撰写In order to study the influence of this microbiome on the training immunity induced by BCG vaccine, we inoculated 321 healthy individuals (aged 18-75, with a median age of 23, of whom 57% were women, and 83% had a body mass index (BMI) between 18.5 and 25) with BCG vaccine. Before participants were vaccinated with BCG vaccine, stool and blood samples were collected. Blood was collected 2 weeks and 3 months after inoculation, as shown in fig. 1A. DNA was extracted from 321 fecal samples (see method), amplified and sequenced. Subsequently, peripheral blood mononuclear cells (PBMC) isolated from the participants at each time point were stimulated in vitro with Staphylococcus aureus or Mycobacterium tuberculosis. IL-6, IL-1β and TNF-α were measured 24 hours later, and IFN-γ was measured 7 days later. In addition, the metabolites in peripheral blood plasma were analyzed at the starting point inspection. Our method includes:(1) assembling all microbial species, (2) testing the regulatory effect of microorganisms on cytokine response, (3) evaluating whether immunomodulatory microorganisms have stronger effects on circulating metabolites than other microorganisms, and (4) determining which metabolites are related to immunomodulatory microbial species.(fig. 1B).
Fig. 1 Overview of study design and ab initio analysis of fecal microbiota samples. A Summary of study design: After collecting fecal and blood samples for analysis of metagenome, metabolomics and cytokine production in vitro, healthy adult volunteers (N=321) were vaccinated with BCG vaccine at the starting point. At the starting point, after 2 weeks and after 3 months, PBMC isolated from the subjects were stimulated in vitro with Staphylococcus aureus and Mycobacterium tuberculosis. B association mining strategy: compare the abundance of metagenomic categories pan genomes (MSPs) of a single metagenomic species with the multiple changes of cytokine expression after 2 weeks and 3 months. Subsequently, the significant MSP in the two models were evaluated for the enrichment of related circulating metabolites, and at the same time, the metagenome assembly was searched to find enzymes to explain the differential metabolic potential. C microbiome map: 345 MSP abundances were quantified by MSPminer software. MSP matched the known taxa at the levels of phylum (345), genus (315) and species (208). The bar chart shows the annotation level of MSP, which is divided into quartiles according to the number of genes. Samples based on Jensen-Shannon distance between relative MSP abundance; The D-tree diagram shows three clusters: the cut-off point (dotted line) is the maximum average difference of the distance between clusters and the distance within clusters (see Figure S1). E multi-dimensional scale (MDS) of microbial abundance in samples based on Jensen-Shannon distance.Figure: The sample is colored by clusters (C1–C3, as shown in D), and the cluster size is indicated in brackets. Three most abundant phylum (Firmicutes, Bacteroides and Actinomycetes) are highlighted. F Relative abundance of gate level detected by each sample: The percentage in brackets reports the abundance within the queue. The number of G MSP is significantly related to the questionnaire variables and stratified by door. Color coding is shown in F. The variation of relative abundance of each individual variable to each MSP was evaluated by Maaslin2 generalized linear model (FDR<0.25).
202 kinds of microorganisms were detected by reference-based MetaPhlAn 2.2 analysis, and more than 1.7 million microbial genes were produced by using MSPminer to reassemble the genome. 345 core metagenomic species’ pan-genomes (MSP) matched a phylum, of which 315 (91.3%) and 208 (60.3%) matched genus and species levels respectively, and more genes led to more accurate annotation (Figure 1C). Ab initio assembly is then used in all downstream analyses; In addition to searching all 202 initial species, this method also increases the size of the analyzed metagenome space by 1.7 times. Hereinafter, MSP is used to refer to a taxon as a gene set that matches a phylum, genus or species.
