Semaglutide: From Molecular Design to Cardiovascular Outcomes

Semaglutide is a synthetic analogue of human glucagon-like peptide-1, with the molecular formula C187H291N45O59 and a molecular weight of roughly 4,114 g/mol. It resolves in PubChem under CID 56843331 and carries CAS 910463-68-2.

The sequence is close to native GLP-1 but not identical, and the differences are the point. Native GLP-1 is cleaved within minutes by dipeptidyl peptidase-4, which cuts near the N-terminus. Substituting α-aminoisobutyric acid - a non-proteinogenic amino acid - at position 8 removes the residue the enzyme recognises, and the peptide survives. A lysine-to-arginine change at position 34 leaves a single lysine available for chemical modification, so acylation goes to one defined site rather than producing a mixture.

That site carries a C18 fatty diacid attached through a spacer. The diacid binds reversibly to serum albumin, and because albumin is large and slowly cleared, the bound peptide is effectively parked in circulation and released gradually. This is why semaglutide's half-life is measured in days rather than minutes. It is worth being precise about the mechanism: albumin binding changes exposure over time, not what the peptide does when it reaches a receptor.

At the GLP-1 receptor, semaglutide reproduces the actions of the native hormone: glucose-dependent insulin secretion, suppression of inappropriate glucagon release, and slowed gastric emptying. The glucose-dependence matters, because it is the reason GLP-1 receptor agonism does not carry the hypoglycaemia risk associated with insulin or sulfonylureas - the insulinotropic effect falls away as glucose falls.

The appetite effect is less straightforward, and the most useful evidence is preclinical. Gabery and colleagues worked in rodents to map where semaglutide actually reaches in the central nervous system. They found the peptide accessed particular brain regions, including areas outside the tightest parts of the blood-brain barrier, and that activity appeared across several distinct circuits rather than converging on one hypothalamic centre. Their conclusion - that body-weight lowering proceeds through distributed neural pathways - is a specific claim about anatomy, not a general statement that the drug acts on the brain.

This distinction deserves emphasis because the paper is frequently cited as though it described human neurobiology. It does not. It is rodent work, and it is cited here as the best available mechanistic account of a phenomenon whose human counterpart is documented at the level of outcome rather than circuit.

Semaglutide has one of the most extensive human evidence bases of any peptide, spread across three named programmes: SUSTAIN in type 2 diabetes, STEP in weight management, and the standalone SELECT cardiovascular outcome trial.

SUSTAIN-6, reported by Marso and colleagues in 2016, was designed as a cardiovascular safety trial and returned more than safety: it reported a lower rate of major adverse cardiovascular events on semaglutide than on placebo in adults with type 2 diabetes at high cardiovascular risk. The same paper reported a higher rate of diabetic retinopathy complications in the semaglutide arm. Whether that reflects the drug or the well-described phenomenon of transient retinopathy worsening after rapid glycaemic improvement has been argued since, but the finding is in the primary report and belongs in any honest summary.

STEP 1, reported by Wilding and colleagues in 2021, took the compound into obesity as an indication in its own right. Over 68 weeks in adults with overweight or obesity and without diabetes, mean body weight changed by approximately −14.9% against roughly −2.4% on placebo. This became the benchmark figure against which the dual and triple agonist results that followed were measured.

SELECT, reported by Lincoff and colleagues in 2023, is the trial that changed how the class is discussed. More than 17,000 participants with overweight or obesity and established cardiovascular disease, but without diabetes, were randomised; the semaglutide group had fewer major adverse cardiovascular events. Because the population excluded diabetes, the result cannot be attributed to glycaemic improvement, which is what earlier cardiovascular data in this class could not rule out.

Across the SUSTAIN, STEP and SELECT programmes, gastrointestinal adverse events dominate the reported profile: nausea, diarrhoea, vomiting and constipation, mostly mild to moderate, concentrated during escalation and diminishing thereafter. Discontinuation for gastrointestinal reasons is consistently reported at a low but non-trivial rate.

Beyond that, the literature documents the retinopathy signal from SUSTAIN-6 discussed above, an increase in gallbladder-related events that appears across weight-loss interventions generally rather than being specific to this class, and a rodent thyroid C-cell finding that has shaped labelling in several jurisdictions without a corresponding human signal emerging in the trial programmes. Distinguishing rodent findings from human observations is essential here; they are frequently reported as if interchangeable.

Discontinuation is the best-documented limitation. Extension work following STEP 1 found that weight and cardiometabolic measures moved back toward baseline after treatment stopped, which frames semaglutide as a maintained intervention rather than a one-time correction. What that means over a lifetime of exposure - rather than over the one to four years covered by the trials - is not established.

Body composition is the second gap. Weight reduction of the magnitude STEP 1 reported includes lean tissue as well as fat, and the trials characterise total weight far more thoroughly than its composition. The clinical significance of the lean-mass component, particularly in older participants, is an active question rather than a settled one.

Third, the mechanism of the cardiovascular benefit in SELECT is not resolved. Weight reduction, blood-pressure change, and direct effects on inflammation or the vasculature have all been proposed, and the trial was designed to detect the outcome rather than to attribute it. Finally, the human counterpart to the rodent circuit mapping does not exist at comparable resolution - what is known in humans is that the effect occurs, not which pathways carry it.