Divergence between circulating trimethylamine and trimethylamine-N-oxide associated with common flavin-containing monooxygenase 3 variation

Summary

Flavin-containing monooxygenase 3 (FMO3) oxidizes trimethylamine (TMA), a gut-derived metabolite of dietary substrates such as choline and betaine, into trimethylamine-N-oxide (TMAO), which has been linked to cardiometabolic and renal diseases. We examined whether two common FMO3 genetic variants, p.Glu158Lys (E158K, rs2266782) and p.Glu308Gly (E308G, rs2266780), were associated with circulating TMA and TMAO in a sub-study of a randomized clinical trial incorporating dietary substrate loads (136 fasting samples from four visits). Each Gly308 allele was associated with approximately 18% lower plasma TMAO concentrations and accounted for approximately one-third of baseline ln(TMAO) variation, independent of rs2266782 and robust to adjustment for estimated glomerular filtrate rate. In contrast, plasma TMA concentrations did not differ by genotype. No consistent genotype effects were observed for blood pressure, although systolic blood pressure showed a genotype × treatment interaction without main genotype or treatment effects. Given the small sample size and sparse genotype representation, these findings should be interpreted as exploratory and may reflect variation in TMA oxidation efficiency. In conclusion, this initial study suggests that individuals carrying the Gly308 variant exhibit lower circulating TMAO without detectable increases in TMA, consistent with preserved short-term substrate handling under controlled conditions and motivating further investigation of the TMA-TMAO pathway in larger, genotype-balanced cohorts.

Graphical abstract

Highlights

• •FMO3 p.Glu308Gly carriers exhibited lower circulating TMAO concentrations
• •Lower TMAO was not accompanied by detectable TMA accumulation
• •Genotype-associated TMAO differences persisted across dietary substrate loads
• •Systolic blood pressure showed a genotype × treatment interaction

biochemistry; biomolecules; molecular biology

Introduction

Flavin-containing monooxygenase 3 (FMO3) is a major hepatic phase I enzyme that oxidizes a broad range of nucleophilic substrates, including xenobiotics and diet-derived compounds.¹ Among these, trimethylamine (TMA) derived from dietary nutrients such as choline and, to a lesser extent, betaine is converted by FMO3 to trimethylamine-N-oxide (TMAO),² a circulating molecule associated with cardiometabolic and renal diseases.^3,4 Under normal physiological conditions, hepatic FMO3 efficiently clears TMA, predominantly leaving TMAO in circulation to be excreted by the kidney. Consequently, circulating TMA concentrations are typically low and volatile, making the precursor difficult to quantify.⁵

The most extreme disruption of FMO3 function is seen in trimethylaminuria (“fish odor syndrome”), a rare monogenic disorder caused by homozygous or compound heterozygous loss-of-function FMO3 mutations. Affected individuals have severely impaired TMA N-oxidation, very low or absent TMAO, markedly elevated TMA in plasma and body fluids, and a characteristic fishy odor due to TMA excretion in urine, sweat, and breath.^6,7 In contrast to rare null mutations, common FMO3 single nucleotide polymorphisms (SNPs) are frequent in the general population. Coding variants such as p.Glu158Lys (E158K, rs2266782) and p.Glu308Gly (E308G, rs2266780) are present at minor allele frequencies of ∼20–30% in many European ancestry cohorts.^8,9 These two variants are often in linkage disequilibrium (LD), making their independent in vivo effects difficult to separate; for instance, in certain subgroups, LD between rs2266780 and rs2266782 is strong.^10,11 Functionally, p.Glu308Gly is characterized as a reduced-function or hypomorphic allele.^12,13 By contrast, p.Glu158Lys (E158K) shows context-dependent effects. In vitro activity varies by substrate, and in vivo associations may simply reflect its linkage with p.Glu308Gly.^12,14 Thus, while rare near-null FMO3 mutations precipitate a dramatic phenotype with very low TMAO and high TMA (i.e., trimethylaminuria), these common alleles may subtly reshape TMAO biology.

Under simple Michaelis-Menten kinetics, reduced FMO3 activity should lower the product (TMAO) while causing a reciprocal rise in the substrate (TMA). If true, this “hydraulic” relationship implies that pharmacologic inhibition of FMO3 could elevate TMA to a degree that increases the risk of iatrogenic trimethylaminuria. Yet, the field’s predominant emphasis on TMAO has left comparatively less attention on how the host manages TMA flux, limiting our ability to anticipate the full range of metabolic processes associated with pathway inhibition. This gap also extends to genetic epidemiology studies, including Mendelian randomization, which often rely on common variants in metabolic enzymes as proxies for long-term differences in pathway activity.^15,16 Existing studies have generally assumed that FMO3 variants specifically perturb TMAO;^15,17 however, without metabolic phenotyping that includes both TMA and TMAO, this assumption remains uncertain.^18,19 Moreover, because TMAO has been linked to cardiometabolic traits and renal function,¹⁹ evaluating related physiological measures such as blood pressure, pulse, and estimated glomerular filtrate rate (eGFR), may help clarify whether common FMO3 variation is associated with broader clinical responses. Therefore, more comprehensive characterization of how these variants relate to circulating TMA, TMAO, and selected physiological phenotypes across dietary precursors may help refine the interpretation of future genetic studies.^12,19

To address this gap, we leveraged a randomized crossover study incorporating controlled dietary challenges, with daily 4-week supplementation of dietary choline (provided as egg, the predominant dietary source of choline) with betaine as well as betaine alone, which function as metabolic “stress tests” given that genetic effects on metabolic flux are often best revealed when the pathway is perturbed. This design offers a distinct advantage. By modulating the precursor pool, we can provide an exploratory assessment of whether genotype is associated with consistent patterns in TMA and TMAO across varying substrate loads.

We hypothesized that FMO3 genetic variation is a determinant of circulating TMAO and related metabolite levels and functional responses to TMA substrate intake. The objectives of this study were to (i) characterize associations between the common FMO3 p.Glu308Gly (E308G) variant and circulating TMA and TMAO across dietary treatment conditions; (ii) examine whether observed associations occur independently of the p.Glu158Lys (E158K) variant; and (iii) explore whether different genotypes modify treatment-related variation in metabolic and clinical phenotypes. Here, we provide a preliminary characterization of how common FMO3 variation relates to substrate-product patterns and physiological outcomes under standardized conditions, offering biological context that may inform the interpretation of future genetic studies.

