Solid diet introduction experiment design

The animal experiment was conducted on a sheep breeding conservation farm (Huzhou, Zhejiang Province, China) from September to November in 2018. The detailed description of animal welfare during lamb experiment is presented in Additional file 9. Forty healthy Hu lambs (11-day-old) were separated from their dams and fed mixed goat milk using a nursing bottle (water: goat milk powder = 10:1). After 3 days of adaptation, the lambs were randomly assigned to four groups, including only milk feeding (M, n = 10), milk plus alfalfa hay feeding (MH, n = 10), milk plus corn-soybean starter feeding (MC, n = 10), and milk plus alfalfa hay and corn-soybean starter feeding (MHC, n = 10). The animals were selected at a comparable initial body weight. The lambs in M group consumed mixed goat milk powder (23.85% crud protein, 25.30% fat, and 36.30% lactose) at will. The lambs in MH, MC, and MHC groups were fed 600 mL/day milk equaling 10% of their initial average body weight [50, 51] and could freely get access to solid feed. The mixed goat milk powder was provided four times daily (07:00, 12:00, 17:00, and 22:00), and the solid diets were provided twice daily (08:00 and 17:00). The nutritional levels of solid diet were shown in Additional file 10: Table S8. At 42 days of age, eight healthy lambs per group were randomly selected and stunned using electric shock. After slaughter using exsanguination, all rumen content was immediately mixed, and 10 g of ruminal content was stored in liquid nitrogen for microbial DNA extraction. The ruminal fluid sample was strained through four layers of sterile cheesecloth and stored in liquid nitrogen for the analysis of metabolome. After measuring rumen volume and weight, one portion of the rumen wall (3 × 3 cm) from the ventral sac was collected for morphology analysis using histomorphometry microscopy, as described previously [52]. After being washed three times in cold phosphate buffer saline, the ruminal epithelium and muscle layer tissue were collected and stored in liquid nitrogen for RNA extraction.

3-IAld and PGD2 infusion experiment design

The 3-IAld and PGD2 infusion studies were conducted on a sheep breeding conservation farm (Huzhou, Zhejiang Province, China) from February to April in 2022. The detailed description of animal welfare during lamb experiment was presented in Additional file 9. Thirty healthy male Hu lambs (11 days of age) were separated from their dams and fed mixed goat milk using an artificial nursing bottle (water: goat milk powder = 10:1). After 3 days of adaptation, 24 lambs with good adaptation to artificial nursing were randomly assigned to three groups with the following treatments: normal saline infusion (Con, n = 8), 3-IAld infusion (3-IAld, n = 8), and PGD2 infusion (PGD2, n = 8). There was no significant difference in initial body weight among three groups. All lambs were provided mixed goat milk powder at their will, and the feeding method was the same as M group lambs. 3-IAld (Aladdin, China) was delivered daily for 28 days by oral infusion at a dose of 30 mg/kg body weight per day in a vehicle of DMSO/saline (1:79). First, the 3-IAld powder was dissolved in vehicle at a concentration of 18.75 mg/mL; then, the 3-IAld group lambs received an oral infusion at 1.6 mL/kg body weight per day. PGD2 (Santa Cruz, CA) was delivered daily for 28 days by oral infusion at a dose of 20 µg/kg body weight per day in a vehicle of DMSO/ saline (1:79). First, the PGD2 powder was dissolved in vehicle at a concentration of 12.5 µg/mL; then, the PGD2 group lambs received an oral infusion at 1.6 mL/kg body weight per day. The Con group lambs received a vehicle at 1.6 mL/kg body weight by oral infusion. The amount of infusion was adjusted weekly according to body weight. The physiological concentrations of 3-IAld and PGD2 in the rumen liquid were 0.21 mg/mL and 0.15 µg/mL. Oral infusion dose was calculated by the physiological concentration of 3-IAld or PGD2 multiplied by the rumen volume (42 days of age), and then divided by body weight (42 days of age) of lambs. Targeted metabolomic analysis of 3-IAld was performed as previously reported [53]. The measure method of absolute concentration of PGD2 was detected by various species PGD2 ELISA kit (Cloud-clone corp., USA). During the feeding and oral infusion period, lambs with diarrhea, pneumonia, and aphtha were excluded. At 42 days of age, six healthy lambs per group were slaughtered and the ruminal epithelium and muscle layer tissue were collected for morphology analysis and RNA extraction.

