A Model for the Rapid Assessment of Solution Structures for 24-Atom Macrocycles: The Impact of β-Branched Amino Acids on Conformation

Abstract

Experiment and computation are used to develop a model to rapidly predict solution structures of macrocycles sharing the same Murcko framework. These 24-atom triazine macrocycles result from the quantitative dimerization of identical monomers presenting a hydrazine group and an acetal tethered to an amino acid linker. Monomers comprising glycine and the β-branched amino acids threonine, valine, and isoleucine yield macrocycles G-G, T-T, V-V, and I-I, respectively. Elements common to all members of the framework include the efficiency of macrocyclization (quantitative), the solution- and solid-state structures (folded), the site of protonation (opposite the auxiliary dimethylamine group), the geometry of the hydrazone (E), the C2 symmetry of the subunits (conserved), and the rotamer state adopted. In aggregate, the data reveal metrics predictive of the three-dimensional solution structure that derive from the fingerprint region of the 1D ¹H spectrum and a network of rOes from a single resonance. The metrics also afford delineation of more nuanced structural features that allow subpopulations to be identified among the members of the framework. Well-tempered metadynamics provides free energy surfaces and population distributions of these macrocycles. The areas of the free energy surface decrease with increasing steric bulk (G-G > V-V ~ T-T > I-I). In addition, the surfaces are increasingly isoenergetic with decreasing steric bulk (G-G > V-V ~ T-T > I-I).

Graphical Abstract

INTRODUCTION

Conducting structure–activity relationships, a hallmark of small-molecule drug design, relies on the assumption that modest changes in chemical composition will not significantly affect global conformation.¹ This belief allows molecular parameters to be rationally fine-tuned across a bounded chemical space vis-á-vis the exploration of aliphatic groups to promote hydrophobic interactions. However, this assumption appears increasingly tenuous for macrocycles that contort between multiple conformations of similar energy.^2–4 As a result, slight changes to the composition can change the distribution of conformations as well as offer access to new conformation space.5–7 Understanding and ultimately engineering conformation are of great interest as the community pursues the development of macrocyclic drugs.^8–12

Through experiment and computation, strategies to understand, map, and control conformation space are emerging.^13–15 Studies survey existing collections^16,17 or known bioactive natural products.¹⁸ Oftentimes, the exploration centers on a single Murcko framework—a conserved arrangement of atoms—that is synthetically accessible or otherwise available.¹⁹ Cyclic peptides attract significant attention in this regard.²⁰ Many of these frameworks are inspired by naturally bioactive molecules including Sanguinamide A²¹ and somatostatin²² or by observed bioactivity including the cyclic pentapeptides advanced by McAlpine.²³ In many cases, modest changes (e.g., methylation) can lead to a conformation change and an obfuscation of causality in a structure–activity relationship particularly with regard to passive cellular transport.

Assigning a molecular basis for causality requires knowledge of the structure, itself a labor-intensive exercise. Employing a framework can facilitate these efforts but still necessitates a determination of whether the change in composition constitutes a desired functional variation or leads to a significant structural change. The development of predictive models that confirm conservation of the molecular shape based on simple experiments—here, a 1D ¹H NMR spectrum and rOe pattern with a single resonance—can greatly accelerate these efforts.

Our longstanding interest in triazine chemistry²⁴ and recent studies of hydrazine-substituted triazines^25,26 led us to explore macrocycles incorporating triazinyl hydrazones as molecular frameworks.^27,28 The ease of synthesis and the numerous sites available to explore functional variation make triazine-based macrocycles a versatile platform for pursuing structure–property relationships. Zerkowski and coworkers showed earlier that static, covalent chemistry could yield similar triazine targets.²⁹

