Single-cell transcriptomics shows dose-dependent disruption of hepatic zonation by TCDD in mice

Abstract

2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) dose-dependently induces the development of hepatic fat accumulation and inflammation with fibrosis in mice initially in the portal region. Conversely, differential gene and protein expression is first detected in the central region. To further investigate cell-specific and spatially resolved dose-dependent changes in gene expression elicited by TCDD, single-nuclei RNA sequencing and spatial transcriptomics were used for livers of male mice gavaged with TCDD every 4 days for 28 days. The proportion of 11 cell (sub)types across 131 613 nuclei dose-dependently changed with 68% of all portal and central hepatocyte nuclei in control mice being overtaken by macrophages following TCDD treatment. We identified 368 (portal fibroblasts) to 1339 (macrophages) differentially expressed genes. Spatial analyses revealed initial loss of portal identity that eventually spanned the entire liver lobule with increasing dose. Induction of R-spondin 3 (Rspo3) and pericentral Apc, suggested dysregulation of the Wnt/β-catenin signaling cascade in zonally resolved steatosis. Collectively, the integrated results suggest disruption of zonation contributes to the pattern of TCDD-elicited NAFLD pathologies.

The spatial organization of the different liver cell types is essential for normal function. The adult liver consists of at least 11 predominant cell types that includes hepatocytes, cholangiocytes, hepatic stellate cells (HSCs), resident macrophages (aka Kupffer cells), and liver sinusoidal endothelial cells (LSECs) (Halpern et al., 2017, 2018; Nault et al., 2021; Xiong et al., 2019). Other single-cell studies have also identified additional cell (sub)types such as spatially resolved LSECs and HSCs, as well as immune cell subtypes with distinct gene expression characteristics (Bleriot and Ginhoux, 2019; Dobie et al., 2019; Halpern et al., 2018). Oxygen and nutrient gradients that run from the portal triad to the central vein further influence the cellular and spatial organization of liver lobules. Nutrient-rich blood from the portal vein, mixed with oxygenated blood from the hepatic artery, impart zone-specific expression of enzymes for β-oxidation and gluconeogenesis in the portal region (zone 1), whereas lipogenesis, glycolysis, and phase I xenobiotic metabolic activities are focused in the central region (zone 3) (Cunningham and Porat-Shliom, 2021; Jungermann, 1995). This spatial organization allows for interdependent metabolic pathways to co-localize and optimize metabolism while avoiding interference and energy waste from opposing pathways (Gebhardt and Hovhannisyan, 2010). Disruption of the functional relationship within this cellular heterogeneity has been associated with several adverse health consequences (Cunningham and Porat-Shliom, 2021; Hall et al., 2017; Panday et al., 2022; Soto-Gutierrez et al., 2017).

In addition to genetics, lifestyle, and diet, accumulating evidence suggests exposure to structurally diverse chemicals and environmental contaminants promotes the development of metabolic disorders such as NAFLD, type II diabetes, cardiovascular disease, and hepatocellular carcinoma (Cave et al., 2010; Lee et al., 2006; Taylor et al., 2013). For example, the persistent organic pollutant and potent agonist of the aryl hydrocarbon receptor (AHR) 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) induces hepatic lipid accumulation (steatosis) that progresses to steatohepatitis with fibrosis (Boverhof et al., 2005; Fader et al., 2017b; Nault et al., 2015a). In humans, TCDD and related compounds are associated with dyslipidemia and inflammation (Pelclova et al., 2006; Taylor et al., 2013; Wahlang et al., 2019; Warner et al., 2013). In mice, AHR activation by TCDD elicits cell-specific and spatially resolved histological and gene expression responses. Specifically, at lower doses of TCDD CYP1A1 induction occurs in the central region while lipid accumulation and inflammation first appears in the portal region (Andersen et al., 1997; Boverhof et al., 2005; Fader et al., 2017b). A recent single-nuclei RNA sequencing (snRNAseq) study reported cell-specific differential gene expression, with a putative loss of portal hepatocytes (Nault et al., 2021). Additional studies are needed to determine if this hepatocyte-specific zonal loss is truly a decrease in cell number or repression of characteristic portal hepatocyte marker expression.

A complex network of cues, receptors, and signaling cascades interact within and between cell subtypes to establish and maintain hepatic zonation. This includes the Wnt/β-catenin, RAS/ERK, Hippo, hedgehog, glucagon, HNF4α, and Dicer signaling pathways (Burke et al., 2009; Cunningham and Porat-Shliom, 2021; Gebhardt and Hovhannisyan, 2010; Kietzmann, 2019). The Wnt/β-catenin signaling pathway is a central driver of liver zonation mediated by hepatic, angiocrine, and extrahepatic secreted factors and their cognate receptors (Burke et al., 2009; Gebhardt and Hovhannisyan, 2010; Rocha et al., 2015; Zhang et al., 2020). In the canonical pathway, secreted Wnt ligands bind to Frizzled (FZD) and low-density lipoprotein receptor-related proteins (LRP) preventing the degradation of β-catenin by a destruction complex comprising adenomatous polyposis coli (APC) and glycogen synthase kinase (GSK3β). R-spondins and leucine-rich repeat-containing G protein-coupled (LGR) receptors further potentiate Wnt/β-catenin signaling (Park et al., 2020; Rocha et al., 2015). The Wnt/β-catenin pathway is also shown to interact with the AHR that is highly expressed in the central hepatic region (Braeuning et al., 2011; Gerbal-Chaloin et al., 2014; Nault et al., 2021; Prochazkova et al., 2011; Vondracek and Machala, 2016; Yang et al., 2022). Developmental AHR deletion results in nuclear β-catenin localization and induction of target genes Axin2, Ccnd1, Myc, and Lef1 as well as the modulation of basal Cyp1a1 expression (Moreno-Marin et al., 2018). Furthermore, β-catenin enhanced the induction of ligand-activated AHR target genes. β-Catenin also formed a complex with the AHR further suggesting interactions between the Wnt/β-catenin and AHR signaling pathways (Moreno-Marin et al., 2018).

