Single-cell multiomics reveals multiple adipogenic pathways and diverse multilineage specializations during embryonic fat tail morphogenesis

Summary

Embryonic adipogenesis remains one of the least understood aspects of adipose biology in mammals due to time sensitivity, limited tissue volume, and ethical concerns. Here, we uniquely applied single-cell multi-omics sequencing to the developing adipose tissues of fat-tailed sheep, characterized by genetically determined, significant fat deposition in the tail during embryogenesis. Our dataset spans all stages of adipogenesis (E50 to E80), revealing three major cellular origins of fat deposition: progenitor and stem cells, connective tissue progenitors, and vascular smooth muscle cells. By integrating scRNA-seq, snATAC-seq, and functional validation, we identified key enhancer-driven gene regulatory networks (eGRNs) governing adipogenesis, with DBI emerging as a critical regulator through its interaction with PPARG. Additionally, we delineated developmental trajectories and unique eGRNs underlying angiogenesis, osteogenesis, chondrogenesis, and myogenesis associated with fat formation. Our findings provide novel insights into embryonic adipogenesis in mammals and reveal critical regulons governing lineage specialization.

Graphical abstract

Highlights

• •Single-cell RNA and ATAC atlas of developing embryonic ovine fat tail tissues
• •Multiple cellular origins contribute to fat deposition
• •Key regulators and the PPARG-DBI regulon involved in adipogenesis
• •Adipogenic, angiogenic, osteogenic, chondrogenic, and myogenic lineage specializations

Gene network; Cell biology; Transcriptomics

Introduction

As a major repository for energy storage and as an important mammalian endocrine organ, white adipose tissue (WAT) is crucial for maintaining metabolic health and systemic energy homeostasis.^1,2,3 The development and expansion of adipose tissue are driven by a combination of increasing adipocyte number (hyperplasia) and size (hypertrophy). Compared to hypertrophic development, which is associated with several metabolic syndromes, hyperplastic adipogenesis – involving the recruitment of progenitors into adipocytes - can prevent metabolic disease, and is linked to maintenance of insulin sensitivity and distinct metabolic responses across adipose depots.^4,5 In recent decades, due to the challenges in detecting and collecting adipose tissue during embryogenesis, research has generally focused on the tissue remodeling process of WAT at postnatal stages or in adult mammals.^2,6,7,8 The developmental origins of adipocytes, the dynamic differentiation processes, and the regulatory systems underpinning embryonic adipogenesis remain some of the least-understood aspects of adipose tissue biology. In addition, accumulating evidence suggests that the pool of stem cells in adipose tissue possesses considerable heterogeneity, distinct metabolic features, and varying adipogenic potential,^9,10,11 which are likely the primary reasons for the anatomically distinct distributions of adipose tissues and make the cellular origins of adipocytes more difficult to decipher.

With advancements in high-throughput sequencing (HTS) technologies, barcoding strategies for efficient multiplexing, and large-scale parallel random cell capture techniques, single-cell RNA sequencing (scRNA-seq) has become a powerful approach for examining, at high resolution, dynamic cellular states and the cellular composition of tissues.¹² However, the large dimensions (up to 300 μm¹³) and buoyant, fragile nature of mature adipocytes limit the application of single-cell sequencing to adipose tissue, necessitating alternative strategies such as stromal vascular fraction (SVF) extraction⁵ or single-nuclei RNA sequencing.¹⁴ Although the single cell atlas of adipose tissues produced by scRNA-seq and snRNA-seq exhibits similar cell type abundance and gene expression profiles,¹⁵ the potential for false positives introduced by batch effect correction or other computational techniques during omics data integration should not be underestimated.

Performing single-cell sequencing on developing or differentiating adipocytes provides an excellent strategy for directly studying the cellular hierarchies and mechanisms governing adipogenesis within WAT in an accurate and unbiased manner. However, challenges such as difficulties in determining sampling times, limited cell quantities, and ethical considerations make the collection of developing adipose tissue from humans, mice, or pigs particularly difficult. Unlike other mammals, where adipogenesis is more evenly distributed, fat-tailed sheep evolved a distinct, genetically programmed adipose accumulation pattern in the tail during embryogenesis, accounting for approximately 10% of body weight. This trait poses significant challenges in modern intensive and semi-intensive production systems, negatively impacting economic efficiency,¹⁶ and also provides a unique opportunity to study adipogenesis in vivo without the confounding effects of postnatal metabolic influences. Our previous studies have demonstrated that large numbers of adipocytes are present during the initial and early stages of fat-tail morphogenesis, specifically at the 14–20 μm diameter stage,¹⁷ highlighting the feasibility of performing single-cell sequencing on developing fat-tail tissues to uncover the molecular mechanisms regulating embryonic adipogenesis and the genetic architecture behind the fat-tail phenotype.

In the present study, we profile global transcriptomic (scRNA-seq) and epigenetic (snATAC-seq) signatures at single-cell resolution in developing ovine fat-tail tissues during embryogenesis, covering all stages of fat deposition (E50–E80). The constructed single-cell multi-omics atlas comprises various cell types from progenitor populations and adipogenic, angiogenic, osteogenic, chondrogenic, and myogenic lineages, revealing remarkable tissue heterogeneity. By mapping lineage specialization and key regulatory factors, our results highlight the multicellular origins of fat deposition and uncover diverse specialization patterns and complex gene regulatory networks (GRNs) shaping the differentiation trajectories of distinct cellular lineages. The work described here introduces and establishes a representative large animal model for investigating prenatal adipogenesis in mammals at single-cell resolution, offering a valuable framework for studying adipose tissue biology in both agricultural and biomedical contexts.

Results

Remarkable tissue heterogeneity of developing ovine fat tail tissues

To investigate the cellular origins of adipose tissue and the gene regulation events underpinning this heritable massive fat deposition trait, we first constructed single-cell transcriptome and chromatin accessibility landscapes for developing tail tissues. These encompassed embryo days E50 to E80, covering the entire stages of adipogenesis with high-density time increments (3 days or 5 days; Figure 1A). Compared to E50, we observed remarkably differentiated adipocytes at E80. A total of 9.20 billion scRNA-seq reads were obtained with a mapping rate of over 90% to the Oar_rambouillet_v1.0 ovine reference genome (Table S1). After removing low-quality cells and doublets, we retained a total of 43,557 clean cells, which were assigned to 37 clusters and annotated as 18 main cell types based on the specific expression of canonical marker genes (Figures 1B, S1A, and S1B). For example, the expression of TOP2A in progenitor and stem cells (PSC), COL6A6 in connective tissue progenitors (CTP), PPARG in committed adipocytes (preadipocytes and adipocytes), and MYOD1 is exclusively expressed in myogenic cell types (Figures S2A and S2B). Notably, some of these cell types can be subdivided at the cluster level, including six clusters for PSC (PSC_C1 to C6), three clusters for CTP (CTP_C1 to C3), two clusters for chondrocytes (C1 and C2), and two clusters for vascular smooth muscle cell (VSMC_C1 and C2). Treating snATAC-seq reads with similar data pre-processing procedures, we obtained 16,042 clean cells (Figures S2C and S2D). We applied scRNA-seq annotations to the snATAC-seq cells. However, due to the sparse nature of ATAC-seq data, the gene expression diversity among clusters of some specific cell types was not mirrored in the chromatin accessibility results, and these clusters were annotated together. Therefore, PSC_C1 to C3 were grouped as PSC subpopulation 1 (PSC_S1), PSC_C4 to C6 as PSC subpopulation 2 (PSC_S2), and Chondrocyte_C1 and Chondrocyte_C2 as Chondrocyte in the snATAC-seq dataset (Figure 1B).

Figure 1.

Single cell atlas of developing ovine fat tail tissue

(A) A workflow detailing sample collection, histological analyses, sequencing strategy, and bioinformatic pipeline. Embryonic tail tissues at nine stages (E50, E55, E60, E63, E66, E69, E72, E75, and E80) were used for scRNA-seq and tissues at four stages (E55, E63, E69, and E75) for snATAC-seq. One sample was obtained for each developmental stage. HE and Oil red staining for the tissues at the initial (E50) and final (E80) time points. Red ∗ indicates differentiated adipocytes with visible fat droplets. Scale bars: 200 μm, 400 μm, 500 μm, and 2,000 μm, respectively (see corresponding images).

(B) A UMAP visualization of 18 annotated main cell types in the scRNA and scATAC datasets. Cells are color-coded according to cell type. The PSC_C1 to C3 clusters were annotated as PSC_S1, the PSC_C4 to C6 clusters as PSC_S2, and the T cell, macrophage, and mast cell as immunocyte in scATAC.

(C) The left heatmap shows the expression of highly expressed genes in each cell type and significantly enriched GO pathways. The right heatmap shows the chromatin accessibility, peak number, and peak types of identified regions in each cell type.

Taken together, six main lineages could be discerned in the constructed scRNA-seq and snATAC-seq landscapes; these encompassed progenitor, VSMC, adipogenic, chondrogenic, osteogenic, and myogenic lineages, demonstrating a pronounced cellular heterogeneity within the developing tail tissue. At approximately 40% of the total, progenitor populations, including both PSC and CTP, were most abundant across all cell types, and this proportion was stable across the nine time points, highlighting the critical role of progenitor cells in cell recruitment during early embryo development and tissue generation (Figures S3A and S3B and Table S2). Preadipocytes first emerged at E63, followed by adipocytes at E69. The population of preadipocytes and adipocytes exhibited a dramatic expansion across the nine stages, increasing to 10.60% and 2.74% by E80, respectively.