Initial efforts produced macrocycles ranging from 22 to 28 atoms—each a different Murcko framework—displaying different molecular shapes ranging from ones that were compact and folded to those that were extended and flat.^27,28 During the course of these studies, however, two isomeric 24-atom macrocycles were prepare—done with glycine and a three carbon acetal (–NHCH₂CONHCH₂CH₂CN–) and the other with β-alanine and a two-carbon acetal (–NHCH₂CH₂CONHCH₂CN–).²⁷ Both showed markedly similar, folded structures in the solid state. To better define the conformational landscape associated with this framework, we chose to prepare additional 24-atom macrocycles that varied in the choice of amino acid. Incorporating β-branched amino acids—valine, threonine, and isoleucine—offers the opportunity to probe steric tolerances. Chart 1 shows these targets as well as the labeling strategy that is adopted for the NMR spectra. These macrocycles are named for the amino acid of interest: G-G, V-V, T-T, and I-I.

Chart 1.

The Murcko Framework, G-G, and the β-Branched Macrocycles^a

^aThe labels associated with the NMR spectra are provided. When the dimethylamine groups (with methyls M1 and M2) are replaced with morpholine groups, an apostrophe is appended to the name. Distances D1 and D2 are the collective variables used in the well-tempered metadynamics calculations. The isomers of I-I result from racemization of the α-stereocenter.

Here, we establish that G-G, V-V, and the isomers of I-I adopt a common three-dimensional shape in solution and in the solid state using NMR spectroscopy and X-ray crystallography, respectively. The structural assignment of T-T rests solely on NMR spectroscopy because no diffraction-quality crystals of T-T have been obtained. Trends in the NMR data reveal that structural predictions are possible and lead to a nuanced model for conformation. This model was applied successfully to both isomers of I-I before X-ray structural data became available to reveal their relationship: they result from racemization of the α-stereocenter leading to an (S,S)-homodimer and an (R,S)-homodimer.

EXPERIMENTAL SECTION

NMR Spectroscopy.

Room-temperature 1D ¹H NMR and ¹³C{¹H} NMR spectra were recorded on a 400 MHz Bruker Avance spectrometer at Texas Christian University. Chemical shifts were referenced to the corresponding solvent resonance (e.g., DMSO-d₆, δ = 2.52 ppm). Low-temperature spectra were acquired on a 500 MHz Varian NMR spectrometer at the University of North Texas in Denton. Structural assignments were made with additional information from COSY and HSQC experiments. Deuterated NMR solvents were used as purchased from either a bottle or ampule.

Well-Tempered Metadynamics Simulations.

The atomistic models were prepared using Antechamber,³⁰ parameterizing the bonded and nonbonded interactions following the generalized Amber force field (GAFF). Atomic point charges were computed using the RESP approach³¹ with HF/(6–31)G* theory by Gaussian 16.³² After an initial minimization and equilibration, a 500 ns-long well-tempered metadynamics (WT-MetaD)³³ simulation was run for each macrocycle with D1 and D2 as collective variables using Gromacs 2021.2³⁴ patched with Plumed.³⁵ After reaching convergence, the deposited bias was reweighted using the Tiwary–Parrinello free energy estimator to obtain the free energy surface.³⁶

Crystallography.

Diffraction data for CCDC 2221530 were collected at Texas Christian University at 100 K on a Bruker D8Quest diffractometer. Diffraction data for CCDC 2233277 were collected at Texas A&M University. Data collection, frame integration, data reduction (multiscan), and structure determination were carried out using APEX2 software.³⁷ Initial structural refinements were performed with XSHELL (v 6.3.1) by the full-matrix least-squares method.³⁸ All nonhydrogen atoms were refined using anisotropic thermal parameters, while the hydrogen atoms were treated as mixed. OLEX2 was used for additional structure refinement and graphical representation.³⁹ Crystals suitable for X-ray diffraction were obtained by dissolving 50 mg of the macrocycle in 3 mL of methanol in a dram vial. The cap was loosely placed on top, and the solvent was allowed to evaporate over two weeks. Crystals were obtained from the resulting residue.

General Chemistry and Synthesis.

Details of the synthesis and characterization of the macrocycles and intermediates appear in the Supporting Information.

RESULTS AND DISCUSSION

Organization of This Report.

