Going Viral: An Investigation into the Chameleonic Behaviour of Antiviral Compounds

Abstract

The ability to adjust conformations in response to the polarity of the environment, i.e. molecular chameleonicity, is considered to be important for conferring both high aqueous solubility and high cell permeability to drugs in chemical space beyond Lipinski's rule of 5. We determined the conformational ensembles populated by the antiviral drugs asunaprevir, simeprevir, atazanavir and daclatasvir in polar (DMSO‐d ₆) and non‐polar (chloroform) environments with NMR spectroscopy. Daclatasvir was fairly rigid, whereas the first three showed large flexibility in both environments, that translated into major differences in solvent accessible 3D polar surface area within each conformational ensemble. No significant differences in size and polar surface area were observed between the DMSO‐d ₆ and chloroform ensembles of these three drugs. We propose that such flexible compounds are characterized as “partial molecular chameleons” and hypothesize that their ability to adopt conformations with low polar surface area contributes to their membrane permeability and oral absorption.

NMR spectroscopy revealed three antiviral drugs in the bRo5 space to behave as “partial” molecular chameleons. Their ability to adopt conformations with low polar surface area is proposed to improve their cell permeability and contribute to oral absorption. A fourth drug was found not to be chameleonic.

Introduction

The Covid‐19 pandemic serves as reminder that viral infections have plagued mankind throughout history. The launch of drugs harnessing the human immunodeficiency virus (HIV) in the 1990s, followed by ones curing hepatitis C virus (HCV) infections in the 2010s, are powerful examples of our recently acquired ability to successfully combat viral pandemics.[ ¹ , ² , ³ ] Both infections have proven challenging to treat as the two viruses have high mutation rates, which quickly can lead to drug resistance. ^[4] For HCV, several genotypes and many subtypes, all with different susceptibilities towards drugs have increased the hurdle to develop efficacious drugs. ^[1] However, by using combinations of drugs directed towards different targets in the two viruses, both pandemics are now under effective control, and there is even hope to eradicate HIV within the coming decade. ^[5]

The efforts to find treatments for HIV and HCV have stimulated drug discovery for targets that are difficult to drug with conventional small molecule ligands, for example targets that have large, featureless and/or flexible binding sites.[ ⁶ , ⁷ , ⁸ ] The NS3/4A protease and the NS5A protein of HCV constitute prominent examples of difficult‐to‐drug targets, just as the HIV protease. The NS3 serine protease, assisted by the co‐factor NS4A, is essential for HCV to process its polyprotein chain. ^[9] The protease has a chymotrypsin‐like fold and is composed of two domains. However, chymotrypsin has several loops that shape the substrate binding groove which are missing for the NS3/4A protease, leaving the NS3/4A active site rather featureless, shallow and non‐polar. ^[1] The HCV NS5A protein is involved in RNA replication, modulation of host‐cell responses and assembly of viral particles.[ ¹⁰ , ¹¹ ] This protein forms dimers upon crystallization, the biological relevance of which is unknown, although it is speculated that the dimerized form is the immature form of NS5A. The fact that drugs inhibiting NS5A are symmetric dimers, or have dimer‐like structures, indicates that stabilization of the dimeric form of NS5A interferes with key aspects of HCV expression and regulation. ^[10] The HIV aspartyl protease is a homodimer in which a channel between the two monomers forms the active‐site. This active site is highly flexible and spans a relatively large area, making it difficult for small molecules to reach all important interactions required for potent inhibition. ^[7] Macrocycles have received increasing interest during the past decade because of their unique ability to modulate targets that have flat and featureless binding sites, such as those of protein‐protein interactions and the NS3/4A protease.[ ⁶ , ¹² , ¹³ , ¹⁴ ] However, even though macrocycles have advantages as ligands for some difficult‐to‐drug targets, non‐macrocyclic drugs are also frequent as they may provide potent inhibitors for large, groove and pocket shaped sites.[ ⁶ , ¹⁵ ]

