Abstract

   Corn silage can usually improve the growth performance and the meat
   quality of ruminants, and subsequently increase the economic benefits
   of farming. However, little is known about the effects of corn silage
   on donkeys. This experiment investigated the effects of corn silage on
   the weight gain, gut microbiota and metabolites of Dezhou donkeys. A
   total of 24 Dezhou donkeys, sourced from the same farm and exhibiting
   similar age and average body weight, were utilized in this experiment.
   The donkeys were allocated into two groups: a control group receiving a
   basic diet and a test group receiving a basic diet supplemented with
   30% corn silage. Each group comprised 12 donkeys, evenly distributed by
   sex (6 males and 6 females). The experiment lasted for 100 days.
   Results showed that dietary supplementation with corn silage
   significantly (P < 0.05) improved the weight gain of Dezhou donkeys at
   the end of the experiment. And the supplementation of corn silage in
   the diet significantly altered the bacterial community composition and
   metabolome in the feces of the donkeys. Notably, the relative abundance
   ratio of Bacteroidetes to Firmicutes was 0.76 in the control group
   compared to 0.96 in the test group. Furthermore, members of the
   Bacteroidetes and Firmicutes phyla were associated with differentiated
   metabolites enriched in the arachidonic acid metabolism and pentose and
   glucuronate interconversion pathways, both of which have been reported
   to be related to animal growth. Specifically, Bacteroidia exhibited
   statistically (P < 0.05) positive correlations with 15S-HpETE, while
   Bacilli demonstrated statistically (P < 0.05) negative correlations
   with D-Xylulose. The findings of this study can advance our mechanistic
   understanding of the remodeling of intestinal microbiota and metabolome
   induced by corn silage, as well as their relationships with the growth
   performance of Dezhou donkeys, which in turn favor the improvement in
   nutrition of Dezhou donkeys.

   Keywords: Dezhou donkey, Bacterial community, Metabolome, Corn silage

   Subject terms: Microbial communities, Animal physiology, Nutrition

Introduction

   Dezhou Donkey (one of five donkeys’ species in China) is an Equus
   animal just like the horse and the pony but with its own physiology and
   metabolism^[34]1,[35]2. Corn silage, a palatable forage with a higher
   yield compared to other forages, is commonly used for ruminants^[36]2.
   It can improve the growth performance and the meat quality of
   ruminants, and then increase the economic benefits of
   farming^[37]2,[38]3. However, only sparse data have been published
   about the application of corn silage for donkeys, to the best of
   authors’ knowledge, only two documents were found^[39]4,[40]5, one of
   which reported that the appropriate amount of corn silage was 25–35%
   for feeding donkeys, which showed maximum growth performance of
   donkeys. Based on previous researches^[41]4,[42]5, in this study, the
   proportion of corn silage added was 30%.

   The quantity and species of ecological community in the gut play an
   important role in the healthy growth and nutrient utilization of equine
   animals. Gut microbiota can ferment nutrients and produce various
   metabolites which have significant impact on the health of monogastric
   herbivores including donkeys. Short chain fatty acids (SCFAs) are
   mainly produced by microbial carbohydrate fermentation and are
   considered to be beneficial for the physiological functions of the
   intestines, such as enteroendocrine hormones, intestinal
   gluconeogenesis and gut immunity^[43]3,[44]6. Up to now, the microbiota
   in different gastrointestinal compartments of healthy donkey (foregut:
   stomach, duodenum, jejunum and ileum; hindgut: cecum, ventral colon,
   dorsal colon, and rectum) have been investigated^[45]7. However,
   knowledge of the temporal changes of microbiota in the gut of donkeys
   fed with corn silage remains limited and requires further
   investigation. Fecal metabolites can reflect the results of nutrient
   digestion and absorption by both the gut microbiota and the host
   gastrointestinal tract, providing a better explanation for the effects
   of the host–microbiota and metabolome interactions on growth
   performance^[46]8. Therefore, in current study, it was hypothesized
   that the feeding of corn silage would impact the growth performance of
   Dezhou donkeys and influenced the intestinal microbiota and metabolome.
   The purpose of this study aimed to perform high output sequencing of
   the 16S rRNA genes V3–V4 regions and metabolome from fecal material to
   explore the temporal microbiota changes in guts of donkeys during oral
   corn silage administration and investigate the relationships among the
   microbiota, metabolome and growth of donkeys.

Materials and methods

Ethical statement

   The animal study was reviewed and approved by Science and Technology
   Ethics Committee of Dezhou University (DezhouUniversity, Dezhou,
   China). All experiments were performed following the ARRIVE guidelines
   ([47]http://arriveguidelines.org) to report animal experiments.

Experimental animals and design

   This study was carried out from April 9 to July 18, 2021 (100 days in
   total) at the Yucheng Huimin Agricultural Technology Co., Ltd.
   (Shandong Province, China). A total of 24 Dezhou donkeys, sourced from
   the same farm and exhibiting similar age (8 ± 0.5 months) and average
   body weight (184 ± 1 kg), were utilized in this experiment. The donkeys
   were allocated into two groups: a control group receiving a basic diet
   and a test group receiving a basic diet supplemented with 30% corn
   silage. Each group comprised 12 donkeys, evenly distributed by sex (6
   males and 6 females). The ingredients and nutrient levels of diets were
   shown in Supplementary Table [48]S1.