Community horizontal profile analysis using core MSP abundance produced three different groups of samples (Figure 1d; Annex 1: Figure S1A). The average cluster composition and sample location in the multi-dimensional scale map reflect the abundance of the three most common gates (Figure 1E). Firmicutes is dominant in Cluster 1, with an average relative abundance of 72.6%, followed by Bacteroides (16.64%, Cluster 2) and Actinomycetes (9.32%, Cluster 3) (Figure 1F). Cluster 1 also showed an increase in the relative abundance of verrucous colitis and Euryarchaeota (Annex 1: Figures S1B-E).The overall composition reflects the developed and western types of communities, which are often dominated by Firmicutes, and the existence of Bacteroides often reflects the changes driven by diet.This was confirmed by the indistinguishability of species level characteristics of an independent Dutch 500FG cohort composed of 473 healthy participants (Annex 1: Figure S1F-G).
All participants in our study were asked to complete a questionnaire containing general health-related questions. Eighteen of the 185 variables were related to three microbial groups (P<0.05, unadjusted Cauchy test), including diet (fruit intake), sex, smoking status, vaccination history and gastrointestinal status (such as Annex 1: Figure S2; Appendix 2: Table S1).
The change of relative abundance of a single species is related to 25 variables (MaAsLin2 generalized linear model, PFDR< 0.25), 10 of which are related to more species (fig. 1g; Appendix 2: Table S1). Especially vaginal yeast infection, sex and daily exercise are related to most microbial species (Figure 1G). Other species-specific variables missed by cluster association include hypertension, adolescent hay fever and antibiotic use.
We measured the production of cytokines in the supernatant of stimulated PBMC to evaluate the training immune response (IL-6, IL-1β and TNF-α after 24 hours of stimulation) and the specific immune response (IFN-γ after 7 days of stimulation). Compared with the starting point,Two weeks and three months after BCG vaccination, IL-1β and TNF-α stimulated by Staphylococcus aureus and IFN-γ stimulated by Mycobacterium tuberculosis gradually increased.(fig. 2a; Appendix 1: Figure S3A). After training, the immunity (expressed by the increase of TNF-α, IL-1β and IL-6 production) and the specific (IFN-γ induced by Mycobacterium tuberculosis) and allogeneic lymphocyte-derived (IFN-γ induced by Staphylococcus aureus) cytokine responses are correlated (Figure 2B, c). In any cytokine response measured after PBMC stimulation in vitro, the microbial group showed no significant difference (for Staphylococcus aureus IFN-γ, the maximum P=0.16, Annex 1: Figure S3B), which prompted us toSearching for potential individual taxa or functional immunomodulators..
Fig. 2 Abundance of microbial species varies with training and stimulating specific immune response. Expression of cytokines in vitro when stimulated by Mycobacterium tuberculosis or Staphylococcus aureus at the starting point (b), 2 weeks (2w) and 3 months (3m) after BCG vaccination.PThe value is determined by paired Kruskal-Wallis test. At 2 weeks (b) and 3 months (c), there was a significant correlation between the changes of cytokine expression, which was measured as a multiple change from the starting point level (Spearman correlation,P<0.05）。 Diagonal cells show significant changes of each individual cytokine relative to the starting point (t test, *P<0.05，***P<0.001）。 D significant MSP related to training immune response with FDR<0.2 and prevalence rate > 20%. The effect of each MSP was estimated by the linear model of the multiple changes of TNF-α, IL-1β and IL-6 with the starting point, and adjusted according to age and sex. E significant MSP related to specific response (IFN-γ stimulated by Mycobacterium tuberculosis), FDR<0.2, prevalence > 20%, and each MSP was not adjusted in the model of use and training and specific response.PA model with the same form of D.F distribution of values. The deviation of uniform distribution shows that MSP has a more significant effect on cytokine expression. G the individual effect of each MSP and time point on the change of cytokine folding is quantified as the explanatory variance multiplied by the effect symbol. The left side reports the association number of each MSP and the confidence of species-level annotations. At the following two time points, the abundance of H MSP 112 (Rosella) decreased by two of the three training immunophenotypes. The relative abundance of MSP is transformed into rank, and the rank is divided into several units by equal frequency division block. The p value in the title refers to the trained immune linear model. The coefficient in is Spearman correlation (*P<0.1，***P<0.001）。 I MSP 091 (Ruminococcus) and MSP 181 (Escherichia slowly) ranked the abundance and specific reaction phenotype. The relative abundance of MSP is transformed into rank, and the rank is divided into several units by equal frequency division block. In the titlePValue refers to the coefficient in the concrete response linear model as followsSpearmanCorrelation (*P<0.1，**P<0.01）。