Historically, small molecule drugs have been designed to adhere to Lipinski's rule of five (Ro5). ^[16] According to the Ro5, drugs are more likely to display satisfactory oral absorption when found in a chemical property space defined by a molecular weight (MW) ≤500 Da, a number of hydrogen bond donors (HBDs) and acceptors (HBAs) ≤5 and ≤10, respectively, and a calculated lipophilicity (cLogP) ≤5. Achieving high oral absorptions requires optimization of multiple properties including solubility, cell permeability and metabolic stability, while still maintaining potent drug‐target binding. During recent years it has become apparent that discovery of ligands for targets that have difficult‐to‐drug binding sites, such as the HCV and HIV proteases, often requires ventures into chemical space beyond that of traditional small molecule drugs, i.e. into beyond Ro5 (bRo5) space.[ ⁶ , ¹⁷ , ¹⁸ , ¹⁹ ] One hypothesis that explains how drugs in bRo5 space can combine potent target binding with satisfactory solubility, cell permeability and oral absorption is that they behave as molecular chameleons.[ ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ ] This hypothesis suggests that bRo5 drugs benefit from possessing an appropriate flexibility that enables them to shield polar moieties from the surrounding environment when passing the cell membrane and expose them again when entering an aqueous environment. ^[26] A large number of studies have revealed that cyclosporin A behaves as a molecular chameleon, ^[27] while studies of the chameleonicity of some other drugs in bRo5 space performed by analysis of crystal structures or using NMR spectroscopy have only been reported recently.[ ²¹ , ²² ] A combination of properties may be used to assess chameleonic behaviour. These include the flexibility of the compound as described by the number of rotatable bonds (NRotB) ^[28] and the Kier's flexibility index (Φ), ^[29] the latter of which is more relevant for macrocycles. ^[30] The ability to form dynamic intramolecular hydrogen bonds (IMHBs),[ ²² , ³¹ ] as well as the variation of solvent‐accessible 3D polar surface area (SA 3D PSA),[ ²¹ , ²² , ³² ] lipophilicity [LogP (MLP)] ^[33] and size, as estimated by the radius of gyration (R_gyr),[ ²¹ , ³² ] between biologically relevant conformations are other important descriptors of molecular chameleons.

Herein, we have used NMR spectroscopy to determine the conformational ensembles in polar and non‐polar environments of four de novo designed antiviral drugs targeting the HCV NS3/4A and HIV proteases, as well as the NS5A protein of HCV. Analysis of the drugs' solution ensembles provided insight into the extent to which they behave as molecular chameleons. As this is the first study to investigate the conformational behaviour of de novo designed drugs in bRo5 space we have also compared our results to those obtained recently for four natural‐product derived antibacterial agents. ^[21]

Results and Discussion

The drug set

The four antiviral drugs chosen for this study are important components of combination treatments for HCV and HIV infections (Figure 1A).[ ¹⁸ , ²⁶ ] Asunaprevir ^[34] and simeprevir ^[35] are structurally diverse peptidomimetics that inhibit the HCV NS3/NS4A protease[ ³⁶ , ³⁷ ] by adopting different binding modes in its active site.[ ³⁴ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ ] Asunaprevir is non‐macrocyclic, while simeprevir has a macrocyclic core comprised of 14 atoms. The peptidomimetic atazanavir ^[42] inhibits the HIV type 1 group M subtype B protease.[ ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ ] Daclatasvir ^[47] is an inhibitor of the NS5A protein of HCV developed from a hit obtained in a phenotypic screen.[ ⁴⁵ , ⁴⁶ ] All four antivirals are cell permeable and orally bioavailable[ ⁴¹ , ⁴⁸ , ⁴⁹ ] while having a molecular weight (MW) between 700 and 750 Da, i.e. above the Ro5 guideline (Figure 1B). ^[16] Hydrogen bond donor (HBD) and acceptor (HBA) counts obey the Ro5, just as the calculated lipophilicities (cLogP), except for that of daclatasvir. However, the topological polar surface area (TPSA) of all four drugs exceeds the upper limit of Veber's rule,[ ²⁸ , ⁵⁰ ] just as the number of rotatable bonds (NRotB) of the three non‐macrocyclic drugs. Kier's flexibility index (ϕ) is a descriptor that provides a better comparison of the flexibility of macrocyclic and non‐macrocyclic compounds than NRotB, which does not account for the flexibility of rings.[ ²⁹ , ³⁰ ] It indicates that the macrocyclic simeprevir has a similar flexibility as asunaprevir and daclatasvir, with atazanavir being more flexible (Figure 1B).

