Single-nucleus transcriptomic atlas of postnatal camel liver development identifies candidate adaptive features

Abstract

Background

Camels possess distinctive physiological traits related to metabolism and environmental resilience, highlighting the need to characterize liver cellular composition in this species.

Results

We used single-nucleus RNA sequencing to generate a developmental atlas of the Bactrian camel (Camelus bactrianus) liver using one individual from each of the newborn, weaning, and adult stages, analyzing 57,036 cells and identifying 19 cell types, including hepatocytes, liver sinusoidal endothelial cells (LSECs), and immune cells. We identified a unique subtype of HGFAC+ LSECs that was not observed in the available cattle or human datasets, characterized by high expression of hepatocyte growth factor activator (HGFAC), which is known to activate hepatocyte growth factor (HGF). Hepatocyte pseudotime ordering was consistent with a putative transition from proliferation-associated states toward metabolically annotated states. In the samples collected in this study, Kupffer cells were more represented in the newborn sample, whereas T-cell subsets were more represented in the adult sample. Exploratory cross-species mapping indicated shared transcriptional features between several adult camel and cattle immune cells.

Conclusions

These results suggest liver biochemical features that may contribute to camel resilience to arid conditions, providing a resource for further studies of camel liver biology and livestock adaptation to harsh environments.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12864-026-13137-4.

Introduction

The liver is a vital organ in mammals that regulates metabolism, detoxification, and immune function [1, 2]. Postnatal liver development represents a critical transition from maternal dependence to autonomous metabolic regulation, accompanied by extensive cellular differentiation and functional maturation [3]. While these processes have been well characterized in model organisms such as mice and humans, they remain poorly understood in livestock species such as the Bactrian camel [4]. The Bactrian camel (Camelus bactrianus) is adapted to arid and semi-arid ecosystems and can cope with dehydration, temperature fluctuations, and irregular nutrient availability [5, 6]. These adaptations are closely associated with hepatic functions involved in energy metabolism and immune regulation. Although genomic studies have identified candidate genes associated with desert adaptation, the cellular and transcriptional mechanisms underlying postnatal liver development in camels remain largely unexplored.

Recent advances in single-nucleus RNA sequencing (snRNA-seq) have enabled the resolution of cellular heterogeneity, transcriptional states, and developmental trajectories in complex tissues [7, 8]. In liver research, such methods have revealed cellular heterogeneity, lineage relationships, and functional zonation across species [9, 10]. However, comparable single-cell or single-nucleus atlases are lacking for camel liver, especially across key postnatal stages such as neonatal, weaning, and adulthood.

In camels, the liver is likely important for maintaining metabolic balance under fluctuating environmental and nutritional conditions [11, 12]. Genomics highlighted divergences between camel lineages, including desert-tolerance genes, whereas cellular transcriptomic data remain limited [6]. Liver sinusoidal endothelial cells (LSECs) and Kupffer cells are key non-parenchymal cell populations that regulate blood filtration, immune tolerance, phagocytosis, and inflammatory responses in the liver [13, 14]. However, their developmental dynamics and functional specialization in camels remain largely unexplored.

This study used snRNA-seq maps of livers from newborn, weaned, and adult camels to identify 19 cell types, describing cell type-specific TF modules, putative hepatocyte trajectories, LSEC subtypes, and predicted intercellular communication networks potentially related to hepatic homeostasis. These findings offer a useful resource for further investigation of camel liver biology and the cellular processes involved in postnatal hepatic development.

Materials and methods

Ethics declarations

All experimental procedures in this study were strictly adhered to local legislations and approved by the Institutional Animal Care and Use Committee of Northwest A&F University (Approval No. XN2024-0416). The sampling process was rigorously monitored by members of the Institutional Animal Care and Use Committee of Northwest A&F University, ensuring the minimal distress of animals during experimentation.

Collection of camel liver

Liver samples were collected from three male Bactrian camels, aged approximately 10 days, 6 months, and 20 months, from a slaughterhouse in Gansu, China. The liver tissues were isolated, quickly rinsed with the pre-cooled solution of 1 × PBS (Ambion, #AM9624), and placed on an ice-cold board for dissection. Each tissue was cut into 5–10 pieces, with each piece weighing 50–200 mg. The liver samples prepared for snRNA-seq were processed using the same procedure and then placed into cryovials (Corning, #430659) and stored in liquid nitrogen until lysis for nucleus extraction.

Single-nucleus suspension preparation

Single-nucleus isolation was performed as previously described [15, 16]. In brief, frozen tissues were homogenized in a homogenizer in 1 mL lysis buffer consisting of 250 mM sucrose (Ambion), 10 mg/mL bovine serum albumin (BSA), 5 mM MgCl2 (Ambion), 0.12 U/µL RNasin Plus (Promega, #N2115), and 1 × Complete Protease Inhibitor Cocktail (Roche, #11697498001). The tissue underwent two additional rounds of homogenization and processing for uniformity, and the homogenized tissue was subjected to a 40 μm cell strainer. The filtered nuclei were pelleted at 500 × g for 5 min at 4 °C, and the pellets were resuspended in 1 mL buffer containing 10 mg/mL BSA, 3 mM CaCl2, 10 mM Tris-HCl, 320 mM sucrose, 0.1 mM EDTA, 2 mM magnesium acetate, 1 mM DTT, 1 × Complete Protease Inhibitor Cocktail, and 0.12 U/µL RNasein, followed by centrifugation at 500 × g for 5 min at 4 °C to pellet the nuclei. The isolated nuclei were then used for library preparation.

