Multiomic data integration unveils molecular mechanisms underlying chestnut and skin tissue differentiation in Dezhou donkeys

Abstract

Background

Donkey chestnuts are hairless, darkly pigmented keratinized structures located on the medial forelimbs, anatomically distinct from surrounding skin. Despite their unique morphology, the molecular mechanisms underlying their formation and function remain poorly understood.

Methods

An integrated multi-omics approach was applied to chestnut and skin tissues collected from four Dezhou donkeys. Bulk RNA sequencing (RNA-seq) was performed on six samples to assess transcriptomic differences, followed by single-cell RNA sequencing (scRNA-seq) to resolve cellular composition, and proteomics to characterize protein expression patterns.

Results

Chestnut tissue displayed exceptionally high keratin expression, including KRT1, KRT5, KRT10, and KRT14 (TPM > 13,000), and was enriched in endothelial cells, pericytes, T cells, and mast cells. Proteomic analysis revealed significant upregulation of collagen I (COL1A1/COL1A2), indicative of active extracellular matrix remodeling. Pathway analysis identified enrichment in peroxisome activity, fatty acid metabolism, PPAR signaling, and PI3K-Akt networks.

Conclusion

These findings collectively demonstrate that donkey chestnut is a metabolically active, specialized barrier tissue with a distinctive cellular and molecular profile. Its unique signatures related to keratinization, pigmentation, and immune cell enrichment suggest potential roles in barrier defense and anti-inflammatory regulation, providing a foundation for further investigation into the biological significance of this understudied structure.

Introduction

The Dezhou donkey (Equus asinus), a prestigious Chinese breed renowned for its superior meat quality and distinctive coat coloration, represents a valuable genetic resource with significant economic and cultural importance in livestock production [1, 2]. Characterized by their robust constitution, exceptional adaptability, and unique phenotypic traits, Dezhou donkeys have garnered increasing attention in both agricultural practices and scientific research [3]. Among the most notable characteristics observed is the distinctive chestnut coloration pattern, which exhibits regional variation across different anatomical sites, particularly when examining the morphological differences between specialized tissues such as chestnuts and adjacent skin [4]. Previously our research team reported the diverse morphological configurations of forelimb chestnuts. Comprehensive morphological and histological analyses also confirmed prominent structural differences between chestnut and adjacent skin tissues [4]. These observed phenotypic differences suggest that elucidating the underlying molecular mechanisms responsible for such tissue-specific variations would provide valuable insights into the developmental and functional characteristics of these specialized structures.

Multi-omics approach has revolutionized our understanding of complex biological systems by integrating data from multiple molecular levels, including genomics, transcriptomics, proteomics, and metabolomics [5, 6]. This comprehensive analytical framework enables researchers to decipher the intricate molecular mechanisms underlying phenotypic variations and tissue-specific characteristics [7]. In livestock species, multi-omics studies have provided unprecedented insights into production traits, disease resistance, and adaptive mechanisms [8, 9].

Tissue-specific gene expression and molecular profiles play crucial roles in determining morphological and functional characteristics in mammals [10]. The chestnut, a specialized keratinized structure found on the inner legs of equids, represents a unique evolutionary vestige with distinct molecular signatures compared to conventional skin tissue [11]. Understanding the molecular differences between these tissues is essential for elucidating the genetic basis of coat color determination and tissue differentiation in donkeys [12].

Previous studies in equine species have identified key genes and pathways involved in coat color determination, including melanocortin 1 receptor (MC1R), agouti signaling protein (ASIP), and tyrosinase-related proteins [13, 14]. However, comprehensive multi-omics comparisons between specialized tissues like chestnuts and skin in donkeys remain limited. The integration of transcriptomic, proteomic, and metabolomic data can provide a holistic view of the molecular landscape governing tissue-specific characteristics and color phenotypes [15].

Advances in high-throughput sequencing technologies and mass spectrometry have made large-scale multi-omics studies increasingly feasible and cost-effective [16]. RNA-sequencing enables comprehensive transcriptome profiling, while proteomics and metabolomics provide insights into functional molecular products and biochemical pathways [17]. The integration of these complementary approaches allows for the identification of key regulatory networks and biomarkers associated with specific phenotypic traits [18].

The aim of this study was to perform a preliminary multi-omics comparative analysis of chestnut tissue and skin tissue in Dezhou donkeys, to explore the molecular mechanisms underlying their tissue-specific characteristics, and to identify key genes, proteins and cells associated with the two distinct tissue types. Through integrated transcriptomic and proteomic analyses, we aimed to provide novel insights into the genetic basis of tissue differentiation and contribute to the broader understanding of equine biology and breeding applications.

Materials and methods

Ethical statement

The study protocol was approved by the institutional Special Committee of Scientific Research Ethic of Liaocheng University (AP2025031726).

Experimental animals and management

Two provincial breeding farms for Dezhou donkeys, both authorized by Shandong Province and located in northwest Shandong, were selected for this study. Both facilities employed similar management protocols, with animals fed twice daily and provided ad libitum access to water as described in previous study [19].

Sample collection and processing

The donkeys involved in this study were processed at a licensed local commercial abattoir for non-research purposes, in accordance with standard commercial livestock slaughter practices. Prior to slaughter, all animals were subjected to a 12-hour fasting period to reduce physiological stress and ensure animal welfare. Each animal was rendered fully and verifiably unconscious by electrical stunning at 220 V applied continuously for 20 s; complete loss of consciousness was confirmed before any subsequent procedure was initiated. The animals were then humanely slaughtered by exsanguination in strict compliance with standard humane slaughter protocols established for livestock. Immediately following humane slaughter at the licensed commercial abattoir, post-mortem tissue samples were collected from the left forelimbs of four female Dezhou donkeys of similar age and body weight. Using a sterile biopsy punch as a standardized post-mortem tissue collection tool, approximately 1 cm³ of tissue was excised from the chestnut region and from the normal skin located 5 cm superior to the chestnut, with dimensions confirmed using digital calipers. A total of seven samples were obtained and designated as follows: chestnut samples (A1–A4) and skin samples (B1–B3). For bulk transcriptome and proteome sequencing, samples A1–A3 and B1–B3 were used, following the principle of three biological replicates per group. For single-cell transcriptome sequencing, sample A2 was initially planned for library construction; however, it was excluded due to unsatisfactory quality control resulting from cell dissociation failure. Sample A4, collected from a fourth Dezhou donkey of identical age and management conditions, was obtained as a necessary replacement to ensure data integrity. Accordingly, single-cell sequencing was performed on samples A1, A4, B1, and B2. The skin samples were collected from exactly the same anatomical sites as those for histological analysis, ensuring consistency and comparability between histological characterization and subsequent transcriptomic and proteomic analyses. Following collection, all tissue samples were immediately rinsed twice with phosphate-buffered saline (PBS) and sectioned into appropriate sizes for downstream analyses. For transcriptome and proteome sequencing, samples A1–A3 and B1–B3 were placed in sterile, RNase-free cryotubes, immediately frozen on dry ice, and transported to Glbizzia Biosciences Co., Ltd. (Beijing, China). For single-cell transcriptome sequencing, samples A1, A4, B1, and B2 were preserved in cryotubes containing tissue preservation solution and transported to the same facility under 4 °C conditions.

RNA sequencing and bioinformatics analysis

Total RNA was extracted from chestnut tissue samples (chestnut tissue group, n = 3) and skin tissue samples (normal skin tissue group, n = 3) using the RNeasy Kit (Qiagen N. V, Germany) according to the manufacturer’s protocol. In this study, we collected chestnut tissue and adjacent normal skin tissue from Dezhou donkeys for multi-omics and phenotypic comparison. It should be noted that the two types of samples are anatomically and functionally distinct specialized tissues. This work is designed for inter-tissue comparative analysis, rather than a typical intervention experiment with treatment and blank control groups.

RNA integrity and genomic DNA contamination were assessed by 1% agarose gel electrophoresis. RNA purity and concentration were determined using a NanoPhotometer spectrophotometer, while RNA integrity was evaluated using the Bioanalyzer 2100 system (Agilent Technologies Inc.,USA). Sequencing libraries were prepared from high-quality total RNA using the NEBNext^® Ultra™ RNA Library Prep Kit for Illumina^® (New England Biolabs, USA) following the manufacturer’s instructions. Library quality was validated, and libraries were pooled based on their effective concentrations and target sequencing depth requirements. Sequencing was performed on the Illumina platform to generate 150-bp paired-end reads. Raw sequencing images were converted to sequence data (reads) using CASAVA base calling software (version 1.6, Illumina Inc., San Diego, USA) with Illumina sequencing identifiers.

