A backbone amide NH sitting in water is hydrogen bonded to solvent. Carrying it into the hydrocarbon core of a bilayer means paying that bond off and getting nothing back, and a cyclic hexapeptide with no N-substituted residues has six of them before a single side chain is counted. That desolvation bill is the passive permeability problem for peptide macrocycles in one line, and it is also why permeability refuses to track mass or computed polarity in any simple way. What sets the cost is not how many donors the molecule owns, but how many of them are still exposed at the moment it enters the membrane.
Where passive diffusion stops being cheap
The classical boundary is a property count: poor absorption or permeation is more likely when a compound has more than five hydrogen bond donors, more than ten acceptors, a molecular weight above 500, or a calculated log P above 5 [1]. A cyclic hexapeptide clears the donor limit on its backbone alone.
For this class specifically, the size ceiling has been located. Reviewing what is known about size limits on cell permeability, Matsson and Kihlberg note that libraries of cyclic peptides displayed a steep drop-off in membrane permeability at molecular weights above 1000 Da, and argue that this cutoff likely constitutes an upper size limit for more druglike compounds as well; the band from 500 to 1000 Da, between the two boundaries, they describe as virtually unexplored [2].
The more useful observation is that the two things you need pull against each other. From an analysis of the structures of 20 approved macrocyclic drugs, Whitty and co-workers proposed that good solubility requires a topological polar surface area of at least 0.2 × molecular weight, while good passive membrane permeability requires a molecular — that is, three-dimensional — PSA in nonpolar environments of no more than 140 Å2, and showed that one or other of those limits is almost invariably violated above 600 Da [3]. Note which descriptor sits on which side. The solubility criterion is computed from the connectivity graph and cannot change; the permeability criterion is a property of a conformation in a named environment. A molecule that satisfies both at once is not presenting the same surface in both places.
The conformational answer
The first evidence came from partitioning rather than from structure. The partition coefficient of cyclosporin A was measured in octanol/water and in heptane/water by centrifugal partition chromatography, and comparison with model compounds showed that its hydrogen-bonding capacity changed dramatically between an apolar solvent, where it is internally hydrogen bonded, and polar solvents, where it exposes its hydrogen-bonding groups to the solvent. Molecular dynamics in water and in CCl4 supported a solvent-dependent conformational change whose interconversion is slow on the simulation timescale [4].
The controlled test on synthetic peptides came later and is the experiment worth knowing. A series of cyclic hexapeptide diastereomers built on the sequence cyclo[Leu-Leu-Leu-Leu-Pro-Tyr] yielded two members whose membrane diffusion rates differed by nearly two log units. Solution NMR and hydrogen/deuterium exchange showed that diffusion rate tracked the degree of intramolecular hydrogen bonding and the exchange rates; the most permeable diastereomer, cyclo[D-Leu-D-Leu-Leu-D-Leu-Pro-Tyr], diffused at a rate comparable to cyclosporin A [5].
Diastereomers share molecular formula, mass, donor and acceptor counts and topological polar surface area exactly. A spread that wide across such a set is a direct demonstration that any descriptor computed from the connectivity graph is blind to the effect by construction — not poorly calibrated for it, blind to it.
Chameleonicity, and how the two states are actually shown
The description that fits is a molecule presenting differently in the two phases: polar groups handed to the solvent in water, shielded by intramolecular hydrogen bonds in the low-dielectric interior. Direct evidence for that behaviour was for a long time confined to cyclosporin A. Danelius and co-workers extended it, showing that the non-peptidic macrocyclic drugs roxithromycin, telithromycin and spiramycin behave as molecular chameleons, with rifampicin showing a less pronounced behaviour. The first three display marked yet limited flexibility and populate significantly less polar and more compact conformational ensembles in an apolar than in a polar environment; the drugs' passive cell permeability correlates with their three-dimensional polar surface area, corroborating two theoretical permeability models that had been developed for cyclic peptides [6]. The analytes in that study were macrolides and rifampicin, not peptides; what it establishes for the peptide case is that the two theoretical models transfer, not that any given peptide behaves this way.
