Our animal experimental procedures were approved by the Institute of Bast Fiber Crops, CAAS (Changsha, China), and Huazhong Agricultural University (Wuhan, China). All experiments in this study were conducted with the Scientific Ethics Committee of the Huazhong Agricultural University in strict accordance (approval number: HZAUMO-2019-059). Three-week-old post-weaning specific pathogen-free C57 Black mice (n = 80, body weight 9.35 ± 1.13 g) were randomly assigned to four groups, each containing 20 male mice. Each group was fed the same diet (AIN-76A) supplemented with 0, 300, 600, and 1,200 ppm ε-polylysine. C57 mice were maintained under 12:12 h light/dark conditions at Huazhong Agricultural University with free access to feed and water. Body weights were measured at the beginning of the experimental period, and then once per week. Fecal samples (n = 12 each group) were collected at weeks four, six, and ten. At the end of the experiment (after about 10 weeks), the mice were anesthetized and sacrificed, and intestine and plasma samples (n = 20 each group) were harvested.

The morphology of the intestine was determined using hematoxylin and eosin. One centimeter of the small intestine and large intestine, as well as the middle position of each bowel segment, were rinsed in PBS, fixed in formalin overnight at 4°C, and embedded in paraffin. Intestinal sections were stained with H&E. Villus height and crypt depth of the small intestine were measured using Image-Pro Plus 6.0 software (Media Cybernetics, Inc., Rockville, MD, United States).

Qubit dsDNA Assay Kit (Cat. No. Q328520, Life Technologies, Carlsbad, CA, United States) was used to extract DNA according to the manufacturer’s instructions. The diluted DNA (1 ng/μL) was used as a template for polymerase chain reaction (PCR) amplification of bacterial 16S rDNA genes. For bacterial diversity analysis, the V3–V4 variable regions of 16S rDNA genes were amplified using the universal primers 343F (5′-TACGGRAGGCAGCAG-3′) and 798R (5′-AGGGTATCTAATCCT-3′) (22). The final amplicon was quantified using the Qubit dsDNA Assay Kit (Life Technologies, Q328520, Grand Island, CA, United States). Equal amounts of purified amplicons were pooled for sequencing on the Illumina MiSeq platform.

Clustering to generate OTUs using Vsearch software (23) with a 97% similarity cutoff. The representative reads of each OTU were selected using the QIIME package (version 1.8.0). All representative reads were annotated and blasted against the Silva database Version 123 (or Greengens) using the RDP classifier (24) with a confidence threshold of 70%.

The abundance of each OTU was normalized relative to the sample with the fewest sequences. Alpha diversity was used to analyze the complexity of species in fecal sample by applying the Chao1, observed species, Shannon, and Simpson indices. Beta diversity was used to evaluate the complexity of species and was calculated using PCoA. Both Alpha and Beta diversity were used in the QIIME software (version 1.8.0). The different phyla and genera of 16S rDNA were analyzed using the Kruskal-Wallis test. P < 0.01 was considered highly significant, and P < 0.05, was considered significant. We confirmed that all species were differentially abundant by LEfSe analysis. Heatmap diagrams and other plots were created in the R environment (v3.1.2). Functional profiles of the gut microbiota were analyzed using PICRUSt (25). OTUs were normalized, and the gene categories were at level 2 of the KEGG orthology groups (26).