The DNBelab C Series Single-Cell Library Prep Set (MGI, #1000021082) was employed [15], following previously established protocols. Specifically, the single-nucleus suspension was first subjected to droplet generation, emulsion breakage, bead collection, reverse transcription, and cDNA amplification to generate barcoded libraries. After the quality control of obtained cDNAs, indexed libraries were constructed according to the manufacturer’s protocol. Subsequently, sequencing was conducted on the DNBSEQ-T1 platform.

Processing of snRNA-seq data

The snRNA-seq data were filtered, and the gene expression matrix was obtained using the DNBelab C Series scRNA analysis software (v.1.0.1, MGI) (https://github.com/MGI-tech-bioinformatics/DNBelab_C_Series_scRNA-analysis software). The reference genomes of camel (Ca_bactrianus_MBC_1.0) used for sequence alignment were downloaded from NCBI. We created a Seurat object using the Seurat [17] R package (v.4.3.0) from three tables including genes, barcodes, and the raw UMI count generated by the DNBelab C Series scRNA analysis software. Cells were only retained if the number of detected genes was greater than 200 and less than 3 500 and the percentage of detected mitochondrial transcripts from MT genes was less than 7 (Supplementary Figure S1A). The “NormalizeData, ScaleData and FindVariableFeatures” functions were used to standardize the raw count matrix and identify the top 2 000 most variable genes. The “RunPCA, RunUMAP, FindNeighbors, and FindClusters” functions were used to reduce dimensionality and for clustering of data, the DoubletFinder R package was used to remove doublets (Supplementary Figure S1B, Supplementary Table S1). “FindAllMarkers and FindMarker” functions were used to identify DEGs across or between clusters with the options “min.pct = 0.25, logfc.threshold = 0.25”. Multiple test correction for P value was performed using the False Discovery Rate (FDR) method, and adjusted P < 0.05 was considered statistically significant. Further, cell-type identities were assigned using canonical cell-type markers.

Analysis of TF regulation

TF regulatory activity was inferred using the command-line workflow of pySCENIC (v.0.12.1) [18], as previously described. Because pySCENIC by default supports only a limited number of species, we first mapped camel genes to their one-to-one human orthologues and then ran pySCENIC using the human cisTarget motif and ranking databases, thereby indirectly inferring TF activity in camel cells. To compare the cell-type specificity of TF regulons, we applied the Regulon Specificity Score (RSS) as previously defined. On this basis, TF modules in camel cells were identified following a four-step procedure: (1) regulon activity inference, where SCENIC was used to estimate the activity of each TF in individual cells using AUCell scores; (2) correlation analysis, in which Pearson correlation coefficients (PCCs) were computed between the activity scores of all TF pairs; (3) calculation of the Connection Specificity Index (CSI), where for a given TF pair (a and b), CSI was defined as the proportion of TFs connected to either a or b whose PCCs with a or b were lower than the PCC between a and b; and (4) module identification and network construction, where the CSI matrix was subjected to hierarchical clustering (Ward’s method) to delineate TF modules and construct a co-regulatory TF network based on TF pairs with high CSI values. TF–TF interactions within and between modules were visualized using Cytoscape, and AUCell scores were further used to assess the activity of each TF module across different cell types or conditions.

Subclustering analysis

Hepatocytes, endothelial cells, and immune cells from snRNA-seq were independently extracted for subclustering analysis. We applied CCA [17] for batch correction and dimension reduction. Briefly, the dimensionality of the data was reduced by principal component analysis (PCA) with 30 components was first performed on the top 2,000 highly variable genes, followed by batch correction on SampleID. Cell clusters were identified using the Seurat (resolution = 0.3 for hepatocytes, resolution = 0.5 for endothelium, and resolution = 1.2 for immune). Finally, cell clusters were annotated based on differential marker gene expression.

Pseudotime trajectory analysis

Monocle3 [19] (v.1.3.1) was used to reconstruct the cellular differentiation trajectory of hepatocytes. Specifically, subsets were separated into large and well-divided partitions using the cluster_cells function and the principal graph was fitted in each partition using the learn_graph function. The resulting principal graph, revealed on the UMAP as ‘skeleton lines’, indicates the differentiation trajectories.

Monocle2 [20] (v.2.26) was used to infer the state transition of the endothelial cells. Genes with q < 0.05 were identified as DEGs using the differentialGeneTest function and sorted according to q value using the setOrderingFilter function. The pseudotime trajectory was constructed using the DDRTree algorithm with default parameters. The dynamic expression changes of selected marker genes in pseudotime were visualized using the plot_genes_in_pseudotime and plot_pseudotime_heatmap functions.

