The experimental design and procedures were approved by the Animal Care and Use Committee of Nanjing Agricultural University. The experiment was carried out using Suzhou Hu sheep at a breeding farm in the Jiangsu province, China. Eight pairs of healthy, 10-day-old lamb twins (Hu sheep, a native Chinese sheep breed) were selected. One kid from each pair remained with the mother and received milk ad libitum without receiving starter feed (M group, n = 8), while the other kid was separated from the mother and received starter feed (M+S group, n = 8) from 4:00 a.m. to 7:00 p.m. every day in a separate pen. During this period, lambs in the M+S group were fed milk for 1 h at each fixed time point (6:30 a.m., 10:30 a.m., and 3:30 p.m.). When the dry matter intake (DMI) of the lambs' starter reached 200 g/animal⁻¹d⁻¹, the amount of starter did not rise any further. The eight lambs in the M+S group maintained a 200 g/animal⁻¹d⁻¹ starter intake for an average of 14 days before sacrifice. All lambs received oat hay (10.05% crude protein, 28.71% crude fiber) and water ad libitum. The ewes were fed three times per day according to the farm's feeding management schedule. None of the lambs in the M and M+S groups had access to the ewes' feed. The DMI of the starter in the M+S group was recorded every day, and the body weights of each lamb was measured weekly (before morning feeding). The experimental starter diets were designed according to the nutrient requirements of Hu sheep lambs (NY/T816-2004; Ministry of Agriculture of China, 2004). The nutrient composition of the starter diet is presented in Table 1 (Liu et al., 2017).

Lambs were stunned by captive bolt and exsanguination at 56 days of age. A representative sample of colon digesta was collected from the proximal colon immediately after slaughter to determine the pH value. Colon digesta from each lamb were homogenized and mixed thoroughly with twice the amount of distilled water. The mixtures were then immediately centrifuged at 12,000 × g, and the supernatants were stored at −20°C until analysis for volatile fatty acids (VFA) and lactic acid. Within 5 min, a segment of the colon tissue was harvested and immediately washed three times in ice-cold, phosphate-buffered saline. A portion of the tissue sample was cut into smaller pieces (~0.5 × 0.5 cm) and immediately frozen in liquid nitrogen for RNA extraction. Another portion of the tissue sample was cut to ~1 × 1 cm and scraped from the underlying tissue using a germ-free glass slide, immediately transferred into liquid nitrogen, and then stored at −80°C until microbial DNA extraction. A final portion was immediately fixed in 4% paraformaldehyde (Sigma, St. Louis, MO, USA) and 2.5% glutaraldehyde for histomorphometric microscopy analysis.

A portable pH meter (HI 9024C; HANNA Instruments, Woonsocket, RI, USA) was used to determine the pH of colonic digesta. Capillary column gas chromatography (GC-14B, Shimadzu, Japan; Capillary Column: 30 m × 0.32 × 0.25 mm film thickness; Column temperature = 110°C, injector temperature = 180°C, detector temperature = 180°C) was used to measure VFA concentration (Qin, 1982). Lactate concentration was detected using a method described by Barker and Summerson (1941).

One gram of colonic mucosal tissue was used for DNA extraction. The DNA was extracted by a PowerSoil DNA Isolation Kit (MOBIO Laboratories, Carlsbad, CA, USA, catalog 12888-100). The solution was precipitated with ethanol, and the pellets were suspended in a 50-μL Tris-EDTA buffer. DNA was quantified using PicoGreen dsDNA reagent kit (Invitrogen Ltd., Paisley, UK) with a Molecular Devices SpectraMax Microplate Reader (Molecular Devices, Sunnyvale, CA, USA).

The V4 regions of bacterial 16S rRNA genes were amplified by PCR (Initial denaturation at 95°C for 2 min, 25 cycles of denaturation at 95°C for 1 min, annealing at 55°C for 1 min, elongation at 72°C for 1 min, and final extension at 72°C for 5 min) using primers 515F (5′-barcode-GTGCCAGCMGCCGCGGTAA-3′) and 806R (5′-barcode-GGACTACHVGGGTWTCTAAT-3′). Amplicons were purified using the Qiagen QIAquick PCR purification kit (Qiagen, Duesseldorf, Germany) according to the manufacturer's instructions and quantified using PicoGreen dsDNA reagent kit (Invitrogen, Paisley, UK). Purified amplicons were pooled in equimolar, and the amplicon size was determined by Aglient 2200 Bioanalyzer (Agilent Technologies, CA, USA). The pooled product was pair-end sequenced (2 × 300) on an Illumina MiSeq platform according to standard protocols.

