Introduction

The liver, a vital metabolic and secretory organ, maintains homeostasis through its remarkable regenerative capacity following liver injury1. Liver regeneration pathways diverge depending on injury modality. In instances of acute liver injury, such as liver partial hepatectomy (PHx), the growth signal for hepatocytes remains unimpeded, allowing for the restoration of injured liver tissue through the process of self-renewal2,3. Conversely, in the case of chronic liver injury, such as that observed in animal models subjected to the 3,5-Diethoxycarbonyl-1,4-Dihydrocollidine (DDC) diet, research has demonstrated that the growth signals for hepatocytes are suppressed. Subsequently, mature hepatocytes transition into liver progenitor-like cells (LPLCs), promoting liver regeneration4. However, the presence of progenitor cells in acute liver injury has long been controversial5.

In previous research utilizing single-cell sequencing technologies in PHx models, it has been observed that a population of hepatocytes exhibits a similar gene expression pattern with fetal hepatocytes during liver regeneration6. Following PHx, a transient increase in the progenitor cell marker Sox9 was detected at the 8-h mark, indicating a potential temporary dedifferentiation of hepatocytes in the PHx model7. The findings indicate that during liver regeneration following PHx, there is a potential capacity for hepatocytes to undergo dedifferentiation and transform into progenitor cells in the PHx model. However, the identification of progenitor cells and the rationale behind the absence of the LPLC-induced regeneration in the PHx model have not been conclusively established.

The process of liver regeneration involves complex crosstalk between hepatocytes and non-parenchymal cells (NPCs), such as Kupffer cells, liver sinusoidal endothelial cells (LSEC), and hepatic stellate cells (HSC), among others. These interactions contribute to a sophisticated regulatory network. Additionally, the activation and differentiation of LPLCs are facilitated through intricate intercellular communications8. For instance, the tumor necrosis factor (TNF)-like weak inducer of apoptosis (TWEAK), secreted by macrophages, interacts with fibroblast growth factor-inducible 14 (Fn14) present on LPLCs, thereby promoting LPLCs proliferation9. Furthermore, inhibition of growth signaling in chronic injury models promotes the transdifferentiation process of LPLCs10. Notably, comparative analyses of LPLCs in acute versus chronic liver injury remain limited.

As a precursor process in gene transcription, chromatin accessibility is instrumental in unraveling additional layers of gene regulation and expression. The analysis of chromatin accessibility provides a means to gather information preceding gene expression, and snATAC-seq serves as a highly effective tool11. Moreover, current research on liver regeneration has predominantly centered on transcriptome-level analyses, whereas temporal dynamics of hepatocyte chromatin accessibility following partial hepatectomy have received less attention.

Therefore, this study aims to investigate temporal chromatin dynamics in hepatocytes following acute liver injury, define the role of hepatic progenitor cells, and identify regulatory factors governing hepatic progenitor cells by snATAC-seq of the mouse PHx model. Through multi-omics integration and cross liver injury model comparisons, we reveal that hepatocytes undergo reprogramming after acute injury, potentially transdifferentiating into hepatic progenitor cells. However, the difference in growth signaling and disrupted intercellular communication limit the contribution of hepatic progenitor cells to regeneration in acute injury.

Results

The chromatin accessibility landscape of hepatic cells in the PHx model using snATAC-seq

To profile the chromatin accessible landscape of hepatic cells and identify potential new cell states during liver regeneration at an epigenetic scale, a total of 34 snATAC-seq libraries were conducted using a canonical mouse 2/3 PHx model, as used in our previous work, which generated scRNA-seq data and spatial transcriptome data7. And the same sampling time points were adopted too, which include homeostasis phase (Day 0 (D0)), priming phase (8 h (8H) after PHx), proliferation phase (Day 1 (D1), Day 2 (D2), and Day 3 (D3) after PHx), and termination phase (Day 7 (D7) after PHx) (Fig. 1A), facilitating for comparing multi-omics data of two studies.

Fig. 1: snATAC-seq landscape of mouse PHx model.
figure 1

A Samples and profiling methods used in this study. D0: before PHx, 8H: 8 h post-PHx, D1: one day post-PHx, D2: two days post-PHx, D3: three days post-PHx, D7: seven days post-PHx. B, C UMAP showing the snATAC-seq cells passing quality control, colored by annotated clusters/sequence libraries. HomeostasisHep1: Homeostasis hepatocytes 1; HomeostasisHep2: Homeostasis hepatocytes 2; HomeostasisHep3: Homeostasis hepatocytes 3; HomeostasisHep4: Homeostasis hepatocytes 4; StressHep5: Stressed hepatocytes 5; StressHep6: Stressed hepatocytes 6; StressHep7: Stressed hepatocytes 7; StressHep8: Stressed hepatocytes 8; ProlifHep9: Proliferating hepatocytes 9. D snATAC-seq gene accessibility for selected marker genes for each snATAC-seq cell types. Color indicates average gene activity score across all cell types and dot size indicates the percentage of cells accessible to the gene in this cluster. HH1: HomeostasisHep1; HH2: HomeostasisHep2; HH3: HomeostasisHep3; HH4: HomeostasisHep4; SH5: StressHep5; SH6: StressHep6; SH7: StressHep7; SH8: StressHep8; PH9: ProlifHep9. E Genomic tracks for chromatin accessibility around the Clec4f and Esam locus (upper) for Kupffer cells subtypes. Bar plots showing the average gene activity score (GAS) and P-value of Clec4f and Esam (Bottom). Error bars represent mean ± standard deviation (SD), with each data point indicating an individual sample library (n = 34). P was calculated by T-test. F Fraction of each sequence library comprising each snATAC-seq cell type.

