Integrated microbiota–host–metabolome approaches reveal adaptive ruminal changes to prolonged high-grain feeding and phytogenic supplementation in cattle

Abstract Diets rich in readily fermentable carbohydrates primarily impact microbial composition and activity, but can also impair the ruminal epithelium barrier function. By combining microbiota, metabolome, and gene expression analysis, we evaluated the impact of feeding a 65% concentrate diet for 4 weeks, with or without a phytogenic feed additive (PFA), on the rumen ecosystem of cattle. The breaking point for rumen health seemed to be the second week of high grain (HG) diet, with a dysbiosis characterized by reduced alpha diversity. While we did not find changes in histological evaluations, genes related with epithelial proliferation (IGF-1, IGF-1R, EGFR, and TBP) and ZO-1 were affected by the HG feeding. Integrative analyses allowed us to define the main drivers of difference for the rumen ecosystem in response to a HG diet, identified as ZO-1, MyD88, and genus Prevotella 1. PFA supplementation reduced the concentration of potentially harmful compounds in the rumen (e.g. dopamine and 5-aminovaleric acid) and increased the tolerance of the epithelium toward the microbiota by altering the expression of TLR-2, IL-6, and IL-10. The particle-associated rumen liquid microbiota showed a quicker adaptation potential to prolonged HG feeding compared to the other microenvironments investigated, especially by the end of the experiment.


Introduction
Ruminants lar gel y depend on the activity of the rumen microbiota to digest and utilize complex dietary pol ysacc harides for the production of energy, protein, and the maintenance of a healthy rumen ecosystem, owing to the symbiotic host-microbiome relationship (Deusch et al. 2017 ).The rumen itself is a complex envir onment, and man y studies hav e highlighted the importance of various niches, hosting a variety of different microorganisms [solid-associated (SAM), liquid-associated (LAM), and epithelialadher ent (EAM) micr obiota], differing in pr eferr ed substr ate for degradation and metabolism.This differential diversity combined with nic he inter action is what allows for an optimal utilization of ingested nutrients (De Mulder et al. 2016, Ricci et al. 2022 ).Despite a significant body of work, our understanding of the metabolic role of these ruminal micr obial nic hes and the extent by which they are affected by diet remains limited, also due to the fact that they are often investigated separately.
A compr ehensiv e c har acterization of the diet × microbiota interactions and the way they affect the host is of particular interest for feeding regimes that are based on the inclusion of large amounts of gr ains, commonl y fed to meet the high energy requirements of lactating cows .T his type of feeding triggers a series of reactions of the ruminal microbiota, that can switch its activity and composition to process the new substrates, ending up with an increased production of fermentation end products, such as volatile fatty acids (VFAs) (Ametaj et al. 2010, Ricci et al. 2022 ).When the ruminal homeostasis is compromised due to the accumulation of VFAs and the consequent decrease of pH, the animals risk to experience subacute ruminal acidosis (SARA) (Zebeli et al. 2012 ).In order to adapt to fluctuations in the pH of the rumen milieu, the epithelium increases its absor ptiv e surface area through the combined expression of genes related to growth and nutrient uptake (Steele et al. 2012a, 2015, Dieho et al. 2016 ).Ho w e v er, these r elativ el y r a pid and pr ofound modifications can alter the integrity of the epithelial barrier, compromising the structure of tight junctions and desmosomes (Steele et al. 2011, Liu et al. 2013, McCann et al. 2016 ).At the same time, the above mentioned dietary challenges may induce changes in microbiota composition and function, leading to production of other microbederived compounds with proinflammatory properties such as biogenic amines, lipopol ysacc harides (LPS) or lipoteic hoic acid, whic h can alter epithelial inflammation pathways resulting in both local and systemic inflammation (Zhang et al. 2016, Zhao et al. 2018 ).These alterations of the microbial composition and activity create a perturbation of the ruminal homeostasis, defined as dysbiosis (Sommer et al. 2017 ).
Giv en suc h pr emises, a healthy rumen depends on the composition of the microbiota and on the metabolites produced by their activity, as well as on the structural and functional integrity of the epithelium.These dynamics have been extensiv el y studied, but usually focusing on isolated aspects, and mainly dealing with only one niche of the rumen microbiota (Ametaj et al. 2010, Fernando et al. 2010, Petri et al. 2020 ).To overcome the limitations of pr e vious studies, and to c har acterize the r esponse of the rumen as a complex multifunctional organ, in this study we c har acterized the micr obiota composition of thr ee ruminal niches , the ruminal metabolome , and the rumen epithelial reaction to a prolonged dietary challenge.By collecting samples consecutiv el y ov er a period of 5 w eeks, w e aimed to define the precise moment for the onset of dysbiosis, as well as to e v aluate the ada ptiv e ca pacity of the rumen alone or in association with a natural remedy.
Among the strategies to counteract the deleterious effects of high grain (HG) feeding, the use of phytogenic feed additives (PFA) has been demonstrated to modulate ruminal microbial composition and its activity, both in vitro and in vivo (Calsamiglia et al. 2007, Bodas et al. 2012 ).Inter estingl y, PFA in shortterm HG feeding have also been reported to alter inflammatory biomark ers, to xin release, and e pithelial gene expr ession, pr oviding positive and encouraging results, such as reduced LPS and biogenic amines concentration, with consequent mitigation of the inflammatory response (Drong et al. 2017, Humer et al. 2018, Petri et al. 2020 ).Ther efor e, although the mec hanisms of action of such phytogenic compounds are still not fully understood, it remains to be determined if the use of PFA during prolonged HG feeding could provide beneficial effects for the ruminal environment.
The aim of this study was to investigate the effects of prolonged HG feeding challenge on various niches (SAM, LAM, and EAM) of the ruminal ecosystem, by e v aluating the structur e and functionality of the epithelium, the composition of the microbiota and the production of metabolites.We also e v aluated the effects of a PFA supplementation, consisting of a blend of menthol, thymol, and eugenol, and we hypothesized that the feed ad diti ve would support the plasticity and adaptability of the ruminal ecosystem, helping to pr eserv e a healthy ruminal en vironment.T he integration of metabolome and microbiota with a panel of genes selected to provide a comprehensive picture of the epithelial reaction allo w ed us to c har acterize the ruminal response to a dietary challenge and a PFA supplementation over a period of 5 weeks , pro viding novel and useful insights on the critical points of the adaptive processes of this ecosystem.

Experiment design and animal housing
The trial was conducted at the r esearc h farm of the University of Veterinary Medicine , Vienna, between J une and September 2019.The experimental pr ocedur e was a ppr ov ed by the Institutional Ethics and Animal Welfare Committee of the University of Veterinary Medicine Vienna and the Austrian national authority according to the European and Austrian laws for animal experiments (protocol number: BMBWF-68.205/0003-V/3b/2019).A total of nine Holstein nonlactating cannulated cows (Bar Diamond, P arma, ID, USA) wer e divided into two groups of four and five animals balanced for body weight (mean body weight: 992 ± 73 kg, mean age: 10.0 ± 0.8 years), in a cross-over design with two experimental runs consisting of 6 weeks each.The animals were fed an onl y for a ge diet (baseline), consisting of 75% gr ass sila ge, 15% corn silage, and 10% grass hay in dry matter basis for 1 week.Then, the co ws w er e tr ansitioned to a HG diet with 65% concentrate via stepwise dail y incr ements of 10% concentr ate ov er 1 week; the HG diet was fed for the following 4 weeks.Details about the diet composition are given in our companion paper (Rivera-Chacon et al. 2022 ).In addition, one group of animals received a blended PFA at 400 mg kg −1 (dry matter basis) (Digestarom ® , a mixture of essential oils and extracts including menthol, thymol and eugenol; BIOMIN Holding GmbH, which is part of DSM), and the second gr oup serv ed as contr ol; the gr oups wer e inv erted between runs, resulting in nine genuine replicates per treatment.The feed additive was dosed through the ruminal cannula during the weeks of for a ge feeding and adaptation to the HG diet, while during the HG feeding the phytogenic blend was included in the concentrate.In the adaptation week the dosage of ad diti v e administer ed thr ough the cannula was adjusted according to the percentage of concentrate included in the diet to reach the targeted daily intake .F eed was mixed and provided once daily (T rioliet T riomatic T15, the Netherlands), and was available ad libitum , together with water and miner al bloc ks.Ev ery cow had access to a single feed bunk and daily feed intake was recorded automatically (Insentec B.V., the Netherlands).Data for feed intake and ruminal pH ar e r eported in our companion paper (Rivera-Chacon et al. 2022 ).For the whole duration of the experiment, the animals were housed in a free-stall barn with deep litter cubicles (2.6 × 1.25 m, straw litter).

Sample collection
Samples were collected at five time points, once at baseline and once per e v ery week of HG.Rumen content and rumen pa pillae were collected 4 h after the morning feeding.To collect samples of SAM and LAM for microbiota analyses, a handful of digesta was sampled from the ruminal mat and the ventral sac of the rumen (Castillo-Lopez et al. 2014 ).The liquid (LAM) was filtered through four layers of sterile gauze and collected in a beaker.The solid digesta (SAM) was sampled with sterile tweezers.Rumen fluid for metabolomics was collected from the v entr al sac of the rumen using a sterile 20 ml syringe.All samples of rumen content were sna p fr ozen and stor ed at −80 • C. For rumen papillae, the rumen was partially emptied and the biopsies were collected following a method pr e viousl y described (Wetzels et al. 2016(Wetzels et al. , P acífico et al. 2022 ) ).The tissue samples for microbiota (EAM) and gene expres-sion anal yses wer e sna p fr ozen and then stor ed at −80 • C. Samples for histology were fixed in 4% formalin solution (Liquid Production GmbH, Germany) until further processing.Right after the sampling, the rumen was refilled with its content.Each part of the equipment w as w ashed and thor oughl y disinfected after every use.