Microbial culture experiment design

Bifidobacterium pseudocatenulatum DSM 20438, Bifidobacterium longum ATCC15707 and Bifidobacterium adolescentis BNCC134301 were obtained from the DSMZ (Braunschweig, Germany), ATCC (Manassas, VA, USA), and BeNa Culture Collection (Beijing, China), respectively. Bifidobacterium pseudocatenulatum DSM 20438 and Bifidobacterium adolescentis BNCC134301 were cultured for 16 h at 37 °C in DSM medium 58 under an anaerobic workstation. Bifidobacterium longum ATCC15707 was cultured for 16 h at 37 °C in ATCC medium 2107 under an anaerobic workstation. Candida albicans BNCC186382 was obtained from the BeNa Culture Collection (Beijing, China), and cultured for 24 h at 37 °C in PYG medium (Beijing, China) under a constant temperature incubator.

At the initial stationary phase of strain growth, 30 mL bacterial solution was centrifuged at 4000 r at 4 °C for 20 min. The supernatant were removed and bacteria were washed using pure M9 minimal microbial growth medium (Sigma, USA). After re-centrifuged with the same condition, 15 mL M9 culture media was added to the tube and divided into three sterile 15-mL tubes (Gbico, USA). Then, the bacteria were cultured for 0, 2, 6, and 12 h under anaerobic conditions (5% CO2, 10% H2, and 85% N2). For the Bifidobacterium species, the M9 medium contained 1 mM Trp (Sigma, USA). For the Candida albicans, the M9 medium contained 1 mM arachidonic acid (Sigma, USA) or lecithin (Sigma, USA). Then, a 200 μL strain medium or medium control was mixed with 800 μL precooled methanol containing 1 μg/mL 4-chloro-phenylalanine (Sigma, USA) and the mixture was stored in a − 80 °C refrigerator for 30 min to remove protein. The mixture was centrifuged at 18,000 × g under 4 °C for 10 min to obtain supernatant. After vacuum concentration and centrifuge treatment, the final samples were obtained for UPLC-Q-TOF/MS analysis.

Cell culture experiment design

Primary ruminal epithelia cells and smooth muscle cells were isolated from the rumen wall of Hu lambs (42 days of age), as described in an early report, with some modifications [54]. Briefly, the ventral blind sac of the rumen wall (10*10 cm) was collected and washed with ice-cold D-hanks containing 100 U/mL of penicillin and 100 mg/mL of streptomycin. Then, the ruminal epithelia and muscle layers were separated and digested repeatedly with 0.25% trypsin. Digestion with trypsin was stopped until enough individual epithelial or smooth muscle cells appeared in the digestion solution. The obtained cell suspension was centrifuged and washed with PBS. Then, the cells were cultured in F12/DMEM medium (GBICO, New York, USA) with 10% FBS (GBICO, New York, USA) and 1 × antibiotic–antimycotic (GBICO, New York, USA). The morphology and identity of primary rumen epithelial cells or rumen smooth muscular cells were authenticated using microscopic observation (Additional file 1, Fig. S9a) and immunofluorescent staining of specific protein markers (Additional file 1, Fig. S9b and c). The specific protein marker keratin 14 (KRT14) of rumen epithelial cells was stained with rabbit anti-KRT14 (1:200; Abclonal; A15069). The specific protein marker α-smooth muscle actin (αSMA) of smooth muscular cells was stained with rabbit anti-αSMA (1:800; Proteintech; 14,395–1-AP). The images were visualized with laser scanning confocal microscopy (Zeiss LSM 900/Axio Observer 7, Jena, Germany). After adherent purification, the cells underwent serum starvation (0.5% FBS) for 24 h.

