Most of the argument about these compounds frames them as metabolic tools. The trial data describes something narrower: drugs that mainly change how much people eat. That distinction drives almost everything downstream, including what the dose ladder buys, what the plateau means, and what the withdrawal data shows.
The evidence base here is unusually strong at one end and absent at the other. Three of these compounds now have phase 3 human trials with thousands of participants. Two have never been tested in a human being. Sorting that gap out is most of the work.
What the appetite drugs did in the trials
STEP 1 (Wilding et al., NEJM 2021, n=1961, 68 weeks) put semaglutide 2.4 mg weekly against placebo in adults with obesity and without diabetes. The result was -14.9% body weight against -2.4%, a difference of 12.4 percentage points. Half the treated group, 50.5%, lost 15% or more of body weight, against 4.9% on placebo.
SURMOUNT-1 (Jastreboff et al., NEJM 2022, n=2539, 72 weeks) tested tirzepatide at three doses: -15.0% at 5 mg, -19.5% at 10 mg and -20.9% at 15 mg, against -3.1% on placebo. In the 15 mg arm, 57% lost at least a fifth of body weight.
Retatrutide’s phase 2 in obesity (Jastreboff et al., NEJM 2023, n=338, 48 weeks) reported -8.7% at 1 mg, -17.1% at 4 mg, -22.8% at 8 mg and -24.2% at 12 mg against -2.1% on placebo.
That -24.2% figure is the number most content in this space still leads with, and it has been superseded. Phase 3 results began arriving in late 2025. TRIUMPH-4 reported in December 2025 in knee osteoarthritis with weight loss up to -28.7%. TRIUMPH-1 reported on 21 May 2026 (n=2,339): -17.6% at 4 mg, -23.7% at 9 mg and -25.0% at 12 mg against -3.9% on placebo at 80 weeks, reaching up to -30.3% at 104 weeks. TRIUMPH-2 and TRIUMPH-3 reported on 23 July 2026. Anyone still citing the phase 2 number as the current ceiling is roughly a year behind, and the phase 3 dose ladder is 4, 9 and 12 mg rather than the phase 2 ladder of 1, 4, 8 and 12 mg.
The mechanism story usually told to explain the ladder is only partly supported. Semaglutide copies GLP-1, so satiety signalling arrives sooner and the stomach empties more slowly. Tirzepatide copies GLP-1 and also the gut hormone GIP. What GIP contributes is genuinely unsettled: GIP receptor agonism and GIP receptor antagonism have both produced weight loss in humans when paired with GLP-1, and no trial in this set isolates the GIP arm. Anyone presenting the additive-agonism story as established is ahead of the field. Retatrutide adds glucagon receptor activity. In humans the well-established glucagon effects are hepatic, covering glycogenolysis, gluconeogenesis and increased hepatic fat oxidation, alongside a rise in energy expenditure whose tissue origin is not settled. Glucagon receptor expression in human fat tissue is low, and direct receptor-mediated lipolysis in human adipocytes is contested, so the popular “tells fat cells to burn stored energy” line is a rodent-flavoured simplification.
Retatrutide did drive dose-dependent heart-rate increases that peaked near week 24 before declining. It is tempting to pin that on the glucagon arm, but heart rate also rises with semaglutide and tirzepatide, neither of which has glucagon activity, so the data can’t carry that attribution.
One comparison gets made constantly and shouldn’t be. The retatrutide and tirzepatide headline numbers come from different trials, populations, durations and phases. The only true head-to-head in this set is SURMOUNT-5 (Aronne et al., NEJM 2025, n=751, 72 weeks), which gave tirzepatide -20.2% against semaglutide -13.7%, with waist circumference down 18.4 cm against 13.0 cm. That trial was open-label rather than blinded.
The lean mass question
Both camps online overstate their case, and one dataset shows why.
The SURMOUNT-1 DXA substudy (Look et al., Diabetes Obes Metab 2025) scanned 160 people at baseline and week 72. Tirzepatide reduced body weight 21.3%, fat mass 33.9% and lean mass 10.9%. Placebo reduced them 5.3%, 8.2% and 2.6%.
The informative part is the split. Of the weight actually lost, roughly 75% was fat and 25% was lean in the tirzepatide arm, and roughly the same ratio held in the placebo arm, consistent across sex, age and weight-loss tertile. The incretins didn’t preferentially spare lean tissue and didn’t preferentially strip it. They removed more total tissue in the same proportion. Neither “preserves muscle” nor “destroys muscle” survives contact with that result.