Figure 1.

A) Structures of the four antiviral drugs for which conformational ensembles were determined in CDCl₃ and DMSO‐d ₆. The peptide backbone is highlighted in purple for all four drugs and the macrocyclic core of simeprevir that is not part of the peptide backbone is highlighted in bold. The IUPAC numbering of the amide protons are given in the structures, full numbering can be found in Figures S1–4. B) Descriptors calculated from the structures of the four drugs using MOE (version 2019.01) compared to the upper limits of the guidelines of Lipinski's rule of 5 and Veber's rule.[ ¹⁶ , ²⁸ , ²⁹ ] MW, molecular weight; HBD, hydrogen bond donor; HBA, hydrogen bond acceptor; cLogP, calculated distribution between octanol and water; TPSA, topological polar surface area; NRotB, number of rotatable bonds; ϕ, Kier flexibility index.

As revealed by the descriptors of the Ro5 and Veber's rules, and as further illustrated by a principal component analysis of the orally administered drugs in DrugBank, ^[51] the four antivirals reside in bRo5 chemical space (Figure 2). They are located somewhat closer to Ro5 space than the four antibacterial agents rifampicin, roxithromycin, telithromycin and spiramycin; the latter three of which we recently found to behave as molecular chameleons. ^[21] The immunosuppressant cyclosporin A, the chameleonicity of which is considered to be essential for its oral absorption, resides at the very outskirts of oral druggable space. The location of the four antivirals in chemical space, close to the four antibacterials but well separated from cyclosporin A and the Ro5 compliant drugs, provides an important driver for investigating if the antivirals behave as molecular chameleons and if chameleonicity may play a role for their cell permeability and oral absorption.

Figure 2.

Principal component analysis (PCA) of the chemical space of the orally administered drugs in DrugBank ^[51] (n=984). The PCA was based on the descriptors of Lipinski's and Veber's rules,[ ¹⁶ , ²⁸ ] as well as on descriptors that representing atom and bond counts, Kier and Hall connectivity, kappa shape indices, and pharmacophore features. ^[52] Drugs that adhere to all of the descriptors of the Ro5 and Veber's rule are shown as blue circles. The four antiviral drugs investigated herein are indicated by olive‐green circles, while the four antibacterial agents studied recently by us are in green and blue. ^[21] The position of cyclosporin A is shown as a violet circle, while the remaining drugs outside of chemical space defined by the Ro5 and Veber's rule are in red. Outliers have been identified with their names in grey; those in italics have low (<1 %) or no oral bioavailability, while that of saquinavir is 4 % (without ritonavir boosting). The first two principal components explain 59 % of the variance in the data set. The contribution of the most important descriptors used to characterize the drugs to each of the principal components is indicated by their position on each of the axes.

Methods

We used NMR spectroscopy to determine the solution ensembles of the four antiviral drugs in chloroform and DMSO‐d₆ . Chloroform has a dielectric constant (ϵ=4.8) close to that of the interior of a cell membrane (ϵ=3.0) ^[53] and ensembles determined in this solvent have provided useful insight into cell permeability.[ ²¹ , ⁵⁴ ] All four antivirals have a very low aqueous solubility (<20 μM for all but asunaprevir, Table 1) which reflects that their lipophilicities are at the high end of the drug‐like range (LogD 3–4, Table 1). ^[19] As all were soluble in DMSO‐d₆ (ϵ=46.7), it was chosen to evaluate whether a polarity increase of the environment changes their conformational ensembles.[ ⁵⁵ , ⁵⁶ ] Use of DMSO‐d₆ should also provide an indication of the ensembles adopted in an aqueous extra‐ or intracellular environment. The four drugs were studied in their non‐charged state as this is generally considered to be the form responsible for cell membrane permeation.

