Cyclic peptides are a distinct class of drug-like molecules bridging the gap between small molecules and biologics, but their ease of synthesis is often highly sensitive to small changes in sequence, topology and reaction conditions. The desired product is a single, correctly folded macrocycle, so any 'diversity' - open-chain impurity, dimer, epimer or truncated species - counts against yield, complicates purification and inflates cost. Optimization is therefore not an academic exercise, but an economic imperative: a route that is of academic interest today could become pre-clinical viable tomorrow if conversion efficiency can be improved by even a modest amount. The most common obstacles are low cyclisation yield, competing oligomerization, racemization at the ligation site and aggregation-driven precipitation; each of these bottlenecks is dependent on inter-related variables that need to be balanced, rather than individually maximized. This review explores the underlying causes of these challenges and provides a concise overview of the practical solutions, born of decades of collective experience, that have been found to be effective in practice.
Fig. 1 Synthesis of cyclic peptides through a combination of solid phase and liquid phase peptide synthesis strategies.1,2
Multigram-scale production of cyclic peptides using standard laboratory protocols has repeatedly proved challenging. Sequences known to cyclize well at the milligram scale may suddenly stop cyclizing when scaling up, or the target may only be obtained after cumbersome chromatographic separation from isomeric and oligomeric byproducts. This unreliability is due to the unique thermodynamic profile of macrocyclization. This intramolecular process is entropically uphill and competes with intrinsically faster, enthalpically-favorable side reactions. The linear precursor is a dynamic mixture of extended, collapsed, and aggregated conformers; only a small fraction transiently sample a cyclisation-friendly, "pre-cyclic" geometry in which the reacting termini or side chains are both proximal and properly aligned. The system will naturally shift towards lower-energy bimolecular processes (dimerization, epimerization, hydrolysis, etc.) without targeted optimization, resulting in reduced yield and/or purification complexity. Awareness of this low-hanging fruit—low cyclisation conversion, oligomer formation, stereochemical drift—allows the development of a rational optimization strategy and moves synthesis away from empiric, "screen-everything" campaigns and towards informed, data-driven, and ultimately scalable protocols.
The synthesis of a cyclic peptide is not a single reaction, but a series of interdependent equilibria, each of which is primed to fail in some way. The linear precursor must first assume a conformation that places the termini in effective contact, overcoming entropic forces and often being prevented by aggregation. The activated carboxylate must then acylate before it epimerizes, hydrolyses, or diffuses to another molecule, while the nucleophile must be deprotonated but unreactive with any other electrophile. Multiple equilibria make the kinetic window narrow: a solution can contain too little activation (in which case the desired ring does not form) or too much (resulting in oligomers, diastereomers, or truncated fragments). Low cyclisation yields, concentration-dependent dimerization, and stereochemical erosion thus appear not as random catastrophes, but as correlated symptoms of the same thermodynamic deficit, each to be treated with orthogonal optimization strategies that must be balanced against one another until a general, robust, and transferable protocol is devised.
Three parameters play a particularly important role in the sub-stoichiometric conversion that most commonly results from a ring closure: the effective concentration, the intrinsic ring strain, and solubility-driven aggregation. The probability that the two termini will meet again in the same solvent cage decreases exponentially with chain length: for short peptides (< seven residues), the entropic penalty is further increased by bond-angle distortion, which raises the transition-state energy. Simultaneously, high hydrophobicity or β-sheet propensity of the sequence can lead to self-association into fibrillar or micellar aggregates that sequester the activated carboxylate away from the incoming nucleophile. Optimization therefore begins with predictive sequence editing: for instance by inserting Gly, Pro or D-amino acids at key positions to nucleate turns, elongating the linker between bulky side chains or moving the cyclisation junction away from β-branched residues. Solvent engineering is also important, either by shifting from polar aprotic solvents that tend to favour aggregation to fluorinated alcohol/water mixtures that can disrupt β-sheets, or by using chaotropic salts that can increase peptide solubility without significantly increasing the rate of thioester hydrolysis. For chemistries that can be performed in solution, pseudo-high-dilution can in some cases be achieved on-resin or in situ inside microdroplets, which allows one to keep the intramolecular pathway available while maintaining bulk concentrations that are synthetically convenient. Careful variation of temperature, pH and catalyst identity can then tune the kinetic balance, often revealing sharp thresholds where conversion suddenly jumps from negligible to near-quantitative with little increase in oligomeric noise.