After introducing details of the synthesis, the focus shifts to the structural elements that define the conformation of this molecular framework. Common characteristics emerge including (1) the efficiency of macrocyclization, (2) common NMR features, (3) C2 symmetry, (4) an (E)-hydrazone, (5) a common site of protonation, (6) a conserved rotamer state, and (7) folded structures in solution and in the solid state. From these criteria, a model to predict the three-dimensional structure is introduced. Subsequently, distinctive elements of this model are advanced to identify subpopulations that vary subtly in conformation. Macrocycle G-G is different from V-V and the isomers of I-I, while T-T shares elements of both groups. A computational analysis of the free energy surfaces is reported.

Synthesis and Nomenclature.

The details of the synthetic method and its execution are described in the Supporting Information. Scheme 1 outlines the route employed. It represents an improvement over those previously published because it requires fewer synthetic steps and only one or two chromatographic separations.^27,28 Briefly, cyanuric chloride is substituted in a stepwise manner in a single reaction flask to yield an acid intermediate named for the amino acid (e.g., G-acid). This protocol is used to install either the dimethylamine or morpholine auxiliary groups, the former more useful for NMR analysis and the latter more amenable to crystallization. After isolation and characterization, the acid is elaborated to the monomer (e.g., G). Treatment of the monomer with trifluoroacetic acid in dichloromethane followed by slow evaporation of the solvent quantitatively yields the macrocycle (e.g., G-G).

Scheme 1.

Synthesis of the Macrocycles^a

^a(a) BOCNHNH₂, NaOH (aq), THF, −10 °C, 0.5 h. (b) Amino acid, NaOH (aq), RT, 6 h. (c) 40% HN(CH₃)₂ (aq), RT, 6 h. (d) EDC-HCl, H₂NCH₂CH₂CH(OCH₂CH₃)₂, DIPEA, RT, DCM, 8 h. (e) 1:1 DCM:TFA, 72 h.

NMR spectroscopy provides the clearest piece of evidence of macrocycle formation. Figure 1 shows the evolution of spectral complexity during the synthesis of V-V. The wealth of resonances for V-acid and acetal V derives from the existence of up to four rotamers resulting primarily from hindered rotation about the triazine-N bonds. The α-H of valine (labeled α) appears as three broad resonances between 4.5 and 4 ppm in V-acid in populations that are different from those of V. In the macrocycle V-V, a single, resolved α-H is observed. Similar patterns are seen in the region between 7 and 9 ppm of the intermediates wherein multiple signals for exchangeable protons are observed in V-acid and V. Single resonances are observed for NNH, C-NH, and α-NH in the spectrum of V-V.

Figure 1.

The 400 MHz ¹H NMR spectra of V-acid, V, and V-V in DMSO-d₆ reveal convergence to a single species upon macrocyclization.

Characteristics Common to the Molecular Framework.

All of the macrocycles share seven characteristics.

Characteristic 1: Efficiency of Macrocyclization.

These macrocycles are products of dynamic covalent synthesis, a method that is often plagued by yielding a diversity of products including oligomers and polymers that result from kinetic traps.⁴⁰ Here, the quantitative formation of macrocycles—independent of steric bulk—is surprising. We infer that these macrocycles represent thermodynamic minima and hypothesize that the acidic environment might template dimerization.⁴¹ Macrocyclization appears to be independent of the concentration across the range explored to date—from 3 to 100 mM (corresponding to 5 to 165 mg/mL). The rate of evaporation is critical to success. Slow evaporation of the solvent with stirring over 3–7 days provides the best results. Efforts to accelerate macrocyclization are ongoing. Accelerating the rate of evaporation by vacuum or air stream results in a mixture of species that we attribute to the aforementioned, kinetically trapped side products, which can be converted to macrocycles by addition of more acid and solvent. The effects of temperature, pH, or cosolvent choice on cyclization rates and/or product distributions have not been observed nor rigorously explored. These effects have been noted in other systems.⁴² Such studies will be pursued in due course.

Characteristic 2: Common NMR Features.

