Characterization of variations within the rumen metaproteome of Holstein dairy cattle relative to morning feed offering

Few studies have utilized proteomic techniques to progress our knowledge of protein-mediated pathways within the rumen microbial community, and no previous research has used these techniques to investigate the patterns or variations of these proteins within this community. It was hypothesized that there would be fluctuations of rumen microbial protein abundances due to feed intake-mediated nutrient availability and that these could be identified using non gel-based proteomic techniques. This study investigated the fluctuations of bovine rumen metaproteome utilizing three mid to late-lactation Holsteins. Rumen fluid was collected at three timepoints on three days relative to their first morning feed offering (0 h, 4 h, and 6 h). Samples were pooled within timepoint within cow across day, analyzed using LC-MS/MS techniques, and analyzed for variations across hour of sampling using PROC MIXED of SAS with orthogonal contrasts to determine linear and quadratic effects. A total of 658 proteins were characterized across 19 microbial species, with 68 proteins identified from a variety of 15 species affected by time of collection. Translation-related proteins such as 50S and 30S ribosomal protein subunit variants and elongation factors were positively correlated with hour of sampling. Results suggest that as nutrients become more readily available, microbes shift from conversion-focused biosynthetic routes to more encompassing DNA-driven pathways.


Results
Metaproteomic analysis using the outlined protocol resulted in identification of 698 proteins across 19 microbial species including multiple strains (see Supplementary Table S1 for list of proteins identified within species and Supplementary Table S2 Table S2). Results were grouped and interpreted based on abundance pattern shifts relative to morning feeding with LOW0 representing proteins that demonstrated an increase in abundance either quadratically or linearly relative to initial sampling (0 h) and HIGH0 representing proteins that began with a higher abundance at 0 h and had either a linear or quadratic decrease in the latter sampling points at 4 h and 6 h.
Of the quantified proteins, there were 68 proteins across 15 microbial species that were affected by time of sampling, including 88.2% that responded linearly, and 30.9% being affected by a quadratic pattern of change.
Microbial proteins with a decrease in abundance from 0 h to 4 h. There

Discussion
Understanding microbial metabolism is a crucial step in the development of strategies to support and sustain maximal nutrient use efficiency in the rumen. Inclusion of proteomic techniques to articulate the underlying shifts in the rumen metaproteome is in its infancy, but gel-based protein fractionation from samples collected in static points of time highlight the breadth of protein identifications that can be achieved 15,27 . Using a combination of rumen-specific fractionation protocols with isobaric labeling techniques, the research reported herein is the first to outline the potential dynamic range of the rumen microbial metaproteome relative to first morning feeding. For this research, 47 composite database searches were completed, encompassing 19 microbial species and numerous strains. While this represents only a small fraction of the rumen microbiota, and further work must be done to ensure inclusive and accurate proportional representation of the rumen microbiota, this research encompasses the largest rumen metaproteomic search to date. Snelling and Wallace 27 were unable to characterize the metaproteome of grazing dairy cattle due to impeding compounds on gels, however, they were able to distinguish 50 unique proteins derived from other ruminants such as lambs and beef cattle. Hart et al. 15 highlighted the 25 most common protein families found within each individual cow and discussed phyla dominance but did not delineate proteomes of specific microbes.
In the present study, there were 43 variants of ribosomal proteins characterized within this research (16 30 s and 27 50 s) across 16 microbial species and is the most represented protein in this study. While some of these identified proteins may be redundant due to overlapping peptide sequences, approximately 20% of bacterial dry weight is made up of ribosomal proteins so it is not unexpected to see that these proteins are the majority of identified proteins 28 . These ribosomal proteins are subunits within a larger bacterial ribosome that serves in mRNA translation to protein, and each subunit serves a specific function. The smaller 30 s subunit decodes the mRNA strand while the larger 50 s subunit assists in peptide bonding of the specified amino acids. Another protein involved heavily in the central dogma, elongation factor Tu, was also a widely represented protein being characterized within 13 microbial species. This protein contributes to the continuous process of sequentially adding amino acids to a peptide chain. Overall, nearly all LOW0 proteins are translation-related. Ribosomal proteins (50 s and 30 s) and their associated variants were the most commonly affected proteins across sampling time with 53.5% of identified ribosomal proteins being affected by time of sampling, and 95.7% of the affected ribosomal proteins being represented in LOW0. Elongation factors (G, R. flavefaciens; Tu, P. ruminis and R. bromii) showed an increase in abundance relative to 0 h which also placed them in the LOW0 group.