In cases where cyclisation is theoretically possible, it may be "out-competed" by intermolecular acylation when the local concentration of activated ends becomes excessive. The resulting, linear oligomers not only waste precious starting material, but also co-elute with the desired product in purification steps to provide false positive results in bioassays. The most direct solution to this is dilution, but this dramatically impacts the throughput of the synthetic process, and is unlikely to work for sequences that are able to associate face-to-face, regardless of the dilution. A more general solution is to combine macro-dilution with micro-concentration: the peptide is attached to a soluble polymer or resin bead, which forces the reactive ends to share a solvent shell while the majority of solution is relatively dilute. Alternatively, the sequence may be engineered to contain self-repelling features (charged sidechains, N-methylated amides or temporary PEG-like inserts) that decrease the probability of two activated ends colliding without affecting the properties of the final macrocycle. Kinetic control is also possible: a change from fast, highly reactive coupling reagents to slower, more selective reagents will reduce the instantaneous concentration of activated species, shifting the balance back towards the unimolecular route. A final solution is found in flow-chemistry platforms, which can continuously separate nascent cyclic product from the unreacted linear precursor and so suppress secondary oligomerization to produce multigram quantities of macrocycle under conditions that would traditionally require liters of solvent.
Stereochemical attrition almost invariably arises at the activated carboxylate during the cyclisation reaction. Formation of oxazolones, base-mediated enolization, or direct generation of enolates all present potential sources of partial or complete epimerization of the C-terminal residue leading to poorly separable diastereomeric macrocycles which can exhibit low levels of bioactivity. The use of an appropriate protecting group strategy is the first line of defence: using acid labile side-chain protections that are not removed under the mildly basic cyclisation conditions, or conversely the use of base labile protections in acid catalyzed cyclizations, will prevent premature exposure of additional nucleophiles on the cyclisation components. Choice of coupling reagent is also critical – phosphonium and uronium salts that are less likely to form oxazolones (HATU, PyAOP) are favored over carbodi-imides when coupling racemization sensitive residues, or if the latter are used, additives that suppress the oxazolone pathway can be employed (Oxyma, ethyl cyanoglyoxylate). Temperature and base strength can be optimized to as low as possible to achieve reasonable conversion, recognizing that stronger bases will favor both acylation and enolization. If racemization is still observed it is possible to move the cyclisation point to a glycine or proline residue, thereby eliminating the stereogenic center, but still allowing for efficient ring closure. Racemization can also occur in the post-cyclisation steps, particularly during global deprotection or oxidative folding steps; radical scavengers, rigorous exclusion of metal ions, and strict control of pH during these steps will preserve the stereochemical integrity set in the ligation.
Optimizing the cyclisation event is less a single adjustment than a concerted calibration of solvent polarity, reagent kinetics, additive cooperativity and phase choice. Because the same elementary step that forges the ring also governs epimerization, oligomerization and aggregation, the coupling system must be tuned to accelerate intramolecular acyl transfer while suppressing every competing pathway. This is achieved by matching the activation mode to the steric demand of the junction, modulating base strength to keep the nucleophile deprotonated yet non-racemizing, and introducing auxiliary molecules that template the peptide into a cyclisation-ready conformation. The following sections dissect these levers in turn, contrasting classic carbodiimide/benzotriazole cocktails with modern oxime/uranium-free protocols, and comparing the convenience of on-resin pseudo-dilution with the flexibility of solution-phase optimization.
