Ethics statement

The experimental design and procedures were approved by the Animal Care and Use Committee of Northwest A&F University (Shaanxi, China) (A20210816).

Sample preparation

The captive FMD used in this study were raised at Shenglinyuan Biotechnology Co., Ltd. in Fengxian County, Shaanxi Province, China. Musk gland tissues were obtained from 6-month-old and 6-year-old male FMD in the non-musk secretory period, both of whom died of hairball disease. The musk glands were washed with phosphate buffer saline and placed in a culture dish. Most of the musk glands were cut into 1 mm³ pieces and immediately placed in liquid nitrogen. The rest of the glands were cut into 1 and 2 mm³ pieces and fixed in paraformaldehyde and in situ hybridization fix solutions, respectively (Servicebio, Wuhan, China).

Hematoxylin and eosin staining

Following overnight fixation in paraformaldehyde, musk gland tissues were dehydrated in an ethanol solution and further incubated with xylene for 30 min. Subsequently, musk gland tissues were embedded in paraffin blocks that were cut at a thickness of 4 μm. The sections were transferred to clean slides for hematoxylin and eosin (H&E) staining, and the slides were deparaffinized in xylene for 40 min and hydrated in an ethanol series. The hydrated slides were incubated in a hematoxylin solution and washed with tap water. The slides were differentiated with 1% hydrochloric acid–ethanol, stained blue with 0.6% ammonia, and then stained with eosin. Finally, the slides were mounted with neutral resin after rinsing with absolute ethanol.

Preparation of single-nucleus suspension

For each sample, 500 μL of precooled lysis buffer was used to lyse the musk gland tissues, and a rubber grinding rod was used to vertically press them. Then, 700 μL of nuclei washing buffer was added, and the tissue homogenate was filtered through a 70-μm cell strainer (Corning, New York, USA) to obtain approximately 1 mL of filtrate. Subsequently, 1 mL of 50% iodixanol was added to the filtrate, and a gradient solution was prepared using 1 mL of 33% iodixanol and 2 mL of 30% iodixanol. Next, 2 mL of iodixanol solution containing nuclei was added to the liquid surface of 30% iodixanol. Following centrifugation, the white nuclei layer was transferred to a new 15-mL centrifuge tube, and nuclei washing buffer was added to resuspend the nuclei. The mixture was centrifuged, and the precipitate was resuspended in the nuclei-washing buffer. After several repetitions, the nucleus suspension was filtered through a 40-μm cell strainer (Corning, New York, USA). After further centrifugation, diluted nuclei buffer was added to resuspend the nuclei precipitate, and the nucleus suspension stained with trypan blue was detected using a microscopic cell-counting plate.

Library construction and sequencing

The single-nucleus suspension and transposase mixture were mixed and incubated according to the manufacturer's instructions (10 × Genomics, Chromium Next GEM Single Cell Multiome ATAC + Gene Expression, protocol CG000338). After entering the nucleus, the transposase preferentially fragmented DNA in the open regions of the chromatin, and sequencing primers were added to the ends of the DNA fragments. A mixed solution of nuclei and enzymes was loaded onto a Chromium Controller (10 × Genomics) to generate Gel Bead-in-emulsions (GEMs).

For the gene expression library, poly(dT) sequences were used to capture pre-mRNA and mature mRNA. After adding labels, the GEMs' structure was ruptured, all cDNA was mixed, and cDNA was amplified using PCR for library construction. P5, P7, the sample index, and sequencing primers were included during the construction of the RNA library by fragmentation, end repair, A-tail addition, ligation, and PCR. For the ATAC library, spacer sequences were used to capture labeled DNA fragments. Then, the GEMs' structure was ruptured, all dsDNA was mixed, and DNA was amplified using PCR for library construction. P7 and the sample index were analyzed using PCR during the construction of the ATAC library. Finally, the resulting libraries were sequenced using an Illumina NovaSeq 6000 sequencing system.