To interpret the results, we broadly partitioned the rumen metaproteome into two conceptual categories, the first category being the above discussed general pathways of transcription and translation. The second category includes proteins with more specific roles in metabolic pathways. While it can be difficult to draw conclusions regarding the substrate or environmental drivers causing an increase in proteins related to translational processing beyond simply surmising that protein synthesis is likely increasing, there were many proteins identified in the current study that are key players in many specific metabolic pathways, highlighting possible shifts in more specific rumen functionality. In contrast to proteins grouped in LOW0, in HIGH0 there was largely a lack of representation of 30 s/50 s ribosomal proteins, elongation factors, and other translation-related proteins. Instead, there were shifts in proteins with more targeted functions, such as fumarate hydrolase, phosphoenol pyruvate carboxykinase, cysteine synthase, nitrogen-fixing protein NifU, GGGtGRT protein, pyruvate, phosphate dikinase, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH). These results highlight the concept that rumen microbiota are independently reactive to their environment but yet are in synchrony with each other. Investigation of only the substrates or products within the rumen would not have likely yielded a comprehensive www.nature.com/scientificreports www.nature.com/scientificreports/ understanding of the individual microbial activities including the extent of synchronicity or similarity in protein profile across the different species within the rumen.
A consideration from the results is that the identified protein shifts illustrate that feeding protocol-and bunk management-induced fasting periods may quickly limit potential productivity of microbes. It is important to note that only 10.3% of analyzed proteins were affected by time of sampling, and could be due to the fact that cows were not fasted or subject to a more significant dietary perturbance or limitation beyond typical daily feeding schedule. Regardless, these results provide insight into the pathways and microbes more readily impacted by substrate-mediated suppression of microbial protein synthesis during non-eating periods 29 , and the impact that it may have on total microbial biomass production in the rumen and the consequent intestinal supply of microbial protein 30 .
Another aspect of rumen function highlighted from the current results is that nutrient deprivation may expose the fundamental metabolic pathways of specific microbes. Identifying the proteins in HIGH0 may give insight into pathways that specific microbes deem vital. As highlighted by 31 , identifying and exploiting the roles of independent microbes or groups of microbes is a current challenge. The inclusion of proteomics in study methodology can more broadly highlight pathways affected by a treatment or physiologic state rather than focusing on a small set of parameters to directly or indirectly assess ruminal changes and can identify metabolic shifts that do not result in a change in microbial diversity. The use of dietary models in combination with these proteomic techniques is also proposed to be a feasible method to better identify the basic roles of the rumen microbes, and what protein-mediated pathways the different microbes divert to in different nutrient scenarios or feed management protocols.
This research is the first published work to report the rumen metaproteome beyond static points in time and demonstrates how proteomic technology can provide a meaningful contribution to the characterization of microbial activity and protein-mediated pathway dynamics. The trial reported herein demonstrates that the rumen protein profile is dynamic and appears to be sensitive to lower nutrient availability. Furthermore, this research supports the hypothesis that inclusion of proteomic technology to characterize the rumen metaproteome and the impact of diet, health, and environment on rumen functionality can provide a useful contribution and further advance our research to maximize production efficiency.