Choice of coupling reagent and activator begins with a survey of the electronic and steric environment surrounding the site of ring closure. In cases of low hindrance around the junction, the low-cost DIC/HOBt combination is appealing: diisopropylcarbodiimide creates a short-lived O-acylisourea which is captured by N-hydroxybenzotriazole to yield an OBt ester, as a crystalline, shelf-stable compound, with reduced risk of base-promoted racemization as no tertiary amine is needed. In the presence of β-branched or N-methyl residues, however, the slower kinetics of OBt ester formation allows for a competing oxazolone side-reaction; the use of HATU instead provides an aminium-type activator, which forms a more electrophilic O-At ester in situ, with a shorter lifetime of the activated species, thus reducing epimerization at the expense of higher reagent cost and moisture sensitivity. The solvent polarity must be balanced against the tendency of the peptide to aggregate: DMF or NMP can solvate charged side chains and help to dissociate β-sheet contacts, while mixtures with dichloromethane or 2,2,2-trifluoroethanol can be varied to achieve solubility while not unduly increasing the rate of thioester hydrolysis in NCL-based procedures. Temperature is kept low (≤30 °C) to avoid loss of stereochemical integrity, but mild heating can be employed for more hindered sequences provided that the strength of the base is reduced to avoid accumulation of oxazolones.
Auxiliary reagents may have a disproportionate effect by guiding the peptide through conformational space or stabilizing high-energy intermediates. The 7-aza analogue of HOBt, HOAt, has been found to have anchimeric assistance for OBt ester decomposition, which in turn reduces the activation barrier for aminolysis, therefore, decreasing both reaction time and racemization probability; additionally, its higher acidity (pKa ≈ 3.3 vs 4.6 for HOBt) guarantees a strong anion formation even at near-neutral pH. If base-free reactions are needed, ethyl diisopropylamine (DIEA) can be used at sub-stoichiometric levels to deprotonate the incoming amine, without increasing the pH much, while stronger bases are used only for stubborn couplings and quenched as soon as the color change is observed. Pseudoproline dipeptides, on the other hand, are a conformational rather than electronic solution: the incorporation of Fmoc-Xaa-Ser(ΨMe,MePro) or ΨPro-Gly motifs transiently interrupts β-strand propagation, which effectively increases the local concentration of reactive termini and thus strongly improves cyclisation efficiency, especially for hydrophobic sequences with a propensity for on-resin aggregation . Halogenated benzotriazoles (6-Cl-HOBt, 6-CF3-HOBt) can be used to further modulate leaving-group ability, as well as decrease the explosive hazard of the dry solid, an attractive option for automated syntheses, where reagent stocks may be kept for long periods of time . In combination, these additives create a synergistic microenvironment that drives the intramolecular acyl transfer over competing hydrolysis or oligomerization pathways.
The decision to cyclize on- or off-resin may depend on throughput/scale and sequence properties. Cyclization on-resin can take advantage of pseudo-dilution effects: physically isolated chains in the polymer matrix can react intramolecularly at high bulk concentrations, which would favour dimerization in the free state. Resin selection is also important: a high-loading polystyrene resin has high mechanical stability for multiple coupling cycles, while a PEG-grafted resin will swell more in polar solvents, increasing accessibility of the activated carboxylate buried within the bead. The anchoring method must also be able to tolerate the cyclisation conditions: safety-catch methods, which release the peptide as a thioester, can directly undergo NCL-type cyclisation on-resin, while Wang or Rink amide methods can be activated by carbodiimide but require subsequent cleavage and solution-phase macrolactamization. Solution-phase optimization on the other hand allows for full flexibility in solvent composition and temperature, allows real-time monitoring of reaction kinetics, and more straightforward scale up (flow chemistry), but requires careful control of low concentrations (often<1 mM) and has the potential for precipitation of protected intermediates. A practical approach may be to perform parallel, miniaturized screens of both methods (on-resin at higher concentration and in solution at higher dilution) by LC-MS to empirically determine which topological approach gives higher conversion and cleaner product profile. Optimization of the preferred phase then allows for additional variables, such as the stoichiometry of base, the identity of the additive, and time, to be varied; in many cases, a hybrid approach (on-resin activation and solution-phase cyclisation) can combine the advantages of both reactions while minimizing the limitations of each.
