Integrative Biomedical Research

1. Introduction

Obesity is no longer described, at least not in the serious literature, as a failure of individual restraint. It is a chronic, relapsing, multi-factorial disease in which genetic predisposition, neuroendocrine regulation of appetite, behaviour and an obesogenic food environment all contribute, and in which the body defends an elevated fat mass with considerable determination (Müller et al., 2022). That reframing matters clinically, because it predicts what decades of practice have confirmed: interventions that do not alter the underlying regulatory set-point tend to be undone by it.

The epidemiology gives the reframing its urgency. Roughly two billion adults were classified as overweight or obese in 2021, a figure projected to approach 3.8 billion by 2050, while the global prevalence of type 2 diabetes mellitus (T2DM), a condition tethered closely to excess adiposity, exceeded 500 million in the same year and is expected to reach 1.3 billion (Xu et al., 2025). Severe obesity does not travel alone. It drags behind it cardiovascular disease, chronic kidney disease and metabolic dysfunction-associated steatohepatitis (MASH), each with its own mortality burden and its own treatment costs (Müller et al., 2022; Xu et al., 2025). Economic projections place the annual global cost of overweight and obesity at approximately $4.32 trillion by 2035 (Xu et al., 2025). One should treat such forecasts with the scepticism any thirty-year extrapolation deserves, but even a substantial overestimate would describe a public health problem of the first order.

Against that backdrop, the therapeutic record until quite recently was thin. Behavioural and lifestyle programmes, delivered well and sustained, typically produce weight reductions in the range of 5 to 10%, and metabolic adaptation reliably erodes much of that over the following year or two (Müller et al., 2022; Østergaard, 2025). Earlier pharmacological agents rarely did better and several were withdrawn for cardiovascular or psychiatric toxicity. Bariatric surgery stood, and to some extent still stands, as the proof that the disease is reversible: 15 to 30% sustained weight loss, remission of T2DM in a large fraction of patients, and reduced cardiovascular mortality (Müller et al., 2022). What surgery could not offer was scale. An operation with perioperative risk, permanent anatomical consequences and a finite supply of surgical capacity was never going to be delivered to several hundred million people. The gap this left, between what was effective and what was deliverable, defined the problem that peptide chemistry has spent the last fifteen years attempting to close (Østergaard, 2025).

Why peptides, though, and not something more conventional? The answer lies in an awkward feature of the targets themselves. Appetite and nutrient sensing are governed largely by class B G-protein-coupled receptors, whose endogenous ligands are peptide hormones and whose binding grooves are long, shallow and extended rather than deep and enclosed (Otvos & Wade, 2023; Wang et al., 2022). Classical small molecules of under 500 to 1,000 Da present contact surfaces of roughly 300 to 1,000 square angstroms, which is simply insufficient for interfaces that typically demand 1,500 to 3,000 square angstroms, and attempts to compensate with higher doses tend to surface off-target and metabolite liabilities instead (Liu et al., 2025; Otvos & Wade, 2023). Monoclonal antibodies solve the affinity problem and, through neonatal Fc receptor recycling, the half-life problem as well, but at roughly 150 kDa they cannot cross membranes, require mammalian expression systems, and commit the patient to injection and the payer to considerable expense (Otvos & Wade, 2023; Ye et al., 2025). Neither end of the size spectrum was well matched to the biology (Figure 2).

Peptides and macrocycles between about 500 and 3,000 Da occupy the intermediate territory, and the metaphor of a "Goldilocks" class, though it has been used often enough to have lost some of its charm, remains apt (Otvos & Wade, 2023; Xiao et al., 2026). These molecules retain enough surface area to engage class B GPCRs at nanomolar to picomolar potency, are assembled synthetically rather than biologically, and are catabolised to their constituent amino acids, which removes an entire category of metabolite toxicity that troubles small-molecule development (Wang et al., 2022; Zhang et al., 2025). Their weakness, and it is a serious one, is fragility. Native glucagon-like peptide-1 (GLP-1) survives about two minutes in plasma before dipeptidyl peptidase-4 (DPP-4) cleaves it, and what escapes the protease is filtered at the glomerulus (Knudsen & Lau, 2019; Lau et al., 2015). Turning such a molecule into a once-weekly medicine required solving a chemistry problem before any clinical question could even be asked (Table 1).

It was solved, and the consequences have been considerable. Lipidation with fatty diacids, anchored through hydrophilic spacers to a lysine side chain, creates a reversible albumin depot that extends circulating half-life from minutes to days (Lau et al., 2015; Østergaard, 2025). On that chemical foundation the pharmacology was then elaborated: first selective GLP-1 receptor (GLP-1R) agonism, then unimolecular dual agonism at GLP-1R and the glucose-dependent insulinotropic polypeptide receptor (GIPR), then triple agonism adding the glucagon receptor (GcgR), and most recently co-agonism at amylin and calcitonin receptors (Coskun et al., 2018; Jastreboff et al., 2023; Dahl et al., 2025). Mean weight loss climbed from roughly 15% to roughly 24% across these generations, which places the best current agents within sight of surgical outcomes (Table 2). Whether that trajectory continues is one of the open questions this review examines.

Success has, predictably, produced its own difficulties. Gastrointestinal intolerance caps dose escalation more tightly than receptor pharmacology does. Oral administration, the route patients actually prefer for chronic therapy, remains constrained by a gastrointestinal environment evolved specifically to destroy peptides, and the one approved oral peptide in this class achieves bioavailability of well under 1% (Buckley et al., 2018; Table 3). And the manufacture of these molecules at population scale collides with a solvent burden that neither regulators nor the industry's own sustainability commitments can comfortably accommodate, with process mass intensity for early-phase synthetic peptides averaging near 33,000 (Kekessie et al., 2024; Table 4).

This review therefore sets out to do four things. It evaluates the structural design principles that make therapeutic peptides viable, examining cyclisation, stapling, N-methylation and diacid acylation as a coherent toolkit rather than as isolated tricks. It synthesises the clinical evidence on multi-receptor poly-pharmacology, comparing dual and triple agonists against mono-agonist benchmarks and asking what each receptor arm actually contributes. It assesses the emerging modalities that sit at the boundaries of the peptide field, including antibody-peptide conjugates, cell-penetrating macrocycles and non-peptide oral GLP-1R mimetics. And it addresses the industrial and environmental question of how any of this reaches the populations described at the outset. Throughout, the aim is integration rather than enumeration: the chemistry, the pharmacology, the delivery and the manufacture are not four separate stories but four constraints acting on the same molecule.

2. Chemistry, Receptor Pharmacology and Translational Delivery of Mid-Sized Modalities

2.1 The Molecular Evolution of a Neglected Chemical Space

For most of the modern pharmaceutical era, drug discovery operated with a curiously binary view of chemical space. On one side sat synthetic small molecules; on the other, macromolecular biologics. The space in between was not so much rejected as regarded as unworkable, largely because of the pharmacokinetic liabilities discussed below (Otvos & Wade, 2023; Wang et al., 2022). That this space has since produced some of the most commercially and clinically significant drugs of the decade is a reminder that "undruggable" is often a statement about available methods rather than about molecules.

