Integrated multi-tissue transcriptome and serum metabolome analysis reveals brain–gut–liver regulatory axes of residual feed intake in ducks

Abstract

Residual feed intake (RFI) is a key indicator of feed efficiency in poultry. Although regulatory links such as the hypothalamus–gut and gut–liver axes have been implicated, most studies remain restricted to single axes or fragmented analyses, and systematic multi-organ integration is lacking. Here, we measured feed efficiency in 1,000 Nonghua ducks and selected 12 individuals with divergent RFI for transcriptomic profiling of the hypothalamus, pituitary, liver, duodenum, jejunum, ileum, and cecum, combined with serum metabolomics. We identified 769 differentially expressed genes (DEGs), with the hypothalamus, liver, and cecum as major contributors, and 28 differential serum metabolites enriched in lipid and amino acid metabolism. Beyond tissue-specific functions, enrichment analysis highlighted several pathways that were repeatedly shared across central and peripheral tissues, including neuroactive ligand–receptor interaction, hormone signaling, steroid hormone biosynthesis, and insulin signaling, suggesting a coordinated regulation of feed efficiency between the brain, gut, and liver. To clarify their relevance, we integrated gene modules with metabolites and identified two candidate cross-organ association frameworks: the MEblack–6-Oxopiperidine-2-carboxylic acid (gut–liver) networks, enriched for liver genes CNTNAP1, SHC3, and RAB36, and cecal genes DCC and CCDC60. The MEblue–LysoPE(18:2(9Z,12Z)/0:0) (gut–brain) networks, enriched for cecal genes FABP6, KCNJ11, and the pituitary gene TRPA1, in which these genes and metabolites may contribute to RFI regulation. Together, these findings provide new insights into cross-organ molecular networks underlying feed efficiency in ducks and establish a valuable resource for future functional studies and breeding strategies.

Introduction

Feed accounts for over 70% of poultry production costs, making feed efficiency (FE) a key determinant of farming profitability and sustainability. Residual feed intake (RFI), defined as the difference between actual feed intake and the predicted intake required for maintenance and growth (Koch et al., 1963), has become a widely accepted indicator for improving FE without compromising growth performance (Lefaucheur et al., 2011; Mondin et al., 2024; Wu et al., 2025). This is largely attributed to its moderate to high heritability (Bezerra et al., 2013). However, despite its importance in breeding programs, RFI is often treated solely as a statistical phenotype, with limited attention given to the underlying biological processes that drive it. Such a phenotype-driven approach, while useful to some extent, fails to uncover the molecular drivers and regulatory pathways responsible for individual differences in feed utilization. Consequently, opportunities to identify functional biomarkers or to enhance the precision of genomic selection remain underexploited. A deeper understanding of the molecular and physiological mechanisms that govern RFI is therefore essential for translating this trait into more predictive and mechanistically informed breeding strategies.

RFI arises from the coordinated actions of multiple organs, each playing distinct roles in regulating feed efficiency. The hypothalamus and pituitary act as central neuroendocrine hubs that integrate internal and external signals to coordinate feeding behavior, energy balance, and hormonal outputs (Waterson and Horvath, 2015; Luquet et al., 2019). Classical lesion studies demonstrated that specific hypothalamic nuclei control feeding responses, with targeted damage leading to either hyperphagia or aphagia (Anand and Brobeck, 1951; Hetherington and Ranson, 1983). Peripheral organs are equally important: the gastrointestinal tract functions as the primary site for nutrient sensing and absorption, while the liver serves as a central hub for nutrient conversion and metabolic feedback (Krogdahl, 1985; Saha and Pathak, 2021). In poultry, gut-derived hormones such as GLP-1 and PYY transmit satiety signals to the brain (Lee, 2023), whereas liver-secreted factors including FGF21 and GDF15 act on the central nervous system to regulate feeding behavior and energy homeostasis (Jena et al., 2023). These findings demonstrate that both central and peripheral organs play key roles in RFI regulation.

Nevertheless, single-organ functions alone are insufficient to explain feed efficiency fully. Increasing evidence highlights the importance of cross-tissue communication, in which central and peripheral organs interact through neural and endocrine pathways to form systemic feedback loops. For example, activation of hypothalamic AgRP neurons has been shown to increase intestinal glucose absorption via autonomic outputs (Steculorum et al., 2016), while hypothalamic signals can also modulate lipid metabolism in adipose tissue, feeding back to influence energy expenditure and feed intake (Joly-Amado et al., 2012). Conversely, peripheral tissues actively communicate with the brain: gut-derived hormones GLP-1 and PYY act on hypothalamic receptors to suppress appetite (Batterham et al., 2003); nutrient-induced gut signals stimulate hepatic secretion of FGF21, which crosses the blood–brain barrier to act on hypothalamic circuits and promote energy expenditure (Liang et al., 2014); and liver-derived GDF15 signals via the brainstem receptor GFRAL to reduce food intake (Mullican et al., 2017). Moreover, gut microbiota-derived short-chain fatty acids (SCFAs) not only regulate intestinal nutrient absorption but also activate GPR41/GPR43 signaling in the liver, thereby affecting glucose and lipid metabolism and feeding back to the hypothalamus (Potthoff et al., 2011; Canfora et al., 2019). These findings reveal a complex integrative signaling network among the brain, gut, liver, and adipose tissue, indicating that RFI is fundamentally shaped by cross-organ regulation. Although previous studies have identified regulatory links such as the hypothalamus–gut and gut–liver axes, these efforts have largely been limited to single axes or local pathways. Comprehensive, multi-organ, integrative studies remain scarce, restricting our comprehensive understanding of the molecular mechanisms underlying the cross-tissue regulation of RFI. This highlights the need for multi-organ and multi-omics approaches to characterize the cross-tissue regulatory networks that shape feed efficiency systematically.

