Urinary TMAO Levels Are Associated with the Taxonomic Composition of the Gut Microbiota and with the Choline TMA-Lyase Gene (cutC) Harbored by Enterobacteriaceae

The primers used in polymerase chain reaction (PCR) for the amplification of the cutC gene were designed as follows. The GenBank database and Conserved Domain Database (CDD) at the National Center for Biotechnology Information (NCBI) were queried to select 52 nonredundant representative bacterial proteins of the choline trimethylamine-lyase protein family TIGR04394 (choline_CutC; EC Number 4.3.99.4), including the CutC enzymes of Desulfovibrio desulfuricans [4], and Klebsiella pneumoniae [22]. Then, the corresponding CDS nucleotide sequences of selected proteins were used to build a UPMGA tree upon ClustalW multiple alignments. According to the obtained dendrogram, sequences were clustered in two groups: One including the cutC sequence of K. pneumoniae, named cut-Kp, and one including the cutC sequence of D. desulfuricans, named cut-Dd (Supplementary Figure S1). Finally, a pair of primers was designed in the most conserved regions of each group of sequences: cut-Dd-F, 5′-CGTGTTGACCAGTACATGTA-3′ and cut-Dd-R 5′-GCTGGTAACCTGCGAAGAA-3′ (expected amplicon of 185 bp); cut-Kp-F, 5′-GATCTGACCTATCTGATTATGG-3′, and cut-Kp-R, 5′-TTGTGGAGCATCATCTTGAT-3′ (expected amplicon of 190 bp).

The two primer pairs designed as described above were used in endpoint PCR with the genomic DNA extracted from 64 bacterial strains (). Reaction mix was prepared in 25 μL, including 0.5 units of DreamTaq Polymerase (ThermoFisher, Fermentas, Waltham, MA, USA), 1× concentration of DreamTaq Polymerase Buffer (ThermoFisher, Fermentas,), 0.25 μM of each primer, 200 μM of deoxyribonucleotide triphosphate (dNTPs), and 0.5 mM of MgCl2. The PCR cycle program used was the following: Initial denaturation at 95 °C for 2 min, followed by 35 cycles of denaturation at 94 °C for 45 s, annealing at 58 °C for 45 s for the cut-Dd couple and 56 °C for 45 s for the cut-Kp couple, and extension at 72 °C for 20 s. A final extension of 7 min at 72 °C was then applied. Table S1

Bacterial strains were grown in the respective culture medium (reported in Table S1) for 48 h. Afterward, the biomasses were collected by centrifugation at 9500 g for 10 min. The cell pellets were then washed with sterile PBS and resuspended in fresh medium with the addition of 0.2% filter-sterilized choline. Bacteria were incubated at 37 °C for 48 h in glass tubes with screw cap. Afterward, supernatants were collected and used for mass spectrometry (MS) and nuclear magnetic resonance (NMR) analyses. The MS analyzes were performed by directly injecting 5 μL of diluted broth cultures after the removal of the bacterial cells by centrifugation and subsequent filtration with a 0.45-μm syringe filter. In detail, the broth cultures were analyzed in full scan in the range from 50 u to 400 u on an HR-MS Orbitrap model Exactive with a HESI-II probe for electrospray ionization (Thermo Scientific, San Jose, CA, USA). The resolution, gain control, mass tolerance, and maximum ion injection time was set to 50 K, 1E6, 2 ppm, and 100 ms, respectively. The MS data were processed using Xcalibur software (Thermo Scientific). Choline and TMA were used as reference standard. Choline and TMA were also directly detected in broth cultures by ¹H-NMR with a 60 MHz benchtop NMR spectrometer Spinsolve 60 Carbon Ultra, Magritek GmbH (Aachen, Germany).

Study participants were recruited within the University campus. In total, four females and 12 males aged 21–45 (mean: 29.8 years) were enrolled (). The inclusion criteria were as follows: Healthy adult volunteers of both sexes who provided signed informed consent of their participation in the study. The exclusion criteria were as follows: Antibiotic consumption in the month preceding the start of the study, consumption of antacids or prokinetic gastrointestinal drugs, episodes of viral or bacterial enteritis in the two months prior to the study, episodes of gastric or duodenal ulcers in the previous five years, pregnancy or breastfeeding, recent history of alcohol abuse or suspected drug use, and any severe disease that may interfere with treatment. Ethical permission was granted by the University of Milan Ethics Committee (ref: opinion no. 37/16, 15 December 2016). Table S2

Three fecal sample were collected weekly over the course of three consecutive weeks from each volunteer. All the participants were asked to follow their regular diet during the three weeks. Concomitantly to the fecal sample, the volunteers provided 24-h urine collection.