Full size image

After data preprocessing and quality control (see Methods), we obtained a total of 102,489 high-quality nuclei (Fig. S1A–C) with median 24,416 fragments and 14.348 TSS enrichment score per nucleus. Firstly, we utilized Latent Semantic Indexing (LSI) within the ArchR package for dimensionality reduction and applied 2-D UMAP to categorize cell types and their nuanced subtypes12. 18 distinct cell populations were identified in all time points with a minimal batch effect (Fig. 1B, C). Cell-type identities were assigned based on the gene activity scores and integrated gene expression of canonical cell-type-specific markers: nine hepatocyte subtypes (HH1 to PH9) highly accessible at Hnf4a, Apoc2, Cyp2ef, Oat, Hamp, Alb, Aldob, and Sox9); HSC highly accessible at Vipr1, Reln, Lhx2, Lrat; two Kupffer cell subtypes (Kc1 and Kc2) highly accessible at Vsig4, Clec4f, Cd5l, Cd163, and Timd4; B cell (Bc) highly accessible at Cd79b, Cd19, Pax5, and Ebf1; Cholangiocyte (Cho/Chol) highly accessible at Krt19, Krt7, and Epcam; Neutrophils (Neu) highly accessible at S100a9 and Retnlg; NKT cells highly accessible at Nkg7, Klrb1c, Cd3g, and Cd3d; LSEC highly accessible at Pecam1, Cdh5, Clec4g, Dnase1l3, and Aqp1; while liver vessel endothelial cell (LVEC) highly accessible at Vwf, Nrg1, Gja5, Adgrg67,13(Figs. 1D, S1D). Through systematic integration of corresponding single-cell RNA-seq data, we acquired cell gene expression signatures. The high degree of temporal synchronization between omics layers confirmed robust dataset alignment (Fig. S1E). The annotated cell populations exhibited higher linked gene expression of these markers too (Fig. S1F). These results demonstrate robust compatibility between the integrated multi-omics datasets. The two Kupffer cell subtypes corresponded to previously characterized groups, a primary cluster with CD206 low expression (CD206low) endothelial cell-selective adhesion molecule negative (ESAM-) population (Kc1) and a minor CD206 high ESAM+ population (Kc2)14,15 (Fig. 1E). However, it should not be overlooked that some current studies suggest that Kc2 represents LSEC-KC doublets16.

The composition of 9 hepatocyte subtypes exhibits distinct temporal characteristics. At the peak of proliferation (D2, D3), PH9 was prominently expanded and diminished at the termination stage, while SH5, SH6, SH7, and SH8 were specifically presented during the priming phase (8H and D1). Conversely, HH1 and HH2 declined during the priming phase (8H). HH3 and HH4 constituted a significant proportion of all hepatocytes at all time points (Fig. 1F). Given the similar cellular composition observed between D1-R1 and 8H sample groups (Fig. 1F), we comparatively analyzed chromatin accessibility profiles across the D1 and 8H cohorts. This analysis revealed a significantly stronger correlation between D1-R1 and 8H samples (Fig. S1G). Therefore, D1-R1 has been categorized as the priming phase. This may be caused by a postponed physiological condition in D1-R1 in response to liver injury.

In conclusion, we had captured and characterized all hepatic cell types with their epigenetic features and mapped the dynamics of their composition during liver regeneration in the mouse PHx model.

Chromatin accessibility landscape of hepatic cell types in the PHx model

To analyze chromatin accessibility variations across hepatic cell types, we called chromatin accessible peaks of each of cell types with Macs2 based on intra-cell-type aggregated pseudo-bulk replicates using ArchR12, and identified marker peaks of all cell types. For hepatocytes, each of the nine subtypes exhibits a distinct set of accessible peaks, with priming stage specific subtypes, SH5, SH6, SH7 and SH8, demonstrating a higher number of differential peaks while stably present subtypes, HH1, HH2, HH3 and HH4, displayed fewer marker peaks (Fig. 2A). This indicates a rapid alteration in the chromatin landscape of hepatocytes during the priming stage, enabling the cells to effectively respond to the liver injury, which is corresponding to the uniqueness of 8H samples in the scRNA-seq data7. Concurrently, we observed minimal effects of cell numbers per type, total identified peaks, and TSS enrichment levels on the quantity of differentially accessible peaks in each cell type (Fig. S2A–C). In the cases of non-parenchymal cells (NPCs), cholangiocyte exhibited fewer marker peaks compared to Kupffer cells, Stellate cells, and endothelial cells. Kc2 demonstrated a mixed chromatin accessibility pattern of Kc1 and LSEC (Fig. 2B). This is consistent with the report that Kc2 expresses LSEC marker genes14,15.

Fig. 2: Characterization of chromatin accessibility between hepatocytes and nonparenchymal cells.
figure 2

A Heatmap reveals differential accessible peaks across different hepatocyte subtypes with P < 0.01. P was calculated by Wilcoxon. B Same as (A) but for nonparenchymal cells (NPCs). P < 0.01. P was calculated by Wilcoxon. C Distribution of gene elements in differential accessible peaks. D Hypergeometric enrichment of TF motifs in marker peaks for different hepatocyte subtypes. P < 0.01. E Hypergeometric enrichment of TF motifs in marker peaks for nonparenchymal cells (NPCs). P < 0.01.

Full size image

Genomic annotation of these marker peaks revealed a similar composition pattern across cell types, where the distant regions from genes and the intronic regions have a higher number of accessible peaks than the exonic and promoter regions have (Figs. 2C, S2D). This disparity might be related to the longer genomic occupation of intergenic and intronic regions17.

We then performed a hypergeometric test-based analysis to identify potential transcription factors by aligning their binding motif sequences with the marker peaks of each cell type12 (Fig. 2D). Motif enrichment analysis of 9 hepatocyte subtypes showed that PH9’s marker peaks enriched in Sp family (Specificity Proteins) motif which have a documented role in controlling the gene expressions related to proliferation and amino acid and protein metabolism18,19. SH5’s marker peaks enriched Nfe2l2 (Nrf2), AP-1 family, Smarcc1, and other transcription factors, which were enriched in SH6, SH7, and SH8 to lower degrees. The Keap1-Nrf2 system is a critical defense mechanism against oxidative stress20. Studies indicate that Nrf2 may play a role in the expansion of oval cells (progenitor cells)21, and a deficiency in Nrf2 can lead to significant delays in DNA synthesis and increased hepatocyte apoptosis in mouse post-PHx22. On the other hand, AP-1 is known to activate DNA binding of NFκB, Stat3, and itself through cytokine-mediated upstream signaling shortly after PHx23, and it was found that the AP-1 transcription factor plays an important role in promoting cardiomyocyte dedifferentiation, proliferation, and protrusion into the injured area by regulating changes in chromatin accessibility in a cardiomyocyte injury model24. The liver X receptors Nr1h2 (LXRα) and Nr1h3 (LXRβ), regulating lipid homeostasis, which were reported preferential in central zonation, enriched in the HH1 and HH2 subtypes25. HH2 also enriched pericentral WNT signaling mediators26, TCF/LEF transcription factors Tcf7l1/2 and Lef127. In addition, our data also suggest that non-parenchymal cells also be regulated by different transcription factors (Fig. 2E).