Based on this rationale, the present study aims to systematically investigate the molecular basis of RFI from a cross-tissue perspective in ducks. Using a large-scale feed efficiency trial of 1,000 Nonghua ducks, individuals with divergent RFI phenotypes were selected for multi-omics analyses. Transcriptome sequencing was conducted across seven key tissues (hypothalamus, pituitary, liver, duodenum, jejunum, ileum, and cecum), combined with untargeted serum metabolomics. Weighted gene co-expression network analysis (WGCNA) and metabolite–trait correlation analyses were performed to construct module (gene)–metabolite–RFI regulatory frameworks and to assess the contributions of different tissues within cross-organ networks. This integrative approach will provide new insights into the hierarchical cross-tissue regulation of feed efficiency and generate a valuable multi-organ, multi-omics resource to support precision molecular breeding in waterfowl.

Materials and methods

Ethics statement

Experimental procedures and animal welfare practices were conducted in accordance with the Guidance on the Humane Treatment of Laboratory Animals. All protocols were reviewed and approved by the Animal Ethics Committee of Sichuan Agricultural University (Approval ID: DKY20170913).

Management of animals and experimental diet

A total of 1,000 Nonghua ducks were selected at 14 days of age and reared in the Waterfowl Performance Testing House at the Teaching and Research Base of Sichuan Agricultural University (Ya'an, China) until 43 days of age. Ducks were floor-reared with free access to feed and water. The experimental diet met the nutritional requirements for meat duck growth, and its composition and chemical analysis are presented in Supplementary Table S1.

Housing conditions and automated feeding and weighing system

All ducks were housed in the same flat-floor duck house under identical environmental and management conditions. The rearing area was functionally divided into a drinking area and a feeding area. Ducks were required to pass through an automated body-weight recording corridor when moving from the drinking area to the feeding area. During each passage, individual ducks were identified using unique RFID wing tags, and body weight was automatically recorded. Movement from the feeding area back to the drinking area was restricted to a one-way corridor, preventing reverse passage and ensuring standardized recording of feeding behavior and body weight.

Water was provided ad libitum via a nipple drinking system. The feeding area was equipped with eight individual feeding stations, each allowing access by only one duck at a time. During each feeding event, the system automatically identified the duck via RFID and recorded the feeding time and feed weight in the feeder at the beginning and end of each visit, enabling calculation of single-meal feed intake. Feed was continuously replenished to ensure ad libitum access throughout the experimental period.

To obtain standardized fasted body weights, feed withdrawal was initiated at 14 and 42 days of age while water remained available. After a 6-hour feed withdrawal period, ducks voluntarily passed through the body-weight recording corridor at 15 and 43 days of age, respectively, and initial and final fasted body weights were automatically recorded.

Measurement of feeding and growth performance

Each duck was tagged with an individual electronic wing band. Feeding behavior and growth performance data (feeding time, frequency, duration, feed intake, and body weight) were recorded daily using an automated feeding and weighing system (platform supported by HSL Technology Co., Ltd., Beijing, China). Fasting body weights were obtained at 14 and 42 days.

Each duck was tagged with an individual electronic RFID wing band. Feeding behavior and growth performance traits, including feeding time, feeding frequency, feeding duration, feed intake, and body weight, were continuously recorded using the automated feeding and weighing system (platform supported by HSL Technology Co., Ltd., Beijing, China) throughout the experimental period.

Average daily gain (ADG), average daily feed intake (ADFI), and metabolic body weight (MBW^0.75) were calculated. RFI was calculated as the residual from the model (Aggrey et al., 2010; Zhang et al., 2019):

$$RFI=ADFI-\left(a+b1\times MB{W}^{0.75}+b2\times ADG\right)$$

where a is the intercept, and b₁, b₂ are the regression coefficients for MBW^0.75 and ADG, respectively.

Sample collection and slaughter phenotyping

Following completion of the feeding efficiency measurement period, final fasted body weights were obtained at 43 days of age, after which RFI was calculated for each individual. Based on individual RFI rankings, 12 Nonghua ducks (6 low-RFI and 6 high-RFI), including both males and females with similar initial body weights, were selected for omics analyses. Selected ducks were euthanized after completion of the RFI calculation, and tissue sampling was conducted accordingly. Whole blood was collected from the wing vein prior to euthanasia, and ducks were humanely sacrificed in accordance with institutional animal care guidelines. Serum was separated by centrifugation (3,000 rpm, 15 min, 4°C) and stored at −80°C for untargeted metabolomics analysis. Seven tissues (hypothalamus, pituitary gland, liver, duodenum, jejunum, ileum, and cecum) were carefully dissected, immediately snap-frozen in liquid nitrogen, and stored at –80°C until RNA extraction for transcriptome analysis.

Serum metabolite extraction and LC-MS/MS analysis

For serum metabolite extraction, 50 μL of each sample was mixed with 200 μL of pre-chilled methanol: acetonitrile (1:1, v/v). After vortexing, the samples were incubated at −20°C for 30 minutes and centrifuged (12,000 rpm, 4°C, 10 min). Supernatants were dried under vacuum and reconstituted in 100 μL of 50% methanol containing 5 ppm 2-chlorophenylalanine, followed by filtration (0.22 μm) and transfer to LC-MS vials. A pooled quality control (QC) sample, prepared by mixing equal aliquots of all samples, was injected periodically to monitor analytical stability.