Raw data underwent quality filtering according to the following criteria: reads containing more than 15 ambiguous bases (N) were removed; reads in which more than 50% of bases had a Phred quality score ≤ 5 were removed; reads shorter than 2 bases were removed; and adapter sequences were automatically detected and removed. Clean reads were aligned to the donkey reference genome (Equus asinus, Ensembl release 110, assembly ASM1607732v2). Raw data underwent quality filtering and sequencing error assessment to obtain clean reads suitable for downstream analysis. Clean reads were aligned to the reference genome using HISAT2 (version 2.2.1) software [20]. Variant site analysis was performed using GATK (version 4.2.2.0) [21], with each variant site statistically characterized based on SnpEff annotation information, focusing on functional, regional, and impact statistics. StringTie software (version 2.1.6) [22] was used for novel gene assembly and structural annotation, followed by functional annotation using the NR database. Gene expression quantification was performed using StringTie to count reads mapping to each gene, including newly predicted genes [23]. Differential expression Genes (DEGs) analysis was conducted using DESeq2 software (version1.34.0) [24] on raw read counts generated from STAR alignment (quantMode GeneCounts), following the standard DESeq2 workflow. Prior to testing, genes with zero expression across all samples were excluded. DESeq2 applies independent filtering by default to remove genes with little chance of showing significant differences, and potential outlier counts are handled using Cook’s distance. P-values were adjusted using the Benjamini–Hochberg method. DEGs were identified using the criteria: |log₂(fold change) | > 1 and adjusted p-value < 0.05. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using clusterProfiler (version 4.2.2) software [25].

Single-cell RNA sequencing (scRNA-seq) and bioinformatics analysis

Tissue samples (~ 0.5 cm³; A1, A4, B1, B2) were removed from MACS^® Tissue Storage Solution (130-100-008; Miltenyi Biotec B.V. & Co. KG, Germany) and washed three times in Dulbecco’s phosphate-buffered saline (DPBS, 14190144; Thermo Fisher Scientific, USA). Samples were minced into ~ 1 mm³ pieces using sterile scissors and transferred to 15 mL centrifuge tubes (Corning Incorporated, USA) containing 1 mg/mL collagenase D (11088866001, Merck, Germany) in Hanks’ balanced salt solution (HBSS, 14185052; Gibco, USA), and incubated at 37 °C for 30 min. Digestion was terminated by adding 10% fetal bovine serum (FBS, Grand Island Biological Company, USA), and the cell suspension was filtered through a 40 μm cell strainer (Falcon^®, Corning, USA). Cells were pelleted by centrifugation at 500 rpm for 5 min at 4 °C, and the supernatant was discarded. The cell pellet was resuspended in 1× DPBS containing 0.04% bovine serum albumin (BSA, Gibco, USA). Cell viability was assessed using acridine orange/propidium iodide (AO/PI, ApexBio, USA) staining and quantified using the Luna FL cell counter (Aligned Genetics, South Korea).

Single-cell libraries were prepared using the Chromium Single Cell 3’ Reagent Kit v3 (10× Genomics) according to the manufacturer’s protocol. Complementary DNA (cDNA) quality was evaluated using fragment analysis (Qsep100; Bioptic Inc, Jiang Su, China). Following library construction, library conversion was performed using the MGIEasy Universal DNA Library Preparation Kit (Shenzhen BGI Intelligent Manufacturing Technology Co., Ltd, Shenzhen, China) for platform compatibility, followed by sequencing on the DNBSEQ-T7 platform (MGI Tech Co., Ltd, ShenZhen, China).

Raw sequencing data were processed using Cell Ranger software 7.2.0 (10× Genomics) with scStar alignment, mapping reads to the reference genome (Equus asinus, Ensembl release 110, assembly ASM1607732v2) and refining annotations using GTF files. Genomic regions were classified as exonic, intronic, or intergenic. Gene annotation was obtained from Ensembl release 110. Quality control metrics, including the number of high-quality cells, median genes per cell, and sequencing saturation, were assessed using Cell Ranger software [26]. The resulting gene count matrix was imported into Seurat v4.0.4 [27] for downstream analysis. Following initial filtering using Cellranger/DNBC4 tools software (version 2.1.2), low-quality cells were further removed using Seurat and DoubletFinder packages. Exclusion criteria included: (1) multiplets identified by DoubletFinder; (2) cells expressing fewer than 200 genes or more than the 95th percentile of gene counts (to exclude potential doublets or cell aggregates); (3) cells with mitochondrial gene expression exceeding 10% of total counts, which are indicative of stressed, damaged, or dying cells; (4) cells with excessive expression of erythrocyte-related genes (in animal samples only), which were minimized by ambient RNA correction using SoupX.

Highly variable genes (HVGs) were identified using the FindVariableFeatures function in Seurat with variance stabilizing transformation (VST) method, selecting the top 2,000 features. The principal component analysis (PCA) was performed for linear dimensionality reduction, and the most significant principal components were evaluated based on their contribution to data variance. By default, the first 50 principal components were selected for downstream analysis. Graph-based clustering was performed using a K-nearest neighbors (KNN) graph constructed based on Euclidean distances in PCA-reduced space. The Louvain algorithm was applied for community detection and cell partitioning into distinct clusters (resolution = 0.4). Cellular distribution was visualized in two-dimensional space using Uniform Manifold Approximation and Projection (UMAP, dims = 1:30, min.dist = 0.1) [28] and t-Distributed Stochastic Neighbor Embedding (t-SNE, dims = 1:30) [29].

The DEGs were identified using the FindAllMarkers function in Seurat (Wilcoxon rank-sum test, p ≤ 0.01; log₂ fold change ≥ 0.25; expressed in ≥ 25% of cells). Cell type annotation was performed through a multi-step approach that integrated multiple complementary methodologies to ensure accurate cellular identification. Initially, literature-based marker genes from relevant single-cell studies were systematically collected, followed by cross-species homologous alignment to identify conserved markers that could reliably distinguish different cell populations. Subsequently, cell types were annotated based on classical anatomical and morphological studies of the target species, leveraging established knowledge of tissue organization and cellular characteristics. Finally, functional annotation was conducted through GO and KEGG pathway enrichment analyses to determine the biological functions and regulatory pathways associated with DEGs, providing additional validation for the assigned cell type identities and ensuring that the molecular signatures aligned with expected cellular functions.

For functional enrichment analysis, DEGs were identified using the FindMarkers function in Seurat with |log₂FC| ≥ 0.25 threshold. Up-regulated genes were defined as having adjusted p-value < 0.05 and log₂FC ≥ 0.5, while down-regulated genes were defined as having adjusted p-value < 0.05 and log₂FC ≤ -0.5. DEG intersections were visualized using UpSetR. Functional enrichment analysis, including GO biological processes and KEGG pathway enrichment, was performed using clusterProfiler [25].

Proteomics Analysis

Tissue samples (A1, A2, A3, B1, B2, B3) were retrieved from − 80 °C storage and homogenized in SDT lysis buffer (4% SDS, Sinopharm Group Co., Ltd, Beijing, China; 100 mM Tris-HCl pH 7.6, Beijing Solarbio Science & Technology Co., Ltd, Beijing, China; 10 mM DTT, Beijing Solarbio Science & Technology Co., Ltd; 1 mM PMSF, Shandong Xiya Chemical Industry Co., Ltd, Shandong, China; 2 mM EDTA, Sinopharm Group Co., Ltd, Beijing, China) for protein extraction. Total protein concentration was determined using the BCA Protein Assay Kit (Beyotime Institute of Biotechnology, Shanghai, China). Proteins were digested with trypsin (Shanghai Promega Biological Products Ltd, Shanghai, China) and desalted using standard protocols. Resulting peptides were acidified to pH 2–3 with 20% trifluoroacetic acid (TFA, Sinopharm Group Co., Ltd, Beijing, China) and purified using C18 columns (Millipore, Billerica, MA, USA). Peptide concentration was quantified using the Pierce™ Quantitative Peptide Assay Kit (Thermo Fisher Scientific, USA).

