The bacterial genomic DNA was extracted from the gut samples and artificial diet using a DNA rapid extraction kit (TransGen, Beijing). Subsequently, the extracted DNA was amplified by PCR with universal primers (338F: 5′-ACTCCTACGGGAGGCAGCAG-3′ and 806R: 5′-GGACTACHVGGGTWTCTAAT-3′). 1% agarose gel electrophoresis was conducted to ensure the quality of the extracted DNA. All samples were sequenced using the Illumina NovaSeq 6,000 platform.

The sequencing data analysis of 16S rRNA was performed based on the method described in Hao et al. (2023). Briefly, the quality of sequencing data was assessed using Trimmomatic and Cutadapt, followed by the calculation of diversity indices and evaluation of sampling depth. The Ace indices, Chao1 indices, Shannon indices, Simpson indices, and operational taxonomic units (OTUs) were used to assess the Alpha diversity of mandarin fish’s gut microbiota. The coverage of every sample library was employed to evaluate the validity of the sequencing results. Principal coordinates analysis (PCoA) was used to assess the beta diversity of the gut microbiota. Each microbiota’s relative abundance was computed at the phylum and genus levels. The evolutionary relationship between the samples was examined by UniFrac. Welch’s t-test and the non-parametric factorial Kruskal-Wallis (KW) sum-rank test were used to compare the two groups. The figure of linear discriminant analysis effect size (LEfSe) was made based on linear discriminant analysis (LDA) to calculate the contribution of each component’s richness to the variation. Hierarchical clustering analysis was performed based on the binary-jaccard distance matrix. The tree structure diagram was constructed according to the Unweighted Pair-Group Method with Arithmetic Mean (UPGMA) algorithm. Then, the functional category abundance information of the microbial community was determined by analyzing the KEGG database using the PICRUSt2 microbial community metabolic function prediction tool.

Based on the 16S rRNA sequencing data, Weissella were the dominant bacteria in the gut of mandarin fish fed an artificial diet. To this aim, Weissella bacteria were isolated and identified from gut in mandarin fish.

The gut contents from Group T were rinsed with 1 mL of sterilized saline. The resultant stock solution was collected and diluted to gradients of 10⁻¹, 10⁻³, and 10⁻⁵. Subsequently, 0.1 mL of liquid from each dilution gradient was inoculated onto De Man, Rogosa, and Sharpe (MRS) solid medium supplemented with 0.2 g/L vancomycin (Kumar et al., 2011). The samples were then incubated at a constant temperature of 30°C for 12 h until colonies appeared. Afterwards, single colonies were randomly selected for transplantation and streaked onto fresh medium three times to ensure the purification of the samples.

The genomic DNA of dominant bacteria was extracted using chloroform method (Sambrook and Russell, 2001), then amplified with primers (27F: 5′-AGAGTTTGATCMTGGCTCAG-3′, and 1492R: 5′-GGTTACCTTGTTACGACTT-3′). After verification via a 1.0% agarose gel, the amplified products were sent to Beijing Zhixu Biotechnology Co., Ltd. for sequencing. The isolated bacteria’s 16S rRNA sequences were analyzed using the National Center for Biotechnology Information (NCBI) database for BLAST analysis. MEGA 11 software and the neighbor-joining method were employed to construct the phylogenetic tree, as reported by Zhang et al. (2016).

The shape and hue of the separated colonies were examined using a scanning electron microscope. The individual colonies were selected for Gram staining examinations, based on the Stain Kit (BKMAMLAB, Changde, China). The pure colonies were cultivated in MRS broth for 12 h at 30°C and 200 rpm. After reaching the exponential phase (OD_600nm = 0.5 ~ 0.7), the strains were rinsed in PBS buffer (pH 7.4), fixed in 2.5% glutaraldehyde, dried using an alcohol gradient, and resuspended in isoamyl acetate. Then, they were ready for morphological analysis using a scanning electron microscope. The biochemical characterization of bacterial cultures was evaluated using VITEK2 system (Ligozzi et al., 2002). The bacterial species was identified based on the observed patterns of biochemical reactions.

Antibacterial activity was determined using the diffusion method, as described in Yadav and Tiwari (2023). The eight indicator bacteria obtained from Hebei Key Laboratory of Nutritional Regulation and Disease Control for Aquaculture, including Escherichia coli, Aeromonas hydrophila, Staphylococcus aureus, Salmonella enterica, Vibrio anguillarum, V. harveyi, V. parahaemolyticus, and V. alginolyticus, were selected for measure the antibacterial activity of dominant strains. The selected Weissella strains were cultured in MRS broth until their OD_600 nm reached 0.7. In brief, the indicator bacteria (1 × 10⁸ CFU/mL) were spread onto LB solid medium with Oxford cups containing three wells filled with Weissella strains (OD600 nm = 0.7) and one well filled with distilled water as a control. The results were interpreted as follows: DIZ < 15.00 mm indicated insensitivity (no significance, ns); 15.00 mm < DIZ ≤ 16.00 mm indicated low sensitivity (+); 16.00 mm < DIZ ≤ 17.00 mm indicated moderate sensitivity (++); and DIZ > 17.00 mm indicated high sensitivity (+++).