DN A extr action, sequencing, and sequences analysis
A total of 90 samples were collected for EAM and SAM over the 5 weeks of experiment, while 89 samples were processed for LAM (one sample was lost due to technical issues).DNA was extracted using DNeasy PowerSoil Kit (Qia gen, German y) with some modifications.Samples wer e pr epr ocessed performing mechanical and enzymatic lysis, as described by Ricci et al. ( 2022 ).The concentration of DNA extracted from each sample was measured with Qubit Fluorometer 4.0 (Life Technologies Cor por ation, USA) using the Qubit DNA HS Assa y Kit (In vitrogen, T hermo Fisher Scientific , Austria).Target 16S rRNA gene amplicon sequencing was performed in an external laboratory (Microsynth, Switzerland), which also completed demultiplexing, trimming of ada ptors, and r eads mer ging.Briefly, the V3-V4 hyperv ariable r egions of the bacterial 16S rRNA gene were amplified with primers 341F-ill (5 -CCTA CGGGNGGCWGCA G-3 ) and 802Rill (5 -GA CTA CHV GGGTATCTAATCC-3 ), with an expected product of ∼460 bp (Klindworth et al. 2013 ).Barcodes and Illumina adaptors were added with 16S Nextera two-step PCR for library prepar ation.Finall y, samples wer e distributed in equimolar pools that were sequenced using a 250-bp paired-end reads protocol for Illumina MiSeq sequencing platform.Merged reads were inspected for quality using FASTQC (Andr e ws and Babr aham Bioinformatics 2010 ) and were further analyzed with software QIIME 2 (v.2020.2) (Bolyen et al. 2019 ).Sequences were filtered for quality (PHRED score 20) and subsequently denoised with deblur (Amir et al. 2017 ).Reads were trimmed at 400 nucleotides for SAM and LAM samples, and at 385 nucleotides for EAM samples.Denoising caused the loss of two samples for LAM ( n = 87) and of one sample for EAM.Furthermore, the latter matrix had three samples below 1000 r eads, whic h wer e discarded ( n = 86).All SAM samples passed the quality filtering and denoising ( n = 90).The output tables were filtered to exclude mitochondrial contamination before taxonomy was assigned with a Naive Bayes classifier trained for the specific 16S rRNA gene tar get r egions a gainst the SILVA 132 99% OTU r efer ence database.To calculate alpha diversity, datasets were rarefied to the lo w est read count that would allow to k ee p the maximum n umber of samples with a Good's cov er a ge index above 0.90 (9002 reads for SAM, five samples discarded, n = 85; 8324 reads for LAM, four samples discarded, n = 83; 6526 reads for EAM, six samples discarded, n = 80) ( Table S1 , Supporting Information ).Diversity was calculated with "diversity core-metrics-phylogenetic" function in QIIME 2. Beta diversity was calculated in R, using phyloseq pac ka ge (1.42.0) (McMurdie and Holmes 2013 ) for weighted and unweighted UniFrac and vegan pac ka ge (2.6.4) (Oksanen et al. 2020 ) for Aitchison distance (based on CLR transformation) (Aitchison et al. 2000 ).

RN A extr action, re verse tr anscription, and qPCR
RN A w as extracted using RNeasy Mini Qiacube Kit (Qiagen) with some minor modifications.About 25 mg of papillae were mixed with 350 μl RLT buffer in 2 ml safe lock tubes containing 0.6 g of ceramic beads.After homogenization in a Fastprep-24 instrument (MP Biomedicals, USA), samples were centrifuged at 10 000 × g for 1 min.The lysate was transferred to a 2-ml tube and centrifuged again at 14 680 × g for 3 min.The following steps were performed as described by the manufacturers, but with centrifugation times incr eased fr om 15 to 30 s.After the final addition of 500 μl of buffer RPE, the column was centrifuged at full speed for 1 min to dry the membrane .T he filter was placed into a new 1.5 ml tube and was left to dry for 1 min.Finally, 30 μl of RNase-free water were added to the filter and incubated for 1 min.RN A w as eluted b y centrifugation at 10 000 g for 1 min and stored at −20 • C. Genomic DNA was r emov ed using DN Ase I (Ambion ® TURBO DN A free), then RN A integrity w as assessed using the Qubit RNA IQ Assay Kit and extracted RN A w as quantified with the Qubit RNA HS Assay Kit (In vitrogen, T hermo Fisher Scientific) on the Qubit Fluorometer 4.0 (Life Technologies Corporation).Reverse transcription was performed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems™, Thermo Fisher Scientific, USA), with the addition of 500 μl of RNAse inhibitor (20 000 U, Biozym, Austria) in a thermocycler (Nexus, Eppendorf, Germany) with the following conditions: 25 • C for 10 min, 37 • C for 2 h, 85 • C for 5 min, and finally 4 • C. Per each sample, 10 μl of template were mixed with 10 μl of 2x mastermix (2 μl of 10x buffer, 0.8 μl of 25x dNTP mix (100 mM), 2 μl of 10x RT random primers, 0.5 μl RNAse inhibitor, 1 μl Mul-tiScribe RT, and 3.7 μl H 2 O, for a total volume of 10 μl), to which were added 80 μl of RNAse free water, to r eac h a concentr ation of 10 ng μl −1 .Gene expression analyses were performed using CFX96 Touch Real-Time PCR Detection System (BioRad, USA), with 10 μl reaction mix (2 μl cDNA as template, 0.8 μl of 100 nM primers, r e v erse and forw ar d r espectiv el y, and Biozym Blue S'Gr een qPCR master mix).Thermocycler conditions were set for 3 min at 95 • C for initial denaturation, follo w ed b y 40 cycles of 95 • C for 5 s and 60 • C for 30 s. Finally, samples were brought to 95 • C for 1 min and to 55 • C for 5 s; melt curv e anal ysis was set to increment of 0.5 • C c ycle −1 .P er each gene, samples were run in two technical replicates and the mean Ct value was used for further calculations.Negativ e contr ol and r e v erse tr anscription contr ol (minus R T) w ere included in each assay.To normalize for mRNA content, hypoxanthine phosphoribosyltr ansfer ase 1 ( HPRT1 ) and tyr osine 3-monoo xygenase/tryptophan 5-monoo xygenase acti v ation pr otein zeta ( YWHAZ ) were used as housek ee ping genes.Primers for the tested genes ar e r eported in Table 1 .Primer design was performed for IL-6 , TLR-4 , MyD88 , EGFR , IGF-1 , HPRT1 , and YWHAZ with Primer3Plus (version: 3.2.6)(Untergasser et al. 2012 ), targeting cDNA regions spanning between two exons (when possible) based on published cow sequences [Ensembl, Genome assembly: ARS-UCD1.2(GCA_002263795.2)].Gradient qPCRs were run per eac h ne w primer pair for estimation of optimal annealing temperatur e and v alidation of primer specificity by melting curv e anal ysis.

Metabolomics analyses
Metabolite profiling was performed as described in Ricci et al. ( 2022 ).For determination of carboxylic acids, sugar phosphates and sugars, 20 μl aliquots of rumen fluid were shaken with 980 μl of acetonitrile/water (80:20, v/v) at 4 • C for 10 min, centrifuged at 14 350 × g for 10 min and the supernatants were diluted 10-fold with acetonitrile/water (20:80, v/v).Analysis was performed by anion exc hange c hr omatogr a phy on a Dionex Integrion HPIC system coupled to a Q Exactiv e Orbitr a p mass spectrometer (both Thermo Scientific).Compounds were quantified based on external calibration curves established between 3 and 9000 ng ml −1 .Apparent recoveries (determined by comparing peak areas of 13 C-labeled internal standards of acetic acid, propionic acid, and butyric acid This study added prior to work-up with peak areas in pure solvent solutions containing the same concentrations) were close to 100%.Biogenic amines were determined by high-performance liquid c hr omatogr a phy coupled to tandem mass spectrometry (LC-MS/MS) after derivatization with phenyl isothioc y anate.Sample pr epar ation was carried out in 96-well plates and LC-MS/MS analysis was performed on an Agilent 1290 series UHPLC system (Ag ilent Technolog ies, Waldbronn, Germany) coupled to a SCIEX 6500 + QTr a p mass spectr ometer equipped with a Turbo V ion source (SCIEX, Foster City, CA, USA) as outlined in Ricci et al. ( 2022 ).Biogenic amines were quantified on the basis of external calibr ation curv es (0.6-1000 ng ml −1 in measur ement solution) pr epar ed on the same plate. 13C-putrescin was added to e v ery sample and standard solution prior to work-up and was used as internal standard for r ecov ery determination.An in-house prepared sample was worked-up and measured three times on each sample pr epar ation day and serv ed as inter-and intr aday quality control sample.Metabolome data for integr ativ e anal yses wer e normalized by row sums and P ar eto scaling.Missing values and zer os wer e r eplaced by the half of the minim um detection limit per each compound.