Multiome data processing

Cell Ranger ARC (version 2.0.1) (https://support.10xgenomics.com/single-cell-multiome-atac-gex/software/pipelines/latest/what-is-cell-ranger-arc) from 10 × Genomics was used for quality control of the raw data. For snRNA-seq data, the Cell Ranger ARC first pruned the reads to remove the TSO and poly(A) sequences. Spliced Transcripts Alignment to a Reference (STAR) (version 2.7.10) (https://github.com/alexdobin/STAR) was used to align Read2 with the reference genome (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA317652/) (Dobin et al. 2013). Regarding snATAC-seq data, Cell Ranger ARC first pruned the reads to remove adaptor and primer sequences. BWA-MEM (version 2.2.1) (https://github.com/bwa-mem2/bwa-mem2) was used to align read pairs to the reference genome (Li 2013). Duplication analysis was used to identify reads with unique alignments, and MAPQ > 30 was used for subsequent analyses. The Unique Molecular Identifiers (UMIs) were filtered and corrected using Cell Ranger ARC, which was employed to identify peaks. After filtering the barcodes by combining the single-nucleus RNA and ATAC data, the final effective cell population was formed.

Cell-type clustering and differential expression gene analysis

Seurat (version 3.0.2) (https://github.com/satijalab/seurat/releases) was used for downstream analysis and visualization (Butler et al. 2018). Particularly, Seurat filtered abnormal cells to retain high-quality cells. The filtering indicators included the number of genes identified (between 200 and 3800) and the total number of UMIs in a single cell (<14 000). For the same type of cell, the number of expressed genes was maintained within a certain range. If the value was too high, multiple cell types were probably wrapped in one GEM, and such barcodes were removed. If the total number of UMIs was too high, two or more cells may have entered the same GEM, and such cells were removed. The clean data were normalized using the LogNormalize function. Seurat reduced the dimensionalities of snRNA-seq and snATAC-seq data using principal component analysis (Chung & Storey 2015) and latent semantic index (Deerwester et al. 1990), respectively. The batch effect correction was performed using a harmony algorithm (Korsunsky et al. 2019), and Seurat constructed a weighted-nearest neighbor (WNN) map based on gene expression levels and chromatin openness (Hao et al. 2021). Finally, the Louvain algorithm was used to cluster the cells based on the WNN map (Rotta & Noack 2011).

The FindAllMarkers function implemented in Seurat was used to identify differentially expressed genes (DEGs) between the different clusters. The Wilcoxon test was performed for each gene, and the gene filtering criteria were as follows: (1) P ≤ 0.01; (2) at least |log₂FC| ≥ 0.36 between two groups; and (3) the proportion of cells expressing the gene in the target cluster or control cluster > 25%.

Differential gene analysis between samples

Based on the cell type and sample information to which the cell belongs, Seurat's Wilcoxon rank sum test was used to find DEGs between different samples in a cell type according to the following criteria: (1) P ≤ 0.01; (2) at least |log₂FC| ≥ 0.36 between two samples; and (3) the proportion of cells expressing the gene in any sample > 25%.

Single-nucleus RNA and ATAC association analysis

The peak was annotated as proximal (0–2 kb upstream of the gene) or distal (2–100 kb upstream of the gene), depending on its position relative to the gene. Seurat's Wilcoxon rank sum test was used to determine differential peaks between different samples in a cell type according to the following criteria: (1) P ≤ 0.01; (2) at least |log₂FC| ≥ 0.36 between two samples; and (3) the proportion of cells that detected the peak in any sample > 25%. Subsequently, the upregulated peaks in the samples were subjected to motif enrichment analysis. TF motif information from the JASPAR database (version 2022) (http://jaspar.genereg.net/) was used to identify TFs (Castro-Mondragon et al. 2022).

