Drug-like properties in macrocycles above MW 1000: Backbone rigidity vs. side-chain lipophilicity

Abstract

Large, macrocyclic peptides can achieve surprisingly high membrane permeability, although the properties that govern permeability in this chemical space are only beginning to come into focus. We generated two libraries of cyclic decapeptides with stable, cross-β conformations, and found that peptoid substitutions within the β-turns of the macrocycle preserved the rigidity of the parent scaffold, whereas peptoid substitutions in the opposing β-strands led to “chameleonic” species that were rigid in nonpolar media but highly flexible in water. Both rigid and chameleonic compounds showed high permeability over a wide lipophilicity range, with peak permeabilities differing significantly depending on scaffold rigidity. Our findings indicate that modulating lipophilicity can be used to engineer favorable ADME properties into both rigid and flexible macrocyclic peptides, and that scaffold rigidity can be used to tune optimal lipophilicity.

Introduction

The discovery and validation of new therapeutic targets is outpacing our ability to find lead compounds against them^[1]. Many of these targets are protein-protein interactions (PPIs), which lack the binding pockets found in enzymes and receptors and are thus considered “undruggable” by conventional small molecules. Cyclic peptides can rival biologics in terms of potency and specificity, even against challenging targets such as PPIs^[2]. Although the size and polarity of most cyclic peptides fail to meet the criteria of Lipinski’s “Rule of 5” for predicting drug-likeness, a growing number of cyclic peptides have been described that exhibit the favorable ADME properties of small molecule drugs, including high passive cell permeability and even oral bioavailability (BA). These exceptional cases, which include natural products such as cyclosporine A (CSA) and griselimycin^[3], in addition to a variety of synthetic compounds^[4], support the idea that macrocycles may provide a “middle way” in the pursuit of challenging intracellular targets^[5].

Achieving permeability in cyclic peptides is far from straightforward, however. Investigations in model systems and natural products have yielded insights into structure-property relationships in cyclic peptides, providing some guideposts for the design of novel macrocyclic scaffolds with favorable drug-like properties^[6]. Backbone stereochemistry, amide N-methylation, and the presence and position of non-proteinogenic residues such as N-alkyl glycines (peptoids), can greatly impact membrane permeability by influencing the degree of exposure of polar NH groups through direct capping, local steric occlusion, or by stabilizing intramolecular hydrogen bond (IMHB) networks^{[4a, 6–7]}.

It has also been observed that large, orally absorbed drugs such as CSA often have “chameleonic” properties, being capable of adopting nonpolar conformations in low dielectric environments (e.g., the cell membrane) while also adopting more polar conformations in aqueous solution^[8]. It has been hypothesized that such environment-dependent flexibility may be necessary to reconcile the opposing requirements for membrane permeability and aqueous solubility in the “beyond-Rule of 5” (bRo5) chemical space^[9]. While these inquiries into the importance of flexibility have been based primarily on surveys of existing drugs, here we describe a complementary approach using an experimental system that treats chameleonicity itself as a controllable variable.

Fouché, et al., recently reported a highly membrane permeable cyclic decapeptide which showed that favorable ADME and PK properties can be achieved in synthetic cyclic peptides with molecular weights above 1000 Da (Figure 1a)^[10]. Peptoids, whose side chains are derived from primary amines, greatly extend the range of chemical functionality that can be introduced into a cyclic peptide. Encouraged by previous studies showing that cyclic peptide-peptoid hybrids (peptomers) can exhibit favorable ADME properties,^[11] we set out to determine whether permeability could be maintained in the Fouché scaffold upon substitution with peptoid residues. Serendipitously, we found that peptoid substitutions in these macrocycles had a specific effect on conformational flexibility, such that solvent-dependent chameleonicity could be tuned depending on the location of the peptoid substitutions within the scaffold. These cyclic peptomers demonstrate that chameleonicity is not a requirement for achieving favorable ADME and PK properties and provide a model system for probing the effect of backbone rigidity on molecular properties in bRo5 macrocycles.