To investigate the effect of 3-IAld on primary ruminal epithelia cell proliferation, the cells were treated with DMSO (control group) or 3-IAld (3-IAld group; Sigma Aldrich, Saint Louis, MO). According to previous studies [48, 55] and our pre-experimental results, we observed that 30 μM concentrations of 3-IAld had the best effect (Additional file 1: Fig. S8a). The cells were cultured for 24 h for RNA extraction and analysis of cell cycle process, cell proliferation ratio, and protein fluorescence intensity. CH223191 (MedChem Express, San Diego, USA) and XAV-939 (MedChem Express, San Diego, USA) are the specific inhibitors of AhR and β-catenin protein. The cells were pretreated with CH223191 (10 uM, 3-IAld + AhR inhibitor group) and XAV-939 (5 µM, 3-IAld + CON inhibitor group) for 1 h before addition of 3-IAld. The control and 3-IAld groups were treated with DMSO instead of the inhibitor. To identify the effect of PGD2 on primary ruminal smooth muscle cell proliferation, the cells were treated with DMSO (control group) or PGD2 (Sigma Aldrich, Saint Louis, MO; PGD2 group). According to previous studies [56, 57] and our pre-experimental results, we found that 0.10 μM concentrations of PGD2 had the best effect (Additional file 1: Fig. S8b). Cells were cultured for 24 h for RNA extraction and analysis of cell cycle process and EdU+ label ratio. The cells were pretreated with KN-93 (10 µM, a specific inhibitor of the CAMK2 protein; 3-IAld + CAMK2 inhibitor group) for 1 h before PGD2 addition. The control and PGD2 groups were treated with DMSO instead of the inhibitor. For all cell culture experiment, the MycoBlue Mycoplasma Detector (Vazyme, China) was used to test and eliminate mycoplasma contamination.

Histological measurements

Rumen papilla morphology and light microscopy histomorphometric analysis were carried out through the method described by previous report [52]. In short, a piece of ruminal tissue (1 × 1 cm) was counted for the density of ruminal papillae, and then, 15 papillae were randomly selected for measuring length and width of ruminal papillae using a sliding caliper. The ruminal epithelial absorption area was calculated as papillae length × width × density × 2. Rumen wall samples were fixed in 4% paraformaldehyde and processed for paraffin imbedding and sectioning. Hematoxylin and eosin staining was used to measure the thickness of ruminal epithelia and smooth muscle layer through Image-Pro Plus 6.0 (Media Cybernetics Inc., Bethesda, MD). Three ruminal papillae per sample were selected for analysis, and five images were captured per papillae. The mean values of 15 replicate images per specimen were calculated, and different specimens were considered as repeated measures.

Metabolomics and analysis

For GC–MS analysis, 100 μL of rumen liquid samples was mixed with 800 μL of methanol. The mixed sample was vortexed for 30 s and placed for 1 h at − 20 °C. Subsequently, the samples were centrifuged at 12,000 rpm at 4 °C for 15 min. The 200 μL supernatant was collected and then evaporated until dry at room temperature. After evaporation, the samples were derivatized by shaking them with 35 μL of methoxyamine hydrochloride (20 mg/mL) in pyridine for 90 min incubation at 37 °C. Then, the samples were trimethylsilylated by adding 35 μL of BSTFA and incubating them for 1 h at 70 °C and 1 h at room temperature. The supernatant was used for GC–MS analysis. A gas chromatography system (Agilent 6890A/ 5973C, Palo Alto, CA, USA) coupled with a Pegasus HT (LECO, Shanghai, China) time-of-flight mass spectrometer (GC-TOF–MS) was used to identify the metabolites fitted with a DB-5MS capillary column (30 m × 0.25 mm × 0.25 μm; J&W Scientific, Folsom, CA, USA). Here, 1 μL aliquot of the sample was injected in splitless mode. Helium was used as the carrier gas. The primary temperature was kept at 70 °C for 2 min before being increased to 200 °C at a rate of 10 °C /min and then raised to 280 °C at a rate of 5 °C/ min and kept at this temperature for 6 min. The column effluent was fully scanned in the mass range 50–550 m/z. The data was performed feature extraction and preprocessed with XCMS in R software, and then normalized and edited into two-dimensional data matrix by excel 2016 software, including retention time, mass-to-charge ratio, and observations and peak intensity. Finally, using NIST library identify matching metabolites though retention time and m/z.