Furthermore, the genes identified as highly expressed within each cell type showed enrichment in functionally relevant pathways; for example, progenitor cells were associated with pathways such as mitotic nuclear division and extracellular collagen matrix, and committed adipocytes showed enrichment in the protein-lipid complex and cellular respiration pathways (Figure 1C and Table S3). These observations provide further support for the accuracy of our cell type annotations. To gain a comprehensive understanding of the specificity of these highly expressed genes and to identify reliable and effective novel markers for further studies, we developed a strategy to evaluate their specificity according to the average expression of the target genes across cell types. This approach categorized genes into four classes: A+ (strong marker), A (general marker), A− (medium marker), and H (other highly expressed genes), with the specificity gradually decreasing from A+ to A− (Figure 2A and Table S4). The A+ marker class was restricted to genes that were specifically overexpressed in a particular cell type and were identified primarily in the PSC population and fully differentiated cell types (Figures S3C, S4, and S5). For example, there are 36, 25, and 19 highly expressed genes identified as A+, A, and A− markers within adipocytes, respectively. Notably, no A+ marker gene was observed in transitional cells, such as preadipocytes, prechondrocytes, preosteoblasts, and satellite cells, suggesting substantial alterations in gene expression and complex gene regulatory events during lineage specialization. In parallel to this, we also observed that the majority of highly accessible regions detected for each cell type from the snATAC-seq data were distributed in the distal and intronic regions of genes across the ovine genome (Figure 1C).

Figure 2.

The lineage hierarchy of progenitor populations and the process of angiogenesis

(A) The evaluated A+ (red), A (green), and A− (blue) class gene markers for specific cell types based on the highly expressed genes in Figure 1C. Labels indicate representative marker genes for each cell type. The radial distance for each gene point represents the log₂ fold change value.

(B) Gene sets used to distinguish progenitor populations at the subpopulation and cluster levels.

(C) A representative tail tissue RNA-FISH image at E80 with DAPI (Blue), TOP2A (red), and MCM5 (green) staining. Three biological replicates. Scale bars: 500 μm.

(D) The PAGA differentiation network was generated for the progenitor populations, and adipogenic, VSMC, osteogenic, and chondrogenic lineages. Line widths indicate the probability of differentiation. And the differentiation subnetwork generated for adipogenic-related cell types, including CTP_C2 and CTP_C3, PSC_C3 and PSC_C4, VSMC_C1 and VSMC_C2, and preadipocyte and adipocyte.

(E) Important genes involved in regulating VSMC differentiation.

(F) The differentiation trajectories constructed for PSC_C3, VSMC_C1, and VSMC_C2 using Monocle2 and scVelo. Cells on the tree are colored by cell type and predicted pseudotime.

(G) Significantly enriched GO terms depending on branch-dependent genes.

Lineage hierarchy of progenitor populations

The identified large cohort of progenitors represents a versatile repository source for multiple lineages, which consist of PSC and CTP, that exhibit specific expression of the TOP2A and COL6A6 genes, respectively (Figure S6A). We defined six partially neighboring clusters (PSC_C1 to C6) and one independent myogenic progenitor (PSC_Myo) cluster for PSC, and three clusters for CTP that are in close proximity (CTP_C1 to C3). These PSC clusters exhibit a high degree of similarity in their gene expression profiles, including conserved expression of the SMC2, CLSPN, and CENPK genes, making them indistinguishable at the cluster level by single genes (Figure S6B). Therefore, through the detection of differently expressed genes (DEGs) and correlation analysis across PSC clusters, we detected one gene set that could be used to classify progenitor populations at the subpopulation and cluster levels (Figure 2B and Table S5). In addition to TPX2 and TOP2A genes that are exclusively expressed in the PSC_S1 and PSC_Myo subpopulations, genes such as CLEC3B, CXCL14, ACTA2, and ACTC1 were observed to be overexpressed in the PSC C1, C2, C3, and PSC_Myo clusters, respectively. Furthermore, depending on the expression levels of MCM5 in the PSC_S2 subpopulation, the SFRP4, IGFBP7, and TMEM132C genes showed further overexpression in the PSC C4, C5, and C6 clusters. With the RNA-fluorescence in situ hybridization (RNA-FISH) assay for TOP2A and MCM5, we observed an overlapping spatial distribution of progenitor cell populations, which were extensively distributed throughout the tissue (Figure 2C). Similarly, based on

the fundamental expression of the PCSK5 and COL6A6 genes in CTP, the CCL11, NUPR1, and CRABP1 genes could be used to characterize the CTP_C1 to C3 clusters, respectively.

The constructed differentiation trajectory for progenitor clusters further highlighted the potential differentiation directions from PSC_S1 to PSC_S2 and then to CTP (Figures S6C–S6E). Although a continuous distribution along the trajectory can be clearly observed, given the similarity in gene expression and the potential complicated interconversion relationships among these progenitor populations, we would cautiously hypothesize that a hierarchical distribution exists in differentiation time, rather than a straightforward, direct differentiation relationship. The PAGA network further suggested the potential differentiation relationships, especially from PSC_C4 to CTP_C2 and CTP_C3, as well as from PSC_C5 to CTP_C1. Furthermore, the CKS2, TOP2A, and TPX2 genes were significantly overexpressed at the beginning of the trajectory. In the PSC_S2 subpopulation, genes such as MCM5, MSH6, and HELLS exhibited high expression levels. Additionally, the expression of genes including ASPN, SFRP2, and CLEC3B progressively increased over pseudotime (Figures S6F and S6G). The dynamic expression patterns of these genes would likely influence the proliferation potential and metabolic capabilities of the different progenitor populations, as well as their commitment to specific lineages.

Our single cell atlas consists primarily of heterogeneous progenitor populations and another five terminal differentiation lineages. This diversity drew our attention to the downstream development of each branch of the progenitor dendrogram, genes and biological pathways involved in lineage commitment and cellular differentiation. We constructed a differentiation network for progenitor populations and terminal differentiation lineages except myogenic related cell types, which can aid in the prediction of specific progenitor clusters corresponding to each lineage (Figure 2D). In this differentiation network, preadipocytes are identified as having three cellular origins; PSC_C3 contributes to the generation of vascular smooth muscle cells; PSC_C2 and PSC_C6 are associated with osteogenesis; and all three clusters of CTP are involved in chondrogenesis. In the following sections, our focus shifts to delineating the differentiation trajectories and identifying the key hub genes that govern lineage specialization and cellular differentiation.

Vascular smooth muscle cell differentiation

To elucidate the cellular origins of adipocytes, initially, we built one independent differentiation subnetwork for cell types related to preadipocytes, which exhibit three cellular origins for adipogenesis, including one cluster of VSMC (VMSC_C2), one cluster of PSC (PSC_C4), and two clusters of CTP (CTP_C2 and C3; Figure 2D). Obviously, except for PSC_C4 and CTP_C2/C3, VMSC_C2 exhibited an additional, distinct upstream trajectory involved in the molecular mechanism of angiogenesis. Vascular smooth muscle cells situated in the medial layer of blood vessels are essential for modulating the contraction and expansion of blood vessels.¹⁸ We observed the overexpression of both VSMC markers (GUCY1B1, AVPR1A, MYH11, ACTA2, and TAGLN) and pluripotent genes (TOP2A, TPX2, and CKS2) in PSC_C3 among the six PSC clusters (Figures 2E and S7A). Using RNA-FISH, we observed a great number of cells located in nascent and growing blood vessels that were positive for TOP2A (Figure S7B). The constructed differentiation trajectory further supports the progenitor role of PSC_C3 for VSMC lineage, followed by VSMC_C1, which splits into two directions: the T1 and T2 branches. The VSMC_C1 cells in the T1 branch ultimately lead to the cells of VSMC_C2 (Figure 2F). Notably, the expression of pericyte marker genes, such as PDGFRB, CSPG4, SLC6A12, and SLC19A1, was absent in these two clusters (Figure S7C).

To explore the molecular basis of this divergence, we applied BEAM analysis to identify branch-dependent genes (bdGenes). TFunctional enrichment of top bdGenes revealed distinct transcriptional programs between the two branches: VSMC_C1 showed enrichment for pathways related to oxidative phosphorylation, oxidation-reduction processes, and transmembrane transport, whereas VSMC_C2 exhibited elevated expression of genes involved in transcriptional activation, growth factor signaling, and extracellular matrix remodeling (Figures 2G and S7C). This bifurcation pattern reflects the well-documented phenotypic plasticity of vascular smooth muscle cells, which can transition between a contractile phenotype—characterized by high expression of cytoskeletal genes and mitochondrial metabolism—and a synthetic phenotype, associated with ECM production, proliferation, and reduced the expression of contractile markers.¹⁹ Notably, VSMC_C2 exhibited the reduced expression of GUCY1B1, AVPR1A, ACTA2, and TAGLN, consistent with features of the synthetic phenotype (Figure 2E). While this transcriptional divergence aligns with classical descriptions of contractile and synthetic states observed in injured or remodeling adult vasculature, we interpret these findings as suggestive rather than definitive evidence, given the developmental context of our samples.

To further examine genes governing VSMC lineage specialization, DEG and differently accessible region (DAR) analyses were conducted across these cell types (Figure S7D and Table S6). The majority of DEGs were downregulated from PSC_C3 to VMSC_C1, including the TOP2A, TPX2, and CKS2 genes, with only six genes (OGN, ASPN, TAGLN, CUTA, PPP1R14A, and HSP90AB1) showing upregulated expression. In contrast to this, approximately 50% of the DEGs identified between VMSC_C2 and VMSC_C1 were upregulated. Through the integration of the significant DEGs, DARs, and bdGenes, we observed the overexpression of several VSMC markers in VSMC_C1, including GUCY1B1, AVPR1A, ABCC9, and ETS1, while RAI14, PLPP1, ANK2, and NRP1 were involved in VSMC_C2 generation, with highly accessible genomic regions in the corresponding VSMC clusters (Figures S7E–S7G).

