Integrated multi-omics analysis reveals breed-specific molecular and metabolic architectures of breast muscle in Cherry Valley, Muscovy, and mule ducks Authors: Lionel Kinkpe, Lihong Gu, Shun Guo, Zixin Xu, Chengjun Hu, Zhigang Hu, Yaya Sun, Dingping Bai, Xinli Zheng, Tieshan Xu, Xia Wang Published: 2026 ABSTRACT AbstractIntegrated transcriptomic and metabolomic analyses of pectoralis major muscle were used to characterize molecular differences among Cherry Valley duck (CVD), Muscovy duck (MD), and their mule hybrid (MU). Birds were reared under identical conditions and evaluated for carcass traits, free amino acids, fatty acids, untargeted metabolomics, and transcriptomics (n = 8 per genotype for phenotypic and metabolomic analyses; CVD = 4, MD = 3, and MU = 4 for transcriptomics). The highest transcriptional difference occurred between the parental lines, with 6,041 differentially expressed genes (DEGs), compared with 2,758 DEGs between CVD and MU and 2,986 between MD and MU. CVD had higher expression of proteasome subunit genes than MD and MU, including genes encoding the 20S core, 19S regulatory particle, and PA28 and PA200 activators. Genes associated with mTOR signalling were also more highly expressed in CVD, accompanied by enrichment of ribosome biogenesis gene sets. Collagen and extracellular matrix genes showed the inverse pattern, with lower expression in CVD and similar expression in MD and MU. MU formed a distinct transcriptional profile from both parental lines. Most DEGs in the parent–hybrid comparisons were more abundant in MU, and oxidative phosphorylation gene sets were more highly expressed in MU than in MD. Free arginine was approximately twice as abundant in MU as in either parent, while total free amino acids did not differ among genotypes. MD had the highest glutamine, asparagine, serine, glycine, and lysine concentrations, while CVD had the highest thrombogenic index and ω-6/ω-3 ratio. These results identify distinct molecular and compositional profiles among the three genotypes, with specific patterns associated with the parental backgrounds and hybridisation.Keywords: Duck, Meat quality, Multi-omics integration, Oxidative phosphorylation Introduction Meat quality is a multifactorial phenotype shaped by the complex interplay of genetic background, metabolic pathways, and post-mortem biochemical processes (Picard and Gagaoua, 2020; Warner et al., 2022). In commercial poultry production, a fundamental challenge persists between genetic selection for rapid growth and high half-eviscerated carcass percentage versus the preservation of desirable sensory and nutritional attributes valued by consumers (Petracci et al., 2015). This yield-quality trade-off is particularly pronounced in duck production, where diverse breeding histories have generated genotypes occupying distinct positions along the phenotypic spectrum. The Cherry Valley duck (CVD, Anas platyrhynchos), a Pekin-type commercial strain intensively selected for production efficiency, represents one extreme; the Muscovy duck (MD, Cairina moschata), renowned for superior meat characteristics, represents another; and the Mule duck (MU), their interspecific hybrid, potentially combines attributes from both parental lineages (Chartrin et al., 2006; Kokoszyński et al., 2020). Reared together, these three genotypes allow the effects of selection history and of hybridisation on muscle composition to be examined in the same experiment. The sensory attributes that define meat quality (tenderness, juiciness, color stability, and flavor) are fundamentally governed by the biochemical composition of muscle tissue (Hughes et al., 2020). Amino acid profiles influence protein content and flavor precursor availability; fatty acid composition determines nutritional value, oxidative stability, and contributes to characteristic flavor development through lipid oxidation during cooking; and metabolite concentrations reflect the metabolic status of muscle at slaughter and during post-mortem conversion (Lee et al., 2023). Phenotypic studies have documented substantial variation in these compositional parameters across duck breeds and show that Muscovy ducks typically exhibit higher amino acid concentrations and more favorable fatty acid profiles than Pekin-type breeds (Wawro et al., 2004; Li et al., 2024). These descriptive characterizations, however, cannot identify the molecular processes associated with these differences. Integrated molecular profiling can help identify candidate gene-expression programs and metabolic pathways associated with breed-specific biochemical and meat-quality characteristics. Protein turnover and intramuscular connective tissue are particularly relevant to differences in muscle development and composition among poultry genotypes. Skeletal muscle protein turnover results from the coordinated processes of protein synthesis and degradation. Protein synthesis is regulated in part through mTOR signalling, while the ubiquitin–proteasome system constitutes a major pathway of intracellular protein degradation. Differences in components of these systems have been reported between poultry lines with contrasting growth characteristics (Nakashima et al., 2009; Saneyasu et al., 2015; Xu and Velleman, 2023b). Protein degradation also contributes to the intracellular free amino acid pool, linking protein turnover with aspects of muscle composition relevant to meat quality. Intramuscular connective tissue represents another important component of the muscle phenotype. Its abundance and organisation, particularly those of collagen, vary among poultry genotypes differing in growth rate (Huo et al., 2021). The contribution of collagen to meat texture is complex in poultry and does not necessarily follow the relationships established in red meat species (Purslow, 2018; Li et al., 2022). Despite the importance of these processes, their transcriptional profiles have received limited attention in comparisons of duck genotypes with markedly different breeding histories. Hybridisation introduces an additional level of variation in the regulation of muscle traits. Studies of avian hybrids have reported several patterns of gene expression relative to the parental populations. Mid-parent expression predominates in some intraspecific chicken crosses (Zhuo et al., 2019; Zhao et al., 2024), while expression levels outside the parental range have been documented in naturally occurring avian interspecific hybrids and in reciprocal chicken crosses (Mugal et al., 2020; Mai et al., 2021). The extent of non-additive expression has also been associated with genetic divergence between parental populations (Runemark et al., 2025). Mule duck provides a particular case, as it results from hybridisation between species belonging to different genera. Quantitative-genetic studies of this cross have reported intermediate values for muscle weight and significant additive effects for growth, with heterosis detected for a limited number of traits (Larzul et al., 2006). Transcriptomic comparisons involving Muscovy duck, common duck, and their hybrids have also identified differences in gene expression among these genetic groups (Liu et al., 2020). However, the extent to which the muscle molecular and compositional characteristics of MU resemble either parental line, remain intermediate, or extend beyond parental values has not been established for protein turnover, extracellular matrix organisation, and the metabolic characteristics associated with meat composition. The combination of transcriptomic and metabolomic analyses provides complementary information on these processes. Transcriptomic analysis identifies differences in gene expression associated with muscle development, protein and lipid metabolism, and cellular energy metabolism (Perrin et al., 2018; de Souza Fonseca et al., 2022). Metabolomic analysis characterises small-molecule composition and provides information on biochemical differences that are closer to the measured muscle phenotype (Ramanathan et al., 2020). Integration of these datasets can identify biological processes supported at both molecular levels and can also reveal cases in which transcriptional and metabolite differences do not coincide (Munekata et al., 2021). Previous studies have documented genotype-associated differences in carcass traits, meat quality, amino acid composition and fatty acid profiles among duck lines (Li et al., 2024). Recent studies have also applied transcriptomic and metabolomic approaches to duck breast muscle, identifying coordinated differences in genes and metabolites associated with meat-quality traits (Xu et al., 2025). In poultry, lipidomic and multi-omics analyses have repeatedly identified glycerophospholipid metabolism among the pathways associated with intramuscular fat deposition, lipid composition and meat-quality variation (Zhang et al., 2025a, 2026b; Li et al., 2026). However, integrated transcriptomic and metabolomic comparison of both parental duck lines and their mule hybrid under the same experimental conditions remains limited. To address these gaps, pectoralis major muscle from CVD, MD, and MU reared under identical conditions was characterised for morphometric, carcass, and biochemical traits, including free amino acid and fatty acid composition. Differentially abundant metabolites and their associated metabolic pathways were examined across the three genetic groups, together with differences in gene expression, with particular attention to protein turnover and extracellular matrix-associated genes. Transcriptomic and metabolomic data were further integrated through joint pathway analysis to assess the molecular and compositional characteristics shared between MU and its parental lines and those that distinguished the hybrid from both parents. Materials and methods Experimental design and animal management A completely randomized design was employed to compare three genetically distinct duck populations: Cherry Valley duck (CVD, Anas platyrhynchos domesticus, Pekin-type commercial strain), Muscovy duck (MD, Cairina moschata, Guanghong strain), and their F1 interspecific hybrid. Male