Results

Sample characteristics and metabolite distributions

The study comprised 34 individuals yielding 136 fasting plasma samples. This genetic sub-analysis was performed post hoc and the sample size was determined by the parent trial rather than by an independent a priori calculation. For rs2266780, genotype counts were Glu/Glu (AA) = 20 (58.8%), Glu/Gly (AG) = 9 (26.5%), Gly/Gly (GG) = 5 (14.7%), consistent with Hardy-Weinberg equilibrium. Given the repeated-measures crossover design under 4-week dietary treatment regimens, the study achieved improved precision for estimating within-person metabolite patterns, but the analysis remained exploratory because of the limited number of individuals within genotype strata, particularly the Gly/Gly group. Baseline (Visit 1) characteristics are summarized in Table 1. Participants had mean age of 25.2 ± 5.2 years, mean body mass index (BMI) of 27.9 ± 2.1 kg/m², and mean systolic blood pressure of 126.8 ± 12.6 mmHg. While 32% of participants exhibited a systolic blood pressure ≥130 mmHg, these phenotypes reflect a representative “real-world” young adult demographic rather than a clinical disease population.

Table 1.

Baseline characteristics of the study participants, shown overall and by flavin-containing monooxygenase 3 p.Glu308Gly (rs2266780) genotype

Characteristic	Overall (n = 34)	Glu/Glu (n = 20)	Glu/Gly (n = 9)	Gly/Gly (n = 5)
Demographics and vitals
Age (years)	25.2 ± 5.2	25.3 ± 4.9	26.8 ± 6.6	22.0 ± 1.9
Male sex, n (%)	17 (50%)	10 (50%)	5 (56%)	2 (40%)
BMI (kg/m2)	27.9 ± 2.1	28.1 ± 2.4	27.6 ± 2.1	27.6 ± 1.4
SBP (mmHg)	126.8 ± 12.6	130.0 ± 13.6	121.2 ± 10.6	124.1 ± 8.7
DBP (mmHg)	76.2 ± 7.9	76.3 ± 8.0	74.3 ± 7.1	79.4 ± 9.3
Pulse (bpm)	73.4 ± 8.4	72.7 ± 10.2	74.9 ± 4.4	73.9 ± 6.7
Renal function
Creatinine (mg/dL)	0.86 ± 0.12	0.86 ± 0.10	0.82 ± 0.12	0.91 ± 0.18
Cystatin C (mg/L)	0.86 ± 0.06	0.86 ± 0.04	0.85 ± 0.10	0.83 ± 0.06
eGFR (mL/min/1.73 m2)a	108.6 ± 9.5	108.2 ± 8.4	110.2 ± 12.0	107.6 ± 11.0
Plasma metabolites
TMAO (μM)	2.59 [2.37, 2.84]	2.95 [2.74, 3.18]	2.22 [1.80, 2.73]	2.05 [1.65, 2.54]
TMA (nM)	13.4 [12.0, 14.9]	13.8 [12.1, 15.8]	12.6 [10.1, 15.8]	13.2 [9.3, 18.8]

Data include 34 participants at Visit 1 baseline. Continuous variables are presented as mean ± SD unless otherwise indicated. Sex is presented as n (%). Plasma concentrations of TMAO and TMA are presented as geometric mean [95% CI]. Abbreviations: BMI, body mass index; eGFR, estimated glomerular filtration rate; SBP, systolic blood pressure; DBP, diastolic blood pressure; TMAO, trimethylamine-N-oxide; TMA, trimethylamine.

^aCalculated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine-cystatin C equation.

Direct renal markers confirmed preserved kidney function at baseline: plasma creatinine was 0.86 ± 0.12 mg/dL, plasma cystatin C was 0.86 ± 0.06 mg/L, and CKD-EPI creatinine-cystatin C eGFR was 108.6 ± 9.5 mL/min/1.73 m², with a range of 87–127 mL/min/1.73 m². Only one participant had baseline eGFR of 87 mL/min/1.73 m², and none had eGFR <60 mL/min/1.73 m². Across all 136 fasting samples, eGFR remained preserved (109.5 ± 9.8 mL/min/1.73 m²; range 87–129). Plasma TMAO exhibited marked inter-individual variability, with concentrations spanning approximately a 65-fold range (0.15–9.5 μM); eGFR was not correlated with plasma TMAO at baseline (Spearman’s ρ = −0.08, p = 0.65) or across all observations (ρ = 0.003, p = 0.97). Plasma TMA concentrations were orders of magnitude lower (median ∼17.5 nM), with a narrower distribution. There was no strong correlation between plasma TMA and TMAO (Spearman’s ρ ≈ 0.16, p = 0.07), suggesting that steady-state TMAO levels are not simply a linear function of circulating TMA.

The FMO3 p.Glu308Gly (E308G) variant was associated with lower plasma TMAO

The Gly308 allele frequency of rs2266780 was ≈0.28. In the baseline model adjusting for age, sex, and BMI, rs2266780 Gly308 allele count showed an inverse association with ln(TMAO) (Figure 1A; Table 2). Each additional Gly308 allele corresponded to a β coefficient of −0.202 on the log scale, equivalent to ∼18% lower geometric mean TMAO per allele (ratio 0.82; 95% CI 0.73–0.91; p < 0.01). The overall model R² was 0.387, and the partial R² for the genotype term was ≈0.33, with an F-statistic of 14.2. This indicates that rs2266780 may be a first-stage predictor of baseline TMAO even after accounting for age, sex, and BMI. This association was unchanged after additional adjustment for eGFR (β = −0.202; SE = 0.050; ratio 0.82; 95% CI 0.74–0.91; p = 0.0004; partial R² = 0.37; F = 16.1), indicating that the genotype-TMAO association was not explained by differences in renal filtration. The plasma TMAO fraction, defined as TMAO/(TMAO + TMA), was also lower in Gly308 carriers than non-carriers (Figure 2).