Histology and imm unohistoc hemistry
P a pillae biopsies for histology were fixed in neutral buffered formalin (4% v/v) and then tr ansferr ed to 70% ethanol and cleared in xylene before embedding in paraffin wax.Histological sections, 3μm thic k, wer e stained with DeadEnd™ Colorimetric TUNEL System (Promega Italia Srl, Italy) for the e v aluation of cellular apoptosis .For each sample , three papillae were evaluated and count of the immunolabeled cells was performed on thr ee micr oscopic fields (Leica DM2500 micr oscope, German y).The same samples wer e imm unostained with monoclonal mouse antihuman Cytokeratin antibody, clone AE1/AE3 (diluted 1:200, Agilent, USA), to e v aluate the thickness of the keratin la yer.T his evaluation was performed on three papillae per each sample, on three microscopic fields (the apex of each selected papilla and in both lateral sides).The same three papillae were considered for the measurement of the stratum corneum thickness.

Sta tistical anal yses
Alpha diversity, metabolite profiles, and histology data were analyzed in SAS (v.9.4).Normality was checked with PROC UNIVARI-ATE and the PROC REG pr ocedur e was used to calculate a linear r egr ession as well as Cook' s distance (Cook' s D).V alues were considered outliers with a Cook's D above 0.08.A linear mixed model was run with the PROC MIXED pr ocedur e, with cow, run, treatment, diet within week, and the interaction between diet and treatment within week as fixed effects.Cow within run and group wer e r andom effects.Measur ements taken on the same cow at different time points were considered as repeated measure, and post hoc Tuk e y corr ection for P -v alues was a pplied.Nonr ar efied feature tables were used to compute differential abundance analysis with Micr obiome Multiv ariable Associations with Linear Models (MaAsLin2) pac ka ge (1.7.3) in R (Mallick et al. 2021 ).Differential abundance was calculated using Centered Log-Ratio (CLR) normalization and LM method, while False Discovery Rate was calculated with Benjamini-Hoc hber g method (Benjamini and Hoc hber g 1995 ).The model was run with diet, treatment, and week as fixed effects and individual animal and experimental run as random effects.To assess the differential abundance between consecutive weeks in HG, the analysis was repeated on subsets of data considering only the 4 weeks in HG for the control and the PFA groups.
The model was run with week as categorical fixed effect, individual animal and experimental run as random effects, and changing the r efer ence le v el in order to assess the changes between consecutiv e weeks.Nonpar ametric MANO VA (PERMANO VA) was used to analyze beta diversity (Anderson 2001 ), through adonis function of vegan package (Oksanen et al. 2020 ).Differences were tested for fixed effects of diet, treatment, week, and their interaction.Functional prediction for the microbiota data was performed through Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2 (PICRUSt2), run using the QIIME2 plugin (v.2019.10), with the default options (av er a ge NSTI was 0.26 for SAM samples, 0.28 for LAM samples, and 0.24 for EAM samples) (Bolyen et al. 2019, Douglas et al. 2020 ).
Principal components analysis (PCA) of metabolites normalized counts was performed to identify the major responsible for distribution variation between the weeks through package stats (4.2.2) (Millard 2013 ).
For gene expression data, one sample from Run 2 was excluded because of technical issues.Rosner's test was applied on the Ct values to detect possible outliers (rosnerTest function of EnvStats pac ka ge, 2.7.0) (Millard 2013 ), and a total of three observations wer e r emov ed fr om the dataset (for EGFR , IFN-γ , and TLR-2 ).Gene r elativ e expr ession was calculated a ppl ying the Ct method (Pfaffl 2001 ), using two r efer ence genes ( HPRT1 and YWHAZ ) to calculate the Ct.The external calibrator was the mean Ct value of eac h gr oup (tr eatment or contr ol) on the first week (baseline) per each run, to normalize data considering the cr ossov er design (Petri et al. 2019 ).Lastl y, the r elativ e expr ession was calculated as value = 2 − Ct .The obtained values per each run were then merged into a single dataset on which statistical analyses were performed.A further statistical test was performed on the relativ e expr ession v alues to identify possible outliers, using the PROC REG pr ocedur e of SAS to calculate Cook's D. Values with a Cook's D abov e 0.10 wer e r emov ed fr om the dataset for downstr eam anal yses (one observation removed from each gene, with the exception of IFN-γ , TLR-4 , IGF-1 , IL-6 , and IL-10 for which two observations wer e r emov ed, and ZO-1 for whic h four v alues wer e r emov ed).The linear mixed model was run with PROC MIXED using the same fixed and random effects described above and with post hoc Tuk e y correction for P -values.Significance was considered for P ≤ .05 and tendencies were discussed for .05< P ≤ .10.
Data integration between microbiota, metabolome and gene expression was performed using unsupervised and supervised appr oac hes.For the first a ppr oac h, normalized counts of the identified metabolites were used to compute canonical correspondence analysis (CCA) between metabolites and microbiome data for LAM and SAM and EAM using v egan pac ka ge.Significance was tested with function ano va.cca().T he supervised a ppr oac h was implemented with the function block.splsda() of pac ka ge MixOmics (Rohart et al. 2017 ), which performs a m ultibloc k sPLS-DA (Sparse Partial Least Squares Discriminant Analysis).Before running the analysis, the microbial datasets were filtered to retain features with a relative abundance above 0.01% across the whole dataset and normalized with CLR transformation.The normalized metabolome and gene expr ession wer e anal yzed in combination with LAM, SAM, and EAM separ atel y and with two different response variables (treatment and week), resulting in six different models (three for each explanatory variable).The optimization of the number of components was obtained through perf() function with 10-fold cr oss-v alidation and 50 r e petitions while the n umber of features to retain was obtained with the tune.block.splsda()function with 3-fold cr oss-v alidation and 50 r epetitions .T he results of the analysis were evaluated to establish the correlations between datasets by comparing the coefficients for each of the components of each model.The loadings per each component were also evaluated to identify the most discriminant variables.The ASVs are discussed if they resulted among the most influential loadings in at least two of the model components.Rele v ance netw orks w ere obtained with function netw ork() with a cutoff of 0.7 and implemented for visualization using pac ka ges igr a ph and ggr a ph (Csárdi andNepusz 2006 , Pedersen 2022 ).Other gr a phs wer e pr oduced with pac ka ge ggplot2 (Wic kham 2016 ).
All alpha diversity indices were affected by the diet in LAM and SAM samples (Table 2 ; Table S1 , Supporting Information ).In LAM samples, all the alpha diversity indices decreased over the first 2 weeks of HG, and significantly increased in the last 2 weeks of experiment.The same pattern was visible for the SAM samples, with the lo w est v alues r eac hed on the second week of HG, and v alues numericall y incr easing in the last week.In EAM samples, Pielou's e v enness and Faith's Phylogenetic Diversity were affected by the diet ( P = .05and P = .03,r espectiv el y).Shannon index tended to be higher in EAM in the PFA group ( P = .10).Both weighted and unw eighted UniF rac distances w ere affected b y the diet and the experimental week in all the ruminal micr oenvir onments ( P < .01 for both in LAM and SAM and unweighted UniFrac in EAM; P < .01 and P = .03for diet and week, r espectiv el y, for weighted UniFrac in EAM).The Aitchison distance matrix results were in agreement with the canonical methods (Fig. 1 ).The phytogenic treatment did not affect any beta diversity matrices.