Table 1.

Aqueous solubility and lipophilicity of the four antiviral drugs.

Compound	Solubility[a] [μM]	LogD7.4 [b]
Asunaprevir	160±49	3.1
Simeprevir	19±3	3.7
Atazanavir	2.3±1.1	4.2
Daclatasvir	0.50±0.13	3.9

[a] Determined in aqueous potassium phosphate buffer at pH 7.4. Values are means±std from three repeats. [b] Determined using a miniaturized shake‐flask procedure. Values are means from three repeats.

The flexibility indicated by NRotB and ϕ for the four compounds (Figure 1B) suggest that they cannot be represented by a single solution conformation. Instead, they are likely to exist in an equilibrium of rapidly exchanging conformations, which can only be accurately represented by a conformational ensemble. ^[57] We therefore used the NMR analysis of molecular flexibility in solution (NAMFIS) algorithm to deconvolute the time averaged ^[58] NMR data into solution ensembles for each drug. ^[57] The NAMFIS algorithm was chosen as it has been validated in numerous studies of structurally diverse compounds both in Ro5 and bRo5 space.[ ²¹ , ⁵⁵ , ⁵⁶ , ⁵⁷ , ⁵⁹ , ⁶⁰ , ⁶¹ , ⁶² , ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁶⁸ , ⁶⁹ ] NAMFIS strives to find the best fit of interproton distances and dihedral angles determined by NMR spectroscopy to back‐calculated data from a probability weighted combination of conformations selected from a theoretical ensemble. ^[57] The nuclear Overhauser effect (NOE) was used to determine interproton distances, ^[70] whereas dihedral angles were derived from the ³ J _HH scalar coupling constants.[ ⁷¹ , ⁷² ] NOE build‐up curves were obtained from a series of NOESY spectra to allow the determination of accurate interproton distances (<3–10 % errors, depending on the solvent viscosity) using the initial rate approximation.[ ⁵⁶ , ⁷³ ] Dihedral angles may be associated with larger errors as the calculation of theoretical ³ J _HH scalar couplings is highly dependent on the constants and the number of terms used for the Karplus equation and generally gives 1–3 Hz errors.[ ⁵⁶ , ⁵⁷ , ⁷¹ ] Generation of a theoretical conformational ensemble with a comprehensive coverage of conformational space for each compound is also important for an accurate description of a solution ensemble.[ ²¹ , ⁶⁸ ] To this end, extensive unrestrained Monte Carlo conformational searches were performed using a high energy cut‐off, five different force‐fields and implicit solvation models for polar and apolar environments. Then the ensembles were combined, redundant conformers were eliminated and crystal structures retrieved from the PDB and the CSD were added. ^[21] This procedure minimizes the risk that the input ensemble to NAMFIS lack theoretically possible conformations, but it does not aim to provide a realistic ensemble for any specific solvent. Validation of the solution ensembles obtained as output from NAMFIS has been performed by comparison of the experimentally observed and back‐calculated interproton distances and dihedral angles, by the addition or subtraction of up to 10 % random noise to the experimental data, and by systematic removal of individual experimental restraints. Experimentally determined NMR data, information about the theoretical conformational ensembles and the output ensembles from NAMFIS for the four antiviral drugs are given in the Supporting Information (Tables S6–22, Figures S5–12).

