What we would need to randomise to answer this properly

Only one randomised trial has done the obvious experiment. In a Danish study of weight-loss maintenance, participants who had already lost weight on a low-energy diet were randomised to exercise alone, a GLP-1 receptor agonist alone, both together, or placebo, and followed for a year with body composition measured throughout. The combination group did better than either component on weight, on fat mass and on the proportion of the loss that was fat. It is a single trial, it used liraglutide rather than a current agent, and it remains the best evidence anybody has for the proposition that training changes the composition of pharmacological weight loss.

The method that measures muscle directly, and why nobody uses it

There is a technique that estimates whole-body skeletal muscle mass rather than inferring it from a subtraction. Deuterated creatine dilution involves an oral dose of labelled creatine, which distributes into the total creatine pool — almost all of which sits in skeletal muscle — with the enrichment of labelled creatinine in a subsequent urine sample giving an estimate of pool size and therefore of muscle mass.¹ It is not an imaging measure and it does not depend on regression equations fitted to a reference population.

Comparisons with DXA are instructive and slightly deflating. The two methods correlate only moderately in older adults, and where they disagree the creatine-dilution figure has been the better predictor of physical function and of incident disability. That is an argument that DXA appendicular lean mass, the standard proxy, is measuring something adjacent to what matters rather than the thing itself.

The method has been available for more than a decade. It has been used in no trial of any drug in this class. It requires a timed urine collection and a mass spectrometry laboratory, which is a modest imposition set against the volume of argument the absence of good muscle-mass data has generated.

The magnetic-resonance substudy in the diabetes programme

The most methodologically interesting composition data in this class did not come from an obesity trial. A magnetic-resonance imaging substudy within SURPASS-3, comparing tirzepatide against insulin degludec in type 2 diabetes, measured liver fat content and abdominal adipose tissue volumes rather than whole-body compartments.² Approximately three hundred participants were imaged, which makes it the largest imaging substudy in the programme.

Liver fat content fell substantially more on tirzepatide than on insulin, as did visceral adipose tissue volume, and the separation between the arms was larger than the difference in total body weight would predict. That is the single most useful composition finding in the class, because it shows the two interventions redistributing tissue differently rather than merely producing different amounts of weight change.

Magnetic resonance is the better instrument for this question by some distance: it measures adipose tissue volumes directly and separates visceral from subcutaneous depots, neither of which DXA does well. It is also expensive, slow and unavailable at most trial sites, which is why the whole-body composition argument is still being conducted on DXA data from a few hundred people.

What the substudies were never powered to detect

An imaging substudy inside a large trial is sized to describe rather than to test. The enrolment is set by how many participating sites have a scanner and by what the sponsor budgeted, not by a power calculation against a composition hypothesis, and the analysis is generally pre-specified as exploratory or descriptive. The consequence is that these substudies can report a mean change with a usable confidence interval and cannot support most of the questions asked of them.

They cannot, for instance, establish whether lean-mass change differs between dose arms, because the per-arm enrolment after splitting is in the low tens. They cannot establish whether it differs by age, sex, baseline adiposity or diabetes status, because those subgroups were not enrolled to be comparable. They cannot describe the distribution of individual responses, because the per-participant least significant change is a substantial fraction of the observed mean effect. And they cannot address function at all, because nobody measured it.

Nor was the imaging repeated when the programmes were extended. The two-year semaglutide extension reported weight, waist circumference and cardiometabolic parameters at week 104 and did not repeat the composition substudy, so there is no imaging at all beyond seventy-two weeks in this class.³ Whatever the trajectory of lean mass is in year two of treatment, nobody has measured it.

None of this is a scandal; it is the ordinary economics of trial substudies. It becomes a problem only when a descriptive group mean is quoted as though it characterised what will happen to an individual, which is now the normal register of coverage on this subject.

The endpoint nobody measured

The clinical question is not how many kilograms of lean tissue a person has. It is whether they can climb stairs, rise from a chair without using their arms, carry shopping, and recover from an illness that keeps them in bed for a week. Those are measurable — grip strength, gait speed, chair-stand time, stair-climb power, the short physical performance battery — and they are measured routinely in geriatrics and sports science. Not one phase 3 trial in this drug class has reported them as a pre-specified endpoint.

That absence is the strongest available criticism of the programmes, and it has been made in the general medical literature by authors who are otherwise unsympathetic to muscle-loss alarmism.⁴ Their argument is worth stating precisely: the concern about lean-mass loss is plausible but unquantified, the instrument used to assess it is a poor proxy for the tissue of interest, and the endpoints that would settle whether it matters are cheap, validated and were simply not collected.

Where function has been measured during substantial weight loss by other routes, the results are mostly reassuring: physical performance usually improves, because carrying less mass is itself a functional benefit. That is a reasonable prior and it is not a substitute for the measurement.

The one trial that combined an agonist with supervised exercise

A Danish randomised trial remains the only controlled test of the obvious question. After an eight-week low-energy diet producing approximately thirteen kilograms of weight loss, participants were randomised for one year to supervised exercise alone, liraglutide 3.0 mg alone, both combined, or placebo.⁵ The combination arm achieved the largest weight reduction and, more relevantly here, the most favourable composition outcome: body fat percentage fell roughly twice as much in the combination group as in either single-intervention group, and the exercise arms preserved lean mass better than the drug-alone arm.

