Cross-kingdom genomic variation in chicken gut microbiomes: insights from China’s diverse local breeds

Abstract

Background

The gut microbiome possesses substantial genetic diversity that supports microbial adaptation, but the genomic variation patterns across its prokaryotic and viral populations remain incompletely characterized.

Results

Through integrated metagenomic and metatranscriptomic analysis of ten indigenous chicken breeds from China, we recovered 1527 representative prokaryotic MAGs, 37,555 representative DNA viral contigs, and 1867 representative RNA viral contigs (primarily comprising Bacillota/Bacteroidota, Uroviricota, and Lenarviricota/Pisuviricota, respectively). By integrating complementary short-read and long-read metagenomics with metatranscriptomics, we identified structural variants (SVs) and single-nucleotide variants (SNVs) in these cross-kingdom genomes. Positive SV-SNV density correlations occurred consistently across all microbial groups, indicating coordinated mutational processes. DNA viruses exhibited the highest variant prevalence (86.9% SNVs, 47.7% SVs), with temperate phages accumulating significantly more variants than virulent phages. Functionally, prokaryotic variants accumulated in carbohydrate metabolism and amino acid metabolism, while viral variants demonstrated broad metabolic hijacking. Horizontal gene transfer (HGT) was characterized by a strong virus-associated signature (69.40% of 536 events) and marked by an asymmetric pattern, with phage-to-bacteria (P-to-B) flow alone constituting 37.50% of all events. Random forest analysis revealed a strong bidirectional predictive relationship between SV and SNV densities across prokaryotic, DNA viral, and RNA viral populations, suggesting coupled genomic instability. Niche breadth emerged as a major driver of SNVs across kingdoms and was positively correlated with variant density. In prokaryotes, HGT events significantly shaped variant patterns. For viruses, genomic GC content was an important factor and consistently showed a negative correlation with SNV density in both DNA and RNA viruses.

Conclusions

These findings demonstrate that coordinated mutational processes and kingdom-specific intrinsic factors drive genomic variation, with viruses serving as key genetic exchange vectors in chicken gut ecosystems.

Download video file ^{(49.2MB, mp4)}

Video Abstract

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s40168-026-02347-3.

Introduction

The gut microbiome represents one of the most complex and dynamic ecosystems in nature, harboring trillions of microorganisms that collectively influence host physiology, immunity, and overall health. In poultry production, the gut microbiome plays particularly critical roles in nutrient digestion, pathogen resistance, and immune system development, directly impacting poultry health and productivity [1]. As global demand for poultry products continues to rise, understanding and manipulating the gut microbiome have emerged as a promising avenue for sustainable agricultural intensification and reduced reliance on antibiotics [2, 3]. Despite the recognized importance of avian gut microbiomes, current research faces significant limitations that hinder the development of effective microbiome-based interventions. Most studies have focused predominantly on bacterial communities while largely overlooking viral populations, which represent a substantial and often underestimated proportion of microbial genetic material [4–6]. Bacteriophages, which are viruses that infect bacteria, are the most abundant biological entities on Earth, with an estimated total of 10³¹ particles, corresponding to a biomass of approximately 200 million tons [6]. This bacteria-centric approach overlooks the critical regulatory roles that viruses play in shaping microbial community structure through predation, horizontal gene transfer (HGT), and metabolic reprogramming [7–9]. Furthermore, the vast majority of studies on the gut virome have primarily focused on taxonomic profiling, providing limited insight into the functional capabilities and evolutionary dynamics that drive community assembly and stability [7, 10].

Warm-blooded animals, such as chickens, harbor one of the most complex gut microbial ecosystems [1]. Within this intricate ecosystem, genomic variation serves as the primary driver of microbial adaptation, community assembly, and functional specialization, making it essential to understand the mechanistic basis of variant occurrence and distribution patterns. While structural variants (SVs) (e.g., insertions, deletions, and rearrangements) and single-nucleotide variants (SNVs) represent distinct mutational processes with fundamentally different evolutionary origins and functional consequences, their potential coordination and differential impacts on gut microbial function have not been systematically investigated [11–13]. While SVs can range from small indels to large chromosomal rearrangements reshaping metabolic pathways, SNVs predominantly influence protein function through amino acid substitutions [14].

In this study, we conducted integrated metagenomic and metatranscriptomic analysis of prokaryotic and viral communities across ten genetically distinct chicken breeds to systematically characterize gut microbiome genomic variation and underlying mechanisms. Our objectives were to (1) establish a comprehensive cross-kingdom microbial genome catalog and characterize taxonomic composition, functional profiles, and community assembly patterns; (2) investigate the distribution, abundance, and functional implications of microbial SVs and SNVs; (3) quantify HGT patterns across breeds to understand cross-kingdom genetic exchange mechanisms; and (4) identify key factors driving variant occurrence and their kingdom-specific mechanisms. This approach provides fundamental insights into evolutionary forces shaping gut microbiome diversity and establishes theoretical foundations for precision interventions in gut health of poultry systems.

Results

Fecal microbiome catalog of prokaryotic, fungal, and viral genomes across ten chicken breeds

The microbial communities in ten different chicken breeds were characterized, as detailed in Supplementary Table S1 and Fig. S1: Labai high-leg chicken (LB), Taliu black-boned chicken (TL), Wuliangshan black-boned chicken (WLS), Yunlong short-leg chicken (YL), Wuding chicken (WD), Chahua chicken (CH), Nixi chicken (NX), Yanjin black-boned chicken (YJ), Lanping silky chicken (RM), and Ake chicken (AK). Through a metagenomic genome binning approach, we successfully recovered 8392 prokaryotic metagenome-assembled genomes (MAGs), 12 fungal MAGs, and 40,564 DNA viral contigs from 116 next-generation sequencing (NGS) and 16 Oxford Nanopore Technologies (ONT) metagenomic datasets derived from fecal samples of ten chicken breeds (Fig. 1A). Additionally, metatranscriptomic analysis of 57 datasets yielded 3446 RNA viral contigs. Following dereplication, we obtained 1527 representative prokaryotic MAGs (1518 bacterial MAGs and 9 archaeal MAGs), 8 representative fungal MAGs, 37,555 representative DNA viral contigs, and 1867 representative RNA viral contigs. High-quality genomes comprised 410 prokaryotic representative MAGs, 3 fungal representative MAGs, 841 DNA viral contigs, and 139 RNA viral contigs. Notably, 242 complete DNA viral genomes and 1 complete RNA viral genome were recovered. The limited recovery of eukaryotic genomes likely reflects the low fungal abundance in chicken feces or the inherent limitations of metagenomic binning methodologies for eukaryotic organisms.

Fig. 1.

Recovery and characterization of genomes from chicken fecal metagenomes and metatranscriptomes. A Number and quality of recovered microbial genomes. For prokaryotic and eukaryotic genomes, high quality was defined as completeness ≥ 90% and contamination ≤ 5%, medium quality as completeness between 50% and 90% and contamination ≤ 10%, and low quality as completeness < 50% or contamination > 10%. For viral contigs, complete was defined as a closed genome. High quality was defined as completeness > 90%, medium quality as completeness between 50% and 90%, and low quality as completeness < 50%. “Undetermined” was used when completeness could not be assessed or was unreliable. B Proportion of taxonomically classified microbial genomes at the phylum level. C, D Prokaryotic (C), eukaryotic (C), and viral (D) compositions at the phylum level. The width of the chord indicates the number of genomes. E, F Host types of DNA (E) and RNA (F) viruses predicted according to their taxonomical classification. Venn diagram depicts the intersections of viral host types. G, H Proportion of metagenomic (G) and metatranscriptomic (H) reads mapping to representative prokaryotic, eukaryotic, and viral genomes

All recovered prokaryotic and eukaryotic genomes achieved 100% taxonomic classification success at the phylum level (Fig. 1B), while viral sequences demonstrated classification rates of 94.6% for DNA viruses and 90.7% for RNA viruses, reflecting current limitations in viral reference databases. The 8 representative fungal MAGs belonged exclusively to the phylum Ascomycota, encompassing four genera: Saccharomyces, Kurtzmaniella, Cyberlindnera, and Debaryomyces (Fig. 1C). The recovered representative prokaryotic genomes in chicken feces were dominated by Bacillota, Bacteroidota, Pseudomonadota, and Actinomycetota (Fig. 1C). Bacillota exhibited remarkable taxonomic diversity with multiple subgroups, including Bacillota_A, Bacillota_B, Bacillota_C, and Bacillota_I, demonstrating high phylogenetic complexity within this phylum in chicken feces. As typical ileal-fecal microbes, the prevalence of Bacillota can be attributed to their ability to form spores that survive under harsh conditions and spread across different hosts, resulting in frequent but transient colonization and a highly diverse population structure [15–17]. Archaea belonging to Halobacteriota, Methanobacteriota, and Thermoplasmatota were also detected, contributing to the overall microbial ecosystem complexity.