Peptide samples were separated using a high-performance liquid chromatography (HPLC, Thermo Fisher Scientific, USA) nanoflow system. Mobile phase A consisted of 0.1% formic acid in water (Sinopharm Group Co., Ltd, Beijing, China), while mobile phase B contained 0.1% formic acid in acetonitrile (Sinopharm Group Co., Ltd, Beijing, China). A dual-column trap-analytical setup was employed, utilizing a trap column (PepMap Neo Trap Cartridge, 300 μm × 5 mm, 5 μm, Thermo Fisher Scientific, USA) and an analytical column (Easy-Spray™ PepMap™ Neo UHPLC column, 50 μm × 15 cm, 2 μm, Thermo Fisher Scientific, USA). The flow rate was maintained at 2.5 µL/min. Following chromatographic separation, samples were analyzed using an Orbitrap Astral mass spectrometer (Thermo Fisher Scientific, USA).

MS raw data were analyzed using DIA-NN (version 1.8.1) [30] with a library-free approach using the MaxLFQ algorithm for protein quantification. The UniProtKB proteome database UP000694387 (containing 48,973 sequences) was used to create a spectral library using deep learning neural network algorithms. Match Between Runs (MBR) functionality was employed to create a spectral library from DIA data, followed by reanalysis using this library. False discovery rate (FDR) was set to < 1% at both protein and precursor ion levels. Differentially expressed proteins (DEPs) were identified using fold change (FC) ≥ 1.5 or FC ≤ 0.6667 and p-value ≤ 0.05. The use of FC ≥ 1.5 as the threshold for proteomics (versus |log₂FC| > 1 for RNA-seq) reflects the inherent differences between the two technologies in quantitative accuracy and noise levels; enforcing identical thresholds would either inflate false positives or lose true biological differences in one dataset.

Result

RNA sequencing results

Sequencing data filtration Analysis

The raw sequencing data obtained from the sequencing process are archived in the FASTQ file format. A statistical analysis was conducted on the RNA-seq data of the six samples discussed in this paper, with the results presented in Table 1. The statistics depicted in the table reveal that the sequencing data volume for each sample exceeds 8.485 gigabytes, and the Q20 values are above 97%, indicating a relatively high sequencing quality suitable for subsequent bio-informatics analysis.

Table 1.

Post-filtration read quality assessment

Samples	Total raw reads	Total clean reads	Total clean bases (Gb)	Clean reads Q20(%)	Clean reads Q30(%)	GC content%)
Chestnut A1	28,320,483	28,320,458	8.46	97.89	94.17	50.41
Chestnut A2	33,312,198	33,312,163	9.95	97.90	94.18	50.17
Chestnut A3	26,865,583	26,865,554	8.03	97.86	94.08	50.13
Skin B1	28,772,418	28,772,394	8.59	98.01	94.43	49.05
Skin B2	27,386,801	27,386,775	8.19	97.60	93.28	49.63
Skin B3	225,743,217	25,743,195	7.69	97.76	93.75	49.45

Sequencing alignment to reference genome

After obtaining clean reads, the chestnut and donkey skin data were processed, with each treatment involving three biological replicates. The alignment results are summarized in Table 2. The table indicates that the average alignment rate for each sample reached 93.03%, with minimal variation, suggesting the data among samples are comparable.

Table 2.

Reference genome alignment statistics

Sample	Total Clean Reads	Total Mapping Ratio	Uniquely Mapping Ratio
Chestnut A1	56,640,916	95.85	92.52
Chestnut A2	66,624,326	95.90	92.93
Chestnut A3	53,731,108	95.16	92.29
Skin B1	57,544,788	95.00	92.49
Skin B2	54,773,550	95.36	93.03
Skin B3	51,486,390	94.92	92.30

Quantitative analysis

For each sample, gene expression levels were quantitatively analyzed using Transcripts Per Million (TPM) normalization, and an integrated expression matrix encompassing all samples was subsequently generated. As shown in Table 3, Keratin exhibited significantly high expression specifically in the chestnut tissue, reaching a TPM value of 13,612.54, highlighting its potential functional relevance in this specialized skin region. The top genes listed in Table 3 were selected based on their expression levels in chestnut tissue, ranked from highest to lowest TPM.

Table 3.

Top expressed genes in chestnut tissue ranked by TPM

Annotation protein	Chestnut A1	Chestnut A2	Chestnut A3	Skin B1	Skin B2	Skin B3
Keratin, type II cytoskeletal 2 epidermal	13612.54	5560.86	11425.12	1131.27	397.36	2131.031
Intermediate filament protein	11909.76	6483.16	9830.06	1590.41	832.85	2555.00
Belongs to the intermediate filament family	5955.28	2709.32	4790.15	1277.78	497.641	2029.39
Elongation factor	3296.73	2855.14	3308.61	2521.65	1710.94	2117.32
small proline-rich protein	2827.76	1799.77	1857.74	592.22	382.41	733.43
loricrin	2528.16	2804.68	1515.54	568.86	309.73	1325.71
Keratin, type II head	2501.15	1581.32	2319.70	63.05	43.96	108.47
Keratin, type I cytoskeletal 14	1908.43	1584.34	1529.92	712.92	477.09	1268.02
Keratin, type II cytoskeletal 5	1750.50	1688.92	1991.44	751.18	648.30	1199.29
Filaggrin	953.82	1036.80	403.72	241.02	135.07	525.73
S-100/ICaBP type calcium binding domain	820.21	674.23	665.15	0.015	0.00489	0.0055
Part of the insoluble cornified cell envelope (CE) of stratified squamous epithelia	580.76	364.45	222.88	2.52	2.97	0.56

The boxplot presented in Fig. 1 illustrates the overall gene expression profiles of chestnut and regular skin tissues. The median lines within each group are highly consistent, indicating strong intra-group similarity. These plots were used for quality assessment of data distribution and outlier detection; no statistical testing between groups was performed at this stage, as group-level differential expression testing was conducted using DESeq2 on raw counts.

Fig. 1.

Box plots showing the distribution of gene expression levels (normalized by TPM) in chestnut and skin tissues. Consistent median lines within each group indicate high intra-group reproducibility, while distinct median positions between chestnut and skin groups reflect inter-group transcriptional differences

To further assess both intra- and inter-group variation, Pearson correlation coefficients based on genome-wide TPM values were calculated and visualized as a heatmap (Fig. 2). This heatmap clearly delineates the degree of similarity among biological replicates and reveals the extent of divergence between groups. Notably, the squared correlation coefficients (R²) within the chestnut and skin groups were consistently above 0.850, demonstrating excellent reproducibility across biological replicates. Moreover, the R² values among chestnut samples were slightly higher than those observed within the skin group, suggesting greater homogeneity within the former. Inter-group correlations ranged from 0.689 to 0.822, indicating a substantial degree of similarity but also the existence of detectable transcriptional divergence between the two tissue types.

Fig. 2.

Heat-map of Pearson correlation coefficients based on genome-wide gene expression profiles (TPM values). The color intensity represents the correlation coefficient (ranging from 0 to 1), with darker blue indicating stronger positive correlation. Samples are grouped by tissue type: chestnut tissues (A1, A2, A3) and skin tissues (B1, B2, B3)

The PCA analysis (Fig. 3) was performed on the TPM data to evaluate global transcriptomic variability. The PCA analysis plot revealed that the chestnut samples clustered closely together, confirming their high reproducibility and suitability for downstream analyses. In contrast, while two of the three skin samples showed good consistency, one sample deviated notably from the others, potentially due to technical variability during sample collection or processing. Despite this, the remaining replicates are deemed sufficient for further interpretation. Importantly, the clear separation between the two groups along the principal components indicates the presence of significant inter-group transcriptomic differences.

Fig. 3.

Principal Component Analysis (PCA) of transcriptomic data from chestnut and skin tissues. Each point represents an individual sample: chestnut tissues (A1, A2, A3, orange dots) and skin tissues (B1, B2, B3, green dots). Chestnut samples cluster closely together, confirming high intra-group consistency

Differential expression gene (DEG) analysis of chestnut and skin

Employing the stringent criteria of |log2(Fold Change) | > 1 and adjusted p-value (padj) < 0.05, DEGs analysis was conducted using the DESeq2 package. The analysis revealed a total of 5,295 DEGs between the chestnut and donkey skin, with 1,884 exhibiting up-regulation and 3,411 showing down-regulation (Fig. 4A). The final result identified a total of 15,823 expressed genes. Among them, 13,114 genes were expressed in both chestnuts and skin, 326 genes were preferentially expressed in the chestnut, and 2,383 genes were preferentially expressed in the skin (Fig. 4B). It should be noted that ‘preferentially expressed’ refers to genes meeting our differential expression criteria with TPM < 1 in the other tissue type; we cannot exclude low-level expression below the detection threshold of our RNA-seq analysis, and future qRT-PCR validation of tissue-specific markers is warranted.