For data analyses, raw Illumina fastq files were demultiplexed, quality filtered, and analyzed using Quantitative Insights into Microbial Ecology (QIIME, v.1.8.0), as described by Caporaso et al. (2010b) and with the following criteria, as described by Mao et al. (2015): Operational taxonomic units (OTU) were clustered with a 97% similarity cut-off using UPARSE (Edgar, 2013), and chimeric sequences were identified and removed using UCHIME (Edgar et al., 2011). The most abundant sequences within each OTU (representative sequences) were aligned to the Greengenes database using PyNAST (Caporaso et al., 2010a) with the default parameters set by QIIME. Taxonomy was assigned to representative sequences using QIIME (Wang et al., 2007) with a confidence value of 0.8 against the Greengenes 16S rRNA gene dataset (v.13.8) (DeSantis et al., 2006). Rarefaction curves and alpha and beta diversity calculations were also performed using QIIME. Principal coordinate analysis (PCA) was used to compare groups of samples based on unweighted UniFrac distance metrics (Lozupone and Knight, 2005), and an unweighted distance-based analysis of molecular variance (AMOVA) was conducted to assess significant differences among samples using the MOTHUR v.1.3.9 program (Schloss et al., 2009).

The colonic tissues were embedded in paraffin, sectioned into 6 μm, and stained with hematoxylin and eosin (H&E). During histomorphometric analyses, the microscopist was blinded to treatment conditions. For each lamb, two slides were prepared and two images were captured per slide, resulting in a total of 32 replicates per measurement per group. Predefined criteria described by Steele et al. (2011) were used to assess colonic injury using Image Pro Plus software (Media Cybernetics, Bethesda, MD, USA). The criteria were as follows: a score of one indicated no lesions or minor lesions; a score of five indicated minor lesions with mucosa sloughing; and a score of nine indicated severe, deep lesions with large amounts of mucosa sloughing.

The tissues were fixed with 2.5% glutaraldehyde for at least 24 h, postfixed in 1% osmium, and embedded in Epon araldite. A glass knife was used to cut semithin sections (0.25–0.5 μm) and ultrathin sections (70–90 nm). To stain semithin sections, 1% toluidine blue and 1% sodium borate were used, while uranyl acetate and lead citrate were used to stain ultrathin sections. A transmission electron microscope (H-7650; Hitachi Technologies, Tokyo, Japan) was used to examine and determine ultrastructures of the colonic tissue.

Total RNA was extracted from the colonic tissue using TRIzol (Takara Bio, Otsu, Japan), as described by Chomczynski and Sacchi (1987). RNA concentrations were then quantified using a NanoDrop spectrophotometer (ND-1000UV-Vis; Thermo Fisher Scientific, Waltham, MA, USA). The absorption ratio (260/280 nm) of all of the samples was between 1.8 and 2.0, indicating high RNA purity. Aliquots of RNA samples were subjected to electrophoresis through a 1.4% agarose–formaldehyde gel to verify integrity. The concentration of RNA was adjusted to 1 μg/μL based on optical density and stored at −80°C. Total RNA (1 μg) was reverse-transcribed using a PrimeScript RT Reagent Kit with gDNA Eraser (Takara Bio, Otsu, Japan) according to the manufacturer's instructions.

The primers of cytokine (Liu et al., 2013), TLR (Charavaryamath et al., 2011), and glyceraldehyde-3-phosphate dehydrogenase (GAPDH; Wang et al., 2009) genes used in the present study were described in previous studies. All of the primer sequences are listed in Table S1. The primers were synthesized by Invitrogen Life Technologies (Shanghai, China). The ABI 7300 Real-time PCR System (Applied Biosystems, Foster City, CA, USA) with SYBR green dye fluorescence detection was used to perform qRT-PCR of the target genes and GAPDH. Amplification conditions were as follows: 95°C for 30 s followed by 40 cycles of 5 s at 95°C and 31 s at 57.5°C (for GAPDH) or 62°C (for the cytokines and TLR). Each sample contained 1–10 ng cDNA in 2 × SYBR Green PCR Master Mix (Takara Bio, Otsu, Japan) and 200 nmol/L of each primer in a final volume of 20 μL. All measurements were performed in triplicate. The negative controls were a reverse-transcription-negative blank of each sample and a no-template blank. The GAPDH (a housekeeping gene) mRNA level was used to normalize the relative amount of each studied mRNA, and the 2^−ΔΔCT method was used to analyze the data (Livak and Schmittgen, 2001).

Statistical analyses were performed using the SPSS software package (SPSS v.16, SPSS Inc.). The normality of the distribution of variables was assessed with the Shapiro-Wilk test. The data found to have a normal distribution were analyzed by the Independent Samples t-test procedure, according to the following model: [Y = u+ C + e], where u is the mean, C is the effect of diet, and e is the residual error. The Kruskal-Wallis test was used to analyze variables found to have a non-normal distribution according to the following statistical model: H =  12n(n+1)∑i = 1kRi2ni-3(n+1), where H is the Kruskal-Wallis test, n is the number of measurements, R_i is the sum of the ranks, and n_i is the number of experiments. Significance was declared at P < 0.05.