Chromatographic separation was performed on a Thermo Vanquish Flex UPLC system (Thermo Fisher Scientific, Shanghai, China) with an ACQUITY UPLC HSS T3 column (100 Å, 1.8 μm, 2.1 × 100 mm). Mobile phases were 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B), with a flow rate of 0.4 mL/min and column temperature at 40°C. The gradient program was as follows: 0–1 min, 5% B; 1–7 min, 5–95% B; 7–8 min, 95% B; 8.1–12 min, 5% B.

Mass spectrometry was conducted on a Thermo Orbitrap Exploris 120 (Thermo Fisher Scientific, Shanghai, China) in positive and negative electrospray ionization modes. MS1 spectra were acquired at a resolution of 60,000 (m/z 100–1000). The top 4 most intense ions were selected for MS/MS fragmentation with a normalized collision energy (NCE) of 30% and a dynamic exclusion duration of 8 seconds. MS/MS spectra were acquired at a resolution of 15,000. QC samples were injected at the start and every 5–10 runs to ensure system stability.

Metabolomics data processing and statistical analysis

Raw data files (.raw) were processed using Compound Discoverer™ 3.3 software (Thermo Fisher Scientific, Waltham, MA, USA). Data processing steps included peak detection, retention time correction, peak alignment, gap filling, and normalization. Metabolite features detected in at least 50% of QC samples were retained for subsequent analysis. Principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were performed using the R package ropls to evaluate sample clustering and detect metabolic differences between groups. Differential expression analysis of metabolites between the HRFI and LRFI groups was performed using the limma package (v3.62.2)(Ritchie et al., 2015), with sex included as a covariate in the linear model to adjust for potential sex-related confounding effects. Differentially expressed Metabolites were defined as differentially expressed (DEMs) based on the criteria of VIP > 1, |log2 fold change| > 1, and P-value < 0.05. Kyoto Encyclopedia of Genes and Genomes (KEGG) (Kanehisa and Goto, 2000) pathway enrichment analysis of differential metabolites was performed using the R package clusterProfiler (v4.2.2)(Wu et al., 2021).

RNA extraction, library preparation, and sequencing

Total RNA was extracted from seven tissues (hypothalamus, pituitary, liver, duodenum, jejunum, ileum, and cecum) collected from 12 Nonghua ducks using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions. RNA concentration and purity were measured using a NanoDrop 2000 spectrophotometer (Thermo Scientific, Waltham, MA, USA), and integrity was assessed by agarose gel electrophoresis or an Agilent 2100 Bioanalyzer (Agilent Technologies Inc., California, USA) with the RNA 6000 Nano kit (5067-1511). Strand-specific libraries were constructed using the NEBNext Ultra II RNA Library Prep Kit for Illumina (New England Biolabs Inc., Ipswich, MA, USA). In brief, mRNA was enriched using Oligo (dT) magnetic beads and randomly fragmented under divalent cation conditions. After end repair, addition of 3′ adenylation, and adapter ligation, 400–500 bp fragments were selected, PCR-amplified, and purified to generate cDNA libraries. A total of 84 libraries (7 tissues × 12 ducks) were prepared. All libraries were sequenced using the DNBSEQ-T7 platform (MGI Tech Co., Ltd., Shenzhen, China) in paired-end 150 bp mode (PE150).

Quality control, alignment, and identification analyses of RNA-seq data

Raw reads were trimmed for adapters and low-quality bases using fastp (v0.20.1)(Chen et al., 2022). Clean paired-end reads were aligned to the duck reference genome (ZJU1.0) using HISAT2 (v2.0.1) (Kim et al., 2015a). SAM files were subsequently converted to BAM files using samtools (v1.21.0)(Danecek et al., 2021). Gene expression counts were obtained with featureCounts (v2.0.1)(Liao et al., 2014). To eliminate potential confounding effects from sex differences, a linear model was constructed that incorporated sex as a covariate, and subsequent differential expression analysis was conducted using the limma package (version 3.62.2) (Ritchie et al., 2015). Genes with an absolute |log2FC| > 1.5 and P-value < 0.05 were considered DEGs. Volcano plots of DEGs were generated using ggplot2 (v3.5.2)(Ginestet, 2011). Gene Ontology (GO)(Ashburner et al., 2000) and KEGG pathway enrichment analyses were conducted using the R package clusterProfiler (v4.2.2)(Wu et al., 2021). To investigate further cross-tissue signaling mechanisms, DEGs were annotated using the UniProtKB database (https://www.uniprot.org). Based on protein functional categories, genes annotated as "secreted proteins" or "receptor proteins" were classified accordingly.

Weighted gene co-expression network analysis

To explore the co-expression patterns of differentially expressed genes, we integrated the DEGs identified from seven tissues (hypothalamus, pituitary, liver, duodenum, jejunum, ileum, and cecum) and performed Weighted Gene Co-expression Network Analysis (WGCNA) using the R package WGCNA (v1.73) (Langfelder and Horvath, 2008). The expression matrix of these DEGs was normalized prior to analysis. The soft-thresholding power was automatically determined using the pickSoftThreshold function to approximate a scale-free topology. The adjacency matrix was then constructed and transformed into a topological overlap matrix (TOM), and hierarchical clustering was applied to identify gene modules (minimum module size = 30). The expression profile of each module was summarized by its module eigengene (ME).