   Two groups of donkeys were fed in individual stall (8 × 16 m) with a
   feeder (3.0 m long) and an automatic water dispenser. The entire
   feeding process was carried out by a specially trained person. Briefly,
   all donkeys were fed for ad libitum consumption, and had ad libitum
   access to water throughout the day. During this study, all donkeys were
   regularly examined by a veterinarian to confirm that they were healthy
   and without any metabolic or gastrointestinal disorder.

   All donkeys were weighed every 20 days, and at the same time point
   fresh feces were collected, handled, stored in 15 mL sterile
   cryopreservation tubes as aseptically as possible in order to prevent
   contamination in any manner, and were immediately placed in liquid
   nitrogen, and then transported to laboratory to store at − 80 °C. All
   samples were sent to OE Biotechnology Co., Ltd. for sequencing
   (Shanghai, China) and the samples at last time point were also needed
   for liquid chromatography–mass spectrometry metabolomics profiling in
   the same company.

   Throughout the entire experiment, no clinical diseases were observed.
   However, due to the pregnancy of two donkeys from the test group during
   the experiment, data related to these two samples were excluded at all
   time points. In addition, at the last time point, the fecal samples of
   two donkeys from the control group were too small to be used for
   bacterial diversity or metabolomics analysis, so data related to these
   two samples were also excluded at last time point. Finally, a total of
   130 samples were used for bacterial diversity analysis, and 20 samples
   used for metabolomics analysis at last time point.

DNA extraction, pyrosequencing and data processing

   Total bacteria DNA was extracted from the fecal samples using MagPure
   Soil DNA LQ kit (Magen, China) according to the manufacturer’s
   protocol. DNA was quantified using a NanoDrop spectrophotometer
   (Thermo, USA) and was stored at − 80 °C until further processing. The
   V3–V4 regions of bacterial 16S rRNA gene were amplified using universal
   primers (343F: 5′-TACGGRAGGCAGCAG-3′, 798R: 5′-AGGGTATCTAATCCT-3′).
   Amplicon quality was visualized using gel electrophoresis, purified
   with AMPure XP beads (Agencourt, USA), and amplified for another round
   of PCR. After purified with the AMPure XP beads again, the final
   amplicon was quantified using Qubit dsDNA assay kit (Thermo, USA).
   Equal amounts of purified amplicon were pooled on Illumina Novaseq 6000
   for sequencing according to the standard protocols. The raw reads were
   deposited into the NCBI Sequence Read Archive database in the
   BioProject with the identifier PRJNA1006207^[49]9.

   Raw sequencing data were in FASTQ format. Paired-end reads were then
   preprocessed using Trimmomatic software^[50]10 to detect and cut off
   ambiguous bases (N). It also cut off low quality sequences with average
   quality score below 20 using sliding window trimming approach. After
   trimming, paired-end reads were assembled using FLASH software^[51]11.
   Parameters of assembly were: 10 bp of minimal overlapping, 200 bp of
   maximum overlapping and 20% of maximum mismatch rate. Sequences were
   performed further denoising as follows: reads with ambiguous,
   homologous sequences or below 200 bp were abandoned. Reads with 75% of
   bases above Q20 were retained. Then, reads with chimera were detected
   and removed. These two steps were achieved using QIIME software^[52]12
   (version 1.8.0). Clean reads underwent primer sequence removal and
   clustering to generate operational taxonomic units (OTUs) using Vsearch
   software^[53]12 with a 97% similarity cutoff. The representative read
   of each OTU was selected using QIIME package. All representative reads
   were annotated and blasted against Ribosome Database Project (16 s/18 s
   rDNA) using RDP classifier^[54]13 (confidence threshold was 60%).

   Bacterial diversity including the number of sequences, the number of
   observed OTUs, Shannon diversity indices and Chao1 richness estimators
   was assessed according to the protocols on the website
   ([55]https://mibwurrepo.github.io/Microbial-bioinformatics-introductory
   -course-Material-2018/introduction.html). The dissimilarity of fecal
   microbiota from different groups was analyzed by detrended
   correspondence analysis (DCA) using the R package vegan. Linear
   discriminant analysis effect size (LEFSe) was performed to determine
   the significantly differential bacteria at the OTU level between two
   groups^[56]14 by using the R package microeco. Microbial ecological
   networks (MENs) were constructed according to the analysis pipeline on
   the website ([57]http://ieg4.rccc.ou.edu/mena/).