ducks were selected from a breeding population of 50 birds per genotype. At 70 days of age, eight birds per genotype were randomly selected for phenotypic characterisation and metabolomic profiling, yielding 24 birds in total. A nested subset (n = 4 per genotype, except MD n = 3 after one library exclusion) was used for transcriptomic analysis. All animals were reared at the HAIANAN FARMS experimental facility (Hainan Province, China) under standardized environmental conditions (temperature: 20 to 25°C; relative humidity: 60 to 70%). All birds received ad libitum access to water and a commercial duck grower diet (crude protein: 16.5%; metabolizable energy: 11.7 MJ/kg) from day 1 until slaughter, which was conducted at the same chronological age across all three genotypes. All animal procedures were conducted in strict compliance with the Animals (Scientific Procedures) Act 1986, EU Directive 2010/63/EU on the protection of animals used for scientific purposes, and the NIH Guide for the Care and Use of Laboratory Animals (8th edition). The study was approved by the Institutional Animal Care and Use Committee at the Experimental Animal Center of Hainan Academy of Agricultural Science (HNXMSY 20210743), ensuring adherence to internationally recognized ethical standards for animal welfare. Slaughter protocol and sample collection Ducks were fasted for 6 h with free access to water to standardize gastrointestinal tract contents prior to slaughter. Live body weight was recorded immediately before processing using a calibrated digital scale (±1 g precision). Birds were humanely slaughtered by electrical stunning (50 V, 400 Hz, 5 s) followed by exsanguination according to commercial procedures compliant with animal welfare regulations. Blood samples (5 mL) were collected during exsanguination into K2-EDTA vacuum tubes (Becton Dickinson, Franklin Lakes, NJ, USA), immediately placed on ice, and centrifuged at 3,000 × g for 15 min at 4°C within 30 min of collection. Plasma aliquots (500 μL) were transferred to cryovials and stored at −80°C until analysis. The complete left Pectoralis major muscle was excised within 30 min post-mortem to minimize post-mortem metabolic changes. Each muscle was divided into three portions designated for: (i) meat quality assessment (2 cm³, stored at 4°C for immediate analysis); (ii) metabolomic profiling (100 mg, flash-frozen in liquid nitrogen); and (iii) transcriptomic analysis (approximately 100 mg, preserved in RNAlater® Stabilization Solution; Thermo Fisher Scientific, Waltham, MA, USA). Frozen samples were transferred to −80°C storage within 2 h of collection and maintained until extraction. Phenotypic characterization Carcass traits and morphometrics Carcass processing followed standardized procedures to enable cross-genotype comparison. Hot carcass weight was recorded immediately after evisceration, with the head and feet attached, following the standard protocol mandated by national regulations. Slaughter performance indices were calculated as: Morphometric measurements were obtained using digital calipers (Mitutoyo 500-196-30, ±0.01 mm precision; Mitutoyo Corporation, Kawasaki, Japan) following the protocols described by Kadurumba et al. (2025). Pre-slaughter measurements included body diagonal length, neck length, keel length, chest width, and hip bone width. Post-slaughter measurements included tibia length and tibia circumference. Meat quality assessment All meat quality measurements were performed on the left Pectoralis major muscle collected from each duck, as described above. Color was measured after a 30-min blooming period at 4°C to allow myoglobin oxygenation equilibration. CIE L* (lightness), a* (redness), and b* (yellowness) coordinates were determined using a TS7700 portable spectrophotometer (Shenzhen Sanenshi Technology Co., Ltd., Guangdong, China) equipped with D65 illuminant and 10° standard observer, aperture diameter 8 mm. Three measurements were taken at random non-overlapping locations on the freshly cut muscle surface and averaged. Early post-mortem pH (pH₅, measured at 45 min post-mortem) was determined using a Testo 205 pH meter (Testo SE & Co. KGaA, Titisee-Neustadt, Germany) fitted with a penetration electrode (3 mm diameter). The probe was inserted to a depth of 2.5 cm at three locations per sample; mean values were used for statistical analysis. The instrument was calibrated using pH 4.01 and 7.00 buffer solutions (Hanna Instruments, Woonsocket, RI, USA) before each measurement session. Moisture content was determined gravimetrically according to AOAC Official Method 950.46 (AOAC International, 2019). Approximately 5 g of minced muscle tissue was dried in a forced-air oven at 105°C until constant weight was achieved (weight change <0.01 g over 2 h intervals; approximately 24 h total). Moisture content (%) was calculated as: Inosine 5′-monophosphate (IMP) concentration, a key flavor nucleotide, was quantified by high-performance liquid chromatography (HPLC) in Pectoralis major muscle samples (n = 8 biological replicates per genotype). Briefly, 2 g of minced muscle was homogenized with 10 mL of 0.6 M perchloric acid, centrifuged at 10,000 * g for 10 min at 4°C, and the supernatant was neutralized to pH 6.5 to 6.8 using 1 M KOH. The neutralized extract was filtered through a 0.22 μm PTFE membrane and analyzed using an Agilent 1260 Infinity II HPLC system (Agilent Technologies, Santa Clara, CA, USA) equipped with a C18 column (Agilent ZORBAX Eclipse Plus, 4.6 * 250 mm, 5 μm) and UV detector at 254 nm. IMP was quantified against an external calibration curve (r² > 0.999) prepared using certified reference standard (Sigma-Aldrich, St. Louis, MO, USA; Cat. No. I4625). Biochemical composition analysis All biochemical analyses described in this section were performed on Pectoralis major muscle from the same 24 birds (n = 8 biological replicates per genotype). Free amino acid profiling. Free amino acids were determined by pre-column phenylisothiocyanate (PITC) derivatisation followed by reversed-phase HPLC-UV. Freeze-dried muscle (0.5 g) was extracted in 3 mL of 0.1 mol/L HCl in a 10 mL centrifuge tube, homogenised, and subjected to ultrasonic extraction for 20 min with one intermediate agitation. The extract was centrifuged at 10,000 rpm at 4°C for 10 min and the supernatant collected; the extraction was repeated three times and the supernatants pooled. An aliquot of 300 µL was combined with 150 µL of PITC derivatisation reagent, vortexed for 1 min and incubated at 25°C for 60 min. Excess reagent was removed by adding 2 mL of n-hexane, vortexing for 15 s, allowing 2 min for phase separation and centrifuging at 10,000 rpm for 5 min, discarding the upper hexane layer; this step was repeated twice. The lower aqueous phase was filtered through a 0.22 µm membrane. Separation was performed on an Agilent HC—C18(2) column (4.6 mm × 250 mm, 5 µm) at 35°C with an isocratic mobile phase of acetonitrile:methanol:water (6:2:2, v/v/v) at 1.0 mL/min; UV detection at 254 nm, injection volume 10 µL. Twenty-one free amino acids were quantified by external calibration against certified reference standards (Sigma-Aldrich, St. Louis, MO, USA) and expressed as µg/g fresh muscle weight: alanine, arginine, asparagine, aspartic acid, cystine, glutamic acid, glutamine, glycine, histidine, isoleucine, leucine, lysine, methionine, phenylalanine, proline, serine, taurine, threonine, tryptophan, tyrosine and valine. Untargeted metabolomic profiling Metabolite extraction A dual-extraction protocol was employed to maximize coverage of both hydrophilic and lipophilic metabolite classes. Frozen Pectoralis major tissue (20 ± 1 mg) was weighed into 2-mL Eppendorf tubes maintained on dry ice. For hydrophilic metabolite extraction, 400 μL of pre-chilled 70% methanol (v/v, HPLC grade; Merck, Darmstadt, Germany) containing 0.1 μg/mL of 2‑chloro-l-phenylalanine (internal standard; Sigma-Aldrich; Cat. No. C6506) was added. For lipophilic metabolite extraction, a separate tissue aliquot received 1 mL of methyl tert‑butyl ether:methanol (3:1, v/v; HPLC grade) containing 0.1 μg/mL of LPC (12:0) as internal standard (Avanti Polar Lipids, Alabaster, AL, USA). Tissue disruption was achieved using a Tissuelyser II ball mill (Qiagen, Hilden, Germany) at 30 Hz for 20 s with two 5-mm stainless steel beads per tube, repeated twice with 30-s intervals on ice. Homogenates were vortexed for 30 s and incubated at 4°C for 30 min to enhance extraction efficiency. Phase separation was achieved by centrifugation at 12,000 × g for 10 min at 4°C. Supernatants (200 μL) were transferred to autosampler vials with glass inserts for immediate analysis. All extraction procedures were conducted on ice to preserve metabolite stability. LC-MS/MS analysis Metabolite profiling was performed using an ExionLC AD Ultra-Performance Liquid Chromatography system (SCIEX, Framingham, MA, USA) coupled to a QTRAP 6500+ triple quadrupole-linear ion trap mass spectrometer (SCIEX) equipped with an electrospray ionization (ESI) source. Sample preparation, LC-MS/MS acquisition, and metabolite quantification were conducted by Metware Biotechnology Co., Ltd. (Wuhan, China). All subsequent data processing, statistical analysis, and interpretation were performed by the authors at Northwest A&F University (NWAFU) Hydrophilic metabolites were separated on a Waters ACQUITY UPLC HSS T3 C18 column (1.8 μm, 2.1 * 100 mm; Waters Corporation) maintained at 40°C. Mobile phase A consisted of 0.1% formic acid in water; mobile phase B consisted of 0.1% formic acid in acetonitrile. The gradient program was: 0 to 2 min, 5% B; 2 to 4 min, 5 to 30% B; 4 to 6 min, 30 to 60% B; 6 to 8 min, 60 to 90% B; 8 to 9 min, 90% B; 9 to 9.1 min, 90 to 5% B; 9.1 to 10 min, 5% B (column re-equilibration). Flow rate was 0.4 mL/min; injection volume was 2 μL. Lipophilic metabolites were separated on a Thermo Accucore C30 column (2.6 μm, 2.1 × 100 mm; Thermo Fisher Scientific) maintained at 45°C. Mobile phase A consisted of acetonitrile:water (60:40, v/v) containing 0.1% formic acid and 10 mM ammonium formate; mobile phase B consisted of acetonitrile:isopropanol (10:90, v/v) containing 0.1% formic acid and 10 mM ammonium formate. The gradient program was: 0 to 2 min, 30% B; 2 to 4 min, 30 to 43% B; 4 to 8 min, 43 to 55% B; 8 to 10 min, 55 to 70% B; 10 to 14 min, 70 to 99% B; 14 to 18 min, 99% B; 18 to 18.1 min, 99 to 30% B; 18.1 to 20 min, 30% B. Flow rate was 0.35 mL/min; injection volume was 2 μL. Mass spectrometry detection was performed in both positive and negative ionization modes. ESI source parameters were: source temperature 500°C; ion spray voltage +5500 V (positive) and −4500 V (negative); curtain gas 35 psi; ion source gas 1 and 2, 60 psi each. Metabolite detection and quantification employed scheduled multiple reaction monitoring (sMRM) with optimized declustering potentials and collision energies for each transition. Metabolite identification was performed by matching retention times (±0.2 min) and MS/MS spectra against the Metware Database (MWDB V2.0; Metware Biotechnology Co., Ltd., Wuhan, China), which contains >25,000 metabolite standards. Quality control was ensured by analyzing pooled quality control samples throughout the sequence, with metabolites showing coefficient of variation >30% in QC samples being excluded from further analysis. Detailed extraction protocols, chromatographic gradients, and mass spectrometry parameters are provided in Supplementary Methods. RNA extraction and quality assessment, Library construction and sequencing and Read processing and alignment Total RNA was extracted from Pectoralis major muscle tissue using TRIzol reagent (Invitrogen) according to the manufacturer's protocol. RNA integrity was verified using an Agilent 2100 Bioanalyzer, with all samples exhibiting RNA Integrity Numbers (RIN) greater than 8.0. Strand-specific cDNA libraries were prepared using the NEBNext Ultra II Directional RNA Library Prep Kit (New England Biolabs) and sequenced on an Illumina NovaSeq 6000 platform (Illumina Inc.) to generate 150 bp paired-end reads with an average depth of 40 million reads per sample. All data processing, quality control, alignment, differential expression analysis, and functional enrichment analysis were performed by the authors using the High-Performance Computing platform of NWAFU. Quality control of raw sequencing data was performed using FastQC (v0.11.9). Adapter sequences and low-quality bases were trimmed using Trimmomatic (v0.39) with parameters: LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36. Cleaned reads were aligned to the duck reference genome (Anas_platyrhynchos_ZS-1_1.0) using HISAT2 (v2.2.1) with default parameters. Gene expression quantification was performed using featureCounts (v2.0.1) and normalized as Fragments Per Kilobase of transcript per Million mapped reads (FPKM). Differential gene expression analysis was conducted using the DESeq2 package (v1.30.1) in R. Genes with |log₂FC| ≥ 2 and false discovery rate (FDR)–adjusted P-value < 0.05 were considered differentially expressed. Data analysis Statistical analyses were performed in R (v4.1.0). Phenotypic data were tested for normality using the Shapiro–Wilk test and for homogeneity of variance using Levene’s test. Traits satisfying both assumptions were analysed by one-way ANOVA followed by Tukey’s HSD test. Arginine, serine, and threonine, which did not satisfy these assumptions, were analysed using the Kruskal–Wallis test followed by Dunn’s post-hoc test. Processed metabolomic data were PQN-normalised, log₂-transformed, and analysed by principal component analysis (PCA) and orthogonal partial least-squares discriminant analysis (OPLS-DA). Differentially abundant metabolites (DEMs) were defined by a variable importance in projection (VIP) score ≥ 1 from a validated OPLS-DA model, P < 0.05 from Student’s t-test, and a fold-change threshold of ≥ 1.2 for general metabolites and ≥ 2 for lipids. A more stringent fold-change threshold was applied to lipids to retain changes of larger magnitude, in accordance with the greater analytical and biological variability reported for lipid species (Liebisch et al., 2019). Metabolite pathway enrichment was assessed by over-representation analysis against the KEGG database, with correction for false discovery rate (FDR). Differentially expressed genes (DEGs) were defined by |log₂FC| ≥ 2 and FDR < 0.05. Supplementary Tables S4–S6 retain the orientation generated by the original differential expression analysis for each comparison. Two complementary approaches were used for transcriptomic enrichment analysis. Over-representation analysis was conducted in Metascape v3.5 (Zhou et al., 2019), with Homo sapiens specified as both the input and analysis species. Gene symbols were deduplicated by retaining the most significant transcript for each symbol, and unnamed LOC loci were excluded. The three DEG lists were analysed separately. In the absence of Anas platyrhynchos annotation in Metascape, duck gene symbols were mapped to their human orthologues. This procedure mapped 1,334 of 6,041 DEGs for CVD vs. MD, 519 of 2,758 for CVD vs. MU, and 523 of 2,986 for MD vs. MU. Enrichment was assessed using Gene Ontology Biological Process, Reactome, KEGG, MSigDB canonical pathways, including the NABA matrisome collections, and WikiPathways. Terms were retained with a minimum overlap of three genes, P < 0.01, and an enrichment factor > 1.5. Redundant terms were grouped according to kappa similarity, and the most significant term from each cluster was displayed. Gene set enrichment analysis was performed using the fgsea package (v1.20.0) on the complete ranked gene list for each comparison, independently of the DEG threshold. Genes were ranked according to signed significance, calculated as sign(log₂FC) * −log₁₀(P) using the unadjusted P-value. Values below machine precision were set to the smallest representable double to prevent infinite ranking scores. Signed significance was selected for ranking owing to the presence of approximately 2,400–2,700 genes per comparison with |log₂FC| > 20 resulting from near-zero counts in one group, which would give these genes disproportionate weight in a fold-change-based ranking. Duplicate symbols were collapsed by retaining the most significant value, and unnamed LOC loci were excluded. Gene sets were obtained from Gene Ontology Biological Process (org.Hs.eg.db v3.14.0 and GO.db v3.14.0) and KEGG pathway annotations retrieved using KEGGREST and were restricted to sets containing 15–500 genes. Statistical significance was determined using the fgseaMultilevel procedure with Benjamini–Hochberg correction. Positive normalised enrichment scores indicate enrichment among genes with higher expression in the first-named breed of each comparison. Complete results are provided in Supplementary Table S9-A for Gene Ontology and Supplementary Table S9-B for KEGG. Transcriptomic and metabolomic datasets were integrated using joint pathway analysis in MetaboAnalyst 6.0. The human KEGG pathway library was used in the absence of a corresponding duck library, and duck gene symbols were mapped to human orthologues. DEGs and DEMs were entered as a single pooled query for each comparison against the “all pathways” universe, encompassing metabolic and gene-only regulatory pathways. Pathway enrichment was assessed using the hypergeometric test, and topological importance was estimated using degree centrality. Pathways were retained when they reached FDR < 0.05 in the joint analysis and contained at least three mapped features. Human disease and organismal pathways were excluded from biological interpretation owing to extensive overlap of their gene membership with proteasome, mitochondrial, and synaptic gene sets that lack direct interpretability in avian skeletal muscle. Impact scores are reported together with enrichment statistics. An impact score of zero indicates that the mapped features had no topological importance under the degree-centrality measure and does not preclude statistical enrichment of the pathway. Differences in annotation backgrounds among over-representation analysis, gene set enrichment analysis, and joint pathway analysis were considered when comparing their outputs, and pathway names were not assumed to be directly equivalent across approaches. All DEGs and DEMs identified in the three pairwise comparisons are provided in Supplementary Tables S1–S6. These tables include gene symbols, log₂FC values, adjusted P-values, metabolite classes, molecular formulas, CAS numbers, and per-sample normalised metabolite intensities and are available through Zenodo (DOI:10.5281/zenodo.22021238). Raw RNA-sequencing data have been deposited in the Genome Sequence Archive (GSA) under accession number CRA040079, and the metabolomic dataset has been deposited in the OMIX repository under accession number OMIX015702. Results Breed-specific phenotypic profiles reveal distinct morphometric, carcass, and biochemical specializations Comprehensive phenotypic characterization under standardized rearing conditions revealed significant breed differentiation across morphometric, carcass composition, and biochemical parameters. Morphometric analysis demonstrated that MD and MU consistently ranked highest for body length, chest width, keel length, live weight, neck length, and submerged length (p < 0.05), outperforming CVD across these structural dimensions (Fig. 1). MD exhibited superior tibia circumference, followed by MU and CVD, while MU achieved the highest tibia length values (p < 0.05), indicating hybrid-specific skeletal conformation. Morphometric measurements across duck breeds. Comparison of body morphometric traits (body length, chest width, hip bone width, keel length, neck length, submerged length, live weight, tibia length, and tibia circumference) among Cherry Valley duck (CVD), Muscovy duck (MD), and Mule duck (MU) (n = 8 per genotype). Data are means ± SD; breeds not sharing a common letter differ (P < 0.05). Carcass composition analysis revealed complementary breed specializations (Fig. 2). CVD ducks had the highest numerical means for dressing percentage, half-eviscerated carcass percentage, and Pectoralis major proportion (p < 0.05). MD ducks were statistically