Multiple cell types contribute to committed adipocytes

In our single cell atlas, committed adipocytes (preadipocytes and adipocytes) were annotated by PPARG, ACSL1, LPL, and ADIPOQ genes, but not the expression of TRPV1 and UCP1,²⁰ which are markers of brown adipocyte progenitors and brown adipocytes, respectively (Figures 3A, 3B, S8A, and S8B). We highlighted A+ markers that were exclusively overexpressed in adipocytes and all the preadipocyte markers (Figure 3C), which exhibit expression patterns that gradually increase following adipogenic induction in the primary ADSC cell model in vitro (Figure S8C), indicating they can also serve as marker genes in further studies. Adipocytes descend from a pool of proliferating progenitors (also known as fibroblast and adipogenic progenitors (FAPs), fibroblast-like progenitors, or mesenchymal stem cells) that reside in the adipose tissue with limited differentiation potential. It has been reported that subcutaneous and visceral WAT have both shared and depot-specific FAPs populations, highlighting the complex nature of adipogenic pathways.⁶ The PDGFRA gene has been extensively deployed to identify FAPs from adipose tissues.^2,4,14,21,22 However, when dealing with complex and heterogeneous tissues, such as the ovine tail tissue examined in this study, PDGFRA was collectively expressed in proliferating progenitors and committed precursors, including PSC, CTP, prechondrocytes, preosteoblasts, and preadipocytes, albeit not in the myogenic lineage (Figure S8B).

Figure 3.

Adipogenic pathways underpinning embryonic fat tail development

(A) UMAP plots show the expression of PPARG and ACSL1 genes.

(B) Re-clustering analysis for predicted adipogenic progenitor populations, preadipocytes, and adipocytes with the expression of identified positive regulators (PPARG, LPL, DBI, CIDEA, and ADIPOQ), negative regulators (LIMCH1 and COL21A1), and previously reported genes (PDGFRA, DPP4, ALDH1A3, and ALDH1A1).

(C) A heatmap shows the A+ marker genes in adipocytes and all marker genes in preadipocytes. Each row represents a specific cell type, while each column represents a gene. Each small square in the heatmap indicates the scaled average expression value of a specific marker gene in a given cell type.

(D) The differentiation trajectories generated for the three adipogenic progenitor populations and committed adipocytes using scVelo (top, RNA velocity) and Monocle2 (pseudotime), respectively.

(E) Representative HE and RNA-FISH images stained with DAPI (Blue), LPL (red), MCM5 (green), and SFRP4 (yellow). LPL, the marker gene for preadipocytes and adipocytes, is shown. The MCM5 and SFRP4 genes were used to identify PSC_C4. Arrows indicate the magnification effect for the two zones in the merged image. M1 and M2 are magnified images of two specified regions in the merged image. Three biological replicates. Scale bars: 100 μm and 200 μm, respectively (see corresponding images).

(F) Representative RNA-FISH images stained with DAPI (Blue), LPL (red), ALDH1A1 (yellow) and GUCY1B1 (green). Also shown is ALDH1A1, the gene specifically expressed in CTP_C2 and CTP_C3, and GUCY1B1, a marker gene of vascular smooth muscle cells. Three biological replicates. Scale bars: 100 μm.

Two phases of adipogenesis consist of commitment from progenitors to preadipocytes and differentiation from preadipocytes to adipocytes. Three adipogenic progenitor populations associated with preadipocytes were predicted using the PAGA trajectory analysis approach (Figure 2D). Therefore, we independently performed pseudotime analysis, RNA velocity, and diffusion map analysis on the predicted adipogenic progenitor populations and committed adipocytes to unveil the dynamic processes of adipogenesis. These complementary approaches jointly delineate a clear lineage progression from PSC_C4, CTP_C1/C2, and VSMC_C2 toward mature adipocytes (Figure 3D and Figure S8D). Specifically, PSC_C4 cells that were positive for MCM5 and SFRP4 are exposed to an environment rich in committed adipocytes (LPL⁺; Figure 3E). Moreover, we observed the co-localization of ALDH1A1⁺ CTP cells and GUCY1B1⁺ VSMC cells (Figure 3F). Regarding the two distinct VSMC subpopulations we identified, only a subset of VSMCs was found in proximity to LPL⁺ committed adipocytes (Figure 3F). Taken together, our results support the hypothesis that multicellular origins contribute to the massive and rapid fat accumulation in ovine tail tissues during embryogenesis.

To discover genes that jointly exert positive or negative effects on adipogenic processes across three adipogenic pathways, we performed DEG analysis between preadipocytes and adipogenic progenitor populations, and between adipocytes and preadipocytes (Figure 4A and Table S7). The results of this showed that the majority of DEGs detected between adipogenic progenitors and preadipocytes were downregulated. This group includes eleven genes with continuously decreasing expression levels from all adipogenic progenitor populations to adipocytes (Figure 4B and S8E). The LIMCH1 gene that encodes the LIM and calponin homology domains-containing protein 1, which is involved in tumor and muscle development and has both LIM and calponin homology domains,²³ was one of the most significantly downregulated genes with concomitant significantly repressed chromatin accessibility (Figures 4C, 4D, and S8F), emphasizing its negative regulatory role in adipogenesis.

Figure 4.

Important regulators governing adipogenesis

(A) The experimental contrasts used for the detection of DEGs: preadipocytes versus three adipogenic progenitor populations, and adipocytes versus preadipocytes. Labels indicate the top DEGs for each contrast.

(B) Venn diagrams show the number of overlapping upregulated DEGs (top) and downregulated DEGs (bottom) across the four contrasts.

(C) The upper plot shows DEGs negatively related to adipogenesis; the middle plot shows DEGs positively related to adipogenesis; and the lower plot shows DEGs specifically expressed in adipocytes.

(D) The expression trends for the LIMCH1, PPARG, and DBI genes along three differentiation trajectories. And a peak plot showing the specifically accessible region of DBI in preadipocytes and adipocytes.

(E) The eGRNs generated for committed adipocytes using TFs (large circle nodes), highly accessible regions (diamond nodes), and highly expressed genes (small circles). The colors of the small circles represent the log₂FC gene expression values in committed adipocytes compared to the other cells. The TF–region links are colored by TF, and the region–gene links are colored gray. The red arrow shows the interaction between PPARG and the DBI-specific peak.

(F) The positively correlated expression patterns for the PPARG and DBI genes in differentiating primary ADSCs isolated from ovine fat tail tissues. D0, D2, D4, and D6 correspond to the four stages of in vitro differentiation. At least three biological replicates at each stage. Data are represented as mean ± SEM.

(G) The enhancer effect of the DBI-specific region (653 bp) on gene expression using the dual-luciferase reporting system in the primary adipose-derived stem cells (ADSCs) isolated from embryonic fat tail tissues. pDBI-Peak, recombinant plasmid using DBI-specific region; pDBI-NC, plasmid used as a negative control for pDBI-Peak; OE-PPARG, recombinant plasmid using the CDS sequence of PPARG; OE-NC, plasmid used as a negative control for OE-PPARG. At least six biological replicates were included for each treatment.

Data are represented as mean ± SEM. Statistical significance was determined using Student’s t test. ∗∗p < 0.01.

In contrast, only thirteen DEGs were jointly upregulated in three adipogenic pathways, including PPARG, DBI, FABP5, FABP4, DNAJB1, PEG3, CLU, ID1, MYLK, PLBD1, RGCC, TGFB2, and one novel gene ENSOARG00020006446 (Figure 4C). Only PPARG (a well characterized adipogenic regulator), DBI, and FABP4 exhibited gradually increased expression from adipogenic progenitors to adipocytes (Figure 4D). The upregulated DEGs from preadipocytes to adipocytes were enriched in lipid localization and fatty acid metabolic related pathways. The downregulated genes were enriched in Wnt signaling and Rho protein signal transduction pathways (Figures S9A and S9B). Several genes, such as ADIPOQ, ADIPOR2, ACSL1, CIDEA, PNPLA2, DGAT2, SCD, and GPAM, were exclusively expressed with specifically accessible genomic regions in adipocytes (Figures 4C and S9C). Overall, the gene regulatory network (GRN) was constructed using dynamic gene expression and chromatin accessibility profiling following adipogenesis (Figure 4E), where DBI, FABP4, MYLK, ADIPOQ, ACSL2, SCD, and other key genes involved in adipogenesis and lipogenesis were transcriptionally regulated by PPARG, the master regulator of adipocyte differentiation. Our results further emphasized the central role of PPARG in embryonic adipogenesis; the transcriptional cascade driving adipogenesis within developing fat tail tissue is mainly initiated and regulated by interactions among PPARG, TCF12, MITF, NFIB, HSF2, NFIL3, TCF7L2, JUNB, and EBF1 transcription factors (TFs).

Notably, among these genes, the DBI gene, identified as an A+ marker gene due to its overexpression in adipocytes (Figure 3C) showed a positive correlation with PPARG. In committed adipocytes, one specifically accessible peak (Chr2:198163303-198163955) was detected downstream of the DBI gene, which could interact with PPARG (Figure 4E). This may represent a putative enhancer (Figure 4D) and supports a role for DBI in adipogenesis. DBI (diazepam binding inhibitor), originally discovered as a secreted polypeptide regulating neural activity in the brain, was also termed acyl-CoA binding protein (ACBP) due to its ability to bind long chain fatty acyl-CoA esters and mediate fatty acid termination and synthesis.^24,25 Several studies have also indicated that DBI affects adipogenesis or lipogenesis in a complex fashion through influencing appetite, mediating the digestive system, and regulating lipid metabolism in adipose tissue and liver.^26,27,28 In the primary in vitro ADSC cell model, both DBI and PPARGR exhibited a continuous increase in expression following adipogenic induction and displayed a positive mutual correlation (Figure 4F). We further developed a dual-luciferase reporter system to evaluate the putative PPARG-binding enhancer at the DBI-specific peak. This showed that the recombinant plasmid containing the DBI-specific peak sequence significantly activated gene transcription activity (Figure 4G). Moreover, PPARG overexpression further enhanced this activation effect (Figure 4G). Overall, our results in the ovine fat tail model highlight the functional importance of the PPARG–DBI regulatory axis in controlling fat deposition and adipose tissue development.