Figure 1.

a) Original cyclic decapeptide scaffold published by Fouché, et al. in 2016. Compound 1 was a side chain variant with exceptional oral bioavailability (>100%); b) Peptoid (X¹ and X²) and peptide (X³) building blocks used in the synthesis of Libraries A and B; c) and d) Designs for cyclic peptomer Libraries A (c) and B (d) showing diversity positions X¹-X³.

Results and Discussion

We designed two cyclic peptomer scaffolds based on a series of highly cell permeable and orally bioavailable cyclic decapeptides^[10]. Variants of this scaffold such as 1 (Fig. 1a) are similar to the natural product gramicidin S in which solvent-exposed amides are capped with N-methyl groups. N-methylation eliminates the desolvation penalty associated with exposed backbone amides while preserving the intramolecular hydrogen bonding network of the natural product. The Fouché peptide also shares the same pattern of stereocenters and overall architecture as the previously described cyclic hexapeptide 1NMe3^[4c]. Compared to 1NMe3, 1 has a 4-residue extension in the central -strands, thus creating an extended array of transannular hydrogen bonds. A variety of N-Me-to-peptoid substitutions in 1NMe3 and related scaffolds have been described which preserve (or in some cases even enhance) the cell permeability of the parent scaffold^{[11b, 11c, 12]}.

In order to investigate the effect of NMe-to-peptoid substitutions in 1, we designed two libraries: One in which the two Phe(NMe) residues in the opposing β-turns are replaced with peptoids (Library A, Fig. 1 b), and one in which the two central Ala(NMe) residues are replaced with peptoids (Library B, Fig. 1c). In each library, the peptoid side chains (X¹, X²) were selected from simple aliphatic and ether-containing amines (N1-N6, Fig. 1d), the X³ amino acid was either Phe or Ala, and the remaining residues were kept the same as in the parent scaffold, giving rise to 18 compounds for each library. Since permeability depends not only on backbone geometry, but also on gross, solvent-exposed lipophilicity (i.e., at the side chain level), the two libraries were designed to sample a wide lipophilicity range, with calculated octanol/water partition coefficients (ALogP) from 0.4 to 4.7. The peptoid R-groups and amino acid side chains were selected so that all library members have unique masses to allow deconvolution of the mixtures by LCMS^[13].

Libraries A and B were generated separately as mixtures using solid phase split-pool peptide synthesis, incorporating peptoid side chains as primary amines using the submonomer method^[14]. The libraries were cleaved, cyclized in solution, and purified by solid phase extraction. PAMPA permeabilities within each library varied parabolically as a function of ALogP, consistent with previous observations (Fig. 2a)^{[4f, 15]}. Permeabilities of both Libraries A and B peaked at ~10 × 10⁻⁶ cm/s; however, while the permeability of Library A (Fig. 2a, green markers) peaked at the polar end of the sampled lipophilicity range, with high permeabilities as low as ALogP = 0.4, the permeability of Library B (Fig. 2a, blue markers) peaked at the higher end of the range, near ALogP ~ 3.8.

Figure 2.

a) Relationship between log of effective permeability coefficient (LogPe) and calculated (atomistic) octanol/water partition coefficient (ALogP) for libraries A (green) and B (blue). Marker size represents % recovery in PAMPA. b) Hydrocarbon/water distribution coefficient (pH 7.4) as a function of ALogP for Libraries A (green) and B (blue). c) and d) Log of PAMPA intrinsic permeability coefficient (LogP0), calculated by correcting LogPe for the fraction of each species in aq. solution as determined using a filtration assay, as a function of ALogP for c) Library A and d) Library B. Each value represents the average of three replicates. See Supporting Information for standard deviations.