Given its simplicity, the fingerprint region of the ¹H NMR spectrum between 7 and 13 ppm is particularly valuable for comparing different macrocycles. Figure 2 shows this region for G-G, V-V, T-T, and I-I in DMSO-d₆, a solvent that provides sharp resonances for the exchangeable protons. With the exception of α-NH, the general order of the appearance of resonances is conserved: moving upfield the sequence is NNH, H⁺, C-NH, and then A (N=CH). The categorization of these protons by color (yellow, blue, and green) is addressed later.

Figure 2.

The fingerprint region of the 400 MHz ¹H NMR spectra of G-G, V-V, T-T, and I-I in DMSO-d₆ shows well-resolved resonances and the appearance of two isomers for I-I. The arrow indicates the position of α-NH for T-T. The colored domains indicate subpopulations, G-G-like (blue), V-V-like (yellow), and intermediate (green).

Characteristic 3: C2 Symmetry in Solution.

The degeneracy observed in the ¹H NMR spectra—as shown in Figure 2 and in the upfield region as well—requires that both subunits of each macrocycle adopt C2-symmetric conformations, including both isomers of I-I.

The ¹³C NMR spectra also reflect the symmetry of these macrocycles. Each spectrum presents a single set of degenerate resonances (Figure 3). Again, two resonances with unequal intensities are observed for I-I. These resonances are most evident for the carbonyl (173 ppm) and consistent with a mixture of two isomers, not a single asymmetric molecule.

Figure 3.

The downfield region of the 100 MHz ¹³C{¹H} NMR spectrum. Trifluoroacetic acid is present in the spectra (q, 159 ppm). The colored domains indicate subpopulations, G-G-like (blue), V-V-like (yellow), and intermediate (green).

Characteristic 4: An (E)-Hydrazone.

More amenable to rOesy than nOesy experiments, the macrocycles show a strong correlation between A and NNH (Chart 2). Indeed, it is the most dominant signal in most cases. Cross peaks indicative of the (Z)-isomer are not observed across a variety of solvents (D₂O, CD₃CN, and CD₃OD). Consistent with this observation, the crystal structures of the morpholine derivatives, G′-G′, V′-V′, and I′-I′, adopt (E)-hydrazones.

Chart 2.

Site of Protonation and the Rotamers Derived from Hindered Rotation about the Triazine-N Bond^a

^aProtonation occurs opposite the dimethylamine group and adjacent to α-NH. ROESY correlations observed (solid arrows) and absent (dotted arrows) confirm that rotamer I is adopted in solution. Rotamer I is observed in the solid state.

Characteristic 5: A Common Site for Protonation.

The site of protonation can be determined by rOesy experiments. An rOe between H⁺ and α-NH places these protons in close proximity. The absence of an rOe correlation between H⁺ and either methyl group of the dimethylamine substituent, M1 or M2, argues for the proton on the opposite end of the triazine (Chart 2). The crystal structures of the morpholine derivatives, G′-G′, V′-V′, and I′-I′, show protonation at this site. Intuition suggests that this site is preferred because the lone pair of the hydrazone –C=NNH– can offer additional electron density to stabilize protonation.

Characteristic 6: A Common Rotamer State.

Hindered rotation about the triazine-N bonds leads to conformational isomers that are apparent in the acyclic intermediates.^42–47 The barrier to rotation for this bond in acyclic systems has been determined to be ~15 kcal/mol.⁴⁶ The macrocycles adopt a single rotamer state that can be assigned by rOesy experiments. The solid arrows in Chart 2 represent rOe correlations that are observed experimentally. The dotted arrows indicate ones that would be expected for the rotamer but are not observed. Only rotamer I satisfies the experimental observations. That is, strong rOe correlations are observed between H⁺ and α-NH, but not between H⁺ and the NNH of the hydrazone. Rotamer I is also observed in the crystal structures of the morpholine derivatives, G′-G′, V′-V′, and I′-I′.

Characteristic 7: Folded Conformations in Solution and the Solid State.