The case for small molecules was never in doubt: low cost of goods, established manufacturing, oral bioavailability, and the ability to reach intracellular targets (de la Torre & Albericio, 2022; Liu et al., 2025). The limitation is geometric. A ligand of 300 to 1,000 square angstroms of contact surface cannot productively engage a protein-protein interface requiring 1,500 to 3,000, and the shallow, extended orthosteric sites of class B GPCRs fall squarely in the latter category (Liu et al., 2025; Otvos & Wade, 2023). Biologics invert every one of these properties. They bind with exquisite selectivity, persist for weeks through FcRn-mediated recycling, and are essentially restricted to extracellular targets, mammalian expression and parenteral delivery (Ye et al., 2025; Zhang et al., 2025).

Peptides and macrocycles bridge the divide, and do so in a way that appears, in retrospect, almost obvious (Figure 2). They inherit the selectivity and the amino-acid-based safety profile of biologics while retaining the synthetic tractability of small molecules, and because they mimic endogenous hormones they engage their receptors with potencies in the nanomolar to picomolar range (Bergamaschi et al., 2026; Xiao et al., 2026). The safety argument deserves emphasis because it is sometimes understated: catabolism to natural amino acids means there is generally no reactive metabolite to worry about, which removes a failure mode that has terminated many small-molecule programmes late in development (Wang et al., 2022; Zhang et al., 2025).

What stood in the way was stability. An unmodified linear peptide is a substrate for every protease it encounters and, once below the approximately 60 kDa glomerular threshold, is cleared renally within minutes (Liu et al., 2025; Zhang et al., 2025). The literature of the past two decades is, to a considerable extent, a record of chemists learning to defend a peptide backbone without destroying the receptor engagement that makes it useful. Macrocyclisation, whether head-to-tail, side-chain lactam or disulfide-mediated, eliminates the free termini that exopeptidases recognise and pre-organises the molecule into its bioactive conformation, which lowers the entropic

Figure 1. Evidence identification, appraisal and synthesis workflow for this narrative review. The five-stage workflow applied in this review, from a priori question framing through database and supplementary retrieval, eligibility appraisal, structured extraction with primary-source verification, and finally thematic synthesis into Tables 1 to 4 and Figures 2 to 5. Searching combined four Boolean concept blocks covering modality, receptor pharmacology, indication, and translation or manufacture, executed across PubMed/MEDLINE, Scopus, Web of Science, trial registries and regulatory repositories to June 2026, and was supplemented by backward and forward citation chasing. No meta-analysis was performed, because the four domains report incommensurable outcomes that no pooled estimate could meaningfully combine.

Figure 2. The mid-sized “Goldilocks” window between small molecules and biologics. Small molecules below roughly 500–1,000 Da offer oral access, membrane permeability and low cost of goods, but present contact surfaces of only 300–1,000 Ų, insufficient for the extended, shallow interfaces of class B G-protein-coupled receptors that typically demand 1,500–3,000 Ų. Monoclonal antibodies of about 150 kDa achieve exquisite selectivity and FcRn-recycled half-lives, but are restricted to extracellular targets, mammalian expression and parenteral administration. Peptides and macrocycles of 500–3,000 Da occupy the intermediate window, inheriting biologic-like selectivity and amino-acid catabolism while retaining synthetic tunability; their one serious inherited weakness, proteolytic lability, is the problem the chemistry of Figure 3 was developed to solve.

cost of binding (Bergamaschi et al., 2026; Liu et al., 2025). Hydrocarbon stapling, introduced through ruthenium-catalysed ring-closing metathesis across one or two turns of a helix, rigidifies the secondary structure and simultaneously improves proteolytic resistance and passive permeability (Walensky & Bird, 2014). Non-proteinogenic residues, D-amino acids, alpha,alpha- disubstituted amino acids and backbone N-methylation, disrupt the hydrogen-bonding networks and lower the polar surface area that otherwise prevent membrane crossing (Bergamaschi et al., 2026; Zhang et al., 2025). These strategies and their measured consequences are set out in Table 1.

It is worth resisting the impression that these are interchangeable options from a menu. They are not. Stapling suits helical epitopes and intracellular protein-protein interactions; macrocyclisation suits compact ligands where the bioactive conformation is known; N-methylation buys permeability at some risk to affinity. The art, and it does remain something of an art, lies in choosing which constraint a given scaffold can tolerate.

2.2 Protraction Chemistry: How a Two-Minute Peptide Became a Weekly Drug

If one had to identify the single chemical advance that made incretin pharmacotherapy commercially and clinically viable, it would be protraction. Native GLP-1 is cleaved by DPP-4 between alanine-8 and glutamate-9 within roughly two minutes of secretion, and even a protease-resistant analogue is filtered at the glomerulus unless something enlarges it (Knudsen & Lau, 2019; Lau et al., 2015). Two problems, then, requiring two distinct solutions (Figure 3).

The first was addressed by substitution at position 2 of the mature sequence. Replacing the endogenous alanine with alpha-aminoisobutyric acid (Aib) or with glycine introduces steric bulk that DPP-4 cannot accommodate, and does so without materially disturbing the receptor-binding geometry, since the modification sits at the periphery of the pharmacophore (Lau et al., 2015; Østergaard, 2025). Semaglutide carries Aib8; tirzepatide carries Aib at both positions 8 and 19; albiglutide used glycine (Coskun et al., 2018; Lau et al., 2015). The elegance of the solution is that it costs almost nothing in potency (Table 1).

The second problem required something more inventive. Human serum albumin is the most abundant protein in plasma and has an endogenous half-life of roughly 19 days, sustained by the same FcRn recycling pathway that protects immunoglobulins (Zhang et al., 2025). A peptide that binds albumin reversibly acquires, in effect, a circulating reservoir: the bound fraction is shielded from proteases, exceeds the renal filtration threshold by virtue of the complex's hydrodynamic radius, and releases free drug gradually as equilibrium dictates (Lau et al., 2015; Østergaard, 2025). Liraglutide demonstrated the principle with a single C16 palmitic acid attached to lysine-26 through a gamma-glutamyl spacer, achieving a half-life of about 13 hours and once-daily dosing (Knudsen & Lau, 2019).

The second generation refined the chemistry considerably. Replacing the monoacid with a dicarboxylic fatty acid, and interposing hydrophilic oligoethylene glycol (OEG) spacers between the peptide and the lipid, produced a substantially stronger and more specific albumin interaction: the distal carboxylate engages a defined pocket on the albumin surface through combined electrostatic and hydrophobic contacts (Lau et al., 2015; Østergaard, 2025). Semaglutide employs a C18 diacid at lysine-26; tirzepatide and retatrutide employ C20 eicosanedioic acid, at lysine-20 and lysine-17 respectively (Coskun et al., 2018; Jastreboff et al., 2023; Østergaard, 2025). Human half-lives of five to seven days followed, and with them once-weekly subcutaneous administration (Table 1; Table 2).