Previously we reported a metric called lipophilic permeability efficiency (LPE), which quantifies the efficiency with which a compound achieves passive membrane permeability at a given ALogP-defined lipophilicity^[15]. LPE is derived from the difference between the experimental hydrocarbon-water partition coefficient, LogD(dd/w), which reflects a molecule’s net hydrogen bond acidity in its membrane-associated conformation^[16], and ALogP, which reflects a compound’s minimum lipophilic character in the aqueous environment^[15]. LPE, defined specifically by the equation LPE = LogD(dd/w) - 1.06ALogP + 5.47, represents a scaffold’s intrinsic membrane permeability by normalizing its membrane partitioning against its gross lipophilicity.

Both aliphatic side chain variants on the same scaffold, as well as different scaffolds with the same net hydrogen bond acidity, have similar LPE values and fall on the same 45° line in the plot of LogD(dd/w) vs. ALogP. The average LPE for Libraries A and B are similar, at 3.58 (s.d. = 0.57) and 3.13 (s.d. = 0.55), respectively, suggesting that both scaffolds achieve similar degrees of IMHB in low dielectric environments (Fig. 2b). These LPE values are close in magnitude to those determined previously for a series of 1NMe3 derivatives (LPE ~ 3.7), which have a similar backbone structure and IMHB pattern.

We have also shown previously that on the ascending (polar) portion of the ALogP vs. permeability curve, the correlation between LogD(dd/w) and membrane permeability is high, whereas the correlation breaks down on the descending portion of the curve where solubility declines sharply (Fig. 2a). This observation underscores the fact that distinct physical phenomena govern the relationship between lipophilicity and permeability in the soluble vs. insoluble regimes, with the correlation between permeability and LogD(dd/w) being strong in the soluble regime but weak in the insoluble regime. Consistent with these observations, the correlations between LogD(dd/w) and LogPe are significantly different for Libraries A and B (Fig. 2c,d). For Library A, the correlation between LogD(dd/w) and LogPe is negligible (Fig. 2c)(R² = 0.16), whereas for Library B the correlation is quite strong (Fig. 2d)(R² = 0.93). These data are consistent with the plots of LogPe vs. ALogP for Libraries A and B, in which most of the Library B compounds fall on the polar, ascending half of the curve, while nearly every Library A compound falls on the insoluble, descending half. Furthermore, most Library B compounds show good recovery in PAMPA even up to ALogP = 4 (Fig. 2a, marker size), while recoveries for Library A fall off significantly above ALogP ~ 2. These observations, and the 2-log unit separation between the ALogP vs. permeability curves for Libraries A and B, suggest that the position of the peptoid substituents confers distinct physical properties to the two scaffolds.

To test whether, like the all-peptide parent scaffold, Libraries A and B sequester their backbone NH groups in IMHB, we synthesized the most permeable representatives of each scaffold, A09 and B08 (Fig. 3), and investigated their NMR temperature shifts in solvents of varying polarity. In CDCl₃, A09 and B08 showed equally low temperature coefficients for all four amide NH groups, consistent with the similarity in their LPE values (Fig. 4a). In polar solvents, however, the two scaffolds exhibited very different behaviour. In both DMSO and H₂O/ACN, A09 maintained a stable conformation, showing only a modest increase in NH temperature shift coefficients with increasing polarity. In contrast, the amide signals of B08 split into multiple families in DMSO, and devolved into a mixture of greater complexity in H₂O/ACN (Fig. 4a). The NMR spectra thus indicate that in both low and high-dielectric environments, A09 is almost completely locked into its “closed” conformer, whereas B08, while also adopting a fully closed conformation in CDCl₃, exhibits significant conformational heterogeneity in higher polarity solvents. Scaffold A thus behaves more like its parent scaffold 1, which, according to a recent molecular dynamics study, adopts a rigid conformation in both chloroform and water^[18]. In contrast, scaffold B behaves more like cyclosporine A (CSA), the prototypical “chameleonic” cyclic peptide that, while rigid and closed in chloroform, adopts multiple, open conformers in water^[19].