Methods
Animals and maintenance. Samples were collected from three lactating Holstein dairy cows (207 ± 53.5 days in milk) housed at the Paul R. Miller Research Complex (The University of Vermont, Burlington, VT, USA). Cows were fed a nutritionally balanced dietary ration (diet chemical composition listed in Supplementary  Table S3) ad libitum and were offered a total mixed ration twice daily (0630 h and 1430 h) and a diet supplement (high grain pellet) four times daily (0645 h, 1045 h, 1730 h, and 2300 h). All feed refusals were discarded prior to morning (0630 h) feeding daily. Cows had ad libitum access to water. Animal use and samplings methods performed in this trial were reviewed and approved by the Institutional Animal Care and Use Committee of the University of Vermont (Protocol #16-029) in accordance with the requirements of the Office of Laboratory Animal Welfare.
As per methods outlined by Steele et al. 35 , digesta grab samples were collected at each timepoint via a rumen cannula from beneath the fiber mat within the ventral sac of the rumen. A minimum of three digesta grab samples were collected per cow at each sampling point for a single representative sample per timepoint. Samples were immediately snap frozen in a dry-ice ethanol bath, transported on dry ice to the laboratory, and stored at −80 °C until processing.
Rumen sample processing. For processing, RF samples were thawed overnight at 4 °C. Once thawed, samples were filtered through 4 layers of cheesecloth (Lion Services Inc., Charlotte, NC, USA), and filtered samples were composited within cow within timepoint across day for a representative sample of 0 h, 4 h, and 6 h for each cow prior to freezing at −80 °C. For centrifugation, the composited filtered samples were thawed on ice and centrifuged at 16,000 × g for 20 minutes at 4 °C. The resulting supernatant was discarded, and the remaining pellets were retained.
The collected pellets were lysed using both chemical and mechanical lysis methods based on both Snelling and Wallace 27 and Yu and Morrison 36 with modifications. Briefly, 1.5 mL of RIPA lysis buffer containing protease inhibitor (Pierce TM Protease Inhibitor Mini Tablets, Thermo Scientific, Rockford, IL, USA) and a 5 mm stainless steel bead (Qiagen, Hilden, Germany) was added to each pelleted sample and samples were homogenized (TissueLyser II, Qiagen, Hilden, Germany) through six repetitions of lysis at 30 Hz for 30 seconds with a 3-min incubation on ice between each repetition, similar to the BeadBeater protocol reported by Luccitt et al. 37 . A subsample of each sample homogenate was pipetted into a clean tube and precipitated overnight at 4 °C in a lysis solution (6 M TCA, 80 mM DTT) (3:1 protein extract to TCA/DTT) similar to the protocol outlined by Snelling & Wallace 27 .
Following the overnight incubation, the samples were vortexed and centrifuged at 16,000 × g for 20 minutes at 4 °C and supernatants were discarded. The retained pellets were then washed four times as per methods of Snelling and Wallace 27 with modifications by Song et al. 38 , where the retained pellets were washed in ice-cold 20% DMSO in acetone, incubated for 1 hour at −20 °C, and centrifuged at 10,000 × g for 5 minutes at 4 °C. The supernatants were then discarded, and the pellets were again washed using the same protocol. This wash protocol was then repeated twice more using 100% ice-cold acetone. After the final wash, the collected pellets were air dried and resuspended in phosphate buffered saline. A new 5 mm stainless steel ball was added to each sample before samples were homogenized in the TissueLyser for 30 seconds at 30 Hz. A universal control (UC) sample was generated by combining equal volumes from each of the 9 samples. Samples were stored at −80 °C until protein quantification of samples was performed using the bicinchonic acid assay (BCA) kit (Pierce, Rockford, IL, USA). tMt isobaric labeling, high pH reversed-phase peptide fractionation and liquid chromatographytandem mass spectrometry (LC-MS/MS). Quantified samples (85 μg) were then labeled using TMT Isobaric Tags as per manufacturer instructions (Thermo Scientific, Rockford, IL, USA). Labeling efficiency of