Microenvironmental effects on adipogenesis

White adipose tissue (WAT) can regulate several aspects of organismal physiology through endocrine functionality; concomitant with this, external factors in the microenvironment can also exert influence on de novo adipogenesis and cell fate decisions. Our single cell transcriptome and chromatin accessibility landscape revealed pronounced heterogeneity within developing ovine fat tail tissues. To further investigate the impact of tissue microenvironment effects on adipogenesis, using CellChat²⁹ we inferred and analyzed potential intercellular communications among all cell types according to the differential expression genes encoding ligand and receptor pairs (LR pairs). The global cell-cell interaction pattern comprised 143 LR pairs from 21 signaling pathways (Figure S10A and Table S8). Some of these pathways exhibits cell-type-specific or lineage-specific preferences (Figure S10B). For instance, CDH signaling mediates intercellular communication among PSC_Myo, satellite cells, myoblasts, and myocytes. The SEMA3 and TENASCIN pathways are primarily involved in signal transduction among osteogenic lineage, chondrogenic lineage, and endothelium or mast cells.

Our results suggested that the ADIPONECTIN, CD46, THBS, WNT, ANGPTL, and VEGF signaling pathways serve as major pathways for mediating communication between committed adipocytes and other cell types (Figures 5A and S11A). Specifically, through ADIPONECTIN, ANGPTL, VEGF, and PTN signaling pathways, committed adipocytes function as source cells expressing ADIPOQ, ANGPTL1, VEGFA, and PTN ligands to interact with a range of cell groups, including PSC_C3 (VSMC progenitors), VSMC_C1, neurons, and endothelial cells. Conversely, committed adipocytes serve as targets in THBS, WNT, and CD46 signaling pathways, interacting with cells, such as progenitor populations, myogenic cells, VSMC, and chondrocytes via encoding receptors JAG1, CD36, FZD4 and LRP6. In particular, THBS signaling allows adipocytes and endothelial cells to extensively receive signals from progenitors and differentiated cells (Figures 5B and S11B). The prominence of THBS, ANGPTL, and VEGF-based intercellular interactions suggested enhanced coordination among the angiogenic system and committed adipocytes during adipocyte differentiation and adipose tissue development.

Figure 5.

Intercellular communication between adipogenic cell types and others

(A) The cell-cell interactions between committed adipocytes and other cells.

(B) The specific intercellular communications mediated by the THBS signaling pathway.

(C) The cell-cell interaction between committed adipocytes and several cell types, including VSMC, adipogenic progenitors, endothelium, and neurons.

(D) Violin plots showing expression levels for genes that encode the ligands and receptors shown in Figure 7C.

Due to the widespread presence of endothelium, vascular smooth muscle cells, and neurons anatomically across WAT deposits and phylogenetically across mammalian species, we further directed our focus toward communication events occurring among committed adipocytes, adipogenic progenitors, and the cell types described above at the level of ligand-receptor pairs (LR pairs) (Figures 5C and 5D). Committed adipocytes seem to predominantly receive rather than send signals, especially signals derived from adipogenic progenitors, which can exert an influence on committed adipocytes through WNT2-(FZD4+LRP6) LR pairs. The intracellular communication and cross-talk between preadipocytes and adipocytes were mediated by THBS3/S4-CD36 and ADIPOQ-ADIPOR2 LR pairs. In addition, the interactions between committed adipocytes and VSMC, endothelium, and neurons mainly occurred in preadipocytes rather than in adipocytes, especially the released signals from preadipocytes to endothelium, which were mediated by multiple LR pairs. In contrast to this, adipocytes were more prone to receiving signals from source cells, such as adipogenic progenitors, VSMC, and endothelium via WNT2-(FZD4+LRP6), THBS1/S3/S4-CD36 and CD46-JAG1 LR pairs. Overall, our results suggest that highly efficient communication strategies established between cellular components of the vascular and adipogenic lineages encourage exchange of biological information to promote mutually dependent differentiation of multiple tissues during ovine fat tail development.

Other cell lineage specialization

Osteogenic, chondrogenic, and myogenic processes also emerged during ovine fat tail development. Preosteoblasts and osteoblasts specifically expressed the BMP3 and BMP7 genes with the negative expression of the NGFR, MCAM, LEPR, and NT5E genes (Figures 6A, S12A, and S12B), results that were similar to those obtained for the cellular origins of human skeletal bone tissue.^30,31 The PAGA-derived differentiation network displayed two putative osteogenic progenitor populations, PSC_C2 and PSC_C6 (Figure 2D). The constructed differentiation trajectory suggested the transition from PSC_C2 to PSC_C6 with the decreased expression of PSC_S1 markers, followed by the cell fate specialization into preosteoblasts consisting of two branches, T1 and T2 (Figures 6A and S12C). The preosteoblasts at the T1 branch continuously differentiated into osteoblasts. Although we did not observe the joint osteogenic and chondrogenic progenitors, termed as embryonic skeletal stem/progenitor cells (eSSPCs) and reported previously for human limb bud development,³¹ CADM1, the phenotypic marker of eSSPCs,³¹ exhibited preferential expression in PSC_C2 and PSC_C6 (Figure 6A). Additionally, the CLEC11A gene, critical for maintaining the adult skeleton, was also overexpressed in osteogenic progenitors and preosteoblasts³¹ (Figure 6A). Integration of the DEG and DAR results demonstrated that several genes may execute distinct functions during osteogenic lineage specialization (Figure S12D and Table S9). For example, the conserved expression of TMEM132C and DPEP1 in the osteogenic cell types and the specific overexpression of TNC and CDH6 in preosteoblasts (Figure S12E). Compared to genes that promote the differentiation trajectory from progenitors toward preosteoblasts (i.e., ASPN, TNC, CDH6,GNG11, CCDC80, and PTPRD), PAPPA2, ASAMTS18, LRRTM2, PTCH1, and several other genes would facilitate bone calcification and mature osteocyte formation (Figures 6A and S12E). Importantly, we observed divergent expression of the CXCL14, LRRTM2, and TRPS1 genes among the T1 and T2 branches coinciding with significantly accessible genomic regions, which may promote the differentiation process from preosteoblasts to osteoblasts (Figures 6A, 6B, and S12F).

Figure 6.

Lineage specialization processes of osteogenesis, chondrogenesis, and myogenesis

(A) A re-clustering analysis for osteogenic- and chondrogenic-related cell types; the UMAP visualizations are shown for the genes relevant to osteogenesis; the differentiation trajectories generated for PSC_C2, PSC_C6, preosteoblasts, and osteoblasts; and the divergent gene expression patterns following the osteogenic differentiation trajectory. SPN, SDH6, and TNC genes are positively related to the T1 branch; CXCL14, LRRTM2, and TRPS1 genes are positively related to the T2 branch.

(B) Representative peak plots showing the specifically accessible regions of the CXCL14, TNMD, and CRABP2 genes. In each case, the blue gene body shows the location of the gene on the reverse strand.

(C) The UMAP visualizations are shown for the genes relevant to chondrogenesis; the differentiation trajectories generated for the prechondrocytes and two clusters of chondrocytes; and violin plots showing expression levels of conserved genes in prechondrocytes and chondrocytes, and specifically expressed genes in Chondrocyte_C1 and Chondrocyte_C2, respectively.

(D) A re-clustering analysis for myogenic progenitors (PSC_Myo), satellite cells, myoblasts, and myocytes with differentiation direction inferred by pseudotime analysis and RNA velocity analysis; UMAP visualization for relevant myogenic regulators; the positive regulator genes involved in myogenesis with distinct gene expression patterns, including overexpressed genes in satellite cells, overexpressed genes in myoblast, continuously upregulated genes during differentiation and specifically expressed genes in myocytes (left to right); representative HE and RNA-FISH images stained with DAPI (Blue), and JSRP1 (red). JSRP1, the marker of myogenic cells. Blue ∗, myogenic cells. Red ∗, differentiating adipocytes. Arrow, the special structure surrounding adipocytes with the positive expression of JSRP1.

Scale bars: 100 μm and 400 μm, respectively (see corresponding images).

Different types of chondrocytes have been reported within cartilage tissue in mammals, including classic chondrocytes, proliferative chondrocytes, perichondrial cells, hypertrophic chondrocytes, and several populations that have been defined molecularly under different conditions.^21,32,33 We identified two distinct chondrocyte clusters, including chondrocyte C1 and C2, with the conserved expression of the classic chondrocyte markers COL11A1 and COL2A1 (Figure 6C). The protein encoded by the COL2A1 gene regulates the biosynthesis of extracellular glycosaminoglycans, acts as a key component of the cartilage extracellular matrix, and is involved in chondrocyte differentiation and endochondral osteogenesis.³⁴ Cluster C1 was characterized by the specific expression of TNMD and was more responsible for the generation of the collagen-enriched extracellular matrix. In contrast to this, cluster C2 was positive for ITGBL1 and exhibited enhanced metabolic ability, as evidenced by the high expression of the A+ class marker genes STMN2, KRT19, and ITGBL11 (Figures 6C, S13A, and Table S10). The results of the trajectory analysis suggested that three CTP population clusters have the potential to differentiate into prechondrocytes (Figures S13B–S13E). The integrated DEG, DAR, and trajectory analyses indicate that CSRP2 and GAS2 would regulate the entire chondrogenic process from CTP to chondrocytes with continuously increased expression (Figures 6B, 6C, S13F, and S13G). Importantly, a previous study has shown that GAS2 was highly expressed in osteo-chondrogenic progenitors for human embryonic skeletal bone, and also co-expressed with COL11A1,³¹ both of which were identified as crucial chondrogenic regulators in our results. In contrast to this, TNMD and CRABP2, which exhibited expression patterns that followed two divergent paths, may influence the cell fate decision from prechondrocytes to chondrocyte C1 and C2, respectively (Figures 6B, 6C, and S13F). Taken together, our results suggest the chondrogenic process from CTP to two distinct types of chondrocytes is controlled by a combination of conserved genes and cluster-specific genes.

