Effect of different forage-to-concentrate ratios on ruminal bacterial structure and real-time methane production in sheep

Emission from ruminants has become one of the largest sources of anthropogenic methane emission in China. The structure of the rumen flora has a significant effect on methane production. To establish a more accurate prediction model for methane production, the rumen flora should be one of the most important parameters. The objective of the present study was to investigate the relationship among changes in rumen flora, nutrient levels, and methane production in sheep fed with the diets of different forage-to-concentration ratios, as well as to screen for significantly different dominant genera. Nine rumen-cannulated hybrid sheep were separated into three groups and fed three diets with forage-to-concentration ratios of 50:50, 70:30, and 90:10. Three proportions of the diets were fed according to a 3 × 3 incomplete Latin square, design during three periods of 15d each. The ruminal fluid was collected for real-time polymerase chain reaction (real-time PCR), high-throughput sequencing and in vitro rumen fermentation in a new real-time fermentation system wit. Twenty-two genera were screened, the abundance of which varied linearly with forage-to-concentration ratios and methane production. In addition, during the 12-hour in vitro fermentation, the appearance of peak concentration was delayed by 26-27min with the different structure of rumen bacteria. The fiber-degrading bacteria were positively correlated with this phenomenon, but starch-degrading and protein-degrading bacteria were negative correlated. These results would facilitate macro-control of rumen microorganisms and better management of diets for improved nutrition in ruminants. In addition, our findings would help in screening bacterial genera that are highly correlated with methane production.


Introduction
Of the total methane emission in China, the emission from ruminants was estimated to be approximately 17%, turning them into one of the largest anthropogenic sources of methane emission [1]. The emission of methane associated with agriculture is expected to see a a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 showed high correlation with methane production were screened in order to serve as the reference for accurate prediction of methane production.

Ethics statement
All research involving animals was conducted according to Guide for the Care and Use of Laboratory Animals which was approved by the ethics committee of Jilin Agricultural University, P. R. China. The ethics committee of Jilin Agricultural University, P. R. China approved this study, and the approved permit number for this study is "JLAC20171104".

Animals and diets
A total of nine rumen-cannulated (cannulated at one year of age) hybrid sheep (Chinese merino fine wool sheep × Dorper sheep) were selected, which was 2 years old and whose average weight was 87.83 ± 8.11kg. Randomly assigned to three groups, these sheep were separately fed at random. Jilin Agricultural University, Changchun, China prepared Guide for the Care and Use of Laboratory Animals which provided guidance for all animal-related procedures.
Chosen as the forage, Leymus chinensis was mixed with the concentrate in three proportions of F:C including 50:50 (L), 70:30 (M) and 90:10 (H). The composition and nutrition levels of the three diets based on the NRC [27] are shown in Table 1. The three diets were fed according to the 3 × 3 incomplete Latin square design over 45d in three periods of 15d each, including 14d of pre-feeding and the 15 th day for sampling. Three distinct flora structures were established under different treatments.

Sampling and DNA extraction
Ruminal fluid (400mL) was collected by pump and pre-warmed thermos before feeding (07:00) and saturated with CO 2 . Filtrated through a double-layered gauze, the collected fluid was used to measure the pH value to confirm the health of rumen. The fluid with the pH value between 5.5 and 7.5 from three sheep of one group was mixed. All the samples with 10mL were respectively stored in sterile centrifuge tubes (without any treatment) with 2mL at -80˚C for DNA extraction. Another 300mL of ruminal fluid from each group was warmed to 39˚C for in vitro rumen fermentation right after sampling.
Microbial genomic DNA was extracted from all ruminal fluid samples with 220mg using the methods of Murray and Thompson [28] and Zhou et al. [29]. Agarose gel electrophoresis was applied to confirm the successful extraction of DNA [30]. The qualified DNA continued to be tested for real-time polymerase chain reaction (real-time PCR) and high-throughput sequencing. A total of 9 samples for the three diets were collected and each sample was tested twice in order to have six replicates for each diet.