For UPLC-Q-TOF/MS analysis, 20 μL of rumen liquid samples was mixed with 200 μL of methanol. The mixed sample was vortexed for 30 s and placed for 1 h at − 20 °C. Subsequently, the samples were centrifuged at 18,000 rpm and 4 °C for 15 min. The 100 μL supernatant was collected and then evaporated until dry at room temperature. The sample was reconstituted in 100 μL of methanol (LC–MS grade, Merck) before analysis on an LC − MS/MS. For less polar compounds, LC separation was conducted on an Atlantis T3 (100 mm × 2.1 mm, 3.0 μm; Waters) using a gradient of solve A (5 m Mammonium formate and 0.05% formic acid buffer) and solvent B (acetonitrile). Next, 5 μL of the sample was injected. The flow rate was 0.25 mL/ min. The gradient was 0–3 min, 5% B; 3–8 min, 5–65% B; 8–10 min, 65–95% B; 10–12.5 min, 95% B; 12.5–13 min, 95–5% B; 13–17 min, 5% B. For polar compounds, LC separation was conducted on a XBridge BEH Amide column (4.6 mm × 100 mm, 3.5 μm; Waters) using a gradient of solve A (15 mM ammonium acetate and 0.3% ammonia buffer, pH = 9) and solvent B (acetonitrile). After this, 5 μL of the sample was injected, and the flow rate was 0.4 mL/ min. The gradient was 0–1 min, 85% B; 1–12 min, 85–30% B; 12–13 min, 30% B; 13–14 min, 30–85% B; 14–27 min, 85% B. The 6545 XTQ-TOF mass spectrometer was operated with a spray voltage of − 3.5 kV in negative mode and + 4 kV in positive mode. Drying gas was set at 9 L/min, and sheath gas was set at 110 L/min. The sheath gas temperature was 325 °C. Fast data-dependent acquisition (DDA) MS/MS experiments were performed with a collision energy map in the mass range 50–1000 m/z. The Progenesis QI (Nonlinear Dynamics, Newcastle, UK) was used for peak picking and alignment. Molecular identification of the assigned biomarkers was accomplished by matching the acquired precursors and fragment ions against several standard metabolome databases, including the Human Metabolome Database (http://www.hmdb.ca/), MassBank (http://www.massbank.jp/index.html), and METLIN (http://metlin.scripps.edu/index.php). Partial metabolite identification was further confirmed by comparison with the available standards.

The web-based tool Metabo Analyst 6.0 (https://www.metaboanalyst.ca/) was used for data normalization and analysis, along with for metabolites functional enrichment. The peak area data was normalized by sample median, log transformation and autoscaling. The differential metabolites were determined by a fold-change threshold of 2 and false discovery rate (FDR) of < 0.05 from the Wilcoxon rank-sum test and Benjamini and Hochberg multiple testing correction.

Clustering of ruminal metabolites via WGCNA analysis

R software package WGCNA 1.69 [58] was used to identify key phenotype-related metabolic modules based on correlation patterns. The Pearson correlation matrix was calculated for all possible metabolite pairs and then transformed into an adjacency matrix with soft thresholding power set to 8 for the best topological overlap matrix. A dynamic tree cut algorithm was used to detect groups of highly correlated metabolites. The minimum module size was set to 5, and the threshold for merging module was set to 0.25 as default. The profile of each metabolite cluster was summarized by the mainly class of metabolites. Each module was assigned a unique color and contained a unique set of metabolites. These resulting modules containing metabolites highly correlated with one another were then used in data integration to identify the relationships between rumen metabolites and rumen development phenotypes.