Urine samples were collected over 24 h in sterile tanks and on the same days that fecal samples were been collected. The volume of collected urine was recorded in order to calculate the daily excretion of trimethylamine oxide (TMAO). Immediately after delivery, part of the urine samples was transferred in 10-mL sterile tubes and stored at −80 °C until analysis.

The cutC gene was quantified in fecal DNA with quantitative real-time PCR (qPCR) with both primer pairs, cut-Dd and cut-Kp. To this aim, DNA was extracted from feces using the kit PowerLyzer^® PowerFecal^® DNA Isolation Kit (MO BIO Laboratories, Inc.), starting from 0.25 ± 0.02 mg of sample according to the manufacturer’s instructions. Primer pairs were tested with a gradient qPCR in a range of eight temperatures in order to find the most efficient annealing temperature using DNA of Streptococcus dysgalactiae 485 and Klebsiella sp. A1.2 as reference DNA. In addition, the amplification efficiency of the two pairs of primers was tested in qPCR experiments with six serial 1:3 dilutions of genomic DNA isolated from Streptococcus dysgalactiae 485, Klebsiella sp. A1.2, and human fecal metagenomic DNA. All DNA (bacterial and metagenomic) serial dilutions were tested with primer concentrations of 0.5 µM, 0.4 µM, and 0.3 µM. Efficiency curves were obtained with Bio-Rad software by setting samples as “standard” and obtaining a curve with efficiency (E) parameter and R2 value. Based on the results of these setup experiments, primers were then used at a final concentration of 0.5 µM, as with this concentration, we obtained an R2 value of 0.98. In addition, two randomly selected fecal DNA samples were tested at the different concentration by adding 70 ng, 50 ng, 25 ng, and 10 ng in qPCR reactions. Based on Ct value comparison between the different DNA concentrations, the cutC gene quantification was subsequently performed using 50 ng of total DNA. The reaction mix contained the SsoFast TM Eva-SuperGreen Supermix 2× (Bio-Rad Laboratories), deionized Milli-Q water (Millipore), and primers. All DNA samples (5 µL in each well) were tested in technical duplicate. The qPCR cycles employed were the following: Initial denaturation at 95 °C for 3 min, followed by 44 cycles of denaturation at 95 °C for 30 s, annealing at 58 °C (for cut-Dd primers) or 58.5 °C (for cut-Kp primers) for 30 s, and elongation at 72 °C for 5 s. A final denaturation ramp between 65 °C and 95 °C for 5 s was performed for the melting curve analysis. Moreover, specificity of qPCR reaction was confirmed by checking the presence of only one amplification and of the expected size in electrophoresis on a 2% agarose gel. A total of 48 fecal samples were analyzed. Each sample was analyzed with each primer set in duplicate. The 2^−ΔΔCt method was used for the relative quantification of cutC gene, using the EUB panbacterial primers [Muyzer] targeting the 16S rRNA gene as reference. Data were reported as relative increase of cutC copy number compared to the level of the sample that showed the highest significant Ct in qPCR set as 1.

The bacterial community structure of the fecal microbiota was analyzed as described elsewhere [23,24], with DNA extracted from feces as described in Section 2.2. In brief, extracted DNA was analyzed through 16S rRNA gene profiling. Sequencing reads were generated at the Institute for Genome Sciences (University of Maryland, School of Medicine, Baltimore, MD, USA) with Illumina HiSeq 2500 rapid run sequencing of the V3–V4 variable region. Sequencing reads were equally distributed among the samples. Sequences were filtered and trimmed based on their quality. We obtained a sequence length of 301 bp for both R1 and R2 sequences with an average quality score (Phred score) higher than 35. Sequencing reads were rarefied at 5000 per sample. Subsequently, sequence reads were analyzed through the bioinformatic pipeline Quantitative Insights into Microbial Ecology (QIIME) version 1.9.1 [25] with the GreenGenes database updated to version 13.5. The relative abundance of bacteria in each fecal sample was reported at the taxonomic levels of phylum, class, order, family, genus, and operational taxonomic units (OTUs). Sequence were deposited in the European Nucleotide Archive (ENA) of the European Bioinformatics Institute under accession code PRJEB34169.