Liquid chromatography-mass spectrometry metabolomics profiling

   Approximately 60 mg of each frozen fecal sample was taken into 1.5 mL
   centrifuge tube and mixed with 20 μL internal standard (0.06 mg/mL,
   2-Chloro-l-Phenylalanine) with 600 μL of precooled methanol and water
   (4:1, v/v). Two small steel balls were placed into tubes and then the
   tube was placed in a – 20 °C refrigerator for 5 min before grinding
   them in a grinder (60 Hz, 2 min). Mix samples were then ultrasonic
   extracted for 10 min in an ice water bath and settled at − 20 °C for
   30 min. 200 μL supernatant was collected after being centrifuge for
   10 min (13,000 × g, 4 °C) and transferred to a clean vial for dry, and
   then resuspended with methanol and water (1:4, v/v; vortexed for 30 s
   and ultrasonic extracted for 3 min) before being settled at -20^◦C
   overnight. 150 μL supernatant was collected after being centrifuge for
   10 min (13,000 × g, 4 °C) and transferred to a clean vial for LC–MS/MS
   analysis. Quality control (QC) samples: take equal amounts of each
   sample and mix them as QC samples^[58]15.

   The treated sample (2 μL) was inject into a 100 × 2.1 mm ACQUITY UPLC
   T3 column (Waters, Milford, United States) packed with 1.8 μm particles
   and preheated at 45 °C for chromatographic separation of metabolites.
   MS data were obtained using a UHPLC-Q Extractive Mass Spectrometer
   (Thermo Fisher Scientific Inc., United States) equipped with
   electrospray ionization (ESI) source operating in either positive or
   negative ion mode. The conditions for separation were as follows:
   mobile phase A composed of water with 0.1% formic acid; mobile phase B
   compose of acetonitrile with 0.1% formic acid; flow rate 0.35 mL/min.
   The gradient eluted (A:B) from the column was used according to the
   following conditions: from 0 to 0.01 min, 95%:5%; from 0.01 to 2 min,
   95%:5% to 95%:5%.; from 2 to 4 min, 95%:5% to 70%:30%; from 4 to 8 min,
   70%:30% to 50%:50%; from 8 to 10 min, 50%:50% to 20%:80%; from 10 to
   14 min, 20%:80% to 0%:100%; from 14 to 15 min, 0%:100% to 0%:100%; from
   15 to 15.1 min, 0%:100% to 95%:5%; from 15.1 to 16 min, 95%:5% to
   95%:5%. The mass scan ranged from 70 to 1000 m/z. The resolution (full
   scan) was 70,000. The resolution (HCD MS/MS scans) was 17,500. The
   spray voltage was 3800 V and − 3000 V for electrospray ionization
   positive and negative ion, respectively. The sheath gas flow rate was
   40 Arb and 35 Arb for electrospray ionization positive and negative
   ion, respectively. The aux gas flow rate was 10 Arb and 8 Arb for
   electrospray ionization positive and negative ion, respectively. The
   capillary temperature was 320 °C for both electrospray ionization
   positive and negative ion.

Metabolic data preprocessing and annotation

   The processing and annotation of metabolic data were similar with the
   previous study reported by Zhang et al.^[59]16. In brief, the raw data
   were imported into Progenesis QI v2.3 (Nonlinear Dynamics, Newcastle,
   United States) for baseline filtering, peak recognition, integration,
   retention time correction, peak alignment, and normalization, employing
   the following parameters: precursor of 5 ppm, product tolerance of
   10 ppm, and product ion threshold of 5%. For the annotation of
   metabolic features, the Human Metabolome Database (HMDB), Lipidmaps
   (v2.3) and METLIN database were applied to align the m/z data and
   identify relevant metabolites. The confidently identified metabolites
   were further validated by the score above 36 and the filtered data was
   used for further analysis. The multivariate statistical analysis of
   metabolites was conducted by applying orthogonal projections to latent
   structures discrimination analysis (OPLS-DA) to visualize the global
   metabolic alterations between two groups. Variable importance in the
   projection (VIP) was calculated in the OPLS-DA module and the variables
   are considered relevant for group discrimination if the VIP > 1. The
   co-expression network analysis was conducted by the weighted gene
   co-expression network analysis (WGCNA) package^[60]17. A signed
   weighted metabolites coexpression network was constructed by setting
   the network type to “signed hybrid”, with min module size of 2, power
   of 8, deep split of 4, and MEDiss threshold of 0.4 using the function
   blockwiseModules for the analysis. To obtain the fecal metabolites
   related to the growth performance of donkeys, the differentiated
   metabolites (P < 0.05) and growth rates of donkeys from the control and
   test group were selected for integrative analysis. Metabolomics data
   have been deposited to the EMBL-EBI MetaboLights database with the
   identifier MTBLS8478.

Correlations between donkey performance, fecal bacteria and metabolites

   Procrustes analysis (PA) was used to evaluate the global similarity
   between the metabolome and bacteria of feces. If the global similarity
   between the metabolome and bacteria of feces was above 0.5, then fecal
   bacteria and all the metabolites in interested module were used to
   discover specific bacterial species that were closely associated with
   metabolites statistically and biologically by using spearman
   correlation analysis from MetOrigin
   ([61]https://metorigin.met-bioinformatics.cn/home/) with default
   parameters. In addition, pathway enrichment analysis of interested
   metabolites were also performed through MetOrgin.

