Spatial transcriptomics supports a role for SOX4-driven signaling throughout the disease course of biliary atresia

Abstract

Background:

Biliary atresia (BA) is a neonatal fibroinflammatory cholangiopathy of infancy and the most common indication for pediatric liver transplantation. We aimed to define the molecular mechanisms responsible for differences in the rate of disease progression among children with BA.

Methods:

We performed spatial transcriptomics (ST) analysis on frozen liver tissue at transplant from 14 children: BA with survival with native liver (SNL) <2 years (BA1, n=3), BA with SNL >2 years (BA2, n=4), non-BA cholestasis (n=4), and non-diseased donors (n=3). Transcriptional signatures were compared between patient groups by tissue region (scar, hepatocyte, cholangiocyte). Findings were validated in larger patient cohorts that included BA samples at diagnosis.

Results:

ST analysis of patients with BA1 showed the most aggressive disease phenotype, characterized by reduced hepatocyte zonation, low expression of homeostatic metabolic signatures, and increased scar heterogeneity enriched for pathways including extracellular matrix remodeling, interferon response, and leukocyte activation. Notably, genes involved in SOX4 hepatocyte-to-cholangiocyte reprogramming were most enriched in patients with BA1. Liver immunohistochemistry with in situ mRNA hybridization showed that patients with BA at diagnosis had increased SOX4 quantification as compared with patients with BA at transplant. Lastly, previously published liver bulk RNA-sequencing data demonstrated higher SOX4 gene-set expression in patients with BA at diagnosis with SNL <2 years.

Conclusions:

Children with BA and worse outcomes exhibit increased SOX4 gene-set expression at diagnosis with greater loss of hepatocyte zonation and immune-driven scar heterogeneity at transplant. Further mechanistic studies are needed to determine whether SOX4-associated biliary reprogramming contributes to maladaptive reparative processes in BA.

INTRODUCTION

Biliary atresia (BA) is a neonatal fibroinflammatory obliterative cholangiopathy that leads to cholestatic liver disease and remains the most common indication for pediatric liver transplantation. ¹ While timely Kasai portoenterostomy (KPE) is critical to attempt to restore bile flow, approximately half of patients require liver transplantation (LT) within the first 2 years of life. ¹ Liver fibrosis in BA progresses more rapidly than other pediatric and adult hepatobiliary disorders, yet the precise mechanisms responsible for this difference remain unclear.

Macrophage involvement in BA has been well documented, with increased portal tract macrophages correlating with poor prognosis after KPE.^2–5 Single-cell RNA sequencing (scRNA-seq) has identified distinct hepatic macrophage subsets in patients with BA at the time of LT, ⁶ including scar-associated macrophages previously implicated in liver fibrosis. ⁷ Recent studies using scRNA-seq and spatial transcriptomics (ST) have highlighted the role of immune dysregulation, showing that impaired macrophage function, deficiency of CX3C motif chemokine receptor 1-expressing T and natural killer (NK) cells, and activation of fibrosis-associated immune pathways contribute to disease pathogenesis.^8,9 However, how these immune pathways interact with cellular apoptosis, cellular senescence, and biliary reprogramming remains largely unknown.

In addition to immune signaling, apoptosis plays a crucial role in BA pathogenesis, with early cholangiocyte apoptosis driven by synergistic interferon-γ and tumor necrosis factor-α-mediated caspase-3 activation. ¹⁰ While apoptosis is a regulated form of cell death necessary for normal development and tissue remodeling, insufficient removal of pro-inflammatory apoptotic cells can occur in chronic disease states such as BA. Evidence for a role for increased apoptosis in BA includes the finding that increased accumulation of apoptotic epithelial cells has been associated with worse biliary drainage after KPE.^10–13 In contrast to apoptosis, senescence occurs when cells enter a state of cell cycle arrest but remain metabolically active. Importantly, hepatic senescence and senescence-associated secretory phenotype emerge early in BA and continue to progress until the time of LT, predominantly affecting cholangiocytes before extending to hepatocytes. ¹⁴

Both apoptosis and cellular senescence are important processes in tissue repair; however, ongoing activation of these processes can drive aberrant wound healing programs characterized by epithelial cell reprogramming. In the liver, cellular plasticity enables transdifferentiation from hepatocytes to cholangiocytes, known as hepatobiliary metaplasia, or the reverse.^15–17 A key driver of early hepatobiliary metaplasia is SRY-box transcription factor 4 (SOX4), which causes epigenetic alterations. ¹⁸ This process is regulated by genes involved in signaling pathways associated with liver injury, including NOTCH, YAP, and TGFB, all of which are required for biliary reprogramming.^15,19,20 In contrast to SOX4, SOX9 is important in later stages of hepatobiliary reprogramming. ²¹ While elevated SOX9 has been identified in BA, ²² characterization of factors that promote early stages of hepatobiliary reprogramming in BA, particularly SOX4, has not been well established.

In the present study, we aim to define the interplay between cellular senescence, immune-driven fibrosis, and early stages of hepatobiliary reprogramming. We integrate prior scRNA-seq with ST analyses and demonstrate that children with BA and survival with native liver (SNL) <2 years had increased scar heterogeneity and the greatest loss of hepatocyte zonation. These patients with poor outcome also had increased SOX4 gene-set expression than children with BA who had SNL >2 years. These data significantly contribute to our understanding of hepatobiliary metaplasia associated with BA disease severity and lay the foundation for future mechanistic studies.