The correlation between microorganisms and two immune system models was evaluated by two linear models, and adjusted according to age and sex to test the influence on training immunity and specific T cell response. After training, the immunity is related to 40 important species (prevalence rate > 20%, FDR<0.2, Figure 2D; Appendix 3: Table S2), and specific immunity is related to three species: Rumen coccus (MSP 091),Eggerthella lenta(MSP 181) and streptococcus thermophilus (MSP 076) (fig. 2e; Annex 3: Table S2). When FDR<0.1, 27 species were related to training immunity. Because the model of training immune response combines more biomarkers (three cytokines) than the specific response (one cytokine), it retrieves more species under the same significance threshold. Nevertheless, both methods have causedPThe uneven distribution of values (Figure 2F) supports the influence of microbiota on BCG vaccine response.
Next, we examined the correlation between species relative abundance and individual cytokine production at two time points after vaccination (Figure 2G). The genome of the genus Rosinia (MSP 112) can only be recognized at the genus level, and it is negatively correlated with the increase of IL-1 and TNF-α production at two time points (P=2.67e?05), FDR=0.0011, Figure 2H). In the specific reaction model, rumen cocci (MSP 091,P=0.0012, FDR=0.17) and Escherichia coli rotavirus (MSP 181,P=0.0030, FDR=0.17) showed consistent negative and positive effects on IFN-γ at two time points (Figure 2I). The specific response model reveals the correlation between a single cytokine and microbial species, although the unadjusted correlation number is between 10 and 23 microorganisms for all cytokine-stimulator combinations (e.g. Annex 4: Table S3).
Detectable changes in cytokine production have produced potential immunomodulatory species, the exact number of which varies with the selected statistical threshold. In order to confirm the correlation of cytokines and narrow the potential pathogenic mechanism, we comparedEffect of immunomodulatory microorganisms on plasma metabolomics. A total of 1607 mass/charge (m/z) peaks matched the characteristics of known compounds, and were divided into 20 groups, and concentrated for 8 main molecular categories.
Canonical correlation analysis confirmed thatThe Strong Influence of Microbiology on the Strength of Compounds, because 15 of the top 25 typical components (CCs) show significant differences in interpretation (e.g. Annex 1: Figure S4D-E). T-SNE prediction of the first 25 CCs shows that there is a strong grouping according to molecular categories, and microorganisms obviously coexist in clusters containing carboxylic acids and derivatives, glycerophospholipid and organic sulfuric acid (Figure 3A). Three immunomodulatory species, Rosickus sp. (MSP 112), Ruminococcus sp. (MSP 091) and Escherichia slowly (MSP 181), strongly affect the composition of the metabolite space, as shown by the nodes in Figure 3A.
Fig. 3 Immunomodulatory species strongly affect serum metabolomics. Canonical correlation analysis (CCA) was used to analyze the abundance of 345 MSP and 1607 annotated metabolites in all 321 samples. T-SNE is used to predict 25 typical components that explain the significant increase in variance in two dimensions. The positions of three immunomodulatory MSP (left) and metabolites (right) are highlighted. The joint deviation between the point and the origin (coordinates 0,0) reflects the increased covariance and grouping between metabolites and MSP. B MSP, which has an influence on cytokine response, has a great influence on the strength of metabolites. Maximum absolute Spearman correlation distribution of each MSP in all metabolites. MSP in at least 20% samples and 2 log-linear models.PSignificant msps of < 0.05 (left), < 0.1 (middle) and < 0.2 (right) are displayed in blue, and the rest of the insignificant msps are displayed in white. The vertical dashed line shows the thresholds of significantly related species and metabolites (PBonferroni<0.05）。 C All 32 significantly related metabolites were 34 MSPs. Color thermogram cells show the same relationship withPBonferroniSpearman correlation of < 0.05. Column (MSP) is the smallest of any one of the two log-linear models.PValue sorting. The asterisk indicates MSP related to cytokine response, as shown in Figure 2.