Some parts of the four drugs were not described sufficiently well by the NOE and ³ J _HH coupling restraints to allow a reliable conformational analysis using NAMFIS (Figure 3). This involved the tert‐butyl glycine residue of asunaprevir (Figure 3A), the orientation of the acylsulfonamide moiety relative to the rest of the molecule in the two NS3/4A protease inhibitors and the thiazole unit of simeprevir (Figure 3A–3B). Atazanavir lacked restraints for the orientation of the biphenyl‐like side chain, and for the adjacent tert‐butyl glycine in chloroform (Figure 3C). For these three compounds, the orientations of the poorly described parts in the conformational ensembles obtained from the NAMFIS analyses were refined by systematically sampling the conformational space around the indicated bonds (Figure 3). Refined conformations were selected from the resulting conformational pool of each drug by two approaches, i.e. i) by retaining all conformations within a 5 kcal/mol energy window from the minimum energy conformation and ii) by selection of only those conformations from the entire pool that contain IMHBs for the amide protons that have a temperature coefficient ≤5 ppb/K as determined by NMR spectroscopy (Table 2).[ ⁷⁴ , ⁷⁵ ] Such IMHB‐refined conformations were found only for asunaprevir (conformations 1, 2, 3 and 6) and atazanavir (conformation 16). Both types of refined conformations were considered to be equally probable in the ensembles of each of the three drugs. The sets of refined conformations provide an estimate of the uncertainty that affects some of the conformations. Daclatasvir is symmetric with regards to the biphenyl group. Due to the fact that NMR‐spectra are time‐averaged, the two symmetric halves cannot be distinguished and only the solution ensembles for the asymmetric half of the molecule could be determined by NAMFIS analysis (Figure 3D). Complete conformations of daclatasvir were constructed by modelling all experimentally determined half‐conformations within each solution ensemble on a scaffold for the central imidazole substituted biphenyl moiety constructed using crystal structures. After PM7‐based geometry optimization all complete conformations within 5 kcal/mol from the minimum energy conformation were included in the solution ensembles.

Figure 3.

Overview of experimentally determined proton–proton distances and dihedral angles that were used to determine the solution conformational ensembles of (A) asunaprevir, (B) simeprevir, (C) atazanavir and (D) daclatasvir in DMSO‐d₆ and CDCl₃. Blue lines indicate proton–proton distances validated in the NAMFIS analysis of the four drugs. Orange lines indicate dihedral angles validated in the analysis of atazanavir. Purple lines indicate proton–proton distances that are crucial for the ensembles of asunaprevir. Removal of any of these distances resulted in that the ensemble failed to validate for asunaprevir. Black arrows indicate the torsional angles that were refined by conformational sampling.

Table 2.

Amide, urethane and acyl sulfonamide proton temperature coefficients for the four antiviral compounds determined by variable temperature ¹H NMR spectroscopy in DMSO‐d₆ and CDCl₃.^[a]

Proton	Temperature coefficient, DMSO‐d6 [ppb/K]	Temperature coefficient, CDCl3 [ppb/K]
Asunaprevir
2’‐NH	1.5	−0.3
1’’’‐NH	−2.8	−4.3
1’’’’‐NH	−3.5	−4.3
Simeprevir
13‐NH	−1.0	−1.3
1’’’‐NH	−2.5	−8.8
Atazanavir
4‐NH	−9.8	−2.0
7‐NH	−4.3	b.s.[b]
12‐NH	−3.5	0.8[c]
15‐NH	−6.5	−1.0
Daclatasvir
2’’’‐NH	−3.8	−3.3

[a] A temperature coefficient having an absolute value below 3 ppb/K indicates the amide proton to be involved in a strong intramolecular hydrogen bond, or being extensively sterically shielded from the surrounding solvent. An absolute value above 5 ppb/K is indicative of a solvent exposed amide proton.[ ⁷⁴ , ⁷⁵ ] [b] The resonance suffered from severe line broadening, preventing determination of the temperature coefficient. [c] The resonance suffered from line broadening and overlap with 7‐NH, and the temperature coefficient may therefore be associated with an increased error.