METHODS

Patient cohort

Frozen liver tissue was obtained from 16 patients from the Ann & Robert H. Lurie Children’s Hospital of Chicago tissue biorepository and embedded in optimal cutting temperature compound for Visium. This patient cohort included 4 non-diseased donor liver tissues (NL), 4 non-BA cholestatic livers at the time of transplant (Non-BA), 4 liver explants from BA patients with SNL <2 years (BA1), and 4 livers from BA patients with SNL >2 years (BA2). Additional patient slides for liver histology analyses were obtained from the pathology archives at Children’s Hospital of Colorado (n=32) for validation studies by RNA-scope. Demographics and laboratory data closest to the time of sample collection were gathered retrospectively from electronic medical records. Comparison between groups was performed by analysis of variance with Bonferroni correction for continuous variables and Pearson chi-square test for categorical variables. The study protocol conforms to the ethical guidelines of the Declaration of Helsinki and Istanbul and was approved by the Institutional Review Boards of Lurie Children’s Hospital of Chicago (IRB 2017-1221) and the University of Colorado Anschutz (IRB 22-2287). The need for informed consent was waived.

Sample preparation, immunofluorescence, and Visium library construction

RNA integrity number was assessed on a liver tissue sample from each patient group in collaboration with the Northwestern Genomics Core. All samples had an RNA integrity number between 8.6 and 9.6. Tissue samples were submitted for 10x Genomics Visium ST library construction in the Core Facility of the Carl R. Woese Institute for Genomic Biology at the University of Illinois at Urbana-Champaign. The complementary DNA amplification, final library construction, and sequencing were performed at the DNA Services Core of the Carver Biotechnology Center. The 10x Genomics protocols were followed and described in detail in Supplemental Methods, http://links.lww.com/HC9/C396. Sequencing was performed on an Illumina NovaSeq. 6000 (Illumina, San Diego, CA). Fastq.gz files were generated and demultiplexed with 10x Genomics software SpaceRanger 1.3.0. Sequencing reads were aligned to the human genome Gencode39. Data is available in the Gene Expression Omnibus (GEO) database at accession GSE338525.

Direct immunofluorescence images were obtained using the Zeiss Axiozoom V16 (Carl Zeiss, Oberkochen, Germany) microscope before permeabilization. The GE slides were stained with Abcam Alexa Fluor 647 Anti-CD8 (Abcam Alexa Fluor 647, rabbit, 1:10 dilution) and Anti-CD68 antibodies (Abcam Alexa Fluor 488, mouse, 1:50 dilution) following the 10x immunofluorescence protocol. Image capture was performed only for Alexa Fluor 488 for the original GE slides.

Analysis of ST data

Data integration, clustering, differential expression, and cell type deconvolution

Among all samples, 2 (BA1_3 and NL_4) were excluded from further analysis due to poor sample quality. Visium ST datasets were merged for each biological condition: BA1, BA2, Non-BA, and NL. Normalization and integration of transcriptomic data for each condition were performed using sample-specific (“local”) Pearson residuals and canonical correlation analysis, respectively, after testing multiple methods.^23,24

Each Visium slice was split into 3 major regions: “Scar”, “Hepatocyte”, or “Cholangiocyte” based on the frequency of hepatic stellate cells (HSCs), hepatocytes, and cholangiocyte signatures obtained from deconvolution using the largest previously published reference dataset ²⁵ (Supplemental Table S1, http://links.lww.com/HC9/C397). Cell subtypes with >20% overlap in their marker genes were combined. Cell-subtype proportions were obtained using cell2location, ²⁶ aggregated to a total proportion of hepatocytes, HSCs, and cholangiocytes, and converted to Z-scores. Spots with Z-scores below −0.5 for all 3 cell types were classified as “dead”. Differential expression for each region type was calculated with the pseudobulk approach using edgeR (v3.36.0). The total expression per gene was summed across all spots assigned to a particular region in each slice. Differential expression between biological conditions was calculated.

Clustering (Seurat, 15 dimensions and 0.6 resolution) and gene ontology enrichment analysis (Gorilla ²⁷ ) of integrated datasets for each biological condition were performed to identify spatially related spots independent of their region classification. Reference data to annotate cell types and processes was obtained from our previously published healthy liver map, ²⁵ manually curated genes from other published liver maps,^7,28 cell types from our own single-cell data from pediatric cholestatic liver tissue ⁶ (Supplemental Figure S1, http://links.lww.com/HC9/C398), KEGG pathways (hsa04210, hsa00220), and data on polyamine synthesis pathways.^29,30 The resulting gene sets were used for enrichment analysis within each spot of our ST datasets (Giotto’s hypergeometric test and PageRank enrichment scoring ³¹ ). Pearson correlations between PageRank scores across all slices in each biological condition were calculated at the spot-level, and differences between diseases were evaluated by the Wilcoxon rank-sum test. Clusters derived from integrated analyses and previously published cell signatures for stellate cells ³² were mapped onto individual tissue slices using the function FindTransferAnchors. Lastly, to assess if observed findings by patient group were secondary to processing artifact, we performed integration (SCTransform) and clustering (Seurat, 15 dimensions and 0.6 resolution) across all samples.