Viral genome architectures were predicted based on taxonomic classification according to the International Committee on Taxonomy of Viruses (ICTV) (Fig. S2). For the recovered DNA viral contigs, double-stranded DNA (dsDNA) genomes comprised the overwhelming majority at 94.0%, indicating the predominance of stable, replicatively robust viral forms in the gut environment (Fig. S2A). Single-stranded DNA (ssDNA) viruses represented only 0.4% of the total. Notably, 5.5% of DNA viral contigs remained unclassified regarding genome architecture, likely reflecting limitations in current viral classification systems or the presence of novel viral genome organizations. The RNA virus genome architecture displayed considerably greater diversity and more balanced distribution patterns (Fig. S2B). Positive-sense single-stranded RNA viruses (ssRNA(+)) dominated at 48.6%, followed by double-stranded RNA (dsRNA) viruses at 34.1%, and negative-sense single-stranded RNA viruses (ssRNA(−)) at 1.9%, while 9.3% remained unclassified.

Viral contigs were taxonomically classified into three DNA realms (Duplodnaviria, Varidnaviria, and Monodnaviria) and one RNA realm (Riboviria) (Fig. 1D). Among DNA viral contigs, 99.2% belonged to the phylum Uroviricota, while RNA viral contigs were dominated by Pisuviricota (56.7%) and Lenarviricota (25.4%). Host prediction based on taxonomic classification according to the International Committee on Taxonomy of Viruses (ICTV) revealed distinct host-specificity patterns (Fig. 1E, F). DNA viruses predominantly targeted prokaryotic hosts (94.0%), with the majority infecting bacteria and fewer targeting archaea, while those associated with vertebrates and other eukaryotes were relatively rare. In contrast, RNA viruses demonstrated broader host ranges, with substantial proportions infecting both prokaryotic and eukaryotic hosts, including vertebrate-associated viruses, reflecting their greater evolutionary plasticity and host adaptability.

Prokaryotic and viral community patterns in chicken feces

In the metagenomic datasets, prokaryotic genomes accounted for the dominant proportion of mapped reads (69.9%), followed by DNA viruses (6.72%) and RNA viruses (0.0177%) (Fig. 1G). Eukaryotic genomes represented a minor fraction (0.437%), while 23.0% of metagenomic reads remained unclassified. In the metatranscriptomic datasets, prokaryotes similarly dominated the mapped transcriptional activity (77.0%), followed by DNA viruses (4.01%) (Fig. 1H). Notably, the relative abundance of RNA viruses was markedly higher in the transcriptomic data (0.85%) compared to the metagenomic data. Eukaryotes accounted for 0.135% of the transcriptomic reads, while 18.0% of transcripts were unclassified.

At the phylum level, the prokaryotic community exhibited a clear dominance of Bacillota, Bacillota_A, Bacillota_I, Bacteroidota, Actinomycetota, and Pseudomonadota across fecal samples of all the chicken breeds (Fig. S3A). The family-level analysis revealed that Lactobacillaceae, Lachnospiraceae, and Bacteroidaceae were among the most abundant bacterial families (Fig. S3B). This distribution pattern suggests a well-balanced microbial ecosystem where lactic acid bacteria contribute to gut health maintenance, while Bacteroidaceae members facilitate nutrient processing [18]. The genus-level composition (Fig. S3C) showed notable abundances of Lactobacillus, Ligilactobacillus, Limosilactobacillus, and Phocaeicola.

As for the DNA viral community, Uroviricota constituted the majority of DNA viruses, reflecting the prevalence of bacteriophages that likely regulate bacterial populations within the gut ecosystem (Fig. S3D). At the class level (Fig. S3E), Caudoviricetes (a class of tailed bacteriophages) showed the highest abundance, which is expected given its role as the most common bacteriophage infecting gut bacteria [19]. The family-level classification of DNA viruses (Fig. S3F, G) revealed a diverse bacteriophage community, with Herelleviridae, Inoviridae, Kyanoviridae, and Straboviridae being the most prominent families. In contrast, the RNA viral communities were composed of phyla typically associated with eukaryotic viruses, including Lenarviricota, Pisuviricota, Duplornaviricota, Kitrinoviricota, and Negarnaviricota, with Lenarviricota and Pisuviricota being dominant (Fig. S3H). The most dominant phyla were Lenarviricota (which includes both eukaryotic viruses and RNA bacteriophages) and Pisuviricota (predominantly eukaryotic viruses with few known RNA bacteriophages). At the class level (Fig. S3I), the community was primarily constituted by Duplopiviricetes (primarily eukaryotic viruses), Howeltoviricetes (eukaryotic viruses), Leviviricetes (RNA bacteriophages), Pisoniviricetes (eukaryotic viruses), and Resentoviricetes (eukaryotic viruses). At the family level (Fig. S3J), the most abundant families were Mitoviridae, Partitiviridae, and Sedoreoviridae, all of which are eukaryotic virus families.

Community assembly patterns of fecal microbiomes in chickens

The neutral community model (NCM) analysis revealed fundamental community assembly mechanisms governing prokaryotic, DNA viral, and RNA viral communities in chicken feces (Fig. 2A–C). The prokaryotic community exhibited a strong fit to the neutral model (R² = 0.603, Nm = 43.12) (Fig. 2A), suggesting that stochastic processes such as ecological drift and dispersal limitation play predominant roles in bacterial community assembly. This finding aligns with previous studies on gut microbiomes, where neutral processes often dominate due to the relatively homogeneous gut environment and continuous microbial turnover [20]. DNA viral communities showed even stronger adherence to neutral dynamics (R² = 0.75, Nm = 193.68, neutral microbial proportion = 60.1%) (Fig. 2B), indicating that viral assembly is largely governed by random sampling and immigration processes (Fig. 2A). The higher migration rate [21] for DNA viruses compared to prokaryotes suggests greater dispersal capability, possibly due to their smaller size and association with mobile bacterial hosts. RNA viral communities displayed an intermediate neutral model fit (R² = 0.715, Nm = 2.22) (Fig. 2C), but with notably lower migration rates, implying more restricted dispersal patterns and potentially stronger host-specific associations.

Fig. 2.

Community assembly of prokaryotes and viruses in chicken feces. A–C Community assembly of prokaryotic (A), DNA viral (B), and RNA viral (C) communities in chicken feces predicted by the neutral community model (NCM). D–F Ecological niche breadth classification of prokaryotes (D), DNA viruses (E), and RNA viruses (F) into generalists and specialists

Ecological specialization patterns reflect niche differentiation strategies

The niche breadth analysis revealed differences in ecological specialization patterns across microbial groups, with prokaryotes demonstrating the highest capacity for both specialized and generalized ecological strategies (Fig. 2D–F). These patterns provide insights into the evolutionary strategies and ecological constraints operating within the chicken gut ecosystem. Prokaryotic communities exhibited the most pronounced ecological differentiation, with combined specialist and generalist taxa comprising 44.5% of the community, which was more than double the proportion observed in viral communities. This represents the highest degree of clear niche differentiation across all microbial groups, with specialists (23.2%) and generalists (21.3%) showing nearly equivalent proportions. The strategy of prokaryotic communities for maintaining both high specialization and high generalization suggests sophisticated ecological partitioning within the chicken gut environment. This pattern indicates that bacteria have successfully evolved to exploit both stable, resource-rich microniches (favoring specialists) and variable, heterogeneous environments (favoring generalists) simultaneously within the same ecosystem. The substantial proportion of specialists (23.2%) reflects the complex spatial and chemical heterogeneity of the chicken gut, where bacteria can evolve specific adaptations to exploit distinct metabolic niches, pH gradients, mucus layers, and substrate availabilities. Moreover, the equally high proportion of generalists (21.3%) indicates that many prokaryotic taxa have evolved broad ecological tolerance ranges, providing crucial functional redundancy and ecosystem stability. The remaining 55.5% of prokaryotic taxa showing non-significant niche differentiation likely represent ecologically transitional populations that maintain intermediate strategies, possibly reflecting ongoing evolutionary processes or taxa that exploit resources across multiple niche boundaries. Both DNA and RNA viral communities demonstrated remarkably similar and significantly lower levels of ecological differentiation compared to prokaryotes. DNA viruses showed only 19.7% combined specialists and generalists, while RNA viruses exhibited 18.8%, representing less than half the differentiation capacity of prokaryotic communities.

Positive correlations between SVs and SNVs across microbial groups

We conducted a comprehensive analysis of chicken fecal microbial variations by integrating NGS and ONT datasets. This approach combines the high sensitivity of NGS for detecting SNVs with the superior capability of ONT for resolving SVs, effectively overcoming the limitations of each standalone technology (Fig. 3A, Fig. S4). Analysis of variant relationships revealed consistent positive correlations between SV and SNV densities across all microbial groups, though with varying strengths (Fig. 3B). Prokaryotic genomes showed the strongest positive correlation (Spearman’s r = 0.66, p < 0.001), followed by DNA viruses (r = 0.48, p < 0.001) and RNA viruses (r = 0.28, p < 0.001). This direct relationship suggests that structural and point mutations act synergistically rather than as alternative evolutionary strategies, indicating coordinated mutational processes across different microbial groups.