TMAO levels in urine samples were determined by ultra-performance liquid chromatography coupled to tandem mass spectrometry (UPLC-MS/MS) (Waters Acquity UPLC system). The analysis method involved the use of a totally porous column with stationary C8 stable bond (Agilent Poroshell C8-SB) and a mobile phase consisting of a gradient acetonitrile and formate buffer (3 mM of ammonium formate and 0.1% formic acid). The UPLC system was equipped with a triple quadrupole detector, which allowed the development of a “multiple reaction monitoring” (MRM) method for the analysis of TMAO. In detail, once thawed at room temperature and after centrifugation at 6000 rpm for 5 min, 25 µL of urine sample were diluted in 950 µL of UPLC mobile phase (1/1 (v/v) acetonitrile/ultra-pure sterile water + 0.025% of formic acid), and 25 µL of deuterated internal standard solution (1 ppm, TMAO-d9, Spectra 2000) were used for the normalization of results [26]. The UPLC samples were prepared mixing 950 µL of mobile phase [1/1 (v/v) acetonitrile/ultra-pure sterile water + 0.025% formic acid), 25 µL of urine sample, and 25 µL of deuterium-labeled methyl d9-TMAO solution (1 ppm; Spectra 2000 S.r.l., Roma, Italy). Mobile phase: 1/1 (v/v) acetonitrile/ultra-pure sterile water + 0.025% of formic acid. The run time per sample was 8 min. Sample freezing and thawing or their prolonged storage at room temperature did not have an impact on the TMAO quantification. A triple set of working standards of TMAO (trimethylamine N-oxide dihydrate, Fluka) at concentrations of 5 ppm, 50 ppm, 100 ppm was prepared according to the method described above, replacing the 25 µL of urine sample with 25 µL of standard solution. The average response factor was used for calculation.

Statistical analyses of data were carried out using R statistic software (version 3.4.2). Concerning cutC gene and TMAO data, intrasubject variability was defined “high” when variance among the three replicates results were higher than twice the median of all variances. Correlation analyses were performed using the Kendall and Spearman formula with the items specified in the text as predictors and dependent variables. Significance was set at p ≤ 0.05, and mean differences in the range 0.05 < p < 0.10 were accepted as trends. To find associations among TMAO levels, bacterial taxa relative abundance, and cutC gene abundance, the machine learning supervised linear mixed model (LMM) algorithm was used. In brief, the LMM was performed using “lmer” function in the “lme4” library [27]. All samples were used in the LMM analysis (n = 48), considering that three measurements were available for each subject. The Akaike’s Information Criterion (AIC) was used to test the goodness of fit of the LMM. The AIC index/value depends on the ANOVA test results between two models: The model that considered the effect of the predictors and the null model.

According to the literature, the ability of intestinal bacteria to convert the choline moiety to TMA is primarily associated with a recently discovered choline utilization (cut) genetic region harboring the cutC gene, which encodes a glycyl radical enzyme catalyzing C–N bond cleavage [15,18]. For this reason, we designed primers specifically targeting the cutC gene. These primers were intended for quantitative PCR (qPCR) experiments, and we avoided the use of degenerations in their sequence. In contrast, to target all putative cutC sequences identified in GenBank, we clustered the putative cutC genes into two groups (named Dd and Kp) according to sequence similarity (Supplementary Figure S1) and designed a pair of primers for each group in the most conserved sequence regions. Group Dd included putative cutC genes from Firmicutes (Anaerococcus, Clostridium, Enterococcus, Streptococcus), Proteobacteria (Desulfotalea, Desulfovibrio, Enterobacter), and Actinobacteria (Olsenella). Group Kp comprised putative cutC gene sequences from Proteobacteria (Aeromonas, Enterobacter, Erwinia, Escherichia, Klebsiella, Pectobacterium, Pelobacter, Proteus, Providencia, Raoultella, Serratia) and Firmicutes (Desulfosporosinus, Enterococcus).

Subsequently, the two primer sets were used in endpoint PCR reactions to test the presence of putative cutC genes within the genomic DNA isolated from the pure cultures of 64 bacterial strains. We obtained an amplicon of the expected size from seven strains. Specifically, strains Streptococcus dysgalactiae 485, S. dysgalactiae 486, and S. dysgalactiae A1.3 gave a band of the expected size with primers cut-Dd. In addition, strains Enterococcus gilvus MD179, Enterococcus hirae MD160, Klebsiella oxytoca MIMgr, and Klebsiella sp. MIMgr were positive with primers cut-Kp (Figure 1A,B). MS and NMR analyses revealed the ability to metabolize choline and produce TMA only for the same seven strains that resulted in positive PCR experiments (Figure 1C and Supplementary Figure S2).

