MetaPro-IQ: a universal metaproteomic approach to studying human and mouse gut microbiota
- 4.3k Downloads
The gut microbiota has been shown to be closely associated with human health and disease. While next-generation sequencing can be readily used to profile the microbiota taxonomy and metabolic potential, metaproteomics is better suited for deciphering microbial biological activities. However, the application of gut metaproteomics has largely been limited due to the low efficiency of protein identification. Thus, a high-performance and easy-to-implement gut metaproteomic approach is required.
In this study, we developed a high-performance and universal workflow for gut metaproteome identification and quantification (named MetaPro-IQ) by using the close-to-complete human or mouse gut microbial gene catalog as database and an iterative database search strategy. An average of 38 and 33 % of the acquired tandem mass spectrometry (MS) spectra was confidently identified for the studied mouse stool and human mucosal-luminal interface samples, respectively. In total, we accurately quantified 30,749 protein groups for the mouse metaproteome and 19,011 protein groups for the human metaproteome. Moreover, the MetaPro-IQ approach enabled comparable identifications with the matched metagenome database search strategy that is widely used but needs prior metagenomic sequencing. The response of gut microbiota to high-fat diet in mice was then assessed, which showed distinct metaproteome patterns for high-fat-fed mice and identified 849 proteins as significant responders to high-fat feeding in comparison to low-fat feeding.
We present MetaPro-IQ, a metaproteomic approach for highly efficient intestinal microbial protein identification and quantification, which functions as a universal workflow for metaproteomic studies, and will thus facilitate the application of metaproteomics for better understanding the functions of gut microbiota in health and disease.
KeywordsGene catalog Gut microbiota Metaproteomics Metagenomics Protein identification Quantification
Intestinal microorganisms, namely the gut microbiota, have been shown to be important in multiple aspects of physiological processes relating to health and disease, including nutrition, metabolism, and immunity . Host genetics and nutrition have been shown to affect the composition of gut microbiota; conversely, the microbial metabolites or the microbes themselves can also regulate host metabolic processes [1, 2, 3, 4]. Disruption of the homeostasis surrounding these host-microbe interactions has recently been shown to participate in the development of many diseases including obesity, diabetes, and inflammatory bowel disease (IBD) . Therefore, the gut microbiota is emerging as an important topic for public health and scientific researchers.
Next-generation sequencing (NGS) has been widely applied in gut microbiota studies using various experimental approaches . Classically, NGS studies have focused on characterizing the taxonomic profile of the gut microbiota through targeted amplicon sequencing (e.g., 16S hypervariable regions) or through shotgun metagenomics. More recently, NGS has been used to characterize the metabolic potential of the gut microbiota through metatranscriptomic analysis . While these approaches are quite useful, they are unable to demonstrate that the predicted biological processes are actually present in the gastrointestinal tract. Metaproteomics, which examines all the expressed proteins in a microbial community, has been shown to provide invaluable functional information for the gut microbiota . Moreover, the application of proteomics and metaproteomics to host-gut microbe interactions will help to provide meaningful information on the roles of microbiota . Although the first shotgun metaproteomic study of the gut microbiota was reported in 2009 , only few follow-up large-scale gut metaproteomic studies have been published. The reasons for this lack of progress have been discussed in several reviews [8, 9, 10, 11, 12] and include (1) the inability to detect low abundant proteins with current mass spectrometers coupled to the high diversity of the gut microbiota and, more importantly, (2) the low efficiency in identifying gut microbial peptide or proteins from acquired mass spectrometry (MS) spectra. The latter is mainly due to the lack of a suitable database (in terms of database coverage and size) for peptide-spectra matching (PSM) which is the key step for current bottom-up proteomic studies.
