Animals and diets

We used cecal content samples obtained at the time of sacrifice in the previous study by Vaziri, et al. Animals, housing conditions, and diets have been previously described [16]. Briefly, 10-wk-old male Sprague-Dawley rats were purchased from Harlan Laboratories (Indianapolis, IN) and fed powdered chow (no. 2020X, Harlan Laboratories) containing 0.7% adenine for 2 weeks to induce CKD. Rats were then randomized to receive semi-purified pelleted diets supplemented with 59% by weight with either the digestible starch, amylopectin (low fiber), or indigestible starch, HAMRS2 (Hi-Maize 260, Ingredion, Westchester, IL), for 3 weeks (n = 9 rats/group). Isocaloric diets were prepared by Harlan Laboratories containing 14.5% protein, 66.9% carbohydrate and 18.6% fat. All animals were provided free access to food and water. Rats were placed in metabolic cages for 24-h urine collection. They were then anesthetized [ketamine (50 mg/kg) plus xylazine (4 mg/kg ip)] and euthanized via cardiac exsanguination between 8:00–11:00 AM. Blood was collected and left at room temperature for 30 min to clot, serum was then collected, frozen on dry ice, and stored at -70°C until processed. Cecal contents were removed, frozen on dry ice, and stored at -70°C until processed. Hydration of cecal contents was determined by weighing a frozen aliquot and then oven drying overnight at 100°C; the sample was allowed to cool and weighed twice to ensure a constant weight was obtained. All experiments were approved by the Institutional Committee for the Use and Care of Experimental Animals of the University of California (Irvine, CA). These animals and samples of their cecal content were used in the previous publication by Kieffer et al. [3]. In the current paper we analyze these samples via novel metaproteomics method. We preserved the original enumeration of animals as in Kieffer et al. [3] with animals 7 to 15 being the DS-fed group and animals 16 to 24 being the RS-fed group.

Soluble protein extraction

In a cold room (4°C), 500 μl of PBS with protease inhibitor cocktail (ThermoFisher, Waltham, MA USA) was added to the 100 mg of the frozen cecal content and vortexed vigorously for 5 min followed by sonication for 5 min using Bioruptor (Diagenode, Denville, NJ). Bioruptor settings: High, 30 second on/off cycles. The lysates were centrifuged at 20,000 x g for 30 min. Pellets were saved at -80°C for future studies of membrane proteins. The supernatant (soluble proteins) were collected and protein concentration was determined using BCA assay (ThermoFisher). Reproducibility of the extraction has been evaluated by total de novo peptides identified and matched using the online metaproteomics tool Unipept [17]. Bacterial-to-host proteins ratio has been used as an extraction/sample quality criteria.

20 μg of protein for each sample was resolved on a 4–20% Tris-Gly gel. The gels lanes were cut in 24 pieces, in-gel digested and analyzed on the Thermo’s Orbitrap-Tribrid-Fusion instrument at the UAMS Proteomics Core. Spectral counting and tandem mass tags (TMTs) were used in parallel for protein quantification.

