One Biomarker, Three Different Claims

Circulating trimethylamine N-oxide, or TMAO, has become an attractive candidate in the search for links between the gut microbiome and atrial fibrillation. The attraction is easy to understand: TMAO is produced through a pathway that begins with nutrients handled by gut microbes, and studies have associated higher circulating levels with cardiovascular disease. A recent systematic review found a positive association between higher TMAO and atrial fibrillation (AF), while experimental work in mice suggests that changing the pathway can change susceptibility to arrhythmia. This series examines what those findings mean—and where they stop. The practical question is not simply whether TMAO is “good” or “bad,” but whether it is an association, a mechanism, or a treatment target.

That distinction matters because AF is not a minor laboratory endpoint. It is the most common sustained cardiac arrhythmia and is linked to outcomes such as stroke, heart failure, and death. A marker that could identify people at higher risk would be clinically useful. A molecule that contributes to the development or persistence of AF could be a mechanistic target. A therapy that safely changes that molecule and improves outcomes would be something stronger still. Moving from the first claim to the third requires evidence that cannot be supplied by one promising measurement or one successful animal experiment.

Why the TMAO–AF question matters

AF develops through interacting influences rather than a single trigger. The atrial tissue may be structurally vulnerable; electrical signaling may become unstable; inflammation, autonomic activity, metabolic state, and kidney function may alter the surrounding conditions. TMAO is interesting precisely because it sits at the intersection of several of these influences. It is a circulating metabolite connected to diet, microbial activity, host metabolism, and clearance by the kidney. That makes it biologically plausible as part of a gut–heart pathway, but it also means that its level may reflect several processes at once.

The clinical evidence already shows why a simple interpretation is inadequate. The 2026 systematic review and meta-analysis included 15 studies in its qualitative synthesis and eight prospective estimates in its primary odds-ratio analysis. Higher versus lower circulating TMAO was associated with AF with a pooled odds ratio of 1.35, with a 95% confidence interval from 1.15 to 1.59. That is evidence of a relationship across the included studies. It is not a guarantee that an individual with a given TMAO level will develop AF, and it does not say that lowering the level would produce the opposite result.

The variation around the pooled estimate is as important as the estimate itself. Between-study heterogeneity was substantial: an I² of 68.3% indicates that the observed results differed more than would be expected from sampling variation alone under a simple common-effect model. The strongest association appeared in postoperative AF after cardiac surgery, whereas the pooled estimate for spontaneous incident AF was smaller. In some high-risk groups, the estimate was null. These differences may reflect the atrial substrate, the acute stress of surgery, the definition of the endpoint, or other differences among the populations. They also mean that “the TMAO risk” is not one fixed quantity waiting to be read from a universal chart.

Kidney function adds another reason for caution. TMAO is cleared by the kidney, so a higher blood level can result from reduced elimination as well as from greater production. A prospective study of 5,469 participants found that TMAO was associated with mortality, especially among people with reduced estimated glomerular filtration rate. The association weakened and lost statistical significance after fuller adjustment for kidney function and urinary albumin excretion. That study was about mortality, not AF, so it does not directly estimate the TMAO–AF relationship. It does show why renal function belongs in the interpretation of a TMAO measurement rather than being treated as a minor footnote.

Association, mechanism, and treatment are not synonyms

The first claim is associational: people with higher TMAO may have more AF, or may have a higher probability of an AF-related outcome, than people with lower levels. An association can be useful for prognosis even when the measured molecule is not the main cause. It can also arise because the molecule shares causes with the outcome. Diet, cardiovascular disease, kidney dysfunction, medication, age, or changes in the microbiome could influence both TMAO and AF. Statistical adjustment can address measured differences, but it cannot automatically settle every causal pathway.

Even the phrase “higher TMAO” contains choices that matter. The meta-analysis compares exposure categories across studies rather than defining one universal concentration at which AF begins. The pooled odds ratio summarizes relative differences between those categories; it does not provide an individual’s absolute probability, because that also depends on baseline risk and the population being studied. Nor does it tell us that a proportional reduction in TMAO would create a proportional reduction in AF. A marker may improve risk classification because it tracks a vulnerable state, while an intervention aimed at the marker may fail if the underlying state remains unchanged. This is why the direction of an association is informative, but its clinical use depends on the comparator, outcome, and context.