Figure 1.

Flavin-containing monooxygenase 3 (FMO3) p.Glu308Gly (rs2266780) genotype-associated variation in plasmatrimethylamine-N-oxide (TMAO) and trimethylamine (TMA) concentrations

Plasma (A) TMAO and (B) TMA concentrations stratified by genotype (Glu/Glu n = 20; Glu/Gly n = 9; Gly/Gly n = 5). The Gly308 allele was associated with a stepwise reduction in circulating TMAO. Despite the significant reduction in product formation shown in (A), the substrate TMA did not show differences. Data shown as boxplots (median and interquartile range) with individual points colored by estimated glomerular filtration rate (eGFR) (mL/min/1.73 m²) calculated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine-cystatin C equation. P values correspond to overall group differences evaluated using Kruskal-Wallis tests; and pairwise differences using separate two-sided Mann-Whitney U tests with Holm adjustment. A total of 34 participants contributed fasting samples across four study visits.

Table 2.

Association of flavin-containing monooxygenase 3 rs2266780 Gly308 allele with plasma TMAO and TMA in baseline and longitudinal models

Outcome/model	β (SE) per allele	Ratio [95% CI]	p value	Partial R2
Plasma TMAO(ln)
Baseline (Visit 1)	−0.202 (0.053)	0.82 [0.74, 0.91]	0.0007	0.33
Baseline + eGFR adjustment	−0.202 (0.050)	0.82 [0.74, 0.91]	0.0004	0.37
Longitudinal	−0.200 (0.068)	0.82 [0.72, 0.94]	0.0034	–
Longitudinal + eGFR adjustment	−0.215 (0.065)	0.81 [0.71, 0.92]	0.0009	–
Plasma TMA(ln)
Baseline (Visit 1)	−0.041 (0.080)	0.96 [0.82, 1.12]	0.61	0.01
Baseline + eGFR adjustment	−0.041 (0.081)	0.96 [0.81, 1.13]	0.61	0.01
Longitudinal	−0.001 (0.045)	1.00 [0.92, 1.09]	0.98	–
Longitudinal + eGFR adjustment	−0.011 (0.048)	0.99 [0.90, 1.09]	0.82	–

Estimates reflect per-allele effects from linear regression models (baseline) and linear mixed-effects models (longitudinal). Baseline models used visit 1 data from 34 participants whereas longitudinal models included 34 participants across four study visits. Ratios and 95% CI indicate exponentiated β estimates and represent multiplicative differences per additional Gly308 allele. P values correspond to tests of the per-allele β estimates in each model. Partial R² is shown for baseline models only. eGFR-adjusted baseline models include estimated glomerular filtration rate (eGFR) calculated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine-cystatin C equation, in addition to age, sex, and body mass index. eGFR-adjusted longitudinal models include eGFR in addition to age, sex, body mass index, treatment, visit, and subject-specific random intercept.

Figure 2.

Assessment of the plasma trimethylamine-N-oxide (TMAO) fraction in relation to flavin-containing monooxygenase 3 (FMO3) p.Glu308Gly (rs2266780) genotype

Plasma TMAO fraction as TMAO/(TMAO + trimethylamine (TMA)), stratified by FMO3 p.Glu308Gly (rs2266780) genotype (Glu/Glu n = 20; Glu/Gly n = 9; Gly/Gly n = 5). The Gly308 allele was associated with a reduction in plasma TMAO fraction relative to Glu/Glu homozygotes. As this fraction was derived solely from plasma concentrations, it does not constitute a direct measure of hepatic FMO3 enzymatic flux. Data are shown as boxplots (median and interquartile range) with individual points colored by estimated glomerular filtration rate (eGFR) (mL/min/1.73 m²) calculated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine-cystatin C equation. P values correspond to pairwise differences evaluated using two-sided Mann-Whitney U tests with Holm adjustment. A total of 34 participants contributed fasting samples across four study visits.

In the longitudinal mixed-effects model including all 136 observations and adjusting for age, sex, BMI, treatment, and visit, the per-allele effect was consistent (β ≈ −0.20; SE ≈ 0.07; p < 0.01), again corresponding to ∼18% reduction in TMAO per Gly308 allele (Table 2). The association remained robust after additional adjustment for eGFR (β = −0.215; SE = 0.065; ratio 0.81; 95% CI 0.71–0.92; p = 0.0009). The dominant model likewise showed that Gly308 carriers had lower TMAO than non-carriers.

Random-intercept models indicated that ln(TMAO) has between-person stability. In a model without genotype, the between-person variance component was 0.060 and the residual within-person variance was 0.204, yielding an intraclass correlation coefficient of ≈0.23. Adding rs2266780 to the model reduced the between-person variance to 0.041 while leaving the residual variance essentially unchanged, indicating that genotype accounted for approximately one-third (∼32%) of the between-individual variation in TMAO across visits. Together, a pattern linking FMO3 p.Glu308Gly (E308G) with lower circulating TMAO in this study represents an initial step toward clarifying genotype-phenotype relationships in future studies.

Genotype-TMAO associations were not accompanied by higher circulating TMA

In this cohort, lower TMAO among Gly308 carriers was not accompanied by higher concentrations of the precursor TMA (Figure 1B; Table 2). In baseline models, the per-allele association with TMA was negligible (β = −0.041; NS), and genotype did not explain the variance (R² ≈ 0.03). Geometric mean TMA concentrations were not different across genotypes with Glu/Glu: 13.8 nM (95% CI 12.0–15.9); Glu/Gly 12.6 nM (9.7–16.4); Gly/Gly 13.2 nM (8.0–21.7). Furthermore, this pattern persisted across all longitudinal treatment phases (ratio 1.00; 95% CI 0.92–1.09). The null TMA association was also unchanged after eGFR adjustment at baseline (ratio 0.96; 95% CI 0.81–1.13; p = 0.61) and across all longitudinal treatment phases using eGFR (ratio 0.99; 95% CI 0.90–1.09; p = 0.82), raising the possibility that TMA handling may have additional buffering capacity beyond the variation captured by FMO3 genotype.