One consequence of protraction is easy to overlook. A long half-life does not only reduce injection frequency; it flattens the plasma concentration profile, and since the gastrointestinal adverse effects of GLP-1R agonism track peak exposure more closely than average exposure, the smoother curve is itself part of the tolerability story (Østergaard, 2025). This is one of several places where a pharmacokinetic decision turns out to have been, quietly, a pharmacodynamic one.

2.3 Unimolecular Poly-Pharmacology and the Question of Biased SignallingSelective GLP-1R agonism established the modern benchmark. Semaglutide at 2.4 mg weekly produced mean weight loss of approximately 15% at 68 weeks in STEP 1, alongside glycaemic control and reductions in cardiovascular risk (Wilding et al., 2021; Davies et al., 2021). Impressive as that was, a ceiling became apparent: further dose escalation was limited less by receptor biology than by nausea, vomiting and early satiety severe

Table 1. Chemical Modification Strategies for Structural Fortification and Pharmacokinetic Optimisation of Therapeutic Peptides and Macrocycles. Five recurring chemical strategies are compared across the mechanism by which each modifies the peptide, its conformational and biophysical consequence, its effect on metabolic stability and systemic exposure, and the marketed or investigational agents in which it appears. The first two rows describe the protraction chemistry that dominates the injectable incretin agents, converting a plasma half-life of under two minutes into one of five to seven days; the remaining rows describe the conformational constraints that dominate the exploratory macrocycle literature, where passive membrane permeability rather than circulating half-life is the limiting problem. Strategies are complementary rather than interchangeable, and the choice among them is dictated by the scaffold, the target compartment and the intended route of administration.

Table 2. Comparative Pharmacology, Protractor Chemistry and Clinical Efficacy of Incretin and Non-Incretin Multi-Receptor Agonists in Obesity and Type 2 Diabetes. Agents are ordered by receptor coverage rather than chronology, from selective GLP-1 receptor agonism through dual and triple incretin agonism to non-incretin amylin co-agonism and to ultra-long-acting and oral modalities. For each agent the table records receptor targets and signalling mode, backbone and protractor chemistry, highest development stage and indication, and the principal efficacy endpoint with its dose and timepoint. Efficacy figures derive from separate trials with differing populations, titration schedules and handling of discontinuation, and only a minority are head-to-head comparisons; the gradient they describe should therefore be read as indicative rather than as direct evidence of superiority.

enough to cause discontinuation (Müller et al., 2022; Østergaard, 2025). If more efficacy was to be extracted, it would have to come from somewhere other than more GLP-1R occupancy.

The answer the field arrived at, drawing on earlier proof-of-concept work in rodents and primates, was unimolecular poly-pharmacology: a single peptide backbone engineered to activate several nutrient-sensing receptors simultaneously, rather than a combination of separate drugs (Finan et al., 2013; Finan et al., 2015). The pharmacokinetic argument for a single molecule is straightforward, since two agonists with different half-lives cannot maintain a fixed ratio of receptor engagement over a weekly dosing interval. The pharmacological argument is subtler and concerns synergy (Figure 4).

Tirzepatide was the first such agent to reach approval. It is a 39-amino-acid synthetic peptide built on the GIP sequence, carrying Aib at positions 8 and 19, a C-terminal amide, and a C20 diacid at lysine-20 through a gamma-Glu-2xOEG linker (Coskun et al., 2018; Sun et al., 2022). Its pharmacology is deliberately asymmetric. Tirzepatide binds GIPR with affinity comparable to native GIP but engages GLP-1R with markedly lower potency than native GLP-1, and it recruits beta-arrestin poorly, which limits receptor desensitisation and internalisation (Willard et al., 2020; Sun et al., 2022). Whether this imbalance is the reason for its efficacy, or merely a property it happens to possess, has not been settled; the honest position is that the structure-activity story is more suggestive than proven. Clinically, weight loss of 20.9 to 22.5% at 72 weeks in SURMOUNT-1 comfortably exceeded GLP-1 mono-agonist benchmarks (Jastreboff et al., 2022; Table 2). Mechanistic work indicates that GIPR engagement acts on adipocyte nutrient handling and lipid storage, operating in parallel with rather than downstream of central GLP-1R satiety signalling (Regmi et al., 2024).

A second dual strategy pairs GLP-1R with the glucagon receptor. Survodutide, mazdutide and pemvidutide combine GLP-1-mediated appetite suppression with glucagon-driven lipolysis, hepatic fatty-acid oxidation and increased energy expenditure (le Roux et al., 2024; Østergaard, 2025). The design tension here is obvious, since glucagon is a counter-regulatory, diabetogenic hormone, and the resolution is that concurrent GLP-1R activation offsets the glycaemic penalty while the hepatic benefit is retained. Survodutide produced 19.0% weight loss at 46 weeks with meaningful reductions in liver fat and improvement in MASH fibrosis endpoints (le Roux et al., 2024; Table 2).

Triple agonism extends the logic once more. Retatrutide (LY3437943) engages GLP-1R, GIPR and GcgR from a single backbone lipidated at lysine-17 with a C20 diacid, and structural analysis indicates that the peptide can accommodate the binding pockets of all three class B receptors (Jastreboff et al., 2023; Østergaard, 2025). In phase 2 it achieved mean weight loss of 24.2% at 48 weeks at the 12 mg dose, with liver fat reduction exceeding 80% in participants with MASLD (Jastreboff et al., 2023; Sanyal et al., 2024). Notably, the dose-response curve had not flattened at 48 weeks, which raises the question of what the true ceiling is and whether we would want to reach it.

Non-incretin co-agonism has advanced in parallel and may prove equally important. Cagrilintide, a long-acting amylin and calcitonin receptor agonist lipidated with a C20 diacid, produced 11.8% weight loss as monotherapy and 22.7 to 24.3% when co-formulated with semaglutide as CagriSema (Kruse et al., 2021; Davies et al., 2025). Zenagamtide, also known as amycretin, is a first-in-class single peptide activating GLP-1R alongside amylin receptors, and delivered 24.3% weight loss at 36 weeks in phase 1b/2a without reaching a plateau (Dahl et al., 2025; Gasiorek et al., 2025). Selective amylin receptor agonists such as eloralintide are following (Briere et al., 2025). The division of physiological labour across these receptor arms is summarised in Figure 4 and tabulated in Table 2.

Molecular architecture has also been pushed in a different direction: not more receptors, but longer duration. Maridebart cafraglutide (AMG133) conjugates a GIPR-antagonist monoclonal antibody to two GLP-1R agonist peptides, exploiting FcRn recycling to reach a half-life of approximately 21 days and once-monthly dosing, with 16.2% weight loss at 52 weeks (Véniant et al., 2024; Jastreboff et al., 2025; Ye et al., 2025). That a GIPR antagonist and a GIPR agonist both produce weight loss is one of the more genuinely puzzling observations in the field, and no explanation currently commands consensus.

