Skip to content
GLP Loss
← Research
Evidence

GLP-1 Body Composition: What the DXA and MRI Substudies Found

In the SURMOUNT-1 scan substudy, about 75% of the weight lost was fat and 25% was lean — on tirzepatide and on placebo alike. The drug changed the size of the loss, not its composition.

Tessa Whitfield9 min read
Where the weight came from, in both armsDXA substudy: 160 scanned out of 2,539 randomizedTirzepatide, 21.3% of body weight lostabout 75% fat mass25% leanPlacebo, 5.3% of body weight lostabout 75% fat mass25% leanWhat the drug changedThe size of the loss. Not what the loss was made of.Fat mass fell 33.9% against 8.2%. Lean mass fell 10.9% against 2.6%.No study in the newest systematic review measured physical function.

The one randomized body-composition dataset in the tirzepatide obesity program scanned 160 of the 2,539 people in SURMOUNT-1 with dual-energy X-ray absorptiometry at baseline and at week 72 — 124 on pooled tirzepatide doses, 36 on placebo, 73% female, mean weight 102.5 kg. Body weight fell 21.3%, fat mass 33.9% and lean mass 10.9% on tirzepatide, against 5.3%, 8.2% and 2.6% on placebo, all at P < 0.001.[1]

Then comes the sentence that does the work. Of the weight lost, approximately 75% was fat and 25% was lean, for tirzepatide and for placebo alike, and those proportions held across subgroups of sex, age and weight-loss tertile.[1] The drug moved the size of the loss. It did not move what the loss was made of. Anyone reading a lean-mass figure as a property of the molecule rather than of losing weight is reading a number the study does not support, and the pooled kilogram estimates behind that reading sit in the muscle article.

Both loud versions of the story fail on the same table

The alarming version says these drugs strip muscle. The placebo arm in the same substudy lost the same fraction of its smaller loss as lean tissue, so whatever is happening is what happens when a body gets lighter, not what happens when a GIP and GLP-1 receptor agonist is injected.

The reassuring version says the drug preserves lean mass. It does not do that either: lean mass fell 10.9%, roughly four times the placebo change, because four times as much total weight came off. Preservation would mean a smaller lean share, and the substudy reports no such thing.

How much lean tissue is too much, and who set that number

A 2026 systematic review screened 8,102 records and assembled 35 randomized studies reporting body composition on liraglutide, semaglutide, tirzepatide or dulaglutide. Only 10 of them (28.6%) had prespecified body composition as a primary outcome, which is the first thing worth knowing about this literature.[2]

The review applied benchmarks set in advance: about 25% of total weight loss for fat-free or lean soft tissue measured by bioelectrical impedance or DXA, and about 15% for skeletal muscle measured by computed tomography or magnetic resonance. Within the incretin arms the median proportion of weight loss attributable to muscle-based indices was 28.3% (IQR, 15.9% to 39.9%), with 65% of arms exceeding their benchmark. By imaging method it was about 29% (IQR 16.6% to 43.1%) for impedance and DXA, and about 25.3% (IQR 16.7% to 27.2%) for CT and MRI.[2]

That looks damning until the comparator arrives. Among the 13 studies reporting the lifestyle or placebo groups — which lost a median of just 2.5% of body weight — 38% also exceeded their benchmark.[2] Nearly half of the non-drug weight loss crossed the same line. The review's own conclusion names both halves: muscle-related losses exceeded the benchmarks in about two thirds of incretin interventions and in nearly half of the non-pharmacologic ones that produced weight loss.

One limitation in that paper matters more than the medians. No study reported objective physical function outcomes.[2] Thirty-five randomized studies scanned people and none of them asked whether the people could do more or less afterward. Whether training changes any of this is a separate and mostly untested question, covered in the exercise article.