Plasma samples (100 μL) were added into 10 μL internal standard (2-chloro-l-phenylalanine in methanol, 0.3 mg/mL) dissolved in methanol, and vortexed for 10 s. Then, the mixtures were vortexed for 1 min with 300 μL mixture of methanol and acetonitrile (2/1, v/v). All samples were ultrasonicated at 25–28°C for 10 min and stored at −20°C for 30 min. The extract was centrifuged at 13,000 rpm at 4°C for 15 min. The supernatant (300 mL) was dried in a freeze-concentration centrifugal dryer. A mixture of 400 μL methanol and water (1/4, vol/vol) was added to each sample, then vortexed for 30 s, and placed at 4°C for 2 min. Samples were centrifuged at 13,000 rpm at 4°C for 5 min. The supernatant (150 μL) from each tube was collected using crystal syringes, filtered through 0.22 μm microfilters and transferred to LC vials. The vials were stored at −80°C until LC -MS analysis. The analysis instrument of metabolic profiling was the AB ExionLc system (AB SCIEX, Framingham, MA, United States) coupled with an AB SCIEX Triple TOF 6600 System (AB SCIEX, Framingham, MA, United States). An ACQUITY UPLC BEH C18 column (1.7 μm, 2.1 × 100 mm) were employed in both positive and negative modes. The binary gradient elution system consisted of (A) water (containing 0.1% formic acid, v/v) and (B) acetonitrile (containing 0.1% formic acid, v/v) and separation was achieved using the following gradient: 0 min, 95% A, 5% B; 2 min, 80% A, 20% B; 4 min, 75% A, 25% B; 9 min, 40% A, 60% B; 14 min, 100% B; 16 min, 100% B; 16.1 min, 95% A, 5% B and 18.1 min, 95% A, 5% B. The flow rate was 0.4 mL/min and column temperature was 45°C. All the samples were kept at 4°C during the analysis. Data acquisition was performed in full scan mode (m/z ranges from 70 to 1,000) combined with IDA mode. Parameters of mass spectrometry were as follows: ion source temperature, 550°C (+) and 550°C (−); ion spray voltage, 5,500 V (+) and 4,500 V (−); curtain gas of 35 PSI; declustering potential, 80 V (+) and −80 V (−); collision energy, 10 eV (+) and −10 eV (−); and interface heater temperature, 550°C (+) and 550°C (−). For IDA analysis, range of m/z was set as 50–1,000, the collision energy was 30 eV.

The acquired LC-MS raw data were analyzed using Progenesis QI software (Waters Corporation, Milford, CT, United States). Metabolites were identified using Progenesis QI (Waters Corporation, Milford, CT, United States) data processing software, based on public databases such as http://www.hmdb.ca/, http://www.lipidmaps.org/, and self-built databases. PCA and OPLS-DA were carried out to visualize the metabolic alterations. Hotelling’s T2 region, shown as an ellipse in the score plots of the models, defined the 95% confidence interval of the modeled variation. Variable importance in the projection (VIP) ranks the overall contribution of each variable to the OPLS-DA model, and variables with VIP > 1 were considered relevant for group discrimination. The differential metabolites were selected on the basis of the combination of a statistically significant threshold of variable influence on projection (VIP) values obtained from the OPLS-DA model and p values from a two-tailed Student’s t-test on the normalized peak areas, where metabolites with VIP values larger than 1.0, and p values less than 0.05, were considered as differential metabolites (27).

Spearman’s rank correlation coefficient analysis was performed to examine the possible connection between the composition of intestinal microflora at the genus level and plasma metabolites in C57 mice.

The effects of dietary supplementation with ε-polylysine on body weight were shown in Table 1. At the beginning of the experiment, there were no significant differences in body weight (P > 0.05), while the weights in the control group and 300 ppm group were significantly higher than those in the 1,200 ppm group from 5 to 7 weeks (P < 0.05). After 8 week, the weight of mice did not differ between the four groups (P > 0.05), but the trend of weight decreased with increasing ε-polylysine supplementation. As shown in Table 2, there was no significant difference in villus length and crypt depth in the duodenum (P > 0.05). The villus length in the 1,200 ppm group was significantly higher than that in the other groups (P < 0.05), but the crypt depth in the 300 and 1,200 ppm groups were significantly lower than that in the control group (P < 0.05). The villus length in the 1,200 ppm group was significantly lower than that in 600 ppm group, and crypt depth in the 1,200 ppm group was significantly lower than that in the control group (P < 0.05). Compared with the control group, damage to the jejunum epithelium was observed in the 1,200 ppm group (Figure 1).

Among the bioinformatic analyses, a total of 77,742, 77,686, 77,120, and 75,051 valid tags were obtained for the four groups, accounting for 90.62% (control group), 91.40% (300 ppm group), 91.93% (600 ppm group), and 91.22% (1200 ppm group) of raw sequences, respectively (Supplementary Data 1). Taxonomic analysis revealed a total of 17 bacterial phyla, 35 classes, 93 orders, 149 families, 319 genera, and 88 species. As shown in Figures 2A–D, there were no significant differences in the alpha diversity indices (Chao1, observed species, and Shannon indices) among the four groups (P > 0.05), but 600 ppm group had a lower Simpson index than the control group and 300 ppm group (P < 0.05), while the Simpson index at 300 ppm was significantly different from that in the 1,200 ppm group (P < 0.05; Figure 2D). The microbial community structure of all samples (beta diversity) was visualized using PCoA (Figure 2E). The two factors (PC1 and PC2) accounted for 8.08 and 6.36% of the sample variation, respectively. The PCoA based on the bacterial OTUs showed that the samples clustered together, which indicated a shift in the gut bacterial community among the four groups (Figure 2E). Our results showed that the abundance of phylum Bacteroidetes was significantly increased by ε-polylysine, and the abundance of phylum Firmicutes was significantly decreased (P < 0.05), which was a domain phylum in all samples (Figures 2F,G and Supplementary Data 2). The abundance of Proteobacteria and Verrucomicrobia was higher in the ε-polylysine groups than in the control group (P < 0.05; Figure 2G). A total of 96 different genera were identified by Wilcoxon text analysis (Supplementary Data 3). The LEfSe analysis revealed that the genera had effects among the four groups (Figure 2H). Ten genera were more abundant in the ε-polylysine group, including Faecalibaculum, Ileibacterium, Parabacteroides, Odoribacter, Bifidobacterium, Alistipes, Coriobacteriaceae UCG-002, Family XIII AD3011 group, Akkermansia, and Prevotellaceae UCG-001.