TMT-labeling

The samples were labeled with TMT-126, 127, 128, 129, 130, and 131 tandem mass tags using the ThermoFisher TMT6plex kit. The samples were prepared in four independent TMT batches. In the first batch we mixed samples 7, 8, 16, 17, 18, and a pooled mixture of all 18 samples. The second batch contained samples 9, 10, 11, 19, 20, and the pooled sample. The third batch contained samples 12, 13, 21, 22, and the pool sample (skipped label 129). The fourth batch contained samples 14, 15, 23, 24, and the pool sample (skipped label 129). The isobaric tags have an identical chemical structure and same total mass but with labile parts under collision-activated dissociation (CID), known as the reporter ions [18]. The reporter ions are detected at distinct m/z. Therefore, by multiplexing the samples labeled with different TMT tags, we can determine both the identity and relative abundance of peptide pairs simultaneously [19]. Samples were prepared for TMT labeling by first diluting 100 μg of each protein lysate to 100μl in 50mM TEAB. The samples were reduced in 200 mM Tris [2-carboxyethyl] phosphine (Thermo) at 55°C for 1 hour. Next, the samples underwent alkylation in 50 mM iodoacetamide (Sigma-Aldrich) for 30 min. Proteins were precipitated by adding 100% (w/v) trichloroacetic acid (TCA) to a final concentration of 20% (v/v), incubating at -20°C for 1 hour, then centrifuging in a microfuge for 15 minutes at 4°C. The samples were washed in 20% TCA, 100% acetone (cold), and in 80% acetone (cold) in order to remove any primary amines. After washing, samples were resuspended in 100 ul of 50mM TEAB with 1 μg/ul of trypsin and incubated at 37°C overnight. After protein digestion, the samples were labeled following the TMT6plex labeling protocol. The samples were mixed at a 1:1:1:1:1:1 ratio. We first tested a 1 μg mixing of each sample by mass spectrometry and used median reporter ion intensity values to adjust mixing amounts.

Basic HPLC and multinotch

The following two buffers were used for the basic HPLC: Buffer A, 0.1% formic acid, 0.5% acetonitrile; Buffer B, 0.1% formic acid, 99.9% acetonitrile. Both buffers adjusted to pH 10 with ammonium hydroxide for offline separation. Tryptic peptides were separated into 36 fractions on a 100 x 1.0 mm Acquity BEH C18 column (Waters) using an UltiMate 3000 UHPLC system (Thermo) with a 40 min gradient from 99:1 to 60:40 buffer A:B ratio under basic pH conditions, and then consolidated into 12 super-fractions. Each super-fraction was then further separated by reverse phase Jupiter Proteo resin (Phenomenex) on an in-line 200 x 0.075 mm column using a nanoAcquity UPLC system (Waters). Peptides were eluted using a 60 min gradient from 97:3 to 67:33 buffer A:B ratio. Eluted peptides were ionized by electrospray (2.15 kV) followed by mass spectrometric analysis on an Orbitrap Fusion Tribrid mass spectrometer (Thermo) using multi-notch MS³ parameters [20]. MS data were acquired using the FTMS analyzer in top-speed profile mode at a resolution of 240,000 over a range of 375 to 1500 m/z. Following CID activation with normalized collision energy of 35.0, MS/MS data were acquired using the ion trap analyzer in centroid mode over a range of 400–2000 m/z. Using synchronous precursor selection, up to 10 MS/MS precursors were selected for HCD activation with normalized collision energy of 65.0, followed by acquisition of MS³ reporter ion data using the FTMS analyzer in profile mode at a resolution of 60,000 over a range of 100–500 m/z.

De novo peptide sequencing

Files that were acquired on the Orbitrap Fusion Tribrid mass spectrometer (in “.raw” format) were submitted to the de novo sequencing using PEAKS Studio v. 8 (Bioinformatics Solutions, Waterloo, ON, Canada). The following parameters were used for the data refinement: Merge Scans–left unchecked; Correct Precursor–mass only; Filter Scans–unchecked. The following parameters were used for the de novo sequencing: Parent Mass Error Tolerance– 5 ppm; Fragment Mass Error Tolerance– 0.5 Da; Enzyme–Trypsin; Fixed Modifications–Carbamidomethylation (C); Variable Modifications–Oxidation (M), Deamidation (NQ); Max Variable PTM Per Petpide– 3; Report # Peptides– 5.

Preliminary taxonomy analysis and exclusion of outliers

De novo peptides identified by PEAKS [21] were filtered using the average local confidence score (ALC%), and the peptides with ALC% above 80 were submitted to the online metaproteomics tool UniPept [17, 22]. In the Unipept/metaproteomics analysis tab, the following parameters were set: Equate I and L—checked, Filter duplicate peptides–unchecked, Advanced missed cleavage handling–checked. The taxonomy information was visualized using a tree diagram provided by UniPept (S1 Fig) and using MetaCoMET (Metagenomics Core Microbiome Exploration Tool [23], Fig 2). For the visualization in MetaCoMET, taxonomy tables were downloaded from UniPept and converted to BIOME files using in-house R-script (available upon request). Ratio between eukaryotic and bacterial proteins was calculated using numbers of spectral counts.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Taxonomy analysis at the phylum level using de novo peptides.

