The 39% Question: Synthesizing the GLP-1 Muscle-Loss Literature

The number

Thirty-nine percent.

That is the figure Harvard Science Review put into circulation on 2026-02-23, framing the STEP 1 trial — Wilding et al., NEJM 2021 — as evidence that thirty-nine percent of the weight people lose on semaglutide is lean mass, not fat. The framing went viral. Mainstream wellness press picked it up the same week. By March, it was a line in every "are GLP-1s safe" YouTube essay and most of the sober peptide-cautious Substack pieces.

/

The number did not come from STEP 1 directly. It is a downstream reframing of the trial's body-composition substudy — the share of total weight loss that DEXA scored as lean mass — popularized by the Harvard SR write-up. And the framing has been pushed back on. Stuart Phillips and others have argued for two decades that DEXA's "lean" compartment conflates contractile muscle with water, glycogen, and connective tissue, which inflates the apparent lean-mass component during rapid weight loss. Newer longitudinal data has reinforced the critique. The SEMALEAN study — Alissou and colleagues, Diabetes, Obesity and Metabolism, January 2026 — followed semaglutide patients with serial DXA. Lean mass dropped roughly three kilograms at seven months, then stabilized through twelve. The proportion of lean mass to total body mass actually increased. Same methodology, longer time horizon, different read. The story is contested at the methodology level. Most of the coverage you have seen this spring did not tell you that.

This issue is the synthesis Issue 1 promised when it tier-tagged this debate as one of the three external arguments the Brief had to answer.

Why this is contested

Walk it through in plain language.

STEP 1 is the underlying RCT. Wilding and colleagues, 2021, New England Journal of Medicine. Roughly two thousand participants, semaglutide 2.4mg weekly versus placebo, sixty-eight weeks, mean weight loss around fifteen percent. The trial's primary endpoint was weight loss; body composition was a substudy on a smaller subset using DEXA imaging. That substudy is where the lean-mass conversation actually lives.

The Harvard SR framing took the body-composition substudy and reported the lean-mass component as roughly thirty-nine percent of total weight lost. That math is internally consistent with what the Wilding paper actually reported — a 9.7% absolute decrease in lean mass against a 15% decrease in total weight. The argument is not over the arithmetic. It is over how to read DEXA when most of the variance lives in the assumptions the imaging modality builds into its compartment model.

Body composition is not measured directly. It is inferred from imaging — DEXA, bioimpedance, sometimes MRI — and the inference depends on assumptions the imaging modality builds into its model. DEXA partitions the body into fat, lean soft tissue, and bone mineral. "Lean" in DEXA includes water, organ mass, and connective tissue, not only contractile muscle. When a person loses weight rapidly, water shifts. Glycogen stores deplete. Some of what DEXA scores as "lean mass loss" is contractile muscle; some of it is water and glycogen the person re-acquires within weeks of stabilizing. MRI partitions differently. Different methodology, different number.

That is the heart of the divergence. The Harvard SR framing and the methodologists who push back on it are not arguing about whether STEP 1 happened. They are arguing about how to read its body-composition substudy. The same trial supports both reads depending on which body-composition model you trust and how much of the early lean-mass drop you are willing to attribute to water and glycogen rather than contractile tissue.

Stuart Phillips, the McMaster muscle physiologist named in Issue 3, has spent two decades pointing at exactly this confound — that "lean mass loss" in caloric-restriction studies routinely conflates true contractile muscle loss with water and glycogen artifacts, and that the magnitude of the artifact is methodology-specific. His framework is the methodological backbone the SEMALEAN-style longitudinal reads draw on, and the reason "lean mass dropped then stabilized" is a more honest sentence than "thirty-nine percent of weight lost is lean mass."

The tier breakdown

Issue 1 set the tiered-evidence framework. Apply it here.

Tier 1 — Randomized controlled trials in humans. STEP 1 itself is Tier 1. The trial happened. Weight loss was real and large. The body-composition substudy is also Tier 1 in the sense that it is a substudy of a Tier 1 trial. The interpretation of the substudy — thirty-nine percent versus the lower figure — is downstream. A reader who only knows the headline number does not know which interpretation they are reading.

Tier 2 — Open-label and observational studies on lean-mass preservation under GLP-1. SEMALEAN sits here. So does a growing literature on adding resistance training and elevated protein intake during a GLP-1 cut. The signal across these studies is consistent: lean mass drops early and stabilizes; when participants train and eat protein at the upper end of the standard one-point-six to two-point-two grams per kilogram range, the lean-mass component of weight loss compresses substantially. These studies are smaller and shorter than STEP 1 and they are not blinded, but the direction is durable.