comparable to CVD for dressing percentage and half-eviscerated carcass percentage. MU hybrids were comparable to MD for leg muscle percentage and to CVD for Pectoralis major percentage (p > 0.05), showing that their carcass characteristics did not consistently follow either parental genotype. Carcass and slaughter performance across duck breeds. Comparison of carcass-composition traits (dressing percentage, half-eviscerated carcass percentage, Pectoralis major percentage, and leg muscle percentage) among CVD, MD, and MU (n = 8 per genotype). Data are means ± SD; breeds not sharing a common letter differ (P < 0.05). Biochemical profiling of breast muscle revealed pronounced breed-specific nutritional signatures (Table 1, Table 2). MD meat exhibited significantly elevated concentrations of key amino acids including glutamine, asparagine, serine, glycine, and lysine, consistent with enhanced protein biosynthetic capacity. MU hybrids showed the highest arginine levels, while CVD ducks had maximal taurine content, revealing distinct amino acid profiles across genotypes. Fatty acid profiling showed that MU had the highest concentrations of total fatty acids, SFA, PUFA, and both the n-6 and n-3 families (P < 0.05), whereas MUFA was highest in MD, with MU not differing significantly from MD (P > 0.05). MU also had the highest concentrations of stearic acid (C18:0) and elaidic acid (C18:1n9t), with MD intermediate and not significantly different from CVD. MD showed the most favourable hypocholesterolemic to hypercholesterolemic ratio, while CVD displayed the highest thrombogenic index and n-6/n-3 ratio, both associated with a less desirable nutritional profile. Table 1Free amino acid content of breast muscle (µg/g fresh muscle weight).ParameterCVDMDMUp_ValueAlanine902.73 ± 106.46891.46 ± 92.98895.70 ± 131.749.7E-01Arginine394.48 ± 165.22b322.74 ± 56.87b707.10 ± 197.88a6.0E-06Asparagine126.72 ± 36.16c278.41 ± 49.01a195.76 ± 66.29b1.4E-06Aspartic_acid51.37 ± 12.1855.47 ± 12.8557.26 ± 22.407.0E-01Cystine65.47 ± 0.8465.13 ± 0.8065.50 ± 1.176.3E-01Glutamic_acid106.69 ± 37.02118.29 ± 24.72120.91 ± 53.666.9E-01Glutamine1135.23 ± 278.58b2563.77 ± 805.32a1581.20 ± 816.40b1.5E-04Glycine100.80 ± 27.18b164.53 ± 32.47a119.27 ± 31.02b1.4E-04Histidine74.41 ± 15.4766.22 ± 16.0985.86 ± 38.072.3E-01Isoleucine49.32 ± 17.1561.51 ± 12.2748.28 ± 17.311.3E-01Leucine75.98 ± 27.3890.51 ± 17.8075.36 ± 24.152.7E-01Lysine107.05 ± 28.12b143.81 ± 31.60a137.69 ± 35.73ab2.8E-02Methionine54.72 ± 14.4359.20 ± 7.9965.78 ± 12.131.1E-01Phenylalanine63.15 ± 15.6374.71 ± 13.5272.08 ± 11.131.4E-01Proline88.06 ± 21.66104.14 ± 44.6595.51 ± 35.625.8E-01Serine91.11 ± 20.44b216.72 ± 73.37a203.89 ± 66.19a2.7E-05Taurine2256.17 ± 532.07a1382.13 ± 398.39b1565.03 ± 448.15b3.7E-04Threonine109.46 ± 29.70109.33 ± 17.34131.42 ± 52.172.9E-01Tryptophan52.30 ± 7.0457.66 ± 6.9552.56 ± 5.661.3E-01Tyrosine73.81 ± 20.7074.12 ± 13.0383.20 ± 16.723.7E-01Valine80.43 ± 26.2281.64 ± 13.9570.20 ± 21.234.0E-01Means ± SD with letters indicating significant differences (p < 0.05). Breeds sharing the same letter within a parameter are not significantly different.All amino acids reported in this table were determined as free (non-protein-bound) amino acids by pre-column phenylisothiocyanate derivatisation followed by reversed-phase HPLC-UV at 254 nm, and are expressed as µg/g fresh muscle weight. Free amino acid content of breast muscle (µg/g fresh muscle weight). Means ± SD with letters indicating significant differences (p < 0.05). Breeds sharing the same letter within a parameter are not significantly different. All amino acids reported in this table were determined as free (non-protein-bound) amino acids by pre-column phenylisothiocyanate derivatisation followed by reversed-phase HPLC-UV at 254 nm, and are expressed as µg/g fresh muscle weight. Table 2Fatty acid content (µg/g fresh muscle weight) its index composition across breeds.ParameterCVDMDMUp-ValueC14_05.69 ± 2.823.92 ± 0.324.30 ± 0.240.06C15_04.79 ± 3.003.63 ± 0.113.77 ± 0.170.256C16_0226.30 ± 50.06 b259.62 ± 59.93 b328.42 ± 56.95 a<0.001C16_1n95.11 ± 1.474.79 ± 0.394.52 ± 0.360.175C16_1n75.45 ± 0.945.58 ± 0.555.60 ± 0.670.870C17_07.92 ± 1.68 ab7.17 ± 0.59 b8.45 ± 0.95 a0.006C18_0198.49 ± 44.85 b226.50 ± 51.59 ab278.38 ± 69.81 a0.007C18_1n9123.53 ± 35.88 b192.84 ± 38.37 a165.22 ± 34.23 a<0.001C18_1n9t32.51 ± 12.63 b43.61 ± 10.68 ab50.85 ± 14.44 a0.008C18_2n6223.16 ± 86.65 b273.70 ± 64.32 b394.14 ± 144.71 a0.002C18_3n38.52 ± 1.12 b8.76 ± 0.61 b15.74 ± 2.80 a<0.001C20_08.10 ± 0.43 ab7.93 ± 0.46 b8.55 ± 0.52 a0.014C20_1n99.02 ± 0.989.01 ± 0.678.98 ± 0.740.992C20_3n98.34 ± 0.33 b9.79 ± 0.91 a8.79 ± 0.41 b<0.001C20_2n615.23 ± 2.65 b16.29 ± 2.05 ab19.22 ± 3.36 a0.006C20_3n622.66 ± 9.4220.14 ± 3.9623.76 ± 7.180.384C20_4n6341.19 ± 77.23311.82 ± 63.41357.33 ± 77.850.371C20_5n39.39 ± 0.30 b9.83 ± 0.41 b13.74 ± 1.39 a<0.001C22_4n651.87 ± 13.62 b75.40 ± 19.87 a58.17 ± 14.53 ab0.020C22_5n619.64 ± 3.49 b29.86 ± 4.22 a29.45 ± 4.81 a<0.001C22_5n317.63 ± 3.74 b26.27 ± 4.49 a32.07 ± 8.38 a<0.001C22_6n325.18 ± 3.61 c42.70 ± 7.60 b53.90 ± 9.22 a<0.001Summary indicesTotal fatty acids1369.71 ± 327.73b1589.15 ± 296.91ab1873.36 ± 415.06a0.008SFA451.29 ± 95.74 b508.77 ± 109.94 b631.88 ± 125.52 a0.002MUFA175.62 ± 49.59 b255.83 ± 48.75 a235.16 ± 44.58 a0.002PUFA734.46 ± 186.52 b814.75 ± 144.27 ab997.53 ± 255.26 a0.015n6_PUFA673.74 ± 179.48 b727.20 ± 135.03 ab882.07 ± 239.53 a0.043n3_PUFA60.72 ± 7.91 c87.55 ± 11.20 b115.46 ± 18.49 a<0.001SFA_PUFA_ratio0.62 ± 0.070.62 ± 0.060.64 ± 0.040.691n6_n3_ratio10.96 ± 1.91 a8.29 ± 0.96 b7.57 ± 1.11 b<0.001AI0.26 ± 0.02 ab0.25 ± 0.03 b0.27 ± 0.02 a0.040TI0.71 ± 0.04 a0.65 ± 0.06 b0.68 ± 0.02 ab0.008hH_ratio3.90 ± 0.30 ab4.13 ± 0.54 a3.68 ± 0.31 b0.044These fatty acid indices quantify nutritional quality and health impacts: SFAs (stability), MUFAs/PUFAs (health benefits), and n-6/n-3 + AI/TI ratios (cardiovascular risks)—critical for evaluating duck meat's dietary value and breeding healthier profiles.(i) Saturated Fatty Acids (SFA)=C14:0+C15:0+C16:0+C17:0+C18:0+C20:0;(ii) Monounsaturated Fatty Acids (MUFA) = C16:1n7+C16:1n9+C18:1n9+C18:1n9t+C20:1n9;(iii) Polysaturated Fatty Acids = C18:2n6+C18:3n3+C20:2n6+C20:3n6+C20:4n6+C20:5n3+C22:4n6+C22:5n3+C22:5n6+C22:6n3;(iv) n-6 PUFA = C18:2n6+C20:2n6+C20:3n6+C20:4n6+C22:4n6+C22:5n6;(v) n-3 PUFA = C18:3n3+C20:5n3+C22:5n3+C22:6n3;(vi) SFA/PUFA Ratio =SFA/PUFA;(vii) n-6/n-3 Ratio= ∑n6 PUFA/ ∑n3 PUFA;(viii) Atherogenic Index (AI)= (C14:0+C16:0) / (MUFA+PUFA);(ix) Thrombogenic Index (TI)= (C14:0+C16:0+C18:00) / [(0.5×MUFA) +(0.5×n6 PUFA) +(3×n3 PUFA)];(x) Hypocholesterolemic/Hypercholesterolemic (h/H) Ratio: h/H= (MUFA+PUFA) / (C14:0+C16:0).(Chen and Liu, 2020; Duan et al., 2023). Fatty acid content (µg/g fresh muscle weight) its index composition across breeds. These fatty acid indices quantify nutritional quality and health impacts: SFAs (stability), MUFAs/PUFAs (health benefits), and n-6/n-3 + AI/TI ratios (cardiovascular risks)—critical for evaluating duck meat's dietary value and breeding healthier profiles. (i) Saturated Fatty Acids (SFA)=C14:0+C15:0+C16:0+C17:0+C18:0+C20:0; (ii) Monounsaturated Fatty Acids (MUFA) = C16:1n7+C16:1n9+C18:1n9+C18:1n9t+C20:1n9; (iii) Polysaturated Fatty Acids = C18:2n6+C18:3n3+C20:2n6+C20:3n6+C20:4n6+C20:5n3+C22:4n6+C22:5n3+C22:5n6+C22:6n3; (iv) n-6 PUFA = C18:2n6+C20:2n6+C20:3n6+C20:4n6+C22:4n6+C22:5n6; (v) n-3 PUFA = C18:3n3+C20:5n3+C22:5n3+C22:6n3; (vi) SFA/PUFA Ratio =SFA/PUFA; (vii) n-6/n-3 Ratio= ∑n6 PUFA/ ∑n3 PUFA; (viii) Atherogenic Index (AI)= (C14:0+C16:0) / (MUFA+PUFA); (ix) Thrombogenic Index (TI)= (C14:0+C16:0+C18:00) / [(0.5×MUFA) +(0.5×n6 PUFA) +(3×n3 PUFA)]; (x) Hypocholesterolemic/Hypercholesterolemic (h/H) Ratio: h/H= (MUFA+PUFA) / (C14:0+C16:0). (Chen and Liu, 2020; Duan et al., 2023). Correlation network analysis uncovered breed-specific regulatory relationships between biochemical composition and meat quality attributes (Fig. 3A–F). In MU hybrids, the atherogenic index (AI) positively correlated with both SFA content and meat redness (a*), while the h/H ratio exhibited negative correlation with redness, suggesting coordinated regulation of lipid metabolism and color stability unique to the hybrid. In MD ducks, amino acids drove quality variation: alanine and threonine inversely correlated with pH, while glutamine, serine, asparagine, and glutamic acid negatively associated with lightness (L*). CVD showed the most restricted correlation structure, with glutamine linked to pH, asparagine to redness, and MUFA to lightness. These breed-specific associations suggest distinct regulatory architectures underlying meat quality determination. Correlation networks linking breast-muscle biochemical composition to meat-quality parameters in each breed. Correlations of amino acids (A–C) or fatty acids (D–F) with meat-quality traits (e.g., pH, lightness [L*], redness [a*]) for MU (A, D), CVD (B, E), and MD (C, F). L* = lightness; a* = redness. Metabolomic profiling reveals breed-specific glycerophospholipid metabolism and small molecule diversity Untargeted LC-MS/MS metabolomics on breast muscle confirmed high technical reproducibility across biological replicates (Supplementary Fig. S1). Principal component analysis (PCA) revealed clear metabolomic separation among breeds (Fig. 4A), validated by orthogonal partial least squares-discriminant analysis (OPLS-DA) with robust model performance (Supplementary Fig. S2). Untargeted metabolomic profiling of breast muscle across breeds. (A) Principal component analysis (PCA) of metabolite profiles of MU, CVD, and MD. (B) Venn diagram of shared and unique metabolites among breeds. (C–E) Volcano plots of differentially accumulated metabolites for MD vs. CVD (C), MD vs. MU (D), and CVD vs. MU (E). (F–H) KEGG pathway enrichment of differential metabolites for MD vs. CVD (F), MD vs. MU (G), and CVD vs. MU (H). Differential metabolite analysis identified 208, 80, and 124 unique metabolites in CVD vs. MU, CVD vs. MD, and MD vs. MU comparisons, respectively, with 63 metabolites commonly altered across all contrasts (Fig. 4B; Supplementary Tables S1-S3). This pattern indicates that