Background
The gut microbiome of an animal consists of bacteria, viruses, fungi and so on. This intricate ecosystem interacts with the adjacent epithelial layer, and the microbes perform metabolic functions, protect against pathogens, and condition the immune system, and through these basic functions, these microbes directly or indirectly affect most of the physiological functions of the host [
1,
2]. In recent years, variations in the bacterial community composition have been shown to correlate with infection outcome, inflammatory bowel disease [
3], diabetes [
4], obesity [
5], and depression [
6], and fecal microbiome transplantation has become an effective treatment for refractory
Clostridium difficile infections and other diseases [
7,
8]. The mechanisms of interaction between the gut bacterial microbiome and the host are very complex, and other components also play crucial roles in this process. In addition to bacteria, viruses are also abundant in the gut [
9] and have been hypothesized to markedly alter the structure and function of the bacterial community [
10‐
12]. Additionally, chronic viral infection can confer increased resistance against pathogenic challenges [
13]. Gut virome alteration has been observed in inflammatory diseases such as inflammatory bowel disease and Crohn’s disease [
14]. The recent advent of high-throughput sequencing methods has made it possible to study these communities and their relationships with health and disease in detail [
15].
Bacterial communities play an essential role in host health, but further research is still warranted to obtain an in-depth understanding of the mechanisms underlying this role. Transfer of whole virome communities between humans was documented in fecal microbiome transplantation [
16], and the difference varied more widely between gut viromes than between gut bacterial microbiomes in humans [
17]. However, the relationship between the bacterial microbiome and the virome has rarely been studied, despite its likely medical importance. Previous research has shown close relationships between single viral species and single bacterial species [
18,
19], and single viral species could trigger shifts in the bacterial microbiome and the virome [
20,
21]. At the same time, enteric bacteria were seen to be required for efficient infection by [
22,
23] or suppression of [
24] viruses, and the richness of the gut bacterial microbiome had an obvious effect on bacteriophage composition [
25]; moreover, the gut virome composition in humans was examined, and bacteriophage diversity was found to be inversely correlated with naturally occurring bacterial diversity in human infants during healthy development [
26]. However, few studies have focused on how the whole virome, a diverse community consisting of eukaryotic RNA and DNA viruses and bacteriophages, interacts with the bacterial microbiome.
Rhesus monkeys are good mammalian research models that are closely related to humans, and the virome composition of these animals was seen to be affected by simian immunodeficiency virus infection [
20]. We hypothesized that there is a close relationship between the whole gut virome and bacterial microbiome, and the bacterial microbiome could be depleted by treatment with an antibiotic cocktail in rhesus monkeys. We then examined the virome composition to detect the direct effects of the bacterial microbiota on the virome. We performed 16S rRNA amplicon sequencing of the fecal bacteria and metagenomic analysis of fecal viromes from rhesus monkeys treated with an antibiotic cocktail. Our results suggest that a majority of bacteria were depleted after the monkeys were treated with antibiotics and that the composition of the whole virome changed drastically. Importantly, alteration of the virome along with shifts in the composition and function of the gut bacterial community and metabolites from gut bacteria may have played an important role in the interaction.
Materials and methods
Animals
The rhesus monkey cohort described in this study was housed at the Institute of Medical Biology, Chinese Academy of Medical Sciences (IMBCAMS). An antibiotic cocktail containing ampicillin, streptomycin, kanamycin, metronidazole, and vancomycin was administered orally at a dose of 15 mg/kg 3 times per day for 2 weeks. Three healthy one-year-old rhesus monkeys were treated with antibiotics, and fresh fecal samples were collected one day before treatment with antibiotics and 5, 8, and 9 days after treatment with antibiotics and stored at − 80 °C for subsequent analysis. Fresh fecal samples from an additional three normal monkeys were collected after the monkeys were treated with antibiotics for 9 days. The bacterial community of each sample was detected by 16S rRNA amplicon sequencing, and the virome communities in the samples collected before treatment with antibiotics and in those collected after treatment with antibiotics for 9 days were detected by deep sequencing. Because metabolome analysis requires 6 biological duplications, the metabolomes of samples collected before treatment with antibiotics and of those collected after treatment with antibiotics for 8 and 9 days were detected by GC-MS and LC-MS. In our analysis, samples collected before treatment with antibiotics were used as the control group, and samples collected after treatment with antibiotics were used as the experimental group.