Statistical analysis

   The data, used in Figs. [62]1, [63]2 and [64]5, were analyzed with SPSS
   20 (SPSS Inc., Chicago, USA), and Mann–Whitney U was applied to compare
   differences between groups at a given point. P < 0.05 was considered
   significant. The statistical methods of data used in Figs. [65]6 and
   [66]7 can be found in the section of “[67]Metabolic data preprocessing
   and annotation” and “[68]Correlations between donkey performance, fecal
   bacteria and metabolites”, respectively. All images were generated with
   origin 8.0 software or Rstudio.

Figure 1.

   [69]Figure 1
   [70]Open in a new tab

   The average daily weight gain of Dezhou donkeys during the period of
   experiment (a) and the growth rate of Dezhou donkeys at the end of
   experiment (b).

Figure 2.

   [71]Figure 2
   [72]Open in a new tab

   The effects of feeding corn silage on the relative abundance of
   bacterial community at phylum level (a) and the relative abundance
   ratio of Bacteroidetes to Firmicutes (b). Different uppercase letters
   means the significance at P < 0.05.

Figure 5.

   [73]Figure 5
   [74]Open in a new tab

   Network topologies over time, including total nodes (a), total links
   (b), proportion of positive links (c), average clustering coefficient
   (avgCC, d), average K (avgK, e) and robustness (f).

Figure 6.

   [75]Figure 6
   [76]Open in a new tab

   Characteristics of fecal metabolites and KEGG pathways between the
   control and test groups. (a) OPLS-DA scores plots; (b) Volcanic map of
   differential metabolites; (c) Clustering dendrograms of metabolites,
   with dissimilarity based on topological overlap, together with assigned
   module colors. There is one metabolite dendrogram per block. (d)
   Module-trait associations. Each row corresponds to a module
   eigenmetabolite, column to a trait. Each cell contains the
   corresponding correlation and P-value. The cell is color coded by
   correlation according to the color legend. (e) The scatter plot of
   metabolite significance for donkey growth vs turquoise module
   membership. (f) Metabolites enrichment analysis results from turquoise
   module.

Figure 7.

   [77]Figure 7
   [78]Open in a new tab

   (a) The BIO-Sankey Network for [79]R07035 metabolic reaction in
   arachidonic acid metabolism; (b) The STA-Sankey Network for [80]R07035
   metabolic reaction in arachidonic acid metabolism. The BIO-Sankey
   Network for [81]R07035 metabolic reaction in arachidonic acid
   metabolism. * indicate statistically significant correlations with
   metabolites. The dark red/green color of nodes indicate significantly
   up/down regulated microbes or metabolites (Fold change (FC) > 1 and
   P < 0.05). The light red/green color of nodes indicate up/down
   regulated microbes or metabolites (FC > 1 and P ≥ 0.05). The dark
   red/green color of bands indicate significant positive/negative
   correlation (P < 0.05). The light red/green color of bands indicate
   positive/negative correlation (P ≥ 0.05). The black color of nodes
   indicate microbes or metabolites in the reference database. The purple
   color of nodes indicate metabolic enzymes. The light gray bands
   indicate reference relationships searched through database.

Results

Dietary corn silage supplementation can improve the growth of Dezhou donkeys

   As illustrated in Fig. [82]1a, the mean daily weight gain of Dezhou
   donkeys in the test group fluctuated over time but was all higher than
   that of the control group at corresponding time points, especially with
   a significant increase during the first three time points. What’s more,
   the overall growth rate of Dezhou donkeys in the text group was higher
   than that in the control group at the end of the experiment
   (Fig. [83]1b).

Bacterial diversity in feces of donkeys fed with or without corn silage

   Individual-based rarefaction curves tended to be gentle which suggested
   that the sampling provided enough OTU coverage (Supplementary Fig.
   [84]S1). The Chao1 richness index and Shannon evenness index of two
   groups at almost all time points were increased over time. In addition,
   on the 40th, 60th, 100th days, the Chao1 richness index and Shannon
   evenness index of the test group were significantly higher than those
   of the control group (Table [85]1). As illustrated in Supplementary
   Fig. [86]S2, Detrended correspondence analysis (DCA) showed that the
   samples in different groups were gradually separated with time going.

Table 1.

   Comparison of the diversity indices of bacterial communities in
   different groups of Dezhou donkeys at different time points.
   Group Chao1 Fold change (test/control) P-value Shannon Fold change
   (test/control) P-value
   Control_0 3546.32 ± 508.57 1.02 0.68 6.37 ± 0.54 1.02 0.59
   Test_0 3635.13 ± 443.00 6.48 ± 0.31
   Control_1 3688.12 ± 456.18 1.05 0.43 6.45 ± 0.42 1.04 0.19
   Test_1 3879.77 ± 574.62 6.70 ± 0.35
   Control_2 3677.60 ± 275.53 1.12 0.003 6.38 ± 0.31 1.06 0.003
   Test_2 4106.77 ± 311.08 6.79 ± 0.21
   Control_3 3863.57 ± 330.09 1.12 0.002 6.32 ± 0.39 1.08 0.0013
   Test_3 4312.04 ± 201.94 6.83 ± 0.15
   Control_4 3771.06 ± 375.61 1.06 0.19 6.49 ± 0.25 1.02 0.35
   Test_4 4007.84 ± 405.91 6.60 ± 0.24
   Control_5 3631.99 ± 358.37 1.14 0.002 6.47 ± 0.27 1.04 0.047
   Test_5 4132.85 ± 251.65 6.74 ± 0.29
   [87]Open in a new tab

   Presented values of Chao1 and Shannon are means ± standard deviation
   (SD).