Metabolic pathway analysis

Spot-level metabolic fluxes were inferred using the Compass algorithm (v0.9.10.2), ³³ on sctransform log-normalized ²³ expression data of each individual spot. Differential flux analysis was performed using the pseudobulk approach, whereby average flux in the fibrotic versus hepatocyte compartments was calculated for each sample. Cholangiocyte regions were excluded from this analysis due to the low number of spots (Supplemental Figure S2, http://links.lww.com/HC9/C398). A Student's t test was used to assess differential activity between biological conditions. As Visium spots capture mixed cell transcriptional profiles, application of Compass in this context does not infer single-cell metabolic flux but instead provides spatially informed estimates of dominant metabolic programs within compartment-enriched regions. See also Supplemental Methods, http://links.lww.com/HC9/C396.

Spatial correlation and GeneSet enrichment scores

We tested the degree of expression for previously published genes associated with Sox4-driven biliary reprogramming by Katsuda et al ¹⁸ in our patient samples and performed pairwise gene–gene Pearson correlations and hierarchical clustering to identify an 8-gene signature (SOX4, SPP1, CCL2, ANXA2, S100A6, EPCAM, KRT7, CD24) that we used to define SOX4/hepatobiliary metaplasia (Supplemental Figure S3, http://links.lww.com/HC9/C398). We defined a “rank enrichment” score (see Supplemental Methods, http://links.lww.com/HC9/C396) to calculate SOX4+ enrichment for each spot and identify SOX4hi spots. We next used a Wilcoxon rank-sum test to identify genes upregulated in the slice-specific SOX4hi regions as defined by the threshold of log2FoldChange (log2FC) >1, and false discovery rate (FDR) <0.01%. ³⁴ Genes were aggregated by biological condition, resulting in 385 genes upregulated in SOX4hi regions in BA1, 1202 in BA2, 204 in normal, and 84 in Non-BA. The overlap between upregulated gene sets was determined to identify 7 mutually exclusive groups with >20 genes for which pathway enrichment analysis was performed using Hallmark, Gene Ontology Biological Process, Reactome, and KEGG pathways in MSigDB databases.

Validation cohort analysis

Liver immunohistochemistry/in situ mRNA hybridization and quantification

Validation studies by RNA-scope were performed on non-cholestatic controls (NC, n=4), cholestatic controls (CC, n=4), and patients with BA, both at diagnosis and transplant (n=12 for both). NC samples came from patients with total bilirubin <1 mg/dL or direct bilirubin <0.5 mg/dL at sample collection with diagnoses of choledochal cyst, liver nodule, or metabolic disorder. CC samples came from patients with direct bilirubin >2.5 mg/dL at sample collection and diagnoses of Alagille syndrome, panhypopituitarism, or indeterminate neonatal cholestasis.

For the detection of cytokeratin 19 (CK19)-positive cells and CYP2E1/CRP mRNA pairs or SOX4/CDKN1A mRNA pairs, we performed immunolabeling in combination with the dual chromogenic RNAScope detection kit according to the manufacturer’s protocol (Advanced Cell Diagnostics, Hayward, CA). Staining was visualized using an Aperio CS2 whole slide scanner (Leica Biosystems, Buffalo Grove, IL), and analysis was performed using custom-made (SIA) Analysis Protocol Packages using Visiopharm software to calculate positive area normalized to total section area (ratios). A zonation scoring scale of 0–5 was established and applied to sections stained with CK19 and CYP2E1/CRP mRNAs. Data analysis was performed with GraphPad Prism (Version 10). Outliers were identified using the Rout method (Q=1%). Continuous variables were compared using ANOVA with Bonferroni correction (>2 groups) or an unpaired t test (2 groups).

Gene-set variation analysis of published liver bulk RNA-sequencing data

To assess if signatures of interest identified in our ST analysis differed in patients with BA at diagnosis by SNL status across a larger patient cohort, we pulled previously published bulk liver RNA-sequencing data from 75 BA infants at diagnosis with SNL >2 years and 80 BA infants with SNL <2 years. ³⁵ Associated clinical data to define patient outcomes were obtained under an ancillary study approved by the Childhood Liver Disease Research Network (ChiLDReN), funded by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), National Institutes of Health (NIH). Gene Set Variation Analysis (GSVA) was performed using the GSVA R package ³⁶ using gene sets for stellate cells, cholangiocytes, SOX4 signature, apoptosis, central hepatocyte, portal hepatocyte, and senescence (Supplemental Table S1, http://links.lww.com/HC9/C397). The Z-scores were compared between patients with BA by SNL status using an unpaired t test (GraphPad Prism Version 10).

RESULTS

Characteristics of the patient cohort

Detailed descriptions of the clinical characteristics by ST group are found in Table 1. Among the total 14 patients included for ST analyses, there was a statistically significant difference in age, with patients in the BA1 being the youngest and BA2 being the oldest among the 4 groups in line with the criteria used to define the cohorts (mean 0.7 y in BA1, 11.8 y in BA2, 4.9 y in Non-BA, and 2.8 y in NL; p=0.03). At the time of liver transplant, the BA1 group had higher total bilirubin with a mean of 27.3 mg/dL as compared with 2.6 mg/dL in BA2 and 7.9 mg/dL in Non-BA (p<0.01). Similarly, BA1 participants had the highest direct bilirubin at a mean of 18.5 mg/dL (p<0.01). There were no statistically significant between-group differences for sex, race, aminotransferase levels, platelet counts, and international normalized ratio (INR) for groups with available clinical data. Samples across all groups had high-quality spatial data output with means of >1000 spots under tissue and >2400 genes per spot for all groups. Immunofluorescence showed diffuse staining for CD68; staining patterns were therefore not used to inform downstream ST analyses (Supplemental Figure S4, http://links.lww.com/HC9/C398).