In solution, the macrocycles present an rOe between the hydrazone, A, and one or both of the auxiliary methyl groups, M1 and M2. The macrocycle G-G shows an rOe between A and both M1 and M2, while the β-branched macrocycles show only an rOe between A and M2. These correlations are shown in the Supporting Information (Figure S68). The distance between these groups requires that the molecule adopt a folded conformation as observed in the solid state.

The Emergence of a Predictive Model.

The ¹H and ¹³C NMR data form the foundation of a model that is predictive of the conformational features of the framework (Table 1). Five resonances in the ¹H NMR spectrum and the network of rOes to A (to NNH to establish the (E)-hydrazone and to M1 to establish folding) convey critical structural information quickly. The carbonyl and A resonances in the ¹³C NMR spectrum are of lesser importance but also corroborate assignment.

Table 1.

Comparison of Features of the Two Structural Models^a

Characteristic	G-G	T-T	V-V	I-I (major)	I-I (minor)
δ 1H NNH (ppm)	12.41	12.58	12.62	12.62	12.61
δ 1H H*(ppm)	11.63	11.62	11.36	11.39	11.37
δ 1H C-NH (ppm)	8.94	9.16	9.22	9.23	9.18
δ 1H A (ppm)	7.47	7.47	7.49	7.48	7.49
δ 1H α-NH (ppm)	7.84	7.47	7.18	7.21	7.19
rOe A to NNH	strong	strong	strong	strong	strong
rOe A to M1	yes	-	-	-	-
rOe A to M2	yes	yes	yes	yes	yes
δ 13C C=O (ppm)	171.4	171.9	172.9	173.0	173.2
δ l3C A (ppm)	147.9	147.9	148.0	147.9	147.9

^aThe cells are colored to indicate G-G-like behavior (blue), V-V-like behavior (yellow), intermediate behavior (green), and common behavior (no color).

Subpopulation Distinctions.

Trends within the data suggest that even more nuanced elements of conformation can be inferred. Two subpopulations emerge as indicated by the colored domains of Table 1 (and Figures 2 and 3). One subpopulation is represented by G-G, the other subpopulation by V-V and the isomers of I-I, whileT-T shares elements of both. These distinguishing characteristics, so-called distinctions, arise from differences in (1) π–π stacking informed by the rOesy network, (2) the dynamic behavior informed by variable-temperature NMR experiments, and the diagnostic chemical shifts of (3) NNH, (4) C-NH, and (5) αNH.

Distinction 1: π–π Stacking.

To obtain crystals suitable for X-ray diffraction, dimethylamine was replaced with morpholine (leading to a shift in nomenclature from G-G to G′-G′, etc.) because these derivatives are more amenable to crystallization^26–28 and behave like the dimethylamine analogues in solution. That is, all adopt an (E)-hydrazone and show identical protonation sites, common NMR features, the same rotamer states, and folded structures. Probing the solution structure with morpholine derivatives using ¹H NMR spectroscopy is less desirable than using G-G, V-V, T-T, and I-I because the axial and equatorial protons of the morpholine ring are not degenerate and crowd the spectrum between 4 and 3 ppm whereas the methyl groups of dimethylamine appear as well-resolved singlets in an uncrowded region of the spectrum.

Figure 4 shows the crystal structures of G′-G′, V′-V′, and the isomers of I′-I′. In the folded conformation, the substituents on the α-C of the amino acid are oriented into axial and equatorial positions. In these crystal structures, the α-H occupies the axial position and the side chain occupies the equatorial position. For G′-G′, equatorial protons do not interact, and the triazine rings do not stack upon each other. Instead, each triazine ring stacks on the hydrazone. In V′-V′ and I′-I′, however, the side chains interact (presumably productively) leading to a translation of the π system from G-G wherein π–π stacking results in a greater overlap of the triazine rings. The likely origins of the isomerism in I′-I′ (and I-I) are revealed in the crystal structure, which contains two macrocycles in a unit cell (CCDC 2221530). One macrocycle comprises isoleucine residues that preserve the stereochemistry (S,S). The other macrocycle derives from alloisoleucine (R,S) resulting from epimerization of the α-carbon under the conditions of the synthesis. An additional crystal structure comprising just the (S,S) isomer has also been solved (CCDC 2233277).