Module–metabolite–phenotype integration

To explore potential regulatory axes linking gene co-expression modules, serum metabolites, and feed efficiency, we performed a two-step correlation analysis. First, Spearman correlation coefficients were calculated between module eigengenes derived from WGCNA and differentially abundant metabolites. Second, the same correlation method (Spearman) was applied to test the associations between metabolites and RFI values. Metabolites significantly associated with both MEs and RFI (P-value < 0.05) were considered shared key metabolites. These metabolites were then used to construct a three-layer regulatory framework of module (genes) – metabolite – RFI. To further characterize the tissue origin of regulatory signals, the proportion of genes from each tissue contributing to a given module was calculated. The two tissues with the highest proportions were interpreted as the major contributors to cross-tissue regulation of the identified metabolite–RFI axis.

Statistical analysis

All phenotypic data were expressed as mean ± standard deviation (S.D.). Differences between groups were assessed using independent-sample t-tests, with statistical significance defined as P < 0.05. Correlation analyses were performed using Spearman correlation coefficients, and results were reported as the correlation coefficient and the P-value. All statistical analyses were conducted using R (version 4.3.3).

Results

Divergence of feeding and growth traits between HRFI and LRFI Ducks

During the entire measurement period (14 to 42 days of age), we successfully collected feeding behavior and growth performance data from 1,000 meat ducks using an automated feeding and weighing system, and performed descriptive statistics on these phenotypes (Table 1, Table 2). In the female duck group, the low residual feed intake (LRFI) group had a significantly lower ADFI than the high residual feed intake (HRFI) group (P < 0.05). In terms of slaughter performance, except for the significantly lower whole eviscerated weight, half-eviscerated weight, and abdominal fat percentage in LRFI female ducks compared to HRFI female ducks (P < 0.05), there were no significant differences in other slaughter performance traits between the HRFI and LRFI groups. These results further confirm that RFI is independent of growth performance. To further characterize the relationships among feed efficiency, feeding behavior, and growth traits, we performed population-level correlation analyses (Supplementary Figures S1 and S2). Consistent with previous findings, RFI showed strong positive correlations with FCR and ADFI, moderate positive associations with feeding duration (TFD) and feeding bouts (TFB).

Table 1.

Comparison of feeding phenotypes between high residual feed intake (HRFI) and low residual feed intake (LRFI) groups in male and female ducks.

	Female		Male
	LRFI	HRFI	LRFI	HRFI
SBW (g)	505.00 ± 10.00	498.00 ± 0.00	496.00 ± 1.73	497.67 ± 12.50
BW (g)	2360.00 ± 274.18b	2608.33 ± 92.92ab	2826.67 ± 116.23a	2645.00 ± 329.13ab
MBW0.75	106.53 ± 1.58	105.42 ± 0.00	105.10 ± 0.28	105.36 ± 1.99
ADG (g/d)	68.33 ± 10.42b	78.16 ± 3.44ab	86.32 ± 4.28a	78.85 ± 11.28ab
ADFI (g/d)	167.43 ± 14.35c	237.72 ± 22.46a	193.60 ± 22.24b	225.67 ± 24.90ab
FCR (g/g)	2.48 ± 0.33bc	3.04 ± 0.16a	2.24 ± 0.17c	2.87 ± 0.15ab
RFI (g/d)	−14.53 ± 17.31bc	38.87 ± 16.45a	−19.57 ± 15.93c	25.62 ± 7.72a

Note: Values are presented as mean ± standard deviation (S.D.). SBW, slaughter body weight; BW, body weight; MBW^0.75, metabolic body weight; ADG, average daily gain; ADFI, average daily feed intake; FCR, feed conversion ratio; RFI, residual feed intake. The a, b, and c represent a significant difference (P < 0.05). Notably, HRFI ducks exhibited significantly higher ADFI and FCR.

Table 2.

Comparison of growth and production performance between HRFI and LRFI groups in male and female ducks.

	Female		Male
	LRFI	HRFI	LRFI	HRFI
Slaughter rate (%)	93.44 ± 1.59	94.03 ± 0.52	92.22 ± 1.91	92.34 ± 1.91
Half-clearance rate (%)	83.70 ± 2.35b	88.89 ± 0.76a	87.29 ± 2.45a	85.44 ± 1.20ab
Full-clearance rate (%)	73.72 ± 3.37b	83.12 ± 4.15a	80.51 ± 2.26a	76.41 ± 4.29ab
Pectoral rate (%)	12.05 ± 2.76	11.52 ± 1.01	11.21 ± 0.65	12.00 ± 1.27
Leg muscle rate (%)	10.65 ± 1.51a	8.48 ± 1.10ab	9.70 ± 1.82ab	7.83 ± 1.19b
Lean meat rate (%)	22.70 ± 3.71	20.00 ± 2.12	20.92 ± 2.24	19.84 ± 2.41
Abdominal fat rate (%)	1.09 ± 0.26b	1.87 ± 0.32a	1.70 ± 0.35a	1.36 ± 0.22ab
Liver rate (%)	3.11 ± 0.41	2.62 ± 0.23	2.96 ± 0.68	2.97 ± 0.22
Liver (g)	56.30 ± 6.15	53.57 ± 4.21	67.94 ± 17.14	64.50 ± 4.31
Heart (g)	14.73 ± 1.04	15.57 ± 0.67	16.32 ± 1.99	15.61 ± 1.91
Muscle-stomach (g)	55.4 ± 8.17bc	46.16 ± 1.07c	67.97 ± 5.67ab	70.45 ± 9.38a
Glandular stomach (g)	8.77 ± 1.65	9.90 ± 0.95	9.36 ± 1.10	7.88 ± 1.14

Note: Values are presented as mean ± standard deviation (S.D.). The a, b, and c represent a significant difference (P < 0.05). Notably, HRFI females exhibited significantly higher semi- and full-eviscerated yields and abdominal fat rate compared with LRFI females.