Figure. 3.

Structures of 6 compounds from Libraries A and B that were synthesized and tested individually. The 6 compounds were chosen as three pairs of isomers with a range of lipophilicities (ALogP = 0.4, 1.9, and 3.7).

Figure 4.

(a) NMR temperature shift results for A09 and B08; each spectrum represents a 5°C shift in temperature from 25°C (bottom) to 50°C (top). b and c). McMD results for A09 and B08; b) principle components PC-1 and PC-2 of the conformational ensembles of A09 and B08 in solvents of increasing polarity; c) Hydrogen bond plot showing percentage of the ensemble with the hydrogen bonds shown (the pattern shown represents the canonical cross-β transannular hydrogen bonds found in the parent structure).

To confirm that scaffolds A and B have different conformational stabilities and that the ensemble observed for B08 in higher dielectric solvents represents a variety of more open conformational states, we investigated their conformations using multicanonical MD (McMD) simulations^[20]. We have used this approach to investigate the dynamics of cyclic hexapeptides and have shown that McMD provides a useful method for mapping the conformational space of cyclic peptides and predicting properties based on the resulting ensembles^[21]. McMD simulations of A09 and B08 in explicit solvents ranging in polarity from chloroform to water showed that both compounds indeed adopt compact folds in chloroform (Fig. 4b). However, while A09 maintains its cross-beta structure to a large extent even in water, the hydrogen bond network of B08 becomes increasingly destabilized with increasing solvent polarity, consistent with the NMR data (Fig. 4b,c). The scaffolds representing Libraries A and B thus provide a means of comparing chameleonic and non-chameleonic compounds in the bRo5 macrocycle space using compounds that otherwise share identical compositions and calculated properties.

To further elucidate the structure-property relationships of the two scaffolds, we synthesized additional representatives of each library, generating a total of three matched pairs that span a wide lipophilicity range: A03/B03 (ALogP = 0.4); A09/B09 (ALogP = 1.9); and A08/B08 (ALogP = 3.7) (Fig. 3). The members of each pair have the same composition, differing only in the position of the peptoid residues. For each of the six compounds, we determined aqueous solubility, PAMPA and MDCK permeabilities, P-gp efflux ratios, microsomal stability, and serum protein binding (Tables 1 and S2). Aqueous solubility tracked roughly with ALogP for both scaffolds: At pH 6.8, the most polar pair (A03 and B03) were very soluble (1900 and 1900 mg/mL, respectively), while the most lipophilic pair (A08 and B08) were orders of magnitude less soluble (7 and 210 mg/mL, respectively), with the chameleonic isomer B08 being 30 times more soluble than its rigid counterpart. For the pair in the middle at ALogP = 1.9, the chameleonic isomer B09 was also significantly more water-soluble than its rigid counterpart A09 (2200 vs. 640 mg/mL, respectively)(Table 1).

Table 1.

In vitro ADME data on pure compounds.

Cpd.	ALogP	aq. sol. pH 1.2 (μg/mL)[a]	aq. sol. pH 6.8 (μg/mL)[a]	LodgD (oct/w)[b]	PAMPA (× 10−6 cm/s)[c]	MDCK (× 10−6 cm/s)[c]	P-gp efflux ratio[c]	microsomal stability (%)[b]	serum protein binding (%)[b]
A03	0.4	1800	1900	2.9	17.2	0.4	6.4	64	69
A09	1.9	590	640	> 4.1	35.7	3.1	46.9	67	98
A08	3.7	2.5	6.8	> 4.4	18.8	24.7	8.5	33	> 99.8
B03	0.4	2000	1900	2.1	< 0.1	0.3	1.4	94	57
B09	1.9	2100	2200	3.6	1.4	0.4	10.3	90	68
B08	3.7	210	210	>4.8	> 50	15.3	19.2	53	97

^[a]N = 1.