Our previous work has suggested a global loss of hepatic function following TCDD treatment, especially functions associated with hepatocyte metabolism (Nault et al., 2017a, 2021). Moreover, it is established that AHR activation by TCDD perturbs mechanisms associated with the maintenance of zonation including HNF4α binding and Wnt signaling (Cholico et al., 2022; Moreno-Marin et al., 2018; Yang et al., 2022). In this study, we aimed to further characterize the cell-cell interactions in the mouse liver that were disrupted by TCDD and resulted in the loss of hepatic zonation. We hypothesize that the loss of liver function and portal hepatocytes following treatment with TCDD involves the disruption of spatially resolved Wnt/β-catenin signaling. To accomplish this, dose-dependent snRNAseq and spatial transcriptomic datasets were integrated to investigate zone-specific effects at lower doses that progress to panacinar steatosis at higher doses.

Materials and methods

 

 

Animals and treatment

Male C57BL/6 mice were received from Charles Rivers Laboratories (Portage, Michigan) at postnatal day (PND) 25 and acclimated until PND 28. Mice were randomly assigned to Innocages (Innovive, San Diego, California) with ALPHA-dri bedding (Shepherd Specialty Papers, Chicago, Illinois) at 30%–40% humidity and a 12-h light/dark cycle (3 mice/cage and dose group). Mice had free access to Harlan Teklad 22/5 Rodent Diet 8940 (Envigo, Indianapolis, Indiana) and Aquavive water (Innovive). On PND 28 and every 4 days thereafter, for a total of 7 administrations (24 days), mice were orally gavaged with sesame oil vehicle (Sigma-Aldrich, St Louis, Missouri), 0.01, 0.03, 0.1, 0.3, 1, 3, 10, or 30 µg/kg TCDD (AccuStandard, New Haven, Connecticut) between Zeitgeber time (ZT) 00 and 01 (ZT00–ZT01). The dose range induced a range of reported histopathologies and transcriptomic responses. On PND 56, 28 days after the initial gavage, mice were euthanized by CO₂ asphyxiation, and livers were immediately collected, snap frozen in liquid nitrogen and stored at −80°C. Only male mice were used in these experiments as previous reports using the same study design and dosing regimen showed similar hepatic pathologies in male and female mice with male mice exhibiting greater sensitivity (Fader et al., 2017b). All procedures were approved by the Michigan State University Institutional Animal Care and Use Committee and this report meets the ARRIVE guidelines (Percie du Sert et al., 2020). Deposited metadata fulfilled the Minimum Information about Animal Toxicology Experiments (MIATE) reporting guidelines (https://fairsharing.org/FAIRsharing.wYScsE, last accessed October 18, 2022).

Single-nuclei RNA sequencing

Nuclei were isolated from frozen samples of the right lobe (∼200 mg) as described previously (https://doi.org/10.17504/protocols.io.3fkgjkw). Two to 3 biological replicates passing QC were analyzed for each dose group with at least 2 replicates providing sufficient power to identify cell types and assess cell-specific differential expression (Datlinger et al., 2021; McGinnis et al., 2019; Nault et al., 2021). Briefly, livers were diced in EZ Lysis Buffer (Sigma-Aldrich), homogenized using a disposable Dounce homogenizer, and incubated on ice for 5 min. The homogenate was filtered using a 70-μm cell strainer, transferred to microcentrifuge tube, and centrifuged at 500 g and 4°C for 5 min. The supernatant was removed, and fresh EZ lysis buffer was added for an additional 5 min on ice following by centrifugation at 500 g and 4°C for 5 min. The nuclei pellet was washed twice in nuclei wash and resuspend buffer (1x phosphate-buffered saline, 1% bovine serum albumin, 0.2-U/μl RNAse inhibitor) with 5-min incubations on ice. Following the washes, the nuclei pellet was resuspended in nuclei wash and resuspended in buffer containing DAPI (10 μg/ml). The resuspended nuclei were filtered with 40-μm strainer and immediately underwent fluorescence-activated cell sorting using a BD FACSAria IIu (BD Biosciences, San Jose, California) with 70-μm nozzle at the MSU Pharmacology and Toxicology Flow Cytometry Core (https://drugdiscovery.msu.edu/facilities/flow-cytometry-core, last accessed October 18, 2022).

Libraries from sorted nuclei were prepared using the 10x Genomics Chromium Single Cell 3′ v3 kit and submitted for 150-bp paired-end sequencing at a depth ≥ 50 000 reads/cell using the HiSeq 4000 at Novogene (Beijing, China). Raw sequencing data were deposited in the Gene Expression Omnibus (GEO; GSE184506). Following the assessment of sequencing quality, CellRanger v3.0.2 (10x Genomics) was used to align reads to a custom reference genome (mouse mm10 release 93 genome build) which included introns and exons to consider pre-mRNA and mature mRNA present in the nuclei. Raw counts were further analyzed using Seurat v4.0.5. Each sample was filtered for (1) genes expressed in at least 3 nuclei, (2) nuclei that express at least 100 genes, and (3) ≤1% mitochondrial genes (Supplementary Figure 1). Additional quality control assessments were performed using the scater package (v1.18.6). The DoubletFinder v2.0.3 package excluded putative doublets from subsequent analyses. Raw and processed data have been deposited in GEO with the accession ID GSE184506 and the Broad Single Cell Portal (SCP1871) which can be used to visualize gene expression and quality control metrics on an individual group, cell, and animal basis.

Clustering, annotation, and analysis of snRNAseq data

Integration, clustering, and annotation were performed using Seurat. Clustering at varying resolutions (0.05, 0.1, 0.15, 0.2, and 0.3) was assessed to determine changes in cell type populations (Supplementary Figure 2). At 0.1 resolution, 11 distinct clusters were identified comparable with previous characterizations of liver cell populations (Halpern et al., 2017, 2018; Nault et al., 2021; Xiong et al., 2019). For cell annotation, a semi-automated strategy (Nault et al., 2021) was used with published data (GSE148339) as reference. Annotations were manually verified using highly expressed marker genes. Initially, unidentified clusters were further examined for marker genes using both literature and data repositories (eg, panglaoDB and Broad Single Cell Portal).