Microbiota composition and differential abundance in response to the dietary challenge
The dietary challenge caused major changes in the microbiota composition in all the three niches at each taxonomic level evaluated (phylum, family, and genus).The shifts in phyla r elativ e abundances ar e pr esented in Figur e S1 ( Supporting Information ).In EAM samples, phylum Euryarc haeota (Arc haea) tended to increase due to the diet (0.7% in forage and 1.3% in HG, P < .01)( Figure S1 , Supporting Information ).The three ruminal microenvir onments r eacted differ entl y to the HG diet, with 42, 71, and 18 families significantly affected in the SAM, LAM, and EAM samples, r espectiv el y.The thr ee most abundant families in rumen digesta (both SAM and LAM) were Lachnospiraceae , Ruminococcaceae , and Prevotellaceae , and were all increased from forage to HG ( P < .01).The most frequent families over all the EAM samples were Lachnospiraceae (31.0%),Ruminococcaceae (11.1%),Clostridiales Family XIII (9.5%), and Campylobacteraceae (7.5%).Families Prevotellaceae (2.5% in for a ge and 6.4% in HG, P < .01)and Clostridiales Family XIII (13.5% in for a ge and 8.5% in HG, P = .03)w ere affected b y the diet.
In SAM samples, 461 genera were identified, of which 22 had a r elativ e fr equency abov e 1%.The most abundant gener a ar e pr esented in Fig. 2 (A).In LAM samples, onl y 21 gener a had an ov er all r elativ e fr equency abov e 1% (Fig. 2 B).The most fr equent gener a over all the EAM samples were classified as Butyrivibrio 2 (11.9%),Campylobacter (7.5%), [ Eubacterium ] nodatum group (3.4%), Desulfobulbus (3.1%), and Ruminococcaceae NK4A214 group (3.0%) (Fig. 2 C).The dietary shift from forage to HG affected 126 and 154 genera in SAM and LAM samples, r espectiv el y, while onl y 51 gener a wer e differ entiall y abundant due to the HG diet in EAM samples (Fig. 3 A).Only 33 genera were affected by the HG diet in all three niches (Fig. 3 B).
When considering the effect of the duration of the HG dietary challenge 83, 74, and 28 genera were found to be differentially abundant in SAM, LAM, and in EAM samples, r espectiv el y.Among the most abundant genera (Fig. 2 ), Succinivibrionaceae UCG-001 was affected by the prolonged HG feeding in all three niches ( P < .01 in SAM, P = .10 in LAM, and P = .05 in EAM).Genus Lachnospiraceae NK3A20 group was differentially abundant in the experimental weeks in both SAM and LAM samples ( P = .08and P < .01,respectiv el y).Gener a Ruminococcus 2 and Acetitomaculum were significantly affected by the weeks in LAM samples ( P < .01),while Candidatus saccharimonas tended to be affected by the weeks only in EAM samples ( P = .07).Genus Kandleria was significantly affected by the weeks in all three matrices, reaching the highest concentrations in the first week of HG and decreasing by the end of experiment ( P < .01 in SAM, LAM, and EAM).Genus Ruminococcaceae UCG-002 tended to decrease at the beginning of the HG challenge in all three niches, but reached the highest relative abundance by the end of experiment in SAM and LAM samples ( P < .01),and on week 3 HG in EAM samples ( P = .08).On the contr ary, Lac hnobacterium increased in all three niches in the first week of HG feeding and tended to decrease by the end of the HG challenge in LAM and EAM samples ( P < .01).In SAM samples, Lachnobacterium decreased betw een w eek 2 (0.013%) and 3 HG (0.009%), to incr ease a gain by week 4 HG (0.016%) ( P < .01).Genus Acidaminococcus showed different trends in each microenvironment analyzed.In SAM, it was not present in the first 2 weeks of experiment, and showed the highest r elativ e abundance in week 2 HG (0.47%, P < .01).In LAM, Acidaminococcus disa ppear ed fr om for a ge feeding to week 1 HG, and then incr eased a gain to r eac h the peak on week 3 HG (0.28%, P = .01).In EAM samples, the r elativ e abundance of Acidaminococcus tended to incr ease gr aduall y with the pr ogr ession of the dietary c hallenge, fr om a r elativ e abundance < 0.01% in the first week of HG feeding to r eac h a r elativ e abundance of 0.19% in week 4 HG ( P = .06).
When specifically analyzing the differential abundance between each consecutive HG week, the differences between the ruminal micr oenvir onments wer e e v en mor e deepened, with distinct genera affected in each niche (Fig. 3 C).Genera Schwartzia and Prevotella 7 were the only two taxa with a significant differential abundance between weeks 1, 2, and 3 HG in all the three niches, r eac hing the highest r elativ e fr equenc y on w eek 3 HG in all the micr oenvir onments anal yzed.In LAM, famil y Bacillaceae was significantl y r educed due to the dur ation of the HG diet.LAM samples sho w ed a higher reactivity betw een w eek 3 HG and w eek 4 HG compared to the other two ruminal nic hes anal yzed, with higher pr oportions of differ entiall y abundant gener a in both the control and PFA groups (Fig. 3 C).The microbiota attached to the rumen w all sho w ed the most stable composition over the 4 weeks of HG diet, with only a few genera significantly different between each week, and only in the control group (Fig. 3 C).

Effect of the PFA on the ruminal microbiota
The PFA supplementation in SAM samples increased the overall r elativ e fr equency of uncultur ed P eptococcaceae (0.06% in PFA and 0.03% in control, P = .04),Desulfuromonas (0.013% in PFA and 0.008% in control, P = .02)and Desulfovibrio (0.18% in PFA and 0.11% in control, P = .06).The treatment tended to decrease the ov er all r elativ e fr equency of genus Coprococcus 2 (0.13% in PFA and   0.23% in control, P = .08),Lachnospiraceae NC2004 group (0.05% in PFA and 0.06% in control, P = .09),Lachnospiraceae UCG-001 (0.06% in PFA and 0.12% in control, P = .06),and Ruminococcus 1 (3.1% in PFA and 3.9% in control, P = .06).The PFA also decreased the abundance of [ Eubacterium ] xylanophilum group (0.12% in PFA and 0.21% in control, P < .01)and Pseudobutyrivibrio (0.25% in PFA and 0.35% in control, P < .01).The PFA supplementation affected 13 predicted pathways in SAM samples ( Figure S2 , Supporting Information ).Ov er all, mor e taxa wer e affected by the pr olonged dietary c hallenge in the control group compared to the PFA, indicating a more stable composition over the 4 weeks of HG feeding (Fig. 3 C).No effect was recorded in the PFA group in the transition between week 3 and 4 HG.
In LAM samples, Lachnospiraceae NK3A20 group tended to increase in the PFA group (5.7%) compared to control (4.5%) ( P = .07).Similarly, Ruminiclostridium 9 (0.3% in PFA and 0.1% in control, P < .01),C. saccharimonas (1.3% in PFA and 0.10% in control, P = .05),and uncultur ed P eptococcaceae (0.03% in PFA and 0.01% in control, P = .06)were more frequent in the PFA group.[ Eubacterium ] xylanophilum group (0.14% in PFA and 0.17% in control, P = .10),and Lachnospiraceae NC2004 group (0.03% in PFA and 0.04% in control, P = .04)were more frequent in the control group.The PFA supplementation also affected 24 predicted pathways in LAM samples ( Figure S2 , Supporting Information ).Like for the SAM, in LAM samples the PFA group sho w ed a more stable composition compared to the control group, with less genera differentially abundant between the four consecutive weeks in HG (Fig. 3 C).
The PFA supplementation did not affect the microbiota composition nor the predicted activity in the EAM samples.

Metabolite profiling
All the metabolites measured in the rumen fluid, a part fr om carnitine, ethylbutyric acid, pyroglutamate, αd -glucose-1-phosphate, and hydr oxyphen yl-pr opionic acid, wer e affected either by the diet or by the interaction between the treatment and the diet (Table 3 ).The variation of metabolite concentration over the weeks is shown in Fig. 4 .The PFA maintained or tended to maintain a stable lo w er le v el of dopamine ( P = .08),kynurenine ( P = .01),glyceric acid ( P = .02),and benzoic acid ( P = .03)compared to the contr ol.Ther e was an interaction between the treatment and the diet for six of the analyzed metabolites: 5-aminovaleric acid (5-A V A) ( P = .02),β-aminobutyric acid (B AB A) ( P = .04),creatine ( P = .09),phenylethylamine ( P = .07),methylbutyric acid ( P = .01),and succinic acid ( P = .09)sho w ed a gr eat v ariation of concentration in the different weeks.In particular, the PFA supplementation seemed to stabilize the concentrations of 5-A V A, kynurenine and succinic acid over the experimental weeks (Fig. 5 ).PCA confirmed the three major VFAs (acetic , propionic , and butyric acid) as major responsible for the metabolite distribution variation between the weeks ( Figure S3 , Supporting Information ).CCA performed between normalized counts of these three VFAs and the microbiome data revealed a significant impact for all thr ee nic hes ( P < .01).Significance was tested also per each VFA, revealing a significant influ-   ) showing the number of genera differentially abundant due to the HG diet in each ruminal niche analyzed (SAM, LAM, and EAM).Mean r elativ e abundances (B) in for a ge and in HG feeding of the 33 gener a that were affected by the diet in all three ruminal niches.(C) Number of genera differentially abundant during the prolonged HG feeding challenge in each ruminal niche analyzed (SAM, LAM, and EAM).The pr oportion of significantl y differ ent gener a between eac h HG feeding week is r epr esented for the contr ol gr oup and for the tr eatment gr oup (PFA).ence of each acid, despite the low variance explained ( P < .01,P = .01,and P = .03for propionic , acetic , and butyric acid, respectiv el y in SAM; P < .01,P < .01,and P = .03for propionic, acetic and butyric acid, r espectiv el y in LAM; P = .03,P = .03,and P = .01for propionic , acetic , and butyric acid, r espectiv el y in EAM) (Fig. 6 ).
All epithelial gr owth-r elated genes wer e affected by the prolonged HG feeding (Fig. 7 ).Insulin-like growth factor 1 ( IGF-1 ) increased between week 2 HG and week 3 HG ( P = .01)and tended to be higher in the control group ( P = .06).There was a trend for interaction between diet and PFA treatment for epidermal growth factor receptor ( EGFR , P = .10),while F igure 4. Heatmap sho wing the c hange in concentr ation of metabolites measur ed in the rumen fluid acr oss the fiv e experimental w eeks.Ro ws w ere scaled to have mean zero and standard deviation one, to normalize the different concentrations of each metabolite.Values range from −1 (lo w est concentr ation measur ed) to 1 (highest concentr ation measur ed).The hier arc hical clustering shows a separ ation between the baseline (For a ge) and HG feeding weeks (1-4 HG).
both EGFR and insulin-like growth factor 1 receptor ( IGF-1R ) tended to be affected by the diet ( P = .08and P = .06,respectiv el y).Tata-box binding protein ( TBP ) was affected by the diet ( P < .01).
Statistical anal yses r e v ealed no effects on the histological parameters evaluated, apart from a tendency for an ef-fect of the individual cow on the thickness of the stratum corneum ( P = .10)( Table S2 , Figures S4 and S5 , Supporting Information ).