In summary, we found a dramatic alteration of chromatin accessibility in priming stage specific hepatocyte subtypes. And the modulation of Nfe2l2 (Nrf2) and AP-1 family genes appears to play a crucial role in the characteristics of SH5.

Cellular plasticity of hepatocytes in PHx model

To investigate the plasticity of hepatocytes following PHx, we conducted a comparative analysis of 9 subtypes of hepatocytes based on gene activity scores to define their cell identity (Fig. 3A). Similar to the pattern observed with accessible chromatin peaks, the priming stage-specific subtypes, SH5, SH6, SH7, and SH8, exhibited the most numerous differentially accessible genes (DAGs) while HH3 and HH4 exhibited fewer DAGs (Fig. 3B–D). The mitogen Pdgfa28,29 and the cell cycle regulator gene Cdc2030 were found to be highly accessible in the PH9 subtype, which is enriched during the proliferation stage (Fig. 3B–C). In the SH5, SH6, SH7, and SH8, there was notable accessibility of Tm4sf4 and Ddr1. Previous research reported that Tm4sf4 plays a pivotal role in the pancreas development and the differentiation of various cell types31, and Ddr1 induces stem cells to differentiate into basal cells32. The highly available genes such as Ppp1r3b33, Klhdc7a34, Gck35, Sdsl36, Sds37, Cyp2f238 of HH1, HH2, HH3, and HH4, which are cell subtypes stably existing or except in the priming stage, are related to metabolism (see Methods).

Fig. 3: Functional transition and spatial distribution of hepatocyte subtypes.
figure 3

A UMAP showing the snATAC-seq cells colored by hepatocyte subtypes. B Fraction of each time-point samples comprising hepatocyte subtypes. C Differential accessibility genes (DAGs) for nine hepatocyte subtypes. D The number of DAGs for nine hepatocyte subtypes. E GO term enrichments for genes with high accessibility and high integrated expression in PH9 (ProlifHep9). F Boxplots showing the gene integrated score (GIS) of Ccnb1 and Mki67 calculated by integration of the PHx scRNA-seq. The lines in the middle of the boxplots mean the median. P < 0.01 (PH9 vs each other hepatocyte subtypes) calculated by T-test. G Same as E but for SH5 (StressHep5). H Gene integrated score (GIS) of Tnfrsf12a, Sox9 and Cdc45 calculated by integration of the PHx scRNA-seq. The lines in the middle of the boxplots mean the median. P was calculated by T-test. I Gene accessibility score (GAS) of metabolic pathway-related genes in nine hepatocyte subtypes. Color indicates average gene accessibility score (GAS) across all cell types and dot size indicates the percentage of cells accessible to the gene in this cluster. ***P < 0.001 (HH1 and HH2 vs other hepatocyte subtypes) calculated by T-test. J Genomic tracks for chromatin accessibility around the Glul, Cyp2e1, Alb and Sds locus (upper) for hepatocyte subtypes. Boxplots showing the gene integrated score of corresponding genes.The lines in the middle of the boxplots mean the median. ***P < 0.001. P was calculated by T-test. K Line plots showing the module score of genes with high accessibility and high integrated expression of each hepatocyte subtype in Stereo-seq data. The color of lines representing time point. The nine points on the curve representing liver zonation (1 and 2 pericentral,3–6 midzonal and 7–9 periportal). Error bars: mean ± SD. Each point represents one zonation layer along the CV-PV axis (Data points correspond to 9 zonation layers (CV-PV axis), each resolved into 10 sub-regions).

Full size image

Based on the integration of scRNA-seq data, we performed an analysis to identify differentially expressed genes (DEGs) using the linked gene expression and performed functional enrichment analyses on the genes common to both DEGs and DAGs across each hepatocyte subtype to elucidate their biological identities. In the case of PH9, there was an observed enrichment of regulation of RNA polymerase, cell division, bicellular tight junction assembly, mitotic cell cycle and other biological processes, indicating its proliferating state (Fig. 3E). The PH9 also exhibits a higher linked gene expression of widely used proliferation markers, Ccnb139 and Mki6740 compared to each other hepatocyte subtypes (P < 0.01) (Fig. 3F). The intersecting genes of SH5 are mainly involved in the positive regulation of cell population proliferation, embryonic development (Fig. 3G). Additionally, the SH5 exhibits a significantly higher linked gene expression of Tnfrsf12a41,42,43, Sox944 compared to each other hepatocyte subtypes (P < 0.01). Elevated expression of cell division cycle protein 45 (Cdc45) was also observed (P < 0.01 vs. all subtypes except PH9). Furthermore, gene accessibility of other progenitor cell markers (Fig. S3A), suggesting that SH5 may be a cluster of hepatic progenitor-like cells. Other priming stage-specific subtypes, SH6, SH7, and SH8’s markers are primarily enriched in the regulation of innate immune response, cellular response to oxidative stress, response to bacterium, implying their transient state responding to loss of a vast mass of liver (Fig. S3B). Our analysis also revealed that HH1, HH2, HH3 and HH4’ marker genes are significantly associated with distinct metabolic processes related to various substances (Fig. S3C, D). Furthermore, we observed a significant decrease of metabolic genes accessibility of HH3 and HH4 (stably existing across all timepoints) compared to HH1 and HH2 (absent in the priming stage) (Fig. 3B, I), suggesting a transient decline in the metabolic functions of liver tissue in priming stage following PHx45 (P < 0.01).