Differential expression analysis was performed using a single-cell Bayesian Test (scBT; https://github.com/satabdisaha1288/scBT), a fit-for-purpose test method that effectively controls the false positive rate when testing multiple dose groups (Nault et al., 2022). Genes were considered differentially expressed when the gene (1) was detected in at least 5% of nuclei in at least 1 dose group, (2) had an absolute fold-change ≥1.5, and (3) had an adjusted p-value ≤.05. Gene set enrichment analysis was performed using bc3net v1.0.4 in R (4.0.3) with KEGG and MPO gene sets obtained from the Gene Set Knowledgebase (GSKB; http://ge-lab.org/gskb/). Gene sets were collapsed when ≥60% of the genes were common and manually annotated based on original annotations.

Bulk and pseudobulk analyses

Experimental bulk RNAseq data from the same liver samples were obtained from Gene Expression Omnibus file GSE203302. Pseudobulk conversion of snRNAseq data was performed using the scater package (v1.18.6). Both datasets were analyzed for differential expression using semiparametric normalization and an Empirical Bayes analysis as previously described (Nault et al., 2015b; Orlowska et al., 2022). Pearson correlation of fold-changes and posterior probability (P1(t)) values was performed between datasets to determine the agreement between technologies.

Hepatocyte pseudospatial analysis

Pseudospatial analysis of hepatocyte snRNAseq data was performed as described previously (Nault et al., 2021). In short, nuclei annotated as hepatocytes were extracted from the complete dose-response dataset and reintegrated using Seurat based on the expression of previously characterized spatially resolved genes (Halpern et al., 2017). Following dimensionality reduction using Seurat, Slingshot was used for trajectory analysis along the spatial continuum (Street et al., 2018). Next, we fit negative binomial generalized additive models (NB-GAM) to individual genes using tradeSeq v1.7.4 to evaluate treatment-related effects that were considered significant (p ≤ .05). Only a small subset of genes were examined by NB-GAM and therefore the p-values were not adjusted. Center of expression (CoE), an indicator of zonal bias, was calculated for each dose and gene with a log normalized expression ≥ 0.5 and expressed in ≥ 50% of nuclei as described previously (Halpern et al., 2017). The maximum CoE travel distance was calculated as the differences between the minimum and maximum CoE values to reflect a dose-dependent change in zonation bias.

Spatial transcriptomics

Spatial transcriptomics was performed using the Resolve BioScience Molecular Cartography system, a probe-based platform with subcellular resolution for detecting gene expression in a tissue section. Two biological replicates for each dose group (0, 0.3, 3, and 30 µg/kg TCDD) were examined by placing 10 µm thick frozen liver sections in O.C.T compound (Sakura, Torrance, California) in 1 of 8 designated custom slide regions by the MSU Investigative Histopathology Laboratory (https://sites.google.com/msu.edu/ihpl/home). Frozen slides were shipped to Resolve BioSciences for analysis following manufacturer’s instructions (protocol version 3.0) (D’Gama et al., 2021; Groiss et al., 2021; Guilliams et al., 2022). In short, sections were thawed and fixed with 4% v/v formaldehyde. A total of 99 genes were probed (Supplementary Table 1) overnight, fluorescently tagged, then manually selected regions of interest (ROI; 2–3 mm²) were imaged. Probes were designed using a proprietary algorithm (Resolve Biosciences) for Ensembl gene targets outlined in Supplementary Table 1 that were verified for off-target hybridization. Probes were manually selected to (1) represent different zones, (2) distinguish different cell types, and (3) examine diverse metabolic functions affected by TCDD as noted in Supplementary Table 1. Imaging, segmentation, preprocessing, and decoding were performed as described previously (D’Gama et al., 2021; Groiss et al., 2021; Guilliams et al., 2022). Raw and SpatialExperiment formatted data (Righelli et al., 2022) was deposited in GEO (GSE206294) as well as the Broad Single Cell Portal (SCP1875).

Visualization and analysis of spatial transcriptomic images

Visualization of spatial expression was performed using custom code developed by Resolve BioSciences for ImageJ and Python (https://github.com/ViriatoII/polylux_python). Baysor (v0.5.0) was used for cell segmentation (Petukhov et al., 2021) that considered both transcript expression localization as well as DAPI images with a scale of 50 and standard deviation of 12.5%. MERINGUE was used to evaluate co-localization of gene expression (Miller et al., 2021). Analysis was performed independently on each tissue section using the spatially aware strategy with a filtering distance of 150. Spatial autocorrelation and spatial cross-correlation were determined for all genes after which groups of spatially co-expressed genes underwent hierarchical clustering to identify putative cell types. Because each section was processed independently, the number of clusters varied between tissue sections. For network analysis, gene pairs demonstrating co-expression (p-value ≤ .05) in at least 2 biological or technical replicates for each dose were considered as directly connected (2 nodes 1 edge). A network was drawn from the interactions showing each gene (node) and interaction (edge) at each dose using igraph v1.2.7. From the network, the number of edges between each pair of nodes was calculated (closeness) and used to identify a spatially resolved network of co-expressed genes.

Results

 

Clustering and cell type annotation

snRNAseq was used to further investigate the dose-dependent effects of TCDD on cell-specific differential gene expression. Doses were selected to induce complete gene expression dose-response curves as well as the previously reported spectrum of pathologies using the same dosing regimen to elicit lipid accumulation (≥0.3 µg/kg), inflammation (≥3 µg/kg), and fibrosis (30 µg/kg) (Fader et al., 2017a, 2017b; Nault et al., 2016). At 30 µg/kg, TCDD modestly increased serum ALT levels with body weight loss ≤15% within treatment groups suggesting minimal overt toxicity following oral gavage every 4 days for 28 days (Nault et al., 2016). A total of 131 613 nuclei passed quality control across all samples with an average of 14 624 (ranged from 8717 to 18 131 nuclei) per dose (Supplementary Tables 2 and 3). Approximately 1665 differentially expressed genes (DEGs) were expressed across all nuclei with a median of 3294 unique molecular identifiers (UMI; transcripts). An average of 18 317 genes were detected in individual samples with negligible mitochondrial gene contamination. There was a noticeable representation of long noncoding RNAs, consistent with our previous independent snRNAseq study (Nault et al., 2021). Conversion of this snRNAseq data to pseudobulk, for comparison with a published bulk RNAseq dataset (GSE203302) that were taken from the same liver samples, demonstrated good agreement between both technologies for the identification of DEGs (Supplementary Figure 3). Among the genes that showed poor correlation in response to TCDD were low abundance genes (eg, Atp1a2) and previously reported cytosol-biased genes (eg, Pgm1, Pgm2, and Slc2a4) (Bahar Halpern et al., 2015) (Supplementary Table 4). Small differences with the use of different technologies and sample source (tissue vs nuclei) are not unexpected.