Real-time PCR for total bacteria, methanogens, protozoa and anaerobic fungi
Real-time PCR were tested on Applied Biosystems StepOne Real-time PCR System based on the methods of Denman and McSweeney [31]. The designed primers were shown in the Table 2. Below is the reaction system (25μL): SYBR Premix Ex Taq (RR420A, Takara) 12.5μL, forward primer 0.5μL, reverse primer 0.5μL, DNA template 2.0μL, sterile distilled water 9.5μL. Below is the reaction conditions: 95˚C 2min; 95˚C 5s, 60˚C 30s, 40 cycles; 95˚C 15s; 60˚C 1min; 95˚C 15s. The relative abundance of rumen microorganisms was expressed as a percentage or a thousandth of 16S rDNA relative to the total rumen bacteria. The formula is as follows: Target = 2 -(Ct target −Ct total bacteria) × 100; The Ct value is defined as the number of amplification cycles that are passed when the fluorescence signal of the amplification product reaches a set threshold during real-time PCR amplification.

PCR amplification of 16S rDNA, amplicon sequence and processing of sequence data
Based on previous comparisons [34][35][36], 16S rDNA had V 4 hyper variable regions which performed PCR amplification for microbial genomic DNA extracted from ruminal fluid samples and were adopted in the rest of the study. PCR primers which flanked bacterial 16S rDNA's V 4 hypervariable region were designed. The forward primer with a barcode was 338F 5'-ACTCCTACGGGAGGCAGCAG-3' while the reverse primer referred to 806R 5'-GGACTA CHVGGGTWTCTAAT-3' based on the approach of Fan et al. [37]. Below is the PCR reaction system (TransGen AP221-02, 20μL): 5×FastPfu Buffer, 2.5 mM dNTPs, Forward Primer (5μM), Reverse Primer (5μM), FastPfu Polymerase, BSA and Template DNA and ddH 2 O with 4μL, 2μL, 0.8μL, 0.8μL, 0.4μL, 0.2μL and 10ng respectively were added up to 20μL totally. Below are PCR conditions: One pre-denaturation cycle, 27 denaturation cycles, annealing, elongation and one post-elongation cycle at 95˚C, 95˚C, 55˚C, 72˚C and 72˚C for 3min, 30s, 30s, 45s and 10min respectively. Separated on 1% garose gels, products of PCR amplicon were  [38][39][40][41]. Sequences with lower mean phred score (20bp), equivocal bases and primer mismatching or sequence lengths below 50bp were removed. The assembly of only the sequences which had an overlap above 10bp and no mismatch was completed in accordance with their overlap sequence. Reads which were unable to be assembled were abandoned. Barcoded and sequencing primers were removed from the sequence which was assembled.

Taxonomic classification and statistical analysis
A web-based program called Usearch (version 7.0, http://drive5.com/uparse/) was applied to carry out taxon-dependent analysis. 16S rRNA gene sequences were used for phylogenetically consistent bacterial taxonomy according to the method of a Bayesian classifier, Ribosomal Database Project Classifier [41]. The Silva Database (Silva_128_16s, http://www.arb-silva.de) was compared to calculate the operational taxonomic units (OTUs) for all samples to show the abundance of bacterial species with 97% of identity cutoff, whereas the species for which the sum of OTUs of all the samples was more than 20 reads were retained. The richness of OTUs for each sample was produced at the level of genus. The length of all the valid bacterial sequences with primers was 440bp on the average. The calculation of abundance at the level of genus was transformed according to log 2 and normalized as the method of Niu et al. [42].
Inter-group OTUs were compared by the generation of a Venn diagram. The bacterial community indices adopted contained Chao and Shannon's coverage. The diversity of bacteria was presented by the quantity of OTUs.