To investigate the zonation characteristics of 9 hepatocyte subtypes, we analyzed the chromatin accessibility of known pericentral zonation markers Glul, Cyp2e1 and periportal zonation markers Alb and Sds across nine hepatocyte subtypes46,47 where we found higher Sds expressions in HH3 and HH4, higher Cyp2e1, Alb expression in HH1, and higher Glul, Cyp2e1 expressions in HH2 and SH8 (Fig. 3J). Additionally, we calculated the module score of intersecting genes between DAGs and DEGs in hepatocyte subtypes using PHx Stereo-seq data7 (Fig. 3K). Upon analysis, it was determined that HH3 and HH4 exhibit PV (Portal vein) characteristics, while HH1 and HH2 demonstrate CV (Central vein) characteristics. The zonation characteristics of other hepatocyte subtypes remain ambiguous. Notably, we found that DAGs and DEGs upregulated in the proportionally expanded hepatocyte subpopulation at 8 h post-hepatectomy (SH5, SH6, SH7, and SH8) were similarly upregulated at the spatial transcriptomic level during the same time window (Fig. 3K).

In summary, during the priming stage after PHx, the liver undergoes a loss of some hepatocytes that perform pericentral metabolic functions (HH1, HH2) while preserving a majority of hepatocytes essential for periportal metabolic activities (HH3, HH4). SH6, SH7, and SH8 appear transiently with stress responding characters in the priming stage. Notably, we have observed a subset of hepatocytes (SH5) exhibiting a higher integrated expression of the marker genes associated with hepatocyte progenitors. And we have also observed a proliferating hepatocyte population (PH9) emerging on days 2 and 3 post-hepatectomy, which generates newly formed hepatocytes. Hepatocytes of distinct functional states demonstrate remarkable plasticity following hepatectomy, sustaining coordinated maintenance of proliferative and metabolic homeostasis during liver regeneration.

Multiomics validation of hepatic progenitor-like cells in the PHx model

To further explore the characteristics of progenitor-like SH5, we conducted comparative analyses between SH5 and subtypes with normal metabolic functions (HH1, HH2, HH3, and HH4) in the aspect of gene activity score and linked gene expression, respectively defined as progenitor-related genes (PRGs) (Figs. 4A, S4A–B, Table. S1). Among the intersecting genes, it was observed that 39 genes (Slpi, Ccl2, Anxa2, Cdkn1a, Cd74, Nid1, Ier3, Cd44, Slc25a24, Tgfb2, App, Lgmn, Cpne8, Cystm1, Mmp14, Fus, Palmd, Alcam, Vim, Sox9, Snx7, Arpc1b, Fkbp1a, Agfg1, Adamts1, Pold4, Itgav, 1110038B12Rik, Cnih4, Tead1, Mvp, Slc25a30, Lims1, Myh9, Bzw2, Tsc22d4, Tpd52l2, Flot1, Dsp) were identified as RRGs (Reprogramming/progenitor-related genes) as previous reported48 (Fig. 4A). We also found the intersecting genes involve in various signaling pathways such as ErbB signaling pathway, mTOR signaling pathway, Toll-like receptor signaling pathway, and TGF-beta signaling pathway (Fig. 4B) which are linked to processes involving cell differentiation, proliferation, and progenitor cell growth. These genes expression score also peaked at 8H samples in the PHx Stereo-seq data (Figs. 4C, S4C). High chromatin accessibility of these progenitor-related genes may represent an initial state of dedifferentiation of mature hepatocytes into progenitor cells in response to liver injury signals.

Fig. 4: Multiomic demonstration of progenitor cell characteristics.
figure 4

A Paired heatmaps of progenitor-related genes with high accessibility (GAS) and high integrated expression (GIS) in SH5 compared with homeostasis hepatocytes (HH1, HH2, HH3, HH4). Those genes were called progenitor cell related genes (PRGs). P was calculated by Wilcoxon. B GO term enrichments for genes with high accessibility and high integrated expression in SH5 compared with homeostasis hepatocytes (HH1, HH2, HH3, HH4). C Line plots showing the module score of PRGs for Stereo-seq data. The color of lines representing time point. The nine points on the curve representing liver zonation (1 and 2 pericentral,3–6 midzonal and 7–9 periportal). ***P < 0.001. P was calculated by T-test. Error bars: mean ± SD. Each point represents one zonation layer along the CV-PV axis (Data points correspond to 9 zonation layers (CV-PV axis), each resolved into 10 sub-regions). D K-means clustering (k = 2) of Stereo-seq data based on PRGs module scores identifies distinct expression domains. Color gradient represents PRG expression levels. Deep red represent high expression (PRGhigh) and light red represent low-expression (PRGlow). E Genes expression for Tnfrsf12a and Sox9 (F) in the Stereo-seq data grouped by PRGs module score. P was calculated by T-test. (***P < 0.001). The horizontal line within each violin plot denotes the average gene expression level. G UMAP showing the hepatocytes clusters in the 8 h post-PHx scRNA-seq grouped by PRGs. H Gene expression for Sox9 and Tnfrsf12a across hepatocytes clusters. Color indicates relative expression across all clusters and dot size indicates the percentage of cells in that cluster expressing the gene. P was calculated by Wilcoxon. I The model score of DAGs of SH5 for scRNA-seq hepatocytes clusters. The lines in the middle of the boxplots mean the median. P was calculated by T-test. (***P < 0.001). J The expression of progenitor-related genes (Portion) in scRNA-seq hepatocytes clusters. Color indicates relative expression across all clusters and dot size indicates the percentage of cells in that cluster expressing the gene. P was calculated by Wilcoxon. K GO term enrichments for genes high expression in Cluster2. L The number of shared DAGs between hepatocyte subtypes in PHx model and LPLCs in DDC model. M The module score of DAGs of LPLCs for hepatocyte subtypes. The lines in the middle of the boxplots mean the median. ***P < 0.001 (SH5 vs other hepatocyte subtypes). P was calculated by T-test.