The others were more abundant in the control group, including Ruminococcus, Prevotella, Prevotellaceae UCG-003, Christensenellaceae R-7 group, Prevotellaceae NK3B31 group, Butyricimonas, Coprococcus 2, Rikenellaceae RC9 gut group, Collinsella, Ruminococcaceae, Negativibacillus, Faecalibacterium, Tyzzerella 4, Anaerotruncus, Escherichia-Shigella, Anaerostipes, Subdoligranulum, Robinsoniella, and Lachnospiraceae NK4A136 group (Figure 2I and Supplementary Data 3).

For the four groups, 76,009, 78,823, 78,996, and 77,816 valid tags were obtained, which accounted for 92.03, 91.92, 91.70, and 92.30% of the raw sequences, respectively (Supplementary Data 1). We examined the taxonomic composition of the fecal microbiota in 6 week-C57 mice, a total of 17 bacterial phyla, 37 classes, 104 orders, 183 families, 381 genera, and 99 species. The Chao1 index in 300 ppm group was significantly higher than that in the other groups (P < 0.05). The observed species index was significantly higher than that in the control group and the 1,200 ppm group (P < 0.05; Figures 3A–D). In the PCoA analysis, PC1 accounted for 7.82% and PC2 accounted for 6.96% (Figure 3E). Bacteroidetes and Firmicutes were the domain phyla, which is the same to 4 weeks (Figure 3F). There were seven different phyla among the four groups (Figure 3G and Supplementary Data 4). The abundance of the phyla Gemmatimonadetes, Proteobacteria, and Fibrobacteres was up-regulated by ε-polylysine. After Wilcoxon text analysis, we identified 140 different genera (Supplementary Data 5). In the control group, the genera Clostridium sensu strcto 1, Ruminococcus 1, Marvinbryantia, Lactobacillus, and Parabacteroides played an important role in LEfSe analysis. The genera Ruminococcaceae UCG 014, Odoribacter, and Intestinimonas affected the ε-polylysine groups (Figure 3H). Ten genera were upregulated, including Collinsella, Faecalibacterium, Ruminococcaceae, Butyricicoccus, Lachnospiraceae UCG-001, Intestinimonas, and Actinobacillus, while the other 13 genera were down-regulated, including Ileibacterium, Coriobacteriaceae UCG-002, Ruminococcus 1, Christensenellaceae R-7 group, Faecalibaculum, Bifidobacterium, Roseburia, Rikenellaceae RC9 gut group, Marvinbryantia, Alistipes, Parabacteroides, Lactobacillus, and Tyzzerella 4 (Figure 3I and Supplementary Data 5).