TABLE 1.

Comparison of patient characteristics at the time of sample collection between groups

Patient variable	BA1	BA2	Non-BA	NL	p
ALT—U/L (n, SD)	205 (3, 19.0)	95.0 (4, 84.7)	742 (4, 731)	—	0.16
AST—U/L (n, SD)	378 (3, 160)	94.5 (4, 60.3)	1112 (4, 958)	—	0.10
GGT—U/L (n, SD)	61.0 (3, 38.0)	250 (4, 225)	120 (4, 194)	—	0.41
Total bilirubin—mg/dL (n, SD)	27.3 (3, 10.5)	2.63 (4, 1.30)	7.93 (4, 7.30)	—	<0.01
Direct bilirubin—mg/dL (n, SD)	18.5 (3, 6.08)	1.60 (4, 0.94)	5.90 (4, 5.59)	—	<0.01
Platelet count—1000 per μL (n, SD)	183 (3, 127)	87.0 (4, 57.7)	229 (4, 104)	—	0.17
INR (n, SD)	1.6 (2, 0.07)	1.2 (4, 0.05)	1.4 (4, 0.21)	—	0.05
Age—years at sample collection (n, SD)	0.7 (3, 0.1)	11.8 (4, 6.4)	4.9 (4, 4.4)	2.8 (3, 1.9)	0.03
Sex					0.76
Male (n)	1	1	2	—
Female (n)	2	3	2	—
Race					0.50
Asian (n)	1	0	0	—
Black (n)	0	1	1	—
White (n)	2	3	1	—
Other (n)	0	0	2	—
Spatial data output	BA1	BA2	Non-BA	NL	p
Spots under tissue (n, SD)	1291 (3, 309)	1067 (4, 214)	1182 (4, 575)	1600 (3, 566)	0.48
Median genes per spot (n, SD)	3250 (3, 712)	2938 (4, 334)	2732 (4, 337)	2456 (3, 221)	0.19

Note: Means with standard deviation (SD) are reported for continuous variables.

Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; BA, biliary atresia; BA1, BA patients with SNL <2 years; BA2, BA patients with SNL >2 years; GGT, gamma-glutamyl transferase; INR, international normalized ratio; n, number; NL, normal; SD, standard deviation.

Region-specific differences in metabolic pathways are present in biological conditions

We first identified differentially expressed processes by tissue region and biological group (Supplemental Table S2, http://links.lww.com/HC9/C399). ST data were classified into hepatocyte-rich regions, scar regions, and cholangiocyte-rich regions using semi-supervised data analysis (Supplemental Figure S2, http://links.lww.com/HC9/C398). As NL patients do not have scar regions, we focused this analysis on the 3 disease groups (Figures 1B, C). Importantly, comparison of both BA groups to non-BA showed enrichment for different metabolic processes across all regions in non-BA patients, with the greatest enrichment for mitochondrial processes in the hepatocyte region (Figure 1B). In contrast, all tissue regions in patients with BA showed increased expression of developmental processes, whereas BA cholangiocyte regions showed upregulation of cell adhesion and epithelial development (Figure 1B).

FIGURE 1.

(A) Overview of workflow for sample preparation, ST library preparation, and downstream analyses. (B, C) Enriched processes by tissue region were determined using Gprofiler. The top 10 processes with p-adjusted <0.05 are shown for each tissue region for all BA versus non-BA (B), and BA1 versus BA2 (C). (D) Additional differences in metabolic pathways within hepatocyte regions for disease groups were determined using the Compass algorithm. ³³ Abbreviations: BA, biliary atresia; BA1, BA patients with SNL <2 years; BA2, BA patients with SNL >2 years; FA, fatty acid; EC, extracellular; NES, normalized enrichment score; ST, spatial transcriptomics.

Further comparison of BA1 versus BA2 patient groups demonstrated relatively higher enrichment of metabolic processes in BA2 patients, including the greatest enrichment for lipid metabolic processes in scar regions compared with small molecule metabolic processes in hepatocyte and cholangiocyte regions (Figure 1C). In comparison to patients with BA2, the scar regions of patients with BA1 showed enrichment for processes associated with fibrosis, such as collagen formation and extracellular matrix organization (Figure 1C). In addition, BA1 hepatocyte and cholangiocyte regions also had increased expression for extracellular matrix organization and additional processes of actin cytoskeleton organization and collagen degradation, respectively (Figure 1C). Taken together, this analysis suggests metabolic dysregulation and an increase in pro-fibrotic gene expression and tissue remodeling are greatest in BA1 patients across the hepatic lobule. This finding is in line with increased disease severity of explanted livers from patients with BA who undergo LT before age 2 years, but may also provide insight into the mechanisms contributing to disease pathogenesis.