Full size image

The module score analysis was performed using PRGs on spatial transcriptomic data obtained 8 h post-hepatectomy, revealing that regions with high PRGs expression were predominantly localized around the hepatic resection margin (Figs. 4D, S4D). This spatial distribution suggests that intense injury signals may facilitate the reprogramming of mature hepatocytes. Furthermore, these PRGhigh regions exhibited significant upregulation of Tnfrsf12a (Fig. 4E) and Sox9 (Fig. 4F). These findings provide spatial transcriptomic evidence supporting the existence of hepatic progenitor cells following partial hepatectomy. We have further validated the presence of progenitor-like cells at 8 h after PHx at the transcriptome level by scRNA-seq data7. To pinpoint a population of hepatocytes resembling SH5 at the transcriptome level, we employed the PRGs presented in Fig. 4A to cluster hepatocytes of 8 h post-PHx in the scRNA-seq data (Fig. 4G) in which Cluster2 expresses the highest progenitor cell marker Sox9 and Tnfrsf12a (Fig. 4H). Using PRGs to score the scRNA-seq data, we found that Cluster2 exhibited the highest module score among all cell clusters (Fig. 4I). In contrast, SH5’s DAGs exhibit significantly high expression in Cluster2 (Figs. 4J, S4E). The differentially expressed genes in Cluster2 were primarily enriched in signaling pathways such as regulation of cell growth and cell differentiation (Fig. 4K). Hence, we recognized Cluster2 as a progenitor-like cell at the transcriptome level.

Meanwhile, compared to LPLCs derived from the mouse DDC-diet model, we found that SH5 shares the highest number of DAGs with LPLCs in comparison to the other 8 subtypes (Fig. 4L). And SH5 showed the highest module score of DAGs of LPLCs (Fig. 4M). These suggest that SH5 has similar chromatin accessibility characteristics to LPLCs.

Overall, through the integrated analysis of the single-cell chromatin accessibility and transcriptomic datasets, a distinct subset of hepatocytes displaying progenitor cell characteristics was identified in the PHx model.

Regulation of transcription factors maintains homeostasis and proliferation potential of hepatic progenitor-like cells

snATAC-seq represents a potent approach for investigating transcriptional regulation49. We performed transcription factor motif enrichment analysis with SH5’s marker peaks and found that progenitor-like cells were predominantly enriched in Smarcc1, Jund, Junb, Nfe2l2, Jun, Atf2, Cebpg, Nfil3, Hbp1, Cebpb motifs (Fig. 5A). Examination using PHx scRNA-seq data revealed higher expression levels of these transcription factors in hepatocytes during the priming (8H) and proliferation stages (D1, D2) compared to other time points (Fig. 5B), suggesting their potential role in hepatocyte proliferation. It has been found that the Nfe2l2 has the potential to stimulate the proliferation of progenitor cells, thereby initiating liver regeneration21. Using the methods reported in the previous study (Method)50, we predicted the target genes of Nfe2l2 from snATAC-seq data and corroborated their expression using the scRNA-seq data (Fig. 5C, D). The investigation revealed that the target genes Jun, Kras, Crk, Wasf2, Rap1b, Arf6, Clic4, Picalm, Serpina3m, Serpina3n, Insig2, Rela, Ptpn1, Errfi1, Hipk2, Rap1a, Rac1, Sdcbp and Slc39a14 had higher expression level in the initial phase of liver injury (8H) and/or the peak of liver proliferation (D1, D2) (Fig. 5C, D). In endometrial cancer, the nuclear portion of the Serpina3 protein can activate the MAPK/ERK1/2 and PI3κδ pathways by stimulating the NF-κβ signaling pathway and AKT phosphorylation to block the transition from G2 to M which stimulates the tumor growth51. Subsequently, we conducted functional enrichment using the target genes of Nfe2l2 and identified that these genes primarily govern the cellular response to small molecules and external stimuli (Fig. 5E). It further elucidated the possible functions of Nfe2l2 in maintaining cell homeostasis and responding to external oxidative stress20,22.

Fig. 5: Transcription factor regulatory network of hepatic progenitor-like cells.
figure 5

A Transcription factor motifs enriched in SH5 calculated by marker peaks. B The expression of enriched transcription factors in PHx scRNA-seq with only hepatocytes. Color indicates relative expression across all time points and dot size indicates the percentage of cells in that time point expressing the gene. P was calculated by Wilcoxon. C Prioritization of gene targets for Nfe2l2. The x axis shows Pearson correlation between TF motif activity and integrated gene expression for all expressed genes across all hepatocytes; the y axis shows TF linkage score (for all linked peaks, sum of motif score scaled by peak-to-gene link correlation). Color of points indicates hypergeometric enrichment of the TF motif in all linked peaks for each gene. D The expression of marked genes in (C) in PHx scRNA-seq with only hepatocytes. Color indicates relative expression across all time points and dot size indicates the percentage of cells in that time point expressing the gene. P was calculated by Wilcoxon. E GO term enrichments for positive target genes of Nfe2l2. F The lollipop plots representing the positive transcription factors, whose TF motif activity and matched gene expression are positively correlated across hepatocyte subtypes. The dot color indicates the P. G Genes shared by proliferating cells and progenitor-like cells in the PHx model with LPLCs in the DDC model showing in the Venn plot. H Regulatory networks of the AP-1 family and its target genes. I GO term enrichments for the target genes of AP-1 family.

Full size image

In the analysis of transcription factor motifs enriched within marker peaks of hepatic progenitor-like cells, we also observed a notable enrichment of the AP-1 family transcription factors, including Jund, Junb, and Jun (Fig. 5A). And these genes were identified as “positive regulators” whose regulation activity have strong positive correlations with either their gene module scores or linked gene expressions12 (Fig. 5F). Furthermore, when compared to proliferating hepatocytes in the PHx model and LPLCs in the mouse DDC-diet model, all three cell types demonstrated motif enrichments for AP-1 family transcription factors. This implies that AP-1 family transcription factors may have a potentially significant role in the proliferation of hepatocytes. We further predicted their target genes with published method50, and constructed a transcriptional regulatory network for AP-1 family transcription factors (Fig. 5H). Functional analysis of the target genes within this regulatory network revealed significant enrichment in tumor necrosis factor signaling, cell morphogenesis, and cytokine-mediated signaling pathways (Fig. 5I).