   At the phylum level, Fibrobacteres, Spirochaetae, Proteobacteria,
   Bacteroidetes and Firmicutes were the dominant phyla (> 1%) in feces of
   two groups over time (Fig. [88]2a, Supplementary Table [89]S2,
   Supplementary Fig. [90]S3). The abundance of Firmicutes in the feces of
   the test group significantly increased by an average of 1.12-fold on
   the 60th day while significantly decreased after 80 days compared with
   that in control group. The abundance of Bacteroidetes were lower in the
   feces of test group before 20 days and significantly lower at initial
   time point compared with control group, but after 40 days, the
   abundance of Bacteroidetes were higher in the feces of test group and
   were significantly higher at 40th and 60th days. The abundance of
   Proteobacteria were higher in the feces of test group after 40 days and
   was significantly higher at 60th day compared with the control group.
   Compared with control group, in test group the abundance of
   Fibrobacteres and Spirochaetae were in a fluctuating pattern over time,
   and the abundance of Spirochaetae was significantly lower at 60 days.
   Interestingly, after 20 days, the relative abundance of Bacteroidetes/
   Firmicutes was higher in the test group than that in the control group
   and was significant after 80 days (Fig. [91]2b).

   Figure [92]3a presented a heatmap depicting the dominant genera (those
   with a relative abundance greater than 1% at least once) in feces
   samples from two groups over time. A total of 25 genera were shown in
   Fig. [93]3a, of which 14 belonged to Firmicutes and 7 to Bacteroidetes.
   Prevotellaceae_UCG-003, Alloprevotella and Prevotella_1 all belonged to
   the phylum of Bacteroidetes. Despite fluctuations in their relative
   abundances, these genera exhibited enrichment in the test group over
   time.

Figure 3.

   [94]Figure 3
   [95]Open in a new tab

   The heatmap of dominant genera (the relative abundance > 1% at least
   one time point) in feces of two groups over time (a) and the heatmap of
   significantly different figures belonging to the phyla of Firmicutes
   and Bacteroidetes between the control and test group (b). The features
   within the red box represent their enrichment in the test group, while
   the features within the yellow box represent their enrichment in the
   control group.

   The relative abundance of Prevotellaceae_UCG-003 increased after
   20 days, with the exception of the 40th time point, and was
   consistently higher in the test group compared with the control group
   at corresponding time points. Alloprevotella exhibited increased
   enrichment after feeding with corn silage compared with the control
   group at the corresponding time point. The relative abundance of
   Prevotella_1 was lower before 40 days but higher after 60 days in test
   group compared with the control group. The relative abundance of some
   genera belonging to Firmicutes in control group were higher than that
   in test group after 80 days, such as Ruminococcaceae_NK4A214_group,
   Ruminococcaceae_UCG-010, Lachnospiraceae,
   [Eubacterium]_coprostanoligenes_group, and Ruminococcaceae_UCG-002.

   Growth-associated bacterial features were identified by using LEfSe.
   The abundance of these features is list in Supplementary Table [96]S3
   and the abundance of features belonging to Firmicutes and Bacteroidetes
   is visualized on a heatmap (Fig. [97]3b). A total of 77 features were
   identified, with 31 features showing higher abundance at most time
   points in the control group, including 22 features belonging to
   Firmicutes, and with 12 features showing higher abundance at most time
   points in the test group, including 10 features belonging to
   Bacteroidetes (Supplementary Table [98]S3). The heatmap (Fig. [99]3b)
   also confirmed the observations mentioned above. For example, 22 out of
   27 features within the Firmicutes phylum exhibited higher abundance in
   the control group, while 10 out of 18 features within the Bacteroidetes
   phylum were more abundant in the test group.

Dietary corn silage supplementation can increase the stability of MENs in
feces of donkeys

   To ascertain the impact of corn silage on the complexity and stability
   of MENs in donkey feces, the composition and structure of fecal
   microbial communities were analyzed in donkeys subjected to a 100-day
   corn silage diet compared with a control group. Twelve MENs
   (Fig. [100]4a) were constructed based on Pearson correlations of
   log-transformed OTU abundances, followed by a random matrix theory-
   (RMT-) based approach. The number of total nodes and links of empirical
   MENs at the last time point were more than that at the initial time
   point (Figs. [101]4a, [102]5a,b) in both groups, and there were
   significant linear correlations between the number of total nodes of
   MENs and time points (Fig. [103]4b). In addition, there were also
   significant linear correlations between the number of large modules and
   time points in test group, rather than in the control group
   (Fig. [104]4c). The overall topology indices (Supplementary Table
   [105]S4) revealed that all curves of network connectivity distribution
   were fitted well with the power-law model (R square of power-law values
   from 0.783 to 0.948), indicative of scale-free networks. Average
   clustering coefficient (avgCC) values decreased over time in test
   group, and was lower than that in control group after 60 days
   (Fig. [106]5d). Average K (avgK) values changed in fluctuating pattern
   before 60 days in test group, while the values were lower than that in
   control group after 60 days (Fig. [107]5e). The proportion of positive
   links was higher than that in control group when feeding donkeys with
   corn silage after 80 days (Fig. [108]5c). The robustness significantly
   increased in test group at 40th day and 80th day compared with the
   control group and showed no significant differences at other time
   points (Fig. [109]5f).