2.4 Oral Delivery: Three Partial Solutions to One Hard ProblemPatients on chronic therapy prefer tablets, and adherence data generally bear that preference out. The gastrointestinal tract, however, is an environment

Figure 3. Sequential chemical fortification of the GLP-1 backbone and the albumin depot mechanism. Panel A traces the two-step transformation of a fragile hormone into a weekly medicine: native GLP-1 is cleaved by dipeptidyl peptidase-4 between Ala8 and Glu9 and cleared renally within minutes; substitution of position 2 with α-aminoisobutyric acid denies the protease its recognition motif at negligible cost to potency; and acylation of Lys26 with a C18 or C20 fatty diacid through γ-Glu and oligoethylene glycol spacers extends half-life to five to seven days. Panel B shows the resulting equilibrium, in which reversibly albumin-bound drug forms a protected intravascular reservoir that releases free peptide to engage class B receptors, flattening the plasma profile in a way that matters for tolerability as well as dosing interval. Panel C summarises the orthogonal conformational constraints applied where permeability rather than protraction is limiting.

Figure 4. Receptor poly-pharmacology: escalation, division of physiological labour and clinical efficacy. Panel A traces the progression from selective GLP-1 receptor agonism through dual GLP-1R/GIPR agonism to unimolecular triple agonism and to non-incretin GLP-1R/amylin co-agonism, with the protractor chemistry and headline efficacy of a representative agent in each class. Panel B assigns the principal physiological contribution of each receptor arm, showing that the receptors are complementary rather than redundant: satiety and insulinotropy, adipocyte lipid handling, hepatic oxidation and thermogenesis, and postprandial satiation respectively. Panel C ranks peak reported weight loss across the class. Note that amycretin matches the tri-agonist with two receptor arms and a shorter exposure, indicating that which physiological axes are engaged matters more than how many receptors are listed. Values derive from separate trials and are indicative rather than head-to-head.

Figure 5. Gastrointestinal barriers to oral peptide absorption and the three strategies deployed against them. Panel A sets out the sequential obstacles that hold baseline oral bioavailability for an unprotected therapeutic peptide below 1%: acid-mediated unfolding, luminal and brush-border proteolysis, the adhesive mucus layer, and epithelial tight junctions whose paracellular pores exclude peptide-sized solutes. Panel B compares the three routes past that barrier, with reported bioavailability, dosing constraints and developmental status for each. The comparison exposes a genuine trilemma rather than a ranking: chemical permeation enhancement is approved and scalable yet wastes over 99% of the administered peptide; mechanical micro-devices deliver the dose efficiently but make every tablet a precision-engineered machine; and non-peptide agonists solve delivery convincingly while forfeiting the multi-receptor breadth that produces the largest weight loss.

specifically adapted to dismantling peptides: gastric acid unfolds them, pepsin and pancreatic and brush-border proteases cleave them, mucus impedes diffusion, and epithelial tight junctions exclude anything much larger than a small molecule (Liu et al., 2025; Xiao et al., 2026). Baseline oral bioavailability for an unprotected therapeutic peptide sits below 1% (Figure 5).

The first route past this barrier is chemical. Oral semaglutide (Rybelsus) co-formulates the peptide with sodium N-(8-[2-hydroxybenzoyl]amino)caprylate (SNAC), which dissolves in the stomach, raises local pH sufficiently to protect the peptide from pepsin, and forms transient non-covalent complexes that fluidise the gastric membrane and promote transcellular absorption in a localised region of the gastric mucosa (Buckley et al., 2018). The approach works, in the sense that it produced an approved oral GLP-1R agonist, but the economics are uncomfortable: bioavailability of 0.4 to 1.0% means the overwhelming majority of a milligram-scale dose is never absorbed, and administration requires a fasting window of 30 minutes before food or drink (Buckley et al., 2018; Østergaard, 2025). Related transient permeation enhancer systems using medium-chain fatty acids and bile salts underpin oral octreotide and experimental bis-lipidated GLP-1 analogues (Zhang et al., 2025; Table 3).

The second route is mechanical, and frankly ingenious. Self-orienting millimetre-scale applicators (SOMA) are ingestible capsules weighted so that they reliably right themselves against the gastric wall, then fire a micro-needle of compacted drug into the submucosa using a spring released by gastric fluid; the liquid-injecting variant (L-SOMA) delivers a solution instead (Abramson et al., 2019; Abramson et al., 2022). Reported systemic bioavailabilities are startling by comparison with chemical enhancement, exceeding 50% for solid insulin and reaching 103% plus or minus 42% for GLP-1 analogues in swine. Dynamic omnidirectional adhesive microneedle systems and balloon-actuated capsules such as RaniPill apply similar logic in the intestine, roughly doubling plasma exposure relative to standard tablets (Chen et al., 2022; Table 3). Receptor-mediated approaches, including conjugation to bile acid derivatives that exploit the apical sodium-dependent bile acid transporter, represent a third and more biological variation on the same theme (Zeng et al., 2025).

The third route abandons the peptide altogether. Rational design has produced non-peptide small molecules that bind an allosteric or transmembrane pocket on GLP-1R and activate it without any peptide bond to defend. Danuglipron and lotiglipron established proof of concept but were curtailed by short half-life and hepatic signals respectively (Østergaard, 2025). Orforglipron, a high-potency partial agonist, achieves oral bioavailability of 21 to 28%, supports once-daily dosing without food or water restriction, and produced 12.4 to 14.7% weight loss in phase 2 and 3 trials (Wharton et al., 2023; Table 3). That figure sits below the multi-agonist peptides, which is precisely the trade-off: convenience is bought at some cost in efficacy, at least for now. The oral small-molecule approach is being extended to other indications as well, including oral peptide-mimetic agents in immunology (Habib et al., 2026).

2.5 Manufacturing at Population Scale and the Environmental Reckoning

A drug that several hundred million people might reasonably take is a manufacturing problem before it is anything else, and this is where the peptide field's success has generated its most uncomfortable literature (Østergaard, 2025). Therapeutic peptides are produced by solid-phase synthesis (SPPS), liquid-phase synthesis (LPPS), recombinant expression, or hybrids of these. Recombinant expression in yeast or Escherichia coli is scalable and comparatively clean, but non-proteinogenic residues, site-specific lipidation and macrocyclic architectures generally cannot be installed biologically, which forces a synthetic or semi-recombinant route for exactly the molecules of interest here (Østergaard, 2025; Table 4).

Fmoc-based SPPS remains the workhorse. Its stepwise resin-bound assembly is flexible and reliable, and its atom economy is poor: large excesses of protected amino acids and coupling reagents, extensive washing between cycles, and cleavage with trifluoroacetic acid (Al Musaimi et al., 2020; Ferrazzano et al., 2022). The field quantifies the consequence using process mass intensity (PMI), the total mass of raw materials, reagents and solvents consumed per kilogram of active pharmaceutical ingredient. For phase 1 synthetic peptides, PMI averages approximately 33,000, meaning 33 tonnes of input per kilogram of product, of which more than 80% is hazardous organic solvent: dimethylformamide, N-methyl-2-pyrrolidone and trifluoroacetic acid (Kekessie et al., 2024). Small molecules sit between 168 and 308; monoclonal antibodies average around 7,700, and most of that is water (Kekessie et al., 2024). Extrapolating to multi-tonne annual demand produces solvent figures that are difficult to state without sounding alarmist (Table 4).