In vitro rumen fermentation
The substrate was made with the feed of group M in the feeding experiment by drying and grinding through a 0.45mm sieve. Collected from different dietary treatments during the feeding experiment, the ruminal fluid was filtrated by four-layer cheesecloth and mixed with preheated artificial saliva [43] at a ratio of 2:1 (buffer: ruminal fluid, v:v). The ruminal fluid (150mL) which was buffered was dispensed into pre-warmed 200-mL incubation flasks. Two grams of each substrate was blended with the buffered ruminal fluid in each incubation flask which was incubated at the temperature of 39˚C for 12h in water. The production ratio of methane was measured by real-time in vitro fermentation system (produced by Jilin Academy of Agricultural Sciences, code Qtfxy-6), which was tested for the effluent gas discharged from each incubation flask. The nitrogen (purity 99.99%) was passed into the incubation flask from the bottom at the speed of 200mL/min. Methane was carried by nitrogen into an AGM10 sensor (Sensors Europe GmbH, Erkrath, FRG) and the concentration of methane was measured and recorded every 6min [26]. The fermentation was terminated by placing flasks on ice. After opening the incubation flask, pH was measured using a PHS-3C pH meter (Shanghai INESA Scientific Instrument Co., Ltd., China), and 2mL of incubation medium was collected for NH 3 -N [44]. Another 1 mL of incubation medium was analyzed for volatile fatty acids (VFAs), including acetic acid (AA), propionic acid (PA), and butyric acid (BA) using gas chromatography (Agilent Technologies 7890A GC System, USA) and the method of Castro-Montoya et al. [45]. The left fluid was dried in a forced-air oven at 60˚C for 72h and placed in sealed containers in order to analyze the in vitro dry matter digestibility (IVDMD) [46].

Experimental feeds and chemical analyses
Collected in plastic bags, the samples for diets were reserved at -20˚C. After the feeding experiment, the samples were warmed at 65˚C to a fixed weight. Thereafter, a 0.45mm sieve and a high-speed universal pulverizer were used to grind them for analysis. The filter bag technique of ANKOM A200 (AOAC 973.18) was adopted to analyze neutral detergent fiber (NDF), acid detergent lignin (ADL) as well as acid detergent fiber (ADF). A Kjeltec 8400 analyzer unit (Foss, Sweden) was applied to measure the content of crude protein (CP) on the basis of the Kjeldahl method (AOAC 984.13). In addition, a Soxhlet apparatus was used to measure the content of ether extract (EE) based on Soxhlet extraction method (AOAC 920.85). Methods of Horwitz et al. [46] were the foundation of all chemical analyses.

Data analysis
Sequences of good quality were deeply studied through its uploading to QIIME [39]. A comparison was made between valid bacterial sequences and sequences present in the Silva Database which classified the abundance calculation of each taxon with the optimal choice of classification [36]. QIIME filed the sequence length. Mothur was used for the generation of abundance and diversity indexes. After the implementation of a pseudo-single relevancy algorithm, there was 97% of OTUs identity cutoff [47][48]. For all the parameters, data was analyzed by the R-Studio software (version 7.2). Methane production was up to the approach of Sun et al. [26]. A one-way analysis of variance (ANOVA) was carried out late in each bioassay to compare selected taxonomic groups (abundant phyla or genera), bacterial community indices observed OTUs or methane production indices. Duncan's test was adopted to perform the mean comparison at the significance level of P < 0.05. Redundancy analysis (RDA) was conducted to assess the association between the nutrients in the feed and the bacterial abundance in the rumen. The relationship among bacterial abundance, methane production, peak concentration (C max ) and the time to peak concentration (T max ) was assessed by means of Pearson's correlations. All the data was presented as means ± S.E. (standard error).

Relative quantification of total bacteria, methanogens, protozoa and anaerobic fungi
Firstly, the results of real-time PCR (Table 3) showed that the numbers of methanogens and protozoa were increased with the decreasing F:C but not significantly. Conversely, the numbers of total bacteria and anaerobic fungi were decreased with the decreasing F:C. And the difference of total bacteria in three groups was significant (P = 0.017), while the difference of anaerobic fungi was not. These results exhibited that the change of F:C has extremely effect only on the number of total bacteria but not on the other kinds of microorganism.