De Novo peptides identified by PEAKS (ALC% score above 80) were exported into UniPept. UniPept output was converted to BIOME file format using in-house written R-script and further plotted using MetaCoMET online tool ([23], https://omictools.com/metagenomics-core-microbiome-exploration-tool-tool). * Samples 9 and 21: in the subsequent analysis these samples were proved to be outliers based on total host-to-bacterial protein ratios and excluded from quantification. Major groups are color-coded as followed (labels are shown for the CKD10 sample): u–unassigned, hot-pink; P–proteobacteria, pink; F–fermicutes, purple; R–chordata (rat), cyan; B–bacteroidetes, light brown.

https://doi.org/10.1371/journal.pone.0199274.g002

Samples derived from animals 9 and 21 had the most extreme host-to-bacterial proteins ratios (Fig 2, S1 and S2 Figs). In the subsequent hierarchical clustering analysis, these samples clustered together in a group separate from the two phenotypes. For these reasons samples 9 and 21 were excluded from the final dataset.

Peptide identification and protein inference using database search

Multi-step database search strategy was employed in PEAKS Studio as follows: The default mode of FDR estimation in PEAKS using fused decoy strategy (which is compatible with the multi-step searches [24] was used. NCBI_nr protein database (downloaded on 08/29/2016, 74,510,638 total entries) was searched using Rattus norvegicus as a taxonomic filter, (80,472 non-redundant protein entries searched, Search 1). The results of the PEAKS DB Search 1 were filtered using 1% false discovery rate for peptide-to-spectrum matches (corresponding average -10lgP ~25 across samples) and minimum 1 unique peptide per protein. Unmatched de novo tags from the Search 1 were further searched against NCBI_nr protein database using Bacterial taxonomic filter (64,001,258 non-redundant protein entries searched, Search 2). Fasta protein sequences from the two searches, for all 18 rat cecum samples, unfiltered (combined peptide-to-spectrum FDR 5%, 0 unique peptides allowed), were combined into a custom fasta database for the final search (~75,000 protein sequences in the final database). Contribution from other taxonomies (e.g. archaea, viruses, green plants and fungi–S1 Fig) was deemed negligible, and was not used in the downstream analyses. The original lists of de novo peptides were re-searched using this custom fasta database using PEAKS DB, PEAKS PTM, and PEAKS Spider searches. The final SPIDER results were filtered using 1% false discovery rate for peptide-to-spectrum matches (corresponding average -10lgP ~20 across 18 samples) and minimum 2 unique peptides. The results were exported as mzIdentML files along with spectral data in mascot generic format for the downstream analyses in Scaffold v. 4.

Database search and Taxonomic analysis with MaxQuant and Andromeda and iBAQ quantification was performed as follows. The “raw” files obtained during the Gel-LC-MS/MS experiment were also searched by MaxQuant/Andromeda search engine, in addition to PEAKS, with subsequent quantification using NSAF and iBAQ (Absolute Label-Free Protein Quantification) methods. The following parameters were set for the MaxQuant search: Fixed modifications Carbamidomethyl (C); Decoy mode revert; Include contaminants True; Top MS/MS peaks per 100 Da. (FTMS) 12; MS/MS deisotoping (FTMS) True; Top MS/MS peaks per 100 Da. (ITMS) 8; PSM FDR 0.01; Protein FDR 0.01; Site FDR 0.01; Min. peptide Length 7; Min. score for modified peptides 40; Use only unmodified peptides and True Modifications included in protein quantification Oxidation (M);Acetyl (Protein N-term); Peptides used for protein quantification Razor. For the full set of MaxQuant parameters see supplemental methods. Spectral counts were normalized using NSAF method–weighing by protein length, normalizing to the total sum per sample, and taking log transform. The transformation and imputation of missing values was performed in Perseus statistical suite. Imputed values were drawn from a normal distribution, with a width of 30% of the valid data and shifted by 2 standard deviation from the mean value of the valid data. To define differentially abundant proteins, moderated T-test (limma Bioconductor package) was performed on the transformed and imputed values. iBAQ values were processed in a similar fashion: relative iBAQ values were calculated by divided by a sample sum followed by log2 transformation and imputation of the missing values in Perseus (same parameters for imputation as with the NSAF method). As with the NSAF method, differential abundance was established using moderated T-test (limma package) applied to the log-transformed and imputed dataset. As a result, 2,842 proteins were quantified. Taxonomic analysis was performed in exactly the same way as with PEAKS-derived proteins.