while breeds share core metabolic signatures, each maintains distinctive metabolomic fingerprints reflecting genotype-specific physiology (Supplementary Fig. S3). In the MD vs. CVD comparison, MD showed upregulation of multiple phosphatidylcholine (PC) and phosphatidylethanolamine (PE) species including PC(O-16:0_16:1), PC(15:0_20:3), PC(O-18:1_22:6), PE(17:0_18:1), alongside ceramide derivatives (Cer(d18:1/15:1(2OH)), HexCer(d18:1/16:0)) and nucleotide sugars (CMP-NANA, UDP-GalNAc). Pathway enrichment analysis identified steroid biosynthesis, glycerophospholipid metabolism, sphingolipid metabolism, taurine and hypotaurine metabolism, glycine/serine/threonine metabolism, and arginine/proline metabolism as significantly enriched (Fig. 4C, F). The CVD vs. MU comparison revealed upregulation in CVD of 2-Hydroxy-2-methylbutanenitrile, Hypaphorine, and d-Valerolactam, with downregulation of multiple PE and PC species containing polyunsaturated fatty acids (PE(P-18:2_22:6), PE(17:0_22:5), PC(O-18:1_22:6)), sphingolipids (Hex2Cer(d18:1/24:0)), and phosphatidylglycerol species (PG(16:1_18:3), PG(18:2_18:3)) (Supplementary Fig. S4). Enriched pathways included ether lipid metabolism, alpha-linolenic acid metabolism, arachidonic acid metabolism, linoleic acid metabolism, autophagy, and glycerophospholipid metabolism (Fig. 4E, H). The MD vs. MU comparison unveiled fundamental differences in membrane lipid homeostasis (Fig. 4D, G). MD showed upregulation of Hypaphorine with downregulation of multiple PE, PC, ceramide, diacylglycerol (DG), and lysophosphatidylcholine (LPC) species. Pathway enrichment identified inositol phosphate metabolism, phosphatidylinositol signaling, fatty acid elongation and degradation, glycerophospholipid metabolism, and arachidonic/linoleic acid metabolism as key distinguishing processes. Cross-comparison pathway analysis identified ether lipid metabolism, glycerophospholipid metabolism, and alpha-linolenic acid metabolism as commonly enriched in both MD vs. MU and CVD vs. MU contrasts, suggesting core metabolic processes responsive to hybrid genetic architecture. Transcriptomic profiling reveals reduced matrisome gene expression in CVD and a hybrid-specific transport and lipid-synthesis signature in MU RNA-seq analysis generated 109.22 GB of clean data with >95% of bases at a Q30 quality score and consistently low singleton rates (<4%; Supplementary Fig. S5). PCA separated the three breeds along PC1 and PC2, which accounted for 35% and 17% of the total variance, respectively, while hierarchical clustering grouped samples by breed, indicating distinct and reproducible transcriptomic profiles (Fig. 5A–B). MU formed a distinct cluster from both parental lines rather than occupying an intermediate position between them. Most of the transcriptome was shared among the three breeds: 20,514 genes (86%) were detected in CVD, MU and MD, whereas 394 (1.6%), 205 (0.86%) and 160 (0.67%) were detected only in CVD, MU and MD, respectively. A further 1,492 genes (6.2%) were detected in CVD and MU but not MD, 941 (3.9%) in MU and MD but not CVD, and 193 (0.81%) in CVD and MD but not MU (Fig. 5C). Baseline transcriptomic profiling of CVD, MU, and MD. (A) PCA of transcriptomic data across the three breeds. (B) Heatmap of gene-expression profiles. (C) Venn diagram showing genes detected in each breed and shared among breeds. (D) Bar plot of up- and down-regulated gene counts per comparison. Pairwise differential expression showed that the two parental lines were the most divergent pair, with 6,041 differentially expressed genes (3,404 upregulated and 2,637 downregulated), whereas both parent–hybrid contrasts involved fewer genes: 2,758 for CVD vs. MU (897 upregulated and 1,861 downregulated) and 2,986 for MD vs. MU (551 upregulated and 2,435 downregulated) (Fig. 5D). In both parent–hybrid contrasts the majority of differentially expressed genes were more abundant in MU. This asymmetry was confined to genes reaching significance; across all expressed genes the median log₂ fold change was approximately zero in every contrast, indicating balanced normalisation. Together with the clustering patterns in Fig. 5A–B, these results show that MU formed a transcriptionally distinct group from both parental lines. Functional enrichment separated the two parental lines along structural and stress-related axes. Genes more highly expressed in MD were enriched for core and associated matrisome components, skeletal system development, cell junction organization and tube development, whereas genes more highly expressed in CVD were enriched for monoatomic cation transport, Class I MHC-mediated antigen processing, response to oxidative stress and negative regulation of apoptosis (Fig. 6A, D). Differential expression and functional enrichment across breed comparisons. (A–C) Volcano plots of differentially expressed genes for CVD vs. MD (A), CVD vs. MU (B) and MD vs. MU (C). (D–F) Functional enrichment of the corresponding gene sets for CVD vs. MD (D), CVD vs. MU (E) and MD vs. MU (F), performed in Metascape across Gene Ontology, Reactome, KEGG, MSigDB canonical pathways including the NABA matrisome collections, and WikiPathways. Bars show the twenty most significant terms ranked by −log₁₀(P). The CVD vs. MU comparison reproduced the matrisome signature with the same orientation relative to CVD. Extracellular matrix organization, ECM–receptor interaction, NABA MATRISOME ASSOCIATED and NABA COLLAGENS were enriched among genes higher in MU, including COL4A6, COL28A1, MATN1, MATN4 and ADAMTSL3, together with trans-synaptic signalling and tube development (Fig. 6B, E). Matrisome terms were therefore enriched in both comparisons involving CVD but absent from the MD vs. MU contrast, indicating that reduced expression of matrix-related genes is a feature of CVD rather than a property specific to either of the other genotypes. The MD vs. MU comparison was dominated by transport and biosynthetic processes, all oriented toward MU. Monoatomic cation transmembrane transport, SLC-mediated transmembrane transport and nitrogen compound transport were enriched, with member genes including SLC38A6, SLC30A3, SLC39A4 and SLC2A2, alongside nucleotide biosynthetic process and negative regulation of apoptosis (Fig. 6C and F). Two terms link this contrast to the metabolomic data. Mitochondrion organization comprised OPA1, MIEF2, SPG7, MSTO1 and the mitochondrially encoded ND1, ND2 and ND5. Neutral lipid biosynthetic process comprised GPAM, AGPAT3, LPIN3, PNPLA2 and TMEM68, which encode consecutive steps of the glycerol-3-phosphate pathway generating phosphatidic acid and diacylglycerol, the precursors of the phosphatidylcholine and phosphatidylethanolamine species that distinguished the breeds metabolically. Gene set enrichment analysis of the complete ranked gene lists identified 171 significant gene sets across the three comparisons, and resolved each contrast along a distinct functional axis (Fig.S6-8). Matrix-related processes separated CVD from both other genotypes. Extracellular matrix organization and collagen fibril organization were enriched among genes more highly expressed in MD than CVD (NES = −1.69, adjusted P = 5.1 × 10⁻³ and NES = −1.90, adjusted P = 0.034), and the same two sets, together with external encapsulating structure organization and extracellular structure organization, were enriched among genes more highly expressed in MU than CVD (NES = −1.95, adjusted P = 2.8 × 10⁻⁴). No matrix-related set distinguished MD from MU. In the opposite direction, CVD was characterised by an expanded ribosome biogenesis and RNA processing programme: ribonucleoprotein complex biogenesis (NES = 2.13, adjusted P = 4.8 × 10⁻¹⁴), rRNA metabolic process, ribosome biogenesis, nuclear export and DNA-templated transcription elongation were all enriched relative to MU, with nuclear export, the ERAD pathway and mRNA transport enriched relative to MD. This reciprocal arrangement (greater capacity for ribosome assembly and protein processing in CVD, greater matrix gene expression in MD and MU) accompanies the highest dressing percentage and Pectoralis major proportion recorded in CVD. Oxidative metabolism distinguished the hybrid from the Muscovy parent. Oxidative phosphorylation (NES = −1.73, adjusted P = 0.048), aerobic respiration (NES = −1.68), cellular respiration (NES = −1.63) and energy derivation by oxidation of organic compounds (NES = −1.59) were enriched among genes more highly expressed in MU, while cytoplasmic translation was enriched in MD. Very long-chain fatty acid metabolic process was similarly enriched in MU relative to CVD (NES = −1.95, adjusted P = 0.027). Integrated multi-omics analysis defines breed-specific regulatory architectures Joint pathway analysis of differentially expressed genes and differentially abundant metabolites identified the pathways separating each pair of genotypes (Figs. 7A–C and Fig. 8A–C), and gene set enrichment analysis on the ranked gene lists established the direction of each pathway (Fig. S9; Supplementary Table S10). Integrated KEGG pathway enrichment of differentially expressed genes (DEGs) and differentially expressed metabolites (DEMs). Joint pathway enrichment for CVD vs. MD (A), CVD vs. MU (B), and MD vs. MU (C). Bar length and colour represent −log₁₀(P), with higher values indicating greater statistical significance. The dashed vertical line indicates P = 0.05. Significance markers denote the false discovery rate after Benjamini–Hochberg correction: ns, not significant; *, FDR < 0.05; **, FDR < 0.01; ***, FDR < 0.001. Integrated gene–metabolite networks across breed comparisons. Networks integrating DEG and DEM for CVD vs. MD (A), CVD vs. MU (B), and MD vs. MU (C). The proteasome pathway was higher in CVD than in either of the other genotypes (NES = 1.84, adjusted P = 0.008 against MD; NES = 2.12, adjusted P < 0.001 against MU). Every proteasome gene that reached significance was more abundant in CVD: 31 of 31 in the CVD vs. MD comparison and 29 of 29 in CVD vs. MU, including the regulatory subunits PSME1 (log₂FC = 5.21), PSME4, PSMD3, PSMD6, PSMD11 and the catalytic subunits