Fig. 3.

Structural variants (SVs) and single-nucleotide variants (SNVs) in prokaryotes and viruses in chicken feces. A Comparison of Illumina (NGS) and Oxford Nanopore Technologies (ONT) for SV and SNV detection in prokaryotic genomes. SV types include deletion (DEL), duplication (DUP), insertion (INS), inversion (INV), and translocation (TRA). B Correlations between the number of SVs and SNVs per 1 kb of genome. The red line represents the regression line, with the shaded region indicating the 95% confidence interval. C Proportion of genomes with SVs and SNVs. The percentage is calculated as the number of genomes containing at least one variant divided by the total number of genomes analyzed. D Number of nonredundant SVs and SNVs in prokaryotic, DNA viral, and RNA viral genomes. E SV and SNV numbers per 1 Kb viral genomes with different life types. *p < 0.05; **p < 0.01; ***p < 0.001

Differential variant distribution patterns among microbial communities

The proportion of genomes containing variants revealed striking differences among microbial groups, with DNA viruses showing the highest variant prevalence (Fig. 3C). DNA viral genomes exhibited the highest SNV prevalence at 86.9%, significantly exceeding both prokaryotic MAGs (26.9%) and RNA viruses (59.0%). For SVs, DNA viruses maintained a moderate prevalence (47.7%), while prokaryotic MAGs showed identical SV and SNV prevalence rates (26.9% each), and RNA viruses demonstrated the lowest SV prevalence at only 7.5%. This hierarchical pattern of variant distribution (DNA viruses > RNA viruses > prokaryotes for SNVs; DNA viruses > prokaryotes > > RNA viruses for SVs) suggests fundamentally different evolutionary constraints and mutational processes operating across microbial groups. The extreme disparity in RNA viral SVs (7.5% vs. 92.5% without SVs) indicates strong selection against large-scale genomic rearrangements in RNA viral populations. We merged and deduplicated SVs and SNVs from both prokaryotic and viral genomes across all samples. Prokaryotic genomes yielded 342,571 SVs and 19,849,753 SNVs. For viral genomes, DNA viruses contained 106,185 SVs and 5,867,044 SNVs, while RNA viruses contained 590 SVs and 136,832 SNVs (Fig. 3D).

Differential variant accumulation patterns across microbial phyla

Prokaryotic phyla demonstrated highly variable variant densities (Fig. S6A). Methanobacteriota demonstrated an extreme SNV-biased strategy with the highest point mutation density (average 71.33 per kb) but relatively modest SV (average 0.31 per kb), suggesting evolutionary optimization for fine-scale adaptive changes under anaerobic conditions. In contrast, Bacillota exhibited the highest structural variant tolerance (average 0.53 per kb) combined with elevated SNV levels (average 27.16 per kb), indicating exceptional genomic plasticity for both point mutations and large-scale rearrangements. This genomic plasticity likely stems from the spore-forming capability of Bacillota, which facilitates extreme condition survival and cross-host transmission [15, 16, 22]. The resulting cycles of transient colonization are likely to impose strong and varied selective pressures, favoring genomes capable of rapid adaptation through both point mutations and major structural changes. Synergistota and Halobacteriota also showed high SNV densities (average 41.64 and 41.78 variants per kb, respectively) but with lower SV accumulation, while most other prokaryotic phyla like Pseudomonadota (average 9.42 SNVs per kb, 0.12 SVs per kb) and Bacteroidota (average 21.34 SNVs per kb, 0.35 SVs per kb) presented intermediate levels.

DNA viral phyla presented more constrained variant patterns compared to prokaryotes, with Artverviricota emerging as the most variable phylum (average 44.22 SNVs per kb, 1.29 SVs per kb), followed by Cressdnaviricota (average 30.79 SNVs per kb, 1.22 SVs per kb) and Hofneiviricota (average 25.77 SNVs per kb, 0.92 SVs per kb) (Fig. S6B). Low-variability DNA viral phyla, including Peploviricota (average 2.89 SNVs per kb, 0.03 SVs per kb) and Cossaviricota (average 1.54 SNVs per kb), demonstrated minimal variant accumulation, suggesting phylum-specific evolutionary constraints on genomic variation. For RNA viral phyla, Kitrinoviricota dominated RNA viral SNV accumulation with the highest density (average 42.60 variants per kb) but showed moderate SV levels (average 0.45 variants per kb), while Pisuviricota exhibited the highest SV density among RNA viruses (average 1.04 variants per kb) with intermediate SNV levels (average 27.97 variants per kb).

Differential viral variant density patterns across host and genomic types

Bacteriophage infecting prokaryotes exhibited a significantly higher mean SV density (0.340 variants per kb) than eukaryotic-infecting viruses (0.148 variants per kb) (p < 0.001) (Fig. S7A). Conversely, the mean SNV density was greater in eukaryotic viruses (23.6 variants per kb) compared to prokaryotic viruses (16.4 variants per kb), although this difference was not statistically significant (Fig. S7B). Among viral categories of genomic architectures, ssDNA viruses showed the highest mean SV density (0.577 variants per kb), which was greater than that of dsDNA, dsRNA, and ssRNA viruses (Fig. S7C). Within RNA viruses, ssRNA viruses also possessed a higher mean SV density than dsRNA viruses (p < 0.001). For SNVs, ssRNA viruses displayed the highest mean density (23.2 variants per kb), significantly exceeding that of dsRNA viruses (p < 0.01; Fig. S7D). In addition, dsDNA viruses exhibited a higher mean SNV density than ssDNA viruses (p < 0.001; Fig. S7D). Specifically, high‑variant‑density viruses belonging to Cressdnaviricota and Hofneiviricota are ssDNA viruses infecting eukaryotes and prokaryotes, respectively. Low‑variant‑density viruses in Peploviricota and Cossaviricota are dsDNA viruses infecting eukaryotes. Kitrinoviricota, which showed high variant density, comprises primarily ssRNA viruses that infect eukaryotes.

Heterogeneous viral variant distribution across prokaryotic host phyla

Virus-prokaryote associations were predicted based on CRISPR spacer alignment as shown in Fig. S8. Analysis of bacteriophage variant patterns in relation to their prokaryotic hosts revealed substantial heterogeneity in variant densities across different bacterial phyla (Fig. S9). SV densities exhibited considerable variation among bacteriophage populations associated with different prokaryotic hosts. Bacteriophages associated with Spirochaetota showed the lowest average SV density (0.072 SVs per kb), consistent with the evolutionarily stable, low-selection-pressure niche of their hosts (0.138 SVs per kb for Spirochaetota), which predominantly reside in the homogeneous cecal mucus layer [22]. In contrast, bacteriophages infecting Bacillota displayed the highest average SV density (0.625 SVs per kb), reflecting the need for these bacteriophages to maintain high adaptive potential in response to the broad genetic variation inherent within their diverse host populations (0.525 SVs per kb for Bacillota). This approximately 8.7-fold variation across host phyla underscores how host-specific ecological and evolutionary dynamics fundamentally shape the genomic plasticity of bacteriophages. The SNV density demonstrated even more dramatic variation across host-associated bacteriophage populations, with densities ranging from an average of 3.68 variants per kb in Verrucomicrobiota-associated viruses to an average of 36.74 variants per kb in Bacillota-associated bacteriophages, representing a tenfold difference. These results reveal kingdom-specific evolutionary constraints governing viral variant accumulation, with Bacillota-associated bacteriophages demonstrating the highest genomic instability across both structural and point mutation categories.

Influence of the bacteriophage lifestyle on variant accumulation patterns

Analysis of bacteriophage variants by lifestyle revealed significant differences between temperate and virulent bacteriophages (Fig. 3E). Temperate bacteriophages accumulated significantly higher SV densities compared to virulent bacteriophages (p < 0.001). Similarly, temperate bacteriophages showed increased SNV densities relative to virulent bacteriophages (p < 0.001). When bacteriophages were compared with prophages, the pattern remained consistent, with prophages (integrated temperate bacteriophages) showing higher variant densities than free bacteriophages for both SVs and SNVs (p < 0.001).

Variant-associated functional enrichment patterns

KEGG functional enrichment analysis of prokaryotic variant-associated genes revealed distinct enrichment signatures between SVs and SNVs (Fig. 4A). Prokaryotic SV-associated genes were significantly enriched (p < 0.05) in three major metabolic pathways. Carbohydrate metabolism demonstrated the strongest enrichment signal (11,110 genes) with the highest significance, followed by energy metabolism (6972 genes) and biosynthesis of other secondary metabolites (2846 genes). Prokaryotic SNV-associated genes exhibited more limited but highly significant enrichment compared to SV patterns, with only two pathways (amino acid metabolism and metabolism of other amino acids) achieving statistical significance.