Fig. 4.

Volcano plot (A) and Venn diagram of DEGs (B). The volcano map shows the trend of DEGs between the chestnut and skin of the Dezhou donkey. While Venn diagrams show the differences and commonalities of DEGs between chestnut and skin

GO enrichment analysis of DEGs

Through the statistical analysis of GO terms, the enrichment of DEGs between donkey skin and chestnut was annotated. Based on cellular components, molecular functions, and biological processes, these DEGs were further annotated and classified. Utilizing a False Discovery Rate (FDR) threshold for the p-value ≤ 0.05, the most significantly associated GO terms among the up-regulated DEGs were identified (Fig. 5). In the cellular component ontology, significant enrichment was observed in 13 GO terms, including cornified envelope, peroxisome, microbody, microbody lumen, and peroxisomal matrix. In the molecular function ontology, significant enrichment was noted in sterol esterase activity, triglyceride lipase activity, CoA hydrolase activity, acyl-CoA hydrolase activity, and structural constituent of skin epidermis. In the biological process ontology, significant enrichment was found in cornification, obsolete coenzyme metabolic process, fatty acid catabolic process, monocarboxylic acid catabolic process, and small molecule catabolic process.

Fig. 5.

Chestnut indicates the GO enrichment analysis of up-regulated DEGs between chestnut tissue and normal skin tissue. (A) A bar-plot of up-regulated DEGs enrichment in cellular components. (B) A bar-plot of up-regulated DEGs enrichment in molecular functions. (C) A bar-plot of up-regulated DEGs enrichment in biological processes. Note: Up-regulation and down-regulation represent increased and decreased expression levels in chestnut tissue compared with normal skin tissue, respectively

KEGG enrichment analysis of DEGs

As shown in Table 4; Fig. 6, the pathway functional annotation of up-regulated genes in the DEGs of chestnut and skin reveals that these genes are enriched in metabolic pathways such as peroxidase, fatty acid metabolism, steroid biosynthesis, Propanoate metabolism, and Cholesterol synthesis, as well as in the PPAR signaling pathway. The PPAR signaling pathway plays a crucial role in lipid metabolism, inflammatory response [31], gut microbiota, and hepatocyte proliferation [32]. The enrichment of coronavirus disease (COVID-19) pathway reflects the database structure of KEGG, which assembled this pathway during the pandemic to represent broad-spectrum innate immune and inflammatory response modules. In the context of donkey chestnut and skin—as surface barrier tissues representing the host’s first line of defense—enrichment of this pathway should be interpreted as activation of general host defense mechanisms rather than a specific response to SARS-CoV-2.

Table 4.

Result of KEGG enrichment of up-regulated DEGs

Gene Pathway	GeneRatio	Pvalue
Peroxisome	22/491	4.02747579253971e-08
Fatty acid metabolism	14/491	3.37857847993207e-06
Steroid biosynthesis	10/491	3.78580494172787e-06
Fatty acid degradation	13/491	1.49176102345496e-05
PPAR signaling pathway	15/491	0.000163851799270458
Propanoate metabolism	8/491	0.000204972891863965
Carbon metabolism	17/491	0.000568897329102854
Glycolysis	12/491	0.00118722628426196
Cholinergic synapse	16/419	0.00415065780117405
Biosynthesis of amino acids	11/419	0.00377405464193166
Pyruvate metabolism	9/419	0.00335428317967734
p53 signaling pathway	12/491	0.00255407689014301

Fig. 6.

The up-regulated DEGs underwent KEGG enrichment analysis. (A) A bar-plot of up-regulated DEGs in KEGG enrichment. (B) A Emap-plot of up-regulated DEGs in KEGG enrichment

Emap-plot presents a hierarchical list of KEGG pathway classifications, integrating metabolic pathways, signaling pathways, disease associations, and statistical information, with lines indicating connections between different biological pathways.

Single-cell RNA sequencing results

Data quality control and cell quantification

Single-cell RNA sequencing data were successfully processed using Seurat v4.0.4 for all experimental samples, comprising chestnut tissue specimens (A1, A4) and skin tissue specimens (B1, B2). Comprehensive quality control analysis revealed consistent and robust cell capture efficiency across all sample groups, as summarized in Table 5.

Table 5.

Single-cell RNA sequencing quality control metrics

Sample	Estimated Number of Cells	Median Genes per Cell	Median UMI Counts per Cell	Fraction Reads in Cell
A1	10,338	1,833	3,347	72.6%
B1	9,459	1,793	2,494	79.4%
A4	11,072	1,529	4,212	64.7%
B2	10,192	1,405	4,012	71.8%

The analysis yielded cell counts ranging from 9,459 to 11,072 cells per sample, demonstrating reproducible cell recovery rates. Median gene detection per cell ranged from 1,405 to 1,833 genes, indicating adequate transcriptome complexity to support reliable downstream analyses, including accurate cell clustering and cellular subpopulation identification.

Barcode rank plots (Fig. 7) provided clear delineation between cell-associated barcodes and ambient RNA background signals. The presence of distinct inflection points in all samples confirmed successful discrimination between viable cells and empty droplets, validating the quality of the single-cell capture process and ensuring reliable downstream computational analysis.

Fig. 7.

Barcode rank plots demonstrating cell-associated barcode identification across experimental samples. Barcode rank plots illustrating the discrimination between cell-containing droplets and ambient RNA background for single-cell RNA sequencing quality control. Each plot displays the relationship between barcode rank and UMI (Unique Molecular Identifier) counts, with inflection points indicating the threshold for distinguishing viable cells from empty droplets. (A) Chestnut tissue sample A1 showing clear separation between cell-associated barcodes and background noise. (B) Skin tissue sample B1 demonstrating effective cell capture and background discrimination. (C) Chestnut tissue sample A4 exhibiting robust cell identification parameters. (D) Skin tissue sample B2 displaying consistent cell-to-background ratio determination. The distinct inflection points observed across all samples confirm successful cell capture and validate the quality of single-cell isolation procedures

Cell filtration and quality assessment

Following initial data processing using Cell Ranger and subsequent ambient RNA correction with soupX, stringent quality control filters were applied using Seurat to identify high-confidence single cells. Cell filtration criteria included gene detection thresholds (minimum 200 genes per cell, maximum at 95th percentile) and mitochondrial gene expression cutoffs to exclude compromised or dying cells.

Post-filtration quality assessment demonstrated robust data integrity across all samples. The distribution of detected genes per cell was predominantly concentrated above 1,000 genes, while total unique molecular identifier (UMI) counts per cell exhibited a characteristic distribution centered within the 0–2,000 range (Fig. 8). These metrics collectively indicate exceptional single-cell transcriptome capture efficiency and minimal doublet contamination, confirming the suitability of the filtered dataset for downstream computational analyses. The stringent quality control pipeline successfully retained cells with optimal transcriptional complexity while effectively removing low-quality cells, ambient RNA contamination, and potential multiplets, thereby ensuring high data fidelity for subsequent clustering and differential expression analyses.

Fig. 8.

Violin plots depicting single-cell RNA sequencing quality control metrics following cell filtration. Violin plots illustrating the distribution of key quality control parameters across experimental samples after stringent cell filtration. Each violin plot displays the probability density distribution of cellular transcriptional metrics, with wider sections indicating higher frequency of cells at those values. nCount_RNA represents the total unique molecular identifier (UMI) count distribution, reflecting overall transcriptional activity per cell. nFeature_RNA indicates the number of detected genes per cell, serving as a measure of transcriptome complexity and cell viability. A Distribution of total UMI counts and detected gene numbers for chestnut tissue sample A1, demonstrating robust transcriptional capture. B Quality control metrics for skin tissue sample B1, showing consistent cellular transcriptional profiles. C Transcriptional parameter distributions for chestnut tissue sample A4, indicating high-quality single-cell data. D Quality assessment metrics for skin tissue sample B2, confirming effective cell filtration and data integrity. The consistent distributions across samples validate successful quality control implementation and ensure reliable downstream computational analyses

Dimensionality reduction and cluster analysis

The PCA analysis was employed for initial dimensionality reduction, followed by graph-based clustering using the Louvain algorithm to identify transcriptionally distinct cellular populations. This computational approach successfully resolved 24 discrete cell clusters across all experimental samples, indicating substantial cellular heterogeneity within the analyzed tissues.