Analysis of DNA sequence data
After quality was controlled preliminarily, 517,492 paired-end 440-bp reads were obtained in total. Each sample got 28,750 sequences averagely. Reads had an overall length of 2.28 gigabases (GB), and each sample had a mean read length of 0.13GB with 191,537, 171,125, and 154,794 raw reads in L, M and H groups respectively (Table 4). Based on 97% species similarity, 132,987, 104,640 and 92,194 OTUs were separately obtained from the samples in L, M and H groups (Table 4). Among all the samples, 708 OTUs were identified, of which 542 existing in all the groups were known as key OTUs (Fig 1A). Key OTUs took up about 76.6% of all OTUs, whereas 6, 5 and 11 OTUs were individually identified in groups L, M and H respectively. Good's coverage was 99.4%, 99.3% as well as 99.2% for L, M and H groups separately, indicating the capture of dominant phylotypes by this study. The three groups were similar in diversity (Fig 2A). The richness (P<0.01) of the rumen microbiota was related to F:C ( Fig 2B).  (Table 5). With the increase of the ratio, the proportion of Proteobacteria and Spirochaetes (Table 5) decreased (P<0.01) and the proportion of Synergistetes (P < 0.01) ( Table 5) increased. At the level of genus, the identification of 150 genera in all the samples was conducted despite F:C (Fig 3). L, M and H groups had 141, 146 and 141 genera respectively and shared 130 genera, whereas Coriobacteriaceae and Gemella were special for group L (Fig 1B). 15

Nutrition index in rumen and its correlation with the rumen microbiota
In terms of the RDA, our dataset changed, which was principally interpreted by the increasing F:C (Fig 4). It suggested that 100% change in bacteria was explained by all the nutrition indices whose order of contribution was CP > ADF > NDF > Starch > EE > ADL ( Table 7). The two sorting axes accounted for 95.48% of the changes based on this model with the first sorting axis explaining a change of 66.37% and 29.11% for the second sorting axis. The rumen microbiota in group L was concentrated in the regions with high CP, starch content and low NDF and ADF contents, whereas the rumen microbiota in group H was concentrated in the regions with high NDF and ADF contents and low CP and starch contents. The rumen microbiota in group M was concentrated in the regions with intermediate nutrient levels. According to the RDA analysis, the relevance of CP accounted for 0.72 of the microbiota (P<0.01) as the main nutrient factor affecting the structure of microbiota. Insignificant, the relevance of EE and ADL was the lowest (R 2 = 0.41, P>0.05; R 2 = 0.36, P>0.05) and they were not significant. Under the different levels of F:C, the different kinds of bacteria bacterial community were established, which could reflect the state of the microflora in the rumen fluid at the beginning of in vitro fermentation.

In vitro rumen fermentation characteristics, real-time methane production and its correlation with the rumen microbiota
After 12 h fermentation, the concentrations of pH, AA and A/P were decreased greatly with the decreasing F:C. Simultaneously, the concentrations of PA, BA, NH 3 -N, and IVDMD were increased with the decreasing F:C. The greatly growth of IVDMD had led to the massive production of VFAs (Table 8). The C max (P<0.01, Table 9, Fig 5) and total production (P<0.05, Table 9) of methane decreased with the increase in F:C, whereas T max (P<0.01, Table 9, Fig 5) of methane increased with the increase of F:C. At the level of phylum, Bacteroidetes, Proteobacteria as well as Spirochaetae showed a positive correlation with the C max and total production, and a negative correlation with T max (Table 10). Firmicutes and Saccharibacteria were positively correlated with T max , but negatively correlated with C max and total production ( Succinivibrio, Syntrophococcus and Olsenella was positively correlated with C max and total production in bacterial abundance, but negatively correlated with T max (Table 11). Ruminococcus, Rikenella, Succiniclasticum, Eubacterium, Papillibacter, Pseudobutyrivibrio, Butyrivibrio, Candidatus, Anaerotruncus and Ruminiclostridium was positively correlated with T max in bacterial abundance, but negatively correlated with C max and total production (Table 11).