Fig. 4.

Functional enrichment analysis of SVs and SNVs in prokaryotic and viral genomes. A KEGG enrichment of SV- and SNV-related genes in prokaryotic genomes. B KEGG enrichment of SV- and SNV-related AMGs in viral genomes. The dotted line indicates the significance threshold (p = 0.05). Gene counts per functional category are shown in circles

SV-associated auxiliary metabolic genes (AMGs) in viruses demonstrated extensive and highly significant functional enrichment across six major metabolic pathways, including carbohydrate metabolism, glycan biosynthesis and metabolism, energy metabolism, lipid metabolism, metabolism of other amino acids, and metabolism of terpenoids and polyketides (p < 0.05) (Fig. 4B). This broad functional enrichment indicates that viral structural rearrangements enable comprehensive metabolic manipulation of hosts across diverse biochemical pathways.

SNV-associated AMGs in viruses exhibited enrichment in six pathways. Carbohydrate metabolism (236 genes) showed the greatest significance (171 genes), followed by glycan biosynthesis and metabolism and lipid metabolism (79 genes). Amino acid metabolism (258 genes), energy metabolism (93 genes), and metabolism of terpenoids and polyketides (35 genes) were also significantly enriched. The consistent pathway representation between viral SVs and SNVs suggests coordinated evolutionary optimization across both SV and SNV mechanisms.

Viral protein functional enrichment of variant-associated genes in chicken fecal microbiomes

Viral proteins were classified into different functional categories according to keywords listed in Table S2 and subjected to functional enrichment analysis to uncover enrichment patterns in SVs and SNVs in the fecal virome of chickens (Fig. 5). tRNA-related genes were significantly enriched (p < 0.001), demonstrating that large-scale genomic rearrangements are under extreme selective pressure to optimize the viral translation machinery (Fig. 5A). Integrase, conjugative system proteins, and transposition and transposase functions showed extraordinary enrichment, indicating that structural variants are critical for viral genome integration and HGT capabilities. The viral replication machinery exhibited coordinated significant enrichment patterns, with DNA binding proteins, polymerase, helicase, replication proteins, and primase achieving notable significance. Additional significantly enriched functions included membrane proteins, CRISPR-Cas systems, toxin-associated proteins, and virulence-associated proteins. Viral SNV-associated genes demonstrated similar enrichment patterns compared to SV-associated genes (Fig. 5B). tRNA-related, replication-related, and HGT-related functions showed remarkable enrichment significance. Notably, tapemeasure proteins demonstrated remarkable enrichment significance only in SNVs, indicating infection optimization.

Fig. 5.

Enriched viral functions of SV- (A) and SNV-related (B) genes in viral contigs. The dotted line represents a p value of 0.05. The number in the circle represents the number of genes in each viral functional category

Cross-kingdom and intra-kingdom transfer dynamics

The cross-kingdom and intra-kingdom gene transfer network across the ten chicken breeds is shown in Fig. S10. A Lactobacillus bacterium and a bacteriophage from the phylum Uroviricota showed the highest connectivity and were the most active participants in HGT, predominantly serving as donors that transferred genetic material to multiple recipients. We further analyzed HGT patterns of phage to phage (P-to-P), phage to bacteria (P-to-B), bacteria to phage (B-to-P), and bacteria to bacteria (B-to-B) transfer. P-to-B HGT events demonstrated the highest overall frequency with a mean of 5.28 ± 3.30 events per sample, significantly exceeding all other transfer types (Fig. 6 A and B). B-to-B transfer showed moderate activity with 4.10 ± 2.64 events per sample, while P-to-P transfer exhibited intermediate frequencies of 3.21 ± 2.10 events per sample. B-to-P transfer demonstrated the lowest activity with only 1.57 ± 0.92 events per sample. Through deduplication, 536 deduplicated HGT events were detected in the fecal microbiome (Fig. 6C). P-to-B transfers constituted the largest proportion (37.50%), followed by B-to-B transfers (30.60%), P-to-P transfers (21.83%), and B-to-P transfers (10.07%). The dominance of P-to-B transfers, alongside other phage-mediated routes, underscores the key role of viral elements in horizontal gene transfer. Together, phage-mediated transfers (P-to-P + P-to-B + B-to-P) accounted for 69.40% of all HGT events in the chicken fecal microbiome. Cross-kingdom genetic exchange (P-to-B + B-to-P) accounted for 47.57% of the total HGT activity, while intra-kingdom transfers (P-to-P + B-to-B) constituted 52.43% of the total HGT activity. However, the dramatic dominance of P-to-B transfers over B-to-P transfers reveals highly asymmetric cross-kingdom genetic flow, with viruses serving as primary genetic donors to bacterial populations rather than balanced bidirectional exchange. Intra-kingdom transfer patterns showed moderate imbalance, with B-to-B transfers exceeding P-to-P transfers by 8.77%, indicating slightly enhanced bacterial genetic exchange compared to viral genetic exchange.

Fig. 6.

Role of prokaryote-virus HGT in generating SVs and SNVs. A Average HGT event numbers of phage to phage (P-to-P), phage to bacteria (P-to-B), bacteria to phage (B-to-P), and bacteria to bacteria (B-to-B) in fecal samples of different chicken breeds. B Comparison among different types of HGT events. *p < 0.05; **p < 0.01; ***p < 0.001; NS not significant. C Proportions of HGT types among 536 deduplicated events. D Proportions of horizontally transferred genes with SVs and SNVs. E Numbers of SV- and SNV-associated horizontally transferred genes involved in various KEGG metabolic pathways in different types of HGT events. KEGG-annotated genes were only identified in SV-associated B-to-B, SV-associated B-to-P, SNV-associated B-to-B, SNV-associated B-to-P, and SNV-associated P-to-B

To assess the prevalence of SVs and SNVs within horizontally transferred genes, we analyzed the percentage of genes involved in each HGT type that were associated with these variants. As shown in Fig. 6D, SVs and SNVs were extremely frequent in HGT-transferred genes. Specifically, the prevalence of SVs was 100% in both B-to-B and B-to-P transferred genes, 99.5% in P-to-P transferred genes, and 96.0% in P-to-B transferred genes. The representation of SNVs was also remarkably high across all HGT types: P-to-B transferred genes showed the highest SNV association at 99.4%, followed by P-to-P (98.5%), B-to-B (97.4%), and B-to-P (96.2%) transfers. In stark contrast, among non-HGT genes, only 70.0% were associated with SVs and 71.5% with SNVs. This comparative analysis clearly indicates that genes acquired via HGT are significantly more enriched for both structural and single-nucleotide variants than genes not involved in horizontal transfer.

Functional distribution of HGT-associated variants

KEGG metabolic pathway analysis of HGT-associated variant genes revealed transfer type-specific specialization patterns (Fig. 6E). Only genes involved in SV-associated B-to-B, SV-associated B-to-P, SNV-associated B-to-B, SNV-associated B-to-P, and SNV-associated P-to-B were annotated by KEGG. The SV- and SNV-associated horizontally transferred genes were primarily involved in amino acid metabolism, followed by moderate activity in biosynthesis of secondary metabolites, carbohydrate metabolism, and energy metabolism.

Given the significant role of viruses in horizontal gene transfer, we classified horizontally transferred genes harboring variants into viral functional categories to analyze associated proteins (Fig. S11). Transposition and transposase genes demonstrated exceptional representation as the largest gene category, with B-to-B transfers showing 176 SV and 175 SNV genes, P-to-B transfers showing 119 SV and 126 SNV genes, B-to-P transfers showing 77 SV and 76 SNV genes, and P-to-P transfers showing 30 SV and 31 SNV genes. Integrase genes showed substantial secondary representation with nearly identical SV-SNV distributions: B-to-B (35 genes for both), P-to-B (22–23 genes), B-to-P (15–17 genes), and P-to-P (1 gene for both). Resolvase genes exhibited virus-mediated transfer preferences, with P-to-B and P-to-P transfers showing 11–12 genes each, while bacterial transfers showed minimal representation (1–3 genes). Assembly functions showed exclusive bacteria-mediated transfer, with membrane proteins and assembly proteins appearing only in B-to-B and B-to-P transfers. Immune evasion functions revealed bacterial transfer dominance, with aminotransferase and antibiotic resistance genes showing equal representation in B-to-B and B-to-P transfers.

Integration functions, including transposition and transposase, integrase, resolvase, and recombination and recombinase, dominated all HGT types, accounting for 51.3 ~ 92.7% of transferred genes with variants across different transfer mechanisms. Lysis functions showed transfer type-specific specialization, with bacterial transfers (B-to-B, B-to-P) demonstrating hydrolase dominance for general cell wall degradation, while viral transfers (P-to-B, P-to-P) showed virulence-associated protein preference for targeted pathogenicity enhancement. Immune evasion functions, including aminotransferase, antibiotic resistance, methylase, and methyltransferase, showed bacteria-centric distributions. Packaging functions demonstrated virus-specific specialization with terminase genes appearing exclusively in P-to-P transfers.