Figure 4.

   [110]Figure 4
   [111]Open in a new tab

   Succession of bacterial networks over time in the feces of donkeys. (a)
   Visualization of constructed MENs in control and test groups from 0 to
   100 days. Large modules with ≥ 5 nodes are shown in different colors,
   and small modules are shown in grey. (b) Temporal changes of the number
   of large modules. (c) Temporal changes of the number of nodes in large
   modules. In each panel, filled red symbols represent networks in test
   group, and filled blue symbols represent networks in control group.
   Slopes (b) and adjusted r^2 and P values from linear regressions are
   shown. *0.01 < P < 0.05.

Profile of fecal metabolites and enrichment of metabolic pathways

   The addition of corn silage increased the growth rate of donkeys at the
   end of experiment (Fig. [112]1b), suggesting the corn silage play an
   important role in promoting growth of donkeys. The metabolites of feces
   are the collection of nutrition digested and absorbed by the gut
   microbiota and host gastrointestinal tract over time, therefore,
   metabolites of feces at the end of this experiment were analyzed for
   extend our understanding the effects of silage corn addition on the
   donkey growth performance. The fecal metabolites of donkeys fed with
   silage corn or not were analyzed by a non-targeted LC–MS/MS
   metabolomics platform, 10,901 metabolites were totally detected in two
   groups (data not shown). Among them, 1169 differentiated (variable
   importance in the projection (VIP) > 1, P < 0.05) metabolites were
   identified (Fig. [113]6b), of which 690 metabolites were up-regulated
   and 479 metabolites were down-regulated. The OPLS-DA was conducted to
   compare the distribution of the fecal metabolites of the two groups
   (Fig. [114]6a). The results showed that the test group was completely
   separated from the control group, suggesting that fecal metabolites
   were typically differentiated by the addition of silage corn.

   The aim was to elucidate potential metabolic modules and pathways
   implicated in the impact of corn silage on donkey growth. The
   information obtained will be valuable in deciphering regulatory
   mechanism of corn silage on donkeys, and discovering or even designing
   new feed additives to further enhance donkey growth. Toward the goals,
   the differentiated metabolites of feces from donkeys fed with silage or
   not were analyzed. As results, a total of 23 modules were identified by
   the signed weighted metabolites coexpression network. The hierarchical
   clustering dendrograms was visualized in Fig. [115]6c. To identify
   modules that are significantly associated with the growth rate and
   further decipher the regulatory mechanism of corn silage on the donkey
   growth, 23 modules were correlated with growth rates and the full
   module-trait correlations were shown in Fig. [116]6d. The results
   showed that the enginmetabolites in turquoise module (121 metabolites
   associated with growth rate of donkeys from test group r = 0.95,
   correlation P-value = 1 × 10^–10) were positively correlated with
   donkey growth. In addition, the scatter plot of metabolite significance
   vs turquoise module membership showed a significant correlation
   (Fig. [117]6e). The turquoise module, which displayed a significant
   correlation with donkey growth, was selected for pathway enrichment
   analysis using MetOrgin. The results showed that metabolites in
   turquoise module were significantly (P < 0.05) enriched in pathway
   “steroid hormone biosynthesis” from host, “arachidonic acid metabolism”
   and “pentose and glucuonate interconversions” from host and microbiota
   co-metabolism (Fig. [118]6f).