Through the analysis of progenitor-like cells, we discovered that the transcription factors such as Nfe2l2 and AP-1 family may play an important role in regulating liver regeneration.

Growth signals and cell communication play important roles in the activation of hepatic progenitor cells

Kupffer cells, HSC, LSEC, and other non-parenchymal cells collectively orchestrate liver regeneration. The autocrine and paracrine activities of cytokines, growth factors, and other signaling molecules within the cellular environment are integral to the process of hepatocyte reprogramming8. To explain the transient appearance of progenitor-like cells in the PHx model rather than their continuous existence as in the DDC-diet model, we compared the DAGs of LPLCs in the DDC-diet model with progenitor-like cells in the PHx model. And we found 564 shared genes and 1357 LPLC-specific genes, in which the shared genes are mainly enriched in the growth factor signaling pathway and ERBB signaling pathway while LPLC-specific genes are primarily enriched in the regulation of cell morphogenesis, cell-cell junction assembly, and tight junction assembly, slightly indicating a cell-cell interaction (CCI) of progenitor-like cells (Fig. 6A). Prior research indicates that intercellular communication is a crucial factor in hepatocyte dedifferentiation via activating progenitor genes8,48. We next utilized published scRNA-seq data of the mouse DDC-diet model13 to compare CCI strength between hepatocytes and LPLCs and it shows that LPLCs exhibited higher levels of incoming and outgoing interaction strength than hepatocytes either at day 17 (D17) after DDC-diet injury or at day 2 (R2) of recovery (Fig. 6B).

Fig. 6: The disparities in the regulation of cell communication, growth signals, and transcription factors between progenitor-like cells and LPLCs.
figure 6

A GO term enrichments for shared genes between progenitor-like cells and LPLCs LPLC-specific DAGs. Inset, the unique and shared DAGs between progenitor-like cells and LPLCs showing in the Venn plot. B The strength of cell communication incoming and outgoing for different cell types in the DDC model snRNA-seq data on D17 (17th day undergoing DDC diet) and R2 (2th day regeneration). The x-and y-axes are the total outgoing or incoming interaction probabilities associated with each cell type, respectively. The size of the dots was correlated with the number of inferred interactions, both outgoing and incoming, for each cell type. C The upset plots illustrating the quantity and comparison of cell interactions between LPLCs and other cell types at D17 and R2 in the DDC model, as well as between Cluster2 (progenitor-like cells in PHx scRNA-seq data) and other cell types in 8H samples (Top). Schematic diagram of the interaction (Bottom). D Cell Interactions between cell types associated with LPLCs. The color representing communication probability. The size of dots representing P. E K-means clustering (k = 2) of Stereo-seq data from PHx (Top) and DDC (Bottom) models based on PRG (progenitor-related genes) and RRG (regeneration-related genes) module scores, respectively. PLH (progenitor-like hepatocytes): High module score regions in PHx model (red). LPLC (liver progenitor-like cells): High module score regions in DDC model (red). SR (surrounding region): 0.5 μm periphery of progenitor zones (white outline). BG (background): Remaining tissue areas (gray). F Violin plots showing differential expression of Pdgfa and Pdgfra across spatial regions in the PHx model (Top) and DDC model (Bottom). ***P < 0.001. P was calculated by T-test. The horizontal line within each violin plot denotes the average gene expression level. G Cell type-specific expression patterns of highly expressed genes in hepatic progenitor cell-enriched regions from the PHx model. H Same as G but from the D17 stage of DDC model. I Same as G but from the R2 stage of DDC model. J The expression of growth factors in the PHx scRNA-seq data (Top) and DDC-diet model scRNA-seq (Bottom). Color indicates relative expression across all time points and dot size indicates the percentage of cells in that time point expressing the gene. ***P < 0.001. P was calculated by T-test.

Full size image

We further compare the CCI between hepatic progenitor-like cells (identified in scRNA-seq of PHx model) and LPLCs in DDC-diet model, and found that the number of ligand-receptor (L-R) pairs in progenitor-like cells interacting with other cell types was lower than that in LPLCs on D17 and at the R2 stage of the DDC model. Notably, LPLCs exhibited a greater number of unique L-R pairs (Fig. 6C). We observed that L-R pairs such as Csf1-Csf1r, which governs the proliferation and differentiation of macrophages52,53, and Cxcl12 (SDF-1) -Cxcr4, known for promoting hepatocarcinoma by enhancing angiogenesis and immune escape54, are specifically present in LPLCs (Figs. 6D, S5A). The interaction between Tnfsf12 (TWEAK) and Tnfrsf12a (Fn14), which is an important signal to promote progenitor cell proliferation9, is also specific in LPLCs (Fig. S5A). Therefore, the absence of these critical CCIs during the priming stage of PHx model may hinder the subsequent activation of progenitor-like cells.

Using spatial transcriptomic data from liver injury models, we defined regions with high expression of PRG or RRG gene signatures as hepatic progenitor cell-enriched zones (PLH in PHx model or LPLC in DDC model) (Figs. 6E, S5B, C). We observed significant synchronous upregulation of Pdgfa-Pdgfra expression in the LPLC-enriched region of the DDC-induced liver injury model. In contrast, no such synchronous significant upregulation of Pdgfa-Pdgfra expression was detected in the PHx liver injury model (Figs. 6F, S5C). Platelet-derived growth factor A (Pgdfa) is a growth factor that interacts with the platelet-derived growth factor receptor α (Pdgfra). Pdgfra, a subtype of the receptor tyrosine kinase PDGFR family, plays crucial roles in cell proliferation, survival, differentiation, and growth55. Furthermore, the other identified CCI pairs showed synchronized upregulation in the DDC-diet model but not in the PHx model (Figs. S5D-E, S6A). Meanwhile, we found that genes highly expressed in the hepatic progenitor cell-enriched region of the PHx liver injury model were primarily upregulated in cholangiocytes and hepatocytes (Fig. 6G). In contrast, genes highly expressed in the LPLC-enriched region of the DDC-induced liver injury model were also significantly upregulated in various immune cells (Fig. 6H, I). This suggests that chronic liver injury leads to the enrichment of diverse immune cells around LPLCs, forming a cellular interaction network that promotes hepatic progenitor cell reprogramming.