Generation of random forests classification models

In the present study, the random forest was used to select differentiating biomarkers among the different groups. A random forest classification was performed using the Random Forest package in R (version 3.6.2) with 500 trees and tenfold cross-validation to obtain robust estimates of the generalization error and feature importance. We used out-of-bag (OOB) error rate to measure the performance of the model. Our dataset was partitioned into a training set (including 70% of the samples) and a validation set (including the remaining 30%). Prism 8 (GraphPad Software, La Jolla, CA, USA) was used to construct a graph of TOP 20 metabolites based on mean decrease accuracy value.

Shotgun metagenome sequencing and analysis

Genomic DNA of the rumen content microbiota was extracted using a DNA Kit (EZNA, Omega Bio-Tek, Norcross, GA) according to the manufacturer’s protocols. The quantity and quality of the microbial DNA were examined using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific) and 1.0% agarose gel electrophoresis. All extracted DNA samples were stored at − 80 °C until subsequent processing. Genomic DNA was used with Illumina’s TruSeq for library preparation. Libraries were pooled, and paired-end sequencing was conducted on an Illumina HiSeq PE 150 Platform. Then, BWA (version 0.7.12) [59] and Fast QC (version 0.11.8) [60] were utilized to delete the adaptors, low-quality reads, and ovis aries and diet (maize, medicago, soybean, and wheat) contaminations in the sequencing raw data. The obtained clean reads were assembled using MEGAHIT (version 1.1.1) [61] based on the option of min-contig-len 500. We used Prodigal (version 2.6.3) [62] to predict the gene function depending on contigs from each sample and took advantage of CD-HIT to cluster the assembled contigs depending on the 95% cutoff sequencing identity. Finally, the pan-metagenome was used to analyze changes in metagenome functions. Entries in all the gene catalogs were subjected to taxonomic and functional assignment using DIAMOND [63] (v.0.9.22) based on BLASTP searches against the NCBI-NR (October 2018; approximately 550 M sequences) and KEGG [64] (v.90.0) databases (parameter: –evalue 0.00001 –max-target-seqs 10). The high-quality reads from each sample were aligned against the gene catalogs using BWA-MEM [59] (v.0.7.17), and abundance profiles of genes (alignment length ≥ 50 bp and sequence identity > 95%) were calculated in transcripts per million (TPM) [65], with corrections for variations in gene length and mapped reads per sample. TPM is calculated as:

$$TPM=\frac}}}}\times \frac}\frac}}}}}\times 10$$

where Ng is the read count, i.e., the average number of reads mapped to the g gene; and Lg is the gene length, i.e., the number of nucleotides in the g gene. The index j stands for the set of all genes determined in a catalog, and g is an index indicating a particular gene [65]. The relative abundances of taxa and KOs were calculated from the abundances of annotated genes [66]. Briefly, for the taxonomic profiles, we used phylogenetic assignment of each annotated gene from the rumen microbial gene catalog and summed the relative abundances of genes from the same phylum or genus to produce the abundance of each phylum or genus. The profile of each KO was calculated using the same process. The relative abundance of a KEGG pathway was calculated from the summation of the relative abundances of its contained KOs. The parameters or codes of software and packages mentioned were descripted in the previous studies [46, 67].

Transcriptome analysis

Total RNA was extracted by TRIzol (Invitrogen Life Technologies, Carlsbad, CA, USA), and concentrations and purity were measured using NanoDrop (NanoDrop Technologies, Wilmington, DE, USA). An RNA Nano 6000 Assay Kit was used to assess integrity (Agilent Technologies, CA, USA). Here, 1000 ng RNA per sample was used as the input material for the cDNA library construction. Sequencing libraries were generated by the NEBNextUltraTM RNA Library Prep Kit for Illumina (NEB, USA) according to the manufacturer’s instructions. To select cDNA fragments that were preferentially 240 bp in length, the library fragments were purified with the AMPure XP system (Beckman Coulter, Beverly, USA) and then amplified by PCR. Finally, PCR products were purified (AMPure XP system), and library quality was assessed using the Agilent Bioanalyzer 2100 system. The library preparations were sequenced on an Illumina platform, and paired-end reads were generated.

