Cyclic Peptide Conformational Analysis by Solution NMR

Closing a ring is a bet on entropy: strip out the terminal degrees of freedom, couple the surviving torsions to one another, and the molecule should spend more of its time in fewer shapes. Solution NMR is where that bet gets settled, because every observable it hands you — chemical shift, scalar coupling, cross-relaxation rate, residual dipolar coupling — is an average over whatever the molecule is doing on the timescale of that observable. For a cyclic peptide the hard part is not extracting a structure. It is establishing how many structures the average is hiding, and whether the one you fitted is among them.

What ring closure does to the spectrum, and what it does not

The constraint is real, and it registers on the slowest internal motion a peptide has. Cyclising the tetrapeptide Ac-Ile-Gly-Gly-Asn-NH₂ greatly lowered cis-amide populations and slowed cis/trans isomerisation relative to its linear counterpart, measured by two-dimensional exchange spectroscopy in DMSO ^[1]. Fewer states contributing to each averaged observable is precisely what makes the rest of the analysis tractable.

The dispersion half of the argument needs more care. A peptide that samples backbone space freely drives its proton shifts toward random-coil values, and the reference set for those values was measured on disordered linear pentapeptides H-Gly-Gly-Xaa-Gly-Gly-OH spanning the twenty coded residues, as a function of temperature and of trifluoroethanol concentration ^[2]. Departure from that baseline is the evidence that a defined structure exists. Two warnings arrive with the same measurement: agreement with earlier peptide work was good for CH protons but showed significant variability for NH protons, and trifluoroethanol on its own produced significant changes in backbone and side-chain amide shifts, which the authors cautioned against interpreting solely as an indicator of helix formation ^[2]. Amide dispersion is the softest structural indicator on the bench. A well-spread NH region is a reason to keep going, not a result.

Ring closure is also not automatically rigidifying, and the prolyl isomerisation literature says so plainly. Going from the protected linear peptide CF₃CO-Gly-Gly-Pro-Ala-OCH₃ to the closely related cyclic compound cyclo(-Gly-Gly-Pro-Gly-Ala-) increased the X-Pro interconversion rate by a factor of about six ^[3]. In disulfide-constrained cyclic peptides of the type Ac-Cys-Pro-Xaa-Cys-NH₂, the constraints imposed by the disulfide conferred a rate enhancement for cis/trans isomerisation ranging from a factor of two to thirteen over the acyclic dithiol forms, which the authors proposed results from intramolecular catalysis in which the NH proton of the Pro-Xaa peptide bond hydrogen bonds to the proline nitrogen in the transition state ^[4]. A macrocycle can hold a torsion still, and it can just as readily hold it near a transition state.

Nor does cyclisation solve the acquisition problem. For modified macrocyclic peptides the standing difficulties are very weak nuclear Overhauser effects, the semiquantitative nature of the rotating-frame Overhauser effect, and the low number of NOEs and ROEs available in N-methylated peptides ^[5]. You are working with an unusually low number of restraints per residue, which is the same reason the ensemble question is hard to settle from distance data alone.

Two sets of peaks: reading amide isomerism

Doubled resonances in a cyclic peptide spectrum usually point to slow conformational exchange, and the sites to check first are prolines and N-methylated amides. How far this can go is worth a number: from a library of 30 N-methylated peptides built on cyclo(-D-Ala-L-Ala₄-), only seven exhibited single conformations ^[6]. If you N-methylate a macrocyclic backbone, the default expectation is more than one conformer in the tube.

The exchange is slow, which is why it resolves into separate peak sets rather than a coalesced average. Interconversion across X-Pro bonds ran at 2.5 × 10^-3 s^-1 at 25 °C in the cis-to-trans direction for the zwitterionic form of H-Ala-Pro-OH, and it is strongly sequence-dependent: replacing Ala by Phe, Tyr or Trp gave roughly tenfold slower interconversion, His or Glu substitution decreased the rate only slightly, and elongating the chain by adding Ala C-terminal to the proline raised the rate by about a factor of twenty ^[3]. Secondary amides isomerise too, just rarely: in oligopeptides including Ac-Gly-Gly-Gly-NH₂ and Ac-Ile-Gly-Gly-NH₂ studied in DMSO, cis bonds occurred at 0.13–0.23% at 296 K with trans-to-cis rate constants of 4–9 × 10^-3 s^{-1 [1]}.

Those numbers came from two-dimensional exchange spectroscopy and magnetisation transfer, which is the right tool set here. An EXSY experiment run with a mixing time comparable to the exchange lifetime raises cross peaks between the two sets of resonances; integrating them against the diagonal gives forward and reverse rate constants, and repeating the measurement over a range of temperatures gives activation parameters, as was done for the disulfide-constrained peptides ^[4]. Populations come from integrating well-resolved signals in a plain one-dimensional spectrum, so the thermodynamics and the kinetics are separable measurements.