Key drivers of genomic variants in the chicken gut microbiome

Focusing exclusively on intrinsic genomic and ecological factors, random forest model analysis was conducted to identify key genomic drivers underlying SVs and SNVs in prokaryotic, DNA viral, and RNA viral populations (Fig. 7). For prokaryotic SVs, SNV density (SNV number per 1 Kb) emerged as the dominant predictor (39.58%, p < 0.01), followed by HGT receptor number (15.79%, p < 0.01), phylogenetic classification at the phylum level (15.17%, p < 0.01), HGT donor number (9.88%, p < 0.01), and GC content (9.11%, p < 0.05), achieving an overall model R² of 0.69 (p < 0.01) (Fig. 7A). For prokaryotic SNVs, SV density (SV number per 1 Kb) dominated the predictive model (54.19%, p < 0.01), indicating a strong bidirectional association between structural and point mutations. Niche breadth was the secondary predictor (24.01%, p < 0.01), followed by HGT receptor number (19.61%, p < 0.01) and HGT donor number (11.11%, p < 0.05) (Fig. 7B). DNA virus variant prediction revealed fundamentally different driver hierarchies compared to prokaryotic patterns. For DNA virus SVs, SNV density maintained the greatest importance (40.76%, p < 0.01), followed by GC content (7.54%, p < 0.01) (Fig. 7C). DNA virus SNVs exhibited dramatically different driver patterns with niche breadth as the overwhelmingly dominant predictor (112.89%, p < 0.01), followed by SV density (99.27%, p < 0.01), host number (47.34%, p < 0.01), GC content (46.16%, p < 0.01), and genomic length (31.22%, p < 0.01) (Fig. 7D). RNA virus variant prediction revealed unique driver patterns distinct from both prokaryotic and DNA virus models. For RNA virus SVs, genomic length was the primary predictor (14.03%, p < 0.01), followed by GC content (11.59%, p < 0.01) and phylogenetic classification (7.39%, p < 0.05) (Fig. 7E). RNA virus SNVs demonstrated niche breadth dominance (49.68%, p < 0.01) similar to DNA virus SNV patterns, followed by phylogenetic classification (27.74%, p < 0.01), SV density (25.45%, p < 0.01), GC content (24.38%, p < 0.01), and genomic length (21.71%, p < 0.01) (Fig. 7F). Bidirectional SV-SNV associations were consistently observed across all microbial kingdoms, with SNV density predicting SV occurrence and SV density predicting SNV occurrence, indicating fundamental genomic instability coupling.

Fig. 7.

Key drivers on SV and SNV occurrence in prokaryotes (A, B), DNA viruses (C, D), and RNA viruses (E, F) revealed by random forest model-based analysis of genomic principal components. SV and SNV densities are expressed as counts per 1 Kb. HGT receptor/donor number refers to the number of microbial receptors and donors involved in HGT events. Niche breadth indicates the range of chicken breeds a microbe can inhabit, quantified using Levin’s niche breadth index. Host number refers to the number of prokaryotic hosts for viruses. Infected phage number refers to the number of bacteriophages infecting prokaryotes. Variable importance was determined by the percentage increase in mean squared error (%IncMSE). *p < 0.05; **p < 0.01; ***p < 0.001

Pearson correlation analysis was further performed to assess the associations between SV/SNV density and key genomic/ecological features in prokaryotes, DNA viruses, and RNA viruses. As shown in Fig. S12, niche breadth was positively correlated with SV and SNV density in both prokaryotes and viruses (p < 0.001). HGT receptor and donor numbers also showed significant positive correlations with variant density in prokaryotes and DNA viruses (p < 0.001). GC content exhibited a more complex pattern: it was negatively correlated with SV density in prokaryotes (p < 0.05) and with both SV and SNV density in DNA viruses (p < 0.001). In RNA viruses, GC content was negatively correlated with SNV density (p < 0.001) but showed a positive correlation with SV density (p < 0.01).

Discussion

Integrated metagenomic and metatranscriptomic analyses successfully recovered 1527 prokaryotic MAGs and 39,422 viral contigs from the ten chicken breeds, revealing complex gut microbiome dynamics. The prokaryotic community was dominated by Bacillota, with beneficial genera including Lactobacillus, Ligilactobacillus, and Limosilactobacillus, consistent with established chicken fecal microbiome profiles essential for carbohydrate fermentation and pathogen exclusion [23]. DNA viral communities were overwhelmingly dominated by Uroviricota (99.2%), primarily bacteriophages that regulate bacterial populations. However, approximately 96.0% of DNA viral sequences remained unclassified at the family level, representing one of the most significant taxonomic gaps in gut virome studies and highlighting the presence of novel viral lineages in avian gut environments. RNA viral communities showed greater phylogenetic diversity across multiple phyla, including Lenarviricota, Pisuviricota, Duplornaviricota, Kitrinoviricota, and Negarnaviricota, with Lenarviricota and Pisuviricota being the dominant phyla. Lenarviricota dominated the RNA viral community, primarily represented by the Mitoviridae family. The prevalence of Mitoviridae likely stems from multiple sources, including fungal infections, plant-derived dietary components, and potentially invertebrate-associated viruses from the chicken environment, given their documented presence across diverse host taxa spanning fungi, plants, and invertebrates [24]. The abundance of Pisuviricota indicates a substantial representation of RNA viruses that commonly infect both prokaryotic and eukaryotic hosts, reflecting their broad host range and adaptive capacity in the gut ecosystem [25].

The prediction that the majority of DNA viruses infect bacteria, specifically Bacillota and Bacteroidia (targeted by Uroviricota bacteriophages), reinforces the observed dominance of tailed phages and suggests tight phage-bacterium co-evolution [26], particularly with the abundant Bacillota and Bacteroidota. The prediction of a subset of phages potentially infecting archaea is intriguing, though the presence and abundance of archaea in the chicken gut warrant further confirmation. The identification of viruses predicted to infect Eukaryota, potentially including vertebrates, plants, protozoan parasites, or fungi, offers a valuable direction for future research into eukaryotic pathogens or commensals within the chicken gut ecosystem. The finding that 24.5% of the recovered RNA viruses were predicted to infect bacteria (Fig. 1F) expands our knowledge of the chicken fecal virome, which has historically been biased toward DNA viruses, and emphasizes the need for combined metagenomic and metatranscriptomic approaches for comprehensive virome characterization.

Prokaryotic communities demonstrated superior niche differentiation with 44.5% combined specialists and generalists, which was 2.4-fold higher than viral communities (~ 19%). The prokaryotes with nearly equivalent specialist (23.2%) and generalist (21.3%) proportions exploit gut ecosystem complexity through specialized niche adaptations and broad tolerance ranges that provide ecosystem stability. These contrasting differentiation patterns directly determined assembly mechanisms across microbial kingdoms. High prokaryotic ecological differentiation capacity enabled a variable neutral model fit across breeds, reflecting dynamic deterministic-neutral interactions where specialized taxa respond strongly to gut environment selection while generalists maintain stochastic dispersal patterns. Conversely, low viral differentiation capacity (~ 19% specialists and generalists combined) constrained communities to predominantly stochastic assembly patterns with consistent neutral fits, as limited niche specialization reduces deterministic selection pressure [27]. This integration demonstrates that ecological differentiation capacity is a factor shaping community assembly mechanisms, as enhanced niche differentiation supports flexible assembly dynamics while restricted differentiation promotes neutral processes.

Positive correlations between SVs and SNVs across all microbial groups indicate parallel rather than alternative mutational patterns, with prokaryotic genomes showing stronger parallel mutation occurrence. The positive correlations imply that genomes experiencing high rates of SV also tend to accumulate more point mutations, suggesting shared underlying mechanisms driving genomic instability or adaptive evolution. This pattern may reflect environmental pressures in the chicken gut that simultaneously promote both large-scale genomic rearrangements and fine-scale sequence changes, creating a mutagenic environment that enhances overall genomic plasticity. Such concurrent emergence of structural and nucleotide-level variations is consistently observed in human gut ecosystems, where the gene exchange between bacteriophages and bacteria drives extensive SVs alongside SNPs, as evidenced by long-read sequencing [13]. The gradient in correlation strength (prokaryotes > DNA viruses > RNA viruses) suggests that larger, more complex genomes exhibit stronger coordination between different types of mutations, while smaller viral genomes show more independent mutational dynamics.

Prokaryotic communities employed contrasting evolutionary strategies through SVs and SNVs for gut adaptation. SVs drive broad metabolic network restructuring with the greatest enrichment in carbohydrate metabolism, followed by energy metabolism and secondary metabolite biosynthesis, reflecting comprehensive metabolic reorganization for competitive advantage in fluctuating nutritional environments. Conversely, SNVs exhibited extreme specialization in amino acid processing pathways with dramatic enrichment intensities, indicating exceptional selective pressure for nitrogen metabolism optimization in protein-rich gut environments. This bifurcated strategy enables large-scale metabolic flexibility through structural rearrangements while maintaining fine-scale optimization through targeted point mutations [14].