PSMA4, PSMB1, PSMB3 and PSMB5. mTOR signalling followed the same direction, being higher in CVD in both comparisons (NES = 1.56, adjusted P = 0.035 against MD; NES = 1.59, adjusted P = 0.016 against MU), with ATP6V1F, ATP6V1G3, CASTOR2, STRADB, TBC1D7, FNIP1 and PIK3R2 among the genes elevated in CVD. Collagen and matrix genes ran in the opposite direction. Protein digestion and absorption, a pathway populated in KEGG mainly by collagens, was higher in MD than CVD (NES = −1.86, adjusted P = 0.008; 21 of 24 genes) and higher in MU than CVD (NES = −2.04, adjusted P = 0.002; all 19 genes), driven by COL4A6, COL18A1, COL28A1, COL3A1, COL5A2 and COL9A3, together with the amino acid transporter SLC7A9 and the sodium–hydrogen exchanger SLC9A3. In the CVD vs. MU comparison, ECM–receptor interaction (NES = −1.85, adjusted P = 0.032; 18 of 20 genes) and cytoskeleton in muscle cells (NES = −1.75, adjusted P = 0.010; 35 of 41 genes) were likewise higher in MU, involving COL4A5, COL4A6, CHAD, FREM1, FREM2, ITGA11, ITGB8, LAMA5, AGRN, FHL2 and SSPN. Two genes ran counter to this pattern and were higher in CVD: CD36 and TRIM63. Nutrient transport and carbohydrate handling separated MU from CVD in the joint analysis. ABC transporters (24 hits, P = 9.9 × 10⁻¹⁰, FDR = 3.5 × 10⁻⁷), protein digestion and absorption (17 hits, FDR = 3.1 × 10⁻⁴), fat digestion and absorption (8 hits, FDR = 1.4 × 10⁻²), starch and sucrose metabolism (9 hits, FDR = 2.2 × 10⁻², impact = 0.697) and galactose metabolism (9 hits, FDR = 2.2 × 10⁻²) were enriched, supported by ABCG5, ABCB5, MTTP, CD36, MOGAT2, CEL, SI and GYS2 on the gene side and by free amino acids (L-valine, L‑serine, l-phenylalanine, l-methionine, l-isoleucine) and sugars (maltose, maltotriose, isomaltotriose, raffinose, dextrin, glucose, glucose-6-phosphate, fructose-6-phosphate) on the metabolite side (Fig. 8B). cAMP signalling was higher in MU than CVD (NES = −1.67, adjusted P = 0.010; 30 of 39 genes). No pathway beyond the receptor-interaction term reached significance in the MD vs. MU joint analysis, indicating that the hybrid and the Muscovy parent differ least at the level of these pathways. Neuroactive ligand–receptor interaction was also higher in MD and MU than in CVD (NES = −1.54, adjusted P = 0.013 and NES = −1.69, adjusted P = 0.007), with no difference between MD and MU. This is the largest pathway tested (224 mapped genes), and the contributing transcripts encode peptide hormone and neurotransmitter receptors (SSTR5, GPR83, GRIN1, GRIA2, CCKAR, NTSR1, TACR3) whose relevance to skeletal muscle is unclear. The main differences among the three genotypes are summarised in Fig. 9. Summary of the muscle characteristics distinguishing Cherry Valley duck (CVD), Muscovy duck (MD) and their mule hybrid (MU), and of how the hybrid relates to each parental line. Muscle cross-sections illustrate the difference in intramuscular connective tissue implied by the collagen and extracellular matrix gene expression data, with CVD showing the lowest expression of these genes and MD and MU the highest. Arrows indicate higher (↑) or lower (↓) gene expression between genotypes; no measurement of enzyme activity, protein abundance or metabolic flux was performed, and the symbols denote transcript abundance only. CVD carried higher expression of proteasome subunit genes, spanning the 20S core, the 19S regulatory particle and the PA28 and PA200 activators, together with higher expression of mTOR pathway genes and enrichment of ribosome biogenesis gene sets, and lower expression of collagen and extracellular matrix genes. MD showed the opposite pattern for both processes and carried the highest concentrations of free glutamine, asparagine, serine, glycine and lysine. The broad arrow from MD to MU denotes the traits for which the two did not differ, namely matrix and protein turnover gene expression. Characteristics of MU that fell outside the range of both parental lines are shown separately: free arginine approximately two-fold higher than in either parent without a change in total free amino acid content, higher GAMT expression than both parents, higher expression of oxidative phosphorylation and cellular respiration gene sets than MD, and higher expression of transport and absorption genes than CVD. MU is positioned off the axis joining the two parental lines to indicate that its molecular profile was not intermediate between them. Connecting lines denote association between the observations shown and do not imply causal relationships. Discussion Distinct muscle molecular profiles among CVD, MD and MU Cherry Valley duck (CVD, Anas platyrhynchos) and Muscovy duck (MD, Cairina moschata) differ in genetic background and in selection history, and the mule duck (MU) is the intergeneric hybrid obtained by mating Muscovy drakes with Cherry Valley females. Reared together on a common diet, the three groups differed in muscle gene expression, in free amino acid composition and in lipid composition, and the pattern of those differences was not the same across the three data layers. CVD was separated from MD and MU by higher expression of genes associated with protein turnover and lower expression of extracellular matrix genes. MD and MU did not differ at pathway level for either process, although four proteasome genes were more highly expressed in MU than in MD: PSMA4, encoding a 20S α subunit, and PSME1, PSME3 and PSME4, encoding the proteasome activators PA28α, PA28γ and PA200. MU nonetheless did not occupy a position between its parental lines. It clustered apart from both, showed broader pathway differences relative to CVD than to MD, and carried free arginine, oxidative metabolism gene expression and transport gene expression that set it apart from MD as well. The sections that follow take these three profiles in turn. As CVD and MD differ simultaneously in species and in selection history, the contribution of each cannot be separated in this design, and the parental contrast is interpreted here as a combined effect. Protein turnover and extracellular matrix gene expression separate CVD from MD and MU The transcriptional difference between CVD and the two other groups was not confined to one side of protein turnover. CVD showed higher expression of proteasome subunit genes, of mTOR pathway genes, and of gene sets for ribosome biogenesis and RNA processing. Genes encoding the 20S proteolytic core, the 19S regulatory particle and the proteasome activators PA28 and PA200 were all more highly expressed, and every proteasome gene reaching significance was higher in CVD, in both the CVD versus MD comparison (31 genes) and the CVD versus MU comparison (29 genes). The commercially selected line therefore showed greater representation of the transcriptional programmes associated with both protein synthesis and protein degradation, which is a different observation from an increase in catabolic gene expression alone. The mTOR component agrees with earlier work in poultry lines differing in growth rate. Cai et al. (2026) reported enrichment of the mTOR pathway in breast muscle of commercial broilers compared with an indigenous breed, Wen et al. (2017) found the phosphorylation response of mTOR to dietary methionine to be strain-dependent in fast- and slow-growing broilers, and satellite cells from modern commercial chicken lines were more dependent on mTOR signalling during proliferation than those from historic lines (Xu and Velleman, 2023a). The proteasome component differs from several earlier reports. Lower expression of atrogin-1 and of the 20S C2 subunit has been described in broilers compared with layers (Nakashima et al., 2009; Saneyasu et al., 2015), and reduced atrogin-1 expression has been linked to selection in commercial broilers (Li et al., 2019). Those studies measured atrophy-associated E3 ubiquitin ligases together with a small number of proteasome subunits, and the two do not respond in parallel. Food deprivation in broilers raised atrogin-1 expression markedly while expression of the 20S C1 and C2 subunits tended to fall (Ohtsuka et al., 2011). The present result describes the subunit complement across the whole particle rather than the atrophy response, and PSMC3, one of the 19S subunits identified here, was previously found to be more abundant in broiler than in layer breast muscle (Harper et al., 1999). At the protein level, ubiquitin-proteasome proteins were associated with greater muscle mass in fast-growing broilers (Di Luca et al., 2024), and ubiquitin-mediated proteolysis was enriched in a gene co-expression module correlated with breast width (Bordini et al., 2022). TRIM63 (MuRF1) was also more highly expressed in CVD, which does not follow the reduced ligase expression reported in selected broilers, and indicates that the components of the ubiquitin-proteasome system did not move as a single unit in this comparison. Proteasomal activity, ubiquitin conjugation and protein abundance were not measured, so the transcript data describe the expression of these components and not the rate at which protein is degraded. Collagen and matrix gene expression ran in the opposite direction, being higher in MD and MU than in CVD and not differing between MD and MU. Chemical measurements in duck follow the same ordering. Huo et al. (2021) found higher breast collagen content in slow-growing Liancheng White ducks than in Cherry Valley ducks. Huang et al. (2023) reported the lowest breast collagen content in Cherry Valley among three duck genotypes with a Cherry Valley crossbred intermediate. Furthermore, Wegner et al. (2026) found a significant genotype effect on intramuscular collagen in a direct Pekin-Muscovy comparison. Collagen synthesis genes were downregulated in broiler breast muscle with lower shear force (Piórkowska et al., 2016), and reduced connective tissue spacing accompanies myofibre hypertrophy in fast-growing lines (Xu and Velleman, 2023b). The transcript data reported here identify the genes underlying a collagen phenotype already described in duck and show that the same ordering extends to the hybrid. This holds for muscle without myopathy: in broilers affected by