The ideal database for proteomics should be composed of all potentially expressed proteins in samples being analyzed with no additional spurious sequences; however, this is challenging for gut metaproteomics due to the enormous diversity and individual variations of gut microbiome . The NCBI nr database is often used in metaproteomics due to its high sequence coverage. Unfortunately, its enormous size (~80 million entries) makes the database searching extremely time consuming and also provides less sensitive peptide identifications when using target-decoy approach for false discovery rate (FDR) filtering . To overcome these weaknesses, customized databases are often employed and are composed of a list of known gut microbial genomes that are manually chosen and combined to generate a synthetic metagenome database [14, 15, 16]. However, it is unclear how representative these synthetic metagenome databases are as compared to the actual metagenome since many intestinal microbes are un-cultivable and/or their genomes have not been sequenced. In addition, the researcher’s choice of which bacteria taxa to include in the database is often arbitrary and may result in biases, which also makes the cross-study comparisons difficult. An alternative to manually curating a custom database is using a matched metagenome database. Briefly, either a “representative” subset or all the samples being investigated are subjected to metagenomic sequencing to compile a database of genes . Since metagenomic sequencing introduces additional costs, it has not been generally performed in currently reported metaproteomic studies. These challenges have undoubtedly contributed to the fact that most metaproteomic publications only identify around 3000 proteins, which is far less than expected. Recently, Jagtap et al.  proposed a two-step database search strategy where a first search is performed against the target-only version of a database to generate a smaller refined database, which is then used for a second classical target-decoy database search . This has been shown to increase the sensitivity of peptide identification and greatly increase the number of peptides and proteins identified. In a recent study by Tanca et al.  over 13,000 peptides corresponding to 9000 proteins were identified for mouse cecum samples by combining a matched metagenome database with the aforementioned two-step strategy. To the best of our knowledge, this is the highest protein identification for a single gut metaproteomic study to date. However, as mentioned above, the matched metagenome approach suffers from the need for metagenomic sequencing and the differences of databases used in different studies. Thus, a comprehensive gut microbial gene or protein database, which covers the expressed proteins for all gut microbial species, will make gut metaproteomics more affordable, comprehensive and comparable, which will largely promote wide application of metaproteomics in microbiome studies.
The Metagenomics of Human Intestinal Tract (MetaHIT) [19, 20] and the Human Microbiome Project (HMP) consortiums  have aimed to compile comprehensive gut microbial gene catalogs based on high-throughput sequencing to facilitate the analysis of multi-omic data. More recently, the close-to-complete human and mouse gut microbial gene catalog databases have been published and made freely available. Thus, in this study, we leveraged these gene catalogs as unified databases for protein identification in gut metaproteomics. We subsequently developed a high-performance and universal approach for gut MetaProteome Identification and Quantification (MetaPro-IQ), which led to the quantification of approximately 120,000 peptides corresponding to >30,000 protein groups (defined as the cluster of proteins identified by the same set or a subset of peptides) in a single experiment. More importantly, since the gene catalog databases are centrally curated and publicly accessible, the current approach can be widely applied for gut microbiota studies to generate comparable results from different researchers. Thus, the MetaPro-IQ shows great potential as a universal approach which would largely promote the wide application of gut metaproteomics.
Results and discussion
Implementation of the MetaPro-IQ workflow for gut microbiota study
A gene/protein database is typically needed for the purpose of peptide identification and is readily available for most model organisms. A typical proteome database contains 5000–20,000 entries which leads to a reasonable database search time and FDR filtering sensitivity . However, the gut microbiota consists of thousands of microbial species and most of them are remain largely unknown or lack any available genome or proteome sequences . This makes the gut metaproteomic studies much more challenging. Therefore, in this study, we used the newly generated, well-annotated gut microbial gene catalog databases (available from http://meta.genomics.cn/ and http://gigadb.org/) [20, 24], to improve the database coverage for the gut microbial proteins. The human gut microbial gene catalog contains 9.9 million genes generated from >1200 metagenomic sequencing samples , and the mouse gut microbial gene catalog contains 2.6 million genes generated from 184 sequenced mouse samples of diverse genetic and environmental backgrounds . These databases are the most comprehensive for human and mouse gut microbial genes to date, and more importantly, the application of these gene catalog databases allows the generation of unified protein lists from different studies, thus enabling easy cross-study comparisons. The latter is also one of the important disadvantages for the current gut metaproteomic studies as aforementioned.