The demonstration on your own compound is a solvent-swap NMR experiment. Solvent-dependent conformational change is easiest to identify by 1D 1H NMR when two or more distinct populations interconvert slowly on the chemical shift timescale — longer lived than roughly 10 ms for typical 1H chemical shift differences between conformers at 600 MHz — because the spectrum then carries at least double the expected number of resonances [7]. In practice that means acquiring the same peptide in a polar solvent such as DMSO-d6 and in an apolar one such as CDCl3, and taking populations from the integrals.
A designed macrocycle, D9.16, shows what a positive result looks like. In CDCl3, two slowly exchanging states were observed in equilibrium: a trans–cis state matching the structures seen in DMSO-d6, by X-ray crystallography and in the trans–cis design model, and a trans–trans state closely matching the trans–trans design model, with all backbone NH groups satisfied. The population of the trans–cis state falls from 100% in the polar solvent to 45% in the non-polar one [7].
Two limits on that readout. It requires slow exchange: a peptide whose conformers interconvert fast on the shift timescale gives one averaged set of resonances in both solvents, and any population shift has to be inferred by other means — hydrogen/deuterium exchange, the observable that carried the diastereomer result [5], or the amide temperature coefficients used below to choose N-methylation sites. And CDCl3 is a stand-in for a membrane interior, not a membrane. It supplies a low dielectric and nothing else: no acyl chain ordering, no headgroup region, no interfacial water.
Designing the switch rather than finding it
The design literature makes the mechanism actionable and, more usefully, separates the two routes to it. In a study of 184 designed macrocycles of 6 to 12 residues carrying noncanonical backbone modifications, experimental structures were determined for 35, of which 29 were very close to the computational models. With that degree of structural control, membrane permeability could be achieved systematically by ensuring that all amide NH groups are engaged in internal hydrogen bonding interactions, and 84 designs across the 6 to 12 residue size range crossed membranes with an apparent permeability greater than 1 × 10-6 cm/s. Designs that retained exposed NH groups were made permeable by the other route: designing an alternative isoenergetic fully hydrogen-bonded state that the lipid membrane favours [8].
That second route is the chameleon built on purpose, and isoenergetic is the operative word. The two states have to be close enough in free energy that the environment, rather than an intrinsic conformational preference, decides which one is populated.
Removing the donor instead of hiding it
Backbone N-methylation attacks the same term from the other side: an amide bearing a methyl on nitrogen has no donor to desolvate. The synthetic route carries a structural message of its own. On-resin selective N-methylation of cyclic peptides showed that both the selectivity and the degree of methylation depended on backbone stereochemistry, which suggests that conformation dictates the regiochemistry of the reaction; the permeabilities of the resulting N-methyl variants were corroborated computationally against a 1,024-member virtual library of N-methyl cyclic peptides [9].
Choosing which amide to cap can be done from NMR rather than from a full structure. Amide temperature coefficients identify the exposed amides in a cyclic peptide, and N-methylation then caps those, modifying the overall solvation properties of the peptide. Five leucine-rich scaffolds with selectively N-methylated derivatives were assessed in vitro by PAMPA and by Caco-2 assay, the approach being offered as a rapid alternative to elucidating three-dimensional structures [10].
What N-methylation does not give you is a dose–response in the number of methyls. A polyalanine cyclic hexapeptide library of 54 members, varying in both the number of N-methyl groups (one to five) and their position, was screened for permeability in the Caco-2 model. Most members were poorly permeable, with Papp below 1 × 10-6 cm/s, lower than mannitol as the marker for paracellular permeability, while 10 analogues reached Papp above 1 × 10-5 cm/s, similar to testosterone as a transcellular marker. No correlation was found between the number of N-methylated groups and the enhanced permeability; instead, 9 of the 10 permeable peptides carried an N-methyl placed adjacently to the D-Ala position. The transport route was not fully characterised, and the authors report that the data suggest a facilitated diffusion [11].
