Uncovering the trimethylamine-producing bacteria of the human gut microbiota

Abstract

Background: Trimethylamine (TMA), produced by the gut microbiota from dietary quaternary amines (mainly choline and carnitine), is associated with atherosclerosis and severe cardiovascular disease. Currently, little information on the composition of TMA producers in the gut is available due to their low abundance and the requirement of specific functional-based detection methods as many taxa show disparate abilities to produce that compound.

Results: In order to examine the TMA-forming potential of microbial communities, we established databases for the key genes of the main TMA-synthesis pathways, encoding choline TMA-lyase (cutC) and carnitine oxygenase (cntA), using a multi-level screening approach on 67,134 genomes revealing 1107 and 6738 candidates to exhibit cutC and cntA, respectively. Gene-targeted assays enumerating the TMA-producing community by quantitative PCR and characterizing its composition via Illumina sequencing were developed and applied on human fecal samples (n = 50) where all samples contained potential TMA producers (cutC was detected in all individuals, whereas only 26% harbored cntA) constituting, however, only a minor part of the total community (below 1% in most samples). Obtained cutC amplicons were associated with various taxa, in particular with Clostridium XIVa strains and Eubacterium sp. strain AB3007, though a bulk of sequences displayed low nucleotide identities to references (average 86% ± 7%) indicating that key human TMA producers are yet to be isolated. Co-occurrence analysis revealed specific groups governing the community structure of cutC-exhibiting taxa across samples. CntA amplicons displayed high identities (~99%) to Gammaproteobacteria-derived references, primarily from Escherichia coli. Metagenomic analysis of samples provided by the Human Microbiome Project (n = 154) confirmed the abundance patterns as well as overall taxonomic compositions obtained with our assays, though at much lower resolution, whereas 16S ribosomal RNA gene sequence analysis could not adequately uncover the TMA-producing potential.

Conclusions: In this study, we developed a diagnostic framework that enabled the quantification and comprehensive characterization of the TMA-producing potential in human fecal samples. The key players were identified, and together with predictions on their environmental niches using functional genomics on most closely related reference strains, we provide crucial information for the development of specific treatment strategies to restrain TMA producers and limit their proliferation.

Keywords: Atherosclerosis; Cardiovascular disease; Functional diagnostics; Gut microbiota; Microbiome; Trimethylamine.

Fig. 1

Results of the multiparametric genome-screening approach for the genes cutC (a) and cntA (b). Obtained unique proteins are depicted along the horizontal axis where they are sorted according to their similarity to the constructed hidden Markov chain models (HMM) represented by the primary vertical axis. The secondary vertical axis depicts both the phylogenetic distance to the top-scoring sequences (triangles) and synteny (diamonds) to the corresponding partner genes cutD and cntB. Taxonomic affiliations of sequences are indicated by the color code. Asterisks indicate sequences lacking conserved amino acid residues. Cutoffs set to discriminate true cutC (HMM score of 906.4) and cntA (HMM score of 440.6) genes are illustrated in the respective plots. For cntA, phylogeny was applied as an additional filtering step (see text). Final databases for cutC and cntA contained 1114 (454 unique proteins) and 6772 (491 unique proteins) candidate genes, respectively. CntA reference sequences derived from Acinetobacter baumannii (Ab) and Escherichia coli (Ec; previously termed yeaW) are highlighted by arrows