Bacterial 16S rRNA amplicon sequencing
DNA was extracted from fecal samples, and PCR was performed with the barcode primers 338F/806R to obtain amplicons of hypervariable regions V3 and V4 for phylogenetic discrimination analysis [
27]. Libraries were pooled by using a TruSeqTM DNA Sample Prep Kit and sequenced using an Illumina MiSeq sequencer. Sequences were assigned to closed-reference operational taxonomic units (OTUs) at a 97% identity threshold using bacterial 16S rRNA amplicon sequences from the Silva 128/16S-bacteria database. The OTU data were rarefied to the smallest effective sample sizes [
28]; rarefaction is a homogenization method that is used to randomly draw OTUs to the same quantity based on a minimum value. The α diversity, which includes the abundance-based coverage estimator (ACE) and Shannon diversity index, was analyzed by Mothur (version v.1.30.1), and statistical significance was evaluated by Student’s t-test.
Kyoto encyclopedia of genes and genomes (KEGG) prediction analysis of the bacterial microbiome [29, 30]
We performed a Kyoto Encyclopedia of Genes and Genomes (KEGG) prediction analysis of the bacterial microbiome using PICRUSt. PICRUSt contains the Cluster of Orthologous Groups of proteins (COG) and KEGG Ortholog (KO) information corresponding to Greengene ID numbers. For metagenome prediction, PICRUSt takes an input OTU table containing identifiers that match tips from the marker gene with corresponding abundances for each of the OTUs across one or more samples. First, PICRUSt normalizes the OTU table based on 16S rRNA amplicon copy number prediction so that the OTU abundances accurately reflect the true abundances of the underlying organisms. The metagenome is then predicted by looking up the precalculated genome content for each OTU, multiplying the normalized OTU abundance by each KO abundance in the genome and summing these KO abundances together per sample. The prediction yields a table of KO abundances for each metagenome sample in the OTU table.
Analysis of similarities (ANOSIM)
Analysis of similarities (ANOSIM) is a nonparametric test that shows whether the difference between groups is greater than that within groups. The analyses were performed in vegan or QIIME in R (version3.2.2) by using the Bray-Curtis algorithm [
31].
Virome DNA and RNA purification and sequencing
Fecal samples were suspended in phosphate-buffered saline (PBS) and filtered through a filter with a pore size of 0.45 μm (Millipore). The supernatant was enriched by a 30-kDa molecular mass filter (Ultra-15 30 K, Millipore). The concentrate was treated with DNase I (TaKaRa) at 37 °C for 30 min to eliminate unencapsulated nucleic acids. Subsequently, total viral DNA was extracted from half of the concentrate using the QIAamp DNA Stool Kit (Qiagen), and at the same time, total viral RNA was extracted from the other half using the QIAamp Viral RNA Kit (Qiagen). The extracted RNA was synthesized into double strands using the NEBNext RNA First Strand Synthesis Module (NEB) and the NEBNext mRNA Second Strand Synthesis Module (NEB). The DNA and double-stranded cDNA were amplified by whole-genome amplification (REPLI-g Mini Kit, Qiagen) and then fragmented into approximately 300-bp fragments by a Covaris M220 instrument. Then, the fragments were amplified into a PE library by the TruSeq™ DNA Sample Prep Kit and fixed to the chip by bridge PCR using the HiSeq 3000/4000 PE Cluster Kit. The constructs were sequenced on the Illumina HiSeq platform using HiSeq 3000/4000 SBS Kits.
For virome analysis, we first sheared the adaptor sequences with Seqprep and removed reads that were shorter than 50 bp and those that contained N bases with Sickle to retain the paired-end reads and single-end reads. We compared these reads to the host (rhesus monkey) genome by BWA and removed the reads belonging to the host. Then, we compared all the clean reads with the U.S. National Center for Biotechnology Information (NCBI) Nucleotide database to identify the sequences that belonged to viruses and the sequences that did not belong to any known genome, such as those of bacteria, fungi or other known microorganisms. Then, contigs were built from these reads. The contigs and all reads that could not be mapped to any known genome in NCBI were compared with the virus protein database in the NCBI nonredundant RefSeq database (including sequences from SwissProt, PIR, PRF, and PDB and coding sequences (CDS) from GenBank and RefSeq) based on amino acid sequences using BLASTP (BLAST version 2.2.31+, e-value: 1e-5). These results constituted our virus database and were used to obtain the nonredundant gene catalog by CD-HIT. All the reads were compared to our virus database to analyze their richness. The spliced read alignments were predicted by MetaGene, compared to the EggNOG database and Virulence Factors database (VFDB) for COG analysis using BLASTP (BLAST version 2.2.31+, e-value: 1e-5) and annotated using VFDB.
