This study collected 16 samples of gut content from three distinct horse breeds (Akhal-Teke, Thoroughbred, and Yili). Thoroughbred and Yili horses were collected in Mu Lei, Xinjiang, China, whereas Akhal-Teke horses were sampled in Urumqi, Xinjiang, China. All horses were maintained under similar husbandry and management conditions and were housed individually in separate stables. Thoroughbred horses were provided ad libitum access to commercial formulated concentrates (either concentrate A: 10% Cool Mix, Connolly’s RED MILLS, Goresbridge, Kilkenny, Ireland; or concentrate B: MDN, Beijing Redrunm Technology Co., Ltd., Beijing, China) together with alfalfa hay. Yili and Akhal-Teke horses were fed mixed pasture forage supplemented with 5 kg/day of concentrate feed (concentrate C: Xinjiang Tiankang Feed Co., Ltd., Xinjiang, China) and alfalfa hay. All animals had free access to water and were supplied with sufficient feed according to local husbandry practices. None of the horses had received antimicrobial treatments (including antibiotics, anthelmintics, or non-steroidal anti-inflammatory drugs) within 2 months prior to sampling. Fresh fecal samples were collected rectally under sterile conditions and immediately transferred into pre-sterilized DNase- and RNase-free 50 mL centrifuge tubes. Samples were transported to the laboratory on dry ice and stored at −80 °C until nucleic acid extraction. Detailed information on the sampled horses and corresponding specimens is provided in Supplementary Table S1.

Total RNA was extracted from 0.5 g of fecal content per sample, following the manufacturer’s protocol, using TRIzol reagent (Invitrogen, USA) (15). Briefly, each sample was thoroughly mixed with 200 mg of 0.1 mm zirconia beads (BioSpec Products, USA) and 1.2 mL TRIzol reagent. The mixture was then homogenized using a FastPrep-24 instrument (MP Biomedicals, USA). RNA precipitation was then carried out using a mixture of isopropanol and a salt solution containing 1.2 M sodium chloride and 0.8 M disodium citrate. To remove residual DNA, 1 μL of DNase I (NEB, USA) was added to the precipitated RNA, and the mixture was incubated at 37 °C for 15 min. The integrity and concentration of total RNA were assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies, USA) (16) and NanoDrop 2000 (Thermo Fisher Scientific, USA) (17), respectively. Approximately 100 ng of purified RNA from each sample was used to construct a metatranscriptomic library with the TruSeq RNA Library Preparation Kit v2 (Illumina, USA) (18), following the manufacturer’s instructions. Library quality was evaluated using a high-sensitivity DNA chip on an Agilent Bioanalyzer. Paired-end sequencing (PE150 mode) was performed on the Illumina NovaSeq 6000 platform (Illumina, USA).

Quality control of all datasets generated from the Illumina sequencing platform was conducted using Fastp (version 0.23.2) (19) with default parameters, producing high-quality reads with a Phred quality score threshold of ≥20 (20). The clean reads were aligned to the horse reference genome (EquCab3.0, accession number GCA_002863925.1, downloaded from the NCBI) using Bowtie2 (version 2.2.4) with the “end-to-end, sensitive, -I 200, -X 400” parameters to remove any potential host contamination reads (21).

Clean data assembly was performed using MEGAHIT (v1.2.9) with default k-mer (subsequence) parameters (22). Contigs longer than 1,000 nucleotides were retained for further analysis (option: –min-contig-len 1,000) (23). The assembly was generated using the following parameters: –k-min 21, −-k-max 141, −-min-contig-len 1000 (24). The assembled contigs were then screened using Barrnap to identify those encoding rRNA genes, and any contigs containing rRNA sequences were removed (25). Open reading frames (ORFs) were predicted using Prodigal (v2.6.3) with the standard genetic code (parameters: -n -p meta) (26). Next, libraries were constructed for diamond comparison based on published RdRP (RNA-dependent RNA polymerase) sequences (23, 27–29). All predicted open reading frames (ORFs) were screened using DIAMOND BLASTp with the following parameters: coverage ≥ 50%, E-value ≤ 1e-10 and score ≥ 70 (23). These potential RdRP proteins were then compared against an HMM (Hid-den Markov Model) model constructed from these sequences using HMMsearch with the following parameters: E-value ≤ 1e-10 (11). This resulted in candidates that were considered to have nearly complete RdRP sequences. Contigs containing RdRP sequences identified by both DIAMOND and HMMsearch analyses were considered to represent RNA viruses with nearly complete RdRP genes.