Non-linear dimensionality reduction techniques, including Uniform Manifold Approximation and Projection (UMAP) and t-distributed Stochastic Neighbor Embedding (t-SNE), provided complementary visualization approaches that demonstrated clear inter-cluster separation and distinct transcriptional expression profiles (Fig. 9A-B). These visualizations confirmed the robustness of the clustering results and validated the presence of transcriptionally coherent cellular subpopulations.

Fig. 9.

Dimensionality reduction visualization and cellular composition analysis of single-cell RNA sequencing data. Comprehensive visualization of single-cell transcriptome clustering and tissue-specific cellular distributions following quality control and dimensionality reduction. (A) Uniform Manifold Approximation and Projection (UMAP) two-dimensional embedding of single-cell transcriptomes, with cells colored by sample origin. Each point represents an individual cell, and spatial proximity indicates transcriptional similarity. The plot demonstrates clear separation of cellular subpopulations while revealing sample-specific clustering patterns. (B) t-distributed Stochastic Neighbor Embedding (t-SNE) visualization providing complementary non-linear dimensionality reduction of the same dataset, confirming robust cluster identification and sample-specific cellular distributions. (C-D) Quantitative analysis of cellular cluster composition across experimental samples, displaying the proportional representation of each identified cluster within chestnut tissue samples (C) and skin tissue samples (D). Bar charts illustrate tissue-specific enrichment patterns and highlight clusters with preferential tissue association. (E) Expression heatmap depicting the top four most highly expressed marker genes for each identified cellular cluster. Rows represent individual genes, columns represent clusters, and color intensity indicates relative expression levels. These molecular signatures provide functional characterization of distinct cellular subpopulations and facilitate biological interpretation of cluster identities. The complementary visualization approaches collectively validate the presence of transcriptionally coherent cellular populations and demonstrate successful integration of multi-sample single-cell datasets

Sample integration was achieved through Canonical Correspondence Analysis (CCA) coupled with the Mutual Nearest Neighbors (MNN) algorithm, which successfully harmonized datasets by identifying and utilizing anchor cells to correct for potential batch effects while preserving biological variation between tissue types.

Comprehensive cluster composition analysis revealed pronounced tissue-specific cellular distributions (Fig. 9C-D). Chestnut tissue samples demonstrated enrichment in clusters 0, 1, 3, and 5, while skin tissue samples exhibited preferential representation in clusters 0, 5, and 6. Notably, cluster 23 contained almost exclusively skin-derived cells, whereas cluster 22 was predominantly comprised of chestnut tissue cells, suggesting tissue-specific cellular populations with distinct transcriptional signatures.

Differential expression analysis identified the top four highly expressed marker genes per cluster (Fig. 9E), providing molecular signatures for cellular subpopulations. Cluster 22 was characterized by elevated expression of MMRN1, STAB2, ENSASG00005022306, and ENSAG00005035409. Cluster 1 exhibited high expression of PTPRM, ENSEAG00005022461, PLCB1, and ARL15. Cluster 5 was distinguished by increased expression of ACKR1, ENSEAG00005027013, ENSEASG00005019018, and JAM2. These gene signatures represent potential chestnut tissue-specific molecular markers and provide insights into the functional characteristics of distinct cellular subpopulations.

Cell type annotation and characterization

Cell type identification was performed through comprehensive analysis of spatial distribution patterns and expression of established lineage-specific marker genes. The 24 transcriptionally distinct clusters were systematically consolidated into nine major cellular populations based on their molecular signatures and known biological functions (Fig. 10A). The annotated cell types comprised: Mast cells (cluster 16), Pericytes (clusters 8, 9, and 13), Endothelial cells (clusters 1 and 5), Keratinocytes (clusters 0, 2, 3, 4, 10, 12, 14, 18, 19, 20, and 23), Fibroblasts (cluster 7), Proliferative cells (cluster 21), Schwann cells (cluster 15), Sebocytes (cluster 11), and T lymphocytes (cluster 17).

Fig. 10.

Cell type annotation and comparative tissue composition analysis. Comprehensive visualization of cell type identification and tissue-specific cellular distributions following single-cell RNA sequencing analysis. (A) t-distributed Stochastic Neighbor Embedding (t-SNE) plot displaying the spatial organization of annotated cell types within the two-dimensional transcriptional landscape. Each point represents an individual cell, with colors indicating assigned cell type identity based on lineage-specific marker gene expression and transcriptional signatures. The visualization demonstrates clear segregation of distinct cellular populations and validates the accuracy of cell type annotation through spatial clustering of functionally related cells. (B) Sample-stratified t-SNE visualization illustrating tissue-specific cellular distributions, with cells colored by sample origin (chestnut vs. skin tissue). This comparative representation reveals tissue-specific enrichment patterns and highlights the differential abundance of cellular populations between groups within the same transcriptional space. (C) Quantitative analysis of cellular composition across experimental samples, presenting the proportional representation of each annotated cell type within chestnut and skin tissue specimens. Bar charts provide statistical assessment of tissue-specific cellular enrichment and depletion patterns, enabling identification of cell types with preferential tissue association. The integrated visualization approach confirms successful cell type annotation and demonstrates significant tissue-specific differences in cellular microenvironment composition, reflecting distinct functional specializations between chestnut and skin tissues

This cell type classification represents the major cellular constituents expected in skin and associated tissue architecture, reflecting both structural components and immune surveillance populations essential for tissue homeostasis and barrier function.

Comparative cellular composition analysis between chestnut and skin tissue groups revealed statistically significant differences in cell type distributions (Fig. 10B-C). Chestnut tissue specimens demonstrated pronounced enrichment of endothelial cells, pericytes, T lymphocytes, and mast cells, suggesting enhanced vascularization, immune surveillance, and inflammatory response capabilities. Conversely, skin tissue samples exhibited predominant enrichment of keratinocytes, reflecting the characteristic cellular architecture of epidermal tissue. The enrichment of immune cell types (mast cells, T lymphocytes) and vascular cell types (endothelial cells, pericytes) in chestnut tissue provides single-cell resolution evidence corroborating the bulk RNA-seq findings of enhanced immune and vascular signaling, and is consistent with the proteomic detection of up-regulated structural and metabolic proteins in the same samples.

The observed keratinocyte predominance in skin samples is consistent with their fundamental role as the primary epidermal cell population. Keratinocytes serve critical functions in maintaining the cutaneous barrier, orchestrating host defense mechanisms, and mediating adaptive immune responses through antigen presentation and cytokine production [26]. The differential cellular composition between tissue types provides molecular evidence for distinct functional specializations and suggests tissue-specific adaptations in cellular microenvironment and physiological demands.

Differential expression analysis and functional annotation

Comparative transcriptomic analysis between chestnut and skin tissue samples was performed using the Wilcoxon rank-sum test with stringent statistical criteria (log₂ fold change > 0.5 or < -0.5, adjusted p-value < 0.05) to identify DEGs. This analysis revealed 7,733 DEGs across all cell types, comprising 3,018 up-regulated and 4,715 down-regulated genes, indicating substantial transcriptional remodeling between tissue types.

Cell type-specific analysis demonstrated pronounced differential gene expression patterns, with the most substantial numbers of up-regulated DEGs observed in endothelial cells, fibroblasts, and keratinocytes (Fig. 11A-E). These findings suggest that these cellular populations undergo significant transcriptional reprogramming in response to tissue-specific microenvironmental demands and functional specialization requirements.

Fig. 11.

