the overexpression of Cd163, Cd5l, Timd4, Clec4f, Vsig4, Slc40a1 observed in our RRV model. Indeed, M2-polarized cells are known to promote liver regeneration by stimulating hepatocyte proliferation²⁶, stimulate angiogenesis²⁷, and clearance of apoptotic cells²⁸. Therefore, the enrichment of M2 KCs observed after steroid treatment in our RRV model represents a crucial component of steroid-induced disease amelioration in BA.

In addition to macrophages and Kupffer cells (KCs), lymphocytes also play an important role in BA. We observed an abundance of cytotoxic CD8⁺ (cCD8⁺) T-cells and CD4⁺ T-cells in RRV-infected mouse livers, both of which were reduced following steroid treatment. This is consistent with recent studies reporting significant cytotoxic T-cell infiltration in livers of patients with BA compared with controls²⁹, as well as increased CD4⁺ T-cells in a murine BA model³⁰. cCD8⁺ T-cells are recognized as key effectors of immune-mediated cholangiocyte injury during ductular reaction^31,32, thereby promoting fibrosis and inflammation, with some cCD8⁺ T-cell subsets reported to be autoantigen‑reactive³². Accordingly, the suppression of specific T‑cell subsets by steroid treatment may constitute an important therapeutic mechanism in BA, potentially targeting the autoimmune component of the disease. Our findings pave the way for further studies to determine whether this effect is mediated directly by steroids on cCD8⁺ T-cells or instead occurs secondarily through modulation of other immune pathways.

Our analysis of other types of lymphocytes such as B-cells was less conclusive, owing partially to the relatively small size of the B-cell cluster obtained from our RRV samples resulting in limited statistical power. Although prior work has shown that B-cell tolerance defects contribute to immunoglobulin autoantibody accumulation in BA²⁹, the design of our RRV mouse experiments does not allow us to perform a comparable BA-versus-control analysis to confirm these findings. Furthermore, without detailed clinical data such as BA patient age and the stages of the disease at diagnosis, it is difficult to determine whether B-cells are primarily involved in disease initiation or instead participate at later stages of BA progression. Further studies examining the role of B-cells in BA aetiology and pathogenesis are warranted.

Apart from immune cells, HSCs play a fundamental role in BA. In response to bile duct injury and inflammatory signals, HSCs differentiate from their quiescent, homeostatic state, into a proliferative, myofibroblast-like state, which secrete large quantities of ECM components and are the main drivers of liver fibrosis in BA. We demonstrate distinct differentiation pathways from HSCs to MyoFbs, and show that steroid treatment promotes the quiescent HSC state leading to a reduction in fibrosis. The large pool of qHSCs in the steroid treated group may also promote a homeostatic environment conducive to liver repair and regeneration, a known feature of qHSCs³³, further contributing to the positive effects of steroid. Notably, some HSCs do become activated in the steroid group, albeit at much lower levels, and this activation is mainly towards the MFB-1 subtype in this group. This MFB-1 subtype may have more beneficial effects compared with the other MyoFb subtypes, as it lacks the pro-inflammatory cytokines associated with the RRV-specific MFB-2, and is instead enriched in pathways related to ECM reorganisation, calcium homeostasis, and integrin signalling, indicating a more positive effect on liver regeneration, as opposed to the inflammatory and immune activating effects of MFB-2.