PCR validation
The abundance results that were similar in 2 or more monkeys were selected, and real-time PCR was used to validate the changes in these viruses (SYBR Premix Ex Taq II, TaKaRa). As the template, we used the DNA and cDNA extracted from fecal samples. The samples that were not detected directly from the DNA or cDNA were subjected to multiple displacement amplification (MDA) (total nucleic acid was amplified by multiple displacement to comprehensively detect both DNA and RNA viruses [
26]) and then analyzed by real-time PCR. For real-time PCR, we used the Ct numbers to show the richness of the virus. Viruses that were not detected were not shown. Primers were designed to amplify specific regions in the
Bdellovibrio phage phiMH2K (5′-AATCCTCAATTCCAGACTTCCA-3′ (F) and 5′-CCATTTCCATAAGTCCGAGTG-3′ (R)), Bacillus phage B103 (5′- TGGCGATGTTGATGATGAC-3′ (F) and 5′-CTTTATTTGCGTCTGTTGTCG-3′ (R)), columbid circovirus (5′-TCAGGAGACGAAGGACACG-3′ (F) and 5′- TGGCATCATACATCGGGAC-3′ (R)), potato virus M (5′-CGCTTCGCTGCTTTCG − 3′ (F) and 5′-CGGACCATTCATACCACCA-3′ (R)),
Marseillevirus marseillevirus (5′-AAAGTCCCAAGTTATCACAAGC-3′ (F) and 5′- TTTCTCGCAGCGTCAATG-3′ (R)), simian sapelovirus (5′- TTCCATCTGCTCTAAATGCTCA-3′ (F) and 5′-CAGCAGTTAGAGCGGGTG-3′ (R)), and Andean potato mild mosaic virus (5′-AAGCCCAACATCGTTCTCC-3′ (F) and 5′- AAGAGGATACGGGAGAAAGG-3′ (R)).
Redundancy analysis (RDA)
Redundancy analysis (RDA) shows the interactions between sample distribution and environmental factors. We used vegan’s RDA analysis in R with the phylum-level abundances of the bacterial microbiome as environmental factors.
Regression analysis
We ran a regression analysis between bacterial microbiome diversity and virome richness with the stats package and plotted the results using the ggplot2 package.
The stool samples were suspended in methanol:H2O (4:1), ground, ultrasonicated, concentrated and dried so that the metabolome could be analyzed by GC-MS and LC-MS.
GC-MS
The derivatized samples were analyzed on an Agilent 7890A gas chromatography system coupled to an Agilent 5975C MSD system (Agilent). An HP-5 MS fused-silica capillary column (30 mm × 0.25 mm × 0.25 μm, Agilent) was utilized to separate the derivatives. Helium (> 99.999%) was used as the carrier gas at a constant flow rate of 6.0 mL/min through the column. The injector temperature was maintained at 280 °C. A volume of 1 μL was injected in splitless mode. The oven temperature was initially 60 °C and was then ramped up to 125 °C at a rate of 8 °C/min, to 190 °C at a rate of 10 °C/min, to 210 °C at a rate of 4 °C/min and to 310 °C at a rate of 20 °C/min; finally, the temperature was held at 310 °C for 8.5 min. The temperatures of the MS quadrupole and ion source (electron impact) were set to 150 °C and 230 °C, respectively. The collision energy was 70 eV. Mass data were acquired in full-scan mode (m/z 50–600), and the solvent delay time was set to 5 min. The acquired MS data from GC-MS were analyzed by ChromaTOF software (v 4.34, LECO, St Joseph, MI). Metabolites were qualitatively assessed by the Fiehn database, which is linked to ChromaTOF software. Briefly, after alignment with the Statistic Compare component, a CSV file was obtained with three-dimensional data sets, including sample information, peak name, retention time, m/z and peak intensities. The resulting data were normalized to the total peak area of each sample in Excel 2007 (Microsoft, USA) and imported into SIMCA (version 14.0, Umetrics, Umeå, Sweden) to define the 95% confidence interval of the modeled variation. The differential metabolites were selected on the basis of the combination of a statistically significant threshold of variable influence on projection (VIP) values obtained from the OPLS-DA model and p values from a two-tailed Student’s t-test on the normalized peak areas, where metabolites with VIP values larger than 1.0 and p values less than 0.05 were included.