Differential gene expression analysis (DEGs) across cell types in chestnut versus skin tissue comparison. Comprehensive visualization of transcriptional differences between chestnut and skin tissues through differential expression analysis using Wilcoxon rank-sum test with stringent statistical criteria (log₂ fold change > 0.5 or < -0.5, adjusted p-value < 0.05). A Global overview of DEGs displaying the total number of up-regulated and down-regulated genes across all analyzed cell types. Bar chart quantifies the magnitude of transcriptional changes between tissue types, demonstrating the extent of cellular reprogramming. B Distribution and quantification of up-regulated DEGs across individual cell types, highlighting cellular populations with the most pronounced transcriptional activation. Bar chart illustrates cell type-specific patterns of gene up-regulation, identifying cellular populations undergoing significant functional enhancement or specialization. C Volcano plot depicting differential gene expression in fibroblasts, with x-axis representing log₂ fold change and y-axis showing -log₁₀ adjusted p-values. Each point represents an individual gene, with significantly up-regulated genes (red) and down-regulated genes (blue) highlighted based on statistical thresholds. The plot visualizes both the magnitude and statistical significance of transcriptional changes in fibroblast populations. D Volcano plot analysis for keratinocytes, illustrating the transcriptional landscape of epidermal cells with color-coded significance levels. The distribution pattern reveals genes with enhanced or diminished expression levels that may contribute to keratinocyte-specific functions in tissue differentiation. E Volcano plot representation of differential gene expression in endothelial cells, displaying vascular cell-specific transcriptional modulation. The visualization identifies genes potentially involved in angiogenesis, vascular homeostasis, and endothelial barrier function that differ between tissue types. The volcano plot format enables simultaneous assessment of both biological significance (fold change magnitude) and statistical confidence (p-value), facilitating identification of genes with robust and meaningful expression differences for subsequent functional analysis

The GO enrichment analysis revealed distinct functional profiles characteristic of each cell type’s specialized biological roles (Fig. 12). Endothelial cells exhibited significant enrichment of up-regulated DEGs associated with structural constituents of ribosomes, transcriptional corepressor activity, and translational processes, reflecting enhanced protein synthesis capacity and transcriptional regulation. Fibroblasts demonstrated enrichment in extracellular matrix structural constituents, collagen binding domains, and extracellular matrix organization pathways, consistent with their primary role in tissue architecture maintenance and remodeling. Keratinocytes showed pronounced enrichment in structural constituents of ribosomes, chemokine activity, and peptide biosynthetic processes, indicating active protein production and inflammatory signaling capabilities essential for barrier function and immune surveillance.

Fig. 12.

Gene Ontology (GO) enrichment analysis of DEGs across major cell types. Comprehensive functional annotation of up-regulated DEGs through GO enrichment analysis, visualized as bar chart across three fundamental GO categories. The analysis reveals cell type-specific functional specializations and biological processes enhanced in chestnut versus skin tissue comparison. A Fibroblast GO enrichment: Molecular Function ontology bar chart revealing molecular activities and binding capabilities enhanced in fibroblast populations. Cellular Component ontology analysis displaying subcellular compartments and structural organizations associated with up-regulated genes. Biological Process ontology visualization demonstrating enriched pathways related to extracellular matrix organization, tissue remodeling, and fibroblast-specific functions. B Keratinocyte GO enrichment: Molecular Function ontology bar chart illustrating enhanced molecular activities in keratinocyte populations, including enzymatic functions and signaling capabilities. Cellular Component ontology analysis showing subcellular localizations relevant to epidermal cell structure and function. Biological Process ontology representation displaying enriched biological pathways associated with keratinocyte differentiation, barrier function, and immune response. C Endothelial cell GO enrichment: Molecular Function (MF) ontology bar chart displaying enriched molecular activities and catalytic functions of up-regulated genes in endothelial cells. Cellular Component (CC) ontology visualization showing subcellular localizations and protein complex associations of DEGs. Biological Process (BP) ontology representation illustrating enriched biological pathways and cellular processes characteristic of endothelial cell specialization

The KEGG pathway analysis further elucidated cell type-specific signaling networks and metabolic programs (Fig. 13). KEGG pathway analysis further elucidated cell type-specific signaling networks (Fig. 13). Endothelial cells displayed enrichment in ribosomal biogenesis and lipid metabolism pathways. The enrichment of ‘coronavirus disease (COVID-19)’ pathways in endothelial cells and keratinocytes reflects the KEGG database annotation framework, where this pathway encapsulates broad innate immune and inflammatory response modules relevant to barrier tissue defense, and should not be interpreted as tissue-specific susceptibility to SARS-CoV-2. Fibroblasts exhibited enrichment in TNF signaling and IL-17 signaling networks, consistent with their role in tissue repair and immune modulation.

Fig. 13.

Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of DEGs across major cell types. Functional pathway enrichment analysis of up-regulated DEGs using the KEGG database to identify significantly enriched metabolic and signaling pathways. Each panel displays bubble plots or bar charts representing pathway enrichment results, with statistical significance indicated by color intensity and pathway importance reflected by the number of associated genes. The analysis reveals cell type-specific activation of distinct biological pathways and signaling networks that contribute to specialized cellular functions in tissue-specific contexts. A KEGG pathway enrichment in Fibroblasts. Pathway analysis revealing functional specializations of fibroblast populations, with emphasis on extracellular matrix organization, tissue remodeling, and inflammatory signaling networks. The enrichment profile demonstrates fibroblast involvement in structural maintenance, wound healing, and immune modulation through pathways such as fibrosis-related processes, tumor necrosis factor (TNF) signaling, and interleukin-17 (IL-17) signaling cascades. B KEGG pathway enrichment in Keratinocytes. Functional pathway analysis of keratinocyte-specific gene expression, highlighting pathways associated with epithelial barrier function, immune defense, and cellular metabolism. The enrichment pattern reflects keratinocyte roles in pathogen recognition, inflammatory response coordination, and maintenance of epidermal integrity through pathways including coronavirus disease responses, ribosomal processes, and IL-17 signaling networks. C KEGG pathway enrichment in Endothelial cells. Visualization of significantly enriched pathways in endothelial cell populations, highlighting metabolic processes, signaling cascades, and disease-associated pathways relevant to vascular function. The enrichment pattern reflects endothelial-specific roles in angiogenesis, vascular homeostasis, barrier function, and pathogen response mechanisms. Notable enriched pathways include ribosomal biogenesis, lipid metabolism, and coronavirus disease-related responses, indicating active protein synthesis, metabolic regulation, and immune surveillance capabilities

The systematic GO analysis across molecular function, cellular component, and biological process categories provides comprehensive insight into the functional specialization of each cell type and elucidates the molecular mechanisms underlying tissue-specific cellular adaptations. The enrichment patterns reflect the distinct roles of endothelial cells in vascular function, fibroblasts in structural maintenance, and keratinocytes in barrier protection and immune surveillance.

Proteomics analysis result

Protein identification and quantification

For six biological samples (A1–A3, B1–B3), label-free quantitative proteomics analysis based on DIA-NN software (using the MaxLFQ algorithm) was conducted. The proteomic workflow yielded 61,396 unique peptide segments corresponding to 6,210 distinct proteins following stringent filtration at 1% false discovery rate (FDR). Comprehensive quality control assessment, including evaluation of peptide length distribution, peptide number distribution per protein, and missed cleavage frequency, confirmed that all analytical parameters met established quality standards for high-confidence protein identification (Fig. 14).

Fig. 14.

Quality control assessment of DIA-NN proteomic analysis. (A)Distribution of peptide segment lengths showing the frequency distribution of identified peptides across different amino acid length ranges. (B)Distribution of peptide abundance per protein illustrating the number of unique peptides identified for each protein across the dataset. (C)Distribution of missed cleavage sites demonstrating the frequency of incomplete tryptic digestion events in identified peptide segments. All quality metrics indicate successful sample preparation and mass spectrometric analysis meeting established standards for high-confidence protein identification and quantification

Differential protein expression analysis

Quantitative proteomic profiling revealed distinct expression patterns characterized by predominant abundance of structural proteins, specifically keratins and collagens, across all analyzed samples (Table 6). The top ten proteins in Table 6 were ranked by their quantitative expression levels in chestnut tissue, sorted from highest to lowest. Keratin type II cytoskeletal proteins demonstrated significantly elevated expression levels in chestnut samples (A1-A3) relative to skin samples (B1-B3), providing molecular evidence supporting the keratinized structural composition of chestnut tissue. Of particular note, Collagen type I alpha 2 chain exhibited markedly elevated expression in sample B1 (expression value: 2,898.34), representing a substantial increase compared to expression levels observed in remaining samples.

Table 6.