wooden breast, intramuscular connective tissue accumulates, with thickened perimysium and higher insoluble collagen (Zhu et al., 2023; Bian et al., 2024). Because the relationship between collagen content and tenderness is inconsistent outside beef (Purslow, 2018; Li et al., 2022), the transcript result is not extended to a prediction about texture. Hybridisation is associated with a molecular profile that is not intermediate between the parental lines MU did not simply lie between its parents. It retained the MD-like expression pattern for matrix and protein turnover genes, clustered separately from both parental lines, and differed from MD in oxidative metabolism gene expression, in free arginine concentration and in transport-associated gene expression. The majority of differentially expressed genes in both parent-hybrid comparisons were more abundant in the hybrid. Departures from parental expression have been described in other hybrids. Expression exceeding both parental values occurs in early-generation hybrids (Runemark et al., 2025), has been reported in naturally occurring avian interspecific hybrids across somatic tissues (Mugal et al., 2020), and accounted for more than two-thirds of differentially expressed genes in breast muscle of chicken reciprocal crosses (Mai et al., 2021). The present pairwise design does not assign genes to additive, dominant or transgressive classes, and that classification against mid-parent values would be needed before the inheritance mode is named. Additivity has been reported as the predominant pattern in intraspecific chicken crosses, with more than 80% of genes at mid-parent values in post-hatch F1 breast muscle (Zhao et al., 2024) and additivity predominant in F1 embryonic tissues (Zhuo et al., 2019). Greater divergence between the parental populations offers one explanation for the different pattern observed here, since MU is an intergeneric hybrid. The phenotypic literature on this cross records intermediacy for muscle weight with significant additive effects for growth and heterosis confined to foie gras weight and muscle lightness (Larzul et al., 2006), and the morphometric measurements reported here are consistent with that. Transcriptional differentiation and phenotypic intermediacy are not incompatible. At pathway level MU resembled MD more closely than CVD. Quantitative-genetic analyses of the same cross point the other way, attributing more of the hybrid’s feeding and meat traits to the common duck parent than to the Muscovy parent (Chapuis et al., 2024) and reporting greater genetic determinism on the common duck side (Marie-Etancelin et al., 2011). Pathway-level similarity in gene expression and the heritability of a production trait are different measurements, and the present data do not resolve the difference. Gene sets for oxidative phosphorylation and cellular respiration were more highly expressed in MU than in MD. Differences in oxidative phosphorylation-associated expression have also been reported in avian crosses. These include chicken breast muscle, where non-additive expression was accompanied by differences in ATP content (Mai et al., 2021), and other chicken tissues showing differential expression of oxidative phosphorylation-related genes (Yuan et al., 2023). In ducks, Zhang et al. (2023) identified oxidative phosphorylation among the enriched pathways in breast muscle transcriptomes from parental lines and their reciprocal crosses. The direction observed here differs from the growth-related pattern reported by Fan et al. (2020), who found higher expression of oxidative phosphorylation-related genes in mallard than in Pekin duck. The most pronounced single difference in the hybrid was in free arginine, which was approximately twice as abundant in MU as in CVD or MD and accounted for a larger share of the free amino acid pool, while total free amino acid content did not differ among the three groups. The arginine difference was therefore not part of a general increase in free amino acids. Birds lack carbamoyl phosphate synthetase I and cannot synthesise arginine de novo, so the muscle pool depends on intake, cationic amino acid transport, release from protein degradation, and consumption by nitric oxide synthase, arginase, arginine decarboxylase and arginine:glycine amidinotransferase (Khajali and Wideman, 2010; Fathima et al., 2024). Neither arginase nor nitric oxide synthase transcripts differed among groups, but avian arginase activity is principally renal and responds to the arginine-lysine antagonism instead of to muscle transcription (Smith and Lewis, 1966; Fernandes and Murakami, 2010), so muscle transcript abundance places no constraint on catabolic flux. GAMT was more highly expressed in MU than in both parents. GAMT acts downstream of the arginine-consuming amidinotransferase reaction and converts guanidinoacetate to creatine (Portocarero and Braun, 2021), so its higher expression indicates differentiation of the creatine synthesis pathway and does not by itself account for the arginine difference; measurement of GATM, guanidinoacetate and creatine would be required for that. Higher muscle arginine in MU coincided with the lowest expression of mTOR and proteasome genes, and CVD showed the opposite combination. Dietary l-arginine increased breast and thigh muscle mass in chickens while decreasing protein synthesis rate, phosphorylated mTOR and atrophy-associated ligase expression, in a manner specific to fast-twitch glycolytic muscle (Wang et al., 2022). That study manipulated dietary arginine within one genotype, whereas the present observation is a constitutive difference among genetic groups, so the correspondence between the two patterns is descriptive and does not indicate that arginine concentration determines the expression of these genes. Arginine content has previously been reported to differ between duck lines, being higher in slow-growing than in fast-growing birds from a Cherry Valley cross (Li et al., 2024); we did not find a comparable report of a hybrid exceeding both parental lines, although the literature on mule duck muscle composition is limited. Transport and absorption pathways also separated MU from CVD in the integrated analysis. The KEGG categories concerned are annotated around intestinal and hepatic physiology, and their appearance in a muscle dataset reflects the shared membership of transport and hydrolase genes, so the individual genes rather than the pathway names support the interpretation. CD36 has the clearest evidence in this context: it differed between broiler and Chinese Silky fowl in pectoralis muscle and selectively transports linoleic and arachidonic acid (Guo et al., 2013), is transcriptionally regulated by RXRA in duck myoblasts, where its overexpression reduced triglyceride and lipid droplet accumulation (Pan et al., 2023), and was among the differentially expressed genes associated with myofibre development in black Muscovy duck breast muscle (Zhang et al., 2024a). Solute carrier expression in duck breast muscle varies by breed and covaries with amino acid content (Zhang et al., 2023a), which connects the carriers identified here to the arginine result, although the specific transporters concerned have not been reported in avian skeletal muscle. Several carbohydrates mapped to these pathways, including raffinose and the malto-oligosaccharides, have no endogenous route in avian muscle and are more likely to be feed-derived or to arise from post-mortem glycogen degradation. Amino acid and lipid composition differentiate the three genetic groups MD carried the highest concentrations of glutamine, asparagine, serine, glycine and lysine. Higher free and flavour-related amino acid concentrations in indigenous and slow-growing genotypes than in commercial lines have been reported in chicken (Dalle Zotte et al., 2020; Gu et al., 2024) and in duck (Li et al., 2024), although Yang et al. (2025) found several amino acids higher in Cherry Valley than in two local duck breeds. Taurine was highest in CVD, which matches the observation of Dalle Zotte et al. (2020) that taurine was richest in a commercial broiler hybrid and undetectable in an Italian purebred, a result those authors were unable to explain. The free amino acid ordering did not follow the expression of proteasome-associated genes. Watanabe et al. (2020) found free glutamate concentration correlated with 20S proteasome C2 subunit and calpain expression in chicken breast muscle under dietary lysine restriction, which would predict the highest free amino acid concentrations in CVD. Their comparison was made within one genotype under a nutritional perturbation, whereas the present comparison is between genetic groups on a common diet. Steady-state free amino acid concentration integrates release, uptake, reutilisation and catabolism, and cannot be read from the expression of degradation-associated genes alone. CVD had the highest thrombogenic index and ω−6/ω−3 ratio. Lower atherogenic and thrombogenic indices were reported in native duck breast than in Pekin (Onk et al., 2019), and the least favourable PUFA/SFA ratio was found in a commercial goose cross compared with local varieties (Wołoszyn et al., 2020). MU had the highest absolute concentrations of total fatty acids and of the SFA, PUFA, n‑6, and n‑3 classes compared with either parent, as also reported in a Cherry Valley * Chinese crested crossbred (Huang et al., 2023). Absolute concentrations follow total intramuscular lipid content, so the proportional composition carries more information about the fatty acid profile than the absolute values. Phosphatidylcholine and phosphatidylethanolamine species separated the three groups and glycerophospholipid metabolism was enriched among the differential metabolites, but the pathway was not enriched in the integrated analysis and the genes of the de novo route did not differ in a coordinated way. Metabolite-level differentiation of this pathway has been reported in duck (Hu et al., 2026), goose (Cao et al., 2024) and chicken muscle (Zhang et al., 2022), whereas coordinated transcript-level differences have been described