each samples was verified to be more than 96% through preliminary MS analysis of individual samples. Equal volumes (75 μL) of each TMT-labeled sample was combined into a new tube and a 100 μL aliquot was vacuum dried to remove the triethyl ammonium bicarbonate (TEAB). The peptides were then fractionated using the high pH reversed-phase peptide fractionation kit (Thermo Scientific, Rockford, IL, USA) as per kit instructions resulting in 8 fractions for LC-MS/MS per original sample. One-tenth of each of the fractionated samples was dried down and resuspended in 2.5% formic acid (FA) in water and 2.5% acetonitrile (CH 3 CN). The LC-MS/MS analysis was carried out on the Q-Exactive Plus mass spectrometer coupled to an EASY-nLC 1200 (Thermo Scientific, Waltham, MA, USA) performed by the VGN Proteomics Facility (Burlington, VT, USA). Peptides were separated using a gradient of 2.5-35% CH 3 CN/0.1% FA over 60 min, 35-100% CH 3 CN/0.1% FA in 1 min and then 100% CH 3 CN/0.1% FA for 4 min, followed by an immediate return to 2.5% CH 3 CN/0.1% FA and a hold at 2.5% CH 3 CN/0.1% FA. The nanospray and data acquisition methods were completed per Scuderi et al. 39 . Briefly, samples were loaded onto a 100 μm × 500 mm capillary column packed with Halo C18 (2.7 μm particle size, 90 nm pore size, Michrom Bioresources, CA, USA) at a flow rate of 300 nL min −1 . The column end was laser pulled to a ~3 μm orifice and packed with minimal amounts of 5um Magic C18AQ before packing with the 3-μm particle size chromatographic materials. Peptides were introduced into the mass spectrometer via a nanospray ionization source with a spray voltage of 2.0 kV. Mass spectrometry data was acquired in a data-dependent "Top 10" acquisition mode with lock mass function activated (m/z 371.1012; use lock masses: best; lock mass injection: full MS), in which a survey scan from m/z 350-1600 at 70, 000 resolution (AGC target 1e 6 ; max IT 100 ms; profile mode) was followed by 10 higher-energy collisional dissociation (HCD) tandem mass spectrometry (MS/MS) scans on the most abundant ions at 35,000 resolution (AGC target 1e 5 ; max IT 100 ms; profile mode). MS/MS scans were acquired with an isolation width of 1.2 m/z and a normalized collisional energy of 35%. Dynamic exclusion was enabled (peptide match: preferred; exclude isotopes: on; underfill ratio: 1%). , and E. cellulosolvens (UP000005753). All 9 raw files were searched against the database as one contiguous input file. Search parameters were as follows: (1) full trypsin enzymatic activity; (2) mass tolerance at 10 ppm and 0.02 Da for precursor ions and fragment ions, respectively; (3) dynamic modifications: oxidation on methionine (+15.995 Da); (4) dynamic TMT6plex modification on N-termini and lysine (+229.163 Da); and (5) static carbamidomethylation modification on cysteines (+57.021 Da). Percolator node was included in the workflow to limit the false positive (FP) rates to less than 1% in the data set. The relative abundances of TMT labeled peptides were quantified with the Reporter Ions Quantifier node in the Consensus workflow and parameters were set as follows: (1) both unique and razor peptides were used for quantification; (2) Reject Quan Results with Missing Channels: False; (3) Apply Quan Value Corrections: False; (4) Co-Isolation Threshold: 50; (5) Average Reporter S/N Threshold = 10; (6) "Total Peptide Amount" was used for normalization and (7) Scaling Mode was set "on Control Average", so that the peptide abundances in the UC labels were set as 100 and the abundances in other channels were scaled accordingly. Non-normalized data is listed in Supplementary Table S4. The normalized scaled abundance values were used for subsequent statistical analyses. For any proteins identified that remained "uncharacterized", either PANTHER Classification System 40 was used to identify to protein name using the accession number, or the FASTA sequence was retrieved from UnitProt 41 and searched through BLAST 42 and the top hit protein was selected as the protein ID if above 99.0% matched identity. PROC IML in SAS was used to determine appropriate weighting for orthogonal contrasts to determine linear and quadratic effects of time.