Viral communities demonstrated sophisticated metabolic hijacking through extensive AMG evolution across both variant types. Viral SVs and SNVs show exceptional carbohydrate metabolism enrichment with coordinated multi-pathway intervention spanning carbohydrate, glycan, energy, lipid, amino acid, and terpenoid metabolism, indicating comprehensive metabolic network takeover. This variation pattern suggests evolutionary convergence toward maximum metabolic manipulation efficiency, where both structural and point variants contribute to optimal host resource extraction. Viral metabolic interventions experience extreme selective pressure due to their absolute dependence on host resources and limited genomic capacity for metabolic autonomy [28]. A coordinated enrichment of replication-assembly machinery components was observed in viruses, revealing integrated optimization of the complete viral reproductive cycle from translation up to particle assembly. The coordinated enhancement across all three systems indicates that viral fitness depends on integrated reproductive machinery optimization, where improved translation capacity must be matched by enhanced replication fidelity and precise assembly mechanisms for successful competitive reproduction in dynamic gut ecosystems [29]. The marked enrichment of integrase and transposition functions across both variant types underscores that HGT is inherently linked to genomic instability, with mobile elements serving as hotspots for both large-scale rearrangements and point mutations to facilitate effective genetic exchange [30]. This consistently high significance indicates that viral mobility is under continuous selective pressure in the dynamic gut microbiome, where rapid adaptation to changing bacterial host populations is essential for viral persistence.

HGT analysis indicated a significant viral contribution to cross-kingdom genetic exchange. Specifically, P-to-B transfers led to genetic innovation (37.50% of total events, 5.28 ± 3.30 per sample) with a 3.4-fold higher frequency than reverse transfers, highlighting the fundamental role of viruses as genetic mobilization networks connecting phages and bacteria. Gut viral communities serve as dynamic genetic reservoirs, continuously redistributing adaptive elements across microbial populations through lysogenic conversion, transduction, or recombination-mediated integration [31]. The consistency of this pattern across diverse chicken breeds indicates that bacteriophage-mediated HGT represents a fundamental ecological process. It serves as a key mechanism for bacterial adaptation to the dynamic gut environment, enabling the rapid acquisition of metabolic capabilities, stress resistance, or colonization factors.

A marked enrichment of SV- and SNV-associated genes encoding virulence-associated proteins was observed in bacteriophage-mediated HGT (P-to-B, P-to-P), indicating a primary driver in enhancing bacterial pathogenicity over general metabolic capacity [32]. The scarcity of metabolic pathway annotations in these transfers further underscores that bacteriophages act as key vectors for the horizontal dissemination of pathogenicity factors. This virulence-focused exchange mediated by bacteriophages provides bacterial hosts with enhanced infectivity, which may in turn facilitate their own replication and dissemination, thereby optimizing traits directly linked to the bacteriophage life cycle [33]. In contrast, bacteria-centric HGT events (B-to-B and B-to-P) exhibit a distinct profile dominated by metabolic pathway genes with variants. This metabolic gene core facilitates comprehensive genetic sharing networks that simultaneously enhance community-wide metabolic efficiency. Coupled with the co-transfer of immune evasion functions, these bacteria-centric networks appear to be finely tuned for optimizing both survival mechanisms and nutrient utilization, underscoring a strategy focused on ecological fitness and resilience [34]. The detection of antibiotic resistance gene (ARG) variants in both B-to-B and B-to-P transfers reveals a dual-pathway dissemination system. Bacterial communities share ARGs horizontally among themselves, where these genes confer resistance. Simultaneously, ARGs can be packaged into phage genomes. These phages then act as passive vectors and mobile genetic reservoirs, facilitating the redistribution of ARGs back to bacterial hosts. This establishes a persistent circulation of resistance genes, enhancing bacterial survival during antibiotic exposure in the gut [9].

Random forest analysis revealed bidirectional SV-SNV associations indicative of fundamental genomic instability coupling, wherein structural variants act as primary genomic disruptors that establish persistent mutational hotspots [35]. Niche breadth emerged as the dominant predictor for viral SNVs, revealing ecological diversification as the primary evolutionary driver of viral variants. Its secondary predictive role in prokaryotic models indicates that ecological optimization contributes substantially to bacterial evolution. Phylogenetic constraints operated kingdom-specifically, with bacterial phyla constraining structural rearrangements while permitting flexible point mutations. GC content showed extreme compositional sensitivity in DNA and RNA viruses due to secondary structure dependencies. Bacteria as HGT receptors drove genomic variation, while host number emerged as a major driver of DNA virus SNVs, indicating that viral host range expansion significantly influences genetic diversification. Substantial P-to-B transfer events indicated that bacteria served as primary recipients of viral genetic material, accumulating both SVs and SNVs through lysogenic integration processes that were enhanced by broader viral host specificity networks. This study revealed the genomic variants across kingdoms in chicken fecal microbiomes. These findings suggest that kingdom-specific evolutionary constraints may provide targets for precision microbiome engineering, stability optimization, and population control in chicken gut ecosystems, thereby advancing our understanding of microbial evolution in poultry systems. Besides, these findings highlight the critical importance of preserving not only the genetic diversity of local chicken breeds but also their associated microbiome characteristics.

Methods

Yunnan indigenous chicken breeds and sample collection

Fecal samples were collected from ten distinct indigenous chicken breeds across various regions in Yunnan Province, China. All chickens were raised and immunized in strict compliance with China’s Feeding and Management Guidelines for Yellow-feathered Chickens (NY/T 1871–2010). During the brooding phase (0–7 weeks), chicks were housed in an open-system facility and provided with starter feed. Subsequently (8–30 weeks), chickens were transferred to cage housing and fed a grower diet. All feed formulations adhered to the Nutrient Requirements for Yellow-feathered Chickens (NY/T 3645–2020). Fresh fecal samples were collected from individual chickens (aged 200 ± 20 days) immediately after defecation during the afternoon. Only feces exhibiting a dry, cylindrical, or elongated morphology, indicative of small-intestinal origin, were selected. Each sample was immediately transferred into a sterile centrifuge tube, flash-frozen in liquid nitrogen, and stored at − 80 °C until further processing. For each breed, fecal samples were collected from 10 individual chickens (5 males and 5 females) to ensure representative sampling. The ten chicken breeds and their corresponding collection locations are detailed in Supplementary Table S1, including Labai high-leg chicken (LB), Taliu black-boned chicken (TL), Wuliangshan black-boned chicken (WLS), Yunlong short-leg chicken (YL), Wuding chicken (WD), Chahua chicken (CH), Nixi chicken (NX), Yanjin black-boned chicken (YJ), Lanping silky chicken (RM), and Ake chicken (AK).

Following nucleic acid extraction, individual DNA samples from all 10 chickens per breed were subjected to Illumina metagenomic sequencing, generating 100 Illumina metagenomic datasets in total (Table S3). For ONT metagenomic sequencing, DNA samples from 5 chickens per breed were pooled, generating two pooled DNA samples per breed for long-read sequencing. Due to insufficient DNA quality in 4 samples, 16 ONT datasets were successfully generated (Table S4). These pooled samples also underwent Illumina metagenomic sequencing for hybrid assembly approaches and comparison of their performance on variant detection (Table S4). For metatranscriptomic analysis, RNA samples from 6 chickens per breed were subjected to sequencing, with 3 samples failing quality control, ultimately yielding 57 metatranscriptomic datasets (Table S5). All metagenomic and metatranscriptomic datasets were deposited in the National Genomics Data Center (NGDC), with detailed information listed in Supplementary Table S3, Table S4, and Table S5.

DNA extraction, and short-read and long-read metagenomic sequencing

Total genomic DNA was extracted from all chicken fecal samples using the Mag-Bind® Soil DNA Kit (Omega Bio-tek, USA) following the manufacturer’s instructions. The DNA concentration was determined by a NanoDrop 2000 (Thermo Fisher Scientific, USA), and its quality was checked through 1% agarose gel electrophoresis.

For Illumina metagenomic library construction, extracted genomic DNA was fragmented to average fragment sizes of approximately 350 bp using the Covaris M220 system (Gene Company Limited, China) with optimized shearing parameters. Paired-end sequencing libraries were prepared employing the NEXTFLEX Rapid DNA-Seq Kit (Bioo Scientific, USA) according to the manufacturer’s instructions. Indexed adapters containing complete sequencing primer hybridization sites were ligated to blunt-ended DNA fragments. High-throughput sequencing of Illumina metagenomic libraries was conducted on an Illumina NovaSeq X Plus (Illumina, USA) at Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China). Illumina sequencing generated approximately 10 Gb of paired-end 150 bp reads per sample.