Data dimensionality reduction and clustering were used to assess hepatic cell populations at various levels of resolutions. Because clustering is influenced by the dataset size, unique assignments were made at various resolution levels to identify the optimal parameters to distinguish unique cell types (Supplementary Figure 2). A resolution of 0.1 produced 11 distinct clusters that identified the previously characterized liver cell types (Halpern et al., 2017, 2018; Nault et al., 2021; Xiong et al., 2019). A semiautomated strategy that involved label transferring based on published data complemented by manual cluster identification using marker genes as outline in the Materials and Methods was used to annotate clusters as specific cell types (Figs. 1A and 1B and Supplementary Table 4). In addition to the 9 previously identified cell types, 2 potentially new cell types were identified. By comparison with other hepatic datasets and examining marker gene expression, one cluster was identified as a dendritic cell type (pDCs) based on the expression of Fnbp1, Wdfy4, and Ciita markers (Figure 1C and Supplementary Figure 4). The other novel cluster was identified as portal fibroblasts based on Gas6 and Msln marker expression (Koyama et al., 2017).

Figure 1.

Clustering and annotation of snRNAseq data from liver samples of male mice gavaged with TCDD every 4 days for 28 days. A, UMAP visualization of 131 613 (N = 2–3 biological replicates) annotated nuclei expression profiles across all doses and treatments. B, Label transfer prediction scores for each cluster was estimated based on published hepatic snRNAseq data (GSE148339). C, Dot plot of marker expression distinguishing individual clusters where the color represents average normalized expression and size represents percent of nuclei expressing the marker gene. D, Relative proportion of individual cell types across TCDD dose levels. Colors represent distinct cell types matched to UMAP colors in (A). The bar size represents mean values (N = 2–3 biological replicates).

TCDD dose-dependently altered the relative proportions for liver cell types (Figure 1D and Supplementary Figure 5). Hepatocytes (all zones) comprised the most abundant cell type representing 68.6 ± 2.0% in vehicle samples. However, in treated liver samples, hepatocytes (all zones) only represented 14.2 ± 5.3% following treatment with 30 µg/kg TCDD. HSCs also exhibited a modest decrease. In contrast, macrophages which only made up 9.9% ± 2.4 of hepatic cells in control livers increased to 38.6 ± 3.5% after treatment with 30 µg/kg TCDD. This large change in relative cell population is consistent with the reported increase in F4/80 staining in TCDD-treated mice (Fader et al., 2017b; Li et al., 2020). In addition, the relative proportions of cholangiocytes, LSECs, B cells, T cells, and neutrophils dose-dependently increased. The proportion of pDCs (0.5%–1.5%), and portal fibroblasts (0.4%–0.6%) did not change, though assessing population changes for rarer cell types was more difficult.

Differential gene expression and functional enrichment

The number of DEGs (Bayes Factor-adjusted false discovery rate [FDR] ≥ 0.05 and |fold-change| ≥ 1.5) for individual cell types ranged from 368 in portal fibroblasts to 1339 in macrophages. In agreement with our previous report (Nault et al., 2021), central hepatocytes were more responsive with 900 DEGs compared with 734 in portal hepatocytes. We examined DEGs for cell-specific (only differentially expressed in 1 cell type) or common (differentially expressed in 2 or more cell types) DEGs (Figure 2A). The largest sets were primarily cell-specific DEGs with only 84 DEGs shared across all cell types. The 84 genes largely consisted of known AhR target genes including Cyp1a1 (7.8-fold; central hepatocytes), Cyp1a2 (8.6-fold; central hepatocytes), Fmo3 (26.2-fold; central hepatocytes), Tiparp (5.8-fold; central hepatocytes), and Nfe2l2 (7.6-fold; central hepatocytes) (Supplementary Figure 6 and Table 6). The next largest multiple cell type DEG intersections occurred between macrophages and pDCs (91 DEGs), both derived from the myeloid lineage, followed by portal and central hepatocytes (80 DEGs) consistent with the clustering in Supplementary Figure 2.

Figure 2.

Set and functional analysis of hepatic differentially expressed genes (DEGs) from male mice gavaged with TCDD every 4 days for 28 days. A, UpSet plot of the 15 largest gene sets, in rank order, based on set analysis that identified both unique and common DEGs (Bayes Factor adjusted FDR ≥ 0.05 and |fold-change| ≥ 1.5) among all identified cell types. Set sizes represent the number of genes identified in only the cells as indicated by black circles. B, Functional analysis of DEGs for each cell type using gene lists from KEGG and MPO from the Gene Set Knowledgebase (GSKB; http://ge-lab.org/gskb/). Gene sets with ≥ 60% overlap were combined and manually annotated. The top 30 enriched functions (adjusted p-value) across all cell types are shown. A complete list is available in Supplementary Table 2.

Differential gene expression within individual cell types reflected their specific physiological roles such as phagocytosis in macrophages, the proliferation and differentiation of B cells, T cells, and pDCs, the vascular function of LSECs, and the expression of extracellular matrix-related genes by HSCs (Figure 2B andSupplementary Table 7). Interestingly, macrophages and neutrophil DEGs were also enriched for glucose/lipid metabolism. This included Acadm (0.74-fold), Atf6 (0.69-fold), Clock (0.52-fold), Hsd17b4 (0.61-fold), Ldlr (0.54-fold), Pex7 (0.70-fold), Ppargc1a (0.64-fold), Pten (0.58-fold) in neutrophils, whereas Lipa (1.79-fold), Nr1h3 (1.51-fold), Pparg (1.58-fold), and Yap1 (0.66-fold) were differentially expressed in macrophages (Supplementary Figure 7). Neutrophils have previously been linked to hepatic glucose and lipid homeostasis (Ou et al., 2017) consistent with abnormal glucose metabolism gene expression in neutrophil DEGs (Figure 2B).