Tier 3 — Animal models on GLP-1 and muscle. Mechanistic work on whether GLP-1 signaling affects muscle protein synthesis directly versus indirectly through caloric restriction. The literature is mixed; the cleanest reading is that most of the lean-mass effect is the caloric deficit, not the receptor.

Tier 4 — Practitioner aggregate. Operators who have run sema or tirz cuts and tracked DEXA pre and post report a wide range of lean-mass outcomes that correlate strongly with whether the person trained and how much protein they ate. This is the lowest tier of evidence and the most consistent signal across all four.

The load-bearing teach: the same headline number sits at Tier 1 in one analysis and at "downstream re-interpretation of a Tier 1 substudy" in another. Tier-tagging is not a way to dismiss evidence. It is a way to read evidence honestly.

The risk model mainstream coverage is missing

Mainstream coverage treats GLP-1 muscle loss as a universal warning. The literature does not support that framing. The mortality-risk literature on lean-mass loss is age- and body-composition stratified, and the stratification is the part nobody is writing.

A simple 2x2.

Read the grid as descriptive, not prescriptive. The mortality-risk literature suggests these patterns; it does not predict an individual outcome. A thirty-two-year-old at twenty-five percent body fat losing thirty-nine percent of weight as lean mass is a different problem than a sixty-eight-year-old at borderline sarcopenia losing the same fraction. The headline number is the same. The risk model is not.

That is the synthesis position the Brief has been holding back from Issue 1. The mainstream peptide commentary frame — "GLP-1 muscle loss is a universal warning" — is wrong because it ignores the quadrant. The contrarian peptide-friendly frame — "muscle loss is overstated, ignore it" — is wrong because the older-and-lean quadrant is real and the older-and-higher-BF literature is genuinely thin. Both frames are reading the same paper at the wrong resolution.

Mitigators — general principles only

The dose math is downstream and lives behind the email gate. The principles are public.

Protein intake. Tier 2 evidence is consistent that elevated protein during a GLP-1 cut reduces the lean-mass component of weight loss. The standard literature range cited in the resistance-training and weight-loss studies is one-point-six to two-point-two grams per kilogram of body weight per day. The mechanism is straightforward — muscle protein synthesis is substrate-limited under a caloric deficit, and the substrate is amino acids.

Resistance training. Tier 2 again. Two to four resistance sessions per week, prioritizing compound movements at moderate-to-heavy loads, consistently compresses the lean-mass component of weight loss across every published study that has measured it. The mechanism is also straightforward — contractile activity is the signal that tells the body which tissue to keep.

Dose pacing. Tier 4, with mechanistic support from Tier 3. Faster weight loss tends to come with a higher lean-mass fraction; slower weight loss tends to preserve more lean mass. The titration ladder a manufacturer publishes for a long-half-life GLP-1 is a steady-state ladder, not a "go faster" ladder. The Issue 3 schedule math is the methodology here. The per-compound dosing — what pace, what dose, what interval, what off-ramp — lives behind the email gate.

What the Brief recommends a reader do with this

Three short bullets.

- Know your quadrant. The 2x2 above is the first read.

- Log a DEXA before starting. Without a pre-scan, the post-scan is uninterpretable. The methodology divergence at the top of this issue is exactly why a personal pre-baseline is the only number that lets you read your own outcome honestly.

- Resistance-train through the cut. Tier 2 is consistent; Tier 4 is consistent; the mechanism is consistent. The schedule is two to four sessions per week, compound movements, moderate-to-heavy loads.

That is what the survivors are doing in the field. Frame it as field practice, not protocol.

Next issue tease

Issue 05 — Protocols are downstream. The Wolverine Stack, re-examined: what BPC-157 + TB-500 literature actually says in 2026. The flagship recovery stack, walked through with the same tier framework. What the receptor work supports, what the field is reporting, what the literature actually does and does not say.

CTA

Reply with your DEXA quadrant. We will segment future issues.

Colby

—

For research use only. Not medical advice. Compounds discussed are sold for research purposes; nothing here is a recommendation to use them on humans or animals. The team behind The Compound also operates an RUO peptide vendor; that relationship is disclosed on the about page and applied to every issue under the Conflict Test.