Correlation analysis was assessed by Spearman's correlation test using GraphPad Prism v.5 (GraphPad Software, San Diego, CA, USA). Significance was declared at P < 0.01.

We observed no significant differences in birth weights (3.43 ± 0.10 vs. 3.31 ± 0.07 kg, P = 0.375) and final body weights at 56 days age (14.89 ± 0.36 vs. 14.44 ± 0.34 kg, P = 0.381) between the M and M+S groups. During the feeding trial, the average total DMI of starter per lamb in the M+S group was 5.54 ± 0.16 kg.

As shown in Table 2, compared with milk-fed lambs, starter-fed lambs had a higher concentration of total VFA (P = 0.001), acetate (P = 0.018), propionate (P < 0.001), butyrate (P < 0.001), and lactate (P < 0.001), but had lower luminal pH (P = 0.002) and acetate to propionate ratio (P < 0.001) in colonic content. Starter feeding did not affect other VFA concentrations significantly (P = 0.485).

After quality control, 697,630 valid reads were obtained in all samples with an average of 43,602 sequences per sample. MOTHUR analysis showed that 7,752 OTU at sequence divergences of 0.03 were classified based on these valid sequences. The average number of OTU was 485 ± 6, with an average coverage of 99.77 ± 0.01%. The Chao1 richness, abundance-based coverage estimator (ACE), and Shannon and simpson diversity indices were 580 ± 9, 573 ± 6, 4.30 ± 0.07, and 0.04 ± 0.01, respectively. We found a total of 18 phyla in all samples. The most dominant phyla were Firmicutes (48.58%) and Bacteroidetes (36.33%), and the next dominant phyla were Proteobacteria (4.00%), Verrucomicrobia (3.91%), and Actinobacteria (1.28%). Unclassified bacteria (3.22%) together with these five phyla represented 97.32% of total reads. The proportion of the phyla Tenericutes, Planctomycetes, Lentisphaerae, Spirochaetae, Cyanobacteria, Fusobacteria, and Fibrobacteres accounted for <1.00% of total sequences. We did not detect the phyla Candidate, Elusimicrobia, Synergistetes, Deferribacteres, and Chloroflexi in all of the samples. We found a total of 218 taxa (at the genus level) in all samples. The dominant bacterial taxa were unclassified Ruminococcaceae (18.20%), Bacteroides (9.96%), unclassified S24-7 (7.84%), and unclassified Lachnospiraceae (6.93%), followed by unclassified Christensenellaceae (5.23%), unclassified Bacteroidales (3.88%), Akkermansia (3.87%), RC9_gut_group (3.67%), Alistipes (3.66%), unclassified Clostridiales (2.73%), Blautia (2.62%), Oscillibacter (2.49%), Phocaeicola (1.86%), Prevotella (1.33%), Phascolarctobacterium (1.33%), and unclassified Defluviitaleaceae (1.16%). The proportion of other taxa was below 1.00% of total sequences. As was shown in Figure S1, the top 50 bacterial taxa of different samples were presented in the heat map.

The rarefaction curves of colonic mucosal bacterial communities (Figure S2, at dissimilarity levels of 0.03) showed that all curves approached a plateau, suggesting that deeper sequencing was not responsible for an increase of OTU across all samples. We used the unweighted UniFrac metric in MOTHUR to evaluate β-diversity across the samples (Figure 1). As shown in the PCA figure, the plots of the M and M+S groups were distinctly separated (Figure 1; axis 1 + axis 2 = 38.1%). The AMOVA analysis shows that starter feeding significantly affected the colonic mucosal bacterial communities (AMOVA, Fs = 3.68, P = 0.001). The effects of starter feeding on the α-diversity of colonic mucosal bacterial communities are shown in Table 3. The results show that starter feeding supplementation increased Chao1 richness (P = 0.034), and no significant difference in OTU numbers (P = 0.203), ACE (P = 0.181), and Shannon (P = 0.373) or simpson (P = 0.331) indices.

At the phylum level (Table 4), we found that compared with the M group, the M+S group had a higher relative abundance of Bacteroidetes (P = 0.027) and Actinobacteria (P = 0.005), with a lower relative abundance of Firmicutes (P = 0.027), unclassified Bacteria (P = 0.046), and Cyanobacteria (P = 0.036). We observed no significant difference in the proportions of the phyla Proteobacteria (P = 0.600), Verrucomicrobia (P = 0.462), Tenericutes (P = 0.172), Planctomycetes (P = 0.141), Lentisphaerae (P = 0.294), Spirochaetae (P = 0.916), and Fusobacteria (P = 0.916) between the M and M+S groups.