The sizes of the abovementioned gene catalog databases are large (>106 entries), which will greatly limit the database search sensitivity . Iterative database search strategy has been previously shown to increase the sensitivity of peptide identification using huge databases, particularly for the metaproteomics and proteogenomics . Thus, the MetaPro-IQ conducted a first-step database search against the whole gene catalog database, which generated a reduced database from the original gene catalog based on the acquired MS spectra in a sample. The reduced database was then used for a classical target-decoy database search to obtain confident peptide/protein identifications [13, 17].
In addition to the peptide/protein identification, quantification is another important aspect for proteomics. Spectra counting-based label-free quantification had been used for protein quantitation in many previous gut metaproteomic studies [7, 14, 15]. However, for data generated from high-resolution mass spectrometers, such as orbitraps, precursor-intensity-based algorithms have been shown to have superior accuracy, particularly when dynamic exclusion is enabled for spectra selection [22, 25]. MaxQuant is a widely used quantitative proteomics software implemented with the advanced MaxLFQ algorithm, which is based on precursor signal intensity and advanced delaying normalization across samples . The MetaPro-IQ approach adopted MaxQuant for the final-step quantification analysis, which enabled accurate quantifications of the identified microbial proteins.
MetaPro-IQ reached comparable performance with matched metagenome strategy
To compare the abilities of the two approaches for extracting functional information, all the quantified proteins were annotated with Clusters of Orthologous Group (COG) categories. Twenty-three COG categories were observed with the matched metagenome approach, which were all found with the MetaPro-IQ approach. There is no obvious difference in the relative abundance of the high abundant COG categories between the two approaches (Fig. 3e). Several COG categories such as B, Z, and X were solely present or with obviously higher observed LFQ intensity using the MetaPro-IQ approach (Fig. 3e). This may result from the lack of low abundant genes in matched metagenome databases due to inadequate sequencing depth. The low abundant genes may have relatively high protein-expression levels which are detectable using mass spectrometers, and thereby were identified by MetaPro-IQ approach.
To examine whether the above observations are dataset dependent, the murine fecal metaproteome dataset (MFM; two replicates with two runs for each replicate) from the study of Tanca et al.  were re-analyzed with the MetaPro-IQ workflow. In total, we quantified 19,497 peptides and 4549 protein groups for replicate 1 and 19,972 peptides and 4630 protein groups for replicate 2. More than 92 % of the peptides were quantified for both replicates (Additional file 2: Figure S1A). Tanca et al.’s study, using a matched metagenome database search strategy, identified 14,085 peptides for replicate 1 and 15,669 peptides for replicate 2 with an overlap of 63 % . Compared to the matched metagenome strategy, the MetaPro-IQ workflow identified more peptides with a better overlap between replicates for their dataset. In addition, a Pearson’s correlation coefficient of 0.89 was obtained between the two replicates, and more than 0.86 between runs (two mass spectrometry runs were conducted for each replicate in Tanca et al.’s study, Additional file 2: Figure S1B–F), which is also in agreement with the findings in their study.
Taken together, the MetaPro-IQ metaproteomic workflow using the gut microbial gene catalog database showed better performance for identifying gut microbial proteins, when compared to the workflow using a matched metagenome database. MetaPro-IQ allows high efficient protein identification from MS spectra in metaproteomics without the need for prior metagenomic sequencing (greatly reduces the experimental cost) and is readily applicable for all researchers from various disciplines.
MetaPro-IQ approach revealed metaproteome response of gut microbiota to diet in mice
More than 94 % of the key microbial proteins have a greater than twofold change between HFD and LFD groups, while thirty eight of them have more than a 100-fold change, representing the major responders to HFD (Additional file 1: Table S5). Proteins S-Fe9_GL0178463 and S-Fe7_GL0107273 were the most markedly changed proteins with more than 1000-fold decrease in the HFD group, and both of them were annotated as hypothetical proteins from Eubacterium plexicaudatum (Additional file 1: Table S6). Further examination of peptide identifications revealed that protein S-Fe9_GL0178463 had high-quality PSMs (scores up to 323) in all samples in the LFD group, but no PSM was obtained for HFD groups (Additional file 2: Figure S4). The MaxQuant software has been designed to identify and quantify peptides from precursor ions without any MS/MS scan, by matching to the high-quality PSMs in paralleled samples based on retention time and mass-to-charge (m/z) ratio . This resulted in an average of three peptides quantified for the samples in HFD group (Additional file 2: Figure S4). The lack of PSM for protein S-Fe9_GL0178463 in the HFD group might be the result of low protein levels in the samples, such that their precursor ions were not abundant enough to be selected by mass spectrometer for fragmentation. However, the precursor ions were still present allowing accurate quantification and thereby selection of key proteins with huge differences between the different groups. This scenario, where certain peptides/proteins are almost totally absent from a single group, is expected to be common in gut microbial studies, where some species may be substantially inhibited by other species or environmental compounds in one group as compared to another.