Spatial heterogeneity in hepatic gene expression

To investigate spatially resolved gene expression cell segmentation of molecular cartography analyses was examined using Baysor, while spatial heterogeneity was characterized by MERINGUE (Miller et al., 2021; Petukhov et al., 2021). Individual cell types were identified based on marker gene expression with cell-cell relationships determined using network analysis (Figure 3 andSupplementary Figure 8). For example, the central hepatocyte marker Glul was largely associated with nuclear receptors (eg, Nfe2l2, Nr1h4 [FXR], Nr1i2 [PXR], Nr1i3 [CAR], Ppard, and Rxra), xenobiotic metabolism (eg, Ahr, Gsta3, Gstm3, and Nqo1), and lipid cholesterol metabolism (eg, Aldh3a2, Cd36, Ces1b, Cyp7a1, Hmgcs1, Mgll, Sqle, Srebf1, and Vldlr) genes at lower doses (0–3 µg/kg). Conversely, the portal hepatocyte marker Cyp2f2 was associated with Cbs, Egfr, G6pc, Gldc, Gls2, Hal, Kynu, Pkhd1, and Sds as expected with the portal region receiving nutrient- and oxygen-rich blood requiring antioxidant defenses (Cbs, Gldc). In addition, the portal zone is responsible for amino acid (Hal, Gls2) and glucose metabolism (G6pc). Cyp2f2 was also associated with cholangiocyte markers Sox9 and Pkhd1 consistent with their co-localization in the portal triad. The HSC marker, Tagln, co-localized with the extracellular matrix expression (eg, Adamtsl2, Col14a1, Col1a1, Col1a2, Col3a1, and Des), Wnt signaling (eg, Cacna2d1, and Ccnd1), and LSECs (eg, Rspo3) genes. Similarly, at 0 and 0.3 µg/kg TCDD, macrophage markers were adjacent to HSCs in agreement with a previous report of a cell-cell interaction (Bonnardel et al., 2019). With increasing dose, these spatial delineations eroded and became more ambiguous (Figure 3D). For example, Cyp2f2 was adjacent to genes associated with xenobiotic metabolism genes such as Gsta3, nuclear receptors (eg, Pparg, Ppara, and Nr1i3 [CAR], Nfe2l2, and Ghr), and other genes that typically co-localized with Glul at lower doses. Overall, genes distant from each other in control samples (Figure 3A) become more closely associated as indicated by the increased area with greater blue intensity (Figure 3D). The data indicate a loss of zonal gene expression across the portal to central axis.

Figure 3.

Network visualization of spatially resolved hepatic gene expression determined using MERINGUE and the Resolve Biosciences Molecular Cartography system. Mice were treated with (A) sesame oil vehicle control, (B) 0.3, (C) 3, or (D) 30 µg/kg TCDD every 4 days for 28 days. Correlations between gene co-expression was determined based on co-expression in at least 2 biological or technical replicates for each dose. Gene pairs that exhibited co-expression were considered directly connected are indicated by 2 nodes and 1 edge. Each gene (node) and interaction (edge) was used to draw a network using igraph v1.2.7. From the network, the number of edges between each pair of nodes was calculated (closeness) and used to identify a spatially resolved network of co-expressed genes. A closeness of NA indicates nodes that could not be connected at all (ie, independent clusters; area is empty). A marker gene for individual cell types is colored to illustrate their position within the network. Circles are shown next to the marker genes in the associated heatmap where the intensity of the color represents the closeness of the genes. Lines around groups of highly clustered genes based on hierarchical clustering for a total of 6 groups. Larger panels are shown in Supplementary Figure 14.

Dose-dependent loss of hepatocyte zonal identity

Hepatocyte nuclei were extracted from the complete nuclei dataset and reprocessed independently using spatially resolved genes for dimensionality reduction (Halpern et al., 2017; Nault et al., 2021). Two clear clusters were identified by UMAP visualization with the leftmost cluster enriched for the central hepatocyte markers, Glul and Gulo, whereas the portal markers, Cyp2f2 and Sds, were over-represented in the rightmost clusters (Figs. 4A and 4C andSupplementary Figure 9). In control mice, Igfbp2, considered a midzonal marker with a portal bias, was also more abundant in the rightmost cluster with modest expression in the leftmost cluster consistent with its midzonal expression (Supplementary Figure 9). Trajectory analysis examined the continuum from central to portal zones following the patterns of marker expression. Spatial transcriptomics corroborated the zonal hepatocyte gene expression, showing a clear separation of Glul expressing hepatocytes from Cyp2f2 expressing hepatocytes (Figs. 4B and 4D).

Figure 4.

Effect of TCDD on zonal gene expression in the male mouse liver following oral gavage with TCDD every 4 days for 28 days. The hepatocyte nuclei subset was extracted from the snRNAseq dataset and re-integrated using spatially resolved genes. A, Slingshot was used to evaluate the continuum from central to portal hepatocytes shown as a color gradient with the average represented by the black line. B, Molecular cartography was used to visualize the spatial distribution of Pkhd1, Sds, Glul, and Cyp2f2 zonal marker genes. Colors represent individual genes and individual spots reflect a single transcript. C, UMAP visualization shows the expression of the portal (Cyp2f2) and central (Glul) hepatocyte markers. D, Dose-dependent expression of portal Cyp2f2 and central Glul as well as AHR target genes (Cyp1a1 and Tiparp) along the pseudospace continuum from portal to central zones. Adjusted p-values were calculated based on a Wald statistic for the TCDD treatment effect. Data represents 2–3 biological replicates per dose for snRNAseq and spatial transcriptomic images are representative of data obtained from 2 biological replicates and 2 technical replicates.

