Dominant remodelling of cattle rumen microbiome by Schedonorus arundinaceus (tall fescue) KY-31 carrying a fungal endophyte

Tall fescue KY-31 is an important primary forage for beef cattle. It carries a fungal endophyte that produces ergovaline, the main cause of tall fescue toxicosis that leads to major revenue loss for livestock producers. The MaxQ, an engineered cultivar, hosts an ergovaline nonproducing strain of the fungus and consequently is nontoxic. However, it is less attractive economically. It is not known how rumen microbiome processes these two forages towards nutrient generation and ergovaline transformation. We have analysed the rumen microbiome compositions of cattle that grazed MaxQ with an intervening KY-31 grazing period using the 16S rRNA-V4 element as an identifier and found that KY-31 remodelled the microbiome substantially, encompassing both cellulolytic and saccharolytic functions. The effect was not evident at the whole microbiome levels but was identified by analysing the sessile and planktonic fractions separately. A move from MaxQ to KY-31 lowered the Firmicutes abundance in the sessile fraction and increased it in planktonic part and caused an opposite effect for Bacteroidetes, although the total abundances of these dominant rumen organisms remained unchanged. The abundances of Fibrobacter , which degrades less degradable fibres, and certain cellulolytic Firmicutes such as Pseudobutyrivibrio and Butyrivibrio 2, dropped in the sessile fraction, and these losses were apparently compensated by increased occurrences of Eubacterium and specific Ruminococcaceae and Lachnospiraceae . A return to MaxQ restored the original Firmicutes and Bacteroidetes distributions. However, several KY-31 induced changes, such as the low abundance of Fibrobacter and Butyrivibrio two remained in place, and their substitutes maintained significant presence. The rumen microbiome was distinct from previously reported faecal microbiomes. In summary, KY-31 and MaxQ were digested in the cattle rumen with distinct consortia and the KY-31-specific features were dominant. The study also identified candidate ergovaline transforming bacteria. It highlighted the importance of analysing sessile and planktonic fractions separately.


INTRODUCTION
In the foregut or rumen of ruminants such as cattle, a complex microbial community anaerobically degrades the feed, generating volatile fatty acids as carbon and energy nutrition for the animals [1]. The process also makes livestock major emitters of methane, a potent greenhouse gas [1]. For near 100 years there has been intense research focused on rumen microbiome metabolism because OPEN ACCESS of this importance, helping not only to optimize the feed utilization efficiencies in economically important ruminants but also to develop microbial processes for bioenergy production [1,2]. In this backdrop, a lack of detailed knowledge of rumen microbiome metabolism in cattle raised on tall fescue is a major gap as it impedes the efforts to utilize this economic grass efficiently for beef and dairy production, to assess the associated impacts on methane emission and for mining this source for novel biocatalysts.
Tall fescue [Schedonorus arundinaceus (Schreb.) Dumort., nom. cons. tall fescue] is one of the primary perennial cool-season forages that feeds up to 20 % of beef cattle in the United States [3]. It is grown on more than 35 million acres in the transition zone of Southeastern US, known as the fescue belt [4]. In Argentina, Uruguay and Australia, this forage is grown on over 8.65, 1.24 and 2.72 million acres of the pastures, respectively [5,6]. The most widely used variety of the grass is KY-31 and it is popular for its resilience towards pests and poor environmental conditions [4]. The resilience of KY-31 is due in part to its symbiotic interaction with a fungal endophyte, Epichloë coenophiala [7]. The fungus secretes a variety of ergot alkaloids that are beneficial to the grass [7]. However, one of these compounds, ergovaline, is toxic to the animals causing tall fescue toxicosis syndrome [7], which results in $2 billion of annual revenue loss for the US beef and dairy industries [8]. The use of engineered varieties of tall fescue that either carry a E. coenophiala strain that does not produce ergovaline (MaxQ) or are free of the endophyte (KY32, Cajun I and Bronson) have been promoted as a solution to this problem [7]. However, these varieties are less attractive economically due to higher costs for the seeds and pasture management, and KY-31 remains widely used in the fescue belt [9,10]. The selection of tall fescue toxicosis-resistant or tolerant cattle remains an underdeveloped option [11,12]. All these efforts do not consider the possibility that the rumen microbes instigate and could mitigate the above-mentioned toxicity, although there are indications of such possibilities. In vitro studies suggest that certain rumen micro-organisms transform ergovaline to lysergic acid [13] and it is thought that lysergic acid enters the blood stream of the cattle causing tall fescue toxicosis [13]. The addition of isoflavone producing grasses to a tall fescue diet reduces the severity of tall fescue toxicosis, and it is thought that isoflavones suppress ergovaline transforming microbes [8]; the mechanism of this suppression is unknown. In search of a microbial basis for the promotion and/or suppression of KY-31 toxicity, the characteristics of the faecal microbiomes of cattle raised on KY-31 and a non-toxic tall fescue