MetaPro-IQ approach provides a universal, easy, and high-performance metaproteomic workflow for studying human or mouse gut microbiota and dramatically improves microbial protein identification and quantification. The workflow was established based on the usage of the close-to-complete gut microbial gene catalog database and an iterative database search. These enabled high protein-database coverage and better sensitivity for peptide identification. More importantly, it allows easy comparison for the results obtained from different studies. The MetaPro-IQ approach also makes matched metagenome sequencing unnecessary for general gut metaproteomic studies. Moreover, the workflow generated quantitative results based on precursor intensity and advanced normalization method, which allows accurate comparison across samples.
The limitation of the MetaPro-IQ also happens to the gene catalog databases since they are mainly generated from stool samples and will not represent all types of intestinal microbiomes, for example, the biopsy microbiomes. For uncommon samples, the gene catalog database could be enhanced with additional matched sample metagenomic sequencings, and thus will improve the performance of MetaPro-IQ. In addition, the current gut microbial gene catalog databases represent well for the gut-related bacteria and archaea; the studies focusing on other taxa such as fungi and virus will need further addition of their protein sequences or alternative databases. However, the concept of MetaPro-IQ is applicable for any research as long as the gene catalog or reference protein database is available.
Mouse stool sample collection and processing
The animal experiments were performed at the Ottawa Hospital Research Institute and conducted in strict accordance with the guidelines on the Care and Use of Experimental Animals of Canadian Council on Animal Care (CCAC). The animal use protocol (2009-012) was approved by the Animal Care Committee at the University of Ottawa. A total of eight male C57BL/6J mice (Charles River, Sherbrooke, QC) were housed individually in the same room at 25 °C with a strict 12-h light/dark cycle. Food and water were available ad libitum. Mice were acclimatized to the facility for 2 weeks and fed a normal chow diet (containing 18 % fat by energy; Harlan Laboratories, Inc., Madison, WI) and then randomly divided into two groups (n = 4/group). One group of mice was fed a high-fat diet (containing 39.7 % fat by energy; TestDiet, St. Louis, MO), and the other group a low-fat diet (containing 15.8 % fat by energy; TestDiet) for 43 days. Body weight for each mouse was monitored weekly. Stool samples were collected at days 0, 14, 29, and 43 and stored at −80 °C until analysis.
For proteomic analysis, bacterial proteins were extracted from the stool samples. Briefly, stool samples (~1 g) were suspended in 1.5 mL ice-cold PBS (pH 7.0) with thorough vortexing. The slurries were centrifuged at 300g, 4 °C for 5 min. Supernatants were carefully collected, and the pellets were subjected to the above procedure three times. All the supernatants for each sample were then combined followed by three more centrifugations at 300g, 4 °C for 5 min to remove debris. The supernatant was then centrifuged at 14,000g, 4 °C for 20 min to pellet bacterial cells. The pellet was then re-suspended in fresh PBS and washed another three times. The bacterial cells were then lysed with 4 % sodium dodecyl sulfate (SDS) and 6 M urea in 50 mM Tris-HCl buffer (pH 8.0) with sonication. SDS was removed by protein precipitation in acidified acetone/ethanol buffer at −20 °C overnight. The precipitated proteins were dissolved in 6 M urea in 50 mM ammonium bicarbonate (pH 8) for trypsin digestion.