Here, we performed taxonomic annotation on 497 RNA virus genomes using VITAP (VMR-MSL37 and NCBI RefSeqs209) (30). We conducted multiple sequence alignment of 515 RdRP-encoding contigs using MUSCLE (v7.520) (31) with the following parameters (−maxiterate 1000 -localpair -thread 1) (32). The resulting alignments were automatically trimmed in the “automated1” mode and further refined using TrimAl (v1.4) (33). An unrooted maximum-likelihood (ML) (34) phylogenetic tree was then constructed using IQ-TREE 2 (35) with the ultrafast bootstrap approximation to estimate the support values for the branches. The optimal models were determined by applying the ModelFinder algorithm (36), which was integrated into the IQ-TREE 2 software (options: -m MFP -B 1000 -bnni). Phylogenetic trees were visualized using Chiplot (37).

Alpha diversity indices were compared among different horse breeds using the Kruskal–Wallis (38) test to assess statistical significance (39). Based on the Bray–Curtis (40) distance index, similarity matrices were calculated for the different groups, and principal coordinate analysis (PCoA) (41) was employed to analyze β-diversity of the viral communities across the three horse breeds (42–44).

To evaluate the prevalence of identified RNA viruses in horses, high-quality reads were mapped to RdRP-containing contigs using Bowtie2 (v2.5.1) in the modified “–very-sensitive-local” mode (parameters: -D 20 -R 3 -N 0 -L 16 -i S,1,0.50 -I 0 -X 2000 –no-unal) (28). The resulting SAM files containing the mapped reads were converted into sorted BAM files using SAMtools (v1.17) for subsequent calculation of genome coverage and estimation of abundance (45). The TPM (Transcripts Per Million) (46) of the viral sequences were calculated using CoverM¹ with filters set at 90% nucleotide identity and 75% read coverage thresholds (47). TPM metric was used to represent the relative abundance of each viral genome (with the “-m tpm” option enabled in CoverM). Viral abundance matrices were first normalized prior to analysis, and differential comparisons were conducted using the MARS model implemented in DEGseq. In addition, to compare the differential RNA viruses in the gut of horses of different breeds, we analyzed the abundance of RNA viruses in the intestines of Thoroughbred, Akhal-Teke, and Yili horses to identify differences between the two groups (using the non-parametric Mann–Whitney U test) and calculated FDR (48) and log₂FC. Finally, we identified differential viruses based on thresholds (FDR < 0.05 and |log₂FC| > 1) (49).

To investigate the potential shared network of RNA viruses between horses, human, and other domesticated animals, we collected a series of gut metatranscriptome datasets from human (60), cow (18), and sheep (6) from the NCBI database (50–52). To ensure the comparability of the data, we screened them using the following criteria. First, the datasets must be metatranscriptomic or RNA virus-related. Second, only gut contents or fecal samples were included. Third, all samples must be collected from healthy individuals (e.g., not from diseased or deceased hosts). A total of 84 datasets were retrieved from public databases.

To assess potential cross-species occurrence of RNA viruses, viral contigs identified from horse metatranscriptomes were used as reference sequences for read mapping against metatranscriptomic datasets from cow, sheep, and human hosts. Mapping was performed using CoverM (mode: covered_fraction) with stringent parameters (≥90% nucleotide identity and ≥75% read alignment length) to reduce false-positive assignments. This strategy follows prior virome analyses employing high-identity thresholds to infer viral presence across host environments. Contigs with mapped reads covering ≥30% of their sequence length were considered putatively present in another host dataset. Shared viral members were then summarized at the family and genus, applying overage fraction cutoffs of 0.3 and 0.9, respectively.

We collected the gut contents from 16 healthy horses from three different breeds, including Yili horses (5), Akhal-Teke horses (6), and Thoroughbred horses (5). Individual RNA libraries were constructed for each sample and subjected to metatranscriptomic sequencing, yielding a total of 893,459,632 raw reads. After removing low-quality bases and adapter sequences and filtering out host-derived reads, 888,227,632 high-quality clean reads were retained for further analysis. The overall Q30 value reached 94.41%, indicating high sequencing quality (Supplementary Table S1).