Regulatory pressure has converged with this arithmetic. European restrictions on reprotoxic DMF and on per- and polyfluoroalkyl substances, a category that encompasses TFA, mean that the current process is not merely wasteful but increasingly unlawful (Østergaard, 2025; Ashley et al., 2025). Three responses have gained traction. Solvent replacement substitutes bio-derived alternatives such as N-butylpyrrolidone, gamma-valerolactone, Cyrene and triethyl phosphate, generally coupled with closed-loop recovery (Al Musaimi et al., 2020; Ferrazzano et al., 2022). Convergent hybrid SPPS/LPPS assembles short fragments separately and couples them in solution, a strategy used for kilogram-scale GMP manufacture of tirzepatide from four fragments and reported to lower commercial PMI per amino acid from roughly 1,463 to 874 (Frederick et al., 2021; Østergaard, 2025). Chemo-enzymatic peptide synthesis (CEPS) goes further, ligating fragments in aqueous media using engineered ligases such as omniligase-1, eliminating organic solvent from the condensation step entirely and demonstrated at multi-gram to kilogram scale for exenatide and liraglutide (Toplak et al., 2019; Zhang et al., 2025). Table 4 compares these routes on process characteristics, waste metrics and scale-up readiness.

3. Methods

3.1 Design and Reporting Framework

This work is a narrative, integrative review rather than a systematic review or meta-analysis, and it is worth being explicit about why. The four questions it addresses span synthetic organic chemistry, receptor structural biology, randomised clinical trials and industrial process engineering. These literatures report incommensurable outcomes, and no pooled effect estimate could meaningfully combine a ligation yield, a beta-arrestin recruitment assay and a 72-week body-weight endpoint. What can be done, and what was attempted here, is to apply systematic search and screening discipline to the retrieval of evidence while synthesising it narratively.. The full workflow is shown in Figure 1. No protocol was registered, which is a limitation we state plainly rather than defend.

3.2 Information Sources and Search Execution

Searches were run in PubMed/MEDLINE, Scopus and Web of Science Core Collection, with the last search executed in June 2026. Three additional sources were interrogated for information that the bibliographic databases index poorly: ClinicalTrials.gov and the EU Clinical Trials Register for trial status, enrolment and endpoint definitions; the FDA Drugs@FDA and EMA EPAR repositories for approval dates, approved indications and labelled pharmacokinetic parameters; and manufacturer pipeline disclosures for agents whose phase 2 and 3 data had been presented but not yet published in full. Where a value appeared only in a conference presentation or a corporate release, it was flagged as such and was not used to support a substantive claim.

The search combined four concept blocks with Boolean AND, each block internally combined with OR. Block 1 (modality): "therapeutic peptide", "macrocyclic peptide", "stapled peptide", "peptidomimetic", "lipidation", "acylation", "N-methylation", "cyclization". Block 2 (target and pharmacology): "GLP-1", "glucagon-like peptide-1 receptor", "GIP", "glucose-dependent insulinotropic polypeptide receptor", "glucagon receptor", "amylin", "calcitonin receptor", "co-agonist", "dual agonist", "triple agonist", "unimolecular". Block 3 (indication): "obesity", "overweight", "type 2 diabetes", "MASLD", "MASH", "NAFLD", "weight loss". Block 4 (translation and manufacture): "oral delivery", "permeation enhancer", "SNAC", "ingestible device", "solid-phase peptide synthesis", "process mass intensity", "green chemistry". MeSH terms were exploded where available in PubMed and mapped to equivalent Scopus and Web of Science index terms; free-text terms were searched in title, abstract and keyword fields. No filter was applied to study design. Language was restricted to English, and the date window ran from 1990 to June 2026, the earlier boundary chosen to capture the foundational work on GLP-1 physiology and on Fmoc-SPPS that precedes the therapeutic literature proper.

Database searching was supplemented by backward and forward citation chasing. Reference lists of all included pivotal trial reports and of recent expert reviews were screened manually (Østergaard, 2025; Zhang et al., 2025; Liu et al., 2025; Müller et al., 2022), and forward citations of the principal discovery-chemistry papers were examined through Scopus (Lau et al., 2015; Coskun et al., 2018; Walensky & Bird, 2014). This step recovered several process-chemistry and device papers that the concept-block search had missed, which is a familiar weakness of keyword strategies applied to engineering literature (Figure 1).

3.3 Eligibility Criteria

Records were eligible if they reported primary or authoritative secondary data in at least one of four domains. First, peptide and macrocycle design chemistry: synthesis, conformational constraint, protraction strategy, or structure-activity relationships bearing on proteolytic stability, permeability or receptor engagement, which supply the content of Table 1. Second, receptor pharmacology and clinical efficacy: in vitro potency and signalling bias, structural characterisation of ligand-receptor complexes, and phase 1 to phase 3 trial outcomes for the agents compiled in Table 2. Third, delivery: formulation, permeation enhancement, device engineering, or non-peptide receptor agonist chemistry with reported bioavailability or clinical outcome data, as tabulated in Table 3. Fourth, manufacture and sustainability: synthetic route description, PMI or E-factor quantification, solvent substitution, enzymatic ligation, or regulatory analysis relevant to scale-up, as compared in Table 4.

Records were excluded if they addressed peptide therapeutics in indications with no transferable chemistry or pharmacology; if they were conference abstracts, preprints or corporate communications not subsequently peer-reviewed, except where used only for pipeline status; if they duplicated a trial cohort already represented by its primary publication, in which case the primary report was retained and later analyses cited only for additional endpoints; or if full text could not be obtained. Non-English records were excluded, and we acknowledge this may have under-represented process-chemistry work published in Chinese and Japanese journals.

3.4 Screening and Selection

Retrieved records were deduplicated on DOI and on title-plus-first-author string matching. Titles and abstracts were then screened against the criteria above, and records surviving that pass were read in full. Screening was conducted by the authors working independently on overlapping subsets, with disagreements about inclusion resolved by discussion against the written criteria rather than by majority; where a record sat genuinely at the boundary, it was included, on the reasoning that a narrative synthesis is more damaged by omission than by breadth. Selection decisions and their stage are represented in Figure 1.

3.5 Data Extraction and Verification

A structured extraction template was applied to every included record. For molecules, the fields were: generic and development name; receptor targets and mode of action, including any reported signalling bias; backbone length and origin; non-proteinogenic substitutions and their positions; protractor chemistry, linker composition and attachment residue; reported human half-life; route and dosing interval; highest development phase and indication; and quantitative efficacy endpoints with their timepoints, comparators and doses. For delivery platforms: mechanism of absorption, payload demonstrated, species in which bioavailability was measured, absolute bioavailability, and clinical status. For manufacturing routes: synthetic strategy, solvent system, PMI or related waste metric with the phase to which it applies, demonstrated scale, and regulatory constraint. These fields populate Tables 1 through 4 directly, so that each tabulated value is traceable to a named source.