Human MLI sample collection and processing
Eligible subjects were under 18 years of age and scheduled to undergo diagnostic colonoscopy (Additional file 1: Table S1). The protocol was approved by the Research Ethics Board of the Children’s Hospital of Eastern Ontario (CHEO). Colonoscopy preparation was done as per standard protocol modified to 1 day . During colonoscopy, once the proximal ascending colon was intubated, any fluid and loose debris was aspirated and discarded. Thereafter, sterile water was flushed onto the mucosa to dislodge adherent mucus from mucosal epithelial cells and the mixture was then aspirated into a sterile container through the colonoscope. The samples were immediately placed on ice and transported to the lab for processing.
To remove any debris in the aspirate sample, a first centrifugation at 700g, 4 °C for 5 min was performed and the supernatant was transferred into a new tube. The bacterial cells were then collected with a centrifugation at 14,000g, 4 °C for 20 min. The pellets were then used for protein extraction according to the procedures described above.
Liquid chromatography-tandem mass spectrometry
In-solution trypsin digestion for bacterial proteins recovered from both mouse stool and human MLI samples was conducted as described previously . Briefly, the proteins were first reduced and alkylated with 10 mM dithiothreitol (DTT) and 20 mM iodoacetamide (IAA), respectively. The urea concentration was then diluted to <1 M with 50 mM ammonium bicarbonate. Trypsin (Worthington Biochemical Corp., Lakewood, NJ) was then added at a protein-trypsin ratio of 50: 1 (w/w) for digestion overnight with agitation at 37 °C. The tryptic digest was desalted with a 10-μm C18 column and eluted with 80 % acetonitrile/0.1 % formic acid. The eluent was then evaporated with a Speed-Vac concentrator and tryptic peptides dissolved in 0.1 % formic acid for mass spectrometry analysis.
Tryptic peptides equivalent to 4 μg of proteins were loaded for liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis on a Q Exactive mass spectrometer (ThermoFisher Scientific Inc.). The separation of peptides was performed on an analytical column (75 μm × 50 cm) packed with reverse phase beads (1.9 μm; 120-Å pore size; Dr. Maisch GmbH, Ammerbuch, Germany). A 4-h gradient was performed from 5 to 35 % acetonitrile containing 0.1 % formic acid at a flow rate of 200 nL/min. The instrument method consisted of one full MS scan from 300 to 1800 m/z followed by data-dependent MS/MS scan of the 12 most intense ions, a dynamic exclusion repeat count of 2, and repeat exclusion duration of 30 s. All data were recorded with the Xcalibur software and exported as.raw format for further analysis.
Metagenomic DNA extraction, sequencing, and gene prediction
Total DNA was extracted from intestinal aspirate samples using the Fast DNA spin kit (MP Biomedicals, Santa Ana, CA) and using a FastPrep-24 (MP Biomedicals). Briefly, MLI samples were thawed and contents pelleted by centrifuging at 14,000g for 10 min in a bench top centrifuge. The pellets were re-suspended in 1 mL of cell lysing solution (CLS)-TC and subjected to two mechanical lysis cycles at speed 6.0 for 40 s. The extracted DNA was then used to construct sequencing libraries using an Illumina TruSeq DNA Sample Prep kit v3 according to the manufacturer’s instructions. The sequencing was performed on an Illumina HiSeq 2000 (generating paired-end 100-bp reads) at the Génome Québec Innovation Centre, McGill University (Montreal, Canada).
Gene prediction was performed on each sample individually using the previously published MOCAT pipeline . Briefly, raw reads were first filtered and trimmed to remove sequencing adapters and low quality reads. Reads with human origin were removed using SOAPAligner 2 against the human genome database (hg19) . The remaining reads were then used for scaftig assembly and assembly revision to generate assembled sequences for gene prediction with the MetaGeneMark algorithm . The resulting gene sequence lists for each of the sample were then compiled into FASTA files and used as a matched metagenome database for benchmarking MetaPro-IQ.
Bioinformatics for metaproteome data analysis
Implementation of MetaPro-IQ approach
The implementation of the MetaPro-IQ approach is illustrated in Fig. 1 and details are highlighted in the “Results and discussion” section. The human and mouse gut microbial gene catalog databases were downloaded from the IGC website (http://meta.genomics.cn/) and the GigaScience Database (http://gigadb.org/dataset/view/id/100114/token/mZlMYJIF04LshpgP), respectively [20, 24].