Function enrichment analysis

GO and KEGG analyses were performed using KOBAS [21]. The GO terms and KEGG pathways of selected genes were enriched in the KOBAS databases, using the enrichGO and enrichKEGG functions, respectively, with a threshold parameter of pvalueCutoff = 0.05. GSVA (v.1.49.6) was used to investigate functional differences between cell subtypes. For Gene Set Enrichment Analysis (GSEA) [22], human Hallmark Gene Sets were downloaded from the MSigDB database, and only human gene sets with a 1:1 orthologous relationship to camel were retained. Further filtering was then performed, retaining gene sets with more than 10 entries. The enrichment analysis parameters were set as follows: nPermSimple = 10,000, pvalueCutoff = 0.05, minGSSize = 20, maxGSSize = 1000, pAdjustMethod = “BH”. Results visualization was performed using the enrichplot [23] R package (v.1.18.4).

Assessment of hepatocyte proliferative activity

To compare the proliferative levels of hepatocytes across species and age groups, we evaluated cell cycle status and proliferation based on the cell cycle gene sets provided in the Seurat package. Briefly, S phase and G2/M phase marker genes were extracted from the built-in dataset cc.genes. updated.2019, and Seurat’s “CellCycleScoring” function was used to calculate the S.Score and G2M.Score for each cell.

Based on these scores, we defined a simplified binary “proliferation” index: cells with S.Score > 0 or G2M.Score > 0 were classified as proliferative, whereas the remaining cells were classified as non-proliferative [24]. Using species as the grouping factor, we then counted the total number of hepatocytes (n) and the number of proliferative cells (k) in each group, calculated the proliferation rate (k/n), and derived its 95% confidence interval using a normal approximation to the binomial distribution. Because biological replication was not available for the camel stages, these cell-cycle scores were used only for descriptive visualization and were not interpreted as formal evidence of inter-individual, developmental-stage, or species-level differences.

Cross-species comparisons of single-cell transcriptomic data

One-to-one orthologs among camel, cattle, and human genes were identified using EggNOG mapper [25], with the following parameters: taxonomic scope = vertebrates, and orthology restrictions = One-to-one orthologue, and other parameters were set as default. The predicted name column was compared across species to identify genes for cross-species comparisons, and only one-to-one orthologs were retained (Supplementary Table S2). Transcriptomic similarity of annotated cell types across species was explored using MetaNeighbor (v.1.181) [26].

Cell-cell communication analysis

Cell-cell communication analysis was performed using the CellChat R package [27] (v.1.1.3) for each developmental stage based on snRNA-seq data. Prior to analysis, cell types represented by fewer than 10 cells in a given sample were excluded. Because camel genes are not directly supported in the CellChat database, ortholog mapping was performed to match camel genes to the reference ligand-receptor database used in the analysis. Normalized expression matrices were used as input for CellChat. Significant ligand-receptor interactions were inferred using the default CellChat statistical framework and were retained at P < 0.05. Interactions between hepatocytes and non-parenchymal cells were extracted for downstream visualization using netVisual_bubble. Because each developmental stage was represented by a single biological sample, the inferred communication networks were interpreted as exploratory and hypothesis-generating rather than definitive.

Multiplexed immunofluorescent staining

Adult camel liver tissue Sect.  (8 μm) were deparaffinized and rehydrated through a graded series of xylene and ethanol. Antigen retrieval was performed by heating the sections in antigen retrieval buffer in a pressure cooker for 2.5 min, followed by natural cooling. Endogenous peroxidase activity was quenched with 0.3% hydrogen peroxide in methanol for 15 min. After washing with PBS (pH 7.4), sections were blocked with 5% BSA for 1 h at room temperature. Sections were then incubated overnight at 4 °C with a mixture of primary antibodies: Rabbit anti-STAB2 (1:200, Immunoway #YN2255) and Rabbit anti-HGFAC (1:200, Affinity #DF9570). After PBS washing, sections were incubated with secondary antibodies for 1 h, counterstained with DAPI, mounted, and imaged by confocal microscopy.

Results

Integrating snRNA-seq maps of the developing liver of camels

To investigate the cellular atlas of postnatal camel liver development, we generated single-nucleus RNA sequencing (snRNA-seq) data from three time points: newborn, weaning, and adult stages. Following rigorous quality control, we retained expression profiles for 57 036 cells (Fig. 1A). We then performed normalization, clustering, dimensionality reduction, and annotation (Supplementary Figure S2A, B), identifying 19 cell types based on marker genes (Fig. 1B, Supplementary Table S3): hepatocytes, liver sinusoidal endothelial cells (LSECs), T cells, NK cells, smooth muscle cells, Kupffer cells, cycling LSECs, vascular endothelial cells, cholangiocytes, plasma cells, HGFAC+ LSECs, monocytes, plasmablasts, mesothelial cells, B cells, cycling Kupffer cells, plasmacytoid dendritic cells, hepatic stellate cells, and cycling hepatocytes. Hepatocytes constituted approximately 50% of all cells, consistent with prior studies on liver composition [28] (Fig. 1C). We also detected three cycling cell clusters—cycling LSECs, cycling Kupffer cells, and cycling hepatocytes—which expressed their respective lineage markers alongside elevated proliferation genes MKI67 and TOP2A (Fig. 1D); In this dataset, cycling Kupffer cells and cycling hepatocytes represented larger fractions in the newborn sample, whereas cycling LSECs represented a larger fraction in the adult sample (Fig. 1E). Similarity analysis across developmental stages revealed extensive transcriptional similarities among immune cells, endothelial cells, and stromal cells in newborn and weaning samples, while epithelial cells and hepatocytes showed transcriptional similarities between weaning and adult samples (Fig. 1F).