Overall experimental design and data quality control

We generated a multi-omics dataset from HRFI and LRFI Nonghua ducks (Fig. 1A), including 84 RNA-seq libraries from seven tissues (hypothalamus, pituitary, liver, duodenum, jejunum, ileum, and cecum) and 12 serum metabolomes, enabling cross-tissue analysis of feed efficiency. RNA sequencing yielded approximately 1237 Gb of raw data. After quality filtering, 1190 Gb of clean data were retained, with an average of 95 million clean reads per sample. The average Q30 score exceeded 95%, and GC content was approximately 43 – 48%, indicating high sequencing quality. Alignment to the duck reference genome (ZJU1.0) achieved mapping rates greater than 95% for all samples, ensuring data reliability for subsequent differential expression and network analyses (Supplementary Table S2). Principal component analysis (PCA) revealed clear clustering by tissue type, with partial separation between HRFI and LRFI groups within each tissue (Fig. 1B). Untargeted LC-MS/MS profiling of serum metabolites detected 793 features (354 in positive and 439 in negative ion modes). Quality control (QC) assessments demonstrated tight clustering of pooled QC samples in PCA and PLS-DA score plots, confirming robust analytical reproducibility and batch stability (Fig. 1C). Collectively, these quality control results verified the suitability of the transcriptomic and metabolomic datasets for downstream analyses.

Fig. 1.

Overview of experimental design and data quality control. (A) Residual feed intake (RFI) distribution of selected Nonghua ducks used for multi-tissue transcriptomic and serum metabolomic profiling; red and blue triangles indicate low-RFI (LRFI) and high-RFI (HRFI) individuals. (B) Principal component analysis (PCA) of transcriptomic data from 84 samples (7 tissues × 12 ducks), showing distinct clustering by tissue type with partial separation between RFI groups. (C) Partial least squares-discriminant analysis (PLS-DA) of serum metabolomics data demonstrating tight QC clustering and analytical reproducibility.

Multi-tissue transcriptomic and serum metabolomic differences between HRFI and LRFI ducks

Transcriptomic analysis revealed significant tissue-specific expression differences between HRFI and LRFI ducks across multiple tissues. Using a threshold of |log2FC| > 1.5 and P < 0.05, a total of 769 differentially expressed genes (DEGs) were identified across seven tissues (Fig. 2A, Supplementary Table S3), with notable variation in DEG numbers among tissues (Fig. 2B). The hypothalamus exhibited the highest number of DEGs (114; 56 upregulated and 58 downregulated), followed by the pituitary (105; 64 upregulated and 41 downregulated), liver (122; 78 upregulated and 44 downregulated), and cecum (236; 117 upregulated and 119 downregulated). Fewer DEGs were detected in the small intestine, including 95 in the duodenum (38 upregulated, 57 downregulated), 57 in the jejunum (32 upregulated, 25 downregulated), and 40 in the ileum (18 upregulated, 22 downregulated). The cluster analysis revealed that the DEGs exhibited high reproducibility within groups and significant differences between groups (Supplementary Figure S3). Untargeted serum metabolomics identified 28 differential metabolites (VIP > 1, |log2FC| > 1, P < 0.05), including 15 in positive ion mode (13 upregulated and 2 downregulated) and 13 in negative ion mode (7 upregulated and 6 downregulated) (Fig. 2C, Supplementary Table S4). These differential metabolites were mainly associated with lipid metabolism and amino acid metabolism pathways.

Fig. 2.

Differential transcriptomic and metabolomic profiles and tissue specificity in HRFI and LRFI ducks. (A) Volcano plot of differentially expressed genes (DEGs) between LRFI and HRFI ducks in seven tissues. Colored points represent DEGs in each tissue, while gray points indicate 2,000 randomly selected non-significant genes (NS) for background visualization. (B) Numbers of differentially expressed genes (DEGs) identified in seven tissues (hypothalamus, pituitary, liver, duodenum, jejunum, ileum, and cecum) between HRFI and LRFI ducks. Upregulated (red) and downregulated (blue) genes are shown separately. (C) Numbers of differential metabolites (DEMs) detected in positive (POS) and negative (NEG) ion modes. (D) UpSet plot illustrating tissue-specific and overlapping DEGs among the seven tissues; most DEGs are tissue-specific, particularly in the cecum. (E) UMAP clustering based on 621 tissue-specific DEGs, showing clear separation by tissue type and partial clustering by RFI group.

Functional annotation of all DEGs using the UniProtKB database revealed 121 genes encoding secreted or receptor proteins, including GABRA4 and CCK, which may serve as potential mediators of inter-tissue signaling (Supplementary Table S5). An UpSet plot illustrated the distribution of DEGs across the seven tissues (Fig. 2D), revealing that 621 genes (80%) were tissue-specific. The cecum contributed the largest number of tissue-specific genes (201), followed by the liver (104) and hypothalamus (101). To further characterize these 621 tissue-specific DEGs, UMAP analysis revealed clear clustering by tissue type: the hypothalamus and pituitary formed distinct clusters, the liver clustered independently, and the four intestinal segments grouped closely together (Fig. 2E). Notably, several tissue-specific DEGs annotated as secreted proteins (HSD3B1, CCK, and IL8) and receptors (ADORA1, GABRA4, and GABRA4) were identified, suggesting potential roles in mediating brain–gut–liver communication.