have been compared, resulting in the identification of candidate micro-organisms [14,15]. Recently, the rumen microbiome of pregnant ewes grazing tall fescue with high and moderate endophyte infection levels in parallel, were analysed [16]. However, these findings do not truly represent the events occurring in the rumen, as the faeces and rumen harbour distinct microbiomes, sharing ~30 % of the species [14,17] and in the study with the ewes [16] the rumen contents were sampled via an orogastric tube insertion, which provides preparations with underrepresentation of the fibre-attached microbial population. We hypothesized that the differences between the microbial systems that KY-31 and MaxQ enrich in cattle rumen are likely broader in nature, going beyond the transformation of ergovaline and covering the overall degradation process occurring both in the planktonic and sessile fractions. Accordingly, we have analysed the composition of the sessile and planktonic populations of the rumen microbiome of beef cattle that have alternatively grazed MaxQ and KY-31 pastures; the samples were collected through a cannula and two groups of animals with different levels of sensitivities to tall fescue toxicosis were studied. This is a rare rumen microbiome study with cattle that grazed a pasture and not fed with stockpiled hay or on a feedlot [18][19][20]. For small farmers in Virginia and elsewhere in the US fescue belt, grazing a pasture is the most prevalent way of raising beef cattle before the animals are moved to feed lot for finishing [21]. The results of our study have revealed novel type microbial successions and identified KY-31 as a dominant remodeler of sessile and planktonic microbial populations in the rumen, targeting both the cellulolytic and saccharolytic segments.

Preparation of the animals
Eight Hereford-Angus cross steers (age: 18 months; initial body weight: 1079.88±31.92 pounds [ − x ± − x σ]) belonging to one of the two tall fescue toxicosis sensitivity groups, susceptible and tolerant, were used in the study. The degree of this sensitivity was assessed by the T-Snip Test (AgBotanica, LLC, Columbia, MO) that targets a proprietary set of single nucleotide polymorphisms in the genes for Dopamine Receptor (DRD2) [11] and Kell Blood Group Complex Subunit-Related Family, Member 4 (XKR4) [12], and provides a tolerance rating (0-5). The cattle with ratings of 0-1 are considered susceptible and 4-5 are most tolerant to tall fescue toxicosis. In our study, the susceptible group (n=4) had the rating of 1 and for the tolerant group (n=4) the ratings were 2-3. Each animal was fitted with a 4-inch cannula (Bar Diamond Parma, ID) at the Virginia-Maryland Veterinary Teaching Hospital (Blacksburg, VA) following the standard cannulation surgical procedure [22].

Grazing and rumen sample collection
A rotational grazing study was conducted by moving the animals between two Virginia Tech Agricultural Research and Extension Centers (ARECs) located 130 miles apart. The cattle were first allowed to graze on Jesup MaxQ [23]  . At a point of each grazing period, as described in the Results, the samples of rumen contents of each animal were collected from three different locations, top of the mat, middle of the mat, and ventral sac. For the top layer, a sample was taken directly by a gloved hand. Samples from the middle of the mat and ventral sac were collected using a custom-made sampling device consisting of a vacuum pump that was connected to a vacuum chamber. Here, each sample was drawn out through the collection tube of the sampling device and deposited directly into the respective sample container. The details of this unit are described in the Supplementary Methods and Fig. S1 (available in the online version of this article). Each sample was 500-700 ml in volume and was mixed to get homogeneity. This large volume allowed a better representation of the rumen content in a sample. The liquid and solid fractions of a sample were separated by filtration through a combination of three layers of sterile cheesecloth that had been wrapped separately and baked at 175 °C and then transported in sealed 15 ml conical Falcon tubes (catalogue number: 62406-200, VWR International, Radnor, PA) on dry ice to the laboratory and stored at −80 °C. All containers, tubes and pipes coming into direct contact of the samples were washed prior to use with Tergazyme, a detergent (Alconox, White Plains, NY), and 2 % phosphoric acid to remove microbial and nucleic acid contaminations.

DNA extraction and 16S rDNA amplicon sequencing
From each rumen sample DNA was purified via bead beating and a modified version of phenol, chloroform, and isoamyl alcohol extraction, isopropanol precipitation and ethanol wash [24]; details are presented in the Supplementary Methods. To determine whether a DNA preparation contained inhibitors that could affect downstream processes, near full length 16S rRNA gene was PCR amplified using universal primers 27F and 1525R [25]. Metagenomic DNA preparations that yielded the desired amplicons were subjected to paired-end sequencing of the hypervariable region 4 (V4) of the 16S rRNA gene at the Environmental Sample Preparation and Sequencing Facility at the US Department of Energy's Argonne National Laboratory (ANL). The amplicons were generated using an optimized set of primers, 515F (Parada) -806R (Apprill), that provides the best coverage of prokaryotic 16S rRNAs [26]. The sequencing was performed on the Illumina MiSeq platform (Illumina, San Diego, CA).