Protein and taxonomic unit quantification using TMT labeling

PEAK Studio v. 8 with a quantitative module was used to analyze the TMT-labeled multiplexed experiments. Default parameters for the TMT MultiNotch MS3 analysis were selected and each of the four of the TMT-mixings was analyzed separately. Custom fasta database derived from the Gel-LC analysis was used. 1% peptide-to-spectrum FDR and 2 unique peptides per proteins were set as filters. Final data assembly to included all TMT-mixings, and ratios of normalized summed reporters intensities to pooled samples were recalculated in Microsoft Excel and exported into R for the downstream analysis. Custom fasta database derived from the Gel-LC analysis was used. 549 proteins were quantified using TMT method. limma Bioconductor package [25] was used as in the previous section to establish significance (p adjusted <0.05) for proteins and taxonomic units. 149 taxonomic units were identified in total using TMT method, with 45 established as significant.

Data sharing

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE [26] partner repository with the dataset identifier PXD008845 and 10.6019/PXD008845

Statistical methods

For protein quantification, Scaffold v. 4 with quantitation module (Proteome Software) was used. Data and the custom fasta database were exported from PEAKS into Scaffold as mzIdentML and mascot generic format files. Scaffold-derived normalized spectrum abundance factors (NSAF [27]) values and Total Unique spectral counts were used for the statistical analysis in R/bioconductor [28]. Taxonomic information from the NCBI-nr database (downloaded Oct. 16, 2016) was used to derive taxonomic units. Sum of spectral counts matching a given taxonomic unit, weighted by a total spectral count per a given replicate, was used as a measure of abundance for that taxonomic unit. Moderated t-test (limma package, Bioconductor) was used to establish differential abundance of taxonomic units. Benjamini Hochberg correction for multiple testing was enforced both for protein and taxonomic unit quantification. Heatmaps and hierarchical clustering of spectral count and TMT datasets were performed using the Bioconductor package ComplexHeatmap [29]. PCA analysis was performed using base R.

Metaproteomics protein signature of cecal contents from rats fed resistant starch or digestible starch diets

Overall, our approach allowed us to identify 9,386 unique proteins sequences across all experimental samples and conditions (1% false discovery rate for peptide-to-spectrum matches and minimum 2 unique peptides, S3 Fig). Combining complementary quantification methods (TMT, PEAKS, and MaxQuant) and requiring that to be quantified, a protein has to have at least 3 non-zero abundance values in either ithe CKD-DS or CKD-RS group, out of the 9,386 identified proteins 5,834 unique proteins were further quantified (S1 Table, tab (a)). The TMT method showed significant bias toward host proteins and yielded significantly lower numbers of quantified proteins (452 vs. 3,007 quantified by PEAKS, Table 1). Notably, more host proteins are reduced in abundance upon resistant starch supplementation (black lanes indicate host proteins, while grey indicate bacterial proteins on the annotation bar next to the heatmap in Fig 3, left). This result is consistent with previous reports that indicate an increase in bacterial biomass upon resistant starch supplementation [3].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The phenotype induced by resistant starch in chronic kidney disease (CKD) rats is well-defined by metaproteomics data.