To perform a more detailed analysis of metabolic pathways within tissue regions by disease group, we applied the Compass algorithm (originally developed for scRNA-seq data) to spot-level Visium data ³³ (Figure 1D and Supplemental Figure S5, http://links.lww.com/HC9/C398). BA1 hepatocyte regions demonstrated greater enrichment for processes associated with oxidative stress, such as fatty acid oxidation, pyruvate metabolism, and reactive oxygen species detoxification when compared with hepatocyte regions of BA2 and non-BA groups^37–39 (Figure 1D). Similar enrichment for fatty acid oxidation was observed in the scar region of BA1 versus BA2, and also increased in non-BA versus BA2 (Supplemental Figure S5, http://links.lww.com/HC9/C398). These findings are in line with the known association between oxidative injury and progressive liver disease in patients with BA, particularly those without effective biliary drainage after KPE. ³⁵

Analysis of the scar regions shows increased scar heterogeneity in BA1 patients, characterized by immune signaling

We next aimed to better define the heterogeneity of the scar region among the 4 comparison groups. Within the scar regions, the level of expression for the T-regulatory cell signature was the most significant difference between groups, being slightly elevated in BA versus Non-BA (Supplemental Figure S6, http://links.lww.com/HC9/C398). BA1 patients had more extensive scar regions and a lower hepatocyte cell signature within these regions than BA2 or Non-BA (Supplemental Figures S2B and S6B, http://links.lww.com/HC9/C398). To assess differences between groups in a more unbiased manner, samples were integrated by patient group, and clustering patterns were compared between groups (Supplemental Table S3, http://links.lww.com/HC9/C400). Clustering identified 4 unique scar clusters among the integrated data from BA1 (iBA1) patients in comparison to only 1 scar cluster among iBA2, 2 in Non-BA, and none in iNL (Figure 2A and Supplemental Figure S7, http://links.lww.com/HC9/C398). In contrast, iNL had the greatest non-scar heterogeneity with 7 clusters, followed by iNon-BA (6 clusters), iBA2 (4 clusters), and iBA1 (4 clusters) (Supplemental Figure S7, http://links.lww.com/HC9/C398). Integration and clustering of all datasets demonstrated common clustering of hepatocyte, cholangiocyte, and scar spots across all patient groups, thereby supporting the biological relevance of the above findings rather than a technical artifact (Supplemental Figure S8, http://links.lww.com/HC9/C398).

FIGURE 2.

(A) Clustering of integrated datasets by BA patient groups shows greater heterogeneity of the scar region in BA1 patients than in BA2. (B) Gene ontology enrichment analysis was performed on upregulated differentially expressed genes within each scar cluster of BA patients. The top 4 most significant processes with p-adjusted <0.05 are shown for each integrated BA (iBA) scar cluster. No processes achieved statistical significance for the iBA cluster 0. (C, D) The spatial relationship of the stellate cell signature ³² and each iBA scar cluster is shown for a representative BA1 (C) and BA2 (D) sample. (E) Differentially expressed genes increased in iBA1 scar clusters are visualized across clusters in iBA1 and iBA2 samples (all genes with p-adjusted <0.05 and Log₂FC >1). In contrast to iBA1, these 8 scar-associated genes do not separate into clusters in iBA2. Abbreviations: BA, biliary atresia; BA1, biliary atresia patient group 1; BA2, biliary atresia patient group 2; iBA, integrated biliary atresia dataset; iBA1, integrated biliary atresia group 1; iBA2, integrated biliary atresia group 2; GO, Gene Ontology; DEGs, differentially expressed genes; Log₂FC, Log₂ fold change.

While increased fibrosis can be associated with patient outcome in BA,^35,40 we hypothesized that heterogeneity within the scar region itself may be associated with outcome. Overall, we demonstrate an increased proportion of spots assigned to scar region in BA1 patients compared with all other patient conditions (Supplemental Figure S2, http://links.lww.com/HC9/C398). Gene ontology enrichment analysis on differentially expressed genes upregulated within BA scar clusters showed similar enriched processes between cluster 4 in iBA1 and cluster 5 in iBA2 (Supplemental Table S4, http://links.lww.com/HC9/C401). These processes included peptide biosynthesis, protein targeting, and localization to the endoplasmic reticulum, and viral transcription (Figure 2B). Top processes within scar clusters unique to iBA1 included the developmental process (clusters 4 and 5) as well as processes of interferon-gamma signaling and regulation of leukocyte cell activation within iBA1 cluster 7. Mapping of each iBA1 and iBA2 scar cluster onto representative tissue slices from BA1 and BA2 patients showed the location of each distinct cluster within the scar (Figures 2C, D). For example, iBA1 scar cluster 5, enriched for developmental process and genes such as SPARCL1 and JAG1, was present in the center of the scar in a representative BA1 patient (Figures 2C, E). In contrast, iBA1 scar clusters 4 and 7 were more diffuse across the main scar area and high in genes encoding matrix metalloproteinases (eg, MMP7 and MMP11) and immune signaling (eg, IL7R and CD2) (Figures 2C, E). In contrast to BA1, the BA2 patient scars did not show distinct spatial regions (Figures 2D, E). Similarly, iNon-BA showed low expression of genes characteristic of the iBA1 scar clusters (Supplemental Figure S7, http://links.lww.com/HC9/C398).