Clean reads were processed by deleting low-quality reads, reads with adaptor sequences, and reads including 0.5% unknown bases in raw reads, which were then mapped to the ovisaries reference genome (Oar v3.1) using TopHat (http://tophat.cbcb.umd.edu/; v2.0.9). Gene expression levels were calculated by fragments per kilobase of transcript per million fragments mapped (FPKM). The DEGs changed by diet were identified by comparing any two groups using the DESeq R package (1.10.1). The DEGs were identified by the FDR < 0.05 based on Benjamini and Hochberg multiple testing correction, as well as a fold-change (FC) of > 1.5 or < 0.667. Then, genes that were differentially expressed in any two groups were identified as DEGs. Finally, the GO enrichment analysis of DEGs was carried out by DAVID (version 6.8). KOBAS (version 3.0) was used to test the statistical enrichment of DEGs in the KEGG pathways.

Cell proliferation, protein fluorescence intensity, and quantitative real-time PCR

Cell proliferation was detected by incorporating EdU according to the EdU 594 cell proliferation kit (Beyotime, Shanghai, China), following the manufacturer’s instructions. The process of cell cycle analysis was according to the previous description by Gui et al. [54]. The cell immunofluorescence assay was performed as the following methods. The cells were fixed with paraformaldehyde for 30 min, permeabilized with 0.3% Triton X-100 for 15 min, incubated in 5% BSA for 1 h, and then incubated with primary antibodies (anti-rabbit AhR antibody, 1:200; anti-rabbit CYP1A1 antibody, 1:100; anti-rabbit β-catenin, 1:200; anti-rabbit CAMK2, 1:200; Proteintech) overnight at 4 °C. Next, the cells were incubated with a goat anti-rabbit antibody conjugated to Alexa Fluor 594 (1:500, Abcam) for 60 min at room temperature. Nuclei were stained with DAPI (1:5000, Invitrogen) for 5 min. The samples were examined with a Zeiss 710 laser scanning confocal microscope. Fluorescence images were collected for further qualitative and quantitative analyses.

The total RNA extract, cDNA synthesis, and real-time quantitative PCR were performed according to Lin et al. [46]. Next, qRT-PCR of all genes was performed using the QuantStudio 7 flex Real-time PCR Instrument (Applied Biosystems, Foster City, CA, USA) with fluorescence detection of SYBR green dye. Amplification conditions were as follows: 30 s at 95 °C followed by 40 cycles composed of 5 s at 95 °C, 34 s at 60 °C, 15 s at 95 °C, 60 s at 60 °C, and 15 s at 95 °C. Glyceraldehyde 3-phosphate dehydrogenase was used as a housekeeping gene to normalize the mRNA levels of each gene. The primers and amplicon sizes of all genes are presented in Additional file 10: Table S9.

Statistical analysis

SPSS version 22.0 (SPSS, Inc) was used for statistical analysis unless otherwise indicated. For body weight and daily nutrient intake of lambs, the statistical analysis was performed using the mixed linear model. The treatment, age, and their interaction were treated as fixed factors and the lamb was considered as a random effect. When significant differences were found among different treatment, a multiple comparison of body weight at 42 days of age was conducted based on one-way ANOVA. Rumen organ index and qRT-PCR results in vivo experiment were tested using one-way ANOVA followed by post hoc Tukey tests in SPSS software. Rumen papillae morphology and rumen wall thickness were measured using the mixed effects models (MIXED) procedure of SPSS, with diet as the fixed effect and lamb as the random effect. The data of microbial incubation was analysis by independent sample t-test, while the results of temporal variation was tested using one-way ANOVA followed by post hoc Tukey tests. For data of cell culture experiment, an independent sample t-test in SPSS software packages was used to assess statistical significance. A value of P < 0.05 was considered statistically significant. PCA (“vegan” package) and correlation analysis (“ggpubr” package) were conducted using related packages in R (version 3.6.2). Using G*Power (3.1.9.6), we performed a post hoc analysis to determine the effect size for animal experiment and our samples could detect with adequate power (80%) based on an alpha = 0.05, and F-test using one-way ANOVA.