Fig. 1.

Single-nucleus RNA sequencing (snRNA-seq) identify hepatic cell types in the developing liver. A UMAP visualization of 57 036 cells/nucleus colored by 19 cell types. B Cell counts (left) and cell type composition by time points using snRNA-seq data for each cell type (right). C Heatmap depicting manually selected marker genes for all identified cell types. D Overview of the MKI67 and TOP2A expression in all cells. E Percentage of cycling hepatocytes (left), LSEC (middle) and immune cells (right). F Heatmap shows the correlation between cell types at different developmental stages of camels based on expression levels

Identification of cell type-specific TFs in liver

To determine the key regulatory factors governing postnatal liver development, we performed gene regulatory network analysis on the same dataset using pySCENIC. By computing regulon specificity scores (RSS), we assessed activities linked to major cell types, identifying 319 cell-type-specific transcription factors (TFs) (Fig. 2A; Supplementary Table S4). These encompass established regulators such as HNF4A, a central modulator of hepatocyte identity and metabolism that controls lipid, glucose, and drug metabolism genes [29]; FOXA1/2/3, which activate liver-specific enhancers to facilitate differentiation and homeostasis [30]; and ONECUT1, which governs bile homeostasis, glucose transport, and lipid metabolism [31, 32].

Fig. 2.

Cell type-specific TFs of major hepatic cell types. A Heatmap shows the cell type-specific highly active transcription factors for each cell class. B The co-regulation heatmap, calculated based on Connection Specificity Index (CSI), shows the activity correlations of TF regulon across all cells. Nine co-regulatory modules (Modules 1–9, marked with red boxes) were identified through hierarchical clustering. C The transcription factor co-expression network diagram of the regulatory modules. Node size represents the degree of connection, color represents module affiliation, and edges represent regulatory relationships. Only CSI > = 0.8 is shown. D Heatmap shows the correlation between each module and each cell class. E Heatmap shows the dynamic changes in the activity of the nine regulatory modules at different developmental stages. The right side enriches some GO terms for each module

Recognizing that TFs often synergistically regulate target genes during development, we applied the Connection Specificity Index (CSI) to compare regulon activity score (RAS) distribution similarities across cell types, delineating combinatorial regulatory patterns. CSI-based similarity clustering yielded nine modules with pronounced co-regulatory traits (Modules 1–9) (Fig. 2B, C).

Examination of module preferences across cell types showed that M1 and M9 were specifically associated with immune processes, as they exhibited elevated Z-scores exclusively in immune cells while remaining low in other cell types. In contrast, M7, despite showing a relatively high Z-score in immune cells, reached its peak enrichment in stromal cells, suggesting that it is not immune-specific but rather associated with stromal-related functions or shared processes across multiple cell types (Fig. 2D). GO enrichment analysis of representative modules further suggested cell type-associated functional tendencies. For example, M3 was linked to epithelial differentiation-related terms, M6 to hepatocyte metabolic and maturation-related processes, M7 to stromal-associated regulatory programs, and M8 to endothelial-related developmental functions (Fig. 2E). Overall, these data indicate that different cell types exhibit distinct regulon programs during postnatal liver development.

Identification of LSEC subtypes and functional changes during development

Endothelial cells (ECs) play an important role in maintaining liver homeostasis [33]. We extracted and reclustered all ECs, including liver sinusoidal endothelial cells (LSECs), and annotated cluster identities based on marker gene expression (Fig. 3A, B). Seven endothelial cell subtypes were identified: central vein ECs, portal vein ECs, lymphatic ECs, cycling LSECs, central LSECs, HGFAC+ LSECs, and PTPRC+ LSECs (Fig. 3C). LSECs comprised the majority of hepatic ECs (Fig. 3B), aligning with prior reports [28]. Compared with portal vein ECs (PVECs), central vein ECs (CVECs), and lymphatic ECs (LECs), LSECs exhibited the highest transcriptional diversity during liver development in this dataset [34]. In camel livers, we identified two special LSECs, HGFAC+ LSECs and PTPRC+ LSECs. HGFAC+ LSECs were characterized by high expression of HGFAC, which encodes a serine protease that activates hepatocyte growth factor (HGF) by proteolytic cleavage; this population was not detected in the human, mouse, or cattle LSEC datasets analyzed here (Fig. 3D, Supplementary Figure S3A–C, Supplementary Table S5). High expression of HGF was also found in multiple LSEC subtypes (Supplementary Figure S3D). In the datasets analyzed, adult camel hepatocytes showed higher cell-cycle-associated scores than the cattle and human hepatocyte datasets (Fig. 3E, F; Supplementary Figure S3E). PTPRC+ LSECs expressed immune-related markers, a pattern compatible with an immune-interactive endothelial state. GSVA enrichment highlighted immune-related pathways in PTPRC+ LSECs, including primary immunodeficiency and glycosylphosphatidylinositol (GPI)-anchor biosynthesis (Supplementary Table S6).