Seurat was used to correlate the snRNA-seq and snATAC-seq data. Candidate target genes were those that were upregulated in the snRNA-seq data and whose proximal and distal chromatin accessibility changed in the snATAC-seq data. These genes might also be regulated by chromatin accessibility. Candidate target genes were divided into proximal and distal candidate target genes according to the positions of the target genes annotated by peaks. Candidate TFs were upregulated during development and whose corresponding motifs were enriched in the upregulated peaks. These TFs may also be involved in transcriptional regulation. Moreover, the candidate TFs were divided into proximal and distal candidate TFs according to the positions of the target genes annotated by the peaks of enriched TF motifs. Candidate target genes and TFs were linked by peaks to obtain core target genes and TFs. Core TFs could bind to peaks that could regulate core target genes, thereby forming a regulatory network.

Weighted gene co-expression network analysis

The weighted gene co-expression network analysis (WGCNA) package (version 1.71) (https://rdocumentation.org/packages/WGCNA/versions/1.71) was used to construct co-expression networks (Langfelder & Horvath 2008). After filtering the genes, gene expression values were entered into WGCNA to create co-expression modules. The expression correlation coefficients of the remaining genes were calculated to determine the appropriate soft threshold for creating gene networks using a scale-free topology model. Correlations between modules as well as between modules and genes were also calculated by WGCNA. Subsequently, the module eigenvalues were used to identify the modules associated with the target cell clusters. For the genes in the target module, eggNOG-mapper (version 2.1.4) (eggnog-mapper.embl.de) was used for Gene Ontology (GO) enrichment analysis to analyze the biological function of the module (Cantalapiedra et al. 2021). Furthermore, the predicted protein sequences were aligned with AnimalTFDB (version 3.0) (http://bioinfo.life.hust.edu.cn/AnimalTFDB/) using hmmscan to obtain the TFs (Hu et al. 2019). Finally, Cytoscape (version 3.3.0) (https://cytoscape.org/release_notes_3_3_0.html) was used to visualize the regulatory network (Shannon et al. 2003).

Cell communication analysis

To infer intercellular communication, the CellChat R package (version 1.1.3) (https://github.com/sqjin/CellChat) was used to evaluate the mode of intercellular communication across biological time from multiple aspects. CellChat integrates ligand–receptor pair information from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database and related literature to construct a comprehensive signaling molecule interaction database (CellChatDB, http://www.cellchat.org/) that considers the known structural composition ofligand–receptor interactions, such as multimericligand–receptor complexes, soluble agonists, and antagonists, as well as stimulatory and inhibitory membrane-bound co-receptors. These ligand–receptor pairs also belong to different pathway types. CellChat analysis uses the gene expression data of cells as input and establishes the probability of cell–cell communication by integrating prior knowledge of the interactions between gene expression and signal ligands, receptors, and their cofactors, thereby predicting intercellular communication (Jin et al. 2021). The ggplot2 R package (version 3.3.6) (https://github.com/tidyverse/ggplot2/releases) was used to visualize the ligand–receptor network. In this study, we mapped ligand–receptor pairs at different developmental stages to investigate the differences in intercellular communication during development.

Gene enrichment analysis

eggNOG-mapper (version 2.1.9) (http://eggnog-mapper.embl.de) was used for the GO and KEGG enrichment analysis (Cantalapiedra et al. 2021). Using the KEGG database, we applied gene set variation analysis (GSVA) (version 3.17) (https://bioconductor.org/packages/GSVA) to score the established pathways between different samples of Clusters 0 and 10 (Hänzelmann et al. 2013).

In situ hybridization analysis and immunohistochemical staining

Musk gland tissues were fixed with an in situ hybridization fixing solution, embedded in paraffin, and sectioned. The slides were deparaffinized in xylene solution and incubated with protease K (20 μg mL⁻¹). Then, 3% methanol–H₂O₂ was used to block endogenous peroxidases, and probes for one of the target mRNAs were hybridized to the slides. After washing off the hybridization solution with Saline Sodium Citrate (SSC), the hybridization solution that contained two-labeled probes, anti-DIG-HRP, and FITC-TSA reagent were sequentially added. Then, probes for another target mRNAs were hybridized to the slides. After washing off the hybridization solution, the hybridization solution that contained three-labeled probes, anti-DIG-HRP, and CY3-TSA reagent were added. Finally, DAPI was added, and the slides were mounted with an anti-fade mounting medium. The probes are listed in Table S1, Supporting Information.