Evaluation of FMO3 variants in relation to TMAO and additional phenotypes

In models including both rs2266780 and rs2266782, only p.Glu308Gly (E308G, rs2266780) remained significantly associated with TMAO (p < 0.001; Table 3), whereas rs2266782 showed no association with either TMAO or TMA. This pattern suggests that the observed genotype-TMAO relationship was attributable to rs2266780, rather than rs2266782.

Table 3.

Association of the flavin-containing monooxygenase 3 rs2266780 Gly308 allele with standardized clinical and metabolic phenotypes

Phenotype	Standardized β (per allele)	95% CI	p value
TMAO	−0.75	−1.15 to −0.36	<0.001
TMA	−0.09	−0.57 to 0.39	0.71
SBP	−0.33	−0.80 to 0.13	0.15
DBP	0.10	−0.38 to 0.58	0.68
BMI	−0.13	−0.61 to 0.35	0.58
Age	−0.18	−0.66 to 0.30	0.44
Creatinine	0.12	−0.36 to 0.60	0.60
Cystatin C	−0.23	−0.70 to 0.25	0.33
eGFRa	0.02	−0.46 to 0.51	0.92

Standardized β coefficients (per Gly308 allele) were estimated using linear regression models. Analyses were performed using visit 1 baseline data from 34 participants. P values correspond to tests of the per-allele β estimates. Abbreviations: TMAO, trimethylamine-N-oxide; TMA, trimethylamine; SBP, systolic blood pressure; DBP, diastolic blood pressure; BMI, body mass index; eGFR, estimated glomerular filtration rate.

^aCalculated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine-cystatin C equation.

We next examined whether rs2266780 genotype was associated with demographic, renal, or cardiometabolic characteristics. At Visit 1 (Table 1), rs2266780 genotype was not associated with age (mean 25.3, 26.8, and 22.0 years for Glu/Glu, Glu/Gly, and Gly/Gly, respectively), BMI (28.1, 27.6, and 27.6 kg/m²), plasma creatinine (0.86, 0.82, and 0.91 mg/dL), plasma cystatin C (0.86, 0.85, and 0.83 mg/L), eGFR (108.2, 110.2, and 107.6 mL/min/1.73 m²), systolic blood pressure (SBP; 130.0, 121.2, and 124.1 mmHg), diastolic blood pressure (DBP; 76.3, 74.3, and 79.4 mmHg), or pulse (72.7, 74.9, and 73.9 bpm). The proportion of male participants was also similar across genotypes (50%, 56%, and 40%). Treatment allocation was independent of genotype by design due to the crossover structure. Across the panel of phenotypes, rs2266780 genotype showed a negative standardized effect on TMAO, whereas estimates for plasma TMA, clinical variables, plasma creatinine, plasma cystatin C, and eGFR spanned the null (Figure 3; Table 3). These results indicate that the genotype-TMAO association was not accompanied by differences in the additional phenotypes assessed in this study.

Figure 3.

Association of flavin-containing monooxygenase 3 rs2266780 Gly308 allele with circulating metabolites and cardiometabolic traits.

Points show standardized β coefficients (per Gly308 allele) for plasma trimethylamine-N-oxide (TMAO), trimethylamine (TMA), systolic blood pressure (SBP), diastolic blood pressure (DBP), body mass index (BMI), age, creatinine, cystatin C, and estimated glomerular filtration rate (eGFR), which was calculated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine-cystatin C equation. Associations were estimated using linear regression models with standardized variables. There was a selective association with TMAO, with no clear relationships with other measured phenotypes. Analyses were performed using visit 1 baseline data from 34 participants.

Mixed-effects analyses of treatment responses and genotype interactions

In the longitudinal mixed-effects models including genotype and visit, the association between rs2266780 and TMAO did not differ across visits: genotype × visit interaction terms were not significant (for Visits 2–4 vs. Visit 1), and including these interactions did not improve model fit. In age- and sex-adjusted models, treatment significantly influenced circulating metabolites (Figure S1). Plasma TMAO concentrations varied by treatment (p < 0.0001), with the highest concentrations observed during betaine + eggs. Plasma TMA concentrations also differed by treatment (p < 0.0001), with higher concentrations during betaine-supplemented conditions compared with baseline/cellulose.

Plasma betaine showed a genotype × treatment interaction (p < 0.05 for rs2266780; p < 0.01 for rs2266782), reflecting genotype-specific variation in betaine response. Consistent with treatment composition, plasma betaine concentrations were elevated during both betaine-supplemented conditions (p < 0.0001), whereas plasma choline was highest during betaine + eggs compared with baseline/cellulose (p < 0.0001) for both rs2266780 and rs2266782, with rs2266782 also showing higher plasma choline than baseline after betaine supplementation alone.

For secondary outcomes, genotype showed no main effects on SBP, DBP, pulse, BMI, plasma creatinine, plasma cystatin C, or eGFR; however, SBP exhibited a genotype × treatment interaction for both rs2266780 and rs2266782 (p < 0.001), while DBP, pulse, BMI, creatinine, cystatin C, and eGFR showed no genotype × treatment interactions (Figures S2 and S3).

Discussion

This sub-study leveraged simultaneous plasma TMA and TMAO measurements within a randomized clinical trial to define their modulation arising from common FMO3 genetic variation. The FMO3 rs2266780 Gly308 allele was associated with lower plasma TMAO across visits, supporting a genotype-linked TMAO set-point. The availability of direct renal markers strengthens this interpretation, wherein baseline creatinine, cystatin C, and eGFR were preserved across genotype groups, and the rs2266780-TMAO association persisted after adjustment for eGFR. Importantly, these exploratory analyses did not identify corresponding differences in plasma TMA, indicating that systemic TMA handling appears to remain largely unchanged within the range of FMO3 genotypes represented in this study.