Figure 4.

The crystal structures of an morpholine analogues G′-G′, V′-V′ and the two isomers of I′-I′ from the top and edge. The labels A, M1 and M2 correspond to specific hydrogen atoms that establish folding as well as subpopulation. The labels “a” and “e” correspond to axial and equatorial positions within the macrocycle. The isomers of I′-I′ derive from epimerization of the α-carbon.

This difference in stacking is reflected in rOesy data: the rOe network of G-G differs from that of the β-branched macrocycles. Both populations show a strong rOe between A and NNH, establishing the (E)-hydrazone. However, evidence for a folded structure is distinct. The macrocycle G-G shows rOes to M1 and M2. The β-branched macrocycles only show an rOe to M2. The crystal structures offer a compelling interpretation of this data. In the solid state, the π–π stacking of triazine-on-hydrazone in G′-G′ places proton A between M1 and M2 (circled in Figure 4). In V′-V′ and the isomers of I′-I′, however, triazine-on-triazine overlap is more substantial, leading to an increased distance between proton A and M1 precluding this rOe but conserving the rOe between A and M2. It is noteworthy that the rOe observed between A and M2 is stronger for the major isomer than the minor isomer, but crystal structure data are not compelling enough to justify a tentative assignment of these isomers.

Distinction 2: The Chemical Shift of H⁺.

The chemical shift of H⁺ differs across the subpopulations, appearing at 11.6 ppm for G-G and ~11.4 ppm for V-V and both isomers of I-I. While the source of the difference remains unclear, two hypotheses are noteworthy. The first hypothesis rests on the crystal structure data: the π system might influence the chemical shift of H⁺ because it appears positioned within the ring current of the triazine for V′-V′ and I′-I′ but more distant for G′-G′. The second (and favored) hypothesis rests on differences in dynamic motion that modulate the availability of the carbonyl to engage in hydrogen bonding. In the crystal structure of G′-G′, the carbonyl points outward toward solvating molecules bridging unit cells, while the carbonyls of V′-V′ and the isomers of I′-I′ point inward to engage in hydrogen bonding with H⁺. With the recognition that crystal packing forces can influence conformation, the upfield shift of H⁺ for V-V and the isomers of I-I is consistent with the increased electron density for H⁺.

In addition, variable-temperature NMR experiments reveal markedly different behavior among this series of macrocycles. We have reported that G-G moves between two folded, enantiomeric states⁴⁸ by way of an extended structure.⁴⁹ Importantly, these two states interconvert the axial and equatorial positions of α-H, B and C protons much like the ring-flip in cyclohexane. Accordingly, at high temperatures, single resonances are observed. At lower temperatures, this resonance decoalesces to reveal both the axial and equatorial positions. Neither coalescence or decoalescence has been observed across the available temperature range for the β-branched macrocycles leading us to posit that this motion is limited in these molecules, and accordingly, the hydrogen bond between H⁺ and the carbonyl is more faithfully maintained.

Variable-temperature (VT) NMR provides an additional distinction between these subpopulations. Protons involved in structure vis-è-vis hydrogen bonding are less sensitive to temperature changes than those that are solvent-accessible.⁵⁰ For proteins, a VT coefficient that is more negative than −4.5 ppb/degree K for an amide-NH conveys that it is exposed to the solvent and not engaged in the structure. Values more positive than −4.5 ppb/degree K convey that the amide is engaged in a hydrogen bond or that another structural feature shields it from the solvent. In evaluating the VT coefficients for these macrocycles, we abandon this numeric benchmark and refrain from making comparisons across proton types. However, restricting the discussion to protons of the same type (e.g., NNH) has value and reveals interesting trends.