Two verification rules were applied. Every quantitative value was traced to the primary report in which it was first published; where a review and its source disagreed, the primary publication prevailed, and several discrepancies of this kind were in fact encountered, most often where rounded figures had propagated through secondary literature. Second, where a single agent had been reported at several doses or timepoints, the dose and timepoint are stated alongside the figure rather than presented as a bare maximum, since peak percentages quoted without context are the most common source of misleading comparison in this field (Table 2).

3.6 Synthesis Approach and Assessment of Certainty

Extracted data were grouped thematically into the four domains, and within each domain organised along a development axis, from mechanism to preclinical demonstration to clinical outcome to industrial feasibility. Cross-domain integration was then performed deliberately, by asking of each chemical strategy what it enabled pharmacologically and what it cost in manufacture, which is the reasoning represented in Figures 2 through 5. No quantitative pooling was performed. Clinical comparisons between agents are drawn from separate trials with differing populations, run-in periods, titration schedules and handling of treatment discontinuation, and should be

Table 3. Oral Delivery Platforms, Permeation Strategies and Device Technologies for Peptide Therapeutics, with Reported Bioavailability and Clinical Status. Platforms are grouped by the level at which each attacks the gastrointestinal barrier: chemical modification of the local absorptive environment, mechanical delivery of the payload past the epithelium, and abandonment of the peptide scaffold in favour of a non-peptide agonist. For each, the table records the absorption mechanism, the payload in which it has been demonstrated, the reported absolute bioavailability with the species in which it was measured, and the current regulatory or developmental status. Bioavailability figures are not directly comparable, since several derive from large-animal models rather than from human pharmacokinetic studies, and the wide dispersion around some device estimates reflects genuine variability in deployment rather than analytical imprecision.

Table 4. Manufacturing Routes, Process Mass Intensity and Green Chemistry Responses in Industrial Peptide Production. Four manufacturing routes and one cross-cutting set of green-chemistry innovations are compared on process characteristics, quantified waste burden and demonstrated scale. Process mass intensity (PMI) is the total mass of raw materials, reagents and solvents consumed per kilogram of active pharmaceutical ingredient, and is the field's standard sustainability metric. The figures make the central industrial tension visible: the chemical features that confer the best clinical pharmacology, non-proteinogenic residues and site-specific lipidation, are precisely those that obstruct the cleanest route, recombinant expression, leaving solvent-intensive synthesis as the default for exactly the molecules in greatest demand.

read as indicative rather than as head-to-head evidence; the only rigorous comparisons available are the small number of active-comparator trials, and these are identified as such where cited. Formal risk-of-bias instruments were not applied across the full corpus, since they are ill-suited to synthetic chemistry and process engineering reports; instead, the strength of each substantive claim is signalled in the text by the evidence supporting it, distinguishing approved-label data, published randomised trial results, early-phase or preclinical findings, and mechanistic inference.

4. Synthesis of Findings: From Molecular Fortification to Industrial Feasibility

4.1 A Convergent Chemical Toolkit

Read across the included literature, the chemistry shows a striking degree of convergence. Independent programmes, pursuing different receptors in different companies, arrived at substantially the same small set of modifications, which suggests these are not stylistic preferences but solutions to constraints that any peptide therapeutic must satisfy (Table 1).

Five strategies recur. Position-2 substitution with Aib or glycine abolishes DPP-4 cleavage between residues 8 and 9 while preserving class B GPCR binding, and appears in semaglutide, tirzepatide and albiglutide alike (Lau et al., 2015; Coskun et al., 2018). Fatty diacid acylation through gamma-Glu-OEG spacers creates the reversible albumin depot that carries half-life from minutes to five to seven days, evolving from the C16 monoacid of liraglutide to the C18 diacid of semaglutide and the C20 diacids of tirzepatide and retatrutide (Knudsen & Lau, 2019; Lau et al., 2015; Østergaard, 2025). Macrocyclisation removes free termini and pre-organises the bioactive conformation, lowering the entropic penalty of binding (Bergamaschi et al., 2026). Hydrocarbon stapling across i,i+4 or i,i+7 helical turns rigidifies secondary structure, masks backbone amides and improves passive permeability (Walensky & Bird, 2014). Backbone N-methylation reduces polar surface area and hydrogen-bond donor count, which is the principal chemical lever on oral absorption for a peptide that must cross a membrane unaided (Bergamaschi et al., 2026; Zhang et al., 2025). The sequence by which these are applied, and their cumulative pharmacokinetic effect, is set out in Figure 3.

One observation deserves emphasis because it is easy to miss when the strategies are listed as a catalogue. The first two dominate the marketed incretin agents and the last three dominate the exploratory macrocycle literature. That division is not accidental: the injectable multi-agonists solved their problem through protraction and had no need of permeability, whereas the macrocycle field is pursuing oral and intracellular targets for which permeability is the whole difficulty. The toolkit is therefore not one toolkit but two, applied to different objectives, and conflating them overstates how much transfer has actually occurred (Table 1; Table 3).

4.2 The Efficacy Gradient Across Receptor Combinations

Arranged by receptor coverage rather than by chronology, the clinical data describe a reasonably orderly gradient, though with two informative exceptions (Table 2; Figure 4).

Selective GLP-1R agonism sets the baseline. Semaglutide at 2.4 mg weekly achieved 15.0% mean weight loss at 68 weeks in STEP 1, with high-dose oral formulations reaching up to 16.6% (Wilding et al., 2021; Østergaard, 2025). Dual GLP-1R/GIPR agonism with tirzepatide raised this to 20.9 to 22.5% at 72 weeks in SURMOUNT-1, with superior HbA1c reduction against GLP-1-selective comparators (Jastreboff et al., 2022; Coskun et al., 2018). Dual GLP-1R/GcgR agonism with survodutide produced 19.0% at 46 weeks, a somewhat lower figure for body weight but accompanied by hepatic fat and fibrosis benefits that the GIPR arm does not deliver (le Roux et al., 2024). Triple agonism with retatrutide reached 24.2% at 48 weeks at 12 mg, with liver fat reduction exceeding 80% in MASLD (Jastreboff et al., 2023; Sanyal et al., 2024).

The first informative exception is non-incretin. Amycretin, activating GLP-1R together with amylin receptors, matched the tri-agonist at 24.3% by 36 weeks, and did so with two rather than three receptor arms and over a shorter exposure (Dahl et al., 2025; Gasiorek et al., 2025). CagriSema, a co-formulation rather than a single molecule, reached 22.7 to 24.3% (Davies et al., 2025; Kruse et al., 2021). Receptor count, in other words, is a poor predictor of efficacy; which receptors, and how hard each is pushed, evidently matters more (Figure 4).

The second exception concerns duration rather than magnitude. Maridebart cafraglutide achieved 16.2% at 52 weeks, below the multi-agonist peptides, but on once-monthly dosing and with weight maintained after treatment cessation in the reported follow-up (Jastreboff et al., 2025; Véniant et al., 2024). Orforglipron similarly trails on efficacy at 12.4 to 14.7%, while offering oral administration without food or water restrictions (Wharton et al., 2023). Both illustrate that the efficacy column of Table 2 captures only one dimension of therapeutic value.