Quantification and statistical analysis of 16S rDNA sequences
Raw sequence data obtained from the ANL were analysed by the QIIME 2-2019.4 pipeline [27] for preprocessing and removal of contaminants. The amplicon sequence variants (ASVs) were generated by DADA2 [28] and clustered to operational taxonomic units (OTUs) using vsearch [29] at 99 % sequence similarity. A pre-trained naïve Bayes classifier was used to annotate the sequences using the silva 132 database [30], and sequences annotated as chloroplast and mitochondria were removed from the dataset. The dataset was imported and analysed by using the R statistical packages [31].
Alpha diversity metrics were calculated using species richness estimator Chao1 of the microbiomeSeq package [32] with samples rarefied to 5598 sequences per sample (Fig. S2). Pairwise ANOVA (P<0.001) was calculated to determine the significance of the difference between the species richness of two groups. Community comparison between samples was performed using nonmetric multidimensional scaling (NMDS) analysis using Bray-Curtis dissimilarity distances, and unsupervised principal component analysis (PCA) using the data normalized into relative abundance with total sum scaling (TSS) and transformed into its logarithmic values using centred log-ratio (CLR) as described [33,34]. Phyloseq [35] and vegan [36] were used in R to calculate sample distances (Bray-Curtis and UniFrac with average as method), perform NMDS analysis, and compute a dendrogram of the hierarchical clustering. The TSS and CLR data transformation, as well as sample ordination on PCA were performed using mixMC [33]. A non-parametric permutational analysis of variance (PERMANOVA) and analysis of similarities (ANOSIM) of adonis function in vegan [36] were used to determine which sample parameters (genotype, rumen sample fraction, rumen depths, tall fescue variety and grazing transition) caused sample clustering (permutation: 999, P<0.05, with Bray-Curtis dissimilatory distance matrix). Relative abundance of the OTUs in various samples were estimated using phyloseq [35] and microbiome [37] packages. Core microbial community in a rumen sample was characterized from TSS normalized data at 95 % prevalence threshold across all samples. The assessment of differential abundance of microbial species between two grazing periods with respect to either sample fractions (solid or liquid) or susceptibility to tall fescue toxicosis was performed using DESeq2 [38] with Wald P-value<0.001, fitType = 'parametric' , and sfType = 'poscounts' . Non-parametric Kruskal-Wallis (P Kruskal-Wallis =0.05) and Wilcoxon test with continuity correction (P Wilcoxon =0.05) were performed to determine the significance differences in the relative abundance of a microbial species across rumen fraction and grazing transition.

Rumen samples and sequences of 16S rDNA V4 amplicons
The first set of samples were collected at the end of the initial 5-month-long MaxQ grazing period (MaxQ-1). The animals then grazed KY-31 for 24 days. At the end of this period (KY-31), a second set of samples were collected, and the animals were placed on MaxQ grazing (MaxQ-2). After 23 days of the MaxQ-2 phase, the last set of samples were collected. Under all these grazing conditions, the rumen pH remained constant (7.00±0.24); the same was the case for the animal's rectal temperature (102.7 °F±0.74; 39.3 °C±0.4) and average daily gain. Sampling in triplicate from three rumen locations, top and middle of the mat and the ventral sac, of eight animals and for three grazing periods produced 216 solid and 216 liquid fractions. Sequencing of the 16S rDNA-V4 elements of the DNA preparations generated from these samples provided 6.2 million sequences with 9403 ASVs. A clustering of the data at the 99 % similarity threshold produced 5547 unique OTUs. Further filtering, including the removal of 12 samples with less than 5200 reads to avoid introducing noise (Table S1), provided a final data set comprised of 5488 OTUs from 420 samples.

Alpha and beta diversity of tall fescue rumen microbiome
Overall, the species richness in the microbiome of the solid fractions of the rumen contents was similar to that of the respective liquid fraction (Fig. 1). In terms of the location within the rumen, the top of the mat harboured a less diverse microbial population than the middle of the mat and the ventral sac ( Fig. 1). The grazing transitions altered the species richness significantly ( Fig. 1), distinguishing the MaxQ-2 set (return from KY-31 to MaxQ) with higher species richness than MaxQ-1 and KY-31 ( Fig. 1); the richness was the lowest under KY-31 grazing (Fig. 1). For both MaxQ grazing periods (MaxQ-1 and MaxQ-2), the solid fraction presented a microbiome with higher species richness than the respective liquid fraction, and the situation was opposite with KY-31 ( Fig. 1). The KY-31 grazing resulted into a less diverse sessile microbial community (Fig. 1). The rumen microbiome during the MaxQ-2 grazing of cattle that were susceptible to tall fescue toxicosis had a slightly higher species richness compared to rest five sets (Fig. 1).