The inhibition of hepatocyte proliferation ability in chronic liver injury leads to cellular reprogramming. In our multi-omics data, we observed the upregulation of growth factor Hgf and EGF signaling pathway (Hep: Egfr, Tgfa; NPC: Hbegf)56 during the initiation and proliferation stages of the PHx model in various cell types, supporting the hepatocyte proliferation (Fig. 6J). However, in the DDC model, the expression of these genes is fully or partially inhibited at D17 and R2 corresponding to the proliferative peak during LPLC transformation (Fig. 6J). Therefore, the stimulation of growth signals in the PHx model may also result in the inability of progenitor-like cells to transition into a further activated state.

Therefore, we employed snATAC-seq and scRNA-seq data to certify the adequacy of growth factors and the inadequate CCI hinders the further activation of progenitor-like cells. Consequently, the primary mechanism of PHx liver regeneration relies on the self-renewal of hepatocytes to compensate for the loss of liver parenchyma.

Discussion

Following injury, liver regeneration can be accomplished by hepatocyte self-renewal or by the dedifferentiation of parenchymal cells, which is followed by the differentiation of progenitor cells4,5. The mainstream view holds that liver regeneration after hepatectomy is mainly contributed by the self-renewal of hepatocytes, although there are some opposing voices5. In our recent work, it showed a high progenitor-related gene expression at 8 h post PHx7. In a rat model of 80% subtotal hepatectomy, a study observed a moderate upregulation of hepatic progenitor cell markers, including Sox9 and Tnfrsf12a, following partial hepatectomy-induced liver injury. Furthermore, the study revealed a significant increase in CK19-positive cells (a specific hepatic progenitor cell marker) within the periportal region, suggesting activation of the progenitor cell compartment during liver regeneration57. To further explore the cell type compositions during regeneration in PHx model, we adopted an alternative perspective by investigating chromatin accessibility in addition to the transcriptome. For hepatocytes, we observed a transformation into three distinct functional types: functional hepatocytes with relatively normal metabolic functions (HH1, HH2, HH3, HH4), proliferating hepatocytes (PH9) and stressed hepatocytes (SH5, SH6, SH7, SH8), which transiently appear in the priming stage. Among the stressed hepatocytes, we found that SH5 exhibited high chromatin accessibility of Sox9 and Tnfrsf12a which are upregulated in hepatic progenitor cells, and consolidate its progenitor-like roles with more hepatic progenitor characteristics and compared its chromatin accessibility profile with LPLCs in a mouse DDC-diet model. In addition, we performed dimension reduction on the published scRNA-seq data at 8 h post-PHx with DAGs and DEGs of these progenitor-like cells and found a SH5-corresponding subcluster (Cluster2).

Then, it was asked why progenitor-like cells disappear in the proliferation stage in PHx model other than persisting continuously as in DDC-diet model. We conducted comprehensive comparative analyses between the progenitor-like cells in the PHx model and the LPLCs in the DDC-diet model in aspects of chromatin accessibility and transcriptome gene expression. LPLCs upregulated DAGs are highly enriched in CCI-related pathways. Cell communication analysis with scRNA-seq data also demonstrated a higher CCI strength of LPLC than progenitor-like cells. The Tnfsf12-Tnfrsf12a, Cxcl12-Cxcr4, Pdgfa-Pdgfra and other L-R pairs that facilitates progenitor cells’ expansion are absent between progenitor-like cells and NPCs in PHx model whereas present between LPLCs and NPCs in DDC-diet model9,52,53,54. Utilizing spatial transcriptomic data, we delineated regions enriched with hepatic progenitor cells and identified synchronized upregulation of gene expression, including the Pdgfa-Pdgfra interaction, in the DDC-induced liver injury model, which was not observed in the PHx model. Furthermore, comparative gene expression analysis revealed a notable absence of immune cell aggregation around potential hepatic progenitor cells in the PHx model. The findings collectively highlight the significance of extracellular communication in facilitating the transition of hepatocytes into hepatic progenitor cells. Progenitor-involved liver regeneration lack an adequate growth factor environment as reported28. We observed high expressions of the growth factors Hgf and EGF signal pathway of endothelial cells, HSC, macrophages and hepatocytes at priming stage in PHx model while be completely or partially repressed during the DDC-injured.

Overall, we present the snATAC-seq atlas of mouse PHx model. Cross-model analysis demonstrated the transient appearance of progenitor-like cells during the priming stage. Additionally, we inferred that adequate growth factors and an infertile extracellular NPC environment limited the further activation of progenitor-like cells. However, evidence supporting the existence of progenitor cells in PHx model should be further substantiated.

Methods

Ethical statement

All animal experiments were carried out following ethical regulations regarding animal research and were approved by the Animal Care and Use Committee of the Guangzhou Institutes of Biomedicine and Health under license number IACUC2018038.

PHx experiment

The mice utilized were male C57BL/6 J of specific-pathogen-free (SPF) grade. They were bred and maintained in a facility with regulated temperature and a 12-h light-dark cycle. For the Partial Hepatectomy (PHx) procedure, mice at 12 weeks of age had two-thirds of their liver mass surgically removed. This surgery was conducted between 09:00 and 12:00 h. In brief, the mice were initially anesthetized with 2% isoflurane (RWD, R510-22) in an induction chamber, then moved to the operating pad where isoflurane inhalation sustained anesthesia. A longitudinal incision was made through the skin and muscle, and retractors were utilized to maintain access to the peritoneal cavity. Silk threads were tied near the base of the median and left lateral liver lobes to obstruct blood flow. After securing these ligatures, the lobes were excised with a surgical scalpel. Subsequently, the peritoneum and abdominal incision were stitched closed, and the animals were placed on a heating pad to recover. Once physical activity was observed, the mice were returned to their cages. They were euthanized at predetermined intervals post-resection. Mice that did not undergo treatment (D0) were used as baseline controls.