Solution conformational ensembles

Asunaprevir. The solution ensembles of asunaprevir consist of five conformations in DMSO‐d₆ and seven in chloroform (Figure 4, Table S17, Figures S5 and S6). Both solution ensembles have one conformation that is populated for about 50 % (conformations 3 and 12). The DMSO‐d₆ ensemble consists of four major (≥10 %) and one minor conformation, whereas three major and four minor conformations are found in chloroform. The high flexibility indicated for asunaprevir by its NRotB count is illustrated by the structural diversity of its conformations. The pairwise RMSD values of the heavy atoms in the peptide backbone varied between 1.12 and 3.02 Å for the conformations in the DMSO‐d₆ ensemble, and between 0.62 and 2.91 Å for chloroform (Table S23). For all heavy atoms, the pairwise RMSD values varied between 2.37 and 4.20 Å for the conformations in the DMSO‐d₆ ensemble, and between 2.14 and 4.46 Å for the conformations in chloroform (Table S24). As judged by these RMSD values, the two ensembles populated by asunaprevir show a similar overall degree of flexibility. The large, quinoline based side‐chain is rigid, but displays some flexibility with respect to the rest of the molecule. No common (congruent) conformation was found between the two solution ensembles; the two most similar conformations between solvents (No. 3 and 12) displayed a large pairwise RMSD of 1.07 Å. The lack of experimental restraints for the N‐terminal tert‐butyl glycine moiety of asunaprevir, and to some extent also for the C‐terminal acylsulfonamide (Figure 3A), results in some uncertainty for the ensembles of asunaprevir (Figure 4B). This uncertainty is exemplified by conformations 1, 2 and 3, for which the energy‐based and IMHB‐based refined conformations differ by RMSDs of 0.78, 1.38 and 1.20 Å, respectively.

Figure 4.

(A) Solution conformations and their populations (in %) for asunaprevir in DMSO‐d₆ (in blue) and in chloroform (in orange). (B) The major conformations (≥10 %) in the solution ensembles of asunaprevir and their populations (in %) in DMSO‐d₆ and chloroform. The output conformations from the NAMFIS analyses are given in grey, the conformations after energy‐based optimization are in green and those selected to contain intramolecular hydrogen bonds are in orange. (C) Overview of intramolecular hydrogen bonds found in the conformational ensembles in DMSO‐d ₆ and in chloroform indicated by blue and red lines, respectively. NHs have been labelled as in Table 2.

The urethane moiety of asunaprevir adopts a cis‐orientation both in DMSO‐d₆ (conformations 3 and 5, 70 %) and in chloroform (conformations 7 and 12, 55 %), as also reported for the crystal structure of asunaprevier (CSD ID: MIYWOI). ^[34] However, when bound to the HCV NS3/4 A protease the urethane is found in the trans‐conformation (PDB ID: 4NWL), ^[34] as in the remaining solution conformations of asunaprevir. The amide bond between the tert‐butyl glycine and proline moieties is trans in most solution conformations, just as in the two crystal structures, but cis in minor parts of the two ensembles [conformation 5 (22 %) in DMSO‐d₆ ; 6, 7 and 9 (in total 14 %) in chloroform]. Cis‐urethanes have been reported on multiple occasions,[ ⁷⁶ , ⁷⁷ ] just as cis‐prolyl amides.[ ⁷⁸ , ⁷⁹ ] The frequency of cis‐prolyl amides, or any cis tertiary amide, depends highly on the substituent on the amide nitrogen atom and the solvent.[ ⁷⁸ , ⁸⁰ ]

The temperature coefficients determined for the three NH‐protons of asunaprevir (Table 2) provide information regarding the extent to which these protons are involved in IMHBs or are sterically shielded from the surrounding solvent. The temperature coefficients suggest that the urethane 2'‐NH is strongly hydrogen bonded and/or shielded in both solvents, whereas the amide 1'''‐NH and the acylsulfonamide 1’’’’‐NH are involved in hydrogen bonds of intermediate strength or only partially shielded. The adjacent tert‐butyl group, and potentially also the nearby tert‐butyl ester, limit the solvent accessibility of the urethane 2’‐NH and most likely contribute to its low temperature coefficient. Energy refined conformations 2, 3, 4, 5, 7, 8 and 10 all display an IMHB for the acylsulfonamide 1’’’’‐NH, reflecting the temperature coefficient observed for this proton in both solvents. In addition, the IMHB‐refined conformations 1, 2, 3 and 11 also display an IMHB for 1’’’’‐NH, and conformation 6 displays one for 1’’’‐NH.