Immunohistochemical analyses were performed on paraformaldehyde-fixed, paraffin-embedded musk gland tissue sections. Briefly, slides were deparaffinized, hydrated, and antigen-repaired and then incubated in a 3% hydrogen peroxide (H₂O₂) solution to block endogenous peroxidase activity. The slides were blocked with 3% BSA and incubated with the primary antibody overnight at 4°C. The slides were then incubated with a secondary antibody for 50 min. The DAB solution was used for chromogenic detection. The slides were counterstained with hematoxylin for 3 min, differentiated with hematoxylin differentiation solution, and returned to blue with a hematoxylin blue-returning solution. Dehydration and transparency were performed by gradually introducing 75–100% ethanol (5 min each) and xylene. Finally, the slides were sealed with sealing glue. The antibodies used are listed in Table S1, Supporting Information.

Identification of cellular heterogeneity during musk gland development

Musk glands were dissected before sexual maturity (6 months) and after somatic maturity (6 years) (Fig. 1a). Histological examination showed major changes in morphology along this timeline. In the 6-month-old musk gland, the acinar lumen comprised cubic or short columnar cells, with a large lumen and sparse acinar distribution. However, in the 6-year-old musk gland, the acinar lumen was small, and the acinar cells were tightly arranged in a ring (Fig. 1b). To characterize the molecular features accompanying this developmental progression, we isolated single nuclei from musk gland tissues and performed single cell multiome ATAC + gene expression analysis using the 10 × Genomics platform (Fig. 1a). After quality filtering, we detected 19 283 genes and 89 233 peaks in the 6-month-old musk gland and 20 931 genes and 108 825 peaks in the 6-year-old musk gland. After removing low-quality nuclei, we obtained 18 701 total single-nuclei transcriptome profiles from 6-month-old and 6-year-old musk gland nuclei (2050 and 16 651 single nuclei, respectively) (Fig. 1c; Fig. S1a–d, Supporting Information). In general, the Pearson correlation coefficient between the average gene openness (snATAC-seq) and average gene expression (snRNA-seq) of the two samples was 0.46 (Fig. S1e, Supporting Information).

We performed UMAP analysis on all single nuclei (Fig. S2a,b, Supporting Information), which identified 19 cell clusters based on their gene expression profiles (Fig. 1d). To further characterize the cell clusters, we calculated the DEGs in each cluster compared to all other clusters (Table S2, Supporting Information). We then created a heatmap of known cell marker genes and top DEGs for each cluster and used a series of known markers to identify individual clusters (Fig. 1e,f). The marker genes were selected based on known annotations in the literature. Consistent with previous studies, 12 cell types were identified (Fig. 2a; Table S3, Supporting Information) (Liu et al. 2023). Clusters 0, 8, 12, and 15 were acinar cells 1 that expressed high levels of the Galnt7, Ca12, and Kcnma1 markers (Li et al. 2016). Cluster 10 consisted of myoepithelial cells with high expressions of Cd44, Krt14, and Tp63 (Indumathi et al. 2013; Wood et al. 2016). Cluster 5 contained sebocytes with high expressions of Krt7 and Krt19 (Zouboulis et al. 1991). Cluster 3 consisted of lipofibroblasts with enriched expressions of lipofibroblast markers (Pparg, Fabp5, and Plin2) (Xie et al. 2018). Cluster 7 consisted of mast cells with enriched expressions of mast cell markers (Vwc2 and Hmgcs2) (Young et al. 2018). Clusters 4 and 6 contained myofibroblasts that expressed high levels of Acta2 and Myh11 (Hsia et al. 2016; Guerrero-Juarez et al. 2019). Cluster 14 contained pre-adipocytes with high expressions of Fabp4, Pparg, and Cd36 (Hu & Yang 2020). Cluster 1 comprised fibroadipogenic progenitors with enriched expressions of Pdgfra and Dcn (Giordani et al. 2019). Clusters 9 and 11 consisted of endothelial cells with enriched expressions of Vwf, Pecam1, and Adgrl4 (Park et al. 2018; He et al. 2020). Clusters 2 and 13 contained macrophages with enriched expressions of macrophage markers (Stat1 and Cd163) (Christmann et al. 2014; Ma et al. 2020). Cluster 18 comprised acinar cells 2 with a high expression of Bmp6 (Heikinheimo et al. 1999). Clusters 16 and 17 contained myoblasts that expressed high levels of Nrxn1 and Ttn (Zeng et al. 2016; Li 2020).