In the present study, we obtained a total of 678,375 quality-controlled reads and 642,011 effective reads based on NovaSeq Illumina-based paired-end amplicon 16S rRNA gene sequencing (V3–V4) (Supplementary Table S2). A total of 2,601 operational taxonomic units (OTUs) were identified by removing background noise and performing species annotation, representing the diversity of bacterial species present in the samples. The dilution curves of samples tended to be flat and were close to the plateau, indicating that the current sequencing depth was sufficient to capture the bacterial community’s biological information (Supplementary Figure S1).

The Chao 1, ACE, Shannon, and Simpson indices were employed to assess the diversity and richness of the gut microbiota in two distinct groups of mandarin fish (Supplementary Table S2). Notably, although the average values for the Ace, Chao1, Shannon, and Simpson indices were higher in Group T compared with Group C, there were no significant differences across all four metrics (p > 0.05) (Supplementary Figure S2). A Venn diagram was constructed to identify the core and different Operational Taxonomic Units (OTUs), revealing that a total of 122 OTUs shared between Group T and C (Figure 3A). The hierarchical cluster analysis was conducted on the beta diversity distance matrix of the gut microbiota, resulting in a tree structure diagram generated by the UPGMA algorithm (Figure 3B). Principal Coordinates Analysis (PCoA) was also employed to explore differences in microbial community composition (Figure 3C). In detail, PC1 accounted for 23.6% of the total bacterial community variation, while PC2 explained 15.22% (Figure 3C). Significant differences were observed between the gut bacterial communities of mandarin fish fed live bait and those fed an artificial diet (p < 0.05) (Figure 3C).

Distinct community structures of gut microbiota were observed in mandarin fish fed with live bait and artificial diet. To eliminate the effect of artificial diet on gut microbiota, we detected the abundance of species at the level of bacterial genera in the artificial diet (Supplementary Table S3). As shown in Figure 4A, Firmicutes was the most abundant bacteria at the phylum level in Group C, followed by Proteobacteria and Bacteroidota. In addition, Firmicutes was dominant, followed by Proteobacteria, Actinobacteota, Acidobacteriota, and Bacteroidota in Group T (Figure 4A). At the genus level, Clostridium_sensu_stricto_4 (41.98%) and Clostridium_sensu_stricto_1 (20.45%) were the dominant bacteria in Group C. However, Weissella (67.70%) was the most common bacteria in Group T (Figure 4B). Notably, there was a significant difference in the relative abundances of Clostridium_sensu_stricto_4 (p < 0.05), Clostridium_sensu_stricto_1 (p < 0.05), and Weissella (p < 0.05) between the two groups. LEfSe analyses showed a higher abundance of 14 and 19 taxa in Group C and T, respectively (LDA score > 4.0) (Figure 4C). As shown in Figure 4D, the biomarkers of Group C consisted of Clostridium_sensu_stricto_4, Clostridium_sensu_stricto_1, Paraclostridium, Edwardsiella, Clostridiaceae, Hafniaceae, Clostridiacles, and Clostrdia. Additionally, the biomarkers of Group T were composed of Lactococcus, Weissella, Evilactobacillus, Acidipilasilvibacterium, Acidobacteriaceae, Lactobacillales, and Bacilli.

The KEGG database was utilized for functional prediction of 16S rRNA gene derived from the gut microbiota of mandarin fish. In the present study, 341 metabolic functional pathways were identified, and the genes of the gut microbiota were categorized into six primary metabolic pathways, involved in metabolism, genetic information processing and environmental information processing (Supplementary Table S4). PICRUSt2 analyses revealed that, despite being from different feeding groups (C and T) of the mandarin fish exhibited similar gene functions (Figure 5). There were significant differences in the abundance of nine pathways including metabolic pathways, biosynthesis of secondary metabolites, and microbial metabolism in diverse environments (p < 0.05) (Figure 5). The relative abundance of several pathways in Group T were higher than that in Group C, including biosynthesis of secondary metabolites, biosynthesis of antibiotics, microbial metabolism in diverse environments, and carbon metabolism (p < 0.05) (Figure 5).