Skeletal muscles are characterized by voluntary control, multiple nuclei, and a striated architecture. All myogenic cell types (myogenic progenitors, satellite cells, myoblasts, and myocytes) shared the expression of lineage markers, including ACTC1, JSRP1, CDH15, and FITM1, and were negative for PDGFRA (Figures 6D, S14A and S14B). CDH15 was previously reported as encoding a universal surface marker for isolating myogenic cell populations from human embryonic and fetal limbs, and CDH15⁺/PDGFRA⁻ cells exhibited enhanced myogenic ability in vitro.³² Genes associated with pluripotency, such as SMC2 and RFC3 and TOP2A, were overexpressed in PSC_Myo, rather than satellite cells (Figure S14A). It has been reported that the postnatal satellite cells that reside in mature skeletal muscle exhibit the minimal expression of cell cycle genes,³² and can be activated to re-enter a myogenic program to replace damaged muscle.³⁵ Using RNA-FISH technology, we observed TOP2A⁺ cells predominantly distributed within muscle cell clusters of ovine fat tail tissues (Figure S14C). In addition to typical myogenic cells exhibiting a concentrated spatial distribution, we also observed some JSRP1⁺ cells, which may act as a supportive structure for committed adipocytes (Figure 6D).

Similar to previous studies, the differentiation trajectory of the myogenic lineage initiated with PSC_Myo and culminated with differentiated myoblasts and myocytes. However, we observed that satellite cells were serving as transitional cells with two branches (T1 and T2) and PAX7-overexpression,³² with satellite cells in the T1 branch continuing to terminal differentiation (Figure 6D). Furthermore, we identified several key genes positively regulating myogenesis with distinct gene expression patterns, including overexpressed genes in satellite cells, such as CLU, PDLIM4, and SPARCL1; upregulated genes in myoblasts, such as SPG21, JOSD2, and RBM24; genes continuously increasing in expression following myogenesis, such as MYH3, TNNT3 and ACTC1; and genes that were exclusively expressed in myoblasts, such JPH2, ACTA1, and TNNI2 (Figures 6D and S14D–S14F). Also, the activities of several other genes were also negatively associated with the entire myogenic process (Figure S14G). Overall, under effectively balanced the expression of these regulators, myogenic progenitors undergo specialization, determination, differentiation, and fusion to form mature myofibers.

Global gene regulatory networks governing lineage specialization

Based on gene expression, chromatin accessibility and motif discovery, we used a single-cell regulatory network inference and clustering (SCENIC) approach to decipher the enhancer-driven GRNs (eGRNs) for each lineage (Figure 7A). The generated eGRNs indicate enrichment in unique TF profiles that regulate lineage specialization. For angiogenesis, we identified shared TFs among two clusters of VSMC, including PRRX1, BACH1, JUNB, NFATC4, and EBF2. On the other hand, NFE2L2, MEOX2, KLF4, and NFIB emerged in the VSMC_C2 cluster, indicating their critical role in the phenotype switching. Identified as positive regulators for VSMC_C2 differentiation, the RAI14 gene was regulated by the JUNB and NFE2L2 TFs, NRP1 was jointly regulated by both KLF4 and NFE2L2, and ANK2 was independently regulated by KLF4 (Figure 7B). In addition, we observed similar eGRNs patterns in the osteogenic and chondrogenic lineages (Figures 7C and 7D). Compared to osteoblasts and chondrocytes, TFs and their cognate genes were abundantly enriched in precursor populations, highlighting the pivotal roles of TFs for initializing lineage specialization and triggering transcriptional cascades. Based on these results, TWIST1 acts as a hub TF and cooperates with TCF4, MEIS1, and EMX2 to regulate gene expression in preosteoblasts. For example, in the regulators of preosteoblasts, CCDC80 was governed by the TWIST1 and TCF4 TFs, and PTPRD was regulated by three TFs: TWIST1, MEIS1, and IRX5. In addition, RUNX2 and RARG influence the expression of osteoblast-specific marker genes, such as PAPPA2, CXCL14, and COL13A1. Members of the HOX and NFI TF families were also specifically detected in prechondrocytes and chondrocytes, including HOXA11, HOXC12, NFIB, NFIC, and NFIX.

Figure 7.

The complex eGRNs underpinning lineage specialization

(A) A combined heatmap/dot plot diagram showing TF expression for the eRegulon on a color scale and cell-type specificity (RSS) for the eRegulon on a size scale. Cell types are ordered based on their gene expression similarity.

(B) Constructed eGRNs for VSMC_C1 and VSMC_C2.

(C) Constructed eGRNs for preosteoblasts and osteoblasts.

(D) Constructed eGRNs for prechondrocytes and chondrocytes.

(E) Constructed eGRNs generated for myogenic progenitors and satellite cells, myoblasts, and myocytes. The large node circles represent TFs; the diamond nodes represent highly accessible regions; and the small circles represent highly expressed genes. The colors of the small circles represent the log₂FC of the gene expression values in target cell types compared to the other cells. The TF–region links are colored by TF, and the region–gene links are colored gray.

The myogenic lineage demonstrated enhanced gene expression regulatory activity, with the majority of highly expressed genes in myogenic progenitor-satellite cells or myoblast-myocytes being regulated by multiple TFs (Figure 7E). Some TFs exhibited stage-dependent properties. For example, the PAX7, MYF5, TCF21, and MSC TFs were remarkably active in myogenic progenitors and satellite cells, but SOX6, TEAD, MYOG, TCF12, and ZBTB18 played significant roles in myoblasts and myocytes. The MYOD1 and MYF6 TFs may execute more critical functions during myogenesis due to their presence across myogenic cells. Taken together, the inferred diverse combinations of TFs participate in the regulatory networks controlling dynamic gene expression states in different cell types and during lineage specialization.

Discussion

Maintaining a dynamic equilibrium in adipose tissue is crucial for maintaining energy storage and expenditure, supporting adipose tissue function, and protecting against metabolic disease. Understanding the origins and developmental hierarchy of adipose cells during embryogenesis is key to comprehending the diversity observed in these tissues. Several sheep breeds, through adaptation to environmental and physiological challenges, have evolved a unique phenotype characterized by the prenatal accumulation of large volumes of fat in their tail tissue, which represents an excellent model for disentangling the cellular pathways and GRNs that underpin embryonic fat deposition in mammals.^16,17 In this study, we successfully and unbiasedly delineated the global transcriptomic and epigenomic landscapes to generate an expression atlas of embryonic ovine tail tissues encompassing the complete stages of adipogenesis at single cell resolution without the extraction of stromal vascular fraction (SVF) or nuclei. Though integrated analyses of the six principal lineages that composed the atlas, our study provides new insights into specific cell fate decisions governing adipogenesis, angiogenesis, osteogenesis, chondrogenesis, and myogenesis, and the key genes governing lineage specialization (Figure 8). Importantly, our work defines a developmental hierarchy of adipose progenitor populations consisting of MCM5⁺/SFRP4⁺ progenitor and stem cells, COL6A6⁺/ALDH1A1⁺ connective tissue progenitors, and GUCY1B1⁺/RAI14⁺ vascular smooth muscle cells. From this, we propose that the multicellular origins of adipocytes anticipate the massive fat deposition observed in the tail tissues of fat tail sheep breeds.

Figure 8.

An infographic showing the complex differentiation relationships for each lineage specialization

Also shown are the key regulators that determine lineage specialization and define marker genes for each progenitor population.

Several strands of evidence suggest that adipogenic patterning is varied across depots.^1,4,7,14 For example, in the mouse, the formation of perigonadal WAT is postnatal, but subcutaneous WAT (sWAT) is initiated during embryonic days 14–18 and, postnatally, the number of adipocytes in sWAT remains relatively stable.¹⁰ The lineage hierarchy of adipogenic progenitors is ultimately responsible for the varied fat distribution patterns and distinct metabolic characteristics across depots in both murine and human adipose tissues. It is well established that different fat depots share either common or distinct cellular origins, and even within the same fat depot, adipocytes can arise from various progenitor populations, as evidenced by the three cellular origins described in this study.² In this context, adipocytes likely have a more complex developmental history. Overall, adipogenic progenitors can be categorized into two classes. The first corresponds to cells with a germline origin that are located within specific mesodermal structures during the early embryonic stage. Using the Cre-LoxP lineage tracing system in the mouse, researchers have discovered a Wt1⁺ lineage with limited expression in the intermediate mesoderm, which contributes to a subset of adipocytes in visceral WAT rather than subcutaneous WAT.⁸ In addition, the Myf5⁺ and Pax3⁺ lineage that emerges in early embryos can give rise to adipocytes in both visceral and subcutaneous WAT depots.^7,36 During organogenesis and tissue development, these progenitor populations undergo lineage decisions, progressively narrowing the spectrum of cell types that can be generated and culminating in the final commitment for adipogenesis. Notably, in the present study, the expression of Wt1, Myf5, and Pax3 was absent in adipogenic progenitor populations because of their early expression properties.

The second class of adipogenic progenitors are heterogeneous FAPs identified in the SVF, which can be recruited to produce adipocytes and constitute a developmental adipogenic source with limited differentiation potential. A previous study identified ALDH1A3⁺/PDGFRA⁺ adipogenic progenitors in human and mouse WAT.⁶ In our study, the expression of ALDH1A3 was evident in both PSC_C 4, PSC_C5, and three CTP clusters. In addition, the three predicted adipogenic progenitor populations exhibited specific overexpression of the ALDH1A1 gene, which encodes another member of the ALDH family (Figure 3B). In another parallel study using murine subcutaneous WAT, researchers identified Dpp4⁺ adipogenic progenitors that co-expressed with Wnt2, reside in the reticular interstitium structure of WAT, and can give rise to Icam1⁺/Pparg⁺ preadipocytes and a CD142(F3)⁺/Clec11a⁺ “adipogenesis regulatory cells.”^5,37 The PSC_C1, PSC_C4 and CTP_C2/C3 groups exhibited relatively high expression of DPP4 (Figure 3B). Other evidence also indicates that certain cellular components of vasculature in adipose, but not in other tissue types, can contribute to adipocyte formation, including pericytes,³⁸ specialized endothelial cells,³⁹ and vascular smooth muscle cells.⁴⁰ The results of our study suggest that one subpopulation of VSMC with a synthetic phenotype can give rise to committed adipocytes. Relevant to this, another study showed that Wt1⁺ mesothelium, one special structure that does not exist in subcutaneous WAT, can also produce adipocytes in visceral fat depots.⁸

DBI, also known as acyl-CoA-binding protein (ACBP), is a small, evolutionarily conserved molecule that binds medium-to long-chain acyl-CoA esters, serving as an intracellular buffer that regulates acyl-CoA availability. In mammals, DBI has been implicated in the regulation of lipid metabolism, lipogenesis, and systemic energy balance.⁴¹ For example, DBI-knockout mice exhibit impaired lipid storage and altered metabolic phenotypes, underscoring its functional role in adipocyte biology.⁴² In our study, DBI expression was dynamically regulated along the adipogenic pseudotime trajectory, with peak expression occurring at the transition from late preadipocytes to early adipocytes. This temporal expression pattern supports a model in which the PPARG–DBI interaction functions downstream of lineage commitment, potentially contributing to terminal lipid accumulation and metabolic maturation in differentiated adipocytes.