Table 2 shows that the VT coefficients across the macrocycles are distinct by subpopulation. It is apparent that V-V, T-T, and I-I behave similarly and collectively different from G-G. This trend is consistent with the difference in the dynamic behavior reported. The exchangeable NHs yield the most consistent picture. Resonances of NNH, C–NH, and α-NH of G-G are more sensitive to temperature and, thus, less protected from the solvent than those same protons of the β-branched macrocycles.

Table 2.

Variable Temperature Coefficients (Δppb/ΔT) for Exchangeable Protons of the Macrocycles Reported in ppb^a

		G-G	T-T	v-v	I–I (maj)	I–I (min)
VT coefficients	NNH	−0.8	0.1	0.6	0.7	0.9
	H+	−2.0	−1.7	−1.7	−2.2	−2.2
	C-NH	−3.8	−1.8	−1.9	−2.0	−1.7
	α-NH	−1.5	0.2	0.3	0.3	0.3
	A	0.6	0.3	0.1	0.3	0.3
	OH	-	0.7	-	-

^aThe areas shaded blue and yellow convey distinguishing characteristics of the subpopulations. Green indicates an intermediate value.

Distinction 3: The Chemical Shift of NNH.

The subpopulations differ in the chemical shift of the NNH resonance in DMSO-d₆. It appears at 12.6 ppm for the β-branched macrocycles and at 12.4 ppm for G-G. This difference might be attributed to the extent to which the hydrazone lone pair contributes to hydrogen bonding, the influence of the π system, or a reflection of dynamic behavior and contributions of the carbonyl. While the molecular basis remains unknown, the chemical shift of NNH is a distinct signature of the subpopulation.

Distinction 4: The Chemical Shift of C-NH.

The subpopulations differ in the position of C-NH in DMSO-d₆ with G-G at 8.95 ppm and V-V and isomers of I-I at ~9.2 ppm. We hypothesize that the downfield shift for the β-branched macrocycles is consistent with the carbonyl hydrogen bonding to H⁺. Polarization of the carbonyl increases the π-character of the amide bond leading to a downfield shift of this amide NH for the β-branched macrocycles.

Distinction 5: The Chemical Shift of α-NH.

The α-NH proton shows the most significant difference when considering the two subpopulations. In DMSO-d₆, it appears downfield at 7.840 ppm for G-G but at 7.2 ppm for V-V and I-I. We infer that the α-NH σ bond of G-G contributes electron density to H⁺ to offset the diminished contributions of the carbonyl group.

The Structure of T-T.

Interestingly, T-T adopts elements of both subpopulations. Resonances for NNH and CNH align with V-V and the isomers of I-I as do the variable temperature coefficients reported. In contrast, the resonance for H⁺ aligns with G-G. The chemical shift of α-NH falls evenly between the two subpopulations. The data suggest that T-T adopts a conformation most similar to V-V and the isomers of I-I. However, stabilization of H⁺ by the carbonyl is significantly reduced—making T-T more like G-G. We hypothesize that the electron density from the carbonyl is reduced at H⁺ by competitive hydrogen bonding between it and the β-hydroxyl group. The chemical shift of the β-OH resonance (5.48 ppm) is consistent with hydrogen bonding.

The Free Energy Surfaces.

To conclude these studies, we employed well-tempered metadynamics (WT-MetaD)³³ to probe the free energy landscape. Classical molecular dynamics calculations can suffer from entrapment in local energy minima,^51,52 while WT-MetaD can sample configurations more than 2 kT higher in free energy. The two collective variables, D1 and D2 (Chart 1), were chosen to capture both extended and folded molecules. D1 is defined as the distance between the β-carbons of the β-branched amino acid or the αproton of glycine. D2 is the distance between protonated nitrogens. These distances become the ordinates used to generate the free energy surface (FES) shown in Figure 5.

Figure 5.