LC-MS
LC-MS was performed on an Ultimate 3000-Velos Pro system equipped with a binary solvent delivery manager and a sample manager coupled with an LTQ Orbitrap mass spectrometer equipped with an electrospray interface (Thermo Fisher Scientific); an Acquity BEH C18 column (100 mm × 2.1 mm i.d., 1.7 μm; Waters) was used. The column was maintained at 45 °C, and separation was achieved using the following gradient: 5% B–25% B from 0 to 1.5 min, 25% B–100% B from 1.5 to 10.0 min, 100% B from 10.0 to 13.0 min; 100% B–5% B from 13.0 to 13.5 min, and 5% B from 13.5 to 14.5 min at a flow rate of 0.40 mL/min, where B was acetonitrile (0.1% (v/v) formic acid), and A was aqueous formic acid (0.1% (v/v) formic acid). The injection volume was 3.00 μL, and the column temperature was set at 45.0 °C. The mass spectrometric data were collected using an LTQ Orbitrap mass spectrometer equipped with an electrospray ionization (ESI) source operating in either positive or negative ion mode. The capillary and source temperatures were set at 350 °C, with a desolvation gas flow of 45 L/h. Centroid data were collected from 50 to 1000 m/z with a resolution of 30,000. XCMS (
http://masspec.scripps.edu/ xcms/xcms.php) was used for nonlinear alignment of time domain data and automatic integration and extraction of the peak intensities. Default XCMS parameter settings were used (major default parameters: profmethod = bin; method = matched filter; step = 0.1) except for full width at half maximum = 8, bandwidth (bw) = 6 and snthresh = 5. Variables with < 30% relative standard deviation (RSD) in QC samples were then retained for further multivariate data analysis. The result was a three-dimensional matrix that included retention time and m/z pairs (variable indices), sample names (observations), and normalized ion intensities (variables). The positive and negative data were merged into a combined data set, which was imported into SIMCA-P+ 14.0 software (Umetrics, Umeå, Sweden). The differential metabolites were selected on the basis of a combination of statistically significant VIP values obtained from the OPLS-DA model and
p values from a two-tailed Student’s t-test on the normalized peak areas, where metabolites with VIP values larger than 1.0 and p values less than 0.05 were included. The differential metabolites were qualitatively assessed using the Human Metabolome Database (
http://www.hmdb.ca/) and METLIN (
https://metlin.scripps.edu/).
Discussion
As reported by Adina Howe, Yatsunenko T and Alejandro Reyes, in the same environment and feeding conditions, the composition of the microbiota and virome could remain stable within an individual [
17,
46,
47]. However, the gut microbial composition could be influenced by multiple interacting factors, such as diet [
46], antibiotic use [
48], age, geographical setting [
47], and several diseases, including chronic inflammation, obesity and diabetes [
4]. In our study, the major reason for depletion of the gut bacterial microbiota was treatment with the antibiotic cocktail. The feeding conditions of the rhesus monkeys were stable in terms of their food and water consumption, and blood samples were monitored routinely, showing that there was no infection during the study period (Additional file
1: Figure S1). The bacterial composition exhibited stable and continuous depletion after treatment with the antibiotic cocktail, and we found that the virome composition changed noticeably and was correlated with the shifts in the bacterial community. Moreover, we found that metabolites produced by the gut bacterial microbiome may play a role in the interrelation. In addition, we found that the composition of the rhesus monkey enterovirus group was similar to that of the human enterovirus group [
26], and our results may be beneficial for research on the composition of the human virome.
When the bacterial microbiome was depleted, ampicillin could kill most bacteria, including gram-positive and gram-negative bacteria; streptomycin could kill most bacilli; kanamycin could kill most gram-negative bacteria; metronidazole could kill most anaerobic bacteria and parasites; and vancomycin could kill most gram-positive bacteria. Of course, the numbers of drug-resistant bacteria are increasing, but we believe that the cocktail of five antibiotics could deplete most of the commensal bacteria in the gut. As expected, the whole gut bacterial microbiome, including gram-positive and gram-negative bacteria (Additional file
2: Figure S2E), was depleted after treatment with the antibiotic cocktail, except for
Escherichia-
Shigella species belonging to Proteobacteria, which were resistant to the cocktail.