The pattern observed in Gly308 carriers may be consistent with a functional effect of this FMO3 variant that lowers TMAO. An extreme comparison is monogenic trimethylaminuria, in which a near-complete loss of FMO3 function produces extremely low TMAO, substantial TMA accumulation, and the characteristic fish odor phenotype.^20,21 Under a simple single-enzyme model, reduced FMO3 function would be expected to decrease net TMAO production and increase TMA. Instead, each Gly308 allele was associated with ∼18% lower TMAO, whereas TMA did not exhibit alterations. Thus, our results suggest that partial impairment inferred from common FMO3 variation may yield lower systemic TMAO exposure without provoking substrate accumulation.

These findings are pertinent given emerging lines of evidence that liver-directed antisense therapies against FMO3 in rodents can lower TMAO and improve metabolic parameters and atherosclerosis, in parallel with Fmo3 deletion phenotypes.^22,23,24 More broadly, the handling of TMA and TMAO is clinically relevant given ongoing interest in pathways that may influence cardiometabolic disease risk.¹⁹ In translating such approaches to humans, monitoring TMA levels in early-phase trials will be prudent to ensure that pharmacologic inhibition does not push the system beyond its homeostatic capacity, particularly in individuals with reduced renal function, where compensatory clearance is already constrained.

The magnitude and direction of the rs2266780 association may offer a benchmark for the degree of FMO3 attenuation that lowers TMAO while avoiding the TMA elevation characteristic of severe FMO3 deficiency. Plasma TMAO exhibited substantial inter-individual variability in this study, spanning approximately 65-fold; yet, this dispersion does not appear to be due to renal function: eGFR showed no correlation with TMAO at baseline or across all observations, and adjustment for eGFR did not attenuate the genotype-TMAO association. The persistence of lower TMAO in Gly308 carriers argues against full compensation by alternative oxidative pathways; however, other FMO isoforms and non-FMO oxidative routes could still contribute partially to buffering capacity.²⁵ Although urinary metabolites were not measured, TMA is known to be eliminated through the kidney and potentially via pulmonary exhalation or dermal loss.²⁶ The absence of plasma TMA accumulation despite lower TMAO suggests that these clearance mechanisms, or upstream feedback processes, possess sufficient reserve capacity within the physiological range.²⁷

An additional, non-mutually exclusive mechanism may include feedback to the gut microbiota, although this study did not assess microbial composition and function. Reduced first-pass hepatic conversion could increase transient portal TMA exposure, which may modulate TMA production through effects on TMA-lyase activity or the abundance of TMA-producing taxa,^28,29 which may display complex dynamics.³⁰ Retroconversion of TMAO to TMA³¹ may be another potential contributor, although the dominant flux from gut to liver makes efficient clearance the more likely explanation for the unchanged TMA concentrations. We report effects per Gly308 allele, whereas some cohorts report effects per “minor allele” without functional alignment, and rs2266780 is in LD with rs2266782, which can invert apparent directions across populations.³²

This study was not designed or powered to support causal inference, thus the observed associations provide initial information relevant to future work.^33,34 Within this cohort, rs2266780 showed a dose-dependent relationship with TMAO. At baseline, the variant explained ∼33% of ln(TMAO) variation after adjusting for age, sex, and BMI, with an F-statistic of 14.2, and longitudinally, explained ∼32% of variation, with stable per-allele effects over time. Crucially, geometric mean TMA concentrations did not differ across genotype groups at baseline and across visits, and there was no evidence of genotype-related clustering near any detection limits. Thus, rs2266780 may warrant further evaluation in larger, more genetically diverse cohorts as a potential instrument for probing TMAO biology, while recognizing the importance of reproducibility of the present findings before any such application.

Our findings also contextualize prior human studies. In the Seattle Kidney Study of an ongoing clinic-based cohort of individuals with chronic kidney disease (n = 339), participants carrying the E158K (rs2266782) allele exhibited higher circulating TMAO concentrations, potentially reflecting the interaction between hepatic genetic variation and reduced renal clearance.³⁵ In contrast, other studies report null or weak associations in cohorts with largely preserved kidney function, when measured cross-sectionally or post-prandially, and all have been restricted to TMAO measurements.^36,37,38 Wilson et al. reported no variation in plasma TMAO by E158K or E308G genotype in a mixed clinical cohort (n = 479),³⁶ and James et al. found limited relationships between rs2266780/rs2266782 genotype and post-prandial TMAO area under the curve.³⁷ In line with these, existing studies demonstrate that TMAO concentrations span >2 orders of magnitude, shaped by dietary patterns, kidney function, and gut microbiota composition.^{39,40,41,42,43} Thus, while rs2266780 accounted for approximately one-third of the between-person variation in TMAO set-point in this study, substantial contributions from other factors beyond genotype remain important.

A notable aspect of the cardiometabolic phenotypes examined in this study is the largely null pattern of associations between genotype and clinical traits, with the exception of SBP. No associations were found between rs2266780 and age, sex, BMI, DBP, or pulse. However, SBP showed a significant genotype × treatment interaction (p < 0.001) in the absence of main genotype or treatment effects. Visually, the betaine + eggs period elicited opposite SBP shifts by genotype, consistent with a context-dependent dietary modifier of SBP rather than a baseline hemodynamic effect of FMO3 variation. Because rs2266780 did not show a genotype × treatment interaction for TMAO, the effect on SBP is unlikely to reflect differential TMAO responses but may potentially extend to one-carbon metabolites. These results align with a recent study demonstrating that dietary composition exerts strong and sometimes counterintuitive effects on TMA and TMAO generation,⁴⁴ which emphasizes that diet may modulate TMAO-related physiology independently of genotype. Given that SBP was a secondary outcome and the post hoc nature of our analyses, the interaction findings require cautious interpretation.