We performed a GO enrichment analysis of the inferred cell types. The results showed that fatty acid oxidation, lipid oxidation, as well as lipid modification pathways were primarily enriched in acinar cells 1. These results are because lipids and fatty acids are the primary components of musk, which is consistent with musk secretion function. Regulation of phosphate metabolic process and inorganic ion homeostasis pathways were enriched in acinar cells 2, which was in good agreement with the fact that inorganic salts are the primary components of musk, in line with its musk secretion function (Fig. 2b; Table S4, Supporting Information). We also examined the relative proportion of each major musk gland cell type contributing to the dataset. As expected, the acinar cell content was the highest, accounting for 40.56% of all cells (Fig. 2c). To further confirm the acinar cells, we performed fluorescence in situ hybridization on 6-month-old and 6-year-old musk gland sections. The results showed that Galnt7 and Kcnma1 were expressed in Cluster 0, whereas Bmp6 was expressed in Cluster 10 (Fig. 2d,e). In addition, in the 6-month-old and 7-year-old samples, Cluster 0 (acinar cells 1) accounted for the largest proportion at 26.93% and 41.87%, respectively (Fig. S2c, Supporting Information). Notably, all clusters comprised nuclei from different developmental periods (Fig. S2d, Supporting Information).

Single cell multiome ATAC + gene expression during Cluster 0 acinar cell development

We calculated the DEGs between the different developmental stages of Cluster 0 acinar cells to further elucidate their developmental process (Table S5, Supporting Information). During Cluster 0 development, we identified 543 upregulated and 877 downregulated genes (Fig. 3a). Moreover, we found that propanoate metabolism-related genes, such as Acaca, Pccb, Ehhadh, and Acss3, and fatty acid metabolism-related genes, such as Elovl6, Acsl1, Elovl1, and Acadl, significantly changed during development (Fig. 3b). Consistent with these results, we observed significant enrichment of propanoate and fatty acid metabolism during Cluster 0 development (Fig. 3c; Table S6, Supporting Information). To investigate the functional heterogeneity of Cluster 0 acinar cells at different developmental stages, we scored the KEGG pathways related to musk secretion in different Cluster 0 samples using GSVA. The results indicated that pathways related to musk secretion were enriched in 6-year-old acinar cells (Fig. S3 and Table S7, Supporting Information).

We used single cell multiome ATAC + gene expression to gain a deeper biological understanding of the relationships among TFs, chromatin accessibility, and upregulated genes during acinar cell development. During Cluster 0 acinar cell development, 44 proximal, 19 distal, and 7 proximal and distal candidate target genes were identified (Fig. 3d; Table S8, Supporting Information). Additionally, 12 proximal and 12 distal candidate TFs were identified (Fig. 3e,f; Table S8, Supporting Information). Associated candidate target genes and TFs, 51 proximal and six distal core target genes, as well as 12 proximal and 12 distal core TFs were identified (Fig. 3g,h; Table S8, Supporting Information). Among the core target genes, seven genes (Hsd17b2, Acacb, Lss, Vapa, Aldh16a1, Aldh7a1, and Sqle) participated in pathways related to musk secretion. Seven core target genes related to musk secretion were regulated by 12 core TFs (FOXO1, CUX2, RORA, RUNX1, KLF6, MGA, NFIC, FOXO3, ETV5, NR3C1, HSF4, and MITF) (Fig. 3i). To confirm the core target genes and TFs, we performed fluorescence in situ hybridization for Acacb, Hsd17b2, Lss, and Sqle, and immunohistochemistry for CUX2 and FOXO1 in 6-month-old and 6-year-old musk gland sections using the Cluster 0 acinar cell marker gene Kcnma1. The results indicated that the core target genes related to musk secretion were upregulated during Cluster 0 acinar cell development (Fig. 4a–d). Furthermore, the core TFs, CUX2 and FOXO1, which regulate core target genes related to musk secretion, were also upregulated during Cluster 0 acinar cell development (Fig. 4e,f).