Left heatmap, cecal content protein abundance values were averaged across three different labeling-free methods: PEAKS spectral counting, MaxQuant spectral counting and intensity based-quantification. The averaged values were log-transformed and normalized and plotted using ComplexHeatmaps R/Bioconductor package. 506 proteins, deemed significant by MaxQuant and by PEAKS are shown (p<0.01). CKD and CKDRS phenotypes are indicated by blue and red colors on the top heatmap annotation bar. Distribution properties are shown in BOX plots for each of the animals at the bottom, indicating similar distributions between different animals. The raw annotation bar is adjacent to the right edge of the heatmap–“host” indicates rat proteins by black color, and microbial proteins by grey color. Right heatmap, taxonomic units abundance values derived from the protein data were log-transformed and normalized and plotted using ComplexHeatmaps. Significant taxonomic groups (p <0.05) are shown.

https://doi.org/10.1371/journal.pone.0199274.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Number of quantified proteins by various methods.

For a protein to be quantified, a p-value had to be calculated (requiring at least 3 non-zero quantitative values in either CKD-DS or CKD-RS group). Total number of proteins quantified, and number of significant proteins using either p<0.05 or p<0.01 thresholds after Benjamini-Hochberg correction for multiple testing is reported for each of the quantification methods used.

https://doi.org/10.1371/journal.pone.0199274.t001

Hierarchical clustering procedures applied to abundance values derived by metaproteomics separated the two phenotypes into two distinct clusters (Fig 3, left heatmap). This separation also held when all proteins were considered, not just differentially abundant ones (S3 Fig). We note that that if rats 9 (CKD group) and 21 (CKD-RS group), were included in the heatmap, it would break the clear separation of the two phenotypes and form a separate cluster (S4 Fig). Upon further scrutiny, these two samples showed significant degradation of bacterial proteins, and as a consequence, low-quality fragmentation mass spectra and high host-to-bacterial protein ratio (S2 Fig). These samples were therefore excluded and not used for the final analysis.

There were 179 host proteins that were differentially abundant between the cecal contents of CKD-DS and CKD-RS rats: 125 were proteins reduced in CKD-RS and 54 increased in CKD-RS (Table 2, S1 Table–complete protein list). Among the 125 host proteins that had lower abundances in CKD RS, approximately half were enzymes. The remaining proteins were related to humoral immune response, several proteins previously associated with epithelial–mesenchymal transition (e.g. thioredoxin, S100-A6) and several proteins, previously reported, directly or indirectly, to be associated with CKD (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Representative differentially-abundant host proteins in cecal contents from CKD rats fed digestible starch (DS) or resistant starch (RS).

https://doi.org/10.1371/journal.pone.0199274.t002

Among the 54 host proteins that were increased in abundances in the CKD-RS group, about one third were enzymes. The largest group of proteins with similar functions was related to the immunoglobulin families. Another group included annexins (Table 2, S1 Table) that function as intracellular Ca²⁺ sensors and participate in cellular membrane repair [35], suggesting that in CKD-RS rats a process of intestinal cell repair, presumably damaged by bacteria and/or bacterial metabolites, is ongoing. Other small groups of proteins with similar function included a group of voltage-dependent ion channel proteins that regulate fluxes across the outer mitochondrial membrane and sodium pump subunit proteins. Mitochondrial impairment has been shown in CKD patients and animal models, particularly in the form of a decrease in mitochondrial DNA and down-regulation of many mitochondrial genes and proteins [36, 37]. Down-regulation of voltage channel proteins in CKD rats could be, in part, causing the increase of oxidative stress in CKD. Similar to previous studies [3, 16], we observed an increase in the tight junction protein claudin-3 (2.6 fold change in CKD-RS).

Interestingly, many of the individual proteins we found to be differentially abundant between CKD-DS and CKD-RS rats were previously identified in other reports on CKD and kidney associated diseases [16, 30, 32, 33, 37, 38]. Importantly, changes in these proteins were previously identified either in plasma or in urine. Identification of these proteins in cecal contents points to a potential utility of stool samples as a source of CKD protein biomarkers; however, the pathophysiological link between the association of the proteins in the GI tract with resistant starch and their association in urine and plasma during CKD is unclear.