Discussion
The rumen bacterial community structure was changed significantly with the different levels of F:C, but the numbers of methanogens, protozoa and anaerobic fungi were not changed obviously. The in vitro fermentation has been changed significantly with the change of F:C. And a close relationship was found between the real-time methane production and bacterial community structure. It seems to provide a new understanding of rumen fermentation model and methane production adjustment.  Firstly, the results of real-time PCR showed that the number of methanogens, protozoa and anaerobic fungi, under the change of F:C, was changed significantly except bacterial. Archaea was considered the producer of methane. But with the change of F:C in this study, the quantity of archaea was stable. It showed no significant linear relationship between the structure of archaea and methane production, which is similar to the research of Lengowski et al. [25]. The relationship between archaea and methane production has been discussed in many studies. Moreover, the main raw materials for methane synthesis, such as hydrogen, carbon dioxide and VFAs, were produced by bacteria [49]. Methane synthesis was a passive behavior of archaea to maintain rumen pressure and pH balance in the case of too high ratios of bacterial synthesis [50][51]. Therefore, methane production was more probably related to the concentration of synthetic raw materials in the rumen and the bacteria producing these materials.
The second goal of this experiment was to explore changes in the genus level of the rumen flora with different F:C. With the increase of the ratio, the proportion of different genera showed significant differences, revealing the effectiveness of experimental gradient design. In Structure of rumen bacteria and real-time methane production in sheep this study, the proportion of Prevotella showed a linearly increasing trend with the increase of protein levels in diets, which was consistent with the results of Xu and Gordon [52]. As a genus, Prevotella has many functions, mainly including promoting protein degradation and assisting other strains in enhancing the utilization of fiber materials in ruminants [53]. Ruminococcus, a cellulolytic bacterium [54], increased with the increasing fiber. Succinivibrio,  Structure of rumen bacteria and real-time methane production in sheep Ruminobacter, amylophilus and Selenomonas were starch-degrading bacteria that could produce acetic acid and succinic acid during starch degradation [55]. Succinic acid was eventually transformed to PA [56] to provide energy for microbial protein synthesis in the rumen. With the decrease of starch content in the diets, the proportion of these three genera decreased significantly in this experiment, indicating the change of carbohydrate fermentation substrate from a non-structural carbohydrate to a structural carbohydrate. Butyrivibrio and Pseudobutyrivibrio were carbohydrate-degrading bacteria producing butyric acid [57]. In this experiment, the proportion of Butyrivibrio and Pseudobutyrivibrio decreased linearly with the increase of starch, but increased linearly with the increase of NDF and ADF in the diets, which showed that Butyrivibrio and Pseudobutyrivibrio were more likely to produce energy by using structural carbohydrates. The proportion of Eubacterium with the function of degrading structural carbohydrates was similar to that of Butyrivibrio [58]. In this study, the third aim was to gain a preliminary understanding of the relationship between nutrition levels and the diversity and richness of rumen microbiota in sheep under various F:C. In this experiment, CP was the most important nutrient factor contributing to the change in bacterial diversity. Bodine and Purvis [59] found that the effect of supplementation of non-structural carbohydrate is largely determined by the level of protein in the diet. Adding protein to the diet can improve the balance of energy and nitrogen and increase digestibility. CP could provide nitrogen resource for the self-replication and enzyme synthesis of bacteria [60]. ADF, NDF and starch were important nutrient factors providing carbon resource for self-replication and energy. However, starch showed a negative correlation with bacterial diversity compared to ADF and NDF because of its easier decomposition as a non-structural carbohydrate. According to the studies of Kononoff and Heinrichs [61] and Drackley et al. [62], the rumen fermentation was mainly in the AA-mode when NDF and ADF contents in the diet were high and mainly in the PA-mode when starch content in the diet was high. The Chao index of OTUs increased with the increase in F:C, showing more strains of bacteria were required by the degradation of NDF and ADF to cooperate than