Community comparison via Bray-Curtis and Euclidean distances ordination (NMDS and PCA, respectively) showed clear separations between the solid (sessile) and liquid (planktonic) fractions of the microbiomes and the tall fescue types used ( Fig. 2a, b). The distinctions between the rumen microbiomes on MaxQ-1, KY-31, and MaxQ-2 were more pronounced for the liquid fractions, and here MaxQ-1 samples were clearly separated (Fig. 2a, b). A similar separation, although to a lesser extent, was observed for the solid fractions. Interestingly, MaxQ-2 had some commonalities with KY-31 but not MaxQ-1 (Fig. 2a, b).
Quantitative assessment of the parameters that caused sample clustering was performed using permutational multivariate analysis of variance (PERMANOVA; 999 permutation and P PERMANOVA =0.0001) and similarities (ANOSIM; P ANOSIM =0.001) of the Bray-Curtis dissimilarity distance. Table 1 presents the respective P-values. The results show that the rumen depth was an insignificant parameter in both tests (P PERMANOVA =0.3045 and P ANOSIM =0.104), indicating that it did not contribute to the separation of the samples. In both PERMANOVA and ANOSIM analysis, the animal genotypes were found to be a significant parameter (P PERMANOVA =0.0001 and P ANOSIM =0.001) ( Table 1), although it did not cause separation on the ordination plot (Fig. 3).

Tall fescue cultivar change-driven alterations in the core rumen microbiome
Covering all grazing conditions the rumen core microbiome was represented by 25 OTUs and contained bacteria with the taxonomic identities as shown in Fig. 4 Table S2. There were remarkable alterations in the locations of the Bacteroidetes, . Analysis of alpha diversity occurred across rumen fractions (solid and liquid), rumen depths, grazing periods, combination of rumen fractions and grazing periods, and combination of susceptibility to tall fescue toxicosis and grazing periods. Prior to species richness analysis, all of the samples were rarefied to 5598 reads, which was the lowest number of reads present in a single sample (Fig. S2). Horizontal bars link the samples compared in statistical analyses. *Represent P<0.001 in a pairwise ANOVA statistical test.
Firmicutes, Fibrobacter, and Lentisphaerae populations within the microbiome due to changes in the tall fescue type (Fig. 4a,  b). We describe this effect for the sessile and planktonic populations separately.   At 95 % sample prevalence, no Euryarchaea OTUs were detected in the core rumen microbiome. The prevalence thresholds that allowed the detection of the Euryarchaeota OTUs were 70, 80 and 85 % for KY-31, MaxQ-1 and MaxQ-2, respectively. At 70 % prevalence, the Euryarchaeota was found to be represented by the Methanobacteriaceae and Methanomethylophilaceae families under all grazing conditions. Further analysis of all the data without a prevalence cutoff showed that methanogen abundance did not change significantly with any one of the parameters studied (Fig. 5), except the KY-31 grazing slightly enhanced the abundance of species from the Methanosarcinaceae family in the liquid fraction (Fig. 5).
The core bacterial microbiomes of steer grazing MaxQ (pre-and post-KY-31) and KY-31 tall fescue pasture had clear differences (Fig. 4b). Prevotella 1 genus of the phylum Bacteroidetes had higher abundance in MaxQ-1 (21 %) compared to KY-31 (17 %) and MaxQ-2 (16 %) grazing; the % values are calculated with respect to the core microbiome OTUs. Transition to KY-31 grazing promoted the growth of RC9 species of the Rikenellaceae family (Bacteroidetes phylum) significantly, as the abundance of the respective OTUs increased from 6.47 and 16.41 % in the liquid and solid fractions under MaxQ-1, respectively, to 12.15 and 25.71 % in the corresponding sets during KY-31 grazing. Compared to the MaxQ-1, the KY-31 grazing provided higher abundances for bacteria belonging to the candidate division vadinBE97 of the Lentisphaerae phylum, and the Firmicutes of NK4A214 and XPB1014 genera from the Ruminococcaceae and Lachnospiraceae families, respectively, and of unclassified Clostridiales order (Fig. 4b). Under KY-31 grazing, the population of Fibrobacter species of the Fibrobacteres phylum, and Succiniclasticum, and Butyrivibrio 2 species of the Firmicutes and the RF16 family of the Bacteroidales order decreased, and for the last group the drop was substantial (about 50%) (Fig. 4b. Table S2).