The broader significance of our findings is that they help nuance the prevailing narrative that “TMAO is bad.” While many observational and experimental studies implicate TMAO as pro-atherogenic, pro-thrombotic, and pro-fibrotic, there is also evidence that TMA itself can have adverse effects, for example by increasing blood pressure and kidney injury markers in a rat model.⁴⁵ In this light, rapid conversion of TMA to TMAO may protect certain tissues from high TMA exposure.⁷ In addition, TMAO has been proposed to have protein-stabilizing effects under osmotic stress and potentially protective roles in pancreatic β-cell function in some contexts.^46,47 These considerations suggest that risk may depend not only on the absolute TMAO concentration but also on the balance between TMA and TMAO, the underlying renal and cardiometabolic milieu, and the presence of other stressors.⁴⁸

Limitations of the study

Several limitations merit comment. Our focus on select genetic variants as modulators of metabolic homeostasis makes the detection of subtle pleiotropic effects or associations with a broad range of clinical outcomes beyond the scope of this sub-study of the parent trial. Future studies should stratify participants by FMO3 genotype and kidney function to test whether TMA accumulation emerges with impaired renal clearance,^49,50 assess sex/hormonal-state effects given the known sexual dimorphism in FMO3 expression,^38,51,52,53 and interrogate gene-diet and gene-gut microbiota interactions as upstream regulators.^12,54 Moreover, our study design did not employ a completely controlled feeding design or select for participants with specific gut microbiota profiles, thus whether genotype effects vary with intake of choline, carnitine, and fish, and different microbial TMA production capacity^12,37,44 remains unknown. Multi-omics approaches integrating dietary assessment, gut microbiota profiling, plasma and urinary metabolomics, and ideally, breath TMA measurements, could help partition the relative contributions of hepatic oxidation, microbial metabolism and excretory pathways to FMO3 variation. Finally, progress in this area will depend on assembling substantially larger and more diverse cohorts (outside of the predominantly European-descent participants in this study) to clarify the role of p.Glu308Gly (E308G) and related variants in TMAO biology. These large-scale efforts may also guide the clinical development of FMO3-targeted therapies in the context of cardiometabolic disease risk and advance precision medicine approaches that leverage subtle individual genetic variation.

In conclusion, this genetic sub-study identifies an exploratory pattern of lower circulating TMAO linked to common FMO3 variation without detectable accumulation of TMA, suggesting the presence of a homeostatic system that potentially decouples substrate and product. These findings provide a more nuanced understanding of the TMA-TMAO pathway and create a foundation for future work in larger cohorts to elucidate how genetic determinants contribute to cardiometabolic disease risk.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Dr. Clara E. Cho (claracho@uoguelph.ca).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •De-identified human data (clinic, genotype and metabolite measurements) have been deposited at Mendeley and are publicly available as of the date of publication at https://doi.org/10.17632/826mhbnbpx.2.
• •All original code has been deposited at Zenodo and is publicly available at https://doi.org/10.5281/zenodo.20008719 as of the date of publication.
• •Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request.

Acknowledgments

We are grateful to all participants for their time and involvement in the study. We thank Dr. Alison Duncan for her advice on the design and execution of the clinical trial. We would like to thank Jeremy Easlick at the Genomics Facility for his technical assistance with end-point genotyping. We also thank Drs. Dyanne Brewer and Armen Charcoglyan in the Mass Spectrometry Facility at the University of Guelph for their methodological input on the LC-MS/MS protocol. We extend our thanks to James Turgeon, Ian Scagnetti, Julian Bommarito, and Nina Andrejic for serving as clinic phlebotomists. We acknowledge Nisa Butt, Luke Krushelnycky, Mali Al-Issa, Jianzhang Dong, Faiza Noor, Erisa Noori, and all short-term undergraduate student volunteers for their help with the data collection. This work was supported by the Natural Sciences and Engineering Research Council of Canada-Discovery Grant. J.A.Z. was supported by a College of Biological Science Summer Research Assistantship. E.M.P. was supported by a Summer Undergraduate Research Assistantship. G.V.S. was supported by a Canadian Institutes of Health Research (CIHR) Canada Graduate Scholarship-Master. C.E.C. holds a CIHR Canada Research Chair Tier II.

Author contributions

Conceptualization, C.E.C.; methodology, J.T., J.A.Z., and J.D.L.; investigation, J.T., J.A.Z., J.D.L., A.S., H.D., G.V.S., E.M.P., S.E.B., and C.E.C.; formal analysis, J.T., A.S., and H.D.; data curation, G.V.S., E.M.P., and S.E.B.; writing – original draft, J.T.; writing – review and editing, J.A.Z., J.D.L., and C.E.C.; funding acquisition, C.E.C.; supervision, C.E.C.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

No generative AI and AI-assisted technologies were used in the writing process.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Buffy coat samples	University of Guelph Human Nutraceutical Research Unit	N/A
Plasma samples	University of Guelph Human Nutraceutical Research Unit	N/A
Chemicals, peptides, and recombinant proteins
Dimethyl sulfoxide	Thermo Fisher	BP231
Trimethylamine-N-oxide	Sigma-Aldrich	317594
Trimethyl-d9-amine N-oxide	CDN Isotopes	D-6092
Trimethylamine hydrochloride	Sigma-Aldrich	T72761
Trimethyl-d9-amine hydrochloride	Sigma-Aldrich	613843
Choline chloride	Thermo Scientific	AC110290500
d13-choline chloride	CDN Isotopes	D-5185
Betaine anhydrous	Thermo Scientific	AC204241000
N-(carboxymethyl)-N,N,N-trimethyl-d3-ammonium chloride	CDN Isotopes	D-6303
Acetonitrile HPLC	Thermo Scientific	A998
Formic acid Optima LCMS	Thermo Scientific	A11750
Ethyl bromoacetate	Thermo Scientific	AAA1044822
Critical commercial assays
DNeasy Blood and Tissue Kit	Qiagen	69506
rs2266782, C___2461179_30	Thermo Fisher	4362691
rs2266780, C___2220257_30	Thermo Fisher	4362691
Taqman Genotyping Master Mix 2×	Thermo Fisher	4371355
Creatinine Assay Kit	Millipore Sigma	MAK475
Cystatin C Assay Kit	Millipore Sigma	RAB0105
Deposited data
De-identified raw human data	Mendeley	https://doi.org/10.17632/826mhbnbpx.2
Software and algorithms
Design & Analysis 2.7.0	Thermo Fisher	SCR_026962
Compass HyStar 5.1	Bruker	https://www.bruker.com
Gen5 3.12	BioTek	SCR_017317
Python 3.13.5	Python Software Foundation	SCR_008394
SAS 9.4	Statistical Analysis System	SCR_008567
Statistical code	Zenodo	https://doi.org/10.5281/zenodo.20008719
Other
Microcrystalline cellulose supplemental powder	Canadian Med Health Supplies	N/A
Betaine supplemental powder	NOW Foods	2226
Large grade A premium white eggs	NoFrills	N/A