BA1 patients have the greatest loss of hepatocyte zonation

Calculation of the spatial association between gene sets representative of cell types and processes by patient group showed spatial separation between the portal and central hepatocyte signatures in NL patients, as expected (Figure 3A). In contrast, portal and central hepatocyte signatures showed close spatial association in all disease groups, supporting a loss of hepatocyte zonation across the lobule (Figure 3A). The closest spatial association between portal and central hepatocyte signatures was observed in BA1 patients (BA1: r=0.85, BA2: r=0.53, Non-BA: r=0.75, NL: r=−0.06). Much of this correlation was due to the exclusion of hepatocytes from scar regions, and when only hepatocyte regions are considered, BA2 patients retained hepatocyte zonation (BA1: r=0.68, BA2: r=−0.14, Non-BA: r=0.33). This is further demonstrated by representative periportal (LBP, CRP, HAMP, CYP2A7) and pericentral (CYP3A4, CYP2E1, ADH1A, GLUL) genes across hepatocyte clusters of the integrated datasets and for representative patients in each group (Supplemental Figure S9, http://links.lww.com/HC9/C398). Among periportal genes, LBP and CRP demonstrated the highest enrichment in BA1 patients but without zonated patterning, in contrast to the clearer spatial separation observed in BA2 patients (Supplemental Figure S9, http://links.lww.com/HC9/C398). Comparison of additional cell signatures within the hepatocyte regions by patient group showed the highest enrichment for the SOX4, senescence, cholangiocyte, stellate, CD8 T cell, and conventional dendritic cell type 2 gene signatures in BA1 patients (Figure 3B). Analysis of the validation cohort by CYP2E1/CRP mRNA expression showed the greatest zonation in patients with BA at diagnosis, and this decreased among patients with BA at transplant (Figure 3C and Supplemental Figure S10, http://links.lww.com/HC9/C398).

FIGURE 3.

(A) Pearson spatial correlation was used to calculate the spatial association between gene sets representative of cell types [cholangiocyte (Cholan), stellate cells, portal hepatocyte (PortalHep), and central hepatocyte (CentralHep)] and processes (SOX4-gene signature, senescence, and apoptosis). Patients with BA1 had the greatest loss of spatial distinction between the portal and central hepatocyte signatures. (B) Cell signature analysis within the hepatocyte region of all groups showed increased SOX4, senescence, cholangiocyte, stellate, and CD8 T cell and conventional dendritic cell type 2 gene signatures in patients with BA1. (C) Further evaluation of zonation in a validation cohort using in situ mRNA hybridization with CYP2E1/CRP mRNA pair showed reduced zonation in patients with BA at transplant (Tx) versus diagnosis (Dx). Among patients with BA at diagnosis, there was no difference in the level of zonation by future survival with native liver (SNL) at age 2 years. (D) Correlation between cholangiocytes-high spots and the senescence and apoptosis gene signatures showed the highest correlation of senescence with cholangiocytes in BA2 patients, whereas correlation with the apoptosis signature was similar between BA groups. Abbreviations: BA, biliary atresia; Dx, diagnosis; mRNA, messenger RNA; SOX4, SRY-box transcription factor 4; SNL, survival with native liver; Tx, transplant.

Analysis of the cholangiocyte regions showed increased SOX4, senescence, cholangiocyte, stellate, and circulating natural killer cell gene signatures in both BA groups (Supplemental Figure S6, http://links.lww.com/HC9/C398). Furthermore, patients with BA1 demonstrated the highest correlation between cholangiocytes and apoptosis, whereas BA2 patients exhibited the highest correlation between cholangiocytes and the senescence signature (Figure 3D). Non-BA patients demonstrated a significant correlation between cholangiocytes and senescence signature but not apoptosis (Figure 3D). Taken together, these findings support the premise that greater parenchymal immune cell infiltrate, hepatobiliary injury, and disruption of the normal hepatocyte phenotype may be occurring in BA1 patients who progress more rapidly to transplant.

SOX4 pathway enrichment is greatest in BA1 patients and associated with developmental pathways

SOX4 and its downstream genes have been implicated in hepatocyte reprogramming to cholangiocytes, a process that may occur in BA.^18,41 We evaluated the spatial expression of genes associated with SOX4 hepatocyte reprogramming and observed broader spatial distribution of this pathway across multiple spatial clusters in BA1 patients (Figures 4A, C). In contrast, BA2 patients showed enrichment for SOX4 genes primarily in one cluster that was characterized by mixed hepatocyte, scar, and cholangiocyte signatures (Figures 4B, C). NL and Non-BA patients showed low overall expression of SOX4 genes (Supplemental Figure S11, http://links.lww.com/HC9/C398). Despite some variability across individual samples within each biological condition, the average rank enrichment score for the 8-gene SOX4 signature was higher in BA1 patients compared with BA2, non-BA, and healthy control patients (Figure 4D). BA1 patients exhibited a higher rank enrichment score for SOX4 in both hepatocyte regions and in scar regions than all other groups (Supplemental Figure S12, http://links.lww.com/HC9/C398). In contrast, BA2 patients only had enrichment for SOX4 in scar regions compared with Non-BA, while hepatocyte regions had similar SOX4 compared with Non-BA and NL samples.

FIGURE 4.