The transition from KY-31 back to MaxQ resulted in another reconstruction of the core microbiome characteristics. Some of the bacterial species, for which the abundance decreased during KY-31, regained their presence under MaxQ-2 to the level that was seen under MaxQ-1. This was the case for the bacteria of the RF16 family (Bacteroidales order), for which the abundance decreased from 9.74 % in MaxQ-1-4.22 % in KY-31 and rose to 9.41 % under MaxQ-2 grazing. Similar was the observation for Succiniclasticum species of the Firmicutes phylum (Fig. 4b). The abundances of Pseudobutyrivibrio and Butyrivibrio 2 species of the Firmicutes phyla and Fibrobacter remained low and did not recover in the MaxQ-2 grazing (Table S2). A non-parametric Kruskal-Wallis and Wilcoxon rank-sum test [39,40] also showed a reduction in the abundance of Fibrobacter in sessile fraction due to shifting from MaxQ to KY-31 and a poor recovery upon return to MaxQ (Fig. 6).

Differential abundance analysis of grazing transition driven shift in microbiome
We performed a differential abundance analysis using DESeq2 [38] to identify the shifts in the microbiome composition. The use of negative binomial generalized linear model allows the comparison of log2 values for the counts for the same OTU in different    (Table S3). A similar analysis was performed with rumen fractions as a parameter (Table S4). The results of these analyses for MaxQ-1 vs KY-31 and KY-31 vs MaxQ-2 are summarized in Fig. 7, both in terms of the total microbiomes as well as sessile and planktonic fractions, and that for MaxQ-1 vs MaxQ-2 appear in Fig. 8. The fibre-adherent community in the rumen during MaxQ-1 grazing had fourfold more abundance for Fibrobacter than that with KY-31, and similarly higher abundance OTUs were seen for the genus Saccharofermentans, family of Lachnospiraceae AC2044, Spirochaetes MVP-15 order, and WCHB1-41 class from Kiritimatiellae class (Fig. 7). For the Bacteroidetes phyla, a transfer from MaxQ-1 to KY-31, enriched the rumen microbiome greatly with an uncultured bacterium from the Bacteroidales order and Rikenellaceae RC9, Prevotella 1 and Paraprevotella species (Fig. 7). For the Firmicutes phyla, this grazing transition enriched only one species that belonged to the genus of Lachnospiraceae FCS020. After a month from the return to MaxQ (MaxQ-2), the abundance of a set of OTUs was elevated and these had shared annotations but not identical sequences with a group of OTUs in MaxQ-1. This observation could be interpreted as the occurrence of two groups of organisms with close but not exactly the same genetic identities in these two sets. This group was comprised of Lachnoclostridium 10, Christensenellaceae R-7, Treponema 2, Anaeroplasma, Saccharofermentans and Lachnospiraceae AC2044 genera and uncultured bacteria belonging to the BS11 gut group of Bacteroidales order (Fig. 7).
An interesting pattern was seen for Prevotella 1 species from Bacteroidetes phyla, as a total of 60 distinct OTUs for this genus were detected in the samples from MaxQ-1 and MaxQ-2 periods. While 34 Prevotella 1 OTUs were found during MaxQ-1 grazing of which 19 were highly abundant, the corresponding values were 26 and 16 in the MaxQ-2 period, respectively. Another significant observation was that certain Eubacterium species, which are Firmicutes, were abundant during the KY-31 grazing irrespective of the rumen fractions (Fig. 7) and Eubacterium ruminantium in particular was present at a high abundance in the MaxQ-2 period (Fig. 8).

Host genotype-associated microbial community characteristics
A differential abundance (DESeq2) analysis showed that Bacteroidetes OTUs were significantly more abundant in the steers that are susceptible to tall fescue toxicosis ( Fig. 9. Table S5), and also higher number of OTUs from the Proteobacteria phyla such as those corresponding to the Sutterella genus and Rhodospirillales and Deltaproteobacteria PB19 orders were also associated with this host genotype. In the rumen of this group the OTUs that had higher abundance during MaxQ-1 and MaxQ-2 grazing but were less so in the KY-31 grazing mostly belonged to Prevotellaceae, Marinilabiliceae and Rikenellaceae families, and some of these were annotated as unclassified Rhodospirillales order from Proteobacteria phylum and Gastranaerophilales from the newly defined phylum of Melainabacteria that is closed-related to Cyanobacteria. The cellulolytic Fibrobacter species were 22-and 59-fold more abundant in susceptible steers during MaxQ-1 and MaxQ-2 grazing, respectively (Fig. 9), and such a distinction was not seen during KY-31 grazing, where the abundances of these organisms were generally low.