In MetaPro-IQ, the first- and second-step database searches were carried out with X! Tandem (release 2015.04.01) [44, 45], and the third step was carried out with MaxQuant software (version 126.96.36.199) . For the X! Tandem database search, each raw file obtained in the current study or from Tanca et al.’s study (downloaded from PeptideAtlas Repository at http://www.peptideatlas.org/PASS/PASS00355) was converted into mgf format with an in-house software platform. The tandem search was performed with up to two miss-cleavages (trypsin/P), carbamidomethylation of cysteine as a fixed modification, and oxidation of methionine as a potential modification. A fragment ion tolerance of 20 ppm and a parent ion tolerance of 10 ppm were used. All matched protein sequences for the first-step search were extracted as the sample-specific database (sample_all-macth.fasta). The X! Tandem outputs of the target-decoy database search (step 2) were summarized with an in-house software to generate an identified protein list at a FDR cutoff of 0.01 and maximum expect value of 0.05. The resulting protein list for all samples was then combined, and duplicates were removed for generating a “combined non-redundant database” to use for protein quantification using MaxQuant. Similar peptide identification parameters with X! Tandem database searches were used for MaxQuant (parameter files were uploaded together with the raw data and result files to ProteomeXchange). For quantification, the LFQ algorithm was used for label-free quantification. Both razor and unique peptides were used for protein quantification, and the minimum ratio count was set as 1. An alignment retention time window of 20 min and match time window of 5 min were applied to match the same accurate masses between different runs. Proteins identified by the same set or a subset of peptides were grouped together as one protein group.
Matched metagenome database search strategy
Matched metagenome database searches were performed with exactly the same parameters as the MetaPro-IQ approach, except that the metagenome database (gene sequences) was used for each sample instead of the whole gene catalog database. To ensure a fair comparison, the same multi-step database search strategy and MaxQuant quantification was performed. The outputs of MaxQuant (namely the results included in the txt folder) were then compared between the matched metagenome and MetaPro-IQ approaches.
Functional annotation into COGs
All the quantified protein sequences were aligned against the COG database (ftp://ftp.ncbi.nih.gov/pub/COG/COG2014/data) with DIAMOND using default parameters (e-value cutoff of 0.001) [47, 48]. The best hit for each query was selected for annotation, and the COG id, name, and category information for each of the matched sequences were extracted from the annotation file of the COG database. If the leading protein (defined as the top rank protein in a group; ranking is based on the number of peptide sequences, the number of PSMs, and the sequence coverage) in a protein group had no match for COG id, the others were checked for the existence of any match. The LFQ intensity of all the protein groups annotated with the same COG category were summed together to represent the COG category abundance for each sample.
Taxonomic analysis was performed based on taxon-specific unique peptides using Unipept (version 3.1) . All of the quantified peptide sequences from MaxQuant were imported into the Unipept for analysis with the “Equal I and L” and “Advanced miss cleavage handling” options allowed. The results were then exported as a table format. The relative abundance of each taxon (namely phylum or species in this study) is represented as the sum of intensity for all the unique peptides found with Unipept.
Multivariate statistical analysis
The protein group results from MaxQuant were first imported into Perseus (version 188.8.131.52) to remove any contaminants, reverse sequences, and those identified only by site. The LFQ intensity was log10-transformed and used for statistical analysis. Only those protein groups with valid LFQ intensity values in at least 29 of the 32 samples (>90 %) were kept for subsequent statistical and multivariate analysis. This resulted in a stringently quantified dataset of 5299 protein groups for the bacteria from 32 stool samples. Hierarchical clustering analysis was also conducted in Perseus with default parameters. Significant proteins between HFD and LFD groups were identified by a two-sample t test with Benjamini-Hochberg FDR correction. Those proteins with a FDR corrected p value (q value) of <0.05 were considered as significant proteins.
PCA was performed in MATLAB (version 2010b, The MathWorks Inc.), and the first two PCs were used for generating the score plot. A prior missing value imputation was done with the nearest-neighbor method in MATLAB with the knnimpute function (http://www.mathworks.com/help/bioinfo/ref/knnimpute.html) .