Da ta integr a tion anal ysis
The m ultibloc k sPLS-DA r e v ealed high corr elations between microbiota and metabolome, regardless of the chosen explanatory v ariable (tr eatment or week) ( Tables S3 and S4 , Supporting Information ).The strongest correlation was observed for LAM for both explanatory variables .T his micr obial nic he also sho w ed the lo w est correlation with the gene expression data.In general, the gene expression data sho w ed overall lo w er correlation coefficients with the other datasets.
Both the loadings and the r ele v ance networks obtained per eac h micr obial dataset and the two r esponse v ariables wer e thoroughly examined to find the k e y discriminant features .T he loadings for each model are presented in Figures S6 -S8 S5 ( Supporting Information ).

sPLS-DA results for the weeks
In the model considering SAM and weeks as explanatory variable, only eight ASVs were identified in two components .T hree were classified as Prevotella 1, and other two as Prevotellaceae UCG-003 and UCG-004.In the model for LAM and weeks, 15 ASVs were identified in more than one component, of which five were belonging to family Prevotellaceae .For EAM, 16 ASVs were found in more than one component, and two were classified as Prevotella .ASVs classified as Rikenellaceae RC9 gut group and Christensenellaceae R-7 group were found to be among the most influential in EAM and in SAM and LAM, r espectiv el y.
Genes CLDN4 , DSG1 , IGF-1 , IL-6 , MyD88 , and TBP were among the most influential discriminants betw een w eeks for SAM.Similarly, genes DSG1 , IGF-1 , IL-6 , and MyD88 were among the most influential in the model considering LAM.Other influential genes for this model were ZO-1 and IGF-1R .In the model considering EAM, most of the genes wer e consider ed discriminant in se v er al components of the model, while genes TLR-2 and MyD88 were considered the most influential.
For the metabolome data in the models considering week as explanatory variable, the most influential compounds for SAM wer e succinic acid, phen ylpr opionic acid, kynur enine, hexanoic acid, d -glucose-6-phosphate , d -fructose-6-phosphate , and butyric acid.When LAM was included in the model, the metabolites that contributed the most to the differences between gr oups wer e pr opionic acid and histamine.For EAM, the most influential metabolites were identified as hexanoic acid, d -galacturonic acid, and butyric acid.
The r ele v ance netw orks obtained using w eek as explanatory v ariable r einfor ced the findings obtained b y confronting the loadings per each model.Only a few correlations were found for genes related with inflammation although some inflammatory genes ( IFN-γ , IL-10 , and TNF-α) were consistently correlated in all the networks.Similarly, TLR-4 , ZO-1 , and IGF-1 wer e negativ el y corr elated in all three networks.Both IGF-1 and ZO-1 were highly correlated with many other variables.Glucose and disaccharides were positiv el y corr elated with EGFR and IGF-1 in the r ele v ance networks obtained for SAM and EAM, r espectiv el y.Acetic acid was negativ el y corr elated with IGF-1 in all the networks.d -glucose-6-phosphate and d -fructose-6-phosphate were present in all the r ele v ance networks and highly correlated with CLDN4 , IFN-γ , IL-10 , and TNF-α.Three ASVs classified as Ruminococcaceae were positiv el y corr elated with d -glucose-6-phosphate and d -fructose-6phosphate, IL-10 and TNF-α in the network for SAM.Benzoic acid, iso-butyric acid, iso-valeric acid, phenylacetic acid, and acetic acid wer e also pr esent and highl y corr elated in all thr ee networks and wer e consistentl y associated with epithelial barrier or epithelial gr owth genes (especiall y ZO-1 and IGF-1 ).Pr opionic acid was identified as major component of the r ele v ance network only in LAM, showing correlations with 36 ASVs.Only ASV_6805 ( Solobacterium ), and ASV_9264 ( Ruminococcaceae UCG-014) wer e shar ed between LAM and SAM networks, the latter showing a strong correlation with d -glucose-6-phosphate and d -fructose-6-phosphate in both datasets.

sPLS-DA results for the PFA
The sPLS-DA was unable to consistently identify microbial featur es r esponsible for the v ariation among tr eatment and contr ol groups in both SAM and LAM.For EAM, six ASVs were identified as main drivers of difference ( Figure S8, Supporting Information ).
In the model considering SAM, ZO-1 was identified by se v er al components as discriminant feature for the PFA.Other discriminant genes for this niche were MyD88 , IGF-1 , IGF-1R , and CLDN4 .Similarly, when LAM was included in the model, ZO-1 was identified among the most important genes for discrimination among groups, together with IL-10 , TLR-4 , MyD88 , IGF-1 , EGFR , and CD14 .When EAM was considered in the model, most of the genes were identified as discriminant in more than five components .T he highest contributions were found for MyD88 and DSG1 .
For the models considering the PFA as explanatory variable and SAM, the highest contribution for the metabolome was given by glyceric acid and creatine, while considering LAM, succinic acid, spermine, and glyceric acid were the most discriminant metabolites.When considering EAM in the model, only few metabolites were found to be main drivers of difference in more than one component.The highest contributions were given by αand β-aminobutyric acid, hydr oxyphenilpr opionic acid, and dgalacturonic acid.
Similar patterns were identified between the r ele v ance networks based on week or treatment explanatory variables.Genes related to epithelial growth or barrier integrity showed stronger corr elations compar ed to the inflammation r elated genes.Gene MyD88 sho w ed positiv e corr elations with cr eatine in the net-F igure 7. Boxplot sho wing the r elativ e expr ession of genes r elated with barrier function, epithelial gr owth, and cellular activity in the rumen epithelium of cows fed a for a ge-based diet (For a ge) and a HG diet for 4 weeks.Results are presented for the control and the treatment (PFA) groups per each experimental week.Significant results are presented in each graph for the overall effect of the diet within week (Diet), of the phytogenic treatment (PFA), and of the interaction between treatment within week and diet (I).P -values are also presented for the significant pairwise comparisons between consecutive weeks and between control and PFA treatment within each week.
F igure 8. Boxplot sho wing the r elativ e expr ession of genes r elated with inflammation function in the rumen epithelium of cows fed a for a ge-based diet (For a ge) and a HG diet for 4 weeks.Results ar e pr esented for the contr ol and the tr eatment (PFA) gr oups per eac h experimental week.Significant r esults ar e pr esented in eac h gr a ph for the ov er all effect of the diet within week (Diet), of the phytogenic tr eatment (PFA), and of the inter action between treatment within week and diet (I).P -values are also presented for the significant pairwise comparisons between consecutive weeks and between control and PFA treatment within each week.works for EAM and SAM and was correlated with se v er al ASVs in all three networks.For SAM, positive correlations were found with ASV_4906, ASV_4905, ASV_4904 (all classified as Prevotella 1), and ASV_9384 ( Ruminococcaceae UCG-014).Similarly, in LAM netw ork MyD88 w as positiv el y corr elated with ASV_5021 ( Prevotella 1), ASV_7959, and ASV_7930 (both classified as Ruminococcaceae UCG-010).In EAM, MyD88 sho w ed correlations with ASV_3832, ASV_3239, ASV_3833 ( Lachnospiraceae ), and ASV_2865 ( Ruminococcaceae CAG-352).
TLR-2 in the network for EAM was positiv el y associated with γ -and α-aminobutyric acid, which was in turn correlated with ASV_1665 ( Rikenellaceae RC9 gut group).In the same network, βaminobutyric acid was negativ el y corr elated with IGF-1 and with 10 ASVs.ASV_4665 ( Prevotella 1) was found both in SAM and LAM, but str ongl y corr elated with IGF-1R in the former and with glyceric acid and CLDN4 in the latter.In SAM, spermine was positiv el y correlated with ASV_4532 ( Prevotellaceae UCG-004) and ASV_9725 ( Acetitomaculum ).In LAM, positive correlations for spermine were found with IFN-γ , TNF-α, ASV_4930 ( Prevotella 1), and ASV_7116 ( C. saccharimonas ).