Molecular processes influenced by dietary resistant starch supplementation in the cecal contents of CKD rats

To define enrichment of functional patterns associated with the differentially abundant proteins found by metaproteomics, we performed Gene Ontology analysis using Blast2Go [39]. There were 179 differentially-abundant host proteins described above, as well as 1198 differentially abundant bacterial proteins, the latter of which included 359 reduced and 839 higher proteins with RS feeding. The list of proteins differentially abundant between CKD-DS and CKD-RS rats was tested for enrichment of GO terms (biological process, molecular function and cellular component) for bacterial and host proteins separately using Fisher exact test. All GO terms significant at the FDR level of 0.05 were considered significantly enriched.

GO categories, under-represented in CKD-RS samples for rat proteins.

GO categories under-represented in the CKD-RS list of rat proteins (Table 3) were related to known biological processes occurring in CKD, such as aldehyde metabolism (indicative of lipid peroxidation processes), and humoral immune response. Reactive aldehydes are predominantly generated as a result of oxidative stress-induced lipid peroxidation, a process elevated in CKD patients [40]. Therefore, reduction in aldehyde biosynthesis and catabolism categories in CKD-RS rats is likely a consequence of the ongoing lipid peroxidation process in CKD rats. Reduction in positive regulation of humoral immune response category in CKD-RS rats first indicates the presence of residual bodily fluids in feces and second, that antimicrobial immune response is activated in CKD rats. Response to nutrients category includes many enzymes with different activities that are involved simultaneously in many GO biological processes, so it is difficult to dissect specific CKD-RS related process resulting in reduction ‘response to nutrients’ category. Two most specific molecular functions, reduced in CKD-RS rats were serine-type endopeptidase inhibitor and metalloexopeptidase activities. Only one cellular component category was reduced in CKD-RS rats, namely membrane-bounded organelle, showing that processes activated in CKD rats involve cellular membranes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Gene ontology categories under-represented in the CKD-RS host proteins.

https://doi.org/10.1371/journal.pone.0199274.t003

GO categories significantly over-represented in CKD-RS rat proteins.

There were no biological processes significantly over-represented in CKD-RS rat proteins at the FDR significance level 0.05. Molecular functions, significantly over-represented in CKD-RS rat proteins (Table 4) were related to the sodium-potassium pump subunits and cross-membrane transport; this may indicate that control of pH, osmotic pressure and cell volume are compromised in CKD rats and improve with the RS-rich diet. The most specific cellular components over-represented in CKD-RS rats were zymogen granule membrane, sodium:potassium-exchanging ATPase complex and extracellular exosome. All these molecular functions and cellular compartments were over-represented because of the presence of several specific proteins in CKD-RS differentially abundant protein list, but the amount of these proteins was not enough to identify enriched biological processes at the given significance level. However, if instead of FDR 0.05 we would consider p-value based significance level 0.001 then five biological processes would become significant: cellular sodium ion homeostasis, cellular potassium ion homeostasis, sodium ion export from cell, potassium ion import across plasma membrane and relaxation of cardiac muscle. All these categories were present because of the several sodium pump subunit proteins in CKD-RS DA protein list. However, the mechanisms of sodium-potassium pump regulation are complicated: for example, it was suggested that cardiotonic steroids mediate signal transduction through the Na/K-ATPase, and its downregulation could be indirectly implicated in profibrotic pathways [41]. Given that it is difficult to predict the consequences of insufficient sodium pump levels without further experiments, this observation warrants future study.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Gene ontology categories over-represented in the CKD-RS host proteins.

