Oral carnitine challenge test
The OCCT was performed through oral administration of 1500 mg of carnitine tablets (GNCTM) to the test participants. Blood and urine samples were collected at baseline and at 24 and 48 h after the challenge. Urine was collected within 2 h of blood sampling. An individual’s TMAO productivity was determined by calculating the AUC of the OCCT curve (OCCT TMAOAUC) or by using the OCCT TMAOMAX level. We defined plasma OCCT TMAOMAX > 10 μM (OCCT pTMAOMAX > 10 μM) as the high-TMAO producer phenotype and vice versa according to a review of the literature5,6,8-10. All participants fasted overnight before the OCCT and were requested to avoid red meat, seafood, or any medication during the test. The carnitine tablets used in this this study contained no animal products according to the manufacturer. More details on our creation of the OCCT can be found in a previous publication11.
Carnitine supplement intervention and sample collection in the healthy study cohort
A total of 56 healthy volunteers, comprising 33 omnivores and 23 vegetarians, were recruited to receive 500 mg of oral supplementation of carnitine fumarate (GNCTM) daily (equals to 2-3 folds of carnitine obtained from regular diet) for 1 month. Eligibility criteria were age of 20–65 years, generally health with no chronic disease or recent illnesses, no apparent bacterial or viral infection, and no obvious gastrointestinal disorder. The participants had no exposure to antibiotics, probiotics, food supplements, or any other medication at least 1 month before the study. The OCCT was performed before and after the carnitine supplement intervention. Plasma and urine samples were collected for biochemical analysis and TMAO metabolites measurement. Each participant collected the fecal samples at home by using a validated stool collector before and after the intervention period29. All study protocols and informed consents of human participants were approved by the Institutional Review Board of National Taiwan University Hospital (201507055MINC), and all participants had signed a waiver of informed consent. The study for healthy subject cohort had been registered in ClinicalTrials.gov as NCT02838732. While a preliminary cross-sectional analysis of baseline data in the trial has been published, this paper presents a much more comprehensive and in-depth analysis for this interventional study.
Measurement of Carnitine, TMAO, g-butyrobetaine, d9-Carnitine, d9-TMAO and d9-g-butyrobetaine
Sample preparation and quantification for carnitine and TMAO in healthy volunteers
50 μL plasma and 50 μL urine samples were extracted with 450 μL and 950 μL methanol containing 200 ng/mL isotopically labeled internal standards (d3-carnitine and d9-TMAO) respectively, and the extraction was performed using the Geno/Grinder 2010 (SPEX SamplePrep., Metuchen, NJ, USA) at 1000 rpm for 3 minutes. The extracts were then centrifuged by using the Eppendorf Centrifuge 5810R at 12000 g for 5 minutes at 4 °C. The supernatants were subjected to liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis.
Target metabolites and their corresponding internal standards were analyzed by using Agilent 1290 UHPLC coupled with an Agilent 6460 triple quadrupole mass spectrometer (Agilent Technologies, Santa Clara, CA, USA). Separation was performed using a MicroSolv Cogent Diamond Hydride column (150 mm x 2.1 mm, 4.2μm, MicroSolv, Eatontown, NJ, USA), and the column was thermostated at 40 °C during analysis. The mobile phase was composed of solvent A (10 mM ammonium acetate and 0.2% formic acid in water) and solvent B (10 mM ammonium acetate and 0.2% formic acid in 90% ACN). A 0.4 mL/min linear gradient elution was used: 0-1 min, 90-75% solvent B, 1-2 min, 75-65% solvent B, 2-4 min, 65-55% solvent B, 4-5 min, 55-40% solvent B; and column re-equilibration with 90% solvent B for 1 min. The injection volume was 5 μL. The positive electrospray ionization mode was utilized with the following parameters: 325 °C for drying gas temperature, 7 L/min for drying gas flow, 45 psi for nebulizer pressure, 325 °C for sheath gas temperature, 11 L/min for sheath gas flow rate, and 3500 V for capillary voltage. Nozzle voltage was set at 500 V. The mass spectrometer was configured in multiple reaction monitoring mode, and the monitored transitions for carnitine were m/z 162.1→43.2 and 162.1→60.2; d3-carnitine were m/z 165.1→43.1 and 165.1→61.2; TMAO were m/z 76.1→58.1 and 76.1→59.1; d9-TMAO were m/z 85.1→66.3 and 85.1→68.3. The concentration of each analyte in the samples were determined from calibration curves by using the peak area ratio of the analyte to its corresponding isotope internal standard.
Sample Preparation and Quantification for Carnitine, TMAO and g-butyrobetaine of Cardiovascular Disease Patient
For optimal extraction, 180 μL or 190 μL of methanol spiked with 200 ng/mL of isotopically-labeled internal standards (d3-carnitine and d9-TMAO) was added to 20 μL of plasma or 10 μL of urine samples, respectively. The samples were vortexed for 3 minutes and then centrifuged in Hermle Centrifuge Z216MK at 15000 rpm for 5 minutes at 0 °C. The supernatants were kept for subsequent LC-MS/MS analysis.