Simeprevir. The solution ensembles of simeprevir consist of seven conformations in DMSO‐d₆ and eight in chloroform (Figure 5, Table S18, Figures S7 and S8). Simeprevir populates four major (≥10 %) and three minor conformations in DMSO‐d₆ , whereas the chloroform ensemble is composed of two major and six minor conformations. Simeprevir shows some flexibility of the macrocyclic core, as indicated by the pairwise heavy atom RMSD values which varied from 0.05 to 1.07 Å in DMSO‐d₆ , and between 0.03 and 1.03 Å in chloroform (Table S25). The RMSD values for all heavy atoms of simeprevir varied between 1.65 and 4.41 Å in DMSO‐d₆ and 1.13 and 4.76 Å in chloroform (Table S26). Simeprevir thus displays a high overall flexibility both in DMSO‐d₆ and chloroform. It is well described by the experimental restraints (Figure 3B); consequently, both the DMSO‐d₆ and the chloroform ensemble were determined with reasonable accuracy. However, energy‐based refinement led to minor uncertainties for the C‐terminal acylsulfonamide in the conformations of the DMSO‐d₆ ensemble as illustrated for conformations 1, 2, 4 and 6 (Figure 5).

Figure 5.

(A) Solution conformations and their populations (in %) for simeprevir in DMSO‐d₆ (in blue) and in chloroform (in orange). The conformations for which the macrocyclic core belong to the same conformational class (RMSD<0.4 Å for the macrocyclic core heavy atoms) are indicated by roman numerals in the tables. (B) The major conformations (≥10 %) in the solution ensembles of simeprevir and their populations (in %) in DMSO‐d₆ and CDCl₃. The output conformations from the NAMFIS analyses are given in grey and the conformations after energy‐based optimization are in green. (C) The intramolecular hydrogen bond found in the conformational ensemble of simeprevir in DMSO‐d ₆ is indicated by a blue line. NHs have been labelled as in Table 2.

The macrocyclic core of simeprevir shows some similarity between the two solution ensembles. Conformations 7 (in DMSO‐d₆ ) and 8 (in chloroform) share the same conformation for the macrocycle core as revealed by clustering using RMSD<0.4 Å as cut‐off (RMSD of 0.18 Å). This conformational class (II) makes up 7 % of the solution ensemble in DMSO‐d₆ and 2 % in chloroform. Conformations 1, 2 and 3 all belong to conformational class I, sharing their macrocyclic core (RMSD≤0.4 Å) and make up 66 % of the solution ensemble in DMSO‐d₆ . In chloroform two additional conformational classes are found; class III, consisting of conformations 9 and 10 (51 %) and class IV, to which conformations 13 and 14 belong (36 %). The flexibility of the 14‐membered macrocyclic ring of simeprevir is further illustrated by the remainder of the conformations in the two solution ensembles. These include three conformations in DMSO‐d₆ , each representing from 5 to 12 % of the ensemble, and three in chloroform (with populations between 2 and 6 %) that have distinctly different cores (RMSD>0.4 Å). In summary, the large number of different conformational classes and their structural diversity illustrate the flexibility of the macrocycle core of simeprevir. The core of simeprevir shows a flexibility that is similar to or higher than that of the macrocyclic, natural product derived drugs roxithromycin, telithromycin and spiramycin studied earlier by us. ^[21]

Similar to the prolyl amide in asunaprevir, the tertiaty amide of simeprevir (the N−Me amide), appears as a cis/trans mixture,[ ⁷⁸ , ⁷⁹ ] with the cis‐conformation being dominant in DMSO‐d₆ (conformations 1, 2, 3 and 5; 51 %) and significant in chloroform (conformations 13, 14 and 15; 39 %). The secondary amide that is part of the macrocyclic core also has a low population of the cis‐conformation in DMSO‐d₆ (conformation 4, 12 %). Although cis‐conformations are far less common for secondary than for tertiary amides, they occur depending on the conditions and the substituent pattern around the amide bond.[ ⁸¹ , ⁸² ] For simeprevir one can assume that population of the cis‐rotamers for both amides is elevated by their location within the macrocyclic ring.