Correlations between donkey performance, fecal bacteria and metabolites

   PA was used to evaluate the global similarity between the metabolome
   and bacterial community of feces (Supplementary Fig. [119]S4). The
   result showed that the correlation coefficient R is 0.67
   (P-value = 1 × 10^–4) which is close to 1, indicating the greater
   overall similarity was existed between the fecal microbiome and
   metabolites. Therefore, fecal bacteria and all the metabolites in
   turquoise module were used to discover specific bacterial species that
   were closely associated with metabolites statistically and biologically
   by using spearman correlation analysis from MetOrigin. According to the
   metabolites from both host and bacteria, two significant metabolic
   pathways were identified between two groups, that is, arachidonic acid
   metabolism and pentose and glucuonate interconversions (Fig. [120]6f).
   In arachidonic acid metabolism pathway, two differential metabolites
   (15-lipoxygenase by [1-14C]-(15S, 5Z, 8Z, 11Z,
   13E)-15-hydroperoxyeicosa-5, 8, 11, 13-tetraenoic acid (15S-HpETE) and
   PGG2) were involved, which participated in a total of four different
   metabolic reactions ([121]R00073, [122]R01590, [123]R01593 and
   [124]R07035). Bio-Sankey and STA-Sankey networks were used to explore
   the biological and statistical correlations between microbiota and
   metabolites for each metabolic reaction. In the Bio-Sankey network,
   Proteobacteria and Bacteroidetes as major phyla positively associated
   with the metabolic reaction [125]R07035, while Firmicutes and
   Actinobacteria as major phyla negatively associated with the metabolic
   reaction [126]R07035 (Fig. [127]7). In STA-Sankey networks,
   intergrative statistical correlation analysis confirmed that
   Bacteroidetes had the statistically (P < 0.05) positive correlations
   with 15S-HpETE, while Firmicutes had the statistically (P < 0.05)
   negative correlations with 15S-HpETE. In pentose and glucuonate
   interconversions pathway, two differential metabolites
   (3alpha-androstanediol glucuronide and d-Xylulose) were involved, which
   participated in a total of seven different metabolic reactions
   ([128]R01432, [129]R01478, [130]R01639, [131]R01896, [132]R01898,
   [133]R05604 and [134]R07143). Based on the Bio-Sankey and STA-Sankey
   networks, Firmicutes was statistically negatively associated with
   d-Xylulose and 3alpha-androstanediol glucuronide in reaction
   [135]R01432, [136]R01639, [137]R01898, [138]R01896 and [139]R01478,
   respectively, while Actinobacteria was statistically positively
   associated with D-Xylulose in reaction [140]R07143 (Supplementary Figs.
   [141]S5–[142]10).

Discussion

   The microbial community in the gut of hosts can be influenced by many
   factors, such as age, sex, weight, health status and so on. Therefore,
   in present study, the feeding trail was studied at the same time, and
   Dezhou donkeys were selected based on their similar body weight
   (184 ± 1 kg), similar age (8 ± 0.5 months), gender ratio (1:1) and no
   clinical diseases. Corn silage is commonly used for ruminants to
   improve their growth performance or meat quality, such as lambs and
   growing beef steers^[143]18,[144]19. Similar to previous study, in this
   study, the mean daily weight gain of Dezhou donkeys also increased
   after feeding corn silage and was about four times higher than that of
   the control group at 20th day, but variations in daily weight gain
   within the same group at different time points were found, consistent
   with findings reported by Liu et al.^[145]20.

   Microbial diversity in feces was demonstrated well correlated with
   different diets, thus, it is considered a potential good indicator when
   investigate the intestinal ecosystem of hosts^[146]21,[147]22. In
   present study, the diversity of the bacterial community (Chao 1 and
   Shannon indexes) was higher in test group than that in control group
   after feeding silage corn for 40 days, with the exception of the 80th
   day, indicating that a dramatic shift in fecal bacterial composition
   between the two groups because of the introduction of corn silage. This
   result was further supported by the DCA, which indicated that the
   microbial community structure in the feces of donkeys was influenced by
   the introduction of corn silage over time. These findings were similar
   with previous research identifying the microbial community structure of
   feces in cattle could be affected by diets^[148]23.

   The Firmicutes/Bacteroidetes ratio (F/B) was generally used as a
   relevant biomarker of the health of gut microbiota. For instance, the
   F/B ratio was reported to be increased in overweight person and tended
   to increase with age from 0 to 69 years old in Ukraine
   population^[149]24,[150]25. Ahmed et al. also reported that the
   decrease ratio of Firmicutes to Bacteroidetes in the gut could inhibit
   the development and progression of diabetes mellitus^[151]26. In
   addition, the F/B ratio was also found to be linked to methanogenic
   process parameters including volatile fatty acids concentration,
   organic loading rate, and methane production^[152]27. In present study,
   feeding corn silage to Dezhou donkeys resulted in a significant
   decrease in the F/B ratio, accompanied by an increase in the richness
   of Bacteroidetes, specifically Prevotella_1, Prevotellaceae_UCG-003,
   and Alloprevotella, which were linked to carbohydrate metabolism and
   the production of SCFAs^[153]3,[154]24. SCFAs, a particularly versatile
   class of microbial metabolites, derived from microbial fermentation of
   dietary fibers and protein, were reported to have impacts on various
   aspects of hosts, such as regulating the absorption of various
   nutrients, providing energy requirements, and improving host
   performance^[155]3,[156]28–[157]31. Thus, the increased relative
   abundance of Bacteroidetes following corn silage consumption might be a
   vital contributor to the improved weight gain of Dezhou donkeys.