Escherichia harbored the most diverse antibiotic resistance genes, including genes resistant to multidrug treatments, tetracycline, aminoglycoside, macrolide-lincosamide-streptogramin B, β-lactams, and sulfonamides [
49]. We maintained the bacteria belonging to
Escherichia-Shigella in plates with the antibiotic cocktail. In the future, we will investigate the specific resistance and antibiotic resistance genes in this bacterium. Notably, our study focused on the interaction between virome composition and the bacterial microbiome in rhesus monkeys and may serve as a model for gut microbiota analysis. Therefore, we used the administration of 5 distinct antibiotics at high dosages and high frequencies for 2 weeks to deplete the whole gut bacterial microbiome. In our results, the richness and diversity of the bacterial community were depleted. Because our study did not involve clinical treatment, the normal dose of antibiotics was not evaluated by our procedure.
People are widely prescribed antibiotics each year [
50], and while antibiotics exert very complex effects on the whole bacterial microbiome [
48], the effects of these drugs on the virome are not clear. Antibiotics can directly affect viruses but do not exhibit a wide range of roles. Minocycline [
51], berberine, abamectin, ivermectin [
52], glycopeptides [
53], and teicoplanin [
54] could inhibit the corresponding viruses. In our study design, an antibiotic cocktail that included ampicillin, streptomycin, kanamycin, metronidazole, and vancomycin was administered orally. No study has yet reported that these antibiotics directly affect viruses.
Based on our results, the richness of these viruses was very low in the gut, and we had to use MDA to perform deep sequencing, although the detection by deep sequencing was very sensitive. We first characterized the shift in the gut virome by deep sequencing, and the samples were amplified by MDA. MDA is used as a general technique in virome research, especially for DNA virome detection [
26]. To a certain extent, the amplification read-out can also represent the virus quantity. However, MDA is not well suited to the detection of RNA viruses. The sequence-independent amplification (SIA) approach is more appropriate than MDA for detecting RNA viruses [
55]. In the future, we can use this approach to precisely detect RNA viruses. In this case, we validated the depletion of the virome composition, including DNA viruses, RNA viruses and bacteriophages, by real-time PCR. Although the number of cycles seemed high, these results were verified via three biological replicates, and the results of the no-template control (NTC) were not detected. In addition, we performed serial dilution of the in vitro transcribed RNA of coxsackievirus A16 to generate a standard curve and found that a Ct of 39.96 represents 23 genomic copies (data not shown). In our opinion, these viruses are components of the gut microbiome with low richness and may be involved in host physiology.
The metabolites produced by gut bacteria play very important roles in host physiology [
38], although the effects of these metabolites on virome composition have rarely been reported. Glycan [
39], glycosaminoglycan [
40], quinone [
41] and arginine [
42,
43] support the inhibition of viruses, while tryptophan [
44,
45] and fatty acids [
36] promote viral survival. Although most metabolites that can inhibit or promote viruses play roles in human viruses, tryptophan could promote the simian immunodeficiency virus in macaques [
45]. In addition, most pandemics originating in animals, such as severe acute respiratory syndrome and pandemic influenza, could start to appear because of ecological, behavioral, or socioeconomic changes [
56]. Many human viruses are zoonotic, and some human viruses, such as human enterovirus 71, can infect animals, especially monkeys [
32]. We believe that metabolites play roles in a broad spectrum of viruses and that changes in the metabolites may correlate with depletion of the virome. In our results, the level of quinone, which decreases the abundance of viruses, was increased in the gut metabolome, and the levels of some amino acids that promote the survival of viruses, such as tryptophan, were decreased. Importantly, glycosaminoglycan, which can reduce the populations of various viruses, was noticeably increased in the KEGG pathways of the bacterial microbiome, but we did not measure glycosaminoglycan levels in the present study. It is very difficult to detect glycosaminoglycan by metabolic scanning because glycosaminoglycan has a very high molecular weight. In the future, glycosaminoglycan levels could be measured by time-of-flight mass spectrometry. First, the polysaccharide needs to be dispelled, followed by detection of the monosaccharide to calculate the polysaccharide levels based on the relationships among the monosaccharides in a specific database. However, this process is very complicated, and the database is not sufficiently large at present. By analyzing the relevant data, we found that depletion of bacteria directly promoted changes in the concentrations of some metabolites, which may play important roles in reducing the abundance of DNA viruses.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.