The identification of the Nr3c1-Klf7 GRN, represents a novel mechanism for steroid-mediated transition between HSC quiescent and activation states. Although previous studies have shown that other members of the Krüppel-like transcription factor family such as Klf4, Klf6, and Klf10 are involved in the negative regulation of HSC activation and fibrogenesis³⁴, there have been no studies on Klf7 in the context of HSC activation/deactivation. The glucocorticoid receptor (GCR), encoded by the Nr3c1, is known to be activated by steroid binding and translocates to the nucleus, where it activates transcription. The Nr3c1-Klf7 GRN, reconstructed from the scRNA-seq data, provides compelling evidence that Nr3c1 activates Klf7, Nfib, and Tcf4. These factors function as key master regulators governing the transition between HSC activation and deactivation, through modulating signalling pathways such as Hippo signalling, TGF-Beta signalling, and those related to ECM remodelling, all of which are strongly linked with HSC activation. The Nr3c1-Klf7 GRN also modulates the expression of many genes involved in chromatin remodelling, such as histone deacetylases, SWI/SNF complex, and histone lysine demethylases, which may play an important role epigenetic regulation of HSC activation/deactivation³⁵.

The Nr3c1-Klf7 GRN was active in the MFB-1 differentiation branch as well as the HSC subpopulations, suggesting that these MyoFbs can dedifferentiate back to the quiescent state. However, the inactivity of the Nr3c1-Klf7 GRN in the MFB-2 branch suggests that these cells exhibit loss of control over the quiescent-activation state transition, suggesting that despite their reduction in the steroid group, they are not directly responsive to steroid treatment. This is further exacerbated by the Fosl1-Hnf1b GRN which activates genes in MFB-2 involved in the negative regulation of apoptosis, may increase the longevity of MFB-2 cells thus prolonging their pathogenic effects. Additionally, the Fosl1-Hnf1b GRN activates genes involved in the inflammatory response and immune cell recruitment, indicating MFB-2 is associated with increased liver inflammation and fibrosis. Although there are currently no reports of the MFB-2 subpopulation in BA, a previous study in a carbon tetrachloride-treated model of liver fibrosis identified a pro-inflammatory MyoFb subpopulation which, like MFB-2, overexpresses the pro-inflammatory receptor gene Cd74, but unlike MFB-2, overexpresses Saa3 and Dmkn. The identification of the previously unreported MFB-2 subtype, characterized by unique pro-inflammatory features in the RRV model, strongly supports its important role in BA progression. Moreover, the absence of the Nr3c1-Klf7 GRN in MFB-2 suggests impaired steroid responsiveness, which may help explain steroid resistance in some BA cases. Therapeutic induction of the Nr3c1-Klf7 GRN or suppression of Fosl1-Hnf1b GRN in HSCs/MyoFbs may present alternative approaches to BA treatment, especially for steroid non-responder patients.

The pro-inflammatory MFB-2 subtype is characterized by the Fosl1-Hnf1b GRN, which may represent a novel patho-mechanism in BA. Although there have been no studies regarding either Fosl1 and Hnf1b directly operating within HSCs/MyoFbs, they are known to play a role in other cell types involved in fibrosis³⁶. Fosl1, for instance, is a driver of EMT in various cancers, and in fibrotic liver disease it has been associated with inflammation and the induction of pro-fibrotic cytokines like TGF-β³⁷. Hnf1b has also been implicated in EMT and aberrant TGF-β signalling, but in the context of renal fibrosis³⁸. In the liver, Hnfb1 is involved in cholangiocyte and hepatocyte development, but there is little evidence for its role in HSCs/MyoFbs and liver fibrosis. Hence, Fosl1-Hnf1b GRN may represent a novel regulatory mechanism responsible for the transformation of HSCs into a pro-inflammatory pathogenic subtype of MyoFbs.

The immune-activating characteristic of MFB-2 is of high interest because immune dysregulation is known to be a common feature of BA, and typically leads to obliteration of bile ducts, causing fibroinflammatory liver damage. Our analysis of ligand-receptor interactions revealed MFB-2 as being the main initiator of the inflammatory response by secreting a large number of ligands which directly activate and recruit macrophages, which once activated, likely act as major effectors of inflammation and fibrosis. The macrophage-activating cytokines released from MFB-2, include Spp1 and App. Spp1 encodes osteopontin, a potential biomarker for BA, which is known to act as a cytokine to promote TH1 cell differentiation³⁹ and suppress T-regulatory cells⁴⁰, resulting in persistent liver inflammation in BA. However, our findings suggest an alternative role for osteopontin in BA, through the activation of macrophages via osteopontin secreted by MFB-2. Osteopontin is known to activate macrophages toward the M1 pro-inflammatory phenotype through signalling pathways such as JAK1/STAT1 pathway integrin receptors pathways⁴¹. Therefore, the MFB-2 subtype likely induces a shift in the immune system towards a highly inflammatory state.