https://doi.org/10.1371/journal.pone.0199274.t004

Many biological processes and molecular functions, associated with CKD, especially at the more advanced stages, were previously characterized by transcriptomics. These included down-regulation in the kidney of regulatory proteins involved in cytoskeleton organization, microtubule assembly and stability, epithelial–mesenchymal transition, extracellular matrix remodeling, cell motility and migration, cell adhesion, apoptosis, cell differentiation, proteolysis, aminoglycan metabolic process and protein N-linked glycosylation [42, 43]. One advantage metaproteomics offers is the simultaneous analysis of bacterial and host proteins, with metaproteomics offering higher resolution of the bacterial portion of the proteome compared to metatranscriptomics. Indeed, in its current stage dual transcriptomics (i.e. simultaneous analysis of host and microbiome transcripts) is more challenging than metaproteomics. The pipeline for dual RNA-seq analysis has the same building blocks as conventional RNA-seq pipeline: reads must be cleaned, mapped, normalized and differentially expressed transcripts identified. However, as recommended by Avraham et al [44] host and pathogen reads should be mapped to the reference genomes and analyzed separately [44]. The latter creates a difficulty of using additional READemption pipeline for mapping bacterial reads, in addition to conventional TopHat pipeline for mapping host reads. READemtion pipeline uses the short read mapper segemehl and its remapper lack [45, 46] unlike Tophat that uses bowtie [47]. Also, when ones choses between transcriptomics and proteomics to characterize a disease–in our opinion proteomics is a better choice because 1) cellular phenotypes are defined by proteins more so than by transcripts, hence proteins are better biomarkers 2) druggable targets are usually proteins and not transcripts, and 3) correlation between protein abundance and transcript abundance is typically 40% to 60% depending on the cell type.

The analysis of significantly under-represented GO categories in CKD-RS for bacterial proteins supports the idea that in CKD-DS samples there is an ongoing process of mucin degradation that is more active than in CKD-RS samples (Table 5), in particular monosaccharide metabolic process and L-fucose catabolic process are significant. These processes may refer to the way bacteria in CKD rats preferably degrade mucins, containing fucosyl residues at the extremity of O-glycosidic chain [48]. The importance of this metabolic process is confirmed by the presence of fucosidase-encoding genes in the list of DA proteins (S1 Table). The idea of preferential foraging on mucin proteins by gut microbiota in CKD-DS rats is further supported by the list of bacterial species, under-represented in CKD-RS, where several mucin degraders are evident (e.g. Mucispirillum schaedleri, R.gnavus, R.torques–S2 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. GO categories under-represented in CKD-RS microbial proteins.

https://doi.org/10.1371/journal.pone.0199274.t005

GO categories over-represented in CKD-RS microbial proteins.

For bacterial proteins, GO categories over-represented in CKD-RS samples (biological processes, molecular functions and cellular components, Table 6.) were all related to active bacterial proliferation, emphasizing a shift in CKD-RS microbiota toward new actively dividing bacterial populations, which are able to thrive on RS.). Active bacterial proliferation can also partially explain the previously observed reduction in harmful tryptophan and tyrosine metabolites in CKD-RS rats (indoxyl and p-cresol, respectively). Instead of conversion to the toxins, tryptophan and tyrosine could be incorporated into newly synthesized proteins.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. GO categories over-represented in CKD-RS microbial proteins.