(A, B) Expression of SOX4 and additional genes associated with hepatocyte transdifferentiation to cholangiocytes (HTC) is shown for integrated BA1 (iBA1) and BA2 (iBA2) datasets. HTC gene expression is present in a larger number of clusters in BA1 patients (A) as compared with BA2 patients (B). (C) Visualization of 3 HTC genes is shown for representative BA1 and BA2 samples. (D) Average rank enrichment score for the 8-gene SOX4 signature is shown for all individual samples. (E) Upregulated genes defined by Log₂FC >1 and FDR <0.0001 were identified in SOX4^hi spots and generated 7 mutually exclusive gene sets with >20 genes. (F) Pathway enrichment analysis was performed on the 7 SOX4^hi gene sets and showed many pathways associated with developmental processes across gene sets. *p<0.05; **p<0.0005; ***p<5×10⁻⁶. Abbreviations: BA, biliary atresia; BA1, biliary atresia patient group 1; BA2, biliary atresia patient group 2; iBA1, integrated biliary atresia group 1; iBA2, integrated biliary atresia group 2; SOX4, SRY-box transcription factor 4; HTC, hepatocyte transdifferentiation to cholangiocytes; Log₂FC (log₂FC), Log₂ fold change; FDR, false discovery rate; SOX4^hi, high SOX4 expression.

Pathway enrichment analysis on the top 7 SOX4-high gene sets showed significant enrichment in 8 or more pathways in 6 of the 7 groups, while no pathways were enriched for BA1 exclusive genes (Figures 4E, F). Despite these gene sets being mutually exclusive, many pathways were shared across gene sets, including pathways associated with developmental processes. The top 2 SOX4-high gene sets from BA2 and BA1&BA2 spots had greater enrichment for genes involved in focal adhesion, cell migration, infection response, cell junctions, epithelial–mesenchymal transition, and epithelium development, whereas metabolic processes and the acute phase response were unique to NL SOX4-high spots. These findings suggest an association between hepatic SOX4 gene signaling and transcriptional programs present in both non-diseased and diseased liver. However, the pathways enriched in SOX4-high regions varied substantially between biological conditions, suggesting that SOX4 activity may be associated with distinct transcriptional programs depending on the surrounding microenvironment.

Higher hepatic SOX4 expression is present in BA participants at diagnosis in a validation cohort

Lastly, we performed liver immunohistochemistry in situ mRNA hybridization across a larger patient cohort, including patients with BA at the time of diagnosis, in addition to transplant (Figure 5). The proportion of positive area for both CK19 and SOX4 was highest in patients with BA at diagnosis (Figures 5A, C). Among these variables, only CK19 differed among patients with BA at diagnosis by outcome, with a higher positive area in BA patients with SNL <2 years (Supplemental Figure S10B, http://links.lww.com/HC9/C398). While the proportion of SOX4-positive area did not differ by SNL status in our small cohort, evaluation of expression for the 8-gene SOX4 signature in previously published whole liver RNA-sequencing data of 155 patients with BA at diagnosis showed higher expression in BA patients with SNL <2 years (Figure 5B). Cholangiocyte and stellate cell signatures were also increased in BA patients with SNL <2 years (Figure 5B).

FIGURE 5.

(A) Detection of CK19 and SOX4 expression was performed for a validation cohort comprised of non-cholestatic infants (NC), cholestatic non-BA controls (CC), BA patients at diagnosis (BA Dx), and BA patients at transplant (BA Tx). BA Dx patients had the highest expression of CK19 and SOX4 across all groups. (B) Analysis of previously published bulk RNA-seq data of BA patients at diagnosis using Gene Set Variation Analysis showed a significant enrichment in the Sox4 gene set among patients with survival with native liver (SNL) <2 years. There were no significant differences among pericentral and periportal hepatocyte signatures by SNL status. (C) Representative images of CK19-positive cells (brown) and SOX4 mRNA (red) are shown for normal control, BA Tx, and BA Dx groups by outcome, with red arrowheads pointing to SOX4-positive areas. Scale bar is 20 μm. Abbreviations: BA, biliary atresia; BA Dx, biliary atresia at diagnosis; BA Tx, biliary atresia at transplant; CC, cholestatic non-biliary atresia controls; CK19, cytokeratin 19; GSVA, Gene Set Variation Analysis; mRNA, messenger RNA; NC, non-cholestatic infants; RNA-seq, RNA sequencing; SNL, survival with native liver; SOX4, SRY-box transcription factor 4.

DISCUSSION

In the present study, we leveraged scRNA-seq and ST to identify important cellular and spatial signatures and pathways associated with more rapid disease progression in children with BA. Specifically, BA1 patients with SNL<2 years demonstrated increased metabolic dysregulation and scar heterogeneity, loss of hepatocyte zonation, and elevated SOX4 pathway expression, suggesting more aggressive tissue remodeling and transdifferentiation, yet failed regeneration associated with early disease progression compared with BA2 and non-BA patients. Importantly, follow-up studies in our validation cohort support SOX4 as a potential marker associated with maladaptive hepatocyte reprogramming and poor prognosis.

We show that reduced expression for metabolic processes important in homeostasis was accompanied by enrichment of fibrotic, immune, and developmental pathways across tissue regions in BA1 patients. Furthermore, hepatocyte zonation, which is critical for metabolic compartmentalization, was most disrupted in the livers of BA1 patients, as evidenced by the spatial overlap between periportal and pericentral hepatocyte gene signatures. This is in contrast to prior data from more than 3 decades ago showing the persistence of metabolic zonation in the hepatocytes of children with BA and compensated cirrhosis. ⁴² In addition, enrichment for the SOX4 gene signature was highest in hepatocyte regions of BA1 patients, suggesting hepatocyte reprogramming may play a role in reduced hepatocyte zonation. The process of hepatocyte-to-cholangiocyte reprogramming has previously been described in murine models and human BA tissues. ⁴³ In fact, recent research has used scRNA-seq technology to demonstrate the trajectory of hepatocyte-to-cholangiocyte reprogramming in BA. ⁴¹ In contrast, our study is the first to implicate SOX4 as a central regulator of hepatobiliary reprogramming in human BA. SOX4-high regions were widespread in the liver tissues of BA1 patients and associated with upregulation of pathways related to epithelial–mesenchymal transition, cell junction assembly, and infection response, suggesting that SOX4 enrichment may mark a maladaptive regenerative program associated with failed restoration of ductal patency. Specifically, the data from a validation cohort of patients with BA at diagnosis with available bulk liver RNA-sequencing data showed that higher cholangiocyte and SOX4 pathway expression was associated with shorter SNL. As SOX4 is known to interact with SOX9 during development and ductal plate remodeling,^44,45 this pathway deserves attention to further define its role in the disease pathogenesis of BA.