The metagenomic DNA isolated from the feces collected at three time points from 16 healthy adults (n = 48) was used in 16S rRNA gene profiling experiments. A total of 12,588,795 filtered high-quality sequence reads were generated with an average of 13,340 ± 8677 (mean ± standard deviation; max-min 11,594–4570) per sample.

We failed to stratify samples according to the 16S rRNA gene profiling data, indicating that fecal bacterial community structure was homogeneous among samples and among subjects (Supplementary Figure S3). In addition, we also observed that the overall composition of the fecal microbiota in each subject remained mostly stable over the three collection time points (Supplementary Figure S3). Globally, 182 bacterial genera were estimated, with a minimum of 36 and a maximum of 98 genera per fecal sample. Bacteroides was the most prevalent genus, followed by four genera of the order Clostridiales (undefined Ruminococcaceae, undefined Lachnospiraceae, Ruminococcus, and Faecalibacterium) (Supplementary Figure S4A). At the family level, most of the reads were ascribed to only three families, i.e., Ruminococcaceae, Bacteroidaceae, and Lachnospiraceae (Supplementary Figure S4B).

In order to investigate the presence of cutC genes in the human gut microbiome, the cut-Dd and cut-Kp primer sets were used in qPCR experiments using the same fecal metagenomic DNA as a template from healthy adults used for microbiota profiling. All analyzed fecal samples gave a positive signal in qPCR with both primer pairs (Figure 2). In general, cut-Kp was detected at a higher relative concentration than cut-Dd (median ΔΔCt of 5.33 and 0.85 for cut-Kp and cut-Dd, respectively) (Figure 2A,B). In addition, with both cut-Kp and cut-Dd, six volunteers out of 16 showed a variance among the three replicates that was higher than twice the median of all variances, indicating a higher intrasubject variability (Figure 2A,B).

Subsequently, we performed correlation analyses between the cutC abundances determined with qPCR and the 16S rRNA gene profiling data to find potential relationships between the choline TMA-lyase genes and specific bacterial taxa of the fecal microbiota. To this end, we used the median relative abundance of bacterial taxa in fecal samples as predictors, whereas the dependent variables considered were the median abundances of cut-Dd and cut-Kp determined by qPCR per subject. We found that cut-Dd was positively correlated with taxa belonging to the phylum Firmicutes, including an undefined Mogibacteriaceae genus, Oscillospira, and the family Christensenellaceae. On the contrary, cut-Dd was negatively correlated with the Firmicutes order Bacillales, the Firmicutes genus Streptococcus, and the Proteobacteria genus Haemophilus (Supplementary Figure S5). Conversely, cut-Kp was positively associated with Proteobacteria. In particular, inside this phylum, a significant correlation was found with the family Enterobacteriaceae (Supplementary Figure S5).

Subjects were asked to collect 24-h urine specimens the same days when the fecal samples were taken. Then, the levels of TMAO were quantified by UPLC-MS in all urine samples, revealing wide variability among the investigated healthy adults, with levels of urinary TMAO excretion ranging from less than 1 mg to more than 175 mg per day (Figure 2). We also observed an evident intrasubject variability in five volunteers whose TMAO excretion showed a variance among the three replicates that was higher than twice the median of all variances (Figure 2C). In particular, four out of the five volunteers with wide intrasubject variability (i.e., S07, S11, S19, and S22) were found to possess high intrasubject variability for cutC gene levels determined in qPCR experiments (Figure 2).

A linear mixed model was used to infer potential significant relationships among the datasets collected from volunteers at the three time points considered (Figure 3). TMAO was significantly associated with the cut-Kp/cut-Dd synergy (p < 0.001). Furthermore, studying the association of the single cut gene types, we observed that the relationship with TMAO was mainly determined by cut-Kp (Figure 3). In addition, we found a significant association between TMAO and 23 operational taxonomic units (OTUs). Conversely, cut-Kp and cut-Dd were significantly associated with 18 and eight OTUs, respectively. Notably, most of the OTUs that were significantly associated with cut-Kp (i.e., 15 out of 18) were also associated with TMAO, confirming the relationship between these two variables. Nine of the identified OTUs belonged to the phylum Bacteroidetes, while the remaining 21 were ascribed to Firmicutes. In addition, 80% of the OTUs (n = 24) belonged to only three families: Bacteroidaceae, Lachnospiraceae, and Ruminococcaceae. In particular, the most significant association (i.e., p < 0.001) referred to Bacteroides caccae, an undefined Lachnospiraceae genus, and several undefined Ruminoccaceae species (for TMAO and cut-Kp), Bacteroides fragilis, and an undefined Clostridiales species (for cut-Kp only) and an Oscillospira species (for cut-Dd) (Figure 3).