Notably, we observed increased chromatin accessibility at a DBI-associated enhancer region specifically in committed adipocytes, accompanied by the presence of a predicted PPARG-binding motif. Dual-luciferase reporter assays validated the transcriptional activity of this enhancer element, and PPARG overexpression further potentiated this effect in ovine adipose-derived stem cells. These results strongly suggest that PPARG might directly upregulate DBI through enhancer-mediated transcriptional activation during adipogenic differentiation. We propose that the PPARG–DBI regulatory axis plays a critical role in supporting the enhanced lipogenic capacity characteristic of the fat-tailed phenotype in sheep, a trait of considerable agricultural and evolutionary interest. These findings offer new insights into how transcriptional regulation and lipid metabolism converge during adipose tissue expansion, with potential implications for both livestock breeding and human metabolic disorders such as obesity and lipodystrophy.

Our results indicate skeletal muscle and VSMCs exhibit conserved expression of ACTA2, followed by the distinct expression of VSMC lineage marker genes (GUCY1B1, AVPR1A, MYH11, and TAGLN) and Myogenic lineage marker genes (ACTC1, JSRP1, and CDH15) (Figures 2E and 6D). Phenotypic switching is a unique property of VSMCs that can also contribute to adipogenesis within fat tail tissue, and which was demonstrated by the trajectory analysis conducted as part of this study (Figure 2F). The plasticity of VSMCs is also evident in the extensive organotypic heterogeneity of arterial VSMCs compared to the relative homogeneity of venous VSMCs.⁴³ While the dynamic development of skeletal muscle in human, mouse, pig, and cattle has been well established, our study provides new insights into understanding the molecular differences between myogenic progenitors and satellite cells, which are positive for PAX7. Regarding human limb myogenesis, PAX3⁺ myogenic progenitors first emerge at early embryonic stages (weeks 5–6) without the differentiation of myoblasts or myocytes, followed by a dynamic shift in expression from PAX3 to PAX7. In the later fetal and adult stages, myogenic progenitors and satellite cells are predominantly positive for PAX7.³² Osteogenesis, chondrogenesis, and endochondral ossification are integral to the development of axial and appendicular skeletal systems, descending from paraxial and lateral plate mesoderm. As an extension structure of vertebrae, the coccyx exhibited unique osteogenesis and chondrogenesis patterns compared to the appendicular skeleton, including the moderate expression of SOX9 in both osteogenic and chondrogenic progenitors (Figures 6A and 6C), and independent origins of osteoblasts and chondrocytes. During long bone morphogenesis, cells that exhibit the positive expression of PRRX1, a low level of SOX9 expression, and high PDGFRA expression undergo cell fate specialization into osteogenic and chondrogenic lineages.³¹

Limitations of the study

Our study elucidated the lineage hierarchy from progenitor populations to adipocytes, vascular smooth muscle cells, osteoblasts, chondrocytes, and myocytes, while also identifying causal genes and specific GRNs underpinning each lineage specialization (Figure 8). Notably, our findings highlight three adipogenic pathways contributing to the embryonic fat deposition trait, with particular emphasis on the critical role of the PPARG-DBI regulon in adipogenic regulation. However, the differentiation trajectories were mainly inferred from single-cell bioinformatics analyses. Although RNA-FISH was used for the spatial localization of progenitor cells, more precise lineage-tracing is needed to definitively confirm cell origins and differentiation pathways. Additionally, the conservation and functional role of the PPARG-DBI regulon in mammalian adipogenesis require further investigation.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Lin jiang (jianglin@caas.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• Data: All data needed to evaluate the conclusions in this article are presented in the article and/or supplementary materials. The raw sequence and processed data generated from developing ovine fat tail tissues have been deposited in the GEO database with the accession number GSE254357.

• Code: The code used to pre-process, analyze the data, and generate the figures of this study has been deposited in the GitHub repository: https://github.com/JGangHan.

• Additional information: Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

Acknowledgments

This research was funded by the National Natural Science Foundation of China (grant no. 32222079), the Earmarked Fund for Modern Agro-industry Technology Research System (grant no. nyb-tx-15), and the Key Research Program in Ningxia Hui Autonomous Region (grant no. 2023BCF01007). J.H. was supported by the UCD-GSCAAS joint PhD Program in Agriculture and Food Science, and the China Scholarship Council. We would like to thank the team at the Institute of Animal Science, Ningxia Academy of Agriculture and Forestry Sciences for their assistance in the care and management of experimental animals.

Author contributions

J.H.: methodology, data curation, formal analysis, and writing - original draft. S.M.: methodology and validation. Q.W.: formal analysis. Z.Z.: resources. Y.Z.: methodology and resources. L.T.: validation. G.Z.: methodology. Y.P.: resources. Q.Z.: resources. X.H.: resources. Y.M.: conceptualization and supervision. D.E.M.: supervision and writing - review and editing. L.J.: funding acquisition, supervision, and writing - review and editing.

Declaration of interests

The authors declare that they have no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Sheep	National Tan Sheep Conservation Farm, China	N/A
Chemicals, peptides, and recombinant proteins
MACS Tissue Storage Solution	Miltenyi	130100008
0.25% trypsin	Gibco	15050065
2.5 mg/mL collagenase IV	Sigma-Aldrich	C5138
15 μg/mL DNase I	Sigma-Aldrich	AMPD
Erythrocyte lysis buffer	Solarbio	R1010
40/70 μm strainer	BD Falcon	N/A
FBS	Gibco	10099141
PBS	Gibco	10010031
Critical commercial assays
Chromium Next GEM Single Cell 3ʹ Reagent Kits v3.1	10x Genomics	1000121
Chromium Single Cell ATAC Library & Gel Bead Kit	10x Genomics	1000110
Deposited data
Raw single cell dataset	GEO	GSE254357
Oligonucleotides
pcDNA3.1(+) vector	Miaoling Bio	P0157
pGL3-Promoter vector	Promega	P0194
LPL-target probe	Wuhan Servicebio Technology	N/A
JSRP1-target probe	Wuhan Servicebio Technology	N/A
GUCY1B1-target probe	Wuhan Servicebio Technology	N/A
TOP2A-target probe	Wuhan Servicebio Technology	N/A
ALDH1A1-target probe	Wuhan Servicebio Technology	N/A
MCM5-target probe	Wuhan Servicebio Technology	N/A
SFRP4-target probe	Wuhan Servicebio Technology	N/A
Software and algorithms
CellRanger	https://github.com/10XGenomics/cellranger	v6.0
Cell Ranger ATAC	https://github.com/10XGenomics/cellranger-atac	v2.1.0
Seurat	https://satijalab.org/seurat/	v4
ArchR	https://www.archrproject.com/	v1.0.1
Monocle2	https://cole-trapnell-lab.github.io/monocle-release/	v2.4.0
Scanpy	https://scanpy.readthedocs.io/en/stable/	v1.9
scVelo	scVelo	v0.2.5
clusterProfiler	https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html	v4.0
CellChat	CellChat	v2
Scenic+	https://github.com/aertslab/scenicplus	v1
DoubletFinder	https://github.com/chris-mcginnis-ucsf/DoubletFinder	v2.0
Cytoscape	https://cytoscape.org/	v3.9.1

Experimental model and study participant details

Animals

Female Tan breed sheep, approximately 12 months old, were maintained in the National Tan Sheep Conservation Farm, Ningxia Province, China. According to our previous studies,^17,44 we optimized our sampling strategy by increasing the density of sample time points and reducing the time intervals between them. This approach allowed us to minimize the number of sample replicates needed, obtaining just one sample for each developmental stage (F50, F55, F60, F63, F66, F69, F72, F75, and F80). The embryonic tail tissues for nine time points were used for scRNA-seq, and four time points (E55, E63, E69 and E75) for snATAC-seq. To minimize potential confounder batch effects introduced during library preparation and sequencing, we refined sample preparation procedures¹⁷ by applying one consistent phased-estrus synchronization procedure. Briefly, this procedure involved: 1) implantation of a controlled internal drug releasing (CIDR) progesterone device (Zoetis, Victoria, Australia) for 11 days; 2) removal of the CIDR device and administration of 1 mL of cloprostenol (0.1 mg/mL); 3) a check for estrus after two days; 3) artificial insemination using 2 mL high-quality semen diluted in vitamin B12 solution from a single ram; and 4) confirmation of pregnancy using a B-ultrasonic machine (Gandaofo, Henan, China) after 30 days. The interval between two successive procedures was either 3 or 5 days, depending on the target time points. Tail tissues were collected 50 days after the last artificial insemination. Sex or gender differences were not considered in this study, as the sex hormone-producing organs are not fully developed during the embryonic stages examined. All animal experimental procedures involved in this study were reviewed and approved by the Animal Welfare and Ethics Committee of the Institute of Animal Science, Chinese Academy of Agriculture Sciences (approval number: IAS2021-73) and the University College Dublin Animal Research Ethics Committee (approval number: AREC-E−23-21-MacHugh).