Next, we studied whether immunomodulatory microorganisms have a stronger influence on individual metabolites than other microorganisms. All paired Spearman correlation tests showed that there were 198 significant correlations between 34 MSP and 32 metabolite intensities (PBonferroni< 0.05) (e.g. Annex 5: Table S4). Although most of the correlations were positive (183/198), the resulting increase in metabolite intensity may be caused by microbial-mediated anabolism or compound decomposition. In order to test whether the species with great influence in the two linear models are more likely to have significant influence on the metabolomics, we used three different thresholds, selected at least 40 species, and calculated the maximum absolute influence on any metabolite. All three settings have produced significant enrichment effects (maximumP<10?11Chi-square test, Figure 3B). This confirmsMicrobial species that have an impact on cytokines also have a strong impact on circulating metabolites..
Of the 198 important metabolite associations, 18 were related to the existence of the genus Rosickus (MSP 112), which was the most among all MSP, and one was related to Escherichia tarda (MSP 181) (Figure 3C). Because of the uncertainty of annotation and the dependence on the abundance of metabolites mediated by pathway, we tried to determine the general mechanism by using KEGG pathway to group metabolites. The metabolism of phenylalanine is most strongly influenced by the genus MSP 112, and there are four metabolites in this pathway. The genus Ross (MSP 112) also affects three metabolites in tryptophan metabolism and secondary metabolite biosynthesis (Figure 3C).
Non-targeted plasma metabonomics detected two kinds of SCFA, acetate and propionate, which are candidate media for microbial products of dietary fiber metabolism and host microbial signals. These two molecules are positively correlated with training immunity, but have nothing to do with specific reaction (Annex 1: Figure S5). Although there is no significant correlation between immunomodulatory microorganisms and acetate or propionate, the positive correlation between Escherichia slow and propionate is the highest.
Butyrate, a SCFA which was previously proved to reduce the production of IL-6, was not detected in the circulation, probably due to its absorption in the intestine. Nevertheless, in one genus of Rosia (MSP 209) and five species of fecal cocci (CoprococcusButyrate kinase was detected in the species, which catalyzed the last step of butyric acid biosynthesis, among which fecal coccus (MSP 030) had the strongest positive correlation with trained immunity, especially the increase of IL-6 after vaccination (e.g. Appendix 1: Figure S6A-C). On the contrary,Fecal coccus (MSP 213) is related to the decrease of training immune response, which is manifested by the decrease of TNF-α expression after 3 months.(Annex 1: Figure S6D). The lack of butyric acid kinase in immunomodulatory Rothschild (MSP 112) may be due to the gap in the assembly genome, or allow the use of metabolic modes other than SCFA in mediating BCG-induced immunity.
The comprehensive evidence from the following aspects: (i) the simulated effect of microorganisms on cytokines, (ii) the correlation between the relative abundance of species at two time points after vaccination and the production of two kinds of cytokines related to training immunity, and (iii) the influence of immunomodulatory microorganisms on plasma metabolomics shows that the metabolism of phenylalanine mediated by Rosella affects training immunity.