High-quality reads were reassembled using MEGAHIT (v1.2.9), and contigs shorter than 1,000 bp were discarded (Supplementary Table S2). Subsequent analyses revealed that a total of 80,765 contigs were assembled from the clean datasets of the 3 horse breeds. The longest contig was 31,464 nucleotides in length. These overlapping clusters were annotated using Prodigal to predict encoded genes. We then identified RdRP proteins using both the BLASTp and HMM-based models (Supplementary Tables S3, S4), yielding 497 contigs carrying RdRP (the intersection of results from both methods). These contigs ranged in length from 1,000 to 29,446 bp, with an average length of 2,123.7 bp (Figure 1). These contigs represent a potential pool of RNA viruses in the horse gut. One giant RNA virus exceeding 20 kb (YL_3_k141_120563) in length was identified among them. This is the largest viral genome (29,446 bp) found in the horse gut to date, belonging to the family Tobaniviridae and the genus Torovirus.

Of the 497 viral contigs recovered from the horse gut metatranscriptomic data, 429 were successfully classified using VITAP. Of these, 381, 232, and 136 contigs were identified at the family, genus, and species levels, respectively. Further taxonomic analysis revealed that the horse gut harbored 136 viruses belonging to 22 families (Supplementary Table S5). These include positive-strand RNA viruses [ssRNA (+)], such as Alphaflexiviridae, Amalgaviridae, Benyviridae, Betaflexiviridae, Bromoviridae, Coronaviridae, Endornaviridae, Fiersviridae, Fusariviridae, Mitoviridae, Phycodnaviridae, Picor-naviridae, Potyviridae, Secoviridae, Solemoviridae, Tobaniviridae, Tombusviridae and Virgaviridae (Figure 1; Supplementary Figure S1).

The Picobirnaviridae family was the most abundant RNA viral family in the horse gut, comprising 134 viral genomes. Notably, RNA viruses belonging to the family Picobirnaviridae were detected in all three horse breeds (Supplementary Table S5). Picobirnaviridae are a class of small, non-enveloped viruses possessing a double-stranded RNA genome consisting of two segments. They are most prevalent in the gut of various vertebrates, including human, and have been widely identified in samples from sources such as algae, human, bats, and monkeys (53). At the genus level, the majority of viruses were classified as Orthopicobirnavirus (115). To further resolve the taxonomic placement of the identified RNA viruses, a comprehensive phylogenetic analysis was conducted based on RdRp protein sequences, incorporating representative reference sequences from established RNA virus families. The majority of viral sequences clustered within recognized RNA virus families, including Picobirnaviridae, Picornaviridae, Partitiviridae, Tombusviridae, and others, enabling family-level taxonomic assignment based on their phylogenetic positions. Viral sequences that did not form stable clades with any known reference groups were classified as unclassified or unidentified RNA viruses, reflecting their high divergence from currently described taxa. At the genus level, a considerable fraction of RdRp sequences could not be assigned to any known viral genera, indicating the presence of a substantial amount of uncharacterized RNA viral diversity. Among the classified taxa, Orthopicobirnavirus was the most abundant genus, followed by Potyvirus and several mitovirus-related genera. At the species level, the dominant taxa were Orthopicobirnavirus beihaiense (71 genomes) and Orthopicobirnavirus hominis (11 genomes).

To further investigate the composition and diversity of intestinal RNA viruses in different horse breeds, we performed a comparative analysis of α-diversity. Although Yili horses exhibited the highest Shannon index and Akhal-Teke horses the lowest, Mann–Whitney U tests revealed no statistically significant differences in α diversity of gut RNA viruses among horse breeds (Figure 2A). Conversely, β diversity analysis based on Bray-Curtis distance revealed that the RNA virusesof Thoroughbred and Akhal-Teke horses were significantly more similar to each other than to Yili horses (p < 0.01) (Figure 2B). Coverage fraction analysis of the metatranscriptomic data from the 3 breeds showed that 36 core RNA viruses were shared across all samples (Figure 2C), providing an assessment of the heterogeneity of gut RNA viruses among horse breeds. Akhal-Teke, Yili, and Thoroughbred horses possessed 126, 130, and 63 unique viral genomes, respectively. Of the 36 shared core viruses, 24 received taxonomic annotation, of which 22 were classified at the family level. These included 12 viruses in the Picobirnaviridae family, 6 in the Partitiviridae family, 6 in the Mitoviridae family, and one each in the Potyviridae, Solemoviridae, and Virgaviridae families (Figure 2C).