Primary cell cultures

The primary ADSCs were isolated from fat tail tissues at E80. Procedures for cell isolation, culturing, and adipogenic induction were conducted based on our previously described protocol.⁴⁴

Method details

Single cell/nuclei suspension, library construction, and sequencing for scRNA-seq and snATAC-seq

The ovine embryonic tail tissues collected at nine developmental stages were stored immediately in 5 mL tubes filled with MACS Tissue Storage Solution (Miltenyi, Cologne, Germany) and placed on ice. These ice-stored samples were then transported to the laboratory within 1 h for independent tissue dissociation and digestion. This process involved three washes in PBS (Gibco, Waltham, USA), mincing into approximately 1 mm³ pieces, and treatment with a mixture of 2.5 mg/mL collagenase IV, 15 μg/mL DNase I, and 0.25% trypsin, while shaking at 37°C for at least 60 min. After sufficient dissociation, the cell/tissue suspension was filtered using a 70 μm strainer (Merck Millipore, Darmstadt, Germany) to remove large clumps and then centrifuged at 400 × g for 5 min. The cell pellet was then resuspended in erythrocyte lysis buffer (Solarbio, Beijing, China) for 5 min at RT to lyse red blood cells, followed by another round of filtering and centrifuging. Finally, the cell pellet was resuspended in 0.04% BSA–PBS and the quality of the single cell suspension was evaluated using a Count Star device (Alit Biotech, Shanghai, China).

The procedures for cell capture, cDNA synthesis, library construction, and sequencing were performed by CapitalBio Technology Co., Ltd (Beijing, China). Briefly, 8,700 cells were loaded on a Chromium Single Cell Controller (10x Genomics, Shanghai, China) to obtain 5,000 target cells and create single-cell gel beads in emulsion (GEMs) using the Chromium Next GEM Single Cell 3ʹ Reagent Kit v3.1 and the Chromium Single Cell A Chip Kit (10x Genomics) according to the manufacturer’s instructions. This was followed by cell lysis, RNA barcoding, and reverse transcription using a C1000 Touch Thermal Cycler (Bio-Rad, Shanghai, China) to generate cDNA. The GEMs were then disrupted, and single-stranded cDNA was isolated and purified using DynaBeads (Thermo Fisher Scientific, Shanghai, China). The cDNA sample was used to generate a sequencing library, which was quality- and quantity-checked on an Agilent 4200 TapeStation System (Agilent Bioscience, Hangzhou, China). Library sequencing was performed on an Illumina NovaSeq 6000 instrument with a PE150 read format and a sequencing depth of at least 100,000 reads per cell, which produced approximately 150 GB of data per sample.

For the analysis of chromatin accessibility, single cell suspensions for the E55, E63, E69, and E75 stages were bisected with one portion used to create single nuclei suspensions following the 10x Genomics Nuclei Isolation method for snATAC-seq. These suspensions were centrifuged at 100 × g for 10 min to obtain a cell pellet, which was then resuspended in chilled Lysis Buffer (10x Genomics) for 3–5 min on ice. After addition of 1 mL of chilled Wash Buffer (10x Genomics) and centrifugation at 500 × g for 5 min at 4°C, the nuclei suspensions were resuspended in chilled Diluted Nuclei Buffer (10x Genomics). The Count Star device was then used to assess the concentration and integrity of the single nuclei suspensions. Following this, the transposition, nuclei capture, library construction, and sequencing procedures were conducted at CapitalBio Technology Co., Ltd. Briefly, the nuclei suspensions were incubated with transposase (10x Genomics) to generate transposed DNA for each nucleus, which were then partitioned into GEMs for library construction using the Chromium Chip E Single Cell Kit and Chromium Single Cell ATAC Library and Gel Bead Kit. Library quality- and quantity-checking was assessed using the Agilent 4200 TapeStation System, ensuring that fragment sizes in each snATAC-seq library ranged between 150 and 1000 bp without adapter sequence contamination. Finally, sequencing was performed on an Illumina NovaSeq 6000 instrument using a PE50 read format and a sequencing depth of at least 25,000 read pairs per nucleus, which produced approximately 25 GB of data per sample.

Bioinformatic workflow of scRNA-seq dataset

The Cell Ranger Toolkit software (v6.0, 10x Genomics) was used to align raw sequencing reads with the ovine reference genome (Oar_rambouillet_v1.0).⁴⁵ The Seurat R package (v4)⁴⁶ was deployed for data normalization, cell quality control, clustering analysis, dimension reduction, detection of DEGs, and visualization. To remove low quality cells, we retained cells with UMI counts ≥300 and applied a soft threshold for the proportion of MT genes, and cells predicted as doublets or multiplets by DoubletFinder (v2.0)⁴⁷ were further removed (Table S1). The obtained clean cells-features count matrix was subjected to global-scaling normalization, highly variable genes (top 2000) detection, scaling or linear transformation, linear dimension reduction using Seurat. Subsequently, the Louvain algorithm, a modularity optimization technique, grouped all cells into 37 clusters, which were further visualized via the Uniform Manifold Approximation and Projection (UMAP) approach with the top 50 dimensions and default resolution. And then, these cluster were manually annotated into 18 main cell types based on the expression of canonical marker genes.

We utilized the "FindAllMarkers" function in Seurat to identify highly expressed genes (HEGs) across all cell types, focusing on those with a log₂ fold-change (log₂FC) > 0.5 and a Bonferroni adjusted p-value <0.5. The identified HEGs were assigned into four categories according to their average gene expression levels across different cell types, including A+, A, A−, and H. The A+ class indicates a strong marker characterized by exclusively high expression in a specific cell type; the A and A− marker classes, representing general and medium markers, respectively, are defined by high expression in a range of 10%–30% of cell types; others were defined as H. The classification of these markers was determined using the following formulae:

$$\mathrm{A}+:Nc=Nc-\sum _{n=2}^{Nc}1\left({MeanExp}_i>0.3\ast {MeanExp}_1\right)$$

$$\mathrm{A}:Nc>\left\{Nc-\sum _{n=2}^{Nc}1\left({MeanExp}_i>0.3\ast {MeanExp}_1\right)\right\}\ge Nc-0.1\ast Nc$$

$$\mathrm{A}-:Nc-0.1\ast Nc>\left\{Nc-\sum _{n=2}^{Nc}1\left({MeanExp}_i>0.3\ast {MeanExp}_1\right)\right\}\ge Nc-0.3\ast Nc$$

Where Nc is the number of all cell types; MeanExp₁ is the average expression of a target gene in a target cell type; MeanExp_i is the average expression of a target gene in non-target cell types; and 1() is an indicator function that has a value of 1 when the condition inside the parenthesis is true, and 0 if false.

For detection of DEGs across cell types within the same lineage, and to identify genes most critical for lineage specification and differentiation, we applied a more stringent log₂FC threshold (|log₂FC| > 0.75), with an adjusted p-value <0.05. The identified HEGs or DEGs were converted to human gene symbols using the AnnotationHub and BioMart⁴⁸ packages. We then conducted gene ontology (GO) enrichment analysis using the clusterProfiler package (v4.0) and the org.Hs.e.g.,.db database.⁴⁹ GO terms that had a p value <0.01 and a q value <0.05 were considered significantly enriched.

Bioinformatic workflow of snATAC-seq dataset

We employed the Cell Ranger ATAC Toolkit software (v2.1.0, 10x Genomics) to align sequencing reads with the ovine reference genome using default parameters. The ArchR R package (v1.0.1)⁵⁰ was then used to generate the cell-peak fragment matrix derived from this mapping. Cells were retained based on specific criteria: a transcription start site (TSS) score ≥2 and a fragment number range of 1,000 to 100,000 (Table S1). Doublets were further predicted and removed using the ArchR default parameters. The obtained clean cells-peaks matrix was subjected to dimension reduction, clustering, and UMAP visualization. Following this, gene scores were calculated by assessing chromatin openness within 100 kb up- or downstream of each gene. This process incorporated an exponential weighting function that accounts for the activity of putative distal regulatory elements in a distance-dependent fashion, while imposing gene boundaries to minimize the contribution of unrelated regulatory elements to the gene score.

Based on imputed gene score matrix, we used our scRNA-Seq data as a reference annotation to train the classifiers, automatically annotating cell clusters of snATAC-seq dataset into the same 18 main cell types identified by the scRNA-seq dataset. However, due to the inherently sparse nature of chromatin accessibility data, exhibited limited resolution in certain lineages. Clusters that were transcriptionally distinct in the scRNA-seq dataset were collapsed into broader subpopulations. Specifically: PSC_C1 to PSC_C3 were grouped as PSC subpopulation 1 (PSC_S1); PSC_C4 to PSC_C6 as PSC subpopulation 2 (PSC_S2); Chondrocyte_C1 and Chondrocyte_C2 were combined as Chondrocytes. In the downstream analyses, including gene regulatory network construction and integration of gene activity with chromatin accessibility, we used a harmonized set of cell type annotations, ensuring comparability between scRNA-seq and snATAC-seq modalities.

ArchR is able to achieve peak calling for each cell type with one iterative overlap peak merging approach. The resultant peak matrix was then used to perform DARs to identify highly accessible regions across cell types with the “getMarkerFeatures” function and default parameters. Peaks that had an FDR ≤0.05 and log₂FC ≥ 1 were categorized as significant DARs.

Differentiation trajectory analysis

We used the Scanpy Python package (v1.9),⁵¹ the Monocle2 R package (v2.4.0)⁵² and the scVelo Python package (v0.2.5)⁵³ to conduct differentiation trajectory analysis. Initially, using Scanpy, we constructed PAGA trajectory for six focused lineages, including progenitor population, vascular smooth cell, adipogenic, osteogenic and chondrogenic; the myogenic lineage was excluded from the combined PAGA network because it had an independent differentiation trajectory. Following this, depending on the constructed global PAGA network, the cell types associated with adipogenesis were extracted to construct one dedicated PAGA trajectory.

