Every other article about GLP-1 receptor agonists and oral medication ends up saying some version of the same thing: the stomach slows, the peak arrives later and lower, the total amount absorbed is broadly unchanged, and it does not matter. Tacrolimus is where that sentence should finally break. It is absorbed badly and inconsistently, its absorption is dominated by food, and the penalty for getting it wrong is a rejected organ. It is also, unusually for this subject, a place where the question has been asked of tens of thousands of people.
Why tacrolimus is the worst case on paper
The prescribing information is blunt about how little of a tacrolimus capsule gets in. Absorption from the gastrointestinal tract “is incomplete and variable,” with absolute bioavailability of 17 ± 10% in adult kidney transplant patients, 22 ± 6% in liver transplant patients and 23 ± 9% in heart transplant patients.[1] A standard deviation ten points wide on a mean of seventeen is a drug whose dose has to be found empirically in each person.
What moves it most is food. A high-fat meal decreased mean area under the curve by 37% and peak concentration by 77%, and lengthened time to peak fivefold. A high-carbohydrate meal decreased them by 28% and 65%. Timing alone mattered nearly as much: given immediately after a meal, peak concentration fell 71% and exposure 39%; given 1.5 hours after, peak fell 63% and exposure fell the same 39%. In 11 liver transplant patients dosed 15 minutes after a high-fat breakfast, exposure fell 27 ± 18% and peak 50 ± 19% against the fasted state. The label therefore instructs that the capsule be taken consistently the same way every day, and that therapeutic drug monitoring be performed for all patients.[1]
A drug that loses more than a third of its exposure to a meal is, on its face, the drug most exposed to something that changes when meals leave the stomach — which is the property the whole class is built on, set out in the oral medications article. Every ingredient of a serious interaction is present: poor bioavailability, a large food effect, a narrow window, and a hard endpoint.
The instruction that would cover it names a different drug
Neither weight-management label mentions tacrolimus, cyclosporine, immunosuppression or transplantation anywhere. What they carry is a category. The tirzepatide label instructs prescribers to “monitor patients on oral medications dependent on threshold concentrations for efficacy and those with a narrow therapeutic index (e.g., warfarin)” when they are co-administered.[2] The semaglutide label asks for “increased clinical or laboratory monitoring for medications that have a narrow therapeutic index or that require clinical monitoring.”[3]
Tacrolimus is the textbook member of that category — dosed to a whole-blood trough range, titrated by a laboratory result, with toxicity and efficacy failure both tied to concentration.[1] The instruction plainly covers it. The example given is a drug whose own interaction literature turned out to be almost empty, which is the subject of the warfarin article.
What the trough data actually show
Unlike most pairings in this class, this one has been looked at repeatedly, because transplant centers were already drawing the blood. A 2026 systematic review and meta-analysis pooled 17 studies covering 54,680 kidney transplant recipients. Tacrolimus levels were unaffected at 6 months and modestly decreased at 1 year without compromising graft function. The same analysis found all-cause mortality lower on a GLP-1 (HR 0.53, 95% CI 0.36 to 0.79, k = 4), major adverse cardiovascular events lower (OR 0.55, 0.47 to 0.66, k = 3), eGFR improved at 6 months by 1.99 mL/min/1.73 m² (0.52 to 3.47) and at 12 months by 2.24 (0.02 to 4.46), HbA1c down 0.54 percentage points (−0.89 to −0.19, k = 13), and adverse events consisting mainly of mild gastrointestinal intolerance in 10% to 20%.[4]
A retrospective chart review of 73 solid-organ transplant recipients treated with semaglutide (n = 39) or tirzepatide (n = 34) reported no significant changes in tacrolimus levels, graft function or mortality in either group.[5] A narrative review of the tirzepatide experience identified 10 eligible reports and restricted quantitative synthesis to the six with disaggregated data, 182 recipients, finding trough fluctuation of −0.2 to +0.2 ng/mL alongside HbA1c reductions of 0.6 to 1.4 percentage points and weight loss of 5.5 to 6.9 kg.[6] A four-center liver transplant cohort of 104 users found immunosuppression trough levels stable, weight down 3.8 kg (−3.9%, P < 0.001), and in the matched analysis of 80 users against 116 non-users a 3.4 kg greater weight loss (P = 0.009) with no increase in T-cell-mediated rejection or major adverse cardiovascular events.[7]
A stable trough is not a measured exposure
Four datasets agreeing is worth something, and it is worth less than it looks, because they all used the same instrument and that instrument is the one least able to detect what the mechanism predicts. Tacrolimus is monitored by a trough drawn 10 to 12 hours after a dose, and the label records that trough correlating with total exposure at 0.93 in kidney recipients, 0.94 in liver and 0.89 in heart.[1] Those correlations were established in people with ordinary stomachs. A delayed gastric emptying effect changes the rate of absorption — the height and the timing of the peak — far more than the amount, and by twelve hours a shifted peak has largely washed out of a trough.