Fig. 2

TMA-producing communities of fecal samples revealed by gene-targeting assays. Panel a depicts gene abundances of cutC (red) and cntA (blue) relative to the total amount of 16S rRNA gene copies of a sample. Volunteers are sorted by descending quantity of cutC, and error bars represent standard deviation on triplicate measurements. The composition of cutC (b) and cntA (c) genes is presented as neighbor-joining trees comprising representative sequences derived from complete-linkage clustering on the nucleotide level using a 90% identity cutoff. Heatmaps are aligned with bars from panel a to illustrate the composition of each gene for all samples (relative abundances of clusters are shown). Clusters that contained reference sequences are indicated by name (Clostridium sp. AT5 (1720194.3.peg.1841); C. hathewayi (566550.8.peg.2440); Clostridium citroniae WAL19142 (742734.4.peg.3562)/Clostridium sp. FS41; Clostridium asparagiforme DSM 15981 (518636.5.peg.3238)/Clostridiales bacterium VE202-15 for cutC and Escherichia/Shigella (1169329.3.peg.2590); Citrobacter (742730.3.peg.4068); Klebsiella (1308539.3.peg.2257) for cntA; specific sequence IDs that were used to build the trees are given in brackets), whereas letters represent clusters that did not include any reference. CutC reference sequences that are closely related to representative sequences of clusters are displayed in the tree as well (Eubacterium sp. AB3007 (1392487.3.peg.725); Dorea sp. 52 (1235798.3.peg.5630); Clostridium clostridioforme AGR2157 (1280695.3.peg.868); Collinsella sp. MS5 (1499681.3.peg.1502); Desulfovibrio desulfuricans subsp. desulfuricans DSM 642 (1121445.4.peg.1354)). Bootstrap values ≥50 and ≥90% are indicated by circles. Only major clusters of highest abundances are shown representing 80 and 99% of all cutC and cntA sequences, respectively. Asterisks highlight the reference sequences belonging to Clostridium XIVa strains

Fig. 3

Co-occurrence network and ordination analysis of the cutC gene community. Panel a displays the co-occurrence network based on the clustering result presented in Fig. 2b. Connected nodes (clusters) are co-occurring (p < 0.01, Spearman’s ρ ≥ 0.5, false-discovery correction q-value <0.01) where only clusters that are detected in at least 50% of samples were considered for analysis. Nodes lacking a letter fulfill criteria for co-occurrence analysis, but are not depicted in Fig. 2 due to their low abundance. Node size illustrates relative abundance of the respective cluster. Color codes represent specific clades (clusters a and b from clade I are illustrated as clade Ia and clade Ib, respectively; clade II: cluster g; clade III: cluster j; XIVa: clusters closely related to Clostridium XIVa strains; others: clusters related to other taxa) presented in Fig. 2b. Additional clusters that are not shown in Fig. 2 are color-coded according to their respective clades. Panel b illustrates non-metric multidimensional scaling analysis (Bray-Curtis dissimilarity) of entire cutC gene communities with abundances of clusters depicted in panel a highlighted for each sample. Numbers refer to corresponding samples shown in Fig. 2

Fig. 4

Screening for TMA-producing genes in metagenomes of fecal samples (n = 154) from the Human Microbiome Project. On top the relative abundance of bacteria containing cutC/D (red) and cntA/B (blue) genes calculated as percentage of the total community indicated by the amount of reads matching the housekeeping gene rplB are shown. Samples are sorted along the horizontal axis according to their cutC/D abundances. Results of the gene-targeted metagenomic assembly method as a fraction of the BLAST (blastn) result are displayed as dark red (n = 6) and light blue (n = 3) for cutC/D and cntA/B, respectively (results for cutC/D exceeding those of blastn is shown in orange). Samples are sorted by descending quantity of cutC/D. Specific numeric labels give values exceeding the maxima of axis. Below associated taxonomic affiliations of respective gene communities are displayed. For CutC, reference sequences were subjected to complete-linkage clustering on the nucleotide level based on 90% identity, whereas cntA reference sequences were binned on the genus level. The EKRS cluster involves sequences from Enterobacter, Klebsiella, Raoultella, and Staphylococcus. Clostridium XIVa I, II, and III refer to distinct clusters composed of Clostridium XIVa strains. Asterisks highlight sequences derived from Clostridium XIVa strains. Samples where no genes encoding TMA production were detected (n = 38) are not shown