The instrument decides the answer, and the instruments disagree

The two benchmarks above are not the same measurement with different tolerances. Impedance and DXA report fat-free or lean soft tissue, a compartment that includes water, glycogen, connective tissue and organ mass, and every kilogram of glycogen carries roughly three of water with it. Computed tomography and magnetic resonance measure a muscle cross-section directly. A 25% figure from one and a 15% figure from the other are different quantities, and a page that converts between them without saying so is inventing precision.

The one randomized DXA comparison against an active drug rather than a placebo makes the same point about interpretation. In a SUSTAIN 8 substudy, 178 adults with type 2 diabetes were scanned and 114 had end-of-treatment data: total lean mass fell 2.3 kg on semaglutide 1.0 mg and 1.5 kg on canagliflozin 300 mg, an estimated treatment difference of −0.78 kg (95% CI, −1.61 to 0.04), while the difference in the proportion of lean mass was 0.14 (95% CI, −0.89 to 1.17).[3] Both intervals include zero. The authors state plainly that without a placebo arm, the specific impact of either treatment on body composition is speculative.

What the MRI substudies saw that a scale cannot

Fat is not one tissue, and the imaging work is where that becomes a number. A prespecified MRI substudy of SURPASS-3 enrolled 296 insulin-naive adults with type 2 diabetes and a fatty liver index of at least 60, randomized to tirzepatide 5, 10 or 15 mg or titrated insulin degludec. From a mean baseline liver fat content of 15.71%, the absolute reduction at week 52 was −8.09% on pooled tirzepatide 10 and 15 mg against −3.38% on insulin degludec, an estimated treatment difference of −4.71% (95% CI, −6.72 to −2.70; P < 0.0001).[4]

The correlations in that substudy are the more interesting result. The liver fat reduction correlated with baseline liver fat at ρ = −0.71, but with visceral adipose tissue at only 0.29, with abdominal subcutaneous adipose tissue at 0.33 and with body weight at 0.34.[4] Weight change explains a small minority of the variance in liver fat change. That is an open-label trial in people with diabetes, so it does not transfer cleanly to a cash-pay weight buyer, and the wider hepatic picture is in the liver article.

A follow-up analysis normalized the same 296 participants against sex- and BMI-matched virtual control groups drawn from the UK Biobank imaging study of 40,172 people, so the question becomes how each depot moved relative to people of the same size. At baseline the participants carried more visceral fat (z-VAT +0.42) and far more liver fat (z-LF +1.24) than matched controls, both at P < 0.001.[5]

By week 52 on tirzepatide, z-VAT fell by 0.18 and z-LF by 0.54, both at P < 0.001 — and z-abdominal subcutaneous fat rose by 0.11 (P = 0.012), from a starting value below the matched average.[5] Relative to people of the same body size, one fat depot went up while two went down. Total fat mass fell throughout. A sentence reporting only the total is true and describes none of that.

Some of the lean tissue that came off was heart

The cardiac magnetic resonance substudy of SUMMIT scanned patients with obesity-related heart failure with preserved ejection fraction, of whom 106 had adequate images at baseline and 52 weeks. Left ventricular mass fell by 11 g (95% CI, −19 to −4) against placebo (P = 0.004), and paracardiac adipose tissue by 45 mL (95% CI, −69 to −22; P < 0.001). The change in left ventricular mass tracked the change in body weight, and the authors read it as reverse remodeling that may contribute to the heart-failure benefit of the parent trial.[6]

Eleven grams of myocardium is lean tissue, and it is reported as an improvement, in a population selected for a cardiac condition in which a thickened ventricle is part of the disease. That is not a finding about a healthy person losing weight. It is the cleanest available demonstration that “lean mass” is a bag containing tissues whose loss means opposite things, which is why the compartment total is a poor place to hang a decision. Skeletal mass raises the same ambiguity about the parts of the body nobody scans, and the bone article covers that separately.

What none of these scans can tell a buyer

Every scan above was performed on branded product at labeled doses inside a trial protocol. Most sellers covered here dispense compounded semaglutide or tirzepatide, which is not FDA-approved and is not reviewed by the FDA for safety, efficacy or quality before it is dispensed; no DXA or MRI substudy has ever been run on a compounded preparation.