Fig. 3.

Identification of LSECs cell subtypes and functional changes during development. A UMAP embeddings of all endothelial cells colored by time point. B UMAP embeddings of all endothelial cells colored by subtype. Bar plot showing time point compositions of each subtype. C Dot plot displaying selected markers for each subtype. The size of the dot represents the proportion of cells in which markers are detected. The color of the dot represents the average gene expression level. D Multiplex immunofluorescence staining for HGFAC+LSEC markers (STAB2 and HGFAC). Scale bar: 50 μm. E Violin plot shows the S score of camel, cattle, and human hepatocytes. F Box plot shows the proliferation score for camel, cattle, and human hepatocytes. G LSECs development trajectory colored by pseudotime and time points. H Heatmap depicting trends of significant genes as a function of inferred pseudotime from G. Genes are clustered into three modules based on expression profiles

For systematic developmental analysis of camel LSECs, we isolated LSECs from ECs and used 2 000 highly variable genes for dimensionality reduction and trajectory inference, resulting in a pseudotime ordering that broadly aligned with the sampled stage labels (Fig. 3G). We identified differentially expressed genes along pseudotime (Supplementary Table S7); the top 500 clustered into three modules via trajectory heatmap analysis, showing temporal distribution across developmental stages (Fig. 3H). Module3 (newborn) was enriched in Rap1 signaling, adherens junction/focal adhesion, axon guidance, actin cytoskeleton regulation, Fc gamma R-mediated phagocytosis, and sphingolipid metabolism (representative genes: ACTG1, ARPC2, CALM1/2), suggesting involvement in adhesion, cytoskeletal organization, and receptor-mediated phagocytic processes. Module1 (weaning) was enriched in ribosome, endoplasmic reticulum protein processing, MAPK signaling, antigen processing/presentation, and necroptosis (RPS and RPL family genes, HSPA1A and HSP90AA1, JUN and FOS, PSME1 and RIPK1), indicating increased translation, protein folding, stress response, and antigen-processing activity. Module2 (adult) was enriched in complement/coagulation cascades, PPAR signaling, fatty acid metabolism/degradation, cholesterol metabolism, and unsaturated fatty acid biosynthesis (C1R and C3, FGA and FGB, ACACA, SCD, FADS1 and FADS2, ACADM, APOA2 and APOB), reflecting a shift toward lipid metabolism and humoral factor synthesis. Overall, the pseudotemporal pattern suggested stage-associated functional changes in LSECs during postnatal development.

Functional heterogeneity of hepatocytes from proliferation to fatty acid metabolism during development

Hepatocytes are the predominant cell population in the liver and perform a wide range of metabolic functions throughout postnatal stages. To explore hepatocyte heterogeneity in postnatal camel liver, we isolated parenchymal cells from newborn, weaning, and adult phases, yielding eight subclusters (hep0-7) through clustering analysis (Fig. 4A, B; Supplementary Figure S4A). These subclusters grouped into two primary expression patterns: one encompassing hep0-3, enriched in p53 signaling, Notch signaling, and focal adhesion, mainly related to proliferation; the other including hep4-7, enriched in mature hepatocyte functions like glycerolipid metabolism, unsaturated fatty acid biosynthesis, and pentose phosphate pathway (Fig. 4C; Supplementary Table S8). Hep0 represented a larger fraction of hepatocytes in the newborn and weaning samples and showed higher CytoTRACE and proliferation-associated scores (Fig. 4D). It also exhibited peak enrichment in mismatch repair pathways responsible for detecting and correcting base insertion, deletion, and misincorporation errors during DNA replication and recombination (Supplementary Table S9). Additionally, in hep0, fatty acid metabolism-related genes showed sample-associated expression patterns, with PPARA and LPIN2 showing lower expression and ACACA and ACADL showing higher expression in the weaning sample within this dataset (Supplementary Figure S4B). Hep7 was largely confined to the weaning phase and was highly enriched in RNA polymerase, ribosome, protein export, and unsaturated fatty acid biosynthesis pathways, which may reflect transient metabolic remodeling during the weaning transition.

Fig. 4.