To further analyze and understand the abundance of gut RNA viruses in different horse breeds, we calculated the average TPM values for each sample to assess viral abundance across the three breeds (Supplementary Table S6). At the family level, intestinal RNA viruses exhibited heterogeneity among the three horse breeds. Akhal-Teke, Yili horse, and Thoroughbred horses harbored 16, 13, and 12 viral families, respectively (Figures 3A–C).

The top five families in Akhal-Teke horses were Picobirnaviridae, Partitiviridae, Mitoviridae, Betaflexiviridae, and Endornaviridae. The top five families in Thoroughbred horses were Virgaviridae, Bromoviridae, Mitoviridae, Partitiviridae, and Picobirnaviridae. The top five families in Yili horses were Virgaviridae, Mitoviridae, Solemoviridae, Picobirnaviridae, and Betaflexiviridae. Notably, the Picobirnaviridae and Mitoviridae were dominant families in the gut of all three horse breeds. Furthermore, 11, 12, and 14 RNA viruses in the guts of Akhal-Teke, Thoroughbred, and Yili horses, respectively. It exhibited log-transformed TPM values exceeding the threshold of 4, with detection rates greater than 1% in each breed. Additionally, AK1_k141_55107, belonging to the Mitoviridae family, exhibited an abundance greater than 1% across all 3 breeds, indicating that horses could potentially serve as potential hosts for this virus (Figure 3D).

We analyzed the differences between the three species based on the abundance in-formation of each RNA virus, identifying a total of 82 significantly different viruses in Akhal-Teke and Yili horses only (Supplementary Figure S2). Compared to the Akhal-Teke horse, there are 71 RNA virus genomes upregulated in the gut of the Yili horse, while 11 were found to be downregulated. The top 5 upregulated RNA viruses were TB_4_k141_293664, TB_2_k141_475750, AK3_k141_389117, TB_3_k141_136870, and TB_3_K141392274, while the top five downregulated were YL_4_k141_22425, YL_5_k141_412583, YL_5_k141_126620, YL_4_k141_203666, and YL_3_k141_89994 (Figure 3E; Supplementary Table S7).

Using coverage scores from read alignments against public datasets from human, cow, and sheep (Supplementary Table S8), we found that of the 497 viruses identified in the horse guts, 6.23% (31/497), 8.65% (43/497), and 3.82% (19/497) were present in metatranscriptome datasets from human, cow and sheep, respectively (Supplementary Figures S3, S4). In contrast, 49% (19/497) were present in the human and cow datasets (Figure 4A), and 3.82% (19/497) were present in human, cow, and sheep metatranscriptome data (Supplementary Figures S3, S4). Conversely, 49% (317/497) of viruses were specific to horse gut samples. A total of 12.67% (63/497) of viruses were detected in both cow and sheep samples from the same ecological niche (Supplementary Figures S3, S4). PCoA based on abundance data revealed structural similarities between RNA viruses shared by human, cow, and sheep, as well as some horse-derived viruses. Specifically, sheep and Akhal-Teke horses, as well as Thorough-bred horses, human and Yili horses, and cow and Thoroughbred horses, respectively, have more similar viral community structures and closer viral sharing networks (Figures 4B,C; Supplementary Figures S3, S4).

The average detection rates of horse gut RNA viruses in individual metatranscriptome data from human, cow, and sheep were 0.91 (55/60), 6.83 (123/18), and 15 (90/6), respectively. These results suggest that horse gut RNA viruses are more prevalent in cow and sheep than in human (Figure 5A). Further network analysis revealed that shared horse intestinal RNA viruses exhibited heterogeneous clustering patterns in human versus domestic animals (cow and sheep) (Figure 5B), which may be influenced by their distinct ecological niches within the food chain. By mapping RNA virus genomes to sample sequencing reads, seven viruses shared by horses, human, cow, and sheep were detected, all belonging to the Picobirnaviridae family. Five of these were classified within the Orthopicobirna virus and Orthopicobirna virus beihaiense, while the remaining two were unclassified at the genus and species levels.

We also assessed the detection of these shared viruses at the family and genus taxonomic levels. At the family level, 57, 37, and 67 horse gut RNA viruses were detected in human, cow, and sheep, respectively. At the genus level, no shared RNA viruses were detected in the Solemoviridae or Potyviridae families across any of the three breeds (Figure 5C). At the genus level, the number of shared viruses decreased to 15, 16, and 11 RNA genus in human, cow, and sheep, respectively. None were detected in the Mitoviridae, Solemoviridae, or Potyviridae families. These results suggest that there is a low frequency of shared viruses between human and domesticated animals such as cow and sheep (Figure 5D).