A caution is necessary here, and it applies to the whole of this section. These figures come from separate trials with different populations, titration schedules, run-in periods and statistical handling of discontinuation. Only a minority are head-to-head. The gradient is real enough to be taken seriously and too loosely constructed to be taken literally.

4.3 What Each Receptor Arm Appears to Contribute

Disentangling the contribution of individual receptors from combination outcomes is difficult, but the preclinical and mechanistic literature supports a provisional division of labour (Figure 4). GLP-1R activation drives central satiety signalling, slows gastric emptying and potentiates glucose-dependent insulin secretion; it also accounts for most of the nausea (Wilding et al., 2021; Müller et al., 2022). GIPR engagement acts substantially on adipocytes, regulating nutrient partitioning and lipid storage, and long-acting GIPR agonism appears to operate in parallel with rather than downstream of GLP-1R satiety pathways (Regmi et al., 2024; Coskun et al., 2018). There is a further suggestion, not yet firmly established, that GIPR agonism attenuates GLP-1-associated emesis, which would help explain why tirzepatide tolerates higher effective exposure. GcgR activation contributes on the expenditure side, stimulating hepatic mitochondrial fatty-acid oxidation and thermogenesis, and is the most plausible source of the disproportionate liver-fat reductions seen with retatrutide and survodutide (Jastreboff et al., 2023; Sanyal et al., 2024; le Roux et al., 2024). Amylin and calcitonin receptor signalling adds postprandial satiation and glucagon suppression through a mechanism largely independent of the incretin axis, which is the most persuasive explanation for why amycretin and CagriSema achieve incretin-comparable results by a different route (Kruse et al., 2021; Dahl et al., 2025).

The GIPR paradox remains unresolved and should be stated rather than smoothed over. Tirzepatide agonises GIPR and produces substantial weight loss; maridebart cafraglutide antagonises it and also produces substantial weight loss (Coskun et al., 2018; Véniant et al., 2024). Proposed reconciliations invoke sustained agonism causing functional receptor desensitisation, or tissue-specific differences in receptor reserve, but these remain hypotheses. It is a useful reminder that clinical efficacy has, in this field, repeatedly outrun mechanistic understanding.

4.4 Delivery Platforms Compared on Their Own Terms

The three approaches to oral delivery are best assessed against different criteria, because they are solving the problem at different levels (Table 3; Figure 5). Chemical permeation enhancement is the only approved route. SNAC co-formulation delivers oral semaglutide with bioavailability of 0.4 to 1.0%, requiring milligram dosing and a strict fasting window (Buckley et al., 2018; Østergaard, 2025). Judged as pharmacology this is a success; judged as process economics it is expensive, since over 99% of a synthesised API is excreted unabsorbed, which interacts unhelpfully with the manufacturing constraints of Table 4. Related transient permeation enhancer formulations support oral octreotide and experimental bis-lipidated analogues at comparable or slightly better efficiency (Zhang et al., 2025).

Mechanical micro-devices post far better numbers: above 50% for solid SOMA delivering insulin, and 103% plus or minus 42% for L-SOMA delivering GLP-1 analogues in swine, with DOAMS approximately doubling plasma AUC against standard tablets and RaniPill exceeding 50% in clinical safety studies (Abramson et al., 2019; Abramson et al., 2022; Chen et al., 2022). The wide confidence interval on the L-SOMA figure is itself informative, and most of these data come from large-animal models. What devices gain in efficiency they give back in complexity: each dose is a small precision-engineered machine, with manufacturing, cost and safety implications that chemical formulation does not carry.

Non-peptide agonists sidestep the problem rather than solving it. Orforglipron achieves 21 to 28% oral bioavailability with no fasting requirement, at the cost of 12.4 to 14.7% weight loss, and the attrition of danuglipron and lotiglipron for half-life and hepatic safety respectively indicates that this chemical space carries its own liabilities, precisely the small-molecule liabilities that peptides avoid (Wharton et al., 2023; Østergaard, 2025). Bile-acid transporter targeting occupies an intermediate position and remains preclinical (Zeng et al., 2025).

4.5 The Manufacturing Constraint

The industrial data assembled in Table 4 point to a conclusion the clinical literature rarely confronts: at current efficacy and current demand, the binding constraint on population-level impact may be synthetic rather than pharmacological.

Fmoc-SPPS carries a phase 1 PMI of approximately 33,000, of which more than 80% is hazardous organic solvent, against 168 to 308 for small molecules and about 7,700, largely aqueous, for monoclonal antibodies (Kekessie et al., 2024). Commercial-scale optimisation improves matters considerably, with hybrid SPPS/LPPS fragment condensation reducing PMI per amino acid from roughly 1,463 to 874 and enabling kilogram-scale GMP manufacture of tirzepatide from four fragments (Frederick et al., 2021; Østergaard, 2025). Chemo-enzymatic ligation in aqueous media removes organic solvent from the condensation step entirely and has been demonstrated at multi-gram to kilogram scale for exenatide and liraglutide without epimerisation (Toplak et al., 2019). Semi-recombinant production, in which a canonical backbone is expressed in yeast and then chemically acylated, achieves a PMI near 7,700, comparable to biologics, and underpins semaglutide manufacture at global scale (Østergaard, 2025).

Green SPPS innovations, replacing DMF and NMP with N-butylpyrrolidone, gamma-valerolactone, Cyrene or triethyl phosphate and combining this with microwave-assisted coupling and continuous flow, report 30 to 60% energy reduction and 50 to 80% solvent volume reduction (Al Musaimi et al., 2020; Ferrazzano et al., 2022; Ashley et al., 2025). These are necessary rather than optional developments, given European restrictions on reprotoxic DMF and on PFAS including TFA (Østergaard, 2025). The uncomfortable arithmetic is that the molecules with the best clinical profiles, bearing non-proteinogenic residues and site-specific lipidation, are precisely those least amenable to the cleanest route, recombinant expression (Table 1; Table 4).

5. An Efficacy Revolution and Its Unfinished Business

5.1 What Has Actually Been Achieved

It is worth stating the achievement plainly before qualifying it. A class of molecules once dismissed as pharmacokinetically hopeless now produces weight loss within a few percentage points of bariatric surgery, by subcutaneous injection once a week, with cardiovascular and renal benefit demonstrated in outcome trials (Wilding et al., 2021; Jastreboff et al., 2023; Dahl et al., 2025). Twenty years ago that would have been a reasonable definition of an implausible result. The route from there to here ran through chemistry rather than through biology: the receptors were known, the hormones were known, and what was missing was a way to make a peptide last (Table 1; Figure 3).

This suggests a general lesson that extends beyond metabolic disease. The "undruggable" designation attached to protein-protein interactions and class B GPCRs described the limitations of the available chemical toolkit rather than any intrinsic property of the targets. Once the mid-sized window between small molecules and biologics became synthetically and pharmacokinetically accessible, targets that had resisted for decades became tractable (Figure 2). Whether the same reasoning transfers cleanly to intracellular protein-protein interactions, where permeability rather than protraction is the obstacle, is a different and harder question.