Functional heterogeneity of hepatocytes from proliferation to fatty acid metabolism during development. A UMAP embeddings of all hepatocytes colored by annotated subtype. B Bar plot showing time point compositions of each hepatocyte subtype. C Heatmap showing the GO terms and KEGG pathways of upregulated DEGs from each hepatocyte subtype using single-sample gene set variation analysis. D Box plot shows the CytoTRACE Score for each hepatocyte subtype. E UMAP plot depicting the developmental trajectory of hepatocyte subtypes over a pseudotime series. F Heatmap depicting trends of significant genes as a function of inferred pseudotime from (e). Genes are clustered info two modules based on expression profiles. Enriched pathways for each module were shown in the right

To elucidate hepatocyte transcriptomic dynamics, we conducted trajectory inference using 2 000 highly variable genes for dimensionality reduction and analysis. Because hep0 was predominantly represented in the neonatal and weaning stages, showed the highest CytoTRACE ranking, and exhibited the strongest proliferation-associated signature, including high expression of MKI67 and TOP2A (Supplementary Figure S4C), it was selected as a putative starting state for trajectory inference. This analysis revealed a bifurcation into two branches: one comprising proliferation-focused hep0-2 and the other progressing toward mature hepatocyte states emphasizing amino acid metabolism (Fig. 4E). Differential expression analysis along pseudotime identified genes with significant changes. Based on pseudo-time patterns, 509 differentially expressed genes were screened and clustered into two modules (Fig. 4F), with enrichment analysis indicating a functional transition from metabolic sensing and structural pathways (e.g., axon guidance, Rap1 signaling, and AMPK signaling) to mature parenchymal roles (e.g., complement and coagulation cascades, cholesterol metabolism, and PPAR signaling).

Characterization of immune cell diversity during postnatal hepatic development

The liver harbors abundant innate and adaptive immune cells critical for host immune responses [35]. From our snRNA-seq data across camel liver developmental stages, we extracted and further subclassified all immune cells, identifying two NK cell subtypes, five T cell subtypes, and three Kupffer cell subtypes (Fig. 5A, B). As development progresses, the composition of immune cells shifts from Kupffer cells to T cells [35] (Fig. 5C). Furthermore, monocytes and macrophages form a continuous expression profile, similar to that in humans and mice [36].

Fig. 5.

Diverse immune cell populations during postnatal liver development. A UMAP embeddings of all immune cells colored by time point. B UMAP embeddings of all immune cells colored by subtype. C Bar plot showing time point compositions of each subtype. D Heatmap depicting manually selected marker genes for each subtype. E Dendrogram of immune cell types in camel and cattle based on the AUROC scores acquired from MetaNeighbor analyses within and between species. F Heatmaps showing different expressions of genes involved in B cell signaling pathways in camel and cattle immune cell types respectively. G The volcano plot shows differentially expressed genes in the Kupffer1 and Kupffer2; red indicates differentially expressed genes highly in the Kupffer1, and blue indicates differentially expressed genes highly in the Kupffer2

We delineated two NK subtypes (Fig. 5D): one subtype corresponded to conventional NK cells that specifically express PDGFD, which enhances NK effector functions in humans via PDGF-D-PDGFRβ signaling that bolster IL-15-mediated survival; the other subtype corresponded to tissue-resident NK cells (trNK cells), which are enriched in TNFSF9, CXCR6, FCERT1G, EOMES and KLRB1, and are consistent with the characteristics of human trNK cells [34]. These trNK cells also highly expressed FGL2, which has been implicated in viral defense regulation [37], and showed peak abundance in adults.

To dissect Kupffer cell (KC) heterogeneity during liver development, we identified this population using canonical markers, including scavenger receptors and immunomodulators (e.g., VSIG4, TIMD4), general macrophage indicators (CD14, CD68, CSF1R), complement genes (C1QA, C1QB), and hemoglobin metabolism molecules (CD163, CD5L). Subclustering revealed three functionally distinct groups: cycling KCs with elevated MKI67 expression, and two conventional KC subtypes expressing classic markers. Cycling KCs peaked in newborns, and they were less represented in the weaning sample and rare in the adult sample within this dataset (Fig. 5C). Conventional KCs decreased overall but persisted at stable levels in adults. Cycling KCs may contribute to the expansion and maturation of the hepatic macrophage pool during late embryonic and early postnatal development. As the body matures, KCs appear to transition from a predominantly proliferative state to a more homeostatic state focused on immune surveillance, which may explain their decreased proportion. These stage-associated patterns may point to candidate processes involved in hepatic immune maturation.

We annotated five T cell subtypes and characterized their developmental shifts. Exhausted CD8 T cells (Tex_CD8T) highly expressed canonical exhaustion markers TIGIT, LAG3, and HAVCR2 (TIM-3) [38]. This subtype was not detected in the newborn sample, was detected in the weaning sample, and represented a larger fraction in the adult sample (Fig. 5A, B). γδ T cells, defined by elevated SOX13 and BLK, declined from newborn to weaning but were most abundant in adult liver. Naive CD4 T cells were characterized by high expression of CCR7, SELL, LEF1, MYC, and TCF7, genes associated with memory T cell migration and homing [39]. Their proportions remained comparable between newborn and weaning stages but were more represented in the adult sample (Fig. 5A, B), consistent with enhanced migratory capacity and long-term surveillance in the mature hepatic immune system. Finally, double-positive CD4CD8 T cells expressed high THEMIS and BCL11B, linked to T cell developmental regulation [40].