The rumen microbiome of steers that are susceptible to tall fescue toxicosis was associated with a higher abundance of Firmicutes. Particularly, the species belonging to the R-7 group genera from the Christensenellaceae family showed a preferential association with the rumen of susceptible steers, being 2-, 59-and 32-fold more abundant during MaxQ-1, KY-31, and MaxQ-2 grazing, respectively, in this set compared to the toxic fescue tolerant animals (Fig. 9). In addition to elevated Firmicutes abundance, a higher association of certain members of the Tenericutes family, specifically the Mollicutes order (Mollicutes RF39, Mycoplasma and Mollicutes NED5E9) was observed for the rumen of susceptible steers.

DISCUSSION
We have determined the differences between the compositions of rumen microbiomes of two groups of cattle that grazed a toxic and a non-toxic cultivar of tall fescue, namely KY-31 and MaxQ and inferred the respective metabolic implications. Previous studies on bovine rumen microbiome under similar nutritional conditions focused primarily on the microbial degradation of ergovaline, which makes KY-31 toxic [8, 14-16, 41, 42] and some of these employed faecal microbiome as a proxy [14,15,42]. We used the grazing on a KY-31 pasture as a perturbation in the middle of MaxQ grazing, and recorded the resultant changes in the rumen microbiome. This effort had two major outcomes. It showed that an analysis targeting the sessile and planktonic components separately uncovered more changes in the microbiome than the one focused on overall microbiome. Thus, an effective study of rumen microbiome cannot rely on the analysis of rumen fluid or whole microbiome only. The other was the finding that KY-31 was a dominant remodeler of the microbiome composition. A return from KY-31 to MaxQ grazing did not restore some of the original MaxQ-type features. One of these cases was a severe reduction in the abundances of the OTUs assigned to certain key rumen cellulolytic bacteria. The location within the microbiome, sessile or planktonic segment, where Firmicutes and Bacteroidetes were found in high abundances, changed when cattle were moved from MaxQ to KY-31 grazing, and the original status in most part was restored upon return to MaxQ. As elaborated at the end of the Discussion, these remodelling events likely represent alterations in the microbial team composition and/or task allocations while leaving a broader ruminal metabolic function unaltered. The observation that some of the KY-31-instituted changes lingered even after a switch to MaxQ grazing represented a major departure from what has been seen with various perturbations in ruminal conditions. For example, for bloat or acidosis, the microbiome fully regains the original characteristics within 7 days of the termination of the disturbance [43,44]. The KY-31 grazing caused an enrichment of bacteria with potential capabilities of degrading ergot alkaloids and apparently lowered the methane production potential of the rumen. The cattle's relative susceptibility to tall fescue toxicosis influenced rumen microbiome composition minimally. These observations were significant, as the 3 week long intervening KY-31 grazing, a terminal MaxQ grazing of similar duration, and much longer first MaxQ period offered sufficient times for establishing stable microbiome compositions; in cattle the sign of tall fescue toxicosis appears in 5-15 days of consuming toxic tall fescue [4] and the faecal bacterial community composition alters within 14-28 days of a shift from a mixed ration to MaxQ or KY-31 [14]. Also, the events identified in this study were distinct from those revealed with faecal microbiome as a proxy [14,15].
Our analyses focused on the rumen fractions identified a potentially major microbiome remodelling event. The MaxQ-1 to KY-31 shift reduced the species diversity of the fibre-attached population and increased that of the liquid-associated population, and MaxQ-1 and MaxQ-2 grazing offered an opposite scenario (Fig. 1). A phylum level observation was even more dramatic (Fig. 4a). A move from MaxQ to KY-31 grazing reduced the relative abundance of Firmicutes and elevated that of Bacteroidetes in the sessile fraction and the opposite change occurred in the planktonic fraction, and a return to MaxQ brought back the original composition, providing a higher abundance of Firmicutes and a lowered level for Bacteroidetes in the sessile fraction and the reverse for the planktonic fraction. These effects were not seen at the total microbiome level as both KY-31 to MaxQ provided similar overall abundances of Firmicutes and Bacteroidetes in the rumen, which in combination constituted >80 % of the bacterial population. However, the reversibility did not hold for certain features that are key to an optimal functioning of the rumen.