Impact of HG feeding on the ruminal microbiota composition and activity
We investigated the microbiota composition in three ruminal micr oenvir onments, whic h hav e been pr e viousl y demonstr ated to host distinct microbial populations (De Mulder et al. 2016, Wang et al. 2017 ).As expected, due to HG feeding, alpha diversity was reduced in all the three ruminal niches, with possible implications on impaired animal health (Hook et al. 2011, Plaizier et al. 2017 ).Our study has also confirmed pr e vious findings, in whic h the epim ur al associated micr obiota was less r esponsiv e to dietary changes (Mann et al. 2018 ).In fact, in addition to the lo w er number of taxa affected by the HG diet in this niche, the relevance network for the weeks found few significant correlations between EAM and the other datasets.
In contrast to a pr e vious study, we found a significant impact of the diet on the Archaeal population in the ruminal papillae (Xue et al. 2019 ).Other studies have reported that methane production is dependent on the percentage and type of concentrates in the diet, suggesting that the increased relative frequency of Euryarchaeota could be explained by the composition of our diet (Hook et al. 2010, Moate et al. 2017 ).Ho w e v er, it also needs to be considered that 16S rRNA gene amplicon sequencing is recognized as a less accurate way to assess the presence of Archaea in the rumen (De Mulder et al. 2016 ).
According to the unsupervised data integr ation anal ysis, the main metabolites responsible for the variation in metabolome composition as well as of the rumen contents microbiota composition wer e acetic, pr opionic, and butyric acid.These VFAs are normally the most concentrated in the rumen, and are considered the main microbial fermentation products (Aschenbach et al. 2011 ).The supervised integr ativ e anal yses confirmed pr opionic and butyric acid as main discriminant of difference between experimental weeks in SAM and especially in LAM.
Ho w e v er, VFAs wer e not the onl y metabolites identified as k e y discriminant by the sPLS-DA.Microbial fermentation, especially in case of dysbiosis, can produce harmful compounds, such as biogenic amines.Of the 20 biogenic amines identified in the present stud y, kyn ur enine, phen ylethylamine, and histamine wer e among the most discriminant features for the HG challenge, although we did not find significant correlations between these metabolites and other variables.Some of these compounds, such as histamine and ethanolamine , ha v e been pr e viousl y r eported as incr eased in cows with SARA, and can contribute to the de v elopment of mor e se v er e and systemic symptoms (Nocek 1997, Ametaj et al. 2010 ).Statistical analyses confirmed that all biogenic amines in our study were affected by the diet, except for carnitine and creatine.
Ov er all, our r esults suggest a shift to w ar ds an imbalanced microbial composition and metabolism due to the HG diet, outlining a picture compatible with ruminal dysbiosis.Since the main drivers for the onset of this condition in the rumen seem to be not only VFAs, but also biogenic amines, such metabolites should be al ways e v aluated when inv estigating dysbiosis in cattle.

The effect of dysbiosis on the epithelial expression of inflammation related genes
Toll-lik e rece ptors , among others , ar e r eceptors r esponsible of the interaction with bacteria, and can promote tolerance as well as trigger an immune reaction and relative inflammation (Malmuthuge et al. 2012 ).TLR-4 recognizes LPS and triggers a reaction that depends on CD14 and results in the production of proinflammatory cytokines .T he expression of both genes has been demonstr ated to incr ease in cows with SARA (Stefanska et al. 2018 ).In accordance with this, CD14 r elativ e expr ession incr eased in parallel with TLR-4 in response to the HG diet.Inter estingl y, the r ele v ance netw orks sho w ed a close connection betw een TLR-4 and genes related to epithelial growth and structure ( IGF-1 and ZO-1 ), but not with CD14 .Such cluster was identified in all the networks and was composed of VFAs and of a different number of ASVs depending on the micr obial nic he, mostl y classified as Lachnospiraceae and Ruminococcaceae , as well as Prevotella 1, and Rikenellaceae RC9 gut gr oup.Inter estingl y, str ong corr elations wer e found with family Bacillaceae in LAM.Although the other families included in this cluster were increased in all the microbial niches due to the HG diet, Bacillaceae decreased in LAM.This family is reputed beneficial for the host and was positiv el y corr elated with butyric and iso-butyric acid, benzoic acid, and iso-valeric acid in the cluster.These organic acids have anti-inflammatory properties, which might explain the negative correlations observed with TLR-4 (Mentschel et al. 2001, Pu et al. 2018, Zhang et al. 2021 ).
After the activation due to CD14 and TLR-4 , the inflammation process can develop following two different pathwa ys .T he MyD88 dependent pathway activates the signaling cascade of NF-kB and consequently of TNF-α and other cytokines (Björkbacka et al. 2004 ).MyD88 was identified as one of the main drivers of difference between experimental weeks.Its expression was str ongl y correlated with ASVs belonging to family Prevotellaceae , and particularly Prevotella 1, indicating the major role play ed b y this Gramnegativ e famil y in the de v elopment of an inflammatory r esponse in relation to HG. Surprisingly, the increased expression of MyD88 due to the diet was not accompanied by an increased expression of interleukins in our study.This could be due to the fact that despite the incr eased expr ession, especiall y in some individuals during weeks 2 HG and 3 HG, by the last week of experiment the av er a ge cytokines r elativ e expr ession r eturned to le v els mor e similar to the for a ge feeding.The r elativ e expr ession of genes such as TLR-4 and TNF-α, as well as of other proinflammatory chemokines, was reduced after repeated exposure to LPS (Kent-Dennis et al. 2020 ).This might indicate the activation of a negative feedback loop and suggests a pr ogr essiv e mec hanism of toler ance to w ar ds LPS, which could explain the results of our experiment (Lv et al. 2017, Petri et al. 2019 ).In contrast to our results, a study with a shorter experimental period sho w ed a significant increment of proinflammatory cytokines (Zhang et al. 2016 ), confirming that the tolerance to w ar d LPS most likely occurs after over 4 weeks of exposure.It needs to be considered that the ov er all abundance of Prevotellaceae increased in the first 2 weeks of HG diet, to decrease again to w ar ds the end of experiment: this might also explain the expression of cytokines over the experimental weeks, given the high pr e v alence of this family in the ruminal environment and the strong association found with MyD88 .Furthermore, since the le v els of LPS were not measured in the present study, it needs to be considered that the concentration might have fluctuated over the weeks also in relation to other Gr am-negativ e bacteria, gener ating the observed response.
The response to LPS can be mediated also by other proteins, generating a MyD88 independent pathway (Björkbacka et al. 2004 ), whic h r esults in the production of Type I and Type II interferons ( IFN-γ ) (Mortellaro et al. 2015 , Lee andAshkar 2018 ).In contrast to the other cytokines, IFN-γ r elativ e expr ession was affected by the diet.Studies in vitro and in mouse models suggested that IFN-γ might be str ongl y corr elated to dysbiosis and that its expr ession is mor e stim ulated by Gr am-positiv e bacteria (Hessle et al. 2000, Bae et al. 2020 ).Ther efor e, although Gr am-negativ e bacteria and LPS play a widely recognized part in SARA pathogenesis, the role of Gram-positive bacteria in ruminal dysbiosis should be further in vestigated.T his seems to be confirmed also by the strong correlation between ASVs classified as Ruminococcus 2 and Ruminococcaceae UCG-014 and IL-10 and TNF-α, since both genes wer e str ongl y associated to IFN-γ in the r ele v ance networks.Notably, d -glucose-6-phosphate and d-fructose-6-phosphate were str ongl y associated with the same genes, suggesting that the epithelial inflammation associated with the HG diet might be indeed related to glucose metabolism.Glucose is metabolized in glucose-6-phosphate and subsequently in fructose-6-phosphate, and in fact the concentration in the rumen of both metabolites increased in week 2 HG.T hus , it seems to be the metabolization of glucose that triggers the inflammatory reaction of the rumen epithelium, rather than the increment of glucose concentration itself (Peiró et al. 2016 ).In fact, the activity of glucose-6-phosphate dehydrogenase, which is the rate-liming enzyme for the pentose phosphate pathway, seems to be closely related to the activation of the inflammasome in response to bacterial infections (Yen et al. 2020 ).Str ong positiv e corr elations between these metabolites and ASVs classified as C. saccharimonas , Ruminococcus 2, and Ruminococcaceae UCG-014 were found for both LAM and SAM.All three genera are mainl y amylol ytic, suggesting that the pr oduction of d -glucose-6phosphate and d -fructose-6-phosphate likely derived from the incr eased starc h degr adation during HG feeding (La Reau andSuen 2018 , Baker 2021 ).The role of glucose-6-phosphate dehydrogenase and glucose-6-phosphate in the onset of the inflammatory response of the ruminal epithelium warrants further research.