Single cell multiome ATAC + gene expression during Cluster 10 acinar cell development

We examined the DEGs between different developmental stages of Cluster 10 acinar cells (Table S5, Supporting Information). During Cluster 10 development, we identified 135 upregulated and 87 downregulated genes (Fig. S4a, Supporting Information). The expression of genes related to propanoate metabolism, such as Pcca, Acacb, Acss3, and Suclg2, significantly changed during development (Fig. S4b, Supporting Information). Consistent with the above findings, propanoate metabolism was significantly enriched during Cluster 10 development (Fig. S4c and Table S6, Supporting Information). We also scored the KEGG pathways related to musk secretion between the different samples in Cluster 10 using GSVA. Notably, pathways related to musk secretion were enriched in 6-year-old acinar cells (Fig. S3 and Table S7, Supporting Information).

During the development of Cluster 10 acinar cells, two proximal and one distal candidate target genes, as well as three proximal and three distal candidate TFs were identified (Fig. S4d–f and Table S9, Supporting Information). Associated candidate target genes and TFs, a total of one proximal core target gene (Cdk8) and one proximal core TF (TFAP2A) were found (Fig. S4g,h and Table S9, Supporting Information).

Construction of a regulatory network during acinar cell development using WGCNA

WGCNA was used to categorize genes with similar expression patterns. Genes were divided into 13 modules, and the interactions between the modules were calculated (Fig. 5a). Gene–module correlations indicated that the gene expression between modules was relatively independent (Fig. S5a, Supporting Information). We used module eigenvalues to plot a heatmap of cell cluster expression patterns to identify the module that was significantly related to Cluster 0 acinar cells. The results indicated that the lightgreen module was significantly related to Cluster 0 (Fig. 5b; Table S10, Supporting Information). Notably, the lightgreen module was primarily enriched in the GO terms of “fatty acid metabolic process,” “cellular lipid metabolic process,” and “regulation of steroid metabolic process” (Fig. S5b, Supporting Information). Therefore, we focused on the lightgreen module to construct a regulatory network during Cluster 0 acinar cell development.

In the lightgreen module, 560 genes and 37 TFs were identified (Fig. S5c,d and Table S10, Supporting Information). During Cluster 0 acinar cell development, seven core target genes related to musk secretion were regulated by 12 core TFs. We screened four target genes (Hsd17b2, Acacb, Vapa, and Aldh7a1) and three target TFs (CUX2, HSF4, and ETV5) using Venn diagrams (Fig. 5c,d). The target genes and TFs were used to construct a gene co-expression network. Genes associated with musk secretion were further screened based on the pathways associated with musk secretion mapped by the DEGs during Cluster 0 acinar cell development. The regulatory network diagram is shown in Fig. 5e. Acsl1, Aldh7a1, and Acadl interacted via the “fatty acid degradation” pathway, while Akr1c1, Hsd17b2, and Srd5a1 interacted in the “steroid hormone biosynthesis” pathway. Finally, Gpat4, Aldh7a1, and Lpin2 interacted in the “glycerolipid metabolism” pathway.