Optimizing a peptide sequence for macrocyclization is a trade-off between entropic penalty, ring strain and the requirements for displaying pharmacophores in a bio-relevant geometry. Rings below 7 residues have a low entropic penalty, but often cannot close since the linear precursor is unable to adopt a "loop-like" conformation; successive trans-peptide bonds lock the reactive termini too far apart while eclipsing interactions push the ground-state energy. Macrocycles from 7 to 20 residues in size, by contrast, have enough conformational flexibility to orient termini or side chains into close proximity while being small enough to avoid the statistical dilution of binding affinity seen with larger rings. Within this range, the position of glycine and proline residues becomes critical. Glycine, by virtue of its absence of a β-carbon, imparts local flexibility to relieve angle strain and can act as a hinge around which the peptide folds. Its α-proton also reduces steric clash during the final bond-forming step. Proline's locked φ angle and increased propensity for cis-amide conformers make it a pre-organized β-turn template that shortens the effective distance between otherwise distant residues and biases the chain into a cyclic geometry. A single or double insertion of such a "turn inducer" residue, typically adjacent to the cyclisation junction, can turn an otherwise unworkable sequence into a high-yielding substrate without modifying the primary pharmacophore. In addition to careful size and turn tuning, the global composition must also be judiciously balanced: extreme hydrophobicity will lead to aggregation in the reaction solvent while over-charging can lead to intermolecular electrostatic contacts that out-compete the intramolecular acylation reaction. Successful ring design is therefore an iterative process between computational pre-organisation metrics, empirical cyclisation screens, and downstream biological read-outs, with glycine and proline serving as the tunable hinges that mediate between synthetic accessibility and functional performance.
Isolation of a cyclic peptide requires orthogonal separation modes that take advantage of the change in hydrophobicity, charge distribution and hydrodynamic radius that the backbone closure imposes on the linear parent. Gradient reversed-phase HPLC is a first option: a low % of acetonitrile per minute can separate cyclic product from linear precursor and deletion byproducts, with the macrocycle normally eluting later due to its more compact, solvent-excluded conformation. Fine-tuning of ion-pairing reagents (formic acid for +ve mode detection or ammonium bicarbonate for volatile buffers) can improve separation without loss of electrospray ionization response. MS gives the first indication of correct identity: the observed m/z should be identical to the theoretical mass less one water (for head-to-tail lactams) or two protons (for disulfide-linked variants). High resolution can also identify any oxidation adducts, dimers or residual protecting groups that may have co-eluted at the same UV. Retention-time profiling is especially useful if a series of analogues are being screened: even small sequence modifications alter the hydrophobic moment to give diagnostic ΔtR patterns that can be used to flag successful substitutions prior to NMR or bioassay. Final confirmation of stereochemistry can be obtained from 1H–13C HSQC spectra in aqueous buffer, which will detect the down-field shift of the newly formed amide carbonyl and disappearance of terminal amine resonances, thus confirming that the correct junction (and no extra atoms) has been made. Ion-mobility separation coupled with LC-MS can be used to distinguish cyclic monomers from early-eluting, monoisotopic dimers which have the same m/z but a different CCS in the gas phase. Combined, these orthogonal analysis steps provide a rapid, low sample consumption workflow that can guide iterative optimization and ensure that only correctly folded, chemically authentic macrocycles progress to biological testing.
A trend in modern cyclic-peptide workflows is the transition from the current "craftsman-style", one-jar-at-a-time processes, to automated platforms where robotics produce hundreds of analogues and in silico methods (cloud-based algorithms) predict which variants are promising and worth synthesizing next. Automation obviously eliminates the human latency of manual pipetting, but has more value in terms of closed-loop integration: synthesizers that alter the delivery rates when a downstream inline UV dip shows incomplete deprotection, or LC-MS modules that send an event to lengthen coupling times if the previous cycle's product shows truncated sequences. When these physical data streams are also introduced into machine-learning models, the system learns to predict failure rather than just report it, reducing months of empirical optimization to days of unattended iteration.