At the genus level (Table 5 and Figure S3), starter feeding caused an increase in the relative abundance of the dominant taxa unclassified S24-7 (P = 0.002), Oscillibacter (P = 0.046), Prevotella (P = 0.009), Parabacteroides (P = 0.002), Bifidobacterium (P = 0.002), Ruminobacter (P = 0.002), and Succinivibrio (P = 0.006). Starter feeding also caused a decrease in the relative abundance of unclassified Ruminococcaceae (P = 0.006), RC9_gut_group (P = 0.027), Blautia (P = 0.002), Phocaeicola (P = 0.036), Phascolarctobacterium (P = 0.009), unclassified BS11_gut_group (P = 0.027), unclassified family_XIII (P = 0.016), Campylobacter (P = 0.016), unclassified Firmicutes (P = 0.002), Pseudobutyrivibrio (P = 0.009), Barnesiella (P = 0.046), Lactobacillus (P = 0.001), unclassified Gastranaerophilales (P = 0.036), Butyrivibrio (P = 0.006), dgA-11_gut_group (P = 0.001), and Dorea (P = 0.012).

For the lambs from the M group, we observed sloughing of the mucosal surface in the representative cross-sections of the colonic tissue (Figure 2A). For the lambs from the M+S group, we observed that the intercryptal surface was covered by an irregular layer of mucus (Figure 2B). The colonic injury scores of lambs in the M+S group were significantly lower than that of lambs in the M group (2.04 ± 0.16 vs. 4.63 ± 0.25, P < 0.001). The lambs from the M group had sparse and irregular microvilli in the ultrastructure of the colon tissue (Figure 2C), while the lambs from the M+S group had clear and organized microvillus clusters (Figure 2D). For the lambs in the M group, intercellular tight junction erosion, and mitochondrial swelling appeared in the cells (Figure 2E); for the lambs in the M+S group, we observed normal and clear cell organelles and tight junction bands in the cells (Figure 2F).

As shown in Figure 3, starter feeding decreased the mRNA expression of cytokines TNF-α (P < 0.001) and IFN-γ (P < 0.001) in the colonic mucosa. We found no significant differences in mRNA expression of IL-1β (P = 0.759), IL-6 (P = 0.472), IL-10 (P = 0.068), and IL-12 (P = 0.986) between the M and M+S groups. Meanwhile, starter feeding also decreased colonic mucosal TLR4 (P = 0.017) mRNA expression, and we observed no significant differences in mRNA expression of TLR2 (P = 0.251), TLR3 (P = 0.938), and TLR5 (P = 0.223).

Figure 4 depicts the relationships between the relative abundance of colonic mucosal bacteria and TLR and cytokine expression. The relative mRNA expression of IL-6 was negatively associated with the relative abundance of the genus Blautia (P = 0.0076), while IL-10 mRNA expression was negatively linked with the relative proportion of the genus Phocaeicola (P = 0.0079). IL-12 mRNA expression was positively correlated with the relative abundance of the genus Alistipes (P = 0.0057), whereas the mRNA expression level of TNF-α was positively associated with the relative abundance of the taxa unclassified Ruminococcaceae (P = 0.0022), dgA-11_gut_group (P = 0.0003), Blautia (P < 0.0001), Lactobacillus (P < 0.0001), Dorea (P = 0.0076), unclassified Firmicutes (P = 0.0012), and Butyrivibrio (P = 0.0017), and negatively correlated with the abundance of the taxa unclassified S24-7 (P = 0.0019), Prevotella (P = 0.0056), Parabacteroides (P = 0.0014), Ruminobacter (P = 0.0027), and Bifidobacterium (P = 0.0077). IFN-γ mRNA expression was positively correlated with the relative proportion of the taxa dgA-11_gut_group (P = 0.0049), Blautia (P = 0.0005), Lactobacillus (P = 0.0001), Phocaeicola (P = 0.0099), Pseudobutyrivibrio (P = 0.0054), Desulfovibrio (P = 0.0095), unclassified Firmicutes (P < 0.0001), Butyrivibrio (P = 0.0012), and Phascolarctobacterium (P = 0.0034), and negatively associated with the abundance of the taxa unclassified S24-7 (P = 0.0005), Parabacteroides (P = 0.0008), Ruminobacter (P = 0.0059), and Bifidobacterium (P = 0.0074). The mRNA expression of TLR4 was positively associated with the taxa unclassified Ruminococcaceae (P = 0.0075), Pseudobutyrivibrio (P = 0.0037), and unclassified Firmicutes (P = 0.0093), and negatively linked with the taxa unclassified S24-7 (P = 0.0017), Parabacteroides (P = 0.0083), and Ruminobacter (P = 0.0080).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.