The effect on phenylalanine metabolites prompted us to study the genome of Rosinia and confirm the existence of related enzymes. We compared the differential enzyme and metabolic potential of Rosinia with phylogenetic and biochemical related taxa through further subgenomic analysis of feces. Using core genes, all 345 identified MSP were grouped into phylogenetic differentiation groups (PhyloPhlAn 3.0, Appendix 1: Figure S7), which revealed thatA branch containing most species of the genus Rothschild.(fig. 4A). The most closely phylogenetic thick-walled bacteria include 5 strains of rumen cocci, 3 strains of fecal cocci and 2 strains of eubacterium rectum MSP. Although Escherichia coli is related to Rosella in butyric acid production, we also include a biologically unique but highly abundant Actinobacillus bifidus (MSP 111). The assembled group consisted of 59 MSP’s, which were from 85,208 assembled genes of Bifidobacterium (7 MSP), Rosickus (13 MSP), Rumex (17 MSP), Eubacterium (15 MSP) and Feces (7 MSP), and the prevalence rate of each genus in the queue was over 95% (Figure 4B).
Fig. 4 The phenylalanine pathway enzyme encoded by Rothschild is a universal and abundant enzyme. A subtree of phylogenetic relationships among detected MSPs contains most MSPs mapped to the genus Rothschild. Total abundance of each genus in queue b. C genes encoding 20 of 77 enzymes of phenylalanine metabolic pathway (KEGG orthologous ko00360) were assembled in the sample, and classified according to MSP of each genus. The thermogram shows the samples with detected genes. D the total abundance of a single gene in phenylalanine metabolic pathway. For each KEGG orthologous group, the statistical data of all assembly homologues were summarized. Compared with the other four genera in the group, the abundance (readings per million mapping, RPKM) of each enzyme encoded by the genus Rothschild increased. For the case of the highest abundance of Rosinia, it shows the minimum of four comparisons at most.PValue. Black dots represent the median, and error bars represent 25th and 75th percentiles. Spearman correlation of MSP in E group corresponds to 16 detected metabolite characteristics (mass-charge, m/z, mass spectrum peak) of phenylalanine metabolic pathway. Only the top matching compound names are displayed, while the ambiguous matches are displayed on the right. For each genus, the strongest effect related to any related MSP was selected and marked as significant subjects.P<0.05。 The effects related to MSP 112 (Rothschild) are highlighted in the black box..
A total of 330 genes correspond to 20 of 77 enzymes in the phenylalanine pathway (KEGG ko00360). Rosella contributed the most genes (92,45.1% average prevalence; Fig. 4c; Appendix 6: Table S5), while other genera in the group contributed less, including Rumen coccus (77,19.2%), Eubacterium (66,36.3%), Fecal coccus (50,35.5%) and Bifidobacterium (45,46.1%).Rosella also encodes phenylalanine (13 out of 20 species).. Among them,hisC、mhpE、yhDR、paaH、paaF、paaIandenrThe content in the genus Rothschild is obviously higher than that in the other four genera (Figure 4D).
A total of 15 m/z peaks were matched with compounds in phenylalanine pathway (KEGG ko00360), which may span as many as 19 compounds due to unclear matching degree. The species of Rothschild are significantly related to 11 of the 15 compounds in the pathway, among which MSP 112 acts on 6 compounds (Spearman correlation)P<0.05, figure 4E). Ruminococcus, faecalis, Bifidobacterium and Eubacterium have little correlation (fig. 4E). To sum up,Compared with related species, the influence of Rothschild on phenylalanine metabolic pathway is increased, which is confirmed by increasing the potential of enzyme and the intensity of metabolites..
Finally, we studied 13 kinds of phenylalanine metabolizing enzymes found in the genus Rothschild and their different effects on circulating metabolites. Among them, 5 enzymes (hisC、mhpE、paaH、yhDRandpaaF) is shared by more than half of MSP, while the other 8 species are more species-specific (Figure 5A).aaaT(L- phenylalanine N- acetyltransferase) andpaaIThe co-occurrence of (acyl-CoA thioesterase) is the specificity of immunomodulation MSP 112, becauseThey are not found in any other MSP belonging to the genus Rothschild.. These enzymes play a role in the pathway from phenylalanine to N- acetyl -L- phenylalanine and phenylacetylglutamine, and their plasma concentration is related to the increase in the abundance of the genus Ross (MSP 112) (Figure 4E).