Functional enrichment analysis of tissue-specific differentially expressed genes

To investigate the transcriptional regulatory features and biological functions underlying residual feed intake (RFI) differences across tissues, we performed GO and KEGG enrichment analyses on 621 tissue-specific differentially expressed genes (DEGs) identified from seven tissues (Supplementary Table S6). The results revealed marked heterogeneity in enriched pathways among tissues.

The DEGs in the central tissues (hypothalamus and pituitary) were predominantly enriched in signaling and immune-related pathways. The DEGs in the hypothalamus showed significant enrichment in steroid hormone biosynthesis, intestinal immune network for IgA production, and neuroactive ligand–receptor interaction, while the pituitary was mainly enriched in the Notch signaling pathway, virion–hepatitis viruses, and tight junction pathways. As a metabolic hub, liver-specific DEGs were primarily enriched in the insulin signaling pathway, neuroactive ligand signaling, and cell adhesion molecules, which are involved in glucose metabolism and energy homeostasis. Intestinal tissues (duodenum, jejunum, cecum, and ileum) were enriched in nitrogen metabolism, neuroactive ligand–receptor interaction, hormone signaling, biosynthesis of amino acids, fatty acid metabolism, and fructose and mannose metabolism, suggesting key roles in amino acid, fatty acid, and carbohydrate metabolism as well as neuroendocrine regulation (Fig. 3A). Based on all KEGG pathways enriched across tissues, the pathway specificity and overlap analysis indicated that tissue-specific functions predominated (Fig. 3B). A total of 32 pathways were enriched exclusively in a single tissue, mainly reflecting the core physiological roles of each tissue, such as neural signaling and hormonal regulation in the hypothalamus and pituitary, energy metabolism and substrate transport in the liver, and amino acid, fatty acid, and carbohydrate metabolism in the intestines. Notably, 17 pathways were shared across multiple tissues, including 12 shared between two tissues, 2 shared among three tissues, and 3 shared among four tissues. These cross-tissue shared pathways were mainly involved in neuroactive ligand–receptor interaction, hormone signaling, and metabolic processes, suggesting the existence of potential brain–gut–liver regulatory axes.

Fig. 3.

Functional enrichment and cross-tissue overlap of tissue-specific DEGs. (A) Bubble plot of KEGG pathway enrichment across tissues; bubble size indicates gene ratio and color represents significance (−log10 P-value). (B) UpSet plot showing overlap of KEGG pathways among tissues; most pathways were tissue-specific, with limited cross-tissue sharing.

Building on this, we further examined the neuroendocrine-related pathways and their genes among the cross-tissue shared pathways (Table 3). The neuroactive ligand–receptor interaction was enriched in the hypothalamus (UTS2B), jejunum (GRIK2), and cecum (ADORA1, DRD3, GABRA4, P < 0.05). The neuroactive ligand signaling pathway was enriched in the duodenum (SLC17A8, P < 0.05), jejunum (GRIK2), cecum (ADORA1, DRD3, GABRA4, P < 0.05), and liver (SLC6A1). The hormone signaling pathway was enriched in the duodenum (GCG, MLN, PENK, CCK, P < 0.05) and cecum (DRD3). The steroid hormone biosynthesis was enriched in the hypothalamus (HSD3B1, P < 0.05) and cecum (SRD5A2). The insulin signaling pathway was enriched in the duodenum (GYS2) and liver (SHC3). Overall, neuroactive ligand–receptor interaction, neuroactive ligand signaling pathway, and hormone signaling pathway repeatedly appeared in both central and peripheral tissues, and the associated genes (GYS2, HSD3B1, ADORA1, DRD3, GABRA4, GCG, CCK) may constitute a core neuro–hormonal signaling network regulating RFI.

Table 3.

Common KEGG pathways identified from enrichment analysis of tissue-specific DEGs across multiple tissues in HRFI and LRFI groups.

KEGG Pathway (ID)	Tissue	Gene	P-vaule	Protein Coding Type
Neuroactive ligand–receptor interaction	Hypothalamus	UTS2B	n.s.	SP
	Jejunum	GRIK2	n.s.	TP
	Cecum	DRD3, ADORA1, GABRA4	<0.05
Neuroactive ligand signaling pathway	Duodenum	SLC17A8	n.s.	TP
	Jejunum	GRIK2	n.s.
	Cecum	ADORA1, DRD3, GABRA4	<0.05
	Liver	SLC6A1	n.s.
Hormone signaling pathway	Duodenum	GCG, MLN, PENK, CCK	<0.05	SP
		GYS2	<0.05	TP
	Liver	SHC3	n.s.
Steroid hormone biosynthesis	Hypothalamus	HSD3B1	<0.05	TP
	Cecum	SRD5A2	n.s.

Note: Pathways related to neuroactive ligand–receptor interaction, neuroactive ligand signaling, hormone signaling, and steroid hormone biosynthesis are shown, along with the corresponding tissues, representative genes, P-values, and protein-coding categories. n.s., not significant. TP, target protein-coding genes. SP, secreted protein-coding genes.

Cross-tissue module–metabolite associations reveal candidate gut–liver and gut–brain axes related to feed efficiency

Weighted gene co-expression network analysis of DEGs identified 16 co-expression modules (Supplementary Table S7, S8). Correlation analysis revealed that 9 modules were significantly associated with serum metabolites. In parallel, 14 differential metabolites were correlated with RFI. Integration of these results identified key metabolites that were simultaneously associated with both gene modules and RFI, thereby outlining module–metabolite–RFI association patterns (Fig. 4A and B).

Fig. 4.