less often (Zhu et al., 2024). Membrane phospholipid composition depends substantially on acyl chain remodelling and on dietary fatty acid supply, neither of which requires a change in expression of the biosynthetic pathway, and phospholipids are hydrolysed after slaughter, with phosphatidylcholine and phosphatidylethanolamine falling and lysophospholipids rising during chilled storage of chicken breast (Lv et al., 2023). Part of the phospholipid signal therefore reflects post-mortem change rather than differences present in the live bird. GPAM, AGPAT3 and LPIN3, which encode consecutive steps of the glycerol-3-phosphate route, were more highly expressed in MU. These gene families are transcriptionally regulated in chicken (Yang et al., 2019), and GPAM influences intramuscular adipocyte differentiation in chicken (Li et al., 2019b). PNPLA2 was elevated alongside them but encodes a lipase whose abundance correlated negatively with intramuscular fat content in cattle (Jeong et al., 2012), so it does not belong to the same interpretation. Membrane lipid composition remains relevant to meat quality independently of how it is transcriptionally controlled: reduced phospholipid abundance accompanied increased neutral lipid and altered flavour precursors in broiler breast muscle (Kong et al., 2023), and phospholipid-bound polyunsaturated fatty acids are the principal substrates for the oxidative reactions that generate volatile compounds during cooking (Zhang et al., 2024). MU had the highest absolute fatty acid concentrations without the most favourable lipid indices, and MD combined higher phosphatidylcholine and phosphatidylethanolamine abundance with a lower thrombogenic index and a higher h/H ratio. Overall interpretation, limitations and implications Three molecular profiles emerge from the combined data. CVD, the line under long-term selection for breast yield, showed the highest expression of genes associated with both arms of protein turnover, the lowest expression of matrix and collagen genes, the lowest concentrations of the free amino acids associated with flavour, and the least favourable thrombogenic index and ω−6/ω−3 ratio. MD occupied the opposite position for matrix gene expression and free amino acid composition and had the most favourable proportional lipid indices. MU shared the MD profile for matrix and protein turnover gene expression, and was distinguished from both parental lines by its free arginine concentration, by higher oxidative metabolism gene expression than MD, and by the transport-associated genes separating it from CVD. The hybrid therefore combined features inherited in a parent-like manner with features found in neither parent, and the two sets belong to different biological processes. Structural and protein-related transcription in MU followed the Muscovy pattern, whereas amino acid, oxidative metabolism and transport characteristics did not follow either parent. That separation is the most useful outcome of comparing a hybrid with both of its parental lines under a common environment, and it identifies the processes in which hybridisation, rather than parental origin, accounts for the molecular phenotype of mule duck breast muscle. For meat quality, the practical consequence is that the traits favoured by selection in CVD and the traits associated with composition in MD and MU are supported by different molecular processes. Higher breast yield in CVD coincided with the transcriptional profile of high protein turnover and reduced matrix gene expression, while the amino acid and lipid characteristics associated with flavour and nutritional value were more pronounced in MD and MU. The hybrid retained the compositional characteristics of the Muscovy parent while carrying its own arginine and oxidative metabolism profile, which is relevant to how mule duck meat is positioned relative to both parental products. Several limitations bound these interpretations. Enzyme activity, protein abundance and phosphorylation were not measured, so the transcript and pathway results describe differences in gene expression and not in metabolic flux; this applies to the proteasome, mTOR, oxidative phosphorylation and lipid biosynthesis results alike. Muscle was sampled at a single time point after slaughter, which limits interpretation of the metabolites subject to post-mortem change, particularly the phospholipids. Inheritance modes were not classified against mid-parent values, and doing so would allow the hybrid expression pattern to be compared directly with published crossbreeding transcriptomes. Reads from all three groups were aligned to a single Anas platyrhynchos reference, and mapping efficiency can differ among the genetic types of this cross (Liu et al., 2020), so results with a uniform direction across a gene set should be considered with that in mind. TAOK1, FZD1, MATN4, SLC35C1 and TMEM68 have no reported expression in avian skeletal muscle and are identified here as candidates for further work. Conclusion The three duck genotypes showed distinct muscle molecular and compositional profiles under common rearing conditions. CVD was characterized by higher expression of genes associated with protein turnover and lower matrix gene expression, while MD showed a distinct amino acid and lipid profile. MU did not display a simple intermediate profile between its parental lines, combining MD-like transcriptional features with distinct patterns in arginine concentration and oxidative metabolism. These findings show that parental background and hybridisation are associated with different components of breast muscle phenotype. As the molecular data represent a single post-mortem time point without direct measurements of enzyme activity or metabolic flux, functional studies are required to determine how these molecular differences contribute to duck meat quality. Funding This study was supported by National Key Research and Development Program of China (2023YFD1300300), Hainan Provincial Key R&D Program (ZDYF2024XDNY158, ZDYF2024YJGG003-5), Chinese Modern Technology System of Agricultural Industry (CARS-42-50, CARS-42-2), and Central Public-interest Scientific Institution Basal Research Fund (NO. 1630032025011). Data availability The datasets generated in this study are publicly available. Transcriptomic data have been deposited in the Genome Sequence Archive (GSA) under accession number CRA040079 and can be accessed via https://ngdc.cncb.ac.cn/gsa/browse/CRA040079 . Metabolomic data are available in the OMIX repository under accession number OMIX015702, accessible via https://ngdc.cncb.ac.cn/omix/release/OMIX015702 . The integrated multi-omics dataset, including the data underlying the analyses reported in this study, has also been deposited in Zenodo and is available at https://doi.org/10.5281/zenodo.22021238. The complete differential-analysis results for differentially accumulated metabolites and differentially expressed genes are provided in Supplementary Tables S1–S6. Ethical statement This study was approved by the Institutional Animal Care and Use Committee at the Experimental Animal Center of Hainan Academy of Agricultural Science (HNXMSY 20210743). All procedures involving animals were conducted in strict accordance with the Regulations for the Administration of Affairs Concerning Experimental Animals (Ministry of Science and Technology, China, revised in March 2017), as well as in compliance with international ethical guidelines, including the Animals (Scientific Procedures) Act 1986, EU Directive 2010/63/EU on the protection of animals used for scientific purposes. Author contributions Lionel Kinkpe: Conception and design of the study, methodology, investigation, data analysis and curation, and preparation of the original manuscript draft. Lihong Gu and Shun Guo: Conception and design of the study, methodology, investigation, and preparation of the original manuscript draft. Zixin Xu and Zhigang Hu: Data analysis and curation, and preparation of the original manuscript draft. Chengjun Hu: Investigation and preparation of the original manuscript draft. Yaya Sun: Investigation, data curation, and preparation of the original manuscript draft. Dingping Bai: Methodology and preparation of the original manuscript draft. Xinli Zheng: Data analysis and preparation of the original manuscript draft. Tieshan Xu and Xia Wang: Conception and design of the study, methodology, investigation, data analysis and curation, and preparation of the original manuscript draft. All authors made substantial contributions to the work, critically reviewed the manuscript for important intellectual content, and reviewed and explicitly approved the final version of the manuscript. All authors agree to be personally and publicly accountable for the work and to ensure that any questions related to the accuracy or integrity of any part of the research are appropriately investigated and resolved. Disclosures The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgments We would also like to thank the computing platform support from High-Performance Computing of NWAFU. Footnotes Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.psj.2026.107719. Contributor Information Tieshan Xu, Email: xutieshan760412@163.com. Xia Wang, Email: xiawang@nwafu.edu.cn. Supplementary materials References Associated Data Supplementary Materials Data Availability Statement The datasets generated in this study are publicly available. Transcriptomic data have been deposited in the Genome Sequence Archive (GSA) under accession number CRA040079 and can be accessed via https://ngdc.cncb.ac.cn/gsa/browse/CRA040079 . Metabolomic data are available in the OMIX repository under accession number OMIX015702, accessible via https://ngdc.cncb.ac.cn/omix/release/OMIX015702 . The integrated multi-omics dataset, including the data underlying the analyses reported in this study, has also been deposited in Zenodo and is available at https://doi.org/10.5281/zenodo.22021238. The complete differential-analysis results for differentially accumulated metabolites and differentially expressed genes are provided in Supplementary Tables S1–S6.