those required by the degradation of CP and starch. These were the changes of microbiota in the rumen. EE and ADL in the diets had no significant effects on the changes in the rumen microbiota. Jenkins [63] found that only about 8% of fat in the rumen was degraded. There might be two reasons: The designed levels of EE and ADL content in the diets were too close to result in the similarity of the microbial community or these nutrients were not main energy resources for bacterial Structure of rumen bacteria and real-time methane production in sheep activity in the rumen so that bacteria were not sensitive to the low levels of EE [64]. Based on the above results, three kinds of rumen bacterial community were proved to be successfully established.
In this study, the final goal was to preliminarily understand the relationship between methane production and the rich and diversity of rumen microbiota in sheep under various F:C. In the anaerobic environment of the rumen, a variety of organic compounds could eventually be transformed to methane through decomposed by a number of microorganisms [5]. Leng and Nolan [14] showed that 80% of the nitrogen available to ruminal bacteria came from ammonia and 20% from amino acids or oligopeptides. With the increase of F:C, C max of methane was delayed. For the diet with higher CP and starch contents, methane production could reach C max more quickly, showing a significant correlation with the rumen microbiota. With lower CP content in diet, bacteria required more time for protein decomposition to provide materials for their reproduction and methane synthesis, which indicated that methane synthesis Table 9. C max , T max and total production of methane in vitro and Pearson's correlations with different forage-to-concentrate ratios. Structure of rumen bacteria and real-time methane production in sheep needed to go through a "start-up" phase before the normal fermentation mode. Previous studies only found that C max of methane occurred at about 2 h after food intake [65][66]. The missing of the delay phenomenon could be attributed to the insufficient frequency of detection. On the other hand, both C max and total production of methane with lower CP and starch contents in the diet during fermentation were less than those in the diets with lower NDF and ADF contents. It was probably because related bacteria like Prevotella and Butyrivibrio could decompose nitrogen compounds to provide sufficient raw materials for the reproduction and synthesis of methanogenic archaea [67]. As indicated by the correlation analysis, fiber-degrading bacteria were positively correlated with T max of methane, but negatively correlated with C max and total production of methane. Compared with fiber-degrading bacteria, starchdegrading and protein-degrading bacteria showed an opposite correlation. More readily available nitrogen and degradable carbohydrates could be preferentially used by microorganisms [68], providing more effective support for methane synthesis. Furthermore, C max and T max of methane could be effective parameters for predicting the type of rumen fermentation, which however remained to be confirmed by further research. Based on nutritional parameters, a new model of methane prediction with a wider range of applications is being developed in accordance with the results of the present study. The genera of bacterial, as the parameters for the prediction models, had been narrowed down. There were significant correlations between specific bacterial at the starting of fermentation and realtime methane production in vitro. However, the dynamic changes of bacterial at the moment such as T max during the fermentation need to be explored in the following study. Additionally, the fermentation in vivo was more complex. For instance, nitrogen cycling in ruminants might provide bacteria with initial nitrogenous material [69]. Thus, further studies are required to confirm the occurrence of this delay phenomenon in vivo and illustrate the process.

Conclusions
With the change in F:C, bacterial community structure and methane production in the rumen showed significant changes. Crude protein was the most important nutrient factor that contributed to the change in bacterial diversity. Among the 150 genera identified in the rumen, the abundance of 22 varied linearly with F:C. These genera would be further screened to serve as effective parameters for the methane prediction model. In addition, during the 12 h in vitro fermentation, as F:C increased, the C max and total production of methane decreased significantly, and T max was delayed by 26-27 min. The fiber-degrading bacteria were positively correlated with this phenomenon, but starch-degrading and protein-degrading bacteria were negatively correlated with it.