Module–metabolite–RFI associations and tissue composition of candidate modules. (A) Heatmap of correlations between WGCNA modules and differential serum metabolites. Numbers in each cell indicate correlation coefficients, with corresponding P-values in parentheses. The color scale represents the correlation strength (red = positive, blue = negative). (B) Correlations between differential metabolites and residual feed intake (RFI). Each dot represents a metabolite, with significance indicated by star symbols (* P < 0.05; ** P < 0.01; *** P < 0.001). (C) Tissue composition of representative modules. The bar plot shows the proportion of genes in each module that originated from individual tissues. Tissue composition was calculated as the percentage of module genes annotated as tissue-specific DEGs. Cecum and liver contributed most to MEblack, while pituitary and cecum contributed most to MEblue. (D) Two representative cross-tissue regulatory axes from module–metabolite–RFI associations. The gut–liver axis (MEblack) links cecal genes (DCC, CCDC60) and hepatic genes (CNTNAP1, SHC3, RAB36) to 6-oxopiperidine-2-carboxylic acid, showing positive module–metabolite correlations, while the metabolite itself is negatively associated with RFI. The gut–brain axis (MEblue) connects cecal genes (FABP6, KCNJ11, LRTM1) and pituitary genes (TRPA1, LRTM1) to LysoPE(18:2(9Z,12Z)/0:0), also positively correlated with the module but negatively associated with RFI. Red arrows indicate gene up- or downregulation in the LRFI group, while green arrows denote positive or negative associations.

Module MEblack was significantly associated with 6-Oxopiperidine-2-carboxylic acid, which in turn showed a significant correlation with RFI, suggesting that this module may be linked to feed efficiency via this metabolite (Supplementary Table S9, S10). Tissue origin analysis further revealed that MEblack was primarily composed of genes from the cecum (32.43%) and liver (29.73%) (Fig. 4C), indicating a candidate gut–liver cross-tissue association related to feed efficiency. Representative genes in this module included CNTNAP1, SHC3, and RAB36 (liver), as well as DCC and CCDC60 (cecum). Similarly, MEblue was associated with LysoPE(18:2(9Z,12Z)/0:0), which in turn showed a significant correlation with RFI (Supplementary Table S9, S10). Tissue origin analysis showed that MEblue was predominantly composed of pituitary (28.99%) and cecum (23.19%) genes (Fig. 4C), supporting a candidate gut–brain cross-tissue association to feed efficiency. Representative genes in this module included FABP6 and KCNJ11 (cecum), TRPA1 (pituitary), and LRTM1 (pituitary and cecum). Taken together, these findings identified two candidate association axes, MEblack–6-Oxopiperidine-2-carboxylic acid–RFI and MEblue–LysoPE(18:2(9Z,12Z)/0:0)–RFI, which may represent potential gut–liver and gut–brain relationships relevant to feed efficiency.

Discussion

This study employed a systems biology framework that integrates multi-tissue transcriptomics and serum metabolomics to elucidate the molecular basis of RFI divergence in ducks. Comparative analyses across seven tissues along the brain–gut–liver axis revealed both tissue-specific transcriptional programs and cross-tissue communication, leading to the identification of two module–metabolite–RFI association frameworks.

Phenotypic comparisons demonstrated that LRFI ducks had lower ADFI and FCR, confirming the effectiveness of the grouping strategy. Importantly, most growth performance traits, including ADG, slaughter rate, breast muscle percentage, leg muscle percentage, and lean meat percentage, did not differ significantly between HRFI and LRFI groups, which is consistent with previous studies (Bai et al., 2022). Moreover, the strong positive correlations of RFI with FCR and ADFI, together with its negative correlations with liver percentage and leg muscle percentage, suggest that RFI reflects a composite attribute of energy utilization and feeding behavior. In females, significant differences in abdominal fat percentage, semi-eviscerated rate, and eviscerated rate may be partly attributable to increased variability from the relatively small sample size and individual variation in fat deposition, as well as indirect selection responses under RFI-based grouping. Collectively, these findings support the view that RFI serves as a reliable indicator for improving feed efficiency without compromising growth potential (Mondin et al., 2024; Wu et al., 2025).

Transcriptome profiling identified 769 DEGs, with most being tissue-specific. These findings highlight organ-intrinsic transcriptional regulation as the major driver of RFI divergence, while a smaller set of shared pathways provides molecular links for cross-tissue integration. In central tissues, DEGs in the hypothalamus and pituitary were enriched in neuroactive ligand–receptor interactions and hormone signaling pathways, consistent with their roles in appetite and energy regulation (Morton et al., 2006; Kim et al., 2015b). Representative examples include UTS2B, a microbiota-regulated neuropeptide expressed in vagal afferents that mediates gut–brain nutrient signaling (Yoshioka et al., 2022), and the FGF21–β-Klotho axis, a key endocrine pathway linking central and peripheral metabolic regulation (Fisher and Maratos-Flier, 2016; Lan et al., 2017). In the liver, DEGs were primarily associated with energy metabolism and hepatokine secretion (Markan et al., 2014; Lin et al., 2017; Kim et al., 2022a). Notably, RRAD, which inhibits glycolysis and promotes fatty acid oxidation, was upregulated in LRFI ducks, suggesting metabolic reprogramming toward more efficient energy utilization (Kim et al., 2022b; Minchenko et al., 2023). Intestinal DEGs were linked to nutrient sensing and gut–brain communication, exemplified by CCK and GLP-1, two classic satiety peptides that act via vagal afferents to regulate appetite and glucose metabolism (Greenberg et al., 1990; Borgmann et al., 2021). Although most DEGs were tissue-specific, we observed a limited set of shared neuropeptide and hormone-related pathways across the hypothalamus, pituitary, and intestine, suggesting that cross-organ integration is mediated by a relatively small number of conserved signaling routes.