Examination of the distribution along the hepatocyte pseudospatial trajectory showed an increased proportion of hepatocytes with gene expression indicative of central hepatocytes suggesting a loss of portal hepatocyte identity and function (Supplementary Figure 9). The spatial expression of several genes such as antioxidant, phase I, and II metabolism genes was changed following AHR activation by TCDD with their expression pattern disrupted compared with the typical localization observed in control lobules (Figure 4 and Supplementary Figure 10). Specifically, Cyp2f2, the portal hepatocyte landmark, was dose-dependently repressed by TCDD in portal hepatocytes but was unchanged in central hepatocytes. Likewise, expression of the central hepatocyte landmark, Glul, was dose-dependently repressed by TCDD suggesting a loss of some central characteristics and functions in the central region (Figure 4C). Similarly, AHR target genes Xdh and Gstm3, which are primarily expressed in the central region in controls, were induced in all zones giving portal hepatocytes central characteristics (Supplementary Figure 10). Although spatially resolved transcriptomics indicated that marker genes such as Cyp2f2 and Glul no longer exhibit definitive zonal distribution at 30 µg/kg TCDD, others including Sds and Igfbp2 localized adjacent to portal triads marked by Pkhd1 (cholangiocytes) revealing some zonal expression is preserved following treatment with TCDD (Figure 4B and Supplementary Figure 9). AhR target genes such as Cyp1a1 and Tiparp also exhibited clear zone-specific differential expression with central induction at lower doses while snRNAseq and spatial transcriptomic analyses showed panacinar induction at higher doses (Figure 4D).

Using CoE to calculate pseudospatial peak expression at each dose (Halpern et al., 2017), the maximum distance traveled was calculated as described in the Materials and methods. Overall, TCDD-elicited a bias toward portal and midzonal (ΔCoE ≥ 0.5) characteristics (Figure 5A). Two broad gene clusters were identified: (GA) portal genes displaying more homogenous expression across all zones and (GB) homogenously expressed genes across all zones exhibiting a more biased central expression pattern. The GB cluster consisted of known AHR induced genes including Cyp1a1, Ahrr, Tiparp, Fabp12, and Gk. No GB genes were predominantly expressed in the central region in control animals and in the portal region at higher doses. B4galt6 exhibited the largest ΔCoE (1.3; Figure 5B) while B4galt5 also showed zone-specific differential expression. Both B4galt5 and B4galt6 are implicated in lactosylceramide (LacCer) synthesis, a lipid associated with NASH, though only a modest increase in LacCer was observed at 3 µg/kg using a similar study design (Nault et al., 2017b). Genes in group GA were largely repressed to levels comparable with the central region consistent with an overall loss of zonal functional organization. For example, Acly, is primarily expressed in the portal/midzonal region in control mice but was repressed to near central region levels starting at 3 µg/kg (Figs. 5C and 5D) that likely contributed to impaired lipogenesis in the region that received the most nutrient rich blood. Overall, these results demonstrated TCDD dose-dependently disrupted zonal gene expression.

Figure 5.

Center of expression (CoE) analysis of hepatocyte gene expression in male mice gavaged with sesame oil vehicle control, 0.01, 0.03, 0.1, 0.3, 1, 3, 10, or 30 µg/kg TCDD every 4 days for 28 days. CoE was calculated as described previously (Halpern et al., 2017). The CoE travel distance (ΔCoE; maximum CoE—minimum CoE) was calculated for each gene. A, Hierarchical clustering identified 2 major clusters (GA and GB) for the top 5th percentile of ΔCoE values. Dose-dependent changes of CoE values (top) and pseudospace distribution of expression (bottom) is shown for (B) B6galt6 and (C) Acly. Adjusted p-values were calculated based on a Wald statistic for an effect of TCDD treatment. D, Dose-dependent molecular cartography analysis for Acly as well as the portal (Cyp2f2, Sds), central (Glul), and cholangiocytes (Pkhd1) marker genes. Scale bar represents 50 μm. Data represents 2–3 biological replicates per dose for snRNAseq and spatial transcriptomic images are representative of data obtained from 2 biological replicates and 2 technical replicates.

Disruption of the Wnt/β-catenin signaling cascade

Wnt/β-catenin signaling is a key mediator of liver zonation (Dobie et al., 2019; Rocha et al., 2015; Xiong et al., 2019). Examination of genes involved in the Wnt/β-catenin signaling cascade showed dose-dependent and zone-specific dysregulation (Figure 6 andSupplementary Figs. 11 and 12). R-spondins (Rspo1 and Rspo3) are ligand agonists of Wnt/β-catenin signaling that were highly expressed in portal fibroblasts, LSECs, and HSCs (Figure 6A). TCDD did not alter LSEC or HSC Rspo expression though Rspo3, which is highly expressed in stellate cells, was induced 1.90-fold but did not achieve significance due to the cytosolic bias of Rspo3 mRNA, and the dose-dependent decrease in the number of stellate cells that limited the utility of the scBT test method (Bahar Halpern et al., 2015; Nault et al., 2022). LRT linear analysis, a more appropriate test, identified significant Rspo3 induction in stellate cells consistent with our spatial transcriptomic analysis (Nault et al., 2022) (Supplementary Figure 12). The induction of Rspo3 following stellate cell activation by carbon tetrachloride is also reported in areas of fibrosis (Dobie et al., 2019; Zhang et al., 2020). The most highly expressed WNT ligands, Wnt2 (1.4-fold, adjusted p-value ≤ .05) and Wnt9a (1.3-fold, adjusted p-value ≤ .05), were expressed in LSECs and portal fibroblasts with modest induction by TCDD that did not meet the fold-change threshold (Figure 6A). Wnt2 showed both AHR enrichment and the presence of a putative DRE implicating potential direct AHR regulation (Supplementary Figure 13).

Figure 6.

Dose-dependent effects of TCDD on the cell-specific and pseudospatial expression of genes associated with Wnt/β-catenin signaling. A, Expression of highly expressed Wnt/β-catenin ligands primarily expressed in nonparenchymal cells where the color of the dot represents the average normalized expression, and the size of the dot represents percent of nuclei expressing the gene. Dose-dependent expression along the hepatocyte pseudospace continuum is shown for receptors and downstream targets (B) Lrp5, (C) Lrp6, (D) Lgr5, (E) Axin2, and (F) Apc. Adjusted p-values were calculated based on a Wald statistic for an effect of TCDD treatment. Data represent 2–3 biological replicates per dose.