Top ten proteins ranked by quantitative expression analysis

Accession	Description	Gene	A1	A2	A3	B1	B2	B3
A0A8C4LPI2	Keratin, type II cytoskeletal	LOC106835243	4312.45	3257.56	2686.81	95.93	768.11	118.10
A0A8C4LA37	Keratin 10	KRT10	2497.96	2317.43	2427.67	405.58	742.20	463.420
A0A8C4LMC7	Keratin, type II cytoskeletal 1	KRT1	1510.36	1150.04	1177.04	338.71	446.44	358.21
A0A8C4LHC7	Keratin 14	KRT14	1341.80	1316.72	1192.19	933.30	937.57	1079.45
A0A9L0IJ82	Galectin	LOC10683607	891.50	901.67	747.81	162.86	196.30	188.79
A0A8C4LM51	Keratin 5	KRT5	828.01	908.55	913.75	399.06	530.25	506.46
A0A8C4N032	Collagen type I alpha 2 chain	COL1A2	572.57	364.52	125.22	2898.34	455.32	591.81
A0A8C4PSR9	Collagen type I alpha 1	COL1A1	568.91	378.27	132.65	2472.00	462.01	567.19
A0A8C4LS50	Actin	LOC106842058	472.89	557.31	512.95	600.96	755.03	682.01
A0A8C4L9M3	Histone	LOC123288462	425.83	546.71	423.87	658.22	721.61	791.61

The PCA analysis revealed robust clustering patterns with tight intra-group cohesion and distinct inter-group separation (Fig. 15A), demonstrating high experimental reproducibility within sample groups and significant biological differences between experimental conditions. Correlation matrix analysis further validated these observations, displaying strong positive correlations within sample groups and minimal correlations between distinct sample groups (Fig. 15B), confirming the biological significance of the observed expression differences.

Fig. 15.

The Differential Protein Expression (DEPs) profiling. A Principal component analysis (PCA) of proteomic expression profiles. Two-dimensional PCA plot illustrating the clustering patterns and sample relationships based on global protein expression data. Individual samples are represented as distinct points, with tight clustering within groups demonstrating high biological reproducibility and clear separation between groups indicating substantial proteomic differences between chestnut and skin tissue samples. B Correlation matrix analysis of inter-sample relationships. Heatmap visualization displaying Pearson correlation coefficients between all sample pairs based on protein expression profiles. Color intensity represents correlation strength, with strong positive correlations within sample groups and weak correlations between different tissue types, confirming experimental consistency and biological distinctiveness

Differentially expressed proteins (DEPs), KEGG and GO pathway analysis

Differential expression analysis identified 1,444 DEPs, comprising 906 up-regulated and 538 down-regulated proteins between chestnut and skin tissue samples (Fig. 16).

Fig. 16.

Volcano plot analysis of differentially expressed proteins (DEPs). Scatter plot displaying the relationship between fold change (x-axis) and statistical significance (y-axis, -log10 p-value) for all identified proteins. Each point represents an individual protein, with significantly up-regulated and down-regulated proteins highlighted according to the established criteria (FC ≥ 1.5 or FC ≤ 0.6667, p-value ≤ 0.05)

GO functional classification revealed that proteins are most enriched in cellular process and metabolic process (biological process), cellular anatomical entity and protein-containing complex (cellular component), and binding and catalytic activity (molecular function) (Fig. 17A).

Fig. 17.

The GO and KEGG pathway of differentially expressed proteins (DEPs). A GO functional classification bar chart. Bar chart visualization showing the functional distribution of all identified proteins in the sample across three Gene Ontology (GO) categories: biological process (BP), cellular component (CC), and molecular function (MF), providing insight into the overall functional landscape and predominant characteristics of the proteome. B KEGG pathway classification of DEPs. Bar chart visualization showing the distribution of DEPs across major KEGG pathway categories. The graph displays the percentage and absolute number of proteins assigned to each functional pathway, providing insight into the predominant biological processes affected in chestnut versus skin tissue. C Comparative KEGG pathway classification of up-regulated and down-regulated DEPs. Side-by-side bar chart comparison illustrating the distribution of up-regulated versus down-regulated proteins across KEGG pathway categories. This analysis reveals pathway-specific directional changes in protein expression between tissue types. D Bubble plot visualization of KEGG pathway enrichment analysis for DEPs. Each bubble represents a significantly enriched KEGG pathway, with bubble position indicating statistical significance (-log10 p-value or enrichment score) and bubble size proportional to the number of DEPs mapped to each pathway. The analysis identifies the most statistically significant biological pathways affected in chestnut versus skin tissue, with larger bubbles indicating pathways containing more DEPs and higher positions representing greater statistical significance of enrichment

KEGG functional annotation revealed that the most represented pathway categories among DEPs were cytoskeletal organization in muscle cells (8.03%) and focal adhesion complexes (4.47%) (Fig. 17B). The cytoskeleton functions as a critical regulator of cellular signaling cascades and controls specific signal transduction pathways [33], and in muscle tissue, cytoskeletal components maintain structural integrity, provide essential intracellular support, and facilitate contractile mechanisms [34], potentially contributing to the enhanced mechanical properties and firmer texture characteristic of chestnut tissue. Focal adhesion complexes mediate effective cell–matrix interactions through transmembrane receptors that connect the actin cytoskeleton to extracellular matrix (ECM) components [35, 36], and their formation and dynamics play essential roles in regulating apoptosis, proliferation, migration, and differentiation [37].

Comparative pathway analysis of up-regulated versus down-regulated DEPs demonstrated predominant up-regulation within the PI3K-Akt signaling pathway (24 up, 15 down), endocytosis pathway (23 up, 10 down), and proteasome pathway (32 up) (Fig. 17C). PI3K-Akt represents a critical regulatory network controlling cell survival, proliferation, and growth responses while preventing inflammatory responses through modulation of innate immune cell recruitment [38–40]. The endocytosis pathway facilitates cellular internalization of extracellular materials and participates in cell migration and signal transduction [41–43], while the proteasome is involved in degrading misfolded proteins and modulating diverse cellular functions [44, 45].

KEGG pathway enrichment analysis further identified significant enrichment of the proteasome and alpha-linolenic acid metabolism pathways among DEPs (Fig. 17D). The co-enrichment of the hypertrophic cardiomyopathy pathway is attributable to shared cytoskeletal and sarcomeric protein components rather than true cardiac pathology, reflecting the highly organized keratin and cytoskeletal architecture characteristic of chestnut tissue.

Discussion

Donkey chestnut tissue is morphologically distinct from adjacent skin, featuring a smooth, hairless, and highly pigmented surface. Building on our prior histological characterization of these differences [4], the present multi-omics study (scRNA-seq, RNA-seq, proteomics) provides unprecedented molecular insights into the biological.

nature of this specialized structure—revealing it as an evolutionarily conserved tissue with unique metabolic, cellular, and functional adaptations, rather than a vestigial remnant.

This study adopted single-cell RNA sequencing to systematically characterize the cellular heterogeneity between chestnut tissue and normal skin tissue of Dezhou donkeys at the single-cell resolution for the first time [46], overcoming the limitations of bulk sequencing which fails to identify cell subpopulations and cell-type-specific gene expression patterns. The core advantages of single-cell RNA sequencing lie in its capacity to resolve cellular heterogeneity in complex tissues, identify rare cell subsets, and capture cell-type-specific transcriptional signatures [47]. This technology has been widely applied in major livestock species including pigs, cattle and chickens, and has achieved substantial progress in research on embryonic development, organogenesis, immune system profiling and identification of functional cells associated with economic traits [48, 49].

In pigs, single-cell transcriptome analysis has been used to construct cell atlases across the entire embryonic development period and clarify the cell differentiation trajectories during early organogenesis [50]. In cattle, it has helped decipher cellular heterogeneity in mammary tissues and identify rare epithelial cell subsets linked to lactation performance [51, 52]. In chickens, relevant studies have revealed that Sertoli cells originate from mesenchymal stem cells during gonadal differentiation of chicken embryos [53]. Nevertheless, the application of single-cell RNA sequencing remains limited in donkeys, an important equine species. Particularly, systematic investigations into the cellular composition and functional characteristics of specialized skin structures such as chestnuts have not been reported to date.

Integrated evidence from three complementary omics approaches preliminarily constructs a molecular framework for the chestnut tissue of Dezhou donkeys, which helps to improve our understanding of this unique anatomical feature in equine animals.