For all its versatility, traditional SPPS remains fundamentally an operationally defined process, and it is here that the second generational wave of synthesizers diverges, by implementing cyclic-peptide assembly as a multi-parameter optimization exercise rather than a pre-established protocol. Common fluidic backbones multiplex parallel reactor blocks to enable the independent but concurrent exploration of activation chemistries, resin types and cyclisation strategies, while ensuring that all comparative datasets are produced without operator bias. Inline UV/Vis monitoring of Fmoc release allows the controller to repeat a deblock cycle or to swap to a stronger base if deprotection fails, with no user input required. When assembling a head-to-tail macrocycle, the same fluidics handle on-resin thioester formation, pseudo-dilution cleavage, and spontaneous macrocyclization in one unattended reaction sequence that bypasses the manual transfers that are otherwise a likely source of oxidation and sample loss. After cleavage, the same platform can automatically load the crude onto an integrated LC-MS, with the retention time and observed mass being added to the metadata for each well position. Cumulatively, the resulting metadata from successive runs forms a dataset that can be fed to predictive algorithms that infer, for example, that sequences with consecutive β-branched residues need lower activation temperatures, or that allyl-based side-chain protection needs longer wash times. The overall effect is a continuously self-improving process where each synthetic campaign both feeds from and refines the next.
Supervised models trained on thousands of published cyclisation outcomes are now able to predict, from primary sequence alone, the probability that a given peptide will cyclize cleanly, remain soluble during purification, and retain biological activity after macrocyclization. Embedding layers convert strings of amino acids into descriptors of physicochemical properties (hydrophobic moment, turn propensity, predicted aggregation propensity, etc. ), while graph neural networks learn long-range couplings that resemble contact maps in protein folding. Ensemble regressors provide not only a point estimate of cyclisation yield, but also confidence intervals to inform the chemist on when to step out of their comfort zone to venture into high-risk, high-reward regions of sequence space. Active-learning loops incorporate robotic feedback: when the synthesizer reports that it achieved an unexpectedly low yield, the updated data point is automatically ingested and the model is retrained overnight so that the next morning's suggestions have already been bias-corrected. Generative algorithms can go even further, by proposing entirely new sequences that sit on predicted Pareto frontiers of yield vs. permeability vs. target affinity, effectively converting optimization from a retrospective activity into a prospective design engine. It's important to note that the models are chemistry-agnostic: they learn that NCL behaves differently from lactam cyclisation, and they weight training examples accordingly so that their recommendations still apply even when the synthetic route changes.
An example of a recent 12-mer cyclic peptide campaign which utilized the integrated workflow is shown. The target contained a β-branched tetrad previously associated with < 20 % cyclisation yield. A machine-learning screen identified three substitutions (Gly in position 4, Pro in position 8, and an i→i+4 lactam constraint between Lys-3 and Glu-7) to relieve backbone strain while maintaining a hot-spot tyrosine. The robotic synthesizer prepared a 48-member array exploring these edits in parallel, each peptide with on-resin vs. solution cyclisation arms. Inline LC-MS conversion was monitored every 30 min and showed that on-resin NCL at 25 °C with 2 % DIEA gave the best conversion and that solution-phase HATU coupling gave a slightly different sequence preference. The top hit was resynthesized without human intervention on a 200 mg scale, affording material of > 95 % purity after a single prep HPLC injection. Total cycle time from design to purified peptide: four days, compared with three weeks for the previous manual optimization.
Iterative synthesis, previously cast in the mold of monotonous trial-and-error experimentation, has been redefined by data and algorithmic guidance. Robotics are used to ensure that all chemical datapoints are fully instrumented with metadata (reaction temperature, solvent lot number, age of activator) and machine-learning models are employed to transmute noisy data into robust sequence rules. The marriage of the two compresses discovery timelines, reduces wasted material, and exposes structure–property relationships that the human mind might otherwise overlook. As the datasets generated by these experiments flow to cloud-based platforms, where anonymized information is shared across laboratories, the cyclic-peptide field is advancing toward a self-reinforcing ecosystem where each failed reaction from a lab bolsters the predictive power of the collective and hastens the path from exploratory sequence to clinical candidate.
Cyclic peptide synthesis often presents unique challenges—from steric hindrance and low coupling efficiency to poor cyclization yields and purification difficulties. Our team specializes in diagnosing and resolving these obstacles to deliver high-quality macrocycles at any development stage.
We offer:
If your cyclic peptide program is facing performance issues, our engineering support and optimization strategies can help you achieve successful outcomes.
References