PAMPA and cell monolayers disagree, and the disagreement carries information
A PAMPA well reports transfer across an artificial lipid membrane with no cells in it. A Caco-2 monolayer adds carrier-mediated transport and a paracellular route, the latter being what mannitol marks in the library above [11]. The cleanest demonstration that this distinction is not academic for peptides used enantiomeric pairs. Three polar and two lipophilic cyclic peptide enantiomer pairs showed identical lipophilicity (log D7.4) and identical passive transcellular permeability by PAMPA, which the authors describe as the expected outcome. The polar pairs nonetheless showed different Caco-2 Papp in both the a–b and b–a directions, as did one of the lipophilic pairs, while the second lipophilic pair was identical within error. The authors read this as the first evidence for the involvement of carrier-mediated transport for peptides, especially those of polar nature [12].
The practical consequence is that a PAMPA/Caco-2 discrepancy is a result, not an assay failure. Equal PAMPA with unequal Caco-2 across a stereochemical pair points at a chiral recognition step somewhere in the cell system; a Caco-2 value far below PAMPA points at directional transport or at the assay itself.
Running both across a series is more informative than either alone. Sixty-two cyclic hexapeptides were characterised in both assays and their permeabilities correlated against conformational parameters obtained by rapid chromatographic and NMR methods, yielding a model in which peptides with very high permeability have high lipophilicity and few solvent hydrogen bond interactions, whereas peptides with very low permeability have low lipophilicity or many solvent interactions [13]. Those are two separable axes, and a series that moves on only one of them will not tell you which is limiting.
One assay-side artefact should be excluded before any of this is read biologically. Poor recovery is a common problem in the Caco-2 assay and leads to underestimation of the apparent permeability of the affected compounds; a systematic investigation of an automated high-throughput Caco-2 workflow found nonspecific binding to plastic surfaces to be the major cause, and mitigated it by preloading collection plates with organic solvent containing internal standard [14]. That work was done on general test compounds rather than on peptides, but lipophilic macrocycles are exactly the profile at risk. A Papp reported without a recovery figure is hard to interpret on this class.
Why one computed polar surface area cannot see any of this
The direct test has been run. An atomistic model that treats conformational flexibility explicitly, by sampling in low (membrane) and high (water) dielectric environments and quantitatively estimating the solvation free energy lost on moving between them, was built before the experiments and without training data; across 11 cyclic peptides measured by PAMPA it gave a linear correlation with R2 = 0.96. In the same comparison, both the number of internal hydrogen bonds that can form in the membrane and the solvent-exposed polar surface area correlated more poorly with PAMPA permeability than the model did [15]. So counting the hydrogen bonds is not sufficient either. What the model gets right is the energetic bookkeeping of the transfer, not the bond census.
An experimental substitute for the descriptor a calculation cannot supply is available. EPSA, a supercritical fluid chromatography technique developed for peptides, assesses the exposed polarity of a peptide and was introduced as a permeability surrogate for design work [16]. It measures what the molecule actually presents under the chromatographic conditions, which is closer to the quantity the permeability threshold is written in than a topological count is.
Flexibility is then a variable in its own right, and it can compensate. Across 24 cyclic decapeptides sharing a single backbone N-methylation pattern and differing only in their side chains, the population of the membrane-permeable conformation in water — extracted from extensive explicit-solvent molecular dynamics and used as a metric for conformational rigidity, or prefolding — was combined with lipophilicity measurements to expose an intricate interplay between polarity/lipophilicity and flexibility/rigidity, with possible compensating effects on permeability [17]. A rigid, prefolded but moderately polar peptide and a more flexible, more lipophilic one can arrive at the same permeability by different routes, which is a standing warning against reading any single descriptor as causal within a series.
What to put on the record
For a permeability number on a cyclic peptide: the assay and its membrane composition, donor and acceptor buffers, incubation time, recovery or mass balance, and the direction if a monolayer was used. For a chameleonicity claim: both solvents, field strength and temperature, whether exchange was slow or fast on the chemical shift timescale, the integrals behind every stated population, and whether the apolar-solvent conformation was confirmed independently of the model that predicted it. Keep the assay attached to the number when reporting — the assay formats themselves are covered in the companion guide to in-vitro permeability models, and the ring-closure chemistry in the macrocyclisation guide. A conformational hypothesis tested in one solvent is a hypothesis about that solvent.