For the sessile fraction of the rumen microbiome, the transfer of cattle from MaxQ to KY-31 grazing caused an increase in the abundance of the Bacteroidetes, such as saccharolytic Prevotella 1 and Paraprevotella species and RC9 of the Rikenellaceae family, and lowered the abundance of the Firmicutes representing major rumen cellulolytic populations with members such as Pseudobutyrivibrio and Butyrivibrio 2 (Figs 4b and 7). The opposite was the case for the planktonic population. There was also a decrease in the abundance of fibrolytic Fibrobacter species in the solid fraction (Fig. 6). This void in the sessile population was filled with an enrichment of cellulolytic firmicutes belonging to the Ruminococcaceae and Lachnospiraceae families (Fig. 4b, Fig. 7), and the Eubacterium genus (E. coprostanoligenes, E. ventriosum and E. ruminantium). The abundance of E. ruminantium, a typical rumen cellulolytic bacterium [52], was elevated substantially during KY-31 grazing and remained so in MaxQ-2. Bloat and acidosis are known to affect Fibrobacter succinogenes and the Firmicutes species such as Ruminococcus albus and Ruminococcus flavefaciens [43,45,53]. These conditions are associated with pH values of 5.0-5.6 [43,54], whereas under our grazing conditions, the rumen pH remained neutral (7.00±0.24). Thus, certain KY-31-associated factor(s) were possibly deleterious to Fibrobacter, Pseudobutyrivibrio and Butyrivibrio, and their loss was compensated by the increased abundances of select Firmicutes.
For the planktonic segment of the microbiome, KY-31 grazing not only altered the level of the cellulolytic and non-cellulolytic carbohydrate hydrolysing bacterial population but also brought novel possibilities for organic and cellular material degradation activities. For example, the increased abundance of planktonic Firmicutes likely not only compensated for the above-mentioned loss of the cellulolytic population in the sessile fraction, but also enhanced the ability to degrade plant-derived organics. A member of this phylum, a Coprococcus species, was present at an enhanced level during KY-31, and these organisms degrade plant toxins such as nitropropionic acid, nitropropanol [55] and phloroglucinol [56]. Similarly, the Anaeroplasma and Mycoplasma species as well as the bacteria from the RF9 family of the Mollicutes order, which belong to the Tenericutes phylum and were present in elevated abundances during the KY-31 grazing, would have provided two benefits. First, their saccharolytic and proteolytic activities likely compensated for the lowered Bacteroidetes abundance in the planktonic community. Second, their bacteriolytic and proteolytic activities were potentially useful in recycling dead cells [57]. In this context, it is noted that not all rumen mycoplasmas are bacteriolytic [57]. Proliferation of acetogens such as Acetitomaculum, sulfate-reducing Desulfovibrio species, and propionate-producing bacteria of the Rikenellaceae and Erysipelotrichaceae families indicated that under KY-31 grazing, the contribution of methanogenesis as a hydrogen sink might have been lowered [58]. This suggestion is consistent with the observation that during KY-31 grazing the rumen carried Methanosarcina, which can survive without hydrogen [59,60], in a higher abundance than that with MaxQ-1 and MaxQ-2 (Fig. 5). Also, since the methanogen abundance did not change much, KY-31 or its degradation products were not likely toxic to these archaea.
The transition of the cattle from KY-31 to MaxQ provided an unexpected outcome. Even after 23 days of MaxQ-2 grazing following the KY-31 period, the rumen microbiome did not return completely to the MaxQ-1 stage (Figs 2 and 8), and exhibited some of the KY-31 stage characteristics (Figs 2 and 4). The core ruminal microbiome during the MaxQ-2 had the highest abundance of Firmicutes, comprising up to 50 % of the total sequence, in both solid and liquid fractions (Fig. 4a, Table S2). One of the special features of this period was the higher abundance of Spirochaetes, specifically the species of the Treponema 2 genus, and organisms belonging to the Pedosphaeraceae family of Verrucomicrobia phylum (Fig. 4). Several organisms such as the Firmicutes of the Tyzzerella, Ethanoligenes and Succiniclasticum genera as well as those belonging to Bacteroidales RF16 order and Rikenellaceae RC9 families of the Bacteroidetes phyla that were present in the MaxQ-1 rumen microbiome in high abundance and were affected due to KY-31 grazing were able to re-establish substantial presence in the rumen under MaxQ-2 grazing. However, the abundances of Fibrobacter and Butyrivibrio 2 species in the sessile fraction, which were severely reduced in the KY-31 period, did not return to the MaxQ-1 type level during MaxQ-2 grazing (Fig. 8). On the other hand, the Treponema species, which were present at high levels during the MaxQ-1 grazing and had a reduced presence at the KY-31 phase gained significant abundance in the sessile fraction with the MaxQ-2 (Fig. 7). This is an intriguing observation as Treponema species are known to improve the cellulolytic function of cellulose degraders, especially Fibrobacter [61,62]. It is likely that Treponema assisted Fibrobacter species during MaxQ-1 grazing and other cellulytic bacteria at the MaxQ-2 stage.
We have developed two hypotheses for the factors underlying the KY-31-instigated changes in the cattle rumen microbiome. The first concerns ergot alkaloids released from the grass, which could have been toxic to the fibre-adhered bacteria and served as high-value nutrients for the liquid fraction microorganisms. The fescue-associated alkaloids are known to inhibit ruminal fibre digestion [63][64][65]. The second hypothesis is that the distinct composition of KY-31 drove the remodelling. However, the factors causing these changes to linger remain hard to rationalize.