We found 59,729, 67,909, 70,719, and 74,870 valid tags for the four groups, which accounted for 79.89, 89.97, 84.83, and 87.36% of raw sequences, respectively (Supplementary Data 1). Taxonomic analysis revealed a total of 23 phyla, 54 classes, 132 orders, 225 families, 417 genera, and 127 species. As shown in Figure 4, the alpha diversity, including Chao1 index, observed species index, Shannon index, and Simpson index did not differ between the groups (Figures 4A–D). The PC1 accounted for 20.81% and PC2 accounted for 7.72%, as observed using PCoA analysis (Figure 4E). The abundance of the phylum Bacteroidetes was significantly decreased by ε-polylysine, and the abundance of phylum Firmicutes was significantly increased by ε-polylysine (P < 0.05), which was opposite to those at 4 weeks (Figure 4F). In addition, we identified six different phyla among the groups, and the abundance of Proteobacteria, Bacteroidetes, and Patescibacteria were down-regulated by ε-polylysine (Figure 4G and Supplementary Data 6). Sixty-four different genera were identified by Wilcoxon text analysis (Supplementary Data 7). After LEfSe analysis, the genera Alistipes, Parabacteroides, and Edaphobacter played an important role in the control group, and the genus Lachnospiraceae NK4B4 group had an effect in 300 ppm group (Figure 4H). In addition, the Rikenellaceae RC9 gut group and Intestinimonas had effects at 600 ppm, and the genera Odoribacter, Blautia, Ruminococcaceae UCG-014, and Holdemania had effects in the 1,200 ppm group (Figure 4H). Furthermore, the results showed several upregulated genera, including Allobaculum, Prevotellaceae UCG-003, Prevotella 1, Lachnospiraceae UCG-010, Parasutterella, Ruminococcus 1, Ruminococcaceae, Christensenellaceae R-7 group, Holdemania, Anaerostipes, Odoribacter, Lachnospiraceae NK4B4 group, Intestinimonas and Blautia, and six down-regulated genera, including Ileibacterium, Ruminococcus 2, Lachnospiraceae UCG-001, Alistipes, Parabacteroides, and Streptomyces (Figure 4I and Supplementary Data 7). We investigated the functional profiles of microbiota by PICRUSt analysis, the relative abundances of genes involved in the metabolism of lipids, carbohydrates, glycans, energy, amino acids, other amino acids, cofactors and vitamins, terpenoids and polyketides, and nucleotides (Figure 4J and Supplementary Data 8). Additionally, the results showed increases in the proportions of genera involved in genetic information processing and cellular processes (Supplementary Data 8).

The metabolomics data of the plasma were shown in Figure 5. In 300 ppm group compared with the control group, PC1 accounted for 17% and PC2 accounted for 12.2% (Figure 5A). The PC1 accounted for 20.5% and the PC2 accounted for 12.2%, between 600 ppm and the control group (Figure 5C). PC1 accounted for 23.4%, and PC2 accounted for 14.4% compared to the 1,200 ppm group compared to the control group (Figure 5E). The results of orthogonal partial least-squares discriminant analysis (OPLS-DA) of the liver between the ε-polylysine groups and the control group were presented in Figures 5B,D,F. The abundance of glycerophospholipids was the highest among all the lipid subclasses (Figure 5G). The abundance of organooxygen compounds, fatty acids, and polyketide subclasses were increased by ε-polylysine, but other metabolites such as sphingolipids and glycerophospholipids were decreased by ε-polylysine (Figure 5G). We further examined the taxonomic composition of the glycerophospholipid metabolites and found that PC, PE, and LysoPC were domain metabolites among all groups. The abundance of PC, PE, lysoPC, LysoPE, 1-Arachidonoylglycerophosphoinositol, 1-(2-methoxy-eicosanyl)-sn-glycero-3-phosphoethanolamine, and OHOHA-PS were decreased by ε-polylysine (Figure 5H and Supplementary Data 9). The PC/PE radio of control group and ε-polylysine had no significant changes (Figure 5I). The different metabolites detail was shown in the Figure 6. Plasma metabolites were annotated and were highly enriched in the pathways belonging to “glycerophospholipids metabolism” and “biosynthesis of unsaturated fatty acids” compared to the control groups (Figures 5J–L).

We explored the association between fecal microbiota and the different metabolites shared by the plasma. Spearman correlation analysis showed that the metabolites containing PC(18:0/0:0), PC(0:0/16:0), PC[O-18:1(9Z)/0:0], PC[20:1(9Z)/0:0], LysoPC(18:0), and LysoPC[22:4(7Z,10Z,13Z,16Z)] were positively correlated with Alistipes (R > 0.6, P < 0.001), and PC(18:0/0:0), PC(17:0/0:0), LysoPC(18:0) were positively associated with Lachnospiraceae UCG-001 (R > 0.6, P < 0.001). In addition, PC[20:3(5Z,8Z,11Z)/0:0] was positively correlated with Streptomyces (R > 0.6, P < 0.001). The metabolites PC(18:0/0:0), PC(16:0/0:0), PC(0:0/16:0), PC[20:3(5Z,8Z,11Z)/0:0], PC(17:0/0:0), PC[O-18:1(9Z)/0:0], PC[20:1(9Z)/0:0], PC[20:4(5Z,8Z,11Z,14Z)/18:1(11Z)], LysoPC(18:0), LysoPC[20:3(8Z,11Z,14Z)], lysoPC [22:4(7Z,10Z,13Z,16Z)], and lysoPC (17:0) were negatively associated with Christensenellaceae R-7 group and Allobaculum, respectively (R < −0.6, P < 0.001). The metabolite PC(18:0/0:0), PC[O-18:1(9Z)/0:0], PC[20:1(9Z)/0:0], and LysoPC(18:0) were negatively correlated with changes in Blautia (R < −0.6, P < 0.001), and PC(16:0/0:0), PC[20:4(5Z,8Z,11Z,14Z)/18:1(11Z)], LysoPC(17:0), and LysoPC[22:4(7Z,10Z,13Z,16Z)] were negatively correlated with Ruminococcaceae UCG-004 (R < −0.6, P < 0.001). PC(O-18:1(9Z)/0:0), LysoPC(17:0), and LysoPC[22:4(7Z,10Z,13Z,16Z)] were negatively associated with changes in Odoribacter (R < −0.6, P < 0.001; Figure 7 and Supplementary Data 10).