Ten strains of bacteria from punctate colonies were isolated for 16S rRNA sequencing. These sequences were blasted in the NCBI database, revealing that nine strains shared identical sequences, while one strain was different. The phylogenetic tree indicated that the nine strains were Weissella confusa and one strain was Weissella cibaria, designated as W. confusa RM125 (NCBI accession number: PP125780.1) and W. cibaria SJ548 (NCBI accession number: PP068943.1), respectively (Figure 6A). W. confusa RM125, W. confusa 2,879 and W. confusa 6,400 clustered together based on their similarity of 97.75% (Supplementary Figure S3). Moreover, W. cibaria SJ548, W. cibaria 1,382 and W. cibaria 2,769 gathered into a cluster with their similarity of 99.17% (Supplementary Figure S4). As shown in Figure 6B W. confusa RM125 and W. cibaria SJ548 shared a similarity of 97.81%.

As exhibited in Figure 7, the images from the scanning electron microscope revealed that the cells of W. confusa RM125 and W. cibaria SJ548 were well organized, displaying rod-like or double rod-like structures with a flat surface and absent of flagella. Specifically, W. confusa RM125 had dimensions of approximately 1.15 μm in length and 0.46 μm in width (Figure 7A), while W. cibaria SJ548 measured approximately 1.59 μm in length and 0.47 μm in width (Figure 7B).

Biochemical analyses using the VITEK2 method revealed a pattern of metabolic activities against W. confusa RM125 and W. cibaria SJ548, with 13 positive and 29 negative results, respectively (Supplementary Table S5). Both W. confusa RM125 and W. cibaria SJ548 strains could utilize N-acetyl-D-glucosamine (NAG), D-mannose (dMNE), Saccharose (SAC) and D-maltose (dMAL) as carbon sources and produce the enzymes including Arginine dihydrolase 1 (ADH1) and Arginine dihydrolase 2 (ADH2). Both these strains were tolerant to Novobiocin (NOVO), Optokhin (OPTO), O129R (O/129) and Bacitracin (BACI), and could grow in the presence of 6.5% NaCl. In contrast, neither strain was able to metabolize Amygdalin (AMY), d-ribose (dRIB), D-raffinose (dRAF), Cyclodextrin (CDEX), D-sorbitol (dSOR), Lactose (LAC), D-mannitol (dMAN), Salicin (SAL), Methyl-B-D-glucopyranoside (MBdG), D-trehalose (dTRE), α-glucosidase (AGLU), D-galactose (dGAL), and Pullulanase (PUL). Besides, neither strain could produce Leucine arylamidase (LeuA), Phosphatidylinositol phospholipase C (PIPLC), L-proline arylaminase (ProA), Tyrosine arylamidase (TyrA), L-aspartate arylaminase (AspA), β-glucuronidase (BGURr), β-galactopyranosidase (BGAR), α-galactosidase (AGAL), Urease (URE), β-galactosidase (BGAL), α-mannosidase (AMAN), Pyroglutamate arylaminase (PyrA), Phosphatase (PHOS) and β-D-glucuronidase (BGUR). Notably, only W. confusa RM125 could utilize xylose and tolerate Polymyxin B (POLYB), whereas only W. cibaria SJ548 could produce Alanine-phenylalanine-proline arylamidase (APPA) and Alanine arylamidase (AlaA). The results of Gram staining showed that W. confusa RM125 and W. cibaria SJ548 were Gram-positive bacteria (Supplementary Table S5).

The Oxford Cup method was employed to assess the inhibitory effects of W. confusa RM125 and W. cibaria SJ548 on eight indicator bacteria, revealing varying degrees of suppression exhibited by both strains. The distinct and nearly circular zones of inhibition were observed surrounding the Oxford cups (Figures 7C,D). The plates were then inspected to measure the diameter of the inhibition zone (DIZ). The DIZ was considered as follows: less than 15.00 mm indicated insensitivity; between 15.01 mm and 16.00 mm, it was classified as low sensitivity; between 16.01 mm and 17.00 mm, it was considered moderate sensitivity; and greater than 17.00 mm, it was classified as high sensitivity.

As illustrated in Table 1, W. confusa RM125 exhibited high sensitivity toward V. harveyi, moderate sensitivity toward S. aureus and V. parahaemolyticus, and low sensitivity toward E. coli, A. hydrophila, S. enterica, V. anguillarum, and V. alginolyticus. In contrast, W. cibaria SJ548 demonstrated high sensitivity toward both V. harveyi and V. alginolyticus, moderate sensitivity toward A. hydrophila, S. aureus, S. enterica, and V. anguillarum, and low sensitivity toward E. coli and V. parahaemolyticus (Table 1).