The second claim is mechanistic: TMAO itself participates in the biological process that makes AF more likely. This requires more than observing a correlation. One wants a plausible pathway, controlled perturbation, and a change in the relevant phenotype when the pathway is altered. The JCI study provides that kind of preclinical evidence. In human subjects, plasma TMAO was independently associated with prevalent AF after adjustment for risk factors, including kidney function. In mice, dietary TMAO or choline accelerated AF onset or increased AF inducibility. Choline supplementation also produced atrial enlargement and electrical changes associated with greater susceptibility.

The same study tested a pathway intervention. Iodomethylcholine, or IMC, inhibits microbial conversion of choline to TMA. In choline-fed mice, it reduced TMA and TMAO and delayed or suppressed the AF phenotype. The experiments further implicated inhibition of M2 muscarinic receptor signaling, with autonomic and electrophysiological changes that included higher heart rate, shorter action-potential duration, and shorter wavelength. Taken together, these results make a causal role biologically plausible in the models studied. They are much stronger than a purely observational association.

But stronger than observational evidence does not mean equivalent to a human therapy. The mouse work did not establish a safe human dose, long-term benefit, or a reduction in clinical AF events. It did not show that a person who lowers TMAO will avoid AF, experience fewer recurrences, or have fewer strokes. Some short-term direct-exposure experiments did not reproduce all of the electrical changes, and the progressive mouse model did not test whether AF recurrence would remain lower after IMC was stopped. Those are not reasons to discard the findings. They are reasons to describe them accurately: the study identifies a promising intervention node and a mechanism worth testing.

The third claim—therapeutic target—is therefore a claim about patient benefit, not just molecular importance. It requires an intervention that changes the pathway in people, a credible comparator, safety information, and an improvement in outcomes that matter to patients. A lower TMAO value by itself would be a laboratory result. It would become clinically persuasive only if the change predicted or accompanied better AF outcomes without unacceptable harms, and if the result held in the population and clinical setting where the treatment would be used.

There are at least two ways a treatment story could stop short. An intervention might lower circulating TMAO without changing the atrial process that matters, especially if the measured level was partly a consequence of reduced kidney clearance rather than the dominant driver of AF. Or it might change a real pathway in a mouse but have a different balance of benefits and harms in people. The available evidence does not tell us which of these possibilities will prevail. It tells us what to test: whether the intervention changes the intended pathway, whether that change persists, whether it is safe, and whether AF incidence or recurrence improves beyond the laboratory measurement. None of the three sources reports that human outcome trial. The gap is therefore specific, not vague: the field has moved beyond a purely correlational idea, but it has not yet crossed the human benefit threshold.

flowchart LR A["Circulating TMAO measurement"] --> B["Human AF association"] B --> C["Controlled mechanistic test in mice"] C --> D["Pathway-lowering test in mice"] D --> E["Human intervention"] E --> F["Patient-important AF benefit"] C -.-> G["Evidence boundary: mouse phenotype is not a human treatment effect"] D -.-> G

Context is part of the claim

A biomarker does not carry a single meaning independently of the setting in which it is measured. A preoperative TMAO value may be useful for studying postoperative AF, where surgery introduces a distinct acute physiological stressor. The same value in a community cohort is answering a different question about incident AF over time. A value measured before catheter ablation may relate to recurrence among people who already have an arrhythmic substrate. These outcomes are connected, but they are not interchangeable.

The meta-analysis makes this visible. The postoperative estimate was an odds ratio of 2.88, based on one study, while seven studies of spontaneous incident AF produced a pooled odds ratio of 1.31. The difference is informative, but one postoperative study cannot establish a perioperative treatment rule. Similarly, the supplementary hazard-ratio analysis combined only two studies with extreme heterogeneity; its random-effects confidence interval included the null. The lesson is not that TMAO is irrelevant. It is that the strength and meaning of the association depend on the question being asked.

This is also why the phrase “TMAO causes AF” is too compressed for the evidence currently available. The animal experiments support a causal pathway in particular models. The human synthesis supports an association whose size varies by context. The renal study shows that the measured concentration can be shaped by clearance and by disease processes that also influence outcomes. A careful conclusion must hold all three statements at once.

The central judgment of this series follows from that tension. TMAO is more than an unexplained correlation: it has a coherent biochemical pathway, experimental evidence of pathway perturbation, and clinically relevant associations. Yet it is not automatically a therapeutic target. To decide whether it can become one, we first need to understand what the number represents—how dietary substrates become TMA, how the liver converts TMA to TMAO, and how the kidney removes it. Only then can the later clinical and experimental results be interpreted without mistaking a composite blood signal for a single cause.

References