We identified two candidate cross-tissue association axes, the intestinal–hepatic metabolic axis and the intestinal–brain neuroendocrine axis, that underlie differences in feed efficiency. These association patterns provide insight into cross-tissue relationships related to RFI variation, offering potential biomarkers and targets for enhancing nutrient utilization. In the MEblack–6-Oxopiperidine-2-carboxylic acid–RFI (intestinal–liver axis), module genes mainly originated from the cecum and liver. The cecum, a major metabolic site at the distal end of the intestine, harbors microbial communities with active roles in amino acid metabolism. The 6-oxopiperidine-2-carboxylic acid is thought to be derived from the oxidation of 6-hydroxypipecolic acid and represents a branch metabolite in lysine metabolism(Duran, 2008; Wempe et al., 2019). We observed a significant negative correlation between 6-OP2C and RFI, suggesting that individuals with low RFI may enhance amino acid catabolism and nitrogen clearance, thereby improving energy efficiency. Similar mechanisms have been reported in ruminants, where low-RFI cattle exhibited higher nitrogen utilization efficiency and reduced protein turnover (Jorge-Smeding et al., 2021). Representative genes within this module, such as SHC3, CNTNAP1, and DCC, are associated with receptor adaptor function and cell adhesion. SHC3 has been linked to amino acid and lipid metabolism in chickens (Liao et al., 2020). CNTNAP1 (CASPR1) functions as a cell adhesion molecule critical for maintaining axonal paranodal structures and plays a key role in neuronal signal conduction (Vallat et al., 2016; Chang et al., 2023). These findings suggest that differences in amino acid metabolism may be accompanied by cross-tissue remodeling of material transport and signaling processes.

In the MEblue–LysoPE(18:2(9Z,12Z)/0:0)–RFI (intestinal–brain axis), module genes were mainly derived from the pituitary and cecum. LysoPE(18:2(9Z,12Z)/0:0), a lysophosphatidylethanolamine metabolite, is broadly involved in membrane remodeling, inflammatory responses, and signal transduction. Previous studies have demonstrated a close association between lysophospholipids, obesity, and type 2 diabetes, as these bioactive lipids modulate insulin sensitivity, inflammation, and energy homeostasis (Yamamoto et al., 2022). In this study, LysoPE(18:2(9Z,12Z)/0:0) levels were significantly negatively correlated with RFI and were linked to pituitary- and cecum-derived gene modules, suggesting a potential cross-tissue association pattern involving "intestinal lipid signaling/endocrine sensing/central regulation". Representative genes in this module include FABP6, KCNJ11, TRPA1, and LRTM1. FABP6 is responsible for bile acid uptake and intracellular transport in intestinal epithelial cells; its loss leads to increased fecal bile acid excretion, disruption of the enterohepatic bile acid cycle, and altered lipid absorption and metabolic status (Praslickova et al., 2012; Habib et al., 2021). KCNJ11 encodes the core subunit of the K-ATP channel, which couples the cellular metabolic state (ATP/ADP ratio) with electrical activity to regulate glucose-stimulated insulin secretion (Proks et al., 2005; Moazzam-Jazi et al., 2022). Notably, LRTM1 was detected in both the cecum and pituitary. LRTM1 shows ventral midbrain-specific expression during embryonic development and serves as a marker of dopaminergic precursors. Mouse models have further demonstrated that Lrtm1 can sense insulin receptor signaling and regulate metabolism and behavior (Samata et al., 2016; Yu et al., 2025). Taken together, our findings suggest that LRTM1 may serve as a potential molecular node with the candidate intestinal–brain association axis, linking intestinal lipid/nutrient signals with central–peripheral metabolic feedback.

Conclusion

This study integrated multi-tissue transcriptomics and serum metabolomics to systematically uncover the molecular basis of residual feed intake (RFI) in ducks. The results demonstrated that RFI is independent of growth traits, with differentially expressed genes showing strong tissue specificity, particularly in the hypothalamus, liver, and cecum. Network integration further identified two candidate cross-tissue association axes — MEblack–6-Oxopiperidine-2-carboxylic acid and MEblue–LysoPE(18:2(9Z,12Z)/0:0)—that were significantly associated with RFI. Overall, this study characterizes cross-tissue molecular association patterns, establishes a multi-tissue transcriptomic atlas, and provides valuable data resources for marker-assisted selection and precision breeding.

Data and model availability statement

The RNA-seq data from this study can be read from GSA sequences (Accession Number: PRJCA047513).

CRediT authorship contribution statement

Shihao Guo: Writing – original draft, Validation, Supervision, Investigation, Data curation, Conceptualization. Yang Xi: Methodology. Jingjing Qi: Software, Project administration. Zhao Yang: Software, Project administration. Xu Han: Validation, Data curation. Weikang Ling: Validation, Data curation. Lili Bai: Methodology, Investigation, Formal analysis. Anqi Huang: Methodology, Investigation, Formal analysis. Shenqiang Hu: Methodology, Investigation, Formal analysis. Jiwei Hu: Methodology, Investigation, Formal analysis. Chunchun Han: Resources, Investigation, Formal analysis. Jiwen Wang: Resources, Investigation, Formal analysis. Liang Li: Supervision, Methodology, Investigation, Funding acquisition, Conceptualization. Hehe Liu: Writing – review & editing, Supervision, Investigation, Funding acquisition, Conceptualization.

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.