The rumen can adapt to the perturbation caused by long term HG feeding
The ada ptiv e mec hanism of pa pillae enlar gement to incr ease the absorbing surface is finely regulated by several receptors that bind a multitude of proteins.Among these, the insulin-like growth factor axis promotes tissue growth by triggering the production of protein kinases (Kaulfu β et al. 2009, Steele et al. 2012a ).This seems to be validated by our study, since HG feeding caused a higher expression of both IGF-1 and its receptor ( IGF-1R ).The upregulation of IGF-1 is associated with an increased uptake of glu-cose as well as with the pr olifer ation of rumen epithelial cells (Baldwin 1999, Shen et al. 2004, Steele et al. 2012b ) and the relevance networks showed correlations of glucose and disaccharides with EGFR and IGF-1 .Ther efor e, the incr eased concentr ation of glucose in the rumen, together with disaccharides, seems to be dir ectl y linked to epithelial growth.
The tendency of IGF-1R to increase in the last weeks of experiment might indicate a higher cellular metabolic activity to w ar ds the end of the trial.This higher activity of the rumen epithelial cells was further suggested by the increased expression of EGFR and TBP over the weeks.The latter gene is involved in the initiation of the transcription processes, suggesting a higher activity of the cells with the pr ogr ession of the experiment (Akhtar and Veenstr a 2011 ).Pr e vious studies hav e demonstr ated that some pr oteins can regulate the expression of TBP , although a relationship with the microbiota has never been investigated (Seto et al. 1992, Kerr et al. 1993 ).
Despite the alterations in epithelial gene expression that we observed due to the diet, and although previous studies documented the adaptations of rumen wall when the animals are fed a HG diet (Steele et al. 2011(Steele et al. , 2015 ) ), we did not find any modification of the par ameters e v aluated thr ough histology and imm unohistoc hemistry.It needs to be considered, ho w ever, that changes in mRNA expression do not necessarily result in translation into proteins, as ther e ar e numer ous post-tr anscriptional mec hanisms that could pr e v ent the proteins from being built (Greenbaum et al. 2003, Pacífico et al. 2022 ).Furthermore, since the samples for histology were collected only in the first and the last weeks of experiment, it is possible that the epithelium had already started r ecov ering, and ther efor e differ ences between for a ge and high-concentr ate feeding were not appreciable.
It has been demonstrated that bacterial metabolites can impair the function of the ruminal epithelial barrier (Gao et al. 2022 ).In our study se v er al biogenic amines of microbial origin, such as cada verine , dopamine , putrescine , and histamine , sho w ed a peak in concentration during the third week of HG.In the r ele v ance netw orks, dopamine w as positiv el y corr elated with TBP and spermine with IGF-1R , but no correlations with the barrier integrity genes were found.In fact, nor DSG1 nor CLDN-4 were affected by the diet and the only gene associated with barrier function whose expression was affected by the HG feeding was ZO-1 .This gene was consistently identified as an important discriminant for the HG challenge by the sPLS-DA analysis, regardless of the microbial niche included in the model.T herefore , the expression of ZO-1 in the rumen epithelium seems to be a strong indicator of the health status of the or gan.After r eac hing the lo w est r elativ e expr ession on week 2 HG, ZO-1 expression increased on the third week of HG feeding suggesting that the barrier integrity can be r estor ed e v en if impaired in the first weeks of a HG challenge .T he negativ e corr elation between IGF-1 and ZO-1 might imply a potential disruption of the epithelial barrier due to the excessiv e pa pillae gr o wth.Ho we v er, by the end of the experiment, the metabolic capacity of the micr obiota has ada pted to the le v els of starc h in the diet and can metabolize it faster, limiting the growth and helping to maintain the integrity of the barrier.In fact, our findings for the digesta microbiota seem to suggest that the rumen started adapting by the fourth week of continuous HG feeding.In particular, in LAM, that r epr esents the micr obial comm unity loosel y associated with feed particles (Tafaj et al. 2004 ), the beta diversity sho w ed a significant number of samples from the last week of experiment clustering together with the samples collected in for a ge feeding.This means that, in some individuals, the microbiota of this niche by the last week of experiment had switched to a composition more similar to the first week of experiment.In an analogous study that analyzed the effects of rhubarb powder on the rumen in cows fed a HG diet, the microbiota composition recovered by the end of the experiment, suggesting an adaptation of the micr oor ganisms to changing conditions (Wang et al. 2017 ).Furthermore, studies in beef cattle have demonstrated that the ruminal microbiota starts to shift to w ar d a stable composition b y the fourth w eek after a change in dietary regime, and that such modifications are stable over time (Clemmons et al. 2019, Snelling et al. 2019 ).It is interesting to notice that, as it is known from shorter studies, also in our prolonged dietary challenge the three ruminal microenvironments sho w ed v ery differ ent ada ptations, and the LAM micr obiota seemed to be the first to start the adaptation process (McCann et al. 2016, Ricci et al. 2022 ).The results of the multiblock sPLS-DA anal yses also r e v ealed a str ong corr elation between this ruminal niche and the rumen metabolome, highlighting the crucial role of LAM in the ruminal metabolic adaptation.
At the metabolome le v el, the c hanges in concentr ation fluctuated over the w eeks.Hierar chical clustering evidenced a similarity between the metabolite concentration on week 1 HG and week 3 HG, and between week 2 HG and week 4 HG.The variation in the metabolome in the first week of HG coincided with substantial changes in the microbiota composition, which indicate the instauration of ruminal dysbiosis, as discussed abo ve .Even though some metabolites and biogenic amines sho w ed lo w er concentrations during the second week of HG, after this time the system seemed to colla pse, a ggr av ating the dysbiosis.This breaking point coincided with a considerable shift in microbiota composition, as shown in the beta diversity graphs, and with the lo w est alpha div ersity v alues, as well as with the highest concentr ation of harmful compounds.Finally, in the last week of experiment, both the microbial composition and activity seemed to stabilize, with numerically higher diversity and richness indices and lo w er metabolite concentrations.All the evidence gathered suggests that the ruminal microbiota can restore its composition and activity during the course of a long term HG challenge, by optimizing the utilization of substrates introduced with continuous HG feeding, and at the same time preserving the integrity of the epithelium (Weimer 2015 ).

Effects of the phytogenic supplementation on the ruminal microbiota composition and the epithelial inflammatory response
The PFA supplementation contributed to maintain a stable microbiota composition of the epim ur al fr action during the HG c hallenge, while it affected specific taxa in the rumen digesta.[ Eubacterium ] xylanophilum gr oup, Lac hnospiraceae NC2004 group and uncultur ed P eptococcaceae wer e affected in a similar way both in LAM and SAM, but there is lack of research on the effects of phytogenic compounds on these taxa.Coprococcus 2 and Lachnospiraceae NC2004 gr oup, whic h wer e both r educed in the PFA gr oup, hav e been related to the metabolization of polyphenols (Patel et al. 1981, Burgos-Edw ar ds et al. 2018, Liu et al. 2020 ).[ Eubacterium ] xylanophilum group can metabolize only a few carbohydrates, and pr efer entiall y contributes to the degradation of hemicellulose (Van Gylswyk and Van Der Toorn 1985 ).The incr eased r elativ e fr equency of Peptococcaceae in the rumen contents might indicate the capacity of the PFA supplementation to preserve a physiological composition of the microbiota, since it has been demonstrated that this family is negatively affected by the presence of starch (Kheirandish et al. 2022 ).Furthermore, the lo w er number of taxa affected by the prolonged HG feeding in the PFA group compared to the control, confirms the potential of the phytogenic supplementation to pr eserv e a more stable microbiota composition in the rumen during a prolonged dietary challenge.
Ne v ertheless, the effects of the PFA supplementation on the r elativ e expr ession of some genes in our study might be the result of shifts in microbiota composition due to the treatment.Inter estingl y, the PFA supplementation decreased the teichoic acid biosynthesis predicted pathway (TEICHOICACID-PWY) as well as the r elativ e fr equency of Gr am-positiv e bacteria suc h as Coprococcus 2, Ruminococcus 1, and [ Eubacterium ] xylanophilum group.This was pr obabl y r elated to the lo w er expression of TLR-2 in the PFA group.In fact, TLR-2 reacts to Gram-positive bacteria, but its role in the interaction with the pr olifer ation of specific taxa especially in HG feeding regimes requires further research (Takeuchi et al. 1999, Petri et al. 2019 ).Our integr ativ e anal ysis did not show an y corr elations between this gene and specific ASVs, although it r e v ealed a positiv e corr elation with αand β-aminobutyric acid, whic h wer e also identified as main drivers of difference between the control and the PFA group in EAM.While α-aminobutyric acid is known to be an agonist for TLR-2 (Santone et al. 2015 ), the role of the β isomer in relation to inflammation in cattle is not well known.β-aminobutyric acid is produced in plants in response to stressing stim uli, ther efor e the higher concentr ation of this metabolite in the PFA group might be due to its presence in the phytogenic supplement itself (The v enet et al. 2017 ).Ho w e v er, β-aminobutyric acid was also positiv el y corr elated with se v er al ASVs with v ery different taxonomic classification, and its concentration was reduced in weeks 1 and 2 HG, suggesting its faster metabolization by the ruminal bacteria for the production of proteins (Maeng et al. 1976, The v enet et al. 2017 ).
In goats fed highly fermentable diet, the proinflammatory cytokine IL-6 decreased in parallel with the anti-inflammatory cytokine IL-10 (Shen et al. 2016 ).Similarly, in our study both cytokines sho w ed the same trend due to PFA supplementation, although they were not affected by the dietary r egime.Furthermor e, the PFA treatment also decreased the expression of MyD88 , although not significantly.It is suggested that this expression pattern might be aimed at incrementing the host tolerance to w ar d some bacterial strains that pr olifer ate when the animals are fed HG diets (Shen et al. 2016 ).MyD88 was in fact identified as one of the main discriminants between control and PFA group in all the sPLS-DA models and sho w ed correlations with taxa significantly increased by the HG diet, such as Prevotellaceae , Lachnospiraceae , and Ruminococcaceae .Furthermore, it needs to be considered that the production of IL-6 might also activate anti-inflammatory pathways: it has been suggested that the expression of this cytokine is aimed at restoring the homeostasis by controlling the inflammatory response (Xing et al. 1998 ).Interestingly, the acute phase pr oteins measur ed in our study were reduced by the PFA supplementation, as reported in our companion study (Riv er a- Chacon et al. 2022 ).Although the anti-inflammatory effects of secondary plant compounds have been reported before in cattle and other species, the mechanisms of action of these blends have not been elucidated y et (P etri et al. 2020, Latek et al. 2022, Wang et al. 2023 ).