COG, Clusters of Orthologous Groups; DTT, dithiothreitol; FDR, false discovery rate; HFD, high-fat diet; HMP, Human Microbiome Project; IAA, iodoacetamide; IBD, inflammatory bowel disease; LC-MS/MS, liquid chromatography-tandem mass spectrometry; LFD, low-fat diet; MetaHIT, Metagenomics of Human Intestinal Tract; MetaPro-IQ, metaproteome identification and quantification; MLI, mucosal-luminal interface; PCA, principal component analysis; PEP, posterior error probability; PSM, peptide-spectra matching; SDS, sodium dodecyl sulfate
We acknowledge the assistance of R. Singleton, A. Mack, and D. McCulloch in making this study possible.
This work was supported in part by the Genome Canada and the Ontario Genomics Institute (OGI-067), CIHR grant number GPH-129340, the Ontario Ministry of Economic Development and Innovation (REG1-4450), the Natural Sciences and Engineering Research Council of Canada (NSERC) council of Canada, and the Foundation Jean-Louis Lévesque. DF acknowledges a Canada Research Chair in Proteomics and Systems Biology. JB is the recipient of a CIHR/CAG Postdoctoral Award.
Availability of data and materials
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://www.proteomexchange.org) via the PRIDE  partner repository with the dataset identifiers PXD003527 and PXD003528, and all the raw reads of Illumina sequencing have been submitted to NCBI sequence read archive (SRA) under the accession of SRP068619.
DF, AS, DM, and XZ designed the study. XZ, JM, JIM, JL, JB, SAD, and CKC performed the experiments. XZ, ZN, RC, and MW analyzed the data. XZ and DF wrote the manuscript, and all the authors contributed to the editing and revision of the manuscript. All authors read and approved the final manuscript.
The authors declare that they have no competing interests.
Consent for publication
Consent to publish was obtained from parents of all the children.
Ethics approval and consent to participate
The animal use protocol (2009-012) was approved by the Animal Care Committee at the University of Ottawa, ON, Canada. The clinical sample collection protocol was approved by the Research Ethics Board of the Children’s Hospital of Eastern Ontario (CHEO), Ottawa, ON, Canada. The consent to participate was obtained for all the participants.
- 16.Kolmeder CA, de Been M, Nikkila J, Ritamo I, Matto J, Valmu L, Salojarvi J, Palva A, Salonen A, de Vos WM. Comparative metaproteomics and diversity analysis of human intestinal microbiota testifies for its temporal stability and expression of core functions. PLoS One. 2012;7(1):e29913.CrossRefPubMedPubMedCentralGoogle Scholar
- 30.Song BJ, Akbar M, Abdelmegeed MA, Byun K, Lee B, Yoon SK, Hardwick JP. Mitochondrial dysfunction and tissue injury by alcohol, high fat, nonalcoholic substances and pathological conditions through post-translational protein modifications. Redox Biol. 2014;3:109–23.CrossRefPubMedPubMedCentralGoogle Scholar
- 36.Sokol H, Pigneur B, Watterlot L, Lakhdari O, Bermudez-Humaran LG, Gratadoux JJ, Blugeon S, Bridonneau C, Furet JP, Corthier G, et al. Faecalibacterium prausnitzii is an anti-inflammatory commensal bacterium identified by gut microbiota analysis of Crohn disease patients. Proc Natl Acad Sci U S A. 2008;105(43):16731–6.CrossRefPubMedPubMedCentralGoogle Scholar
- 37.Kanauchi O, Fukuda M, Matsumoto Y, Ishii S, Ozawa T, Shimizu M, Mitsuyama K, Andoh A. Eubacterium limosum ameliorates experimental colitis and metabolite of microbe attenuates colonic inflammatory action with increase of mucosal integrity. World J Gastroenterol. 2006;12(7):1071–7.CrossRefPubMedPubMedCentralGoogle Scholar
- 49.Webb-Robertson BJ, Wiberg HK, Matzke MM, Brown JN, Wang J, McDermott JE, Smith RD, Rodland KD, Metz TO, Pounds JG, et al. Review, evaluation, and discussion of the challenges of missing value imputation for mass spectrometry-based label-free global proteomics. J Proteome Res. 2015;14(5):1993–2001.CrossRefPubMedPubMedCentralGoogle Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.