Experimental model and study participant details

The study protocol (ID: 24-03-006) was approved by the University of Guelph Research Ethics Board. Thirty-four adults (17 male, 17 female) of any race and ethnicity with ages ranging from 18 to 70 years and a body mass index (BMI) range of 25–35 kg/m² (representative of the typical North American population) were recruited from Guelph, Canada and surrounding areas from January-June 2025. Participants were excluded if they were <18 years or >70 years, had a BMI <25 kg/m² or >35 kg/m², were pregnant or planning to become pregnant during the course of the study (females), were users of hormone therapy, follow a vegan diet, were smokers or recreational drug users, had chronic illnesses, alcoholism, or other metabolic diseases (including trimethylaminuria), had abnormal blood chemistry or hematology results indicative of organ dysfunction, have used prebiotics, probiotics, dietary fiber supplements in the past 2 months, have taken antibiotics within the past 2 months, or have food allergies. This genetic sub-analysis was performed post hoc, and our sample size calculation for the parent trial was based on our previous studies indicating that n = 34 provides 80% power to detect a 10% difference in plasma choline (dietary precursor of TMA) concentrations, assuming a variance of 1.1 and α = 0.05.⁴¹ Three participants dropped out of the study due to scheduling/travel conflicts thus the sample size was fulfilled by recruiting additional participants. The two FMO3 variants examined rs2266780 (c.923A>G; p.Glu308Gly, E308G) and rs2266782 (c.472G>A; p.Glu158Lys, E158K) are common missense polymorphisms with reported minor allele frequencies of 20–30% in populations.^8,9 Under these frequencies, a cohort of 34 participants is expected to include individuals in all three genotype categories (0, 1 and 2 minor alleles) for each SNP, permitting analysis under an additive genetic model. Given the reduced within-subject variability afforded by the crossover design, the available sample supported detection of moderate genotype-related differences in TMAO and related metabolites, while acknowledging that estimates for the minor allele homozygous group remain less stable. Therefore, this genetic analysis should be interpreted cautiously.

The parent study (Figure S4) included a randomized crossover design where each participant was exposed to three treatment conditions, each lasting four weeks, separated by a one-to two-week washout period: daily intake of 3 g cellulose powder (Canadian Med Health Supplies, Surrey, BC, Canada) with no eggs, daily intake of 3 g betaine powder (NOW Foods) with no eggs, and daily intake of 3 g betaine powder (NOW Foods, Bloomingdale, IL, USA) with 3 whole eggs (NoFrills, Guelph, ON, Canada). All supplemental powders were similar in appearance, provided in a single serving sachet that can be mixed into a beverage or food of choice. The random allocation sequence was generated by the study administrator using a random number generator and was concealed in sequentially numbered, sealed envelopes until interventions were assigned. The sequence order was coded, and participants, sample collectors, and outcome assessors were blinded to the interventions. All participants were instructed to return empty sachets and cartons to the research unit to confirm adherence. All participants provided written informed consent prior to enrolling in this study.

Method details

Study visits

Participants arrived at the Human Nutraceutical Research Unit (University of Guelph) between 0700 and 1000 for each of the four visits, following a 10-h overnight fast. On the preceding day, participants were instructed to avoid grapefruit juice and indole-containing vegetables and fish,^42,55 known to affect FMO3 activity and/or TMAO response. Compliance with these restrictions was verified using a 24-h food diary. Upon arrival, participants rested quietly in a seated position for 15 min, after which systolic blood pressure (SBP), diastolic blood pressure (DBP), and pulse rate were measured twice using a validated automated oscillometric device (Omron Healthcare, Hoffman Estates, IL, USA), with participants seated, the arm supported at heart level, and an appropriately sized cuff applied to the bare upper arm. Subsequently, baseline blood samples were obtained by a trained phlebotomist via standard venipuncture. Whole blood was drawn in EDTA-coated tubes, immediately placed on ice, and centrifuged at 2,000 × g for 10 min at 4°C. Plasma was distributed into coded cryogenic vials and stored at −80°C until metabolite quantification and measurement of creatinine and cystatin C for eGFR calculation. Buffy coat was aliquoted into coded cryogenic vials containing 50 μL DMSO, gently inverted, and stored at −80°C until SNP analyses.

DNA extraction

Genomic DNA was extracted from buffy coat using a QIAGEN DNeasy Blood & Tissue Kit following the manufacturer’s protocol. The quantity and quality of genomic DNA samples were measured using an Epoch Spectrophotometer (Agilent BioTek, Santa Clara, CA, USA).

TaqMan SNP determination

Targeted genotyping for FMO3 rs2266780 and rs2266782 was performed using endpoint allelic discrimination TaqMan assays on a QuantStudio 7 Pro Real-Time PCR System in the Genomics Facility at the Advanced Analysis Center (University of Guelph) as previously described.⁵⁶ PCR reactions were prepared with 20 ng DNA, TaqMan SNP Genotyping Assay Mix (Thermo Fisher Scientific) and TaqMan Genotyping Master Mix (Thermo Fisher Scientific). Thermal cycling conditions were as follows: polymerase activation at 95°C for 10 min; 40 cycles of 95°C for 15 s and 60°C for 1 min, with pre- and post-read steps built in at 60°C for 35 s. Samples were prepared in duplicate with negative water controls and in-run standards. Genotype calls were performed using Design & Analysis Software 2 version 2.7.0 (Thermo Fisher Scientific).

Measurement of plasma TMAO, TMA, choline and betaine concentrations

Targeted metabolite analyses were conducted using previously established methodologies^57,58 to enable simultaneous quantification of plasma TMA, TMAO, choline, and betaine concentrations. The liquid chromatography-tandem mass spectrometry (LC-MS/MS) platform comprised a UHPLC Thermo Ultimate 3000 (Thermo Fisher Scientific, Wilmington, DE, USA) coupled to an EVOQ Qube LC-TQ Triple Quadrupole Mass Spectrometer (Bruker Daltonics, Bremen, Germany) operated with electrospray ionization in positive-ion mode.