The exceptional enrichment of keratin genes in chestnut tissue points to enhanced structural integrity compared to conventional skin [11]. At the transcript level, KRT1, KRT5, KRT10, and KRT14 reach TPM values exceeding 13,000, while at the protein level, Keratin type II cytoskeletal proteins and KRT10, KRT1, KRT14, and KRT5 dominate the proteomic profile with markedly higher abundance in chestnut samples relative to skin. This convergent evidence across both transcriptomic and proteomic analyses strongly corroborates the structural keratinization hypothesis for chestnut tissue. This hyperkeratinization likely confers superior mechanical resilience, serving as a protective adaptation during locomotion or environmental stress.

Complementing this structural specialization, single-cell analysis revealed that chestnut tissue is enriched in endothelial cells, pericytes, and immune cells, indicating enhanced vascularization and immunosurveillance. This cellular enrichment is consistent with the bulk RNA-seq findings of enhanced immune and vascular gene signatures, and the proteomic detection of elevated actin and structural proteins. The alignment of all three omics layers in highlighting vascular and immune features provides robust, multi-level evidence for these functional characteristics of chestnut tissue.

A key finding is the pronounced up-regulation of peroxisome-related pathways, coupled with activated PPAR signaling. Peroxisomes are critical for fatty acid β-oxidation, cholesterol biosynthesis, and reactive oxygen species detoxification [54], and their enhanced activity in chestnut tissue implies multiple functional benefits. For instance, melanogenesis is an energy-intensive process [55], and the observed metabolic up-regulation (including carbohydrate, carbon, and pyruvate metabolism) likely provides the biochemical support for sustained melanin production—underpinning the tissue’s distinctive pigmentation and potential photoprotective properties. Additionally, cholesterol biosynthesis gene enrichment reinforces epidermal lipid architecture, correlating with chestnut’s thicker epidermis [56]. PPAR signaling also mediates anti-inflammatory responses [57], suggesting chestnut tissue has enhanced capacity for inflammation resolution and homeostasis. This peroxisome/PPAR metabolic signature is evident in both the bulk RNA-seq KEGG enrichment and the proteomic pathway data, suggesting that the transcriptional upregulation translates to functional protein-level changes, reinforcing the biological validity of these findings.

Proteomic evidence of enhanced cytoskeletal organization further supports chestnut’s superior mechanical properties [33], regulation of focal adhesion pathways suggests specialized cell-matrix relationships that contribute to the tissue’s resilience and elasticity [35].

Intriguingly, both proteomic and single-cell analyses identified enrichment of coronavirus disease-related pathway annotations and mast cell/T lymphocyte populations, respectively. As discussed above, the COVID-19 pathway enrichment in KEGG should be interpreted as activation of broad innate immune and inflammatory modules, reflecting the exposed anatomical position of chestnut tissue rather than SARS-CoV-2-specific biology. Mast cells act as innate immunity sentinels, orchestrating inflammatory responses and tissue remodeling, further supporting an integrated defense system that combines structural, metabolic, and immunological specializations.

Taken together, the convergence of transcriptomic, single-cell, and proteomic data across three independent experimental platforms provides mutually corroborating evidence: (1) the structural keratinization signature is supported by elevated keratin gene expression in bulk RNA-seq, keratinocyte-specific transcriptional profiles in scRNA-seq, and high keratin protein abundance in proteomics; (2) the enhanced metabolic activity (peroxisome/PPAR) observed in bulk RNA-seq is echoed in proteomic pathway enrichment, with single-cell data further clarifying that these signals originate from coordinated activity in keratinocytes and endothelial cells; (3) immune and vascular features identified in bulk RNA-seq are precisely localized to mast cells, T lymphocytes, and endothelial cells via single-cell analysis, with proteomic upregulation of PI3K-Akt and endocytosis pathways providing functional validation. These inherent consistencies across omics layers constitute a robust, multi-level framework supporting our conclusions on chestnut tissue specialization.

Limitations and future directions

Although this study preliminarily analyzed the molecular characteristics of chestnut and skin tissues, several limitations should be acknowledged. First, the number of experimental animals and samples was relatively limited. Only four Dezhou donkeys were included for multi-omics analysis with a small scale of biological replicates, which may affect the population representativeness and generalizability of the results. Notably, three biological replicates per group represents the minimum threshold for statistical analysis, and no a priori statistical power analysis was conducted to determine the adequacy of this sample size. This constitutes a critical methodological limitation, and the omics findings reported here should be interpreted as hypothesis-generating rather than definitive. Future work should expand the sample size and include individuals of different ages and feeding groups for further validation. Second, this is a preliminary exploratory study. We only compared molecular differences between chestnut and normal skin at the transcriptomic, proteomic, and single-cell transcriptomic levels, without independent external sample validation or in vitro functional verification of key candidate genes and proteins. The reliability of identified molecular pathways still needs further experimental confirmation, such as qRT-PCR and Western blot verification in subsequent molecular biology studies. Third, the present study only focused on Dezhou donkeys, and comparative analysis across different equine breeds was not performed. The breed-specific differences in chestnut tissue characteristics remain unexplored.

Future research directions are proposed as follows: to investigate the temporal and spatial expression patterns of key regulatory genes during embryonic development and reveal the developmental basis of chestnut formation; to carry out comparative studies among different equine breeds and species to clarify the evolutionary conservation of chestnut structures; to evaluate the mechanical properties, wound healing capacity and antibacterial activity of chestnut tissues, so as to provide functional evidence for the potential advantages predicted by omics data; and to conduct systematic multi-omics integration using professional bioinformatics tools for deeper mechanistic interpretation.

Conclusion

This preliminary multi-omics analysis revealed that donkey chestnuts represent highly specialized keratinized tissues with distinctive molecular, cellular, and functional characteristics that extend far beyond their morphological uniqueness. The integration of transcriptomic, single-cell, and proteomic data provides convergent evidence supporting enhanced metabolic activity through coordinated up-regulation of peroxisome, fatty acid metabolism, and PPAR signaling pathways, coupled with specialized cellular composition featuring enriched endothelial, immune, and structural cell populations.

The molecular evidence suggests that chestnut tissue functions as an evolutionarily adapted structure with enhanced barrier function, metabolic capacity, and potential anti-inflammatory properties. The exceptional keratin expression levels and modified cytoskeletal organization provide molecular foundations for the observed superior mechanical properties, while the enriched metabolic pathways are consistent with enhanced pigmentation and cellular homeostasis. These findings advance our understanding of equine biology and provide a molecular framework for future investigations into tissue specialization and evolutionary adaptation. The identified molecular signatures and cellular characteristics establish chestnut tissue as a unique biological system worthy of continued scientific investigation, with potential applications in veterinary medicine, comparative biology, and biotechnology.

Authors’ contributions

Conceptualization, methodology, supervision, writing—original draft, project administration: W.C., M.Z.K. and C.W (Chunming Wang).; formal analysis and interpretation, sample collection, software: data curation, validation, writing—review and editing: W.C., J.L., A.U., M.D., Y.P., C.W (Changfa Wang), M.Z.K., and C.W (Chunming Wang), resources and funding: C.W. All authors have read and agreed to the published version of the manuscript.

Funding

The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This research was funded by the Key R&D Program of Shandong Province (2025LZGC033), the National Key R&D Program of China (grant number 2023YFD1302004), Liaocheng Municipal Bureau of Science and Technology, High-talented Foreign Expert Introduction Program (GDWZ202401), The Shandong Province Modern Agricultural Technology System Donkey Industrial Innovation Team (grant no. SDAIT-27), Livestock and Poultry Breeding Industry Project of the Ministry of Agriculture and Rural Affairs (grant number 19211162), The National Natural Science Foundation of China (grant no. 31671287), The Open Project of Liaocheng University Animal Husbandry Discipline (grant no. 319312101-14), The Open Project of Shandong Collaborative Innovation Center for Donkey Industry Technology (grant no. 3193308), Research on Donkey Pregnancy Improvement (grant no. K20LC0901), and Liaocheng University scientific research fund (grant no. 318052025).

Data availability

The RNA-seq data generated in this study have been deposited in the NCBI SRA database under BioProject accession number PRJNA1348122.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the institutional Special Committee of Scientific Research Ethic of Liaocheng University (AP2025031726).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.