The durations are the other ceiling. The DXA substudy ran 72 weeks, the MRI substudy 52 and the cardiac substudy 52. Nothing here describes what several years of continuous treatment does to any compartment, and the protein and resistance-training protocols sold alongside these drugs were not randomized arms in any of it — which is the gap the diet article works through, and the reason a program listed on the tirzepatide board claiming body-composition benefits is describing something no scan has tested.

Frequently asked

What share of the weight lost on a GLP-1 is fat?
In the SURMOUNT-1 scan substudy, roughly 75% of the weight lost was fat mass and roughly 25% was lean mass. The same split appeared in the placebo group, which lost 5.3% of body weight against 21.3% on tirzepatide. The proportions held across subgroups of sex, age and total weight lost.
Do these drugs protect lean mass?
Nothing in the substudy data shows that. Lean mass fell 10.9% on tirzepatide against 2.6% on placebo, and the fraction of total loss coming from lean tissue was the same in both arms. A protective effect would show up as a smaller lean share on the drug, and it does not.
Is the lean-mass loss worse than with dieting?
The newest systematic review applied prespecified benchmarks and found that 65% of incretin arms exceeded theirs, against 38% of the lifestyle or placebo groups that lost weight. Those comparator groups lost a median of only 2.5% of body weight, so the comparison is not matched for the amount lost, and no randomized trial has matched it.
What did the MRI studies add that DXA could not?
They separated the fat depots. In the SURPASS-3 MRI substudy, liver fat fell 8.09 percentage points on pooled tirzepatide 10 and 15 mg against 3.38 on insulin degludec, and that change correlated only weakly with weight change. A follow-up analysis against BMI-matched UK Biobank controls found visceral and liver fat falling relative to matched peers while abdominal subcutaneous fat rose.
Does losing lean mass mean losing strength?
The scans cannot answer that. Of the 35 randomized studies in the 2026 systematic review, none reported an objective physical function outcome. Lean soft tissue as DXA measures it also includes water, glycogen, connective tissue and organ mass, so a kilogram of it is not a kilogram of usable muscle.

Sources

  1. [1] Look M, Dunn JP, Kushner RF, et al. (2025). Body composition changes during weight reduction with tirzepatide in the SURMOUNT-1 study of adults with obesity or overweight. Diabetes Obes Metab. PMID 39996356
  2. [2] Batsis JA, Gavras A, Gross DC, et al. (2026). Effect of Incretin-Based and Nonpharmacologic Weight Loss on Body Composition : A Systematic Review. Ann Intern Med. PMID 41996180
  3. [3] McCrimmon RJ, Catarig AM, Frias JP, et al. (2020). Effects of once-weekly semaglutide vs once-daily canagliflozin on body composition in type 2 diabetes: a substudy of the SUSTAIN 8 randomised controlled clinical trial. Diabetologia. PMID 31897524
  4. [4] Gastaldelli A, Cusi K, Fernández Landó L, et al. (2022). Effect of tirzepatide versus insulin degludec on liver fat content and abdominal adipose tissue in people with type 2 diabetes (SURPASS-3 MRI): a substudy of the randomised, open-label, parallel-group, phase 3 SURPASS-3 trial. Lancet Diabetes Endocrinol. PMID 35468325
  5. [5] Cariou B, Linge J, Neeland IJ, et al. (2024). Effect of tirzepatide on body fat distribution pattern in people with type 2 diabetes. Diabetes Obes Metab. PMID 38528819
  6. [6] Kramer CM, Borlaug BA, Zile MR, et al. (2025). Tirzepatide Reduces LV Mass and Paracardiac Adipose Tissue in Obesity-Related Heart Failure: SUMMIT CMR Substudy. J Am Coll Cardiol. PMID 39566869

Where to get it

Best GLP-1 injections

Every injectable seller we can verify, with the price each one publishes and an honest read of what the trials measured.

Compare providers →

More in Evidence