The Frizzled family of receptors (Fzd1, 3, 4, 5, 6, 7) exhibited low levels of expression in hepatic nuclei with negligible changes following TCDD treatment. Conversely, the co-receptors Lrp5 and Lrp6 were highly expressed in central and portal hepatocytes, respectively, with Lrp1 and Lrp4 evenly distributed across all zones (Figs. 6B and 6C and Supplementary Figure 11). Lrp1 was modestly repressed in central hepatocytes while Lrp4 was induced in both portal and central hepatocytes. Although R-spondin receptors potentiate WNT signaling, Lgr4 and Lgr5 were the most abundantly expressed receptors in portal and central hepatocytes, respectively. Lgr4 was not altered by TCDD while Lgr5 exhibited modest dose-dependent repression in the central zone (Figure 6D and Supplementary Figure 11). In agreement with altered zone-specific Wnt/β-catenin activation, the downstream target Axin2 which is largely restricted to central hepatocytes at lower doses, exhibited greater dose-dependent induction in the central zone (Figure 6E and Supplementary Figure 12). Additionally, Apc, a member of the β-catenin destruction complex that supports portal hepatocyte maintenance was dose-dependently induced in all zones with the highest induction occurring in central hepatocytes (Figure 6F).

Under homeostatic conditions and following liver injury, the IGFBP2-mTOR-CCND1 axis drives the proliferation of midzonal hepatocytes (Wei et al., 2021). Our spatial transcriptomic analysis shows that Ccnd1 expression at low doses is limited to the portal triad with widespread portal and central expression at 30 µg/kg TCDD (Supplementary Figure 12). Although Igfbp2 and Ccnd1 co-localize, Ccnd1 is more commonly co-localized with Axin2 consistent with induction via Wnt/β-catenin signaling pathway rather than the IGFBP2-mTOR-CCND1 axis. The inhibition of liver regeneration and induction of Ccnd1 is consistent with the perturbation of Wnt/β-catenin signaling and the G1 cell cycle arrest elicited by TCDD (Jackson et al., 2014).

Discussion

The development and progression of hepatic steatosis to steatohepatitis with fibrosis involves interactions between distinct cell types as well as the disruption of the spatially organized and temporally separated metabolic functions of the liver lobule. We integrated snRNAseq and spatial transcriptomic datasets to investigate the role of AHR activation in the disruption of zonal gene expression in liver lobules following dose-dependent treatment with TCDD. In contrast to our previous report that examined only 16 015 nuclei (Nault et al., 2021), pDCs and portal fibroblasts were also identified using this larger dose response snRNAseq dataset that included 131 613 nuclei. Although pDCs did not represent a significant proportion of cells in the mouse liver, increased numbers are reported in acute decompensated cirrhosis samples with higher levels of interferon (Cardoso et al., 2021). Portal fibroblasts play a key role in maintaining the portal tract and produce extracellular matrix when activated (Karin et al., 2016). However, the most striking cell population change was the increased proportion of macrophages from 10% in controls to 39% in treated liver samples. Consequently, the relative proportion of hepatocytes decreased but the results do not indicate hepatocyte loss or death. Although the nuclei isolation protocol may be biased (eg, preferably capture macrophage nuclei), the observed increase in macrophages is consistent with increased F4/80 staining as well as other reports of immune cell infiltration in NAFLD models (Fader et al., 2017b; Li et al., 2020; Miura et al., 2012; Xiong et al., 2019).

Functional analysis of DEGs is in agreement with immune cells playing a central role in TCDD-elicited NAFLD pathologies. Macrophages, neutrophils, and pDCs were the 3 most responsive cell types followed by hepatocytes. Enrichment of lysosomal-related genes in macrophages is consistent with lysosomes contributing to the internalization of cholesterol in Kupffer cells (KCs) driven in part by Cd36 (Bieghs et al., 2012). Cd36 is a direct target of TCDD-activated AHR that leads to cholesterol and cholesterol ester accumulation in the liver (Dornbos et al., 2019; Lee et al., 2010; Nault et al., 2017b). Foam-like KCs secrete chemokines that recruit other immune cell types including neutrophils in addition to the recruitment by leukotrienes induced by TCDD (Doskey et al., 2020; Takeda et al., 2017). Interestingly, neutrophil infiltration is believed to contribute to NAFLD progression due to dysregulation of diurnal rhythm. Although we have previously shown that TCDD dose-dependently disrupts diurnal regulation, in this study Clock, a master circadian regulator, was repressed in neutrophils from treated livers (Crespo et al., 2020; Fader et al., 2019).

In addition to cellular heterogeneity, spatial organization is a key determinant of liver function, (Braeuning et al., 2011; Gerbal-Chaloin et al., 2014; Kietzmann, 2019; Soto-Gutierrez et al., 2017). Although AHR activation at lower TCDD doses regulates gene and protein expression primarily in the pericentral region, steatosis and inflammation initially occur in the periportal region (Andersen et al., 1997; Boverhof et al., 2005; Fader et al., 2017b). Moreover, snRNAseq data suggests dose-dependent induction of AhR target genes including Cyp1a1 and Cyp1a2, as well as antioxidant defense genes initially in central hepatocytes. Spatial transcriptomic analyses also distinguished hepatocyte zonation based on markers associated with zone-specific metabolic pathways. However, TCDD elicited a dose-dependent loss of zonal identity. For example, Acly and Cyp2f2, 2 portal markers, were dose-dependently repressed. Cyp1a2 is also proposed as a marker of central hepatocytes in a model of acetaminophen toxicity (Umbaugh et al., 2021), but as a direct target of the AHR, it’s zone-specific expression is lost following TCDD treatment. In contrast, other markers maintained their zone-specific expression such as the portal expression of serine dehydratase (Sds) associated with gluconeogenesis. This suggests a functional gradient was retained for some pathways, though TCDD caused its expression to further concentrate around the portal region.