The Monocle2 approach was more useful for inferring the precise pseudotime of cell state transitions during specific lineage specialization. In summary, the process involved the extraction of targeted cell types, removal of lowly expressed genes (expressed in less than ten cells), and estimation of size factors and dispersions with default parameters. Lastly, the dynamical expression patterns of target genes were illustrated and visualized using the “plot_genes_in_pseudotime” and “plot_pseudotime_heatmap” functions. To detect key regulators driving lineage specialization or bifurcation events, we applied the Branched Expression Analysis Modeling (BEAM) method implemented in Monocle2. BEAM uses a branch-aware generalized linear model to identify genes whose expression patterns differ significantly between pseudotime branches.

By evaluating the ratio of unspliced to spliced transcripts across the genome, the RNA velocity approach can be used to reveal cellular differentiation pathways, dynamic developmental trajectories, and root cell populations. To perform this technique, the output BAM files from Cell Ranger were used as input files for Velocyto (v0.17.17)⁵⁴ using default parameters. Subsequently, the LOOM files containing the count matrix of spliced and unspliced mRNA were used to perform RNA velocity analysis using scVelo. In addition, we performed diffusion map analysis using the scanpy.tl.diffmap function, followed by visualization with scanpy.pl.diffmap, both implemented in the Scanpy package. This approach provided an additional unsupervised embedding of cell states and helped validate the relationships among target celltypes by capturing continuous developmental transitions.

Gene regulatory network (GRN) construction

To enhance the integration of scRNA-seq and snATAC-seq, we employed the SCENIC approach⁵⁵ to construct the eGRNs. Our implementation of the SCENIC pipeline involved the following steps. (1) Preprocessing of scRNA-seq data via Scanpy⁵⁶ with default parameters, including normalization, high variable features identification, scaling, and dimension reduction and UMAP embedding. (2) Preprocessing of snATAC-seq data via pycisTopic (v1.0.2)⁵⁵ with default parameters, including generation of pseudo-bulk ATAC-seq profiles, calling and merging peaks, and quality control. Genome coordinates and corresponding gene information of fragments were annotated using the ovine Oar_rambouillet_v1.0 reference genome. Using the count matrix of ATAC-seq fragments over consensus peaks and all metadata from snATAC-seq cells to create one cisTopic object and conduct topic modeling with automatic selection of the optimal number of topics. (3) Evaluation of DARs per cell type, which were identified as candidate enhancer regions. (4) Generation of a sample-specific motif database for this study using the consensus peaks generated above and motif collection database of cisTarget (v1.0.2),⁵⁵ which were used to detected motifs that were enriched within candidate enhancer regions. (5) Merging the scRNA-seq and snATAC-seq datasets into a pseudo multi-omics dataset, which was then used to create one SCENIC object incorporating the motif enrichment results, using this to calculate region-gene correlations and TF-gene correlations, identify enhancer-regulons and construct the final eGRNs. Finally, eGRNs generated with Scenic+ were visualized using the Cytoscape software tool (v3.9.1).⁵⁷

The following key criteria or thresholds were applied during eRegulon construction.

• (1)Motif Enrichment (run_pycistarget): motif scanning was performed using the run_pycistarget function from scenicplus.wrappers, based on the v10nr_clust motif annotation database. We set run_without_promoters = True, thus considering only distal regulatory elements (enhancers). Only motif-enriched regions were retained, ensuring direct TF–target relationships.
• (2)Regulatory Network Construction (build_grn): eRegulons were built using build_grn from scenicplus.grn_builder. Minimum number of target genes per eRegulon: min_target_genes = 10. Motif–gene correlation threshold: rho_threshold = 0.05. p-value threshold for regulatory significance: adj_pval_thr = 1 (i.e., no additional filtering beyond motif presence and expression). Only direct eRegulons (i.e., motif present in accessible regions linked to target genes) were retained in the final network (keep_extended_motif_annot = False was enforced during downstream filtering). eRegulons were merged by default if multiple motifs of the same TF converged on similar target sets (merge_eRegulons = True).
• (3)Enrichment Scoring (score_eRegulons): We calculated gene- and region-based enrichment scores using the score_eRegulons function. AUC threshold for signature enrichment: auc_threshold = 0.05.

Intercellular communication analysis

The CellChat R package (v2) was used for intercellular communication analysis with the human ligand-receptor pairs (LR pairs) database.²⁹ Initially, using a log₂FC > 0.5 and a p value <0.01, a total of 297 highly variable genes that encode LR pairs were identified by CellChat across all cell types except an “Unknown” cell population. These genes were used to estimate potential intercellular communication probabilities at both LR pair level and pathway level. Any interaction occurring in fewer than ten cells was excluded from our analysis. The resulting intercellular communication networks were visualized as heatmap, circular plot, hierarchy plot and dot plot using the “netAnalysis_signalingRole_heatmap”, “netVisual_circle”, “netVisual_aggregate” and “netVisual_bubble” functions, respectively.

Cellular assays

Primary adipose-derived stem cells (ADSCs)

The primary ADSCs were isolated from fat tail tissues at E80. Procedures for cell isolation, culturing, and adipogenic induction were conducted based on our previously described protocol.⁴⁴ RT-qPCR. We detected the expression of adipocyte marker genes (ACACA, ACSL1, FABP4, FASN, GPAM, ADIRF, ADIPOQ, ACAT1, PPARG, and DBI) in differentiating primary ADSCs at D0, D2, D4, and D6 (see Table S11 for primer information). Trizol reagent (Invitrogen, Waltham, USA) was used to lyse cells and extract total RNA at each differentiation stage. RNA concentration and quality were evaluated using an Agilent 2100 Bioanalyzer instrument (Agilent, California, USA). The PrimeScript RT reagent Kit (Takara, Shiga, Japan), TB Green Premix Ex Taq Kit (Takara, Shiga, Japan) and an ABI 7500 Real-Time PCR System (Applied Biosystems, Waltham, MA, USA) were used to conduct the RT-qPCR assays.

Plasmid recombination

DNA oligonucleotides (oligos) corresponding to the coding sequences (Table S11) of the PPARG gene were synthesized by Tsingke Biotechnology (Beijing, China), and ligated into pcDNA3.1(+) vector (Promega, USA) between the KpnI and XhoI sites (OE-PPARG). The blank pcDNA3.1(+) vector (OE-NC) acts as a negative control. To evaluate the potential effect on gene expression activity of one 653 bp DBI-specifically accessible region (Chr2:198163303-198163955, Oar_rambouillet_v1.0), the fragments covering the DBI-Peak were amplified with forward primer (GGTCACCTCAGAGCTTCCTC) and reverse primer (AGATTGCTGCCTAAGAGCAGAG), which targeted Chr2:198163003-198164255 in Tan sheep DNA samples. The PCR products were then Sanger-sequenced. Finally, DNA oligos corresponding to the DBI-Peak were synthesized by Tsingke Biotechnology (Beijing, China), and ligated into the pGL3-Promoter vector (Promega, Madison, USA) between the MluI and XhoI enzyme sites (pDBI-Peak). The blank pGL3-Promoter vector (pBlank) was used as a negative control for the pDBI-Peak. Lipofectamine 3000 (Invitrogen, California, USA) was used with these plasmids according to the manufacturer’s protocol. Dual-Luciferase reporter assay system. The pGL3-promoter vector (0.5 μg) (pDBI-Peak or pBlank) was co-transfected with 0.1 μg pRL-TK vector (Promega) and co-transfected with pcDNA3.1 vector (OE-PPARG or OE-NC) into cultured cells according to the manufacturer’s protocol.

Histological analyses

After fixation in 4% paraformaldehyde solution, tissues were thoroughly washed with PBS to remove any residual fixative. A subset of tissues were then placed in a 30% sucrose solution at 4°C until they sank, indicating complete tissue infiltration. The tissues were subsequently rapidly frozen in embedding medium and stained with oil red solution. Additionally, other tissues were sequentially dehydrated in a graded concentrations of ethanol solutions, cleaned with xylene and then embedded in paraffin wax for HE staining and RNA-FISH assays. For HE staining, paraffin sections underwent a systematic process of dewaxing, hydration, pretreatment, and staining in hematoxylin and eosin solution. The sections were then treated with absolute ethanol and sealed with neutral gum. To accurately identify the location of specific cell types within tissues, we conducted RNA-FISH assays for LPL (a committed adipocyte marker), JSRP1 (a myogenic cell marker), GUCY1B1 (a VSMC marker), TOP2A (a PSC_S1 and PSC_Myo marker), ALDH1A1 (a CTP marker), MCM5 and SFRP4 (both PSC_C4 markers). The specific probes for these genes were synthesized by Wuhan Servicebio Technology (Wuhan, China) and detailed information is provided in Table S11.

The RNA-FISH procedure involved several steps. Paraffin sections were treated with dewaxing solution twice, passed through a graded ethanol series, exposed to repair solution and protease K (Servicebio, Wuhan, China). Pre-hybridization solution (Servicebio) was then applied to create an optimal chemical environment for hybridization, and to enhance the specific and effective binding of probes to target mRNA molecules. Subsequently, paraffin sections were hybridized with probes for the target gene, corresponding branch probes (Servicebio), and signal probes (Servicebio) to achieve specific hybridization signaling. The above step was repeated to simultaneously localize multiple target genes. Sections were then thoroughly washed to remove hybridization solution and incubated with DAPI counterstain. Finally, a fluorescent Microscope (Nikon, Japan) was used to detect corresponding signals and collect images with the following wavelengths: DAPI (blue), excitation wavelength 330–380 nm and emission wavelength 420 nm; FAM488 (green), excitation wavelength 465–495 nm and emission wavelength 515–555 nm; CY3 (red), excitation wavelength 510–560 nm and emission wavelength 590 nm; CY5 (yellow), excitation wavelength 608–648 nm and emission wavelength 672–712 nm.

Quantification and statistical analysis

Statistical analyses to determine significance were conducted using Student’s t test or two-way analysis of variance (ANOVA). A p-value of less than 0.05 was considered statistically significant. Detailed statistical information is provided in the method details and figure legends.

Additional resources

No additional resources were generated in the study.

Published: September 3, 2025