   Dietary interventions are thus a potential tool to impact the health of
   hosts by modulating the gut microbes, the gut microbes could in turn be
   used to identify those microorganisms that benefit from dietary
   interventions^[158]32–[159]34. Prevotella, a prevalent genus within the
   Prevotellaceae family, has been shown to increase the ability of host
   gut microbes to digest complex polysaccharides from the diet and
   produce succinate^[160]34. Moreover, many strains belonging to
   Prevotellaceae possess the ability to decarboxylate succinate to
   propionic acid^[161]35,[162]36. Alloprevotella was known as a
   SCFAs-producing bacteria and usually found in the gut microbiota of
   host using traditional Chinese medicine, which was negatively related
   to diabetes and metabolic syndrome^[163]37,[164]38. Although there was
   a trend toward enrichment of Bacteroidetes in the test group with time
   going, some of the members belonging to the Fimicutes were also
   increased, such as the unclassified_UCG_005, indicating that these
   microbial species belonging to different phyla have the similar
   biological function and belong to the same “guild”^[165]39.
   Ruminococcaceae and Lachnospiraceae belonging to the phylum of
   Firmicutes were reported to have the ability to decompose substrates
   that were indigestible for hosts, such as cellulose and hemicellulose
   components of plant, and then convert into SCFAs which could be
   absorbed and utilized by hosts^[166]40, but had a lower relative
   abundance in the gut of donkeys after feeding corn silage. Therefore,
   Firmicutes and Bacteroidetes had contrastive population dynamics in
   response to diets containing with corn silage or not in this study,
   which implies that some species from these two phyla compete for the
   same niche in the gut.

   Previous studies reported that the microbial diversity was related to
   the stability of the ecological network, and then was used to reflect
   the homeostasis of the intestine^[167]41,[168]42. For example, Su et
   al. suggested that a diet low in selenium can regulate microbial
   community and enhance their stability, ultimately promoting fish
   growth^[169]43. Our study also showed that dietary corn silage could
   enhance the microbial stability. In this study, the positive links in
   the microbial community of donkeys fed with corn silage after 80 days
   were more than that in control group. In addition, there were more
   negative links than positive links in the control group. These results
   were similar with the previous study, which reported that
   metacommunity-level foundation predators might enhance regional
   ecological stability by preventing the outbreaks of competitively
   superior prey species in each local community^[170]44.

   Compared with donkeys in control group, compounds (15-HpETE and PGG2)
   involved in the arachidonic acid metabolism pathways were determined to
   significantly enrich in the fecal metabolites of the donkeys fed with
   corn silage. Arachidonic acid (ARA, 20:4n-6) is an omega-6 (n-6) long
   chain-polyunsaturated fatty acid. It was reported that there was a
   significant and positive correlation between human body weight and ARA
   in plasma^[171]45. And a growing number of studies are suggesting that
   ARA take part in various physiological processes, such as reproduction,
   health, growth, stress resistance and so on^[172]46,[173]47. It was
   reported that the rabbit 15-lipoxygenases could be suicidal inactivated
   by 15-HpETE by covalent modification of the active site
   peptides^[174]48, suggesting 15-HpETE may have an important role in
   regulation of accumulation of ARA. What’ more, the abundance of
   Bacteroidetes had the statistically (P < 0.05) positive correlations
   with 15S-HpETE, while the abundance of Firmicutes had the statistically
   (P < 0.05) negative correlations with 15S-HpETE. Thus, the analyses of
   both fecal microbiota and metabolites showed that the ratio of
   Bacteroidetes to Firmicutes in feces might be one of the key reasons
   for promoting the growth of donkeys.

   Compounds (3alpha-androstanediol glucuronide and d-Xylulose) associated
   with pentose and glucuonate interconversions pathway also significantly
   enriched in the fecal metabolites of the donkeys fed with corn silage,
   indicating that the addition of corn silage could affect glycan
   biosynthesis and metabolism which may be contribute to the growth of
   donkeys. The result was similar with the findings of Tang et al., who
   found that the enrichment of metabolites in glycan biosynthesis and
   metabolism play an important role in the growth and development of
   Mesona chinensis Benth^[175]49. In addition, the abundance of
   Firmicutes had significantly negative correlations with
   3alpha-androstanediol glucuronide and d-Xylulose. It thus suggested
   that the remodeling of microbiota and metabolome in the feces might be
   crucial for the growth of the donkeys in the test group compared with
   those in the control group.

   Unfortunately, no SCFAs were detected out in this study, which may be
   related to that the content of SCFAs were too low to detected out by
   LC–MS/MS. When further discussing the underlying mechanisms
   contributing to the observed diet—related changes in microbial
   community composition and metabolome of Dezhou donkeys, it is
   apparently difficult to distinguish correlation from causality in the
   findings observed. Indeed, the microbiota, microbial metabolites and
   the host influence each other reciprocally by various ways among
   themselves during the experiment. Therefore, more in—depth researches
   are needed for a better understanding the underlying mechanisms of
   feeding corn silage to promote the growth of Dezhou donkeys.

Conclusions

   In summary, dietary supplementary of corn silage promoted the growth of
   Dezhou donkeys, and altered the bacterial diversity and metabolome in
   the feces, especially, the relative abundance ratio of Firmicutes to
   Bacteroidetes, members in which were also related with the
   differentiated metabolites enriched in arachidonic acid metabolism and
   pentose and glucuonate interconversions pathways.

Supplementary Information

   [176]Supplementary Information.^ (1.5MB, pdf)

Acknowledgements