5.2 Poly-Pharmacology and the Limits of Adding Receptors

The escalation from mono- to dual to triple agonism produced real gains, but the pattern of those gains argues against extrapolating it further (Table 2; Figure 4). Moving from one receptor to two added roughly seven percentage points of weight loss; moving from two to three added around two more (Wilding et al., 2021; Jastreboff et al., 2022; Jastreboff et al., 2023). That amycretin reached the same territory with GLP-1R and amylin receptors alone, over a shorter exposure, strongly suggests that the relevant variable is which physiological axes are engaged rather than how many receptors are listed (Dahl et al., 2025; Gasiorek et al., 2025).

There is a further reason for caution. As receptor coverage broadens, off-target physiology broadens with it. GcgR activation raises hepatic glucose output, which GLP-1R activation must offset; sustained amylin receptor agonism affects gastric motility; and the cumulative effect of engaging four or five nutrient-sensing pathways at once has not been characterised over the multi-year horizons on which these drugs will actually be used. A quadruple agonist is chemically feasible. Whether it would be clinically sensible is a separate matter, and the honest answer is that nobody yet knows.

The most consequential unresolved question is the one that receptor addition does not address. Gastrointestinal adverse effects, not receptor occupancy, constrain dose escalation across the entire class (Müller et al., 2022; Østergaard, 2025). An agent that produced current efficacy with materially less nausea would expand the treatable population more than an agent that produced two further percentage points of weight loss in those who can already tolerate therapy. Tolerability, not potency, is where the remaining headroom lies.

5.3 Body Composition, Durability and What the Percentages Conceal

Percentage weight loss has become the field's headline metric, and it is an imperfect one. It does not distinguish fat mass from lean mass, and at the magnitudes now being achieved the lean mass component is not negligible, particularly in older patients for whom sarcopenia carries its own mortality risk. Combination with amylin analogues, and interest in myostatin and activin pathway modulation alongside incretin therapy, both reflect a field beginning to take this seriously, though the clinical data remain immature (Kruse et al., 2021; Davies et al., 2025).

Durability is the second concealed variable. If obesity is a chronic disease defended by a regulatory set-point, then treatment is indefinite, and weight regain on discontinuation is the expected rather than the anomalous result. This makes the post-treatment maintenance reported for maridebart cafraglutide interesting out of proportion to its efficacy figure (Jastreboff et al., 2025; Véniant et al., 2024). Whether that observation reflects the very long half-life, something about GIPR antagonism, or the particular follow-up design, cannot be determined from the available data (Table 2).

5.4 Delivery as the Access Question

The oral delivery problem is often framed as one of patient convenience. That framing understates it. Injectable therapy requires cold chain, device manufacture, disposal infrastructure and, in many settings, clinical administration, all of which constrain where a drug can realistically be deployed. A genuinely effective oral agent is an access intervention rather than a comfort feature (Table 3; Figure 5).

None of the three current routes fully delivers this. SNAC-based formulation works but wastes more than 99% of the API, which is difficult to reconcile with the manufacturing arithmetic of Table 4 (Buckley et al., 2018; Kekessie et al., 2024). Ingestible micro-devices achieve excellent bioavailability in large-animal models but add per-dose mechanical complexity that no chronic global therapy has previously carried (Abramson et al., 2019; Abramson et al., 2022; Chen et al., 2022). Non-peptide agonists solve delivery convincingly and forfeit the multi-receptor breadth that generates the largest effects, while reintroducing the small-molecule hepatic and off-target liabilities that peptides were valued for avoiding (Wharton et al., 2023; Østergaard, 2025).

If one had to guess, the most likely resolution is convergence rather than victory for any single approach: orally bioavailable macrocycles or peptidomimetics carrying enough conformational constraint and N-methylation to cross the epithelium unaided, while retaining multi-receptor engagement (Bergamaschi et al., 2026; Zhang et al., 2025). That is a chemistry problem of considerable difficulty and there is no guarantee it is solvable. It is, however, where the exploratory literature is pointing (Table 1).

5.5 Manufacturing, Sustainability and Equity

The manufacturing discussion is frequently relegated to an appendix in clinical reviews, and this seems to us a mistake. A PMI of 33,000 is not merely an environmental statistic; it is a statement about how many people can be treated (Kekessie et al., 2024; Table 4). Multi-tonne annual demand for a molecule made this way implies solvent handling on a scale that regulatory trajectories in Europe are actively closing off (Østergaard, 2025; Ashley et al., 2025).

The responses are genuine and measurable. Hybrid SPPS/LPPS approaches have taken tirzepatide to kilogram-scale GMP production with materially lower PMI per residue (Frederick et al., 2021). Chemo-enzymatic ligation removes organic solvent from fragment condensation (Toplak et al., 2019). Semi-recombinant routes bring peptide manufacture close to biologics in waste profile (Østergaard, 2025). Yet a structural tension persists, and it is worth naming: the chemical features that confer the best pharmacology, non-proteinogenic residues and site-specific lipidation, are the features that block the cleanest manufacturing route (Table 1; Table 4). Designing for manufacturability from the outset, rather than optimising the process after the molecule is fixed, would address this, and the field has been slow to adopt that discipline.

There is an equity dimension that follows directly. The populations with the fastest-growing obesity and T2DM prevalence are largely in low- and middle-income countries, where cold chain, injection infrastructure and per-patient cost are binding constraints. If cost of goods remains dominated by a solvent-intensive synthesis and delivery remains parenteral, these therapies will disproportionately benefit populations that are not where the burden is growing fastest (Xu et al., 2025). Process chemistry and oral delivery are therefore not peripheral technical concerns but the mechanisms by which clinical efficacy becomes public health benefit.

5.6 Limitations of This Study

Several limitations bear on how much weight the foregoing can carry. Comparisons between agents draw on separate trials with heterogeneous populations, titration schedules and analytical conventions, and only a minority are head-to-head (Table 2). Several of the most recent agents are represented by phase 1b/2a or phase 2 data, with the well-documented tendency of early-phase effect sizes to attenuate in larger populations. English-language restriction may have under-represented process-chemistry literature published elsewhere. No protocol was registered and no formal risk-of-bias assessment was applied across the corpus, for the reasons given in the Methods. And a narrative review, however disciplined its search, retains selection judgement that a systematic review would constrain more tightly. We have tried to make those judgements visible rather than invisible.

6. Conclusion

Peptides and macrocycles have moved from a neglected middle ground to the centre of metabolic drug development, and the reason is chemical rather than biological. Protraction through fatty diacid acylation, protease resistance through position-2 substitution, and conformational constraint through cyclisation, stapling and N-methylation turned fragile hormones into weekly medicines. On that foundation, unimolecular poly-pharmacology lifted weight loss from roughly 15% to roughly 24%, within reach of surgical benchmarks. What remains unfinished is instructive. Tolerability rather than potency now caps dosing; oral delivery remains partially solved at best and a process mass intensity near 33,000 sits awkwardly beside multi-tonne demand. The next decade's decisive advances will likely come from permeability chemistry and sustainable manufacture rather than from further receptor combinations. Whether these drugs become a global intervention or a selectively available one depends on that unglamorous work.

References