Effects of the phytogenic supplementation on the microbial metabolism and ruminal epithelium structure
Although the PFA treatment affected some r elativ el y high abundant taxa, it did not significantly impact the overall ruminal microbiota composition, as sho wn b y the microbial diversity.On the other hand, the phytogenic supplementation sho w ed effects on the micr obial pr edicted activity and on the measured metabolites .T his might be explained as the phytogenics are more effective to w ar d the microbial metabolism, rather than affecting microbiota composition (Hassan et al. 2020 ).
In fact, the PFA treatment impacted se v er al metabolites and biogenic amines in our study.For example, the stabilization of concentration of 5-A V A over the weeks could prevent the formation of 5-aminovaleric acid betaine, a potentially harmful compound derived from the fermentation of 5-A V A by the microbiota (Haikonen et al. 2022 ).The increment in methylbutyric acid due to the PFA, especially in the last weeks of experiment, could impr ov e the fermentation in the rumen, as supported by pr e vious studies in which the supplementation with 2-methylbutyric acid promoted the growth of cellulolytic bacteria in vitro and impr ov ed feed digestion in beef cattle (Dehority et al. 1967, Wang et al. 2012 ).Additionall y, the PFA tr eatment maintained stable lo w er le v els of dopamine, whic h might contribute to incr ease rumen contractions, since such metabolite has been demonstr ated to r educe rumino-r eticular motility in sheep (Buéno et al. 1983 ).T his , in turn, could enhance the clearance of VFAs and contribute to the buffering of ruminal pH (Riv er a- Chacon et al. 2022 ).
Similarly, the PFA supplementation reduced the concentration of kynurenine, a potentially harmful metabolite derived from the metabolization of tryptophan (Mándi andVécsei 2012 , Bae et al. 2020 ).The production of kynurenine is stimulated by several cytokines, including IFN-γ (Mándi and Vécsei 2012 ), but the lo w er concentr ations measur ed in the rumen did not correspond to a reduced epithelial expression of interferon in the PFA group.The PFA supplementation also increased the L-tryptophan biosynthesis pathway (TRPSYN-PWY) in LAM samples, but it is possible that the metabolism of kynurenine at the epithelial le v el was not shifted toward the production of toxic metabolites.It has in fact been observed that most of the ingested tryptophan is processed in the intestinal epithelium as well as in the liver through the kynurenine pathway, which results in de novo NAD + biosynthesis (Castro-Portuguez and Sutphin 2020 ).Dysregulation of tryptophan-kynurenine metabolism and NAD + synthesis may pr omote mitoc hondrial malfunction, and consequentl y alter the pr oduction of ATP (Castr o-Portuguez andSutphin 2020 , Castr o-Portuguez et al. 2020 ).It is known that the tightness of the epithelial barrier of the gastrointestinal tract is strictly ensured by a corr ect expr ession of ZO-1 , whose activity is ensur ed by a constant production of ATP by the mitochondria (Rossi 2022 ).In our study, the imm unohistoc hemical e v aluations sho w ed that the intrinsic apoptotic pathway was not activated in the ruminal epithelial cells, suggesting that the mitochondrial activity was preserved.T hus , the physiological expression of tight junctions was maintained, ensuring the integrity of the epithelial barrier of the rumen (Steele et al. 2011, McCann et al. 2016 ).Considering this, although our model did not show a significant impact of the PFA supplementation on the r elativ e expr ession of ZO-1 , the mor e stable concentration of kynurenine observed in the PFA group might have helped to preserve a regular expression of this tight junction.
Studies in other species demonstrated the potential beneficial effects of phytogenic compounds on the gut barrier function (Bachinger et al. 2019, Latek et al. 2022 ) and the PFA supplementation in our study reduced the relative expression of CLDN-4 and DSG1 .Ho w e v er, this is in contrast with pr e vious findings in cattle, in which phytogenic substances did not impact genes related with the barrier function (Petri et al. 2020 ), indicating that other el-ements, such as the experimental design, might have contributed to the altered gene expression in the rumen.Furthermore, the rele v ance netw ork sho w ed a correlation betw een CLDN4 and glyceric acid.This metabolite was reduced by the PFA supplementation and is part of the pentose phosphate pathway, implying a possible correlation between CLDN4 expression and the utilization of starch in the rumen.
The integr ativ e anal ysis performed with the tr eatment as explanatory variable provided results largely overlapping with the findings for the model considering the weeks .T his might indicate that the strong influence of the prolonged HG feeding on the sPLS-DA model could possibly mask some of the effects of the PFA.Further r esearc h is warr anted to establish the mec hanisms of action of the phytogenic supplementation on the ruminal microbiota and epithelial structure.

Conclusion
The prolonged feeding of HG diet caused ruminal dysbiosis, which was noticeable in the rumen digesta, both in LAM and SAM, especially after the second week of HG feeding, and decreased in se v erity by the last week of experiment.The increased expression of genes related with inflammation and with epithelial growth indicated a reaction of the epithelium to the challenging conditions .T he period between weeks 2 and 3 HG was recognized as the breaking point for the homeostasis of the ruminal ecosystem, and our integr ativ e anal ysis identified ZO-1 , MyD88 and Prevotella 1 as main drivers for the ruminal response .T he changes in barrier integrity genes as well as the lack of significant alterations of the rumen wall structure by the fourth week of experiment suggest a process of adaptation of the ruminal environment to the prolonged HG feeding.The metabolic adaptation of the rumen to the diet seems to be led mostly by the microbial activity of the LAM fraction.The PFA supplementation sho w ed the potential to aid this ada ptiv e pr ocess by altering the micr obial activity, specificall y r educing harmful metabolites, suc h as dopamine and 5-A V A, and by shifting the epithelial gene expression to increase the tolerance to w ar d the micr obiota, thr ough the alter ation of the expression of TLR-2 , IL-6 , and IL-10 .Future studies should be aimed at investigating the mechanism of action of the phytogenic compounds.

Figure 1 .
Figure 1.Principal component analysis based on the Aitchison distance measured for three ruminal microenvironments (LAM = liquid associated microbiota; SAM = solid associated microbiota; and EAM = epithelial adherent microbiota) in cows fed a forage-based diet (Forage) and a HG diet for 4 weeks.Results are presented for the control and the treatment (PFA) groups .T he distance matrix was affected by the diet and the weeks in all three ruminal niches ( P < .01 for LAM, SAM, and EAM for both diet and week).

Figure 2 .
Figure 2. Bar-gr a ph showing the mean r elativ e fr equency of the most abundant genera across all SAM (A), LAM (B), and EAM (C) samples.Results are presented for the control and the treatment (PFA) groups for the forage-based diet week (Forage) and for the four HG diet weeks.

Figure 3 .
Figure 3.Effect of the HG diet on the ruminal microbiota composition.Venn diagram (A)  showing the number of genera differentially abundant due to the HG diet in each ruminal niche analyzed (SAM, LAM, and EAM).Mean r elativ e abundances (B) in for a ge and in HG feeding of the 33 gener a that were affected by the diet in all three ruminal niches.(C) Number of genera differentially abundant during the prolonged HG feeding challenge in each ruminal niche analyzed (SAM, LAM, and EAM).The pr oportion of significantl y differ ent gener a between eac h HG feeding week is r epr esented for the contr ol gr oup and for the tr eatment gr oup (PFA).

Figure 5 .
Figure 5. Variation in concentration over the experimental weeks of the metabolites affected by the phytogenic treatment (PFA).Concentrations were measured in the rumen fluid of cows fed a forage-based diet (Forage) and a HG diet for 4 weeks.Results are presented per each metabolite for the control and the PFA groups per each experimental week.

Figur e 6 .
Figur e 6. CC A plot.CC A was performed between microbiome data for SAM, LAM, EAM, and the three metabolites identified as main responsible for the variation in ruminal metabolites concentration.Results are presented for the control and the treatment (PFA) groups for the forage-based diet week (For a ge) and for the four HG diet weeks.

Figure 9 .
Figure 9. Rele v ance networks obtained from the sPLS-DA using week as explanatory variable .T he networks integrate the gene expression (genes), metabolome (metabolites), and microbiota (ASVs) datasets for SAM, LAM, and EAM, r espectiv el y.The associations are shown for a cut off ≥ | 0.7 | .

Figure 10 .
Figure 10.Rele v ance networks obtained from the sPLS-DA using treatment (PFA) as explanatory variable .T he networks integrate the gene expression (genes), metabolome (metabolites), and microbiota (ASVs) datasets for SAM, LAM, and EAM, r espectiv el y.The associations ar e shown for a cut off ≥ | 0.7 | .

Table 1 .
Primers used for gene expression analysis.

Table 2 .
Alpha diversity matrices calculated for solid-(SAM), liquid-(LAM), and epithelial-(EAM) associated microbiota.Mean values and standard error of the mean are presented at the baseline and per each week of HG feeding for control and treatment (PFA) group .
(F aith's PD = Faith's phylogenetic diversity; Observed ASVs = Observed amplicon sequence variants).1 P -values for the effect of phytogenic treatment (PFA), diet within week (Diet) and of the interaction between treatment within week and diet (I).a, b Values with different superscripts indicate a significant difference ( P ≤ .05) between consecutive weeks.x, y Values with different superscripts indicate a tendency for difference (.05 < P ≤ .10) between consecutive weeks.

Table 3 .
Metabolites measured in rumen fluid.Mean values and standard error of the mean are presented at the baseline and per each week of HG feeding for control and treatment (PFA) gr oup.Concentr ation is giv en in μg ml −1 .

Table 3 .
Continued 1 P -values for the effect of phytogenic treatment (PFA), diet within week (Diet) and of the interaction between treatment within week and diet (I).a, b Values with different superscripts indicate a significant difference ( P ≤ .05) between consecutive weeks.x, y Values with different superscripts indicate a tendency for difference (.05 < P ≤ .10) between consecutive weeks.c, d Values with different superscripts indicate a significant difference ( P ≤ .05) between Control and treatment (PFA) groups within the same week.