SEMALEAN (Alissou et al., Diabetes Obes Metab) complicates it from a different direction. It followed a prospective real-world cohort of 115 enrolled and 106 completed on semaglutide 2.4 mg with DXA plus function testing. Fat mass fell 14.3% at month 7 and 18.9% at month 12, with visceral fat significantly reduced. Lean mass fell 3.0 kg by month 7 and then stabilised. Handgrip strength rose 3.7 kg at month 7 and 4.1 kg at month 12, and sarcopenic obesity prevalence dropped from 49% at baseline to 33% at month 12. The study also found resting energy expenditure down 244 kcal per 24 hours at month 7, partially recovering by about 140 kcal at month 12, which is the most interesting metabolic-adaptation finding in the paper and rarely gets quoted.
SEMALEAN is single-arm and uncontrolled, so the handgrip improvement carries familiarisation and regression-to-the-mean components that a controlled design would separate out. Holding that caveat, lean kilograms on a scan and physical function are still not the same variable, and some of what DXA classifies as lean tissue is water, glycogen and connective tissue supporting fat mass that no longer exists.
The retatrutide body-composition substudy (Coskun et al., Lancet Diabetes Endocrinol 2025) is frequently attached to the obesity trial above, which is an error. It is a substudy of the phase 2 trial in people with type 2 diabetes, ran 36 weeks rather than 48, and its seven arms include a dulaglutide active comparator. It enrolled 189 with only 103 usable paired scans. Fat mass fell 26.1% at 8 mg pooled and 23.2% at 12 mg, a non-monotonic result that most likely reflects small-sample noise. The authors reported a lean-loss-to-weight-loss ratio similar to other obesity drugs.
Why the plateau isn’t a failure
Weight loss in these trials doesn’t run linear to the end. It decelerates, and where it flattens tracks dose and compound. The SEMALEAN pattern is instructive: fat mass kept falling between month 7 and month 12, from 14.3% to 18.9%, while lean mass stopped falling after month 7. What is coming off changes as the cycle runs.
The dose-response at the top of the ladder is shallow. In SURMOUNT-1, moving from 10 mg to 15 mg bought 1.4 additional percentage points of weight loss while carrying the full side-effect curve. That shape is worth knowing before reading any argument built on escalating past the studied range.
What the withdrawal data shows
This is the finding most at odds with how these compounds get marketed.
The STEP 1 off-treatment extension (Wilding et al., Diabetes Obes Metab 2022) followed 327 of the original participants for 52 weeks after both drug and lifestyle support stopped. They had lost 17.3% at week 68. By week 120 they had regained 11.6 percentage points, about two-thirds of what was lost, ending at -5.6% from baseline. Blood pressure, lipids and HbA1c reverted toward baseline alongside the weight.
That describes what the intervention is. It produces a maintained state rather than a completed one. No equivalent withdrawal data has been published for retatrutide at any dose.
The two compounds with no human data
HGH fragment 176-191 has no human trials of the unmodified fragment, and ClinicalTrials.gov returns no registered studies of it. Every human number in circulation belongs to AOD9604, which is the same fragment with an additional tyrosine residue at the N-terminus. Six randomised double-blind placebo-controlled trials of AOD9604 ran between 2001 and 2006 in approximately 900 subjects, the largest n=502 over 24 weeks. Routes were intravenous, at 25 to 400 µg/kg single doses, and oral capsule, at 0.25 mg/day up to 54 mg/day. No published human subcutaneous dosing exists for either molecule, which is the route the fragment is sold for.
The efficacy results were never peer-reviewed. The single peer-reviewed paper from that program (Stier, Vos & Kenley, J Endocrinol Metab 2013, 3(1-2), 7-15) reports safety, IGF-1 (non-significant at every dose, p 0.38 to 1.00) and glucose data, with no weight outcome at all, despite both phase IIb trials being designed to measure weight reduction. Metabolic Pharmaceuticals publicly announced that the phase IIb missed its endpoint and terminated the obesity program in 2007. The results going unpublished is a weaker claim than concealment, and it is the claim the record supports.
There is a plausible mechanism for why the rodent work didn’t translate. Heffernan et al. (Endocrinology 2001, 142(12), 5182-89) showed that in beta-3 adrenergic receptor knockout mice, chronic hGH and AOD9604 failed to produce the weight loss and lipolysis seen in wild-type animals. Beta-3 receptors are considerably more functionally significant in rodent fat tissue than in human. The mouse data was real. It may have been about a pathway humans don’t have much of.
5-Amino-1MQ blocks NNMT, an enzyme inside fat cells that consumes the cell’s NAD+ and methyl groups. Blocking it leaves fat cells with more NAD+ and shifts them toward burning rather than storing. The mechanism is elegant and has been tested only in mice.
The 2024 study (Babula et al., Diabetes Obes Metab, n=8 per group, 28 days) gave diet-induced obese mice 32 mg/kg/day subcutaneously and reported body weight gain of 0.9 g against 5.4 g in controls, with fat mass gain of 1.3 g against 4.7 g. At 10 mg/kg/day the effect disappeared, with 5.2 g of weight gain against the controls’ 5.4 g. Lean mass showed no significant treatment effect in any arm. Those figures sit behind a paywall and are not corroborated outside the paper itself, so weigh them accordingly.