Two cautions on interpretation. First, not every minor peak set is a cis bond: in the DMSO study, multiple minor conformations were detected for most peptide bonds, and the cis isomers had to be distinguished from minor trans isomers that appear because of an adjacent cis peptide bond, using their thermodynamic and kinetic properties ^[1]. Second, the exchange measurement licenses a statement about how many states are slowly interconverting and how fast, and nothing about the geometry of the minor state unless you can assign it separately — an assignment that competes for the same scarce cross peaks already noted for N-methylated macrocycles ^[5].

Amide temperature coefficients, and their fine print

Ramp the temperature, track each amide proton shift, and the slope reports how exposed that proton is. The calibration everyone quotes was built from 793 amides across 14 proteins: temperature gradients more positive than −4.6 ppb/K carry a hydrogen-bond predictivity above 85%, rising to over 93% within the −4 to −1 ppb/K window, with an inverse proportionality between the coefficient and hydrogen-bond length ^[7]. The same analysis is where the structural fine print lives. For hydrogen bonds of similar length, gradients in α-helices are on average about 1 ppb/K more negative than in β-sheets, so a number of helical amides in bonds shorter than 2 Å fall below −4.6 ppb/K; conversely, because their hydrogen bonds are longer, 90% of amides in 3₁₀ helices and 98% in β-turns sit more positive than −4.6 ppb/K ^[7]. Ring currents move the number as well: in seven of eight cases, non-hydrogen-bonded amides strongly deshielded by neighbouring aromatic rings showed coefficients more positive than −2 ppb/K ^[7].

Those figures were derived on proteins, which matters when the analyte is a turn-rich macrocycle with aromatic side chains packed against its own backbone. Two further constraints are load-bearing. Amide and CαH coefficients measured in bovine pancreatic trypsin inhibitor and hen egg-white lysozyme spanned roughly −16 to +2 ppb/K and −4 to +3 ppb/K respectively, and the temperature dependence was generally linear only up to about 15 °C below the global denaturation temperature; the conclusion drawn was that combining temperature coefficients with exchange rates gives a far more reliable hydrogen-bonding indicator than either alone, and that the previously observed unreliability of temperature coefficients in peptides may arise from losses of peptide secondary structure on heating ^[8]. That last clause is the trap for a flexible macrocycle: you heat the sample to make the measurement, and the structure you are probing may not survive the ramp. Bracketing the ramp with a return to the starting temperature, and checking that the shifts come back, costs one experiment.

The solvent can also take the observable over outright. In two α-helical peptides differing in amphipathicity, studied in water/trifluoroethanol, coefficients of −10 to −6 ppb/K — values usually read as indicating no intramolecular hydrogen bond — were seen even for hydrophobic residues in the middle of the helices, while hydrogen isotope exchange indicated intact intramolecular hydrogen bonds over most of the peptide length; the authors suggested that the amide-proton shift pattern in these solvents is dominated by bifurcated intermolecular hydrogen bonding between backbone carbonyl groups and trifluoroethanol ^[9]. Used inside those limits the measurement is fast and genuinely useful: temperature coefficients have been applied to cyclic peptides specifically to identify exposed amides, offered as a rapid alternative to elucidating a full three-dimensional structure ^[10].

The honest reading of a less-negative coefficient is that the proton's shift is insensitive to temperature, which is consistent with sequestration from solvent. Whether that sequestration comes from an intramolecular hydrogen bond, from a side chain lying across the amide, or from a neighbouring ring current is a separate question, and pairing the gradient with an exchange measurement is what separates them ^[8].

NOEs, and what an average actually is

Cross-relaxation depends on the inverse sixth power of the interproton distance, so a measured NOE reports an average of r^-6 across the ensemble, weighted toward whichever conformer brings the pair closest. The consequence was stated cleanly a long time ago: where the individual conformers contributing to the average are not known, the average derived from spectroscopic data represents a virtual structure devoid of physical meaning ^[11]. A single model refined against exchange-averaged restraints can be a shape the molecule never adopts, because the average of two real conformers need not itself be a real conformer.

This is not an argument that NOE restraints are imprecise. On the flexible model compound 4-propylaniline — a small molecule, not a peptide — NOE measurements alone established interproton distances, and hence conformational detail, to within a few percent of their ensemble-averaged values, reducing the demand for additional restraints such as Karplus analysis of scalar couplings ^[12]. The precision is real; it is precision about an average. The failure mode is not noisy distances. It is a well-fitted single structure standing in for a distribution.