Briefly, plasma samples (100 μL) were combined with 200 μL acetonitrile and 10 μL internal standard containing d9-TMA hydrochloride (Sigma-Aldrich, Inc., St. Louis, MO, USA), d9-TMAO (Cambridge Isotope Laboratories, Tewksbury, MA, USA), d13-choline chloride (CDN Isotopes, Pointe-Claire, QC, Canada), and d3-betaine ammonium chloride (CDN Isotopes, Pointe-Claire, QC, Canada). The mixture was vortexed and centrifuged at maximum speed for 5 min, and supernatants were transferred to vials with glass inserts. To each, 2 μL concentrated ammonium hydroxide and 30 μL of 20 mg/mL ethyl bromoacetate in acetonitrile (EBA) were added to derivatize volatile TMA as previously described⁵⁸ to the more stable TMA-EBA adduct, followed by incubation at room temperature for 45 min. Samples were diluted 1:1 with water:acetonitrile:formic acid (1:1:0.0005). Injections of 10 μL samples were made onto a Prevail Silica analytic column (150 × 2.1 mm, 5 μm; Grace, Columbia, MD, USA) fitted with a matching guard column. The autosampler and column were kept at 5°C and 25°C, respectively.

Detection was performed in multiple-reaction monitoring mode, with transitions of m/z 146.1 → 118.1 for TMA-EBA, m/z 155.1 → 127.1 for d9-TMA-EBA, m/z 76.3 → 58.4 for TMAO, m/z 85.3 → 66.4 for d9-TMAO, m/z 104.2 → 60.2 for choline, m/z 117.2 → 69.2 for d13-choline, m/z 118.2 → 58.4 for betaine, and m/z 121.2 → 61.4 for d3-betaine. Intra-assay coefficient of variation (CV) values ranged from 2 to 14%, with the larger CV for TMA reflecting the inherent volatility of this compound. Data acquisition was carried out using Compass HyStar (Bruker Daltonics).

Measurement of plasma creatinine and cystatin C concentrations

Plasma creatinine and cystatin C concentrations were measured by enzyme-linked immunosorbent assay (ELISA) using the Creatinine Assay Kit (MAK475, Millipore Sigma, Burlington, MA, USA) and the Cystatin C Assay Kit (RAB0105, Millipore Sigma), respectively, in accordance with the manufacturers’ instructions. All samples were assayed in duplicate, and concentrations were interpolated from standard curves generated on each plate.

Estimation of glomerular filtration rate

Estimated glomerular filtration rate (eGFR) was calculated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine-cystatin C equation,⁵⁹ which removes race coefficients and incorporates age and sex as required. Plasma creatinine and cystatin C concentrations obtained from ELISA were entered into the equation, and eGFR values were expressed in mL/min/1.73 m².

Quantification and statistical analysis

Hardy-Weinberg equilibrium was assessed using an exact test. Alleles were aligned to dbSNP/ClinVar on the + strand; the G allele encodes Gly308 and the A allele encodes Glu308, and Gly308 count was modeled additively as the effect allele.

Plasma TMAO and TMA concentrations (primary outcomes) were natural log-transformed. For baseline cross-sectional analyses (Visit 1), linear regression models were fitted with ln(TMAO) or ln(TMA) as outcomes and rs2266780 Gly308 allele count as the primary predictor, adjusted for age, sex, and BMI.

Linear mixed-effects models for ln(TMAO) and ln(TMA) were fitted with a subject-specific random intercept. The main predictor was rs2266780 Gly308 allele count (0, 1, or 2). Models included age, sex, BMI, treatment (baseline, cellulose, betaine, and betaine+eggs), and visit (1–4) as fixed effects. The genotype coefficient thus represents the average per-allele difference in TMAO (or TMA) across all visits, adjusting for covariates and within-person correlation. To evaluate potential confounding by renal function, primary baseline and longitudinal genotype-metabolite models were repeated with CKD-EPI creatinine-cystatin C eGFR included as an additional covariate.

Baseline demographic or clinical variables were compared across genotype groups using one-way analysis of variance or χ² tests. Primary genotype-metabolite association models were conducted using Python 3.13.5. Overall effects were assessed using Kruskal-Wallis tests; pairwise comparisons were evaluated using two-sided Mann-Whitney U tests with Holm adjustment. Spearman’s correlation was used to assess associations among metabolites and between metabolites and clinical variables. Genotype, treatment, and genotype × treatment effects for repeated-measures plasma metabolites (TMA, TMAO, choline, and betaine), renal markers (creatinine, cystatin C, and eGFR), and clinical variables (SBP, DBP, pulse, and BMI) were tested using repeated measures mixed-effects models in SAS 9.4, with age and sex as covariates. To simplify groups, individuals with no variant allele were compared with individuals carrying either one or two copies of the variant allele within each SNP. When overall effects were significant, least-squares means were compared using Tukey’s post hoc tests. P-values <0.05 were considered statistically significant; NS denotes not significant. Continuous variables are presented as mean ± SD. Plasma TMAO and TMA concentrations are presented as geometric mean [95% CI] based on log-transformed values, unless otherwise stated.

The Gly/Gly group was small (n = 5) and the Glu/Gly group included 9 participants, aligned with the expected prevalence of genotype distributions.⁶⁰ Given the sparse genotype representation, this genetic sub-study should be considered exploratory. We estimated the proportion of variance in log-transformed plasma TMAO associated with the genotype (partial R²) and reported the corresponding F-statistic from a linear regression model. These metrics were used only to summarize the strength of the observed genotype-phenotype association within this cohort. To contextualize other sources of influences, we examined associations between rs2266780 and participant characteristics (age, sex, and BMI) as well as study design factors (treatment and visit). We also assessed whether rs2266780 showed detectable associations with plasma TMA, creatinine, cystatin C, eGFR, SBP, DBP, and pulse.

Additional resources

This trial was registered at clinicaltrials.gov as NCT06758856.