The metabolites PE[20:4(5Z,8Z,11Z,14Z)/0:0], PE[20:4(8Z,11Z,14Z,17Z)/0:0], PE(22:0/0:0), and PE[22:6(4Z,7Z,10Z,13Z,16Z,19Z)/0:0] were positively associated with changes in Streptomyces (R > 0.6, P < 0.001). PE(21:0/0:0) was positively associated with changes in Alistipes and Lachnospiraceae UCG-001 (R > 0.6, P < 0.001). PE[20:1(11Z)/0:0] was positively correlated with that of Alistipes (R > 0.6, P < 0.001). However, the genera Christensenellaceae R-7 and Allobaculum were negatively correlated with PE[20:4(5Z,8Z,11Z,14Z)/0:0], PE[22:6(4Z,7Z,10Z,13Z,16Z,19Z)/0:0], PE(18:0/0:0), PE(21:0/0:0), LysoPE[0:0/22:6(4Z,7Z,10Z,13Z,16Z,19Z)], PE[20:4(8Z,11Z,14Z,17Z)/0:0], PE(22:0/0:0), and PE(20:1(11Z)/0:0) (R < −0.6, P < 0.001). The change in Blautia was negatively associated with PE(18:0/0:0) and PE(21:0/0:0) (R < −0.6, P < 0.001), and the change in Ruminococcaceae was negatively correlated with PE(21:0/0:0), PE(22:0/0:0), and PE[20:1(11Z)/0:0] (R < −0.6, P < 0.001). Moreover, the metabolites containing PE[22:6(4Z,7Z,10Z,13Z,16Z,19Z)/0:0] and PE[20:4(8Z,11Z,14Z,17Z)/0:0] were negatively associated with Odoribacter, Parasutterella, and Lachnospiraceae UCG-010 (R < −0.6, P < 0.001). PE(21:0/0:0) and LysoPE[0:0/22:6(4Z,7Z,10Z,13Z,16Z,19Z)] were correlated with changes in Lachnospiraceae UCG-010 (R < −0.6, P < 0.001). LysoPE(0:0/18:0) was negatively associated with Allobaculum (R < −0.6, P < 0.001), and LysoPE[0:0/22:6(4Z,7Z,10Z,13Z,16Z,19Z)] was negatively correlated with Parasutterella (R < −0.6, P < 0.001; Figure 6 and Supplementary Data 10).

Furthermore, 1-Arachidonoylglycerophosphoinositol was positively correlated with Alistipes and Streptomyces (R > 0.6, P < 0.001), but negatively correlated with Christensenellaceae R-7 group, Ruminococcaceae UCG-004, and Allobaculum (R < −0.6, P < 0.001). OHOHA-PS was positively associated with Alistipes, Lachnospiraceae UCG-001, and Streptomyces (R > 0.6, P < 0.001), whereas OHOHA-PS was negatively associated with Odoribacter, Ruminococcaceae UCG-005, Christensenellaceae R-7 group, Ruminococcaceae UCG-004, Allobaculum, and Lachnospiraceae UCG-010 (R < −0.6, P < 0.001). 1-(2-methoxy-eicosanyl)-sn-glycero-3-phosphoethanolamine was negatively correlated with changes in the Christensenellaceae R-7 group (R < −0.6, P < 0.001). Other low correlation results (R > −0.6 or R < 0.6) were shown in Figure 6 and Supplementary Data 10.