Fig. 5 Comprehensive analysis of the influence of Rostrum on phenylalanine metabolic pathway (KEGG ko00360). A the total abundance of MSP-encoded enzymes in the genus Rothschild, with MSP 112 highlighted in the black box. B observe the enzymes and compounds in phenylalanine pathway. The compounds and enzymes in the blue and orange groups are colored accordingly. There is a significant Spearman correlation between the enzymes encoded by the genus Rothschild and the strength of phenylalanine metabolic complex. Only the top matching compound names are displayed. The complex enzyme groups are represented by orange and blue bars. The black line emphasizes the separation between the two groups. Correlation between the total abundance of C enzyme and the average intensity of compounds grouped by blue and orange bands in B. D The zero distribution of the correlation between the selected compound and the random group of six (blue group) or five (orange group) genes from the genome of Rosinia. Observed Spearman correlations (e.g. C) are marked with black vertical lines, and the zero distribution is shown in gray. Phenylalanine metabolic pathway (KEGG ko00360). The detected compounds and enzymes are blue and orange.
The correlation between the abundance of a single enzyme and the strength of a compound reveals two different pathway patterns. Firstly, L- tyrosine, L- phenylalanine and phenylpyruvic acid (m/z fuzzy matching: 2- hydroxy -3- phenylacrylate, trans -3- hydroxycinnamate) are accompanied by five common enzymes (hisC、mhpE、paaH、yhDRandpaaF) and hipO (grouped by blue bars in fig. 5B). Secondly,HPD、paaK、aaaTandpadEIt is related to the increase of phenylpropionate, hippurate (m/z fuzzy matching: phenylacetylglycine) and phenylacetylglutamine (grouped by orange bars in Figure 5B). The latter compounds also decreased strongly with the existence of five common enzymes, which indicated the antagonistic mechanism between the two groups of compounds and enzymes.
We evaluated the significance of the observed correlation, thus verifying that the difference in the intensity of each group of compounds is indeed more likely to be caused by related enzymes than by other genes. Spearman correlation between enzyme abundance and compound strength shows a stronger effect than any single compound enzyme combination (Figure 5C, upper left and lower right). On the contrary,The correlation between groups showed weak to negative correlation.(fig. 5C, upper right and lower left). Finally, the observed correlation was compared with the zero distribution of 10,000 randomly selected correlations from 6 (blue group) or 4 (orange group) of all 22,832 Rossella genes. . In fact,Compared with the random genome, all four observed effects are extreme (fig. 5D), which confirms that the opposite compound-enzyme association is unlikely to be caused by other genes of the genus Rothschild.. Fig. 5E shows the complete set of detected compounds and enzymes, and shows the group separation. These results reveal two mechanisms, each of which is characterized by a group of mutually exclusive enzymes and compounds, and summarize how the difference of species level of Rosinia affects the circulating phenylalanine metabolites.
A study used the supernatant of a single microbial species growing on a specific culture medium for targeted metabolomics research, and provided the species of Rothschild (including species adjacent to Rothschild MSP 112 at the species level:Roseburia inulinivorans) additional evidence of the ability to change compounds in the phenylalanine metabolic pathway (fig. 4A).Roseburia inulinivoranAt least 2-fold changes of two compounds were observed in S supernatant:yhDRandhisCGlycolic acid encoded downstream increased, while inmphESuccinic acid encoded upstream decreases (e.g. Annex 1: Figure S8, Figure 5D). Although we did not detect these compounds in the plasma metabolomics method, the change direction was consistent with the related genes, and supported the positive role of Rosella in phenylalanine metabolism pathway.
discuss