Two things follow. There appears to be a dose threshold below which the compound does nothing. And oral bioavailability was reported at 3.5%, with a Cmax of 14.5 ng/mL at 30 mg/kg oral against 7,010 ng/mL at 25 mg/kg subcutaneous. That gap is a pharmacokinetic observation in mice rather than a demonstrated efficacy failure, and the poor oral absorption is reported as species-specific, occurring in mice but not rats. The capsule is the format most buyers encounter, and nobody has measured what it does in a human.
What the trials dosed
The parameters below are the studied protocols, recorded here as what researchers administered rather than as guidance for anyone.
Semaglutide was given at 2.4 mg once weekly, subcutaneous, reached through a 16-week escalation starting at 0.25 mg. STEP 1 ran 68 weeks on drug, with GI-related discontinuation at 4.5% against 0.8% on placebo.
Tirzepatide was given at 5, 10 or 15 mg once weekly, subcutaneous, with a 20-week escalation. SURMOUNT-1 ran 72 weeks, with adverse-event discontinuation between 4.3% and 7.1% against 2.6% on placebo. SURMOUNT-5 used a maximum tolerated dose of 10 or 15 mg.
Retatrutide’s phase 2 in obesity used 1, 4, 8 or 12 mg once weekly, subcutaneous, for 48 weeks. The phase 3 TRIUMPH ladder is 4, 9 and 12 mg. The phase 2 trial separately compared a 2 mg starting dose against 4 mg and found the lower start meaningfully reduced GI adverse events, which is the most directly useful escalation finding in the dataset and runs against the practice of starting high.
5-Amino-1MQ has mouse data only, at 32 mg/kg/day subcutaneous for effect and 10 mg/kg/day for none, over 28 days. An earlier protocol is often cited as 20 mg/kg three times daily for 11 days, which isn’t verifiable in the primary text. There is no human dose because there is no human trial.
HGH fragment 176-191 has no published human subcutaneous dose. The human range that exists belongs to AOD9604 by intravenous and oral routes.
What the trials monitored is also informative: GI tolerability during escalation, which is where discontinuation clusters. Heart rate for retatrutide specifically, given the dose-dependent rise. And DXA where it was measured at all, which was rarely.
Where this evidence is thinner than it looks
Body composition data is far weaker than the headline weight data even for approved drugs. STEP 1’s DXA substudy was n=140, semaglutide 95 and placebo 45, and explicitly exploratory. SURMOUNT-1’s was 160 of 2539 randomised. The retatrutide substudy had 103 usable scans across seven arms. None were powered for body composition. Every confident claim online about what fraction of your loss is muscle rests on these, and that includes the 75/25 figure above, which should be held as the best available estimate rather than as a settled number. The ratio also matters differently depending on the person: the same 25% carries different consequences for someone older or already sarcopenic, where absolute lean kilograms lost is the variable that counts.
No long-term body-composition data exists for any of them. The longest published DXA follow-up is 72 weeks. What lean mass, bone density and muscle function look like at three or five years, or across repeated on-off cycling, is unknown.
There are no adequately powered randomised trials of any of these compounds combined with structured resistance training and adequate protein. That combination is the single most-repeated recommendation in this space. It is physiologically reasonable and it is not trial-supported in incretin users specifically.
Grey-market product is not trial product. A 2026 preprint using a large EHR network reported compounded retatrutide users losing 7.2% of body weight at 6 to 12 months against 15.5% for trial participants whose records appeared in the same network, no better than matched tirzepatide users. Several of the figures in that preprint, including the network denominator and the p-value, could not be independently verified, and it has not been peer-reviewed. Product identity and potency are one explanation for the gap. Adherence, titration discipline, monitoring intensity and confounding by indication are all at least as plausible, and the design can’t separate them.
Retatrutide’s cardiovascular outcomes remain unpublished, which both the heart-rate signal and the energy-expenditure mechanism make worth waiting for. TRIUMPH-Outcomes (NCT06383390) is the cardio-kidney outcome trial and has not reported. Retatrutide is not approved by any regulator.
What to read next
Start with the STEP 1 extension paper (DOI 10.1111/dom.14725) rather than the original STEP 1, because the withdrawal data reframes the parent trial. Then the SURMOUNT-1 DXA substudy (DOI 10.1111/dom.16275) for the 75/25 split, and SEMALEAN (DOI 10.1111/dom.70141) for the function and energy-expenditure data that complicate it. Read Stier 2013 for the AOD9604 program and note what is absent from it. For retatrutide, read the TRIUMPH-1 phase 3 report rather than the 2023 phase 2 paper, and watch TRIUMPH-Outcomes for the first cardiovascular safety readout.
This article is a review of published research. It is not medical advice, and nothing here is a recommendation to use any compound. Several compounds discussed have no human data and are sold for laboratory research use only.