MFB-2 also overexpresses the App gene, which encodes the amyloid precursor protein (App). Widely studied in Alzheimer’s disease, App has also been implicated in fibrotic liver diseases, including BA, where it may be secreted by hepatocytes and cholangiocytes. App serves as a biomarker for BA progression and is associated with the fibroinflammatory process⁴². Our analysis shows App produced by MFB-2, activates macrophages via the Lrp1 receptor. We therefore suggest that in addition to liver epithelial cells, MFB-2 is one of the primary sources of App in BA livers, and propose that App on the cell surface of MFB-2 cells is cleaved into amyloid-beta peptides, which interact with hepatic macrophages to promote or enhance their pro-inflammatory function in BA.

Our findings in the RRV model for BA has identified novel disease mechanisms involving MyoFB subtypes. However, as the RRV model may not completely recapitulate every aspect of BA in human, it is important to provide evidence from BA patient livers. Through the reanalysis of scRNA-seq data from a recent study¹⁷, we show that the rare pro-inflammatory MFB-2 subtype is present in human BA liver samples, together with the other MyoFb subtypes, thereby providing evidence for their clinical relevance in BA patients.

Conclusions

This study uncovered the effects of steroid treatment on the immune-mesenchymal niche in BA livers. Steroid promoted M2 polarization of KCs towards an anti-inflammatory phenotype with a positive influence on fibrosis resolution and liver regeneration. We observed a shift in the HSC population, with steroid promoting the quiescent state and ECM remodelling activity of activated HSCs, while reducing the pro-inflammatory MyoFb subpopulation. By unravelling the patho-mechanisms underlying BA progression and steroid-mediated disease amelioration, these findings will pave the way for improving BA treatment. This could lead to the development of novel BA treatment applicable for a broader range of patients, as well as enhancing patient stratification based on their immune-mesenchymal profiles.

Abbreviations

BA

Biliary atresia

ECM

Extracellular matrix

EMT

Epithelial-to-mesenchymal transition

GRN

Gene regulatory network

HSCs

Hepatic stellate cells

KC

Kupffer cell

MyoFb

Myofibroblast

RRV

Rhesus rotavirus

scRNA-seq

Single cell RNA-sequencing

TF

Transcription factor

UMAP

Uniform manifold approximation and projection

Author contributions

Conceptualization: PHYC, VCHL, CSMT, PKHT; Methodology: PHYC, VCHL, CSMT; Software: PDB; Investigation: PDB, FL, ZW; Data acquisition & analysis: PDB, FL, ZW; Writing—original draft: PDB; Writing—review & editing: VCHL, CSMT; Visualization: PDB; Supervision: PHYC, VCHL; Project administration: PHYC, VCHL, KKYW, PKHT; Funding acquisition: KKYW, VCHL PHYC, CSMT, PKHT.

Funding

Theme-based Research Scheme 2021 (T12-712/21-R); Health and Medical Research Fund (08191976).

Data availability

Primary scRNA-seq data has been deposited to the Genome Sequence Archive under accession PRJCA054904. Publicly available scRNA-seq data is available from the Gene Expression Omnibus under accession GSE176189. R and Python code will be made available on request.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

All the animal experiments were performed according to the animal ethics approvals (CULATR 5720–21) from the Faculty of Medicine, The University of Hong Kong, Hong Kong. All methods performed in the animal experiment were carried out in accordance with relevant guidelines and regulations, and in accordance with ARRIVE guidelines.