Subsequently, we compared adult camel and cattle immune cells using cattle liver scRNA-seq data [41], extracting adult camel and cow datasets where cattle KCs also formed three clusters (Supplementary Figure S5). MetaNeighbor analysis suggested that several camel immune subtypes share transcriptional features with their cattle counterparts. Notably, camel exhausted CD8 T cells clustered tightly with NK cells rather than with other T cell populations (Fig. 5E). This conserved mammalian pattern indicates adoption of an NK-like cytotoxic program by exhausted T cells under chronic antigenic stimulation [42]. Cattle Kupffer3 clustered with camel cycling KCs on the B cell branch (Supplementary Table S10). Differential analysis between this branch and others enriched for B cell signaling pathways, with genes like PLCG2 and CARD11 highly expressed in cattle Kupffer3 and camel cycling Kupffer (Fig. 5F). Camel Kupffer1 clustered closely with cattle Kupffer1, and camel Kupffer2 with cattle Kupffer2, suggesting shared transcriptional features between these annotated adult immune subsets in the datasets analyzed. Differential expression analysis of Kupffer1 and Kupffer2 yielded approximately 1980 differentially expressed genes (Fig. 5G; Supplementary Table S11). Kupffer1 highly expressed resident KC classics (VSIG4, CD163, MARCO, TIMD4), with GSEA enriching GAVISH_3CA_METAPROGRAM_MACROPHAGES_LIPID_ASSOCIATED, HE_LIM_SUN_FETAL_LUNG_C2_APOE_POS_M2_MACROPHAGE_CELL, HSIAO_LIVER_SPECIFIC_GENES, and lysosomal/phagocytic pathways, primarily handling lipid-iron metabolism, debris clearance, and immune tolerance. Kupffer2 overexpressed inflammatory stress genes (S100A8, CXCL8, FCN1, ALOX5AP, LYZ, FOS, JUN, DUSP1, NR4A1), with GSEA enriching HALLMARK_TNFA_SIGNALING_VIA_NFKB, GAVISH_3CA_METAPROGRAM_MACROPHAGES_MONOCYTE_SECRETED, and acute stress pathways.

Predicted intercellular interactions between hepatocytes and NPCs during postnatal liver development

Intercellular communication between hepatocytes and major non-parenchymal cells (NPCs) is essential for liver function under both physiological and pathological conditions. Using CellChat on our snRNA-seq data, we predicted putative ligand–receptor interactions across newborn, weaning, and adult camel liver stages, focusing on hepatocyte-centric signaling via secreted factors, cell–cell contact, and ECM–receptor pathways. In the CellChat output, the newborn sample showed the highest predicted hepatocyte-related interaction strength and number (73 outgoing and 39 incoming hepatocyte pairs).

When hepatocytes acted as receivers, CellChat suggested potential newborn-enriched interactions from hepatic stellate cells and cholangiocytes through ECM–receptor and laminin-axis signaling, which may be associated with hepatocyte adhesion, survival, and microenvironmental support (Fig. 6A). As signal senders, hepatocytes were predicted to engage NRXN3–NLGN1 interactions with hepatic stellate cells, liver sinusoidal endothelial cells, and vascular endothelial cells. Although best characterized in synaptic transmission, this evolutionarily conserved neurexin–neuroligin pair may mediate neuroendocrine-like crosstalk in the liver [43, 44] (Fig. 6B). Additionally, we detected newborn-enriched hepatocyte outgoing signals PLG–PARD3 and F2–PARD3 toward hepatic stellate cells, cholangiocytes, and mesothelial cells—interactions first reported in porcine liver and implicated in coagulation–fibrinolysis balance, dysregulation of which contributes to cirrhosis [45]. Homotypic CDH2–CDH2 (N-cadherin) signaling was also prominent between hepatocytes and hepatic stellate cells, cholangiocytes, and mesothelial cells, suggesting a possible role in cell–cell adhesion and migratory coordination previously linked to enhanced angiogenesis and proliferation in other systems [46].

Fig. 6.

Cell-cell communications between hepatocytes and non-parenchymal cells across time points. A-B Dot plot showing significant ligand-receptor interactions between non-parenchymal cells (cholangiocytes, hepatic stellate cells, liver sinusoidal endothelial cells, mesothelial cells, smooth muscle cells and vascular endothelial cells) and hepatocytes across time points in the aspect of hepatocytes as the signaling targets (A) and as the signaling sources (B). The dot color indicates the communication probability. The area within the blue box in (A) represents ECM–receptor signaling communication

Collectively, these analyses identified candidate hepatocyte–NPC communication axes in the sampled camel liver nuclei, providing hypotheses for future validation of postnatal liver remodeling.

Discussion

In this study, we used single-nucleus transcriptomics to construct a foundational atlas of the Bactrian camel liver across postnatal stages, including newborn, weaning, and adult stages. We also identified an HGFAC+ LSEC subpopulation in camel liver that was not detected in the human, mouse, or cattle datasets analyzed. In addition, we profiled expression changes across developmental stages in hepatocytes, liver sinusoidal endothelial cells, and immune cells during postnatal development.