Three qualifications belong with that result. The exercise was supervised and substantial — two group sessions and two individual sessions weekly, with a vigorous-intensity target — which is not what most people mean by adding exercise. The agent was liraglutide at 3.0 mg daily, producing considerably less weight loss than the current agents, so whether the interaction scales to a twenty per cent reduction is unknown. And the trial began after weight had already been lost, so it is a maintenance study rather than an induction study.

With those stated, it is the best evidence in the field and it points in the direction the general advice already points.

What the older-adult trials found

The closest analogue to rapid weight loss in an older, heavier population predates this drug class entirely. In a randomised trial of adults aged sixty-five and over with obesity, assigned to diet, exercise, both or a control condition for a year, the combination produced the largest improvement in physical function, and the exercise component attenuated the loss of lean mass and of bone mineral density that diet alone caused.⁶ Diet alone improved function too — carrying less mass helps — but by less, and at a measurable skeletal cost.

That trial is the template for how the question should be asked in this class: randomise the co-intervention, measure function as a primary endpoint, measure bone, and follow for long enough for the skeleton to respond. Its population, older and heavier and losing weight quickly, resembles a large share of current incretin users far more closely than the young resistance-trained cohorts from which most consumer advice descends.

The Journal cites it frequently for that reason and notes the obvious limitation: the weight loss achieved was roughly a tenth of body mass over a year, which is half or less of what the current agents produce. Whether the protective effect of training holds at twice the rate of loss is not established.

The prescription survives scrutiny; the reasoning often does not

Two claims are routinely bundled together and only one is well supported. The weaker claim is that resistance training during pharmacological weight loss builds or maintains muscle mass. In a substantial energy deficit, training generally attenuates the loss rather than preventing it, and net accrual is unusual outside of untrained beginners and the specific controlled-feeding conditions of the trials cited earlier. The stronger claim is that training preserves strength and physical function even where mass declines, which is consistently observed and is mechanistically sensible: a large part of early strength change is neural rather than structural.

The distinction has practical consequences. Somebody training hard, eating well, and watching their DXA appendicular lean mass fall by two kilograms across nine months has not failed at anything, and may be measurably stronger than at baseline. If the expectation set for them was mass preservation, they will read a normal outcome as a failure and may respond by eating more or training in ways that suit the metric rather than the goal.

The Journal reports the training recommendation and reports what it is expected to achieve, which is function first and mass second.

Is there a drug-specific skeletal effect?

Two hypotheses compete and both are underpowered. The first is that incretins are neutral for bone beyond making their users lighter, so any density change is the ordinary consequence of reduced mechanical loading. The second is that GLP-1 receptor signalling has direct skeletal effects — receptors have been reported on osteoblast lineage cells, and GLP-1 influences the entero-osseous axis and calcitonin secretion — which could be protective, harmful, or negligible.

The evidence cited for a protective effect is an early study of weight-loss maintenance in which liraglutide treatment was associated with preserved bone mineral density relative to a diet-alone comparison, interpreted at the time as a direct skeletal benefit.⁷ That finding sits awkwardly beside the later secondary analysis in which the agonist arm did worse than the exercise arms, and the two are not straightforwardly reconcilable: different agents at different doses, different comparators, different durations, small samples throughout.

The Journal reports the question as open, which is unsatisfying and accurate. What would settle it is a randomised bone endpoint with imaging that is not confounded by soft-tissue change, in a population whose weight loss is matched across arms. Nothing of that description is under way.

How the Journal reports a body-composition figure

Four things accompany every composition number in these pages. The instrument, because DXA, magnetic resonance, bioimpedance and creatine dilution are not interchangeable and the choice frequently determines the sign of the result. The sample size of the substudy rather than of the parent trial, because the parent trial size is irrelevant to the composition finding and quoting it is misleading. The definition used — total lean mass, lean soft tissue, appendicular lean mass or fat-free mass — because these differ by several kilograms in the same person. And whether the figure is a proportion of body mass or an absolute quantity.

Where a source omits any of the four, we say so rather than guessing, and where we have had to convert between definitions we show the conversion. This is more cumbersome than the alternative and it is the only way we have found to write about this subject without producing sentences that are technically true and practically misleading.

Readers who find a figure in these pages that lacks its instrument and its sample size have found an error, and the standards desk would like to hear about it at standards@compoundjournal.com.

What the testing services can and cannot tell you here

A category confusion arrives in the Journal postbag with some regularity, and it is worth addressing directly. The four independent testing services this market relies on — Janoshik, Medutest, PeptideMeter and VendorInvestigate — analyse the contents of a vial. They report chromatographic purity, identity by mass, sometimes peptide content, and in the case of the verification services, what they were able to establish about a supplier. None of them measures anything about a person.

A certificate stating 98.7 per cent purity for a batch supplied by WWB, SSA or KP is silent on that customer’s body composition, and a low-purity result does not explain a disappointing DXA scan. The two questions are answered by different instruments in different buildings, and conflating them produces a particular kind of dead end in which somebody spends several hundred pounds on analytical testing to investigate a clinical question.

The reverse confusion also occurs: a satisfactory laboratory panel or a favourable body-composition scan is offered as evidence that a vial contained what its label claimed. It is not evidence of that either. Compounds sold for research use only are not approved for human use, and nothing in this section should be read as advice about using them.

Readers should be sceptical of any body-composition figure quoted without its instrument, and sceptical of their own scans taken less than six months apart on different machines. The measurement error in this field is not a technicality; it is comparable in size to the effects being discussed, and it is the reason the same substudy tables support opposite conclusions in different hands.