ONT sequencing libraries were constructed using the following protocol: (1) Magnetic bead-based size selection of target DNA fragments followed by DNA damage repair and end-repair procedures, with A-base addition to 3′ termini and subsequent magnetic bead purification; (2) Barcode addition using the EXP-NBD 114 kit (Oxford Nanopore Technologies, UK) followed by additional magnetic bead purification; (3) Adapter ligation using the SQK-LSK109 Ligation Sequencing Kit (Oxford Nanopore Technologies, UK). High-throughput sequencing of ONT sequencing libraries was conducted on the Nanopore PromethION R9.4 (Oxford Nanopore Technologies, UK) platform at Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China). ONT sequencing yielded approximately 10 Gb of long-read data per sample.

RNA extraction and metatranscriptomic sequencing

Total RNA was extracted and purified from chicken fecal samples using the Soil RNA Extraction Kit (Majorbio, Shanghai, China) following the manufacturer’s procedure. The RNA concentration, purity, and integrity were comprehensively evaluated using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA) and an Agilent 5300 Bioanalyzer system (Agilent Technologies, USA) to ensure high-quality RNA suitable for downstream transcriptomic applications.

Ribosomal RNA removal from metagenomic and metatranscriptomic datasets was accomplished using the RiboCop rRNA Depletion Kit for Mixed Bacterial Samples (Lexogen, Vienna, Austria). Strand-specific cDNA synthesis and library construction were conducted using the Illumina® Stranded mRNA Prep, Ligation Kit (Illumina, USA), according to the manufacturer’s protocol. Metatranscriptomic sequencing was performed on the Illumina NovaSeq X Plus sequencer (Illumina, USA) employing paired-end sequencing methodology at Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China). Approximately 10 Gb of transcriptomic data per sample were sequenced.

Illumina and ONT metagenomic data assembly

Metagenomic and metatranscriptomic datasets were processed using an established analysis pipeline [9, 36]. Integrated analysis of Illumina and ONT metagenomic sequencing data was implemented to achieve enhanced genome-centric resolution of microbial community structure and function. Illumina metagenomic reads underwent adapter trimming and quality control using Fastp (v0.23.2) [37], followed by assembly using MEGAHIT (v1.2.9) [38] with k-mer sizes ranging from 21 to 127. ONT long-read data were subjected to quality filtering using Filtlong (https://github.com/rrwick/Filtlong), followed by assembly using metaFlye (v2.9.2) [39]. Assembled ONT contigs underwent sequential polishing procedures using Medaka (v1.7.3) (https://github.com/nanoporetech/medaka) for initial consensus correction, followed by Racon (v1.4.3) (https://github.com/lbcb-sci/racon) for iterative consensus improvement, and polishing with Pilon (v1.24) [40] using Illumina short reads for base-level accuracy enhancement.

Prokaryotic MAG recovery and taxonomic classification

Prokaryotic MAGs were recovered using two binning strategies: one method involved binning the Illumina metagenomes via MetaWRAP (v1.3) [41], while the other method utilized BASALT (v1.1.0) [42] for hybrid binning of both Illumina and ONT data. The completeness and contamination of prokaryotic MAGs were assessed using CheckM2 (v1.1.0) [43]. MAGs with completeness ≥ 50% and contamination ≤ 10% were retained for further analysis. Prokaryotic MAG dereplication was performed to generate representative MAGs using dRep (v3.4.3) [44] at a 95% average nucleotide identity [45] threshold with the following parameters: “–S_algorithm fastANI -cm larger -pa 0.9 -sa 0.95 -nc 0.3 -comp 50 -con 10.” Taxonomic classification of prokaryotic MAGs was performed using GTDB-Tk (v2.4.0) [46].

Eukaryotic MAG recovery and taxonomic classification

Eukaryotic MAGs (specifically targeting fungi) were identified from the preliminary MAG set recovered using MetaWRAP’s binning module. All preliminary MAGs underwent taxonomic classification via the Contig Annotation Tool (CAT) (v6.0.1) [47], retaining only those designated as eukaryotes for subsequent eukaryotic MAG analysis. MAGs were filtered based on size (> 1 Mb) and quality metrics using EukCC (v.2.1.3), applying the criterion: completeness – contamination ≥ 30. Finally, dereplication of eukaryotic genomes was performed with dRep (v3.4.3) at a 95% ANI threshold.

DNA viral contig identification and taxonomic classification

Viral sequences were systematically identified from metagenomic assemblies using a comprehensive multi-tool approach. Assembled contigs underwent initial length filtering with a minimum threshold of 2.5 kb. Putative viral contigs were identified through parallel implementation of three viral detection tools: geNomad (v1.8.1) [48], VirSorter2 (v2.2.4) [45], and DeepVirFinder (v1.0) [49]. For geNomad classification, contigs were retained as candidate viral sequences if they achieved a virus_score ≥ 0.8 and contained at least one viral hallmark gene. VirSorter2 candidates required maximum scores ≥ 0.8, while DeepVirFinder candidates needed scores ≥ 0.8 with p values < 0.05 to minimize false positive predictions. Quality assessment and validation of candidate viral contigs were performed using CheckV (v1.0.1) [50] to evaluate genome completeness and contamination levels. Contigs flagged by CheckV as not-determined were excluded from downstream analyses to ensure a high-confidence viral dataset. Additionally, CheckV detected prophages and removed host-derived contaminations from viral sequences. Viral contig dereplication and clustering were performed using the method described by Nayfach et al. [51] with thresholds of 95% ANI and 85% alignment fraction to generate representative viral contigs.

RNA viral contig identification and taxonomic classification

RNA virus identification from metatranscriptomic datasets was performed using a pipeline designed to detect RNA-dependent RNA polymerase (RdRP)-containing viral sequences. Raw metatranscriptomic reads were subjected to quality control and adapter trimming using Fastp (v0.23.2). A rRNA reference database comprising SSU rRNA (including 16S and 18S variants) from SILVA, LSU rRNA (23S and 28S) from SILVA, and 5S/5.8S rRNA sequences from Rfam was constructed to eliminate ribosomal contamination via sequential alignment using minimap2 (v2.28) [52]. Additionally, host contamination was removed by aligning reads against the chicken reference genome (GCA_000002315.5) using minimap2 to exclude host-derived transcripts from downstream viral identification procedures. Quality-filtered and decontaminated metatranscriptomic reads were assembled using rnaSPAdes (v3.15.5) [53]. Open reading frames (ORFs) within assembled contigs were predicted using Prodigal-gv (v2.11.0) [48].

RNA virus identification was accomplished through the detection of RdRP proteins, which serve as universal markers for RNA viruses. An RdRP reference protein database was constructed by downloading all available RdRP protein sequences from the NCBI GenBank database. Predicted proteins from assembled contigs were searched against the RdRP reference database using DIAMOND BLASTp (v2.1.9) [54] with stringent parameters: “–evalue 1e-5 –id 70 –min-score 60 –query-cover 50” to identify putative RdRP-containing sequences. Complementary RdRP detection was performed using profile hidden Markov models (HMMs) through hmmsearch (HMMER v3.4) [55] against curated RdRP HMM profiles derived from the previous study [56]. This approach enhances detection sensitivity for divergent RdRP sequences that may be missed by sequence similarity searches alone. Candidate RNA virus contigs containing putative RdRP proteins were subjected to validation using PalmScan2 (v2.0) [57] to verify the presence of characteristic palmprint motifs (A, B, and C) within the RdRP core domain. These well-conserved palmprint motifs are essential structural features of authentic RdRP proteins and serve as definitive markers for RNA virus identification. Contigs lacking complete palmprint motifs were excluded as potential false positives to ensure high-confidence RNA virus predictions. Viral sequence dereplication and clustering were performed following the method described by Nayfach et al. [51] with thresholds of 95% ANI and 85% alignment fraction to generate representative viral contigs.

Viral taxonomic classification and host prediction

Taxonomic assignment of viral contigs was performed using a hierarchical classification strategy adhering to ICTV nomenclature standards. For DNA viral contigs, primary taxonomic classification was conducted using geNomad (v1.8.1). DNA viral contigs that remained unclassified following geNomad analysis were subjected to secondary annotation using CAT (v6.0.1) with taxonomic assignment based on homology searches against a customized viral database derived from the NCBI NR database. For RNA viral contigs, primary taxonomic classification was performed using VITAP (v1.7.1) (https://github.com/DrKaiyangZheng/VITAP). RNA viral contigs that remained unclassified after VITAP analysis were subjected to complementary annotation using CAT (v6.0.1).