Liver sinusoidal endothelial cells (LSECs) are not only structural components of the sinusoid, but also specialized regulators of angiocrine signaling, immune tolerance, and antigen presentation [47–49]. In our atlas, multiple LSEC subtypes were identified, including an HGFAC+ population, a PTPRC+ population, and a cycling LSEC population. The HGFAC+ LSEC subtype may represent a local endothelial program for HGF activation, because HGFAC is a serine protease that converts pro-HGF into its active form which is known to contribute to tissue repair and regeneration [50, 51]. This suggests that camel HGFAC+ LSECs may contribute to sinusoidal remodeling and the maintenance of hepatocyte growth potential throughout postnatal development. The persistence of HGFAC+ LSECs into adulthood may reflect a mechanism for preserving regenerative capacity under repeated dehydration and nutrient fluctuation, which place chronic stress on the liver. The PTPRC+ LSEC population may instead reflect an immune-interactive endothelial state, which can cross-present antigen and induce T-cell tolerance in the liver [48, 52]. Finally, the cycling LSEC population, marked by MKI67 and TOP2A, suggests that the sinusoidal endothelium remains developmentally active after birth rather than rapidly becoming quiescent. These diverse liver sinusoidal endothelial cells may contribute to growth-factor activation, immune regulation, and proliferative remodeling. This may help maintain hepatocyte stability and hepatic homeostasis under fluctuating nutritional and environmental stress.

Camel postnatal liver maturation may reflect a need to coordinate growth with metabolic adaptation under fluctuating water and nutrient availability. Early postnatal hepatocyte states were enriched for proliferative programs, whereas later states showed stronger metabolic specialization. In particular, the hep0 cluster was enriched in newborn and weaning stages and displayed high expression of PPARA [53] and LPIN2 [54], together with later upregulation of ACACA [55, 56] and ACADL [57], suggesting a stage-dependent reorganization of lipid metabolism during the transition from neonatal dependence to post-weaning feeding. Consistent with previous cross-species hepatocyte comparisons, camel hepatocyte-enriched genes were associated with fatty acid degradation and PPAR signaling pathways, including ACSL4, ACSL5, and ACOX1, suggesting potential differences in long-chain fatty acid metabolism [58–60]. Exploratory comparison with cattle and human datasets showed higher cell-cycle-associated scores in the adult camel dataset, suggesting that adult camel livers may possess greater proliferative and metabolic plasticity, potentially aiding in liver adaptation under environmental stress.

Although liver immune dynamics have been well studied in model species, they remain poorly characterized in camels. Using snRNA-seq, we found that Kupffer cells represented the largest immune fraction in the newborn sample, whereas T-cell subsets were more represented in the adult sample. This pattern is consistent with the postnatal shift from myeloid-dominated neonatal immunity to lymphoid-mediated surveillance [61]. Cross-species comparison suggested shared transcriptional features between several camel and cattle immune cell types. In addition, cattle Kupffer cells from the single-cell GTEx project clustered closely with camel counterparts [41]. Similar patterns have been reported between humans and mice [36], suggesting that core Kupffer cell programs related to immune tolerance, lipid and iron metabolism, and inflammatory defense are evolutionarily conserved across mammals.

Overall, this study provides a snRNA-seq atlas of the postnatal Bactrian camel liver. However, since only one individual sample was collected for each developmental stage, the observed differences in developmental stage expression profiles are limited to comparisons of the collected datasets. Cross-species comparisons were also used mainly for annotation support and hypothesis generation. Future replicated and functional studies are needed to validate the candidate cell states, marker genes, and communication axes identified here.

Conclusion

In summary, this snRNA-seq atlas provides a foundational resource for studying cell-type diversity and developmental transcriptional changes in postnatal camel liver. These findings may help guide future investigations into camel liver biology and postnatal adaptation.

Authors’ contributions

XW contributed to research design. HS contributed to animal research, data analyses, and drafting of the article. AZ and HL participated in the data analyses. TS, YL and WW participated in animal research and article review. XW provided advice on study design, division of labor, and drafting and revision of the paper. All authors read and approved the final version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (Inter-organizational cooperative research; W2512088), and the Xinjiang Provincial “Open Bidding for Selecting the Best Candidates” Project.

Data availability

The camel datasets from newborn and weaning stages were deposited in the SRA under BioProject ID PRJNA1470435, while the adult camel datasets are available in the SRA under BioProject ID PRJNA1119173. The analysis code is available in the Science Data Bank (https://www.scidb.cn/s/NnAvA3).

Declarations

Ethics approval and consent to participate

All experimental procedures in this study were strictly adhered to local legislations and approved by the Institutional Animal Care and Use Committee of Northwest A&F University (Approval No. XN2024-0416). The sampling process was rigorously monitored by members of the Institutional Animal Care and Use Committee of Northwest A&F University, ensuring the minimal distress of animals during experimentation.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.