Free energy surfaces (left) and population plots (right) for the four macrocycles. Each contour of the FES corresponds to 1 kT. The circles reflect the data observed in the crystal structures. The relative populations of each minimum are indicated. The population comprises all structures within 3 kT of the local minima corresponding to red and orange domains

The regions on the left side of the FES plots (low D1) correspond to those structures with significant side chain–side chain interactions. Regions on the right are populated with structures that orient the side chains distant from one another. Distances derived from the crystal structures are indicated with circles.

The results are consistent with expectations: these macrocycles are flexible and can adopt many different conformations within a few kT of each other, a situation ideal for the design of molecular chameleons. The FES of G-G shows multiple shallow minima with regions between them populated with other structures of similar energy. In contrast, the FES of the β-branched macrocycles has three well-defined minima.

The FES also shows that increasing steric bulk reduces the size of the conformational footprint: G-G (4.1 nm²) > V-V (2.9 nm²) ~ T-T (2.8 nm²) > I-I (2.3 nm²). In terms of percent differences relative to G-G (100%), the area of V-V is 69% of G-G while T-T at 68%. Both exceed that of I-I at 55%.

These FESs can be used to calculate the relative probabilities $P\left(r\right)$ to determine populations of structures using eq 1.

$$P\left(r\right)\propto e^{-G\left(r\right)/kT}$$ (1)

The resulting probability maps are shown in Figure 5 (right). The populations indicated are based on including structures within 3 kT of the minima (the red and orange domains of the FES). The shallow FES of G-G is reflected in the population (15%) that exists on the surface not defined by local minima. These populations cannot readily relate to specific structures observed in solution, only the D1 and D2 coordinates, because each minimum is populated by a range of different structures ranging from those that are compact with significant π–π overlap to those that are more extended.

CONCLUSIONS

The data derived from the solution and the solid state provide a clear picture of attributes conserved across this molecular framework. From a 1D ¹H NMR spectrum and the measurements of rOe correlations with A, a nuanced determination of conformation can be obtained that allows assignment of specific subpopulations. Table 3 summarizes the spectroscopic features and the elements reported.

Table 3.

Comparison of Features of the Two Structural Subpopulations of This Molecular Scaffold

characteristic	V-V subpop.	G-G subpop.	reports on...
δ 1H NNH (PPm)	12.62	12.41	NNH hydrogen bonding
δ 1H H+ (ppm)	11.36	11.63	C=O orientation/dynamics; π−π stacking
δ 1H C-NH (ppm)	9.22	8.94	C=O hydrogen bonding/dynamics
δ 1H A (ppm)	7.5		macrocyclization
δ 1H α-NH (ppm)	7.18	7.84	hydrogen bonding; π−π stacking
rOe A to NNH	strong		(E)-hydrazone
rOe A to Ml		present	folding; π−π stacking of triazine-on-hydrazone
rOe A to M2	present		folding; π−π stacking of triazine-on-triazine
δ 13C C=O (ppm)	172.9	171.4	hydrogen bonding of CO to H+
δ 13C A (ppm)	148.0	147.9	(E)-hydrazone

The lessons learned have clear value as we consider strategies for drug design wherein small differences in shape can greatly influence physicochemical properties. Indeed, we propose that the 24-atom macrocycle represents a molecular framework amenable to substitution of the α-amino acid without loss of conformation. Functionalization at the α-carbon places groups more proximate than functionalization at the auxiliary positions of the triazine (e.g., replacing dimethylamine or morpholine). Additionally, the choice of glycine or the β-branched alternatives will influence the dynamic motion affording the framework. Probing the generality of this model is currently in progress.

In addition, these studies contribute to a better understanding of these macrocycles, which are “stable” in a descriptive sense. They can be stored at room temperature in organic solvents and acidic water without signs of hydrolysis. They survive conventional silica gel chromatography. Ongoing efforts address the mechanistic basis for the quantitative cyclization (a result that we hypothesize to be due to templating through the formation of a network of hydrogen bonds), the chiral sorting that appears to be occurring and the reaction conditions required to minimize racemization.⁴¹

Data Availability Statement

All the data needed to reproduce our calculations can be found in the linked Zenodo repository https://doi.org/10.5281/zenodo.7575450.