The pro-fibrotic gene expression and tissue remodeling across the hepatic lobule seen in BA1 patients in our study are consistent with prior studies showing a link between increased fibrotic gene expression and patient outcomes. ³⁵ More specifically, the BA1 hepatocyte region had increased expression for actin cytoskeleton organization, which aligns with prior data showing a negative correlation between SNL and expression of α-smooth muscle actin on liver biopsy samples of patients with BA after KPE. ⁴⁶ Another study reported that the disappearance of periductal α-smooth muscle actin expression after successful KPE was associated with reduced progression of fibrosis, collagen accumulation, and serum levels of bile acids and bilirubin. ⁴⁷ In addition to an overall increase in fibrotic pathways, we demonstrate marked heterogeneity in the scar regions of BA1 patients compared with all other patient groups. Scar heterogeneity in BA1 patients was characterized by increased expression of immune pathways similar to prior findings in BA. ⁹ However, our study uniquely includes sub-groups of patients with BA and other non-BA cholestatic diseases to suggest that immune-driven expansion of the scar region may be most characteristic of patients with BA and rapid disease progression after KPE. These findings support the need for ongoing research directed at interrupting the early pro-fibrotic pathways across the hepatic lobule to potentially improve patient outcomes.

One of the potential triggers for persistent immune-driven liver injury may be secondary to increased levels of oxidative stress. Prior evidence also suggests the positive correlation of oxidative damage with BA liver inflammation and cirrhosis. ⁴⁸ In line with this mechanism, we show that BA1 hepatocyte regions have greater enrichment for processes associated with oxidative stress when compared with hepatocyte regions of BA2 and non-BA groups. Ongoing tissue injury can prevent beneficial wound healing and lead to unresolved fibrosis, as evidenced by enrichment for genes involved in extracellular matrix remodeling in BA1 patients. Among genes involved in extracellular matrix remodeling, we show that matrix metalloproteinase-7 (MMP7) is centrally located within the scar region of BA1 patients. While prior studies support the diagnostic role for serum MMP7 levels, ⁴⁹ our study supports the hepatic scar region as the source of systemic MMP7.

Despite the novelty of our approaches, the findings of this study need to be considered within the context of some limitations. First, the relatively small cohort size may have limited our statistical power to identify statistically significant differences for between-group comparisons in our ST analyses. Second, the cross-sectional nature of liver samples obtained either at diagnosis or the time of LT limits our ability to perform temporal analysis of disease progression. Moreover, the differences in bilirubin and age between the BA1 and BA2 cohorts at the time of transplant may confound the interpretation of our findings. Patients with BA1 exhibited more severe cholestatic liver injury that may impair hepatocyte metabolism, and the younger age of these patients may also contribute to differences in developmental and metabolic transcriptional programs as compared with the liver from older patients. ⁵⁰ Although validation analyses support an association between SOX4 pathway activation and poor clinical outcome independent of age at transplant, future studies are warranted to fully distinguish age-related developmental biology from disease-specific maladaptive reparative responses. In addition, the clinical heterogeneity of the non-BA cholestatic group in terms of diagnosis may have contributed to the variability seen in some analyses. Although we observed robust enrichment for developmental and fibrotic pathways, the current study did not incorporate mechanistic validation, which will be important to establish causal relationships in future work. Lastly, because ST analyses were performed on explant tissues at a single time point representing end-stage disease, these findings are correlative and do not establish whether SOX4 pathway activation is causative in fibrosis progression or instead reflects a downstream response to ongoing cholestatic injury.

Taken together, our study demonstrates that children with BA and rapid disease progression (ie, SNL <2 y) exhibit greater metabolic dysregulation with loss of hepatocyte zonation in addition to increased scar heterogeneity characterized by immune activation. Importantly, we identified SOX4 as a potential biomarker associated with maladaptive hepatobiliary reprogramming and poor prognosis. These findings emphasize the paramount importance of the hepatic microenvironment and cellular reprogramming in the disease progression of BA. Future studies will be necessary to define whether and how these pathways contribute to hepatobiliary injury and may help identify new treatment modalities to prolong SNL in children with BA.

Supplementary Material

hc9-10-e00992-s001.docx ^{(86.9KB, docx)}

hc9-10-e00992-s002.xlsx ^{(80.4KB, xlsx)}

hc9-10-e00992-s003.pdf ^{(3.2MB, pdf)}

hc9-10-e00992-s004.xlsx ^{(15.2KB, xlsx)}

hc9-10-e00992-s005.xlsx ^{(1.4MB, xlsx)}

hc9-10-e00992-s006.xlsx ^{(82.5KB, xlsx)}