For LC-MS/MS analysis, 20 μL of each sample was injected into Sciex Exion LC AC system coupled with SCIEX Triple TOF 5600 mass spectrometer (AB SCIEX, Canada). Separation was achieved with HILIC column (250 x 4.0 mm, 5μm, Fortis, UK) maintained at 40 °C. The mobile phase A was 0.1% formic acid in water and mobile phase B was 0.1% formic acid in acetonitrile. The total running time was 12 minutes: 0-1 min, 50% solvent B, 1-9 min, 50-40% solvent B, 9-10 min, 40% solvent B, 10-10.1 min, 40-50% solvent B, followed by column re-equilibration with 50% solvent B for 1.9 min. The flow rate was 0.5 mL/min. The electrospray was set in positive ionization mode with the following parameters: 30 psi for curtain gas supply, 500 °C for the capillary temperature, 5500 V for the spray voltage floating, and 80 V for the declustering potential. The concentrations of each analyte in samples were determined from calibration curves using peak area ratio of the analyte to its corresponding isotope internal standard.
Sample Preparation and Quantification for d9-Carnitine, d9-TMAO and d9-g-butyrobetaine of Humanized Gnotobiotic Mice Model
Twenty microliters of plasma and 10 μL of urine samples were extracted with 180 μL and 190 μL methanol containing 50 ng/mL isotopically labeled internal standards (d3-carnitine and 13C3-TMAO) respectively, and then vortex for 3 minutes. The extract was then centrifuged by using Hermle Centrifuge Z216MK at 15000 rpm for 5 minutes at 0 °C. The supernatant was subjected to Sciex Exion LC AC system coupled with SCIEX Triple TOF 5600 mass spectrometer (AB SCIEX, Canada).
Target metabolites and their corresponding internal standards were analyzed using the same method as for the samples of cardiovascular disease patients. The concentrations of d9-carnitine and d9-TMAO in the samples were determined from the calibration curves using the peak area ratio of the analyte to its corresponding isotope internal standard. The concentration of d9-g-butyrobetaine was calculated from the calibration curve constructed by plotting the peak area versus the concentration of d9-g-butyrobetaine.
16S rRNA microbiome sequencing
Fecal samples were transported to our lab in cold storage (4-7 °C) within 24 hours of collection. The feces were then aliquoted and stored at -80 °C for microbiome analysis and other experiments. Fecal genomic DNA was extracted by using the Mobio PowerFecal DNA Isolation Kit according to the manufacturer’s instructions and quantified using the NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific). A two-step polymerase chain reaction (PCR) workflow was conducted for library preparation in accordance with procedures described in the Illumina 16S sample preparation guide. The 16S rRNA gene V3-V4 region was amplified using a primer overhanging adapter (forward = 5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGC CTACGGGNGGCWGCAG-3′ and reverse = 5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAG ACAGGACTACHVGGGTATCTAATCC-3′). Dual indices and Illumina sequencing adapters were attached through PCR by using a Nextera XT Index Kit according to the manufacturer’s instructions. After each PCR process, PCR cleanup was performed using AMPure XP beads to purify V3-V4 amplicon from the free primer and primer dimer. The sizes of PCR products were verified using the Bioanalyzer DNA 1000 chip. Library quantification was performed for quality control before sequencing by using the Agilent Technologies 2100 Bioanalyzer. The pooled libraries were then sequenced on the Illumina MiSeq platform with v3 reagents for paired-end sequencing (2 × 300 bps).
Bioinformatics analysis
16S-amplicon processing pipeline
The 16S-amplicon processing pipeline was modified from 16S Bacteria/Archaea SOP v1 of Microbiome Helper workflows31. Paired-end reads were assembled by using PEAR v0.9.832. Assembled sequences were quality-filtered by thresholds of sequence length ≥400 bp and quality score of 90% bases of reads ≥20. All quality-filtered reads were analyzed with the pipeline of QIIME (v1.9.1) pipeline33,34. OTUs were assigned using a closed-reference OTU picking approach, which referenced picks against SILVA (NR132) database35,36 by using UCLUST algorithm37 with 97% of sequence identity. The generated OTU table was filtered by removing singletons as well as low-confidence OTUs which, which underwent bleed-through removal between MiSeq runs. The final OTU table was rarefied into 61,600 reads/sample.
Mining low-abundance sequences by using the BLAST algorithm
I. massiliensis failed to pass the confidence-filtering of the OTU table because of its low read abundance in the sequence library. We retrieved the 16S rRNA gene sequence of I. massiliensis strain Marseille-P2843 from NCBI GenBank database (accession number NR_144749.1)23. The absolute read counts of I. massiliensis were then extracted from quality-filtered reads by using blastn algorithm38 (e-value < 1e-5, identity ≥ 97%) by searching against the 16S rRNA gene sequence of I. massiliensis.