A comparison of our observations with those of others [14,15,42] suggested that faecal microbes are unlikely biomarkers for tall fescue toxicosis. In two studies where cattle were moved from a mixed ration to MaxQ and KY-31 in parallel [14,15] or placed on KY-31 only [42], Coprococcus one and several other Firmicutes belonging to the Ruminococcaceae and Lachnospiraceae families, Paraprevotella species from Bacteroidetes phylum, and Tenericutes of Mycoplasmataceae family were found in higher abundances in the faeces of KY-31 tall fescue-fed cattle [14,15,42]. Similar changes were observed in our MaxQ-1 to KY-31 shift (Fig. 7). The faecal microbiome of cattle grazing tall fescue also contains Proteobacteria, Euryarchaeota and Lentisphaera phyla at abundances similar to those observed by us for the rumen [14,15,42]. However, unlike the situation with the rumen microbiome studied here, in a faecal microbiome Actinobacteria and Firmicutes are found in significantly higher abundances, while Bacteroidetes, Fibrobacteres and Tenericutes are underrepresented [14,15,42]. Similar is the case with alfalfa [45] and a mix of bermudagrass and Dallisgrass [50] (Table S6). Such differences are expected as the foregut deals with heterogeneous plant materials and in the hindgut undigested plant fibres are degraded [50,[66][67][68]. Also, the hindgut selects for microbes that can tolerate bile salts [67], which may inhibit many Gram-negative bacteria such as Bacteroidetes and Fibrobacteres and help to enrich Gram-positive bacteria such as Actinobacteria and Firmicutes [67]. In Holstein steer, the foregut and hindgut microbiomes share only 30 % of the species [14,17].
An in vitro enrichment study identified certain Clostridiaceae and Prevotellaceae as potential ergovaline degraders [41], and a comparison of faecal microbiomes of cattle grazing toxic and nontoxic tall fescue led to similar assigments for certain Paraprevotella and Coprococcus 1 species and several members of the Ruminococcaceae, Lachnospiraceae and Mycoplasmataceae families [14,15,42]. The planktonic segment of the rumen microbiome of pregnant ewes is thought to carry ergovaline degrading bacteria from the Lachnospiraceae and Veillonellaceae families of the Firmicutes phylum and Coriobacteriaceae family of Actinobacteria phylum [16]. From the observed enrichment in the rumen as the cattle were shifted from MaxQ-1 to KY-31 grazing, we have identified Coprococcus 1, Paraprevotella, Prevotella 1, RC9, Mycoplasma, Anaeroplasma and several members of the Ruminococcaceae and Lachnospiraceae families as potential ergovaline degraders. These similarities between the hypotheses arising from rumen and faecal microbiome studies (current report and [14][15][16]42]) suggest that ergovaline is metabolized in the rumen and large-intestine through similar processes. Ergovaline generated from KY-31 in the rumen is known to be available in the cattle's large intestine [13,69].

CONCLUSION
The compositions of rumen microbial communities in the cattle that grazed MaxQ and KY-31 tall fescue pastures, were distinct. Interestingly, this distinction was captured only when the comparison was performed at the sessile and planktonic levels separately and not at the whole microbiome level. The MaxQ grazing helped to enrich a cellulose metabolizing community made up of Fibrobacter, Pseudobutyrivibrio and Butyrivibrio 2, and under KY-31 grazing the same overall function was provided by a consortium of Eubacterium species and certain members of Ruminococcaceae and Lachnospiraceae families. A similar change occurred in the saccharolytic team. There was an apparent enhanced need for Tenericutes for the digestion of KY-31. A major observation was that some of the effects of KY-31 grazing, such as the reduction of Fibrobacter and enrichment of substitute cellulolytic bacteria, lingered even after the cattle were moved to MaxQ grazing. It was hypothesized that the alkaloids produced by the fungal endophyte of KY-31 as well as the distinct compositions of MaxQ and KY-31 were the drivers for these effects. Often, beef cattle are raised by grazing tall fescue until the age of 6-12 months, and then, are placed on feedlots until the age of 18-21 months for finishing [21]. Our results suggest that the cattle grazing KY-31 and MaxQ arrive on feedlots with distinct microbiomes, and the KY-31-specific characteristics linger. Thus, a fuller understanding of the metabolic details of the MaxQ-and KY-31-specific microbiomes will help to develop strategies for mitigating tall fescue toxicosis as well as improving the feed utilization efficiencies at both the grazing and finishing steps. It will also reveal new biomass degradation systems of applied value. The next steps for obtaining such details would be the generation and analysis of metagenome assembled genomes (MAGs), transcript, protein and metabolite level analysis, and metabolic modelling.