Zonation of hepatic lipid accumulation depends on the model and exhibits species differences. It is dependent on the functional differences between portal and central hepatocytes (Hijmans et al., 2014; Schleicher et al., 2017). In human NAFLD and alcoholic fatty liver disease (AFLD), lipids accumulate in the central region consistent with the upregulation of lipogenesis (a central biased pathway) and the uptake of mobilized lipids originating from peripheral tissue stores. In contrast, lipid accumulation in pediatric NAFLD exhibits a periportal bias or no zonal preference in the absence of hepatocyte ballooning (Carter-Kent et al., 2011; Hijmans et al., 2014). In ob/ob, db/db, and high-fat diet-fed mouse models, fat accumulation occurs centrally (Flach et al., 2011; Hijmans et al., 2014; Wiegman et al., 2003), whereas diets high in carbohydrates and the western diet primarily induce portal fat accumulation (Ghallab et al., 2021; Hijmans et al., 2014). Because zonation of steatosis appears to be model dependent, further studies are needed to elucidate the underlying mechanisms responsible for central versus portal steatosis and whether these factors are additive in panacinar steatosis. Furthermore, mice and humans exhibit zonal differences in metabolic activity, particularly for genes related to lipogenesis (Massalha et al., 2020). In this study, TCDD disrupted zonation with periportal hepatocytes losing characteristic functions and adopting central functions (ie, pericentralization). More specifically, TCDD repressed β-oxidation and increased triglyceride synthesis, functions typically associated with the portal and central region, respectively (Cholico et al., 2021; Lee et al., 2010; Nault et al., 2017b). TCDD-elicited steatosis has been attributed to dietary sources consistent with elevated levels of fatty acids in the portal circulation following feeding and not impaired metabolism in the portal region (Angrish et al., 2012; Nauli and Matin, 2019; Nault et al., 2017b; Schleicher et al., 2017).

Several mechanisms are implicated in the distribution of enzymatic activities along the portal-central axis that determines the metabolic characteristics of hepatocytes. The Wnt/β-catenin pathway plays a key role in determining this functional distribution (Braeuning et al., 2011; Gerbal-Chaloin et al., 2014; Nault et al., 2021; Prochazkova et al., 2011; Vondracek and Machala, 2016; Yang et al., 2022). Deletion of β-catenin in hepatocytes, as well as Lgr4 and Lgr5 deletion, increase NAFLD severity in mice fed a methionine choline deficient diet or high-fat diet (Behari et al., 2010; Saponara et al., 2021). Abundantly expressed Wnt ligands, Wnt2 and Wnt9a showed only modest induction with TCDD. Wnt2 was identified as abundantly expressed by central LSECs with its knockout eliciting a “periportal” hepatocyte phenotype (Halpern et al., 2018; Hu et al., 2022). More importantly, Wnt2 in LSECs is co-expressed with Rspo3, another Wnt/β-catenin signaling member implicated in the maintenance of central hepatocyte characteristics (Halpern et al., 2018; Rocha et al., 2015). However, in TCDD-treated mice, Rspo3 was primarily expressed and induced in stellate cells as reported in other hepatic scRNAseq datasets (Dobie et al., 2019; Xiong et al., 2019). Levels of R-spondins, particularly Rspo3, are increased in human hepatic fibrotic lesions and induced following the activation of stellate cells in vitro. HSCs also exhibit zonation, with central HSCs producing collagen that is associated with elevated Rspo3 expression (Dobie et al., 2019). In our dataset, very few HSCs with low Rspo3 expression were identified possibly due to a nuclei isolation bias or given the difference in the age between mouse models (ie, 4–8 weeks vs. 10–16 weeks). Spatial transcriptomics suggested the presence of zonated HSCs further implicating AHR-mediated induction of Rspo3 in HSCs as another factor contributing to disrupted lobular zonation following treatment with TCDD. Compensation of hepatocyte function is observed in response to liver injury by partial hepatectomy and acetaminophen overdose also involved the Wnt/β-catenin pathway, whereas carbon tetrachloride is reported to induce Axin2 expression in midlobular hepatocytes (Walesky et al., 2020; Zhao et al., 2019). Consistent with the acetaminophen and carbon tetrachloride injury models (Walesky et al., 2020; Zhao et al., 2019), our data implicate LSECs in Wnt/β-catenin signaling during compensation and reveals a putative role for HSCs.

In summary, dose-dependent AHR activation by TCDD elicited cell- and zone-specific dysregulation of gene expression along the portal-central axis of hepatic lobules. Spatial transcriptomics confirmed the initial loss of periportal hepatocyte characteristics progressed to disruption of positionally defined functions throughout the liver lobule following treatment with higher doses of TCDD. β-Catenin activation in the portal region and impaired activation in the central region is consistent with Apc expression and its role as a zonal gatekeeper (Benhamouche et al., 2006). The loss of portal-specific functions is also in agreement with impaired lipid metabolism (ie, β-oxidation), and initial increases in lipid accumulation in the periportal zone. The progression of pathologies likely involves aberrant Wnt and R-spondin signaling within LSECs and HSCs that contributes to metabolic reprogramming and hepatotoxicity that contribute to an inflammatory environment and further tissue damage. In contrast to other hepatotoxicants such as acetaminophen and carbon tetrachloride that are metabolically activated in the central region by Cyp2e1 eliciting centrilobular necrosis, TCDD elicits little to no necrosis within the dose range used in this study. Nevertheless, compensation of hepatocyte function via Wnt/β-catenin signaling by LSECs and HSCs appears to be common among multiple hepatotoxicants. Further studies using cell-specific and zone-specific ablation of AHR and Wnt/β-catenin pathway members are required to further explore the disruption of enzymatic functions along the portal-central axis as a novel mechanism contributing to hepatotoxicity. Collectively, this study begins to elucidate the dose-dependent cell-cell interactions elicited by TCDD that underlie the progression of steatosis to steatohepatitis with fibrosis that parallel the pathologies associated with NAFLD.

Supplementary data

Supplementary data are available at Toxicological Sciences online.

Data availability

Spatial transcriptomic data were deposited in the Gene Expression Omnibus (GEO) under the accession ID GSE206294 and Broad Single Cell Portal (SCP) at SCP1875. Dose-response single-nuclei RNA sequencing data are available on GEO under the accession ID GSE184506 and SCP at accession ID SCP1871. Single-nuclei transcriptomic data for label transfer were obtained from GEO under the accession IDs GSE148339 and are also available from SCP with ID SCP1851. Supplementary tables can be obtained from https://doi.org/10.5061/dryad.547d7wmc5.