So the honest reading is not that there is no pharmacokinetic interaction. It is that no published study has drawn a full concentration profile in a transplant recipient on a GLP-1, that the parameter everyone measured is the one a rate effect moves least, and that a stable trough is compatible with a peak that has moved substantially. What the data do establish is the thing that matters clinically: the number the dose is titrated against did not drift enough to require a dose change, and graft function did not suffer.
The single case report proposing the mechanism withdraws it
One published case raises the absorption hypothesis directly. A 59-year-old liver re-transplant recipient developed acute disorientation and vomiting four days after starting dulaglutide. A low tacrolimus trough of 2.1 µg/L was recorded on admission, but the authors note it reflected four days off the drug before sampling, and that the concentration during the symptomatic episode was never measured. The patient also had hypomagnesemia at 0.68 mmol/L, a glomerular filtration rate of 49 mL/min and extensive portal-systemic shunting, all risk factors for calcineurin inhibitor neurotoxicity. Their own conclusion is that the case “does not prove this mechanism due to the absence of contemporary tacrolimus measurement and the presence of multiple alternative explanations.”[8] A literature of one, which disclaims itself, is not evidence of an interaction.
The outcome finding reverses under a subgroup
The largest question is not whether a trough drifts but whether these drugs help or harm transplant recipients, and the answer changes depending on which recipients. A retrospective cohort linked a national registry to Medicare claims and identified 18,016 first-time kidney transplant recipients who had type 2 diabetes at transplantation; 1,969 of them (10.9%) filled at least one GLP-1 prescription afterwards. Against non-users, GLP-1 use was associated with 49% lower death-censored graft loss (adjusted subhazard ratio 0.51, 95% CI 0.36 to 0.71; P = 0.0001) and 31% lower mortality (adjusted hazard ratio 0.69, 0.55 to 0.86; P = 0.001). Five-year cumulative incidence of graft loss was 6.0% against 10.7%, and mortality 17.0% against 25.8%, in the survival-time-matched cohort.[9]
An overlapping group then ran the same analysis on the adjacent population — recipients whose diabetes began after the transplant. Among 7,681 first-time adult recipients with post-transplant diabetes mellitus, 430 were prescribed a GLP-1, most commonly dulaglutide (46.1%), a median of 1.7 years after diagnosis. GLP-1 use was not associated with any difference in mortality or graft failure.[10]
Same disease, same organ, same claims data, opposite conclusion. The second study is much smaller, its exposure is rarer, and its follow-up is shorter, so the difference may be power rather than biology. But a page quoting the 0.51 without the null result would be describing a benefit that has been demonstrated in one transplant population and looked for, and not found, in the next one along.
The signal both studies found is not the one anyone was watching
Safety endpoints in the larger cohort were rare and unassociated with GLP-1 use, with one exception: diabetic retinopathy, adjusted hazard ratio 1.49 (1.11 to 2.00; P = 0.008).[9] The post-transplant diabetes cohort found the same thing at 1.80 (1.11 to 2.91), with no increased risk of pancreatitis, biliary complications or medullary thyroid cancer, and advised that care be taken when initiating.[10] The meta-analysis found no increase in hypoglycemia, pancreatitis or infections.[4]
Two independent analyses landing on retinopathy, in a population under constant surveillance for graft rejection, is a more interesting result than another null trough. It is also consistent with what the semaglutide label already records about rapid glycemic improvement and retinopathy progression, covered in the eye health article. Nothing in the immunosuppression data pointed anywhere; the one place something pointed was the retina.
What this changes at a cash intake
For a transplant recipient the practical hazard is simpler than pharmacokinetics and larger. Both GLP-1 labels record severe gastrointestinal reactions and acute kidney injury following dehydration from vomiting or diarrhea. In a person taking tacrolimus twice a day, a vomited dose is a missed immunosuppressant dose, and a stretch of reduced oral intake is a stretch of unpredictable absorption of a drug the graft depends on. Gastrointestinal intolerance in 10% to 20% of recipients[4] is mild as an adverse event rate and is not mild as a threat to a dosing schedule — which is why the sick-day framing in the illness article applies here with more force than anywhere else on this site.
Every dataset above came out of a transplant center, where the trough was already being drawn, the graft function was already being tracked, and somebody knew both prescriptions existed. None of that is a property of the drug; it is a property of the setting. A prescription obtained through a checkout reproduces the molecule and none of the monitoring, and the labeled instruction to increase laboratory monitoring for a narrow-therapeutic-index drug is addressed to a prescriber who, in that transaction, does not exist. The wider pattern is described in the telehealth article, and who should be excluded at all in the contraindications article.
Finally, the product distinction. Compounded semaglutide and tirzepatide are not FDA-approved, and the FDA does not review them for safety, efficacy or quality before a pharmacy dispenses them. Every study described above used a branded product inside a transplant program, and the reassurance those studies provide is a property of that combination rather than of the molecule name printed on a vial. How claims here are established is set out in the methodology.