^[b]N = 2.

^[c]N = 3.

(See Supporting Information for individual values where N = 2 and standard deviations where N = 3).

PAMPA permeability studies on the six pure compounds were performed initially under the same conditions as those used in the original library screen, using a concentration of 1 μM in the donor well based on the estimated concentration of each compound in the original library. Under these conditions, no compound was observed in the acceptor well for any of the six compounds, yielding P_app values of < 0.01 × 10⁻⁶ cm/s, in stark contrast to the permeabilities of the same compounds observed in their original mixtures. We hypothesized that the pure compounds were precipitating in the donor well or adhering to the walls of the apparatus, and that in mixtures, other compounds were acting as blocking agents to help prevent aggregation or adsorption. Therefore, we re-investigated the PAMPA permeabilities of the pure compounds using detergents in the donor and acceptor wells as “sink” conditions. Using these conditions, the permeability trends were similar to those observed in the libraries. At the polar end of the continuum, A03 (18 × 10⁻⁶ cm/s) was much more permeable than B03 (< 0.1 × 10⁻⁶ cm/s), whereas at the lipophilic end this trend was reversed, with B08 (> 50 × 10⁻⁶ cm/s) being more permeable than A08 (18 × 10⁻⁶ cm/s)(Table 1). Interestingly, the PAMPA permeabilities of the scaffold A compounds were high across the entire ALogP range, while the permeabilities of scaffold B trended upward along with ALogP.

The MDCK permeabilities roughly followed the same trends observed in the PAMPA studies, with some differences due, most likely, to the high P-gp efflux ratios for many of them. For both scaffolds, increasing ALogP led to lower microsomal stability and higher plasma protein binding, although compounds from the more rigid Library B were, on average, more metabolically stable and less protein bound than compounds from the more chameleonic Library A.

Based on their permeabilities in PAMPA and MDCK cells, we investigated the in vivo pharmacokinetics of 5 of the 6 pure library members in mice (B03 was not included in the oral PK study due to its low PAMPA and MDCK permeabilities). Oral absorption was low for all 5 compounds (Table 2), a result that was not entirely unexpected given the compounds’ high efflux ratios and generally poor microsomal stabilities. To test whether metabolism and efflux were the main factors contributing to the low oral BA, mice were given an initial dose of the CYP inhibitor 1-aminoazabenztriazole (ABT), the P-gp inhibitor elacridar (GF120918), or a combination of the two inhibitors. Both CYP and P-gp inhibition significantly increased oral bioavailability for all 5 compounds, and for the three Library A compounds, P-gp and CYP inhibition had a synergistic effect. Combining the inhibitors also produced oral bioavailabilities that reflected the compounds’ relative PAMPA permeabilities. For A03, even in the absence of CYP or P-gp inhibition, oral BA increased as a function of dosage, from F = 1% at 3 mg/kg to F = 50% at 200 mg/kg (Table S5). Such significant dosage-dependent oral BA indicates saturation of metabolic and efflux processes and is consistent with the observed increase in oral BA in the presence of CYP and P-gp inhibitors.

Table 2.

Oral bioavailabilities in mouse, with no inhibitors, or pretreated with CYP and P-gp inhibitors. The mean value from two independent experiments is shown in bold, with the two individual values given in parentheses.

Cpd.	No inhib. (%)	+ CYP inhib.[a] (%)	+ P-gp inhib.[b] (%)	+ CYP+ P-gp inhib. (%)
A03	1 (1/1)	2 (2/2)	4 (5/3)	39 (39/40)
A09	2 (2/3)	5 (3/6)	22 (25/19)	105 (97/114)[c]
A08	5 (5/5)	31 (31/32)	7 (7/8)	59 (62/57)
B03	-[d]	-[d]	-[d]	-[d]
B09	1 (1/1)	2 (1/2)	3 (3/2)	14 (14/13)
B08	6 (6/7)	148 (147/149)[c]	13 (12/14)	101 (103/99)[c]

^[a]1-aminobenztriazole (ABT), cytochrome P450 (CYP) inhibitor.