Biodiversity and statistical analyses
Gut microbial community analyses were conducted with the R package vegan39. A Mann-Whitney U test in R software40, with α = 0.05, was used for statistical analyses. Multiple-testing P values were adjusted with FDR by using “p.adjust” function in R software. Alpha diversity indices and Shannon index were calculated by “diversity” function; observed OTU was counted by “specnumber” function; and the Chao1 index was calculated by “chao1” function of fossil package41. For beta diversity, dissimilarities among microbial communities were measured by Bray-Curtis distance and principal coordinates analysis; ADONIS (permutational multivariate analysis of variance using distance matrices) was used to test the heterogeneity of gut microbial composition among sample groups42. The associations between TMAO-producing phenotypes (Log10 urine TMAOMAX) and gut microbial abundance were measured using Pearson’s correlation coefficients by using log10-transformed OTU abundances. The top 2.5% highly correlated OTUs (39 of the 1637 OTUs) were selected for subsequent phylogenetic analysis and random forest modelling. To profile the OTU abundance alterations among samples/groups or across TMAO concentrations, a heatmap was generated by using pheatmap R package43. The statistical analysis for comparing serum biochemical values and TMAO levels were evaluated by Student’s t test, Wilcoxon rank-sum test and Mann-Whitney U test accordingly and appropriately using Prism 8.0 software based on data distribution and characteristics.
Phylogenetic analysis
Phylogenetic analysis of associated TMAO-producing OTUs was conducted using was conducted using an online phylogeny analysis pipeline (Phylogeny.fr)44,45. The "One Click" mode was selected for the following analyses: the candidate 16S rRNA gene sequences were first submitted in the fasta format; sequences were aligned with MUSCLE46, and aligned sequences were curated with Gblocks47; the phylogenetic tree was constructed with PhyML by using the maximum likelihood method48; a constructed phylogenetic tree was visualized and rendered by FigTree v1.4.349.
Cardiovascular disease patient validation cohort
A total of 50 patients with catheterization-proven CVD (>50% stenosis for at least one coronary vessel) were recruited as an independent cohort for validating the microbiome-based TMAO producer prediction model (The baseline characteristics of CVD validation cohort were described in Table S2). The patients with CVD received a simplified OCCT by urine sampling and were categorized according to urine OCCT TMAOMAX level (uOCCT TMAOMAX). Because the urine and plasma TMAO levels were highly correlated, a cutoff value of OCCT uTMAOMAX ≥ 162.79 mmol/mol creatinine, corresponding to OCCT pTMAOMAX > 10 μM was defined for high-TMAO producers on the basis of the equation of the linear regression plot between urine and plasma TMAO levels (Supplementary Figure 6A). Stool samples were collected using the same process as that described for microbiome analysis in the healthy study cohort. The study protocol and informed consent obtained from patients with CVD were approved by the Institutional Review Board of National Taiwan University Hospital (201712030RIND) and all participants in this study had signed a waiver of informed consent. The study for CVD patient cohort had been registered in ClinicalTrials.gov as NCT03781011.
Random forest classification modelling for TMAO producer phenotypes prediction
Random forest classification models were built for predicting high- or low-TMAO producing phenotypes. A healthy cohort (112 samples from 56 subjects receiving OCCT for two times) was used for building predication models with randomForest R package50. An external CVD cohort with 50 participants served as a validation cohort for the prediction model. Two prediction models were built, one with 39 high-TMAO-correlated OTUs and the other with the whole OTU dataset (1637 OTUs). Leave-one-out cross-validation was performed for estimating model accuracy by using caret R package51. The importance of features was then ranked using the values of “mean decrease accuracy.” Finally, the model accuracy was measured according to the AUROC by using the ROCR package in R software52.
Humanized gnotobiotic mouse experiments
C57BL/6JNarl male germ-free mice (8-12 weeks of age) were used to establish four study groups of hGM model through fecal microbiota transplantation (FMT) from donors of two OCCT-defined high-TMAO producers (O15 and O32) and two OCCT-defined low-TMAO producers (V12 and V25). Each hGM group was established by gavaging germ-free mice with 200µL of human fecal suspension, which was prepared by using a 1g aliquot of feces diluted in 10 mL of reduced phosphate-buffered saline (supplemented with 0.05% L-cysteine) in an anaerobic chamber. After the colonization had stabilized (at least 2 weeks after FMT), the hGM received oral carnitine challenge test by gavage administration with d9-carnitine (150 µL from a 150 mM stock) dissolved in water. Plasma was collected from the submandibular vein at baseline and at indicated times.
The hGM experiments were performed using the facilities of the National Laboratory Animal Center, National Applied Research Laboratories, Taiwan, and were approved by the Institutional Animal Care and Use Committee (IACUC2017O04). Their germ-free mice maintenance protocol is as follows: mice are maintained in a vinyl isolator (CBC, Madison, WI) at the room the temperature of 22 ± 1 °C and 55–65 % relative humidity with a 12-h light/12-h dark cycle. Mice are fed a commercial diet (5010 LabDiet, Purina Mills, St. Louis, MO) and sterile water ad libitum. To confirm GF status, microbiological assays were performed on a monthly basis by culturing feces, bedding, and drinking water in thioglycollate medium (DIFCO, Camarillo, CA).