The practical response is to fit a bundle rather than a structure. One implementation folds NOE-derived interproton distances into a distance-geometry conformer generator as loose upper or lower bounds in the molecular distance bounds matrix, generates a large conformer set, then applies subselection to yield the bundle that best reproduces the NOE data; it was benchmarked on a set of 24 mostly cyclic peptides for which NOE-derived distances and reference solution structures obtained by other software were available ^[13]. A second route is to improve the restraints themselves rather than the fitting. Exact NOEs measured in viscous solvent mixtures, chosen to replicate different cellular environments, gave a backbone atom RMSD of 0.10 Å for cyclosporin A and resolved distinct conformational states for the multiply backbone N-methylated macrocycle omphalotin A ^[5].

Residual dipolar couplings

NOEs are short-range, and a set of them can be satisfied by more than one global arrangement of a ring. Residual dipolar couplings, measured after weakly aligning the sample so that dipolar interactions no longer average to zero, constrain internuclear vector orientations against a common alignment frame, and so report on the relative disposition of parts of a macrocycle that no NOE connects. Two studies show the shape of the analysis. A conformational deconvolution methodology combined RDCs, J-couplings and intramolecular hydrogen-bond analysis with molecular dynamics simulations and density functional theory calculations, run in both a low-dielectric solvent (chloroform) and a high-dielectric solvent (DMSO), to study the solvent-dependent conformational change hypothesis experimentally ^[14]. The second combined J-couplings, NOEs, amide proton temperature coefficients and RDCs to determine a minimal conformational ensemble for the natural cyclic peptide heterophyllin B in methanol at density functional theory accuracy, with conformer sampling by CREST and energetic sorting by CENSO ^[15].

Take that framing seriously: the stated deliverable is a minimal conformational ensemble. The openly accessible full text is more specific — both isotropic and anisotropic observables consistently supported a single backbone conformation for heterophyllin B, and the two-member ensemble is a proline side-chain result, conformers III and VI weighted 73% and 27% and differing in the ring conformation of one proline, with the anisotropic state formed in methanol using the liquid-crystal-forming oligopeptide AAKLVFF ^[15]. RDCs are averages as well, and their value is not that they escape averaging but that they add orientational information the isotropic data does not contain — the stated motivation being that NMR-based conformational analysis of cyclic peptides is often limited by sparse isotropic solution-state data ^[15]. They are an addition to the isotropic set, not a replacement for it: both studies report J-couplings and NOEs alongside the anisotropic data ^{[14, 15]}.

The conformation you report is the conformation in that solvent

Every number above carries a solvent inside its scope, and for macrocyclic peptides the solvent is not a background condition. Comparing ROESY measurements on cyclosporin A in cyclohexane-d₁₂ and n-hexane-d₁₄, the major conformations in those apolar solvents were essentially the same as in CDCl₃ except for some side-chain rotamers; the same work concluded from simulated free-energy landscapes that the experimentally determined structures of cyclosporin A were only a part of the conformational space ^[16]. That is the general position. The solvent selects a region of the landscape, and your experiment characterises that region.

Where the change in environment is large, the backbone rearranges. In the omphalotin A work already described, the transition from a polar to an apolar environment was accompanied by a rearrangement of its hydrogen-bonding network together with a cis-to-trans isomerisation of an ω torsion angle within a type VIa β-turn ^[5]. Solvent also reaches the thermodynamics of the isomerism itself: the exchange study of secondary amide isomerisation, comparing peptides against N-methylacetamide, revealed a unique enthalpy profile in DMSO ^[1]. DMSO is the default solvent for a poorly soluble macrocycle, and a conformational study run only in DMSO has answered a question about DMSO.

The paired-solvent design — chloroform against DMSO, the same observable set in each ^[14] — is the pattern worth copying. Two solvents at opposite ends of the dielectric range, and the comparison reported rather than one of the two chosen after the fact.

What to record

Ring-closure chemistry and the general mechanics of a solution structure calculation are separate topics. What this particular analysis needs on the record is narrower: solvent and deuteration, temperature and concentration, field strength, the number of resolved resonance sets with their integrated populations, and the mixing time and experiment type behind every distance restraint. State whether the deposited coordinates are a single structure or a weighted ensemble, and if a single structure, say what evidence excludes exchange averaging. Report the temperature range over which each coefficient was linear, and whether an exchange measurement supports the hydrogen-bond assignments that coefficient implies ^[8]. A cyclic peptide reported as one refined structure, with no population statement and no second solvent, has been reported as a hypothesis rather than a measurement.