Host type predictions and viral genome architecture classifications were determined based on taxonomic assignments cross-referenced with the Virus Metadata Resource (VMR) database (https://ictv.global/vmr/current) maintained by ICTV. This approach enabled systematic categorization of viral contigs according to their predicted host ranges and genomic characteristics based on established viral taxonomy standards. Bacteriophage-specific host prediction was enhanced through CRISPR-spacer matching analysis to identify precise virus-host interactions within the chicken gut microbiome. CRISPR arrays were extracted from reconstructed prokaryotic MAGs using CRISPRCasTyper (v1.8.0) [58]. Extracted CRISPR spacers were aligned against viral contigs using BLASTn with parameters specifically optimized for short sequence matching: “-task blastn-short -perc_identity 100 -penalty −1 -gapopen 10 -gapextend 2 -word_size 7.” Phage-prokaryote infection associations were identified when alignments met stringent criteria of ≥ 15 bp alignment length and ≥ 90% coverage. Bacteriophage lifestyle prediction was performed using PhaTYP (v2.1.12) [59] to classify phage-like contigs as either virulent (lytic) or temperate (lysogenic) based on genomic features and protein content analysis.

Microbial abundance calculation and ecological analysis

Relative abundances of representative MAGs were determined by aligning quality-filtered metagenomic reads against MAG sequences using CoverM (v0.6.1) (https://github.com/wwood/CoverM) with the following parameters: “genome –min-read-percent-identity 95 –min-read-aligned-percent 75 –min-covered-fraction 50 -m tpm.” Relative abundances of representative viral contigs were calculated using CoverM (v0.6.1) with the following parameters: “contig –min-read-percent-identity 95 –min-read-aligned-percent 75 –min-covered-fraction 75 -m tpm.” This approach provides normalized transcript per million (TPM) values enabling direct comparisons of microbial abundance patterns across samples.

Community assembly processes were evaluated using neutral community modeling (NCM) to assess the relative contributions of deterministic versus stochastic forces in shaping the microbial community structure. NCM analysis was applied separately to prokaryotic, DNA viral, and RNA viral communities to elucidate differential assembly mechanisms across these distinct biological components. Model fitting performance was quantified using the coefficient of determination (R²) calculated through the MicEco R package [60], where positive R² values indicate significant neutral model adherence. The migration parameter [21], representing the product of metacommunity size and migration frequency, along with the estimated migration rate (m), provides insights into dispersal limitations and demographic stochasticity effects on community assembly patterns.

Ecological niche characterization was performed using Levin’s niche breadth analysis to quantify habitat specialization patterns among microbial taxa across different chicken breeds. Niche breadth calculations were implemented using the EcolUtils R package. The statistical significance of niche specialization indices was assessed through permutation-based hypothesis testing using 1000 randomizations to establish robust confidence intervals. Taxa exhibiting habitat specialization values exceeding the upper 95% confidence threshold were classified as ecological generalists, while those falling below the lower 95% confidence limit were designated as habitat specialists.

SV detection using long reads

Representative prokaryotic MAGs (high-quality: ≥ 90% completeness, ≤ 10% contamination) along with DNA and RNA viral genomes served as references for variant detection. ONT reads were aligned to reference genomes using minimap2 (v2.28) with parameters optimized for ONT data (-ax map-ont). To identify high-confidence SVs in microbial genomes, we implemented a multi-caller approach utilizing four specialized long-read SV detection algorithms: Sniffles (v2.6.1) [61], cuteSV (v2.1.2) [62], and SVision-pro (v2.4) [63]. SV calls from individual detection tools were merged using SURVIVOR (v1.0.7) [64]. Only SVs detected by at least two independent callers with a minimum SV length of 30 bp were retained for downstream analysis.

SV detection using short reads

SV detection using Illumina short reads was performed to enhance detection sensitivity. Illumina reads were aligned to reference sequences using BWA-MEM (v0.7.18) [65]. Three short-read SV callers were employed: Manta (v1.6.0) [66], LUMPY (v0.3.1) [67], and GRIDSS (v2.13.2) [68]. SV calls from individual tools were merged using SURVIVOR (v1.0.7), retaining only SVs with a minimum SV length of 30 bp supported by at least two detection methods. The resulting SV sets from short-read and long-read approaches were integrated to generate a catalog of microbial SVs across all samples using SURVIVOR (v1.0.7).

SNV detection

SNVs were identified using a reference-based approach. For long reads, ONT reads were aligned to prokaryotic and viral reference genomes using minimap2 (v2.28). For short reads, Illumina reads were aligned to reference genomes using Bowtie2 (v2.5.4) [69]. SNV profiles were generated using inStrain (v1.3.1) [70] based on long-read and short-read alignments to references.

HGT detection in chicken fecal microbiome

HGT events across microbial kingdoms (prokaryotes and viruses) within the chicken fecal microbiome were identified using LocalHGT (v1.0.1) [71], a computational framework specifically designed for detecting recent HGT events with precision breakpoint resolution (insertion/deletion sites) from metagenomic data. This analysis was performed using representative prokaryotic and viral genomes as input sequences. Based on donor-recipient taxonomic relationships, HGT events were classified into four transfer categories: phage-to-phage (P-to-P), phage-to-bacteria (P-to-B), bacteria-to-phage (B-to-P), and bacteria-to-bacteria (B-to-B). Network visualization of HGT events was performed using Cytoscape (v3.10.3) [72] to illustrate the complex gene transfer relationships among microbial communities.

Functional annotation and enrichment analysis

ORFs within prokaryotic MAGs and viral contigs were predicted using Prodigal (v2.6.3) and Prodigal-gv (v2.11.0) [48], respectively. The functional annotation of predicted proteins was performed through a multi-database approach, querying against PFAM [73], VOGdb (http://vogdb.org/), and eggNOG [74] databases using a combination of eggNOG-mapper (v2.1.12) with an E value threshold of 0.001 and a minimum score of 60. Annotated genes were functionally classified into primary categories, including packaging, replication, infection, regulation, assembly, lysis, integration, immune evasion, transfer RNA, transduction, conjugation, and transformation. Each primary category was further subdivided into specific functional subcategories through keyword-based classification (Table S2) following a previously reported method [13].

Further functional annotation of ORFs was conducted using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database through KofamScan (v1.3.0) [75], which searches protein sequences against KOfam HMM profiles using profile HMMs for precise functional assignment. This annotation strategy enabled detailed metabolic pathway characterization of both prokaryotic and viral protein-coding sequences, providing insights into the functional potential and metabolic capabilities within the chicken gut microbiome ecosystem.

For viral contigs, AMGs were identified based on their genomic context relative to viral hallmark genes detected by geNomad. Specifically, genes located between pairs of viral marker genes within each viral contig were extracted and subjected to functional annotation analysis. Predicted genes that received KEGG annotations corresponding to metabolic pathways were classified as putative AMGs, representing virus-encoded functions that potentially modulate host cellular metabolism during infection.

Statistical analysis and random forest modeling analysis

Statistical analyses and data visualization were performed using R (v4.4.1). Differences between treatment groups were assessed using Kruskal–Wallis one-way analysis of variance, followed by pairwise Wilcoxon signed-rank tests for post-hoc comparisons. Statistical significance was set at p < 0.05 for all analyses.

Random Forest modeling was implemented using the randomForest package (v4.7–1.2) and rfPermute (v2.5.5) following the methodology described by Liao et al. [76] to identify key predictors of SV and SNV occurrence across microbial kingdoms. Key parameters were set to ntree = 500 and nrep = 1000. Models were initially trained on 70% of the data for parameter optimization, with the remaining 30% serving as the validation set. Final models incorporated all datasets and were evaluated through 1000 permutations to assess their statistical significance and cross-validated R² values. Variable importance was determined by the percentage increase in mean squared error [14] using out-of-bag estimates, where higher MSE percentages indicated greater predictor importance.

Authors’ contributions

Jiayu Zhang: Conceptualization; investigation; methodology; formal analysis; visualization; writing—original draft; writing—review and editing; data curation; funding acquisition. Le Xu: Conceptualization; investigation; methodology; formal analysis; writing—original draft; writing—review and editing; validation; data curation. Xuehai Ge: Investigation; formal analysis; data curation. Xiannian Zi: Investigation; methodology; resources; formal analysis. Shiyu Chen: Investigation; formal analysis; data curation. Chen Liu: Investigation; validation; formal analysis. Kun Wang: Resources; funding acquisition. Jinping Zhou: Investigation. Tengfei Dou: Resources. Jonathan W.C. Wong: Writing – review & editing; funding acquisition; supervision. Qiuye Lin: Writing – review & editing; project administration; supervision. Xiangtao Kang: Conceptualization; methodology; resources; supervision; writing – review & editing. Zhenhui Cao: Conceptualization; methodology; resources; funding acquisition; project administration; supervision; writing – review & editing.

Funding

This project was supported by the Key Project of Yunnan Province Agricultural Joint Special Project (No. 202301BD070001-136), the Yunnan Province Young and Middle-aged Academic and Technical Leader Reserve Talent Project (No. 202305AC160040), the Yunnan Fundamental Research Projects (No. 202401AU070079), the Guangdong Basic and Applied Basic Research Foundation (No. 2024A1515140076), and the Dongguan University of Technology Top Talent Professor Start Up Fund (No. 221110133).

Data availability

All the sequencing data have been deposited in NGDC under BioProject accession number PRJCA043783.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.