snATAC library construction and alignment

The snATAC-seq libraries were prepared using the DNBelab C Series Single-Cell ATAC Library Prep Kit (MGI, #1000021878). Briefly, the barcoded snATAC-seq libraries were generated through a workflow that included transposition, droplet encapsulation, pre-amplification, emulsion breakage, capture bead collection, DNA amplification, and purification. Paired-end sequencing was carried out on the DIPSEQ T1 sequencer at China National Gene Bank (CNGB). And then we used bwa (https://github.com/lh3/bwa) to align snATAC-seq data to the mm10 reference genome. d2c (https://github.com/STOmics/d2c) was used to generate fragment files.

snATAC-seq quality control and clustering analysis

After aligning the ATAC-seq fragment data, we proceeded to process it using the ‘ArchR’ R package (v.1.0.2)12. Subsequently, we computed the number of unique sequenced fragments and transcription start site (TSS) enrichment for each cell (TSS enrichment >4, Unique fragments >2000). This calculation serves as a signal-to-noise metric for ATAC-seq data. To remove doublet cells, we use the addDoubletScores and filterDoublets functions in ArchR (filterRatio = 8). In addition, we conducted cell filtration in the region between two distinct clusters on the UMAP.

We then used ArchR’s implementation of iterative LSI dimensionality reduction using the addIterativeLSI function (varFeatures = 15,000, dimsToUse = 1:30). Using addHarmony function to remove the batch effect between different samples and sequence libraries. And then we used addClusters and addUMAP functions to identified clusters and gained a two-dimensional representation of the data (resolution = 0.8, nNeighbors = 30, minDist = 0.5, metric = cosine).

The analysis of DAGs of different cell types was used the getMarkerFeatures function with GeneScoreMatrix and GeneIntegrateMatrix representing the accessibility around each gene and presumed gene expression by integrating with scRNA-seq, respectively.

To integrate the snATAC-seq and corresponding scRNA-seq, we used the ArchR’s function addGeneIntegrationMatrix with sampleCellsATAC = 25000, sampleCellsRNA = 25,000, and nGenes = 3000.

Calling peaks and transcription factor motif annotation

To get the accessible peaks in snATAC-seq, we used the ArchR’s addPeakMatrix function. And getMarkerFeatures. The gene elements annotation of different accessibility peaks was gained from ArchR project. And then we used addMotifAnnotations function to annotate transcription factor motif information (motifSet = cisbp).

Functional enrichment analysis

The differentially accessible and predicted differentially expressed gene sets across distinct cell types, obtained using the getMarkerFeatures function in ArchR, were subjected to Gene Ontology and pathway enrichment analysis (P < 0.01, log2FoldChange > 0.25). The analysis was performed using the R packages topGO (https://bioconductor.org/packages/release/bioc/html/topGO.html) and clusterProfiler58,59,60. We focused on the most statistically significant pathways (P < 0.05) and visualized them using the R package ggplot261.

Add module score analysis

Following the identification of marker genes for different cell types, the resulting gene sets were used to score each other cell types using the function AddModuleScore in R package Seurat62.

scRNA-seq re-cluster analysis

Clustering the hepatocyte 8 h post-PHx in scRNA-seq was performed by R package Seurat62. We used the marker genes found in SH5 as features to re-cluster the cells. First, hepatocytes at 8 h post-PHx were subset using the subset function. These cells were then subjected to dimensionality reduction and clustering via the Seurat pipeline, including RunPCA, FindNeighbors, FindClusters, and RunUMAP. For PCA computation, we used marker genes identified in SH5 as features. Key parameters were set as follows: dimensions (dims) = 1:10, resolution = 0.5. The top 25 differentially expressed genes (DEGs) in Cluster2, identified by the FindMarkers function of Seurat and overlapping with SH5-associated genes, were visualized in the figures.

The ranking of transcription factor (TF) motifs

The differently accessible peaks of SH5 compared with HH1, HH2, HH3, and HH4 were calculated using getMarkerFeatures function of ArchR. And then ArchR’s peakAnnoEnrichment function was applied to annotate and calculate the enrichment level (cutoff = “FDR < = 0.1 & Log2FC >= 0.5”)12. Visualization was sorted by P using the R package ggplot261.

Identification of potential regulatory target genes of TF regulators

To identify potential target genes of a TF, we collected the method published in ‘Integrated single-cell chromatin and transcriptomic analyses of human scalp identify gene-regulatory programs and critical cell types for hair and skin diseases’ and get the script from GitHub (https://github.com/GreenleafLab/scScalpChromatin).

Identification of positive regulators

The TFs whose gene expression is positively correlated to changes in the accessibility of their corresponding motif termed as positive regulators. We used ArchR’s correlateMatrices function with useMatrix1 = GeneScoreMatrix and useMatrix1 = GeneIntegrationMatrix to find the positive regulators12.

Delineation of spatially enriched hepatic progenitor cell niches

Using the PRGs identified in this study and RRGs in LPLCs, we performed AddModuleScore analysis on the spatial transcriptomic data from both the PHx and DDC models, respectively. Regions with high module scores were then classified as hepatic progenitor cell-enriched zones (PLH/LPLC) via K-means clustering (k = 2). Subsequently, the 0.5 μm periphery of these enriched zones was defined as the surrounding region (SR), while the remaining areas were designated as the background region (BR). For the convenience of presentation, minimal values (Gene expression > 0) were filtered out in the violin plots of gene expression, and the P were calculated using the T-test.

Cell-cell communication analysis

CellChat was performed to explore cell-cell communications. The normalized gene expression matrix served as the input for CellChat analysis using default setting. The CellChat database CellChatDB.mouse was used63,64.

Statistics and reproducibility

Differentially accessible genes (DAGs) and peaks (DAPs) were identified by applying a significance threshold of P < 0.01. In the spatial transcriptomic analysis of liver zonation, differential gene expression patterns along the central vein (CV) to portal vein (PV) axis were evaluated by stratifying the tissue into 90 layers based on zonation gene scores, followed by t-test statistical analysis for significance assessment. All statistical analyses were performed as described in the respective methods sections, with specific tests indicated for each experimental approach.

Reporting summary

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