https://doi.org/10.1371/journal.pone.0199274.t006

Cecal microbiome composition revealed by metaproteomics at high resolution

Potential mechanisms of RS action

The question that remains to be fully understood is the mechanism(s) through which the RS diet exerts such a dramatic shift in control of biological processes and functions that are different between CKD-DS and CKD-RS rats. In part, it might be explained by bacterial population that shifts from mucin foraging to RS as a source of nutrients, relieving the host system from the constant flow of toxins that traverse the gut barrier. Other effects may involve reduction of oxidative stress. In the past, it was shown that kidneys from CKD patients have an impaired mitochondrial respiratory system [37], decreased DNA mitochondrial copy number [36], loss of mitochondrial membrane potential and lower ATP production [49]. The hypothesis that new bacterial populations influence mitochondrial biogenesis and activity (for example, through the influx of short chain fatty acids) is an attractive one. Recently it was shown that butyrate supplementation improved mitochondrial biogenesis in mice [50], presumably via inhibition of histone deacetylase (HDAC) that may down-regulate expression of PGC-1α, associated with mitochondrial dysfunction [51–53]. There is growing evidence that RS prevents colonic DNA damage via the production of SCFA, especially butyrate [54, 55]. In fact, it has been known, for almost a decade, that butyrate has a central role in maintaining gut epithelial integrity via involvement in key biological processes, such as being a source of energy for colonocytes, promoting fatty acid oxidation, having anti-inflammatory activity, limiting oxidative stress and inducing cell cycle arrest [56, 57]. In all these processes butyrate putatively functions by blocking substrate access to active sites in HDACs. Butyrate is a microbial fermentation product and butyrate producers are polyphyletic, belonging to different bacterial species, most known are members of Lachnospiraceae and Ruminococcaeae [58]. Butyrate is synthesized by those microorganisms via pyruvate and acetyl-coenzyme A (CoA) by breakdown of complex polysaccharides (such as RS). We observed that RS supplementation, at least for the Ruminococcaeae family, results in a shift from fibrolytic family members to amylolytic family members that are well known butyrate-producers (Ruminococcus bromii, Ruminococcus torques). Typically, in the absence of sufficient dietary fibers, commensal and pathogenic bacteria start to forage on mucin glycans to harness carbon and energy [59, 60]. While we did not examine if microbes lowered with RS feeding are mucin degraders we did find indirect evidence in support of that from GO categories reduced with RS feeding (monosaccharide metabolic process and L-fucose catabolic process). We therefore hypothesize that there is a relative shortage of fiber in CKD-DS and the RS diet shifts gut microbial communities, from bacteria foraging on mucins and ammonia and contributing to leaky gut phenotype to bacteria utilizing RS instead, and producing butyrate as a byproduct of fermentation.

Recently it was found that out of 3,184 sequenced bacterial genomes, mostly from the Human Microbiome Project, 225 were likely to be butyrate producers [58]. From the list of bacteria, up-regulated in CKD-RS, only Eubacterium rectale, Clostridium botulinum and Lachnospiraceae bacterium strains are present in this list of potential butyrate producers, presumably because rat and human gut microbiomes differ. It would be interesting to evaluate quantitatively the amount of mucin degraders and butyrate producers caused by the diet shift by genomics and transcriptomics, to reduce biases due to proteomics undersampling (e.g. bias against low abundance proteins).

To summarize, dietary RS supplementation in rats ameliorates chronic kidney disease coincident with a massive shift in gut microbial communities (Fig 4). Identified organisms and proteins point toward a higher abundance of butyrate-producing bacteria, and reduced abundance of mucin-degrading bacteria. It is suggested that the bacterially-derived butyrate leads to reduction of oxidative stress and inflammation, as well as to improvement in other biological processes that are impaired in CKD. It is also possible, that the function of the gut epithelial barrier is improved during the RS supplementation due to the reduced degradation of mucins. Finally, resistant starch supplementation leads to the active bacterial proliferation and the reduction of harmful bacterial metabolites. The fact that simple change of available source of nutrients (from ammonia/mucins to RS) leads to system-level changes underscores the importance of diet during disease management and highlights the potential role of the gut microbiome in disease progression.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Working model of how RS diet changes gut microbiome composition, and in turn alleviates CKD.

Dietary resistant starch supplementation results in a higher abundance of butyrate-producing bacteria and reduced abundance of mucin-degrading bacteria. Bacterially-derived butyrate mediates reduction of oxidative stress and inflammation. Integrity of gut epithelial barrier is also benefiting from reduced abundance of mucin-degrading bacteria. In addition, resistant starch supplementation leads to the active bacterial proliferation and the reduction of harmful bacterial metabolites, events that ameliorate CKD-induced oxidative stress and local and systemic inflammation. In fact, recent clinical trial has revealed that dietary supplementation of resistant starch, amylose, attenuates elevation of biomarkers of oxidative stress and inflammation in patients with end stage renal disease maintained on hemodialysis [61]-.

https://doi.org/10.1371/journal.pone.0199274.g004