Ligand–receptor interaction prediction during acinar cell development

A public ligand–receptor database was used to infer the intercellular communication during musk gland development. We revealed the number and strength of interactions between cell types in the 6-month-old and 6-year-old musk glands (Fig. 6a,b; Fig. S6a,b and Table S11, Supporting Information). To identify the significantly changing interactions between cell types, we compared the number and strength of interactions between cell types. As expected, the number of interactions decreased and the strength of interactions weakened between Clusters 0 and 10 during musk gland development (Fig. 6c,d; Table S12, Supporting Information). Based on functional similarities, we identified four groups of signaling pathways (Fig. 6e). Notably, Cluster 1 primarily included pathways related to cell growth, development, and organ formation, such as PTPRM, PECAM1, and SEMA6 (Craig & Brady-Kalnay 2015; Villar et al. 2019; St Clair et al. 2019) (Fig. 6f). We compared the information flow of each signaling pathway between 6-month-old and 6-year-old samples. Certain developmental-related pathways, including WNT, BMP, TGFβ, and ANGPT (Liu & Xie 2012; Mulligan & Cheyette 2012; Morikawa et al. 2016; Magro-Lopez & Muñoz-Fernández 2021), were enriched in the 6-month-old sample (Fig. 6g; Fig. S6c, Supporting Information). We also compared the probability of communication mediated by the ligand–receptor pairs of Clusters 0 and 10 in the two samples. The results revealed 15 potential interactions between the ligands and receptors (Fig. 6h). NAMPT interacted with INSR in Clusters 0 and 10 and participated in the positive regulation of cell proliferation (Samal et al. 1994; Jensen et al. 2007). The ligand HBEGF interacted with its multi-subunit receptor, EGFR/ERBB2, and participated in the positive regulation of cell proliferation (Miyazaki et al. 1995; Gaudet et al. 2011). FGF2 interacted with FGFR1 and positively regulated cell proliferation (Mansukhani et al. 1990; Ornitz et al. 1996). The ligand BMP6 interacted with its multi-subunit receptors, BMPR1A/BMPR2 and BMPR1A/ACVR2A. BMP6 is a bone morphogenetic protein 6, which participates in aldosterone biosynthesis and secretion (Inagaki et al. 2006). APP, which is an amyloid-β precursor protein that plays a role in the cholesterol metabolic process (Liu et al. 2007), interacted with CD74 in Clusters 0 and 10.

Expression patterns of genes related to fatty acid degradation and steroid hormone biosynthesis pathways during acinar cell development

snRNA-seq and snATAC-seq association analysis was used to explore changes in transcriptional activity and identify key TFs during acinar cell development. During Cluster 0 acinar cell development, DEGs were enriched in 18 KEGG pathways associated with musk secretion, including propanoate metabolism, fatty acid metabolism, fatty acid degradation, and steroid hormone biosynthesis pathways. WGCNA further showed that fatty acid degradation, steroid hormone biosynthesis, and glycerolipid metabolism pathways played crucial roles in the development of Cluster 0 acinar cells. In the present study, we demonstrated the expression patterns of genes related to fatty acid degradation and steroid hormone biosynthesis during Cluster 0 acinar cell development (Fig. 7). Ten DEGs were detected in the fatty acid degradation pathway, including Aldh7a1, Aldh16a1, and Acsl1. Among these genes, we found that Aldh7a1 was regulated by the core TFs FOXO1, RORA, NR3C1, and KLF6, and Aldh16a1 was regulated by the core TFs FOXO3, RORA, ETV5, and RUNX1. Notably, this pathway was associated with several pathways related to musk secretion, such as fatty acid biosynthesis, glycerolipid metabolism, and synthesis and degradation of ketone bodies pathways (Fig. 7a). Regarding the steroid hormone biosynthesis pathway, Hsd17b2 (17-β-hydroxysteroid dehydrogenase type 2) underwent significant changes, was regulated by the core TFs FOXO1, CUX2, RORA, RUNX1, KLF6, and MGA, and directly affected testosterone content (Fig. 7b).