^[b]GF120918 (GF), P-glycoprotein (P-gp) inhibitor.

^[c]Oral BA values of >100% are likely due to higher clearance in i.v. compared to oral routes as a result of saturable intestinal transporters^[23].

^[d]B03 was not included in the oral PK study due to its low PAMPA and MDCK permeabilities.

For the most lipophilic compounds, A08 and B08, inhibition of CYP-mediated metabolism caused a greater increase in oral BA than inhibition of P-gp efflux. For the more polar compounds, P-gp inhibition had a more significant effect, and the synergy between CYP and P-gp inhibition was also greater than for the more lipophilic compounds. Consistent with their observed PAMPA permeabilities, the oral BAs of the scaffold A compounds (with CYP and P-gp inhibition) were remarkably high across the entire ALogP range, with a maximum BA of >100% for A09 in the middle of the range (oral BA values of greater than 100% have been observed for compounds that exhibit high efflux ratios^[23]). Also consistent with the permeability trends observed for scaffold B, oral BA was at a maximum at the high end of the ALogP range (compound B08). Thus, to a large degree the oral bioavailabilities of these compounds reflect their in vitro ADME behaviour when efflux and metabolism are factored in.

Conclusion

The results described here show that high permeability and good aqueous solubility can be achieved in both rigid and chameleonic macrocyclic scaffolds, even with molecular weights above 1000 Da. Members of the more rigid scaffold were permeable at a much lower bulk lipophilicity (ALogP) than was required of the more flexible scaffolds. For chameleonic macrocycles such as those defined by scaffold B, permeability could be achieved, albeit at lipophilicities that exceeded the solubility limit of their more rigid counterparts.

CSA and griselimycin represent a diverse class of cyclic peptide natural products that engage intracellular targets with high affinity and, despite their large size and peptidic nature, exhibit good passive permeability and oral bioavailability. These compounds generally have a preponderance of aliphatic residues and few if any side chains that contain functional groups. These compounds interact with diverse protein targets through a combination of van der Waals contacts mediated by their side chains, along with polar contacts which are made almost exclusively with backbone C=O and NH groups. Model systems have shown that even a single side chain bearing a strongly ionizable group or solvent-exposed hydrogen bond donor can diminish permeability by orders of magnitude, suggesting that this class of natural products may provide useful design principles for generating cell permeable cyclic peptides capable of binding novel targets with high affinity. While the number of commercially available amino acids containing no ionizable groups or hydrogen bond donors is relatively low, the available amines (i.e., peptoid monomers) that fit this description number in the thousands. These include many types of heterocycles, substituted aromatics, ethers, and even tertiary amines, which could, in principle, make polar contacts with a target without significantly diminishing permeability.

Flexibility may therefore be desirable for achieving high affinity against a given biological target by allowing the backbone to remain hidden in the membrane while adopting more open states in water that can form hydrogen bonds with residues in the target^{[8a, 22]}. The small conformational penalty associated with the membrane partitioning of scaffold B also points toward a small entropic penalty that might be incurred upon target binding, even if the target-bound state is among the many “open” conformers observed experimentally and computationally in water.

Model systems such as the two isomeric scaffolds described here continue to provide insight into the physico-chemical requirements for achieving drug-like properties in the bRo5 chemical space. In this study we have learned that environment-dependent chameleonicity such as that observed in the classic bRo5 drug CSA can be modulated by strategic backbone substitutions. Moreover, although water solubility, CYP metabolism, plasma protein binding, and passive membrane permeability correlate strongly with a scaffold’s chamelonicity, achieving favorable ADME and PK properties is possible for both rigid and flexible scaffolds.