The two side‐chains of the macrocyclic core of simeprevir, i.e. the C‐terminus of the peptide backbone and the quinoline moiety, displayed a large degree of flexibility (Figures 5B, S7 and S8). The temperature coefficient of the amide proton (13‐NH) within the macrocyclic ring is low in both solvents, just as for the C‐terminal acylsulfonamide proton (1’’’‐NH) in DMSO‐d₆ , indicating these protons to be involved in IMHBs or sterically shielded from the solvent (Table 2). The amide proton within the macrocyclic ring (13‐NH) is involved in an IMHB in conformations 9, 10 and 12 of the chloroform ensemble. In DMSO‐d₆ conformations 1, 3, 6 and 7 (48 %) all display an IMHB for the C‐terminal acylsulfonamide proton 1’’’‐NH, which is in line the low temperature coefficient (−2.5 ppb/K).

Atazanavir. The solution ensembles of atazanavir consist of nine conformations in DMSO‐d₆ and eight in chloroform (Figure 6, Table S19, Figures S9 and S10). Four major and five minor conformations were found in DMSO‐d₆ , while the chloroform ensemble consists of three major and five minor conformations. The pairwise RMSD‐values of the heavy atoms of the peptide backbone of the conformations varied between 0.16 and 3.67 Å in the DMSO‐d₆ ensemble and between 0.22 and 3.63 Å in chloroform (Table S27). For all heavy atoms, the RMSD‐values varied between 0.80 and 5.31 Å in DMSO‐d₆ and between 0.61 and 4.95 Å in chloroform (Table S28). As revealed by these high RMSD values atazanavir has a high flexibility in both solution ensembles, which surpasses that of asunaprevir and simeprevir somewhat. This observation agrees well with atazanavir having the highest number of rotatable bonds and Kier flexibility index of the four drugs (Figure 1B). Atazanavir is described somewhat better by the experimental restraints in DMSO‐d₆ than in chloroform (Figure 3C), resulting in that the DMSO‐d₆ ensemble could be determined with good accuracy (Figures 6B and S9).

Figure 6.

(A) Solution conformations and their populations (in %) for atazanavir in DMSO‐d₆ (in blue) and in chloroform (in orange). The four conformations for which the peptide backbone belong to the same conformational class (RMSD <0.5 Å for the heavy atoms) are indicated by the roman numeral I in the tables. (B) The major conformations (≥10 %) in the solution ensembles of atazanavir and their populations (in %) in DMSO‐d₆ and CDCl₃. The output conformations from the NAMFIS analyses are given in grey and the conformations after energy‐based refinement are in green. (C) Overview of intramolecular hydrogen bonds found in the conformational ensembles in DMSO‐d ₆ and in CDCl₃ indicated by blue and red lines, respectively. NHs have been labelled as in Table 2.

Just as for simeprevir, the ensembles of atazanavir contained conformations that were closely related between DMSO‐d₆ and chloroform (Figure 6A). Conformations 1a (DMSO‐d₆ ) and 1b (chloroform), both of which originate from energy‐based refinement of conformation 1 from the NAMFIS analysis, as well as conformations 2 (DMSO‐d₆ ) and 10 (chloroform) all have a similar peptide backbone (RMSD <0.5 Å) and belong to one conformational class (I). All the other conformations found for atazanavir are unique conformations, i.e. their pairwise backbone heavy atom RMSD‐values differ by >0.5 Å. Class I conformations represent 33 % of the ensemble in DMSO‐d₆ and 9 % in chloroform.

Atazanavir has four amide protons and one hydroxyl group, which all could form IMHBs. The temperature coefficients of the amide protons indicate that three (4‐NH, 12‐NH and 15‐NH) are involved in strong IMHBs, or significantly sterically shielded in chloroform, while two (7‐NH and 12‐NH) may form IMHBs of moderate strength or are moderately shielded in DMSO‐d₆ (Table 2). In the chloroform ensemble, IMHBs involving 4‐NH are found in conformations 14 and 16, 12‐NH forms IMHBs in conformations 12 and 14, while 15‐NH is hydrogen bonded in conformation 16. In addition, 10‐OH is involved in IMHBs in conformations 9, 12, 13 and 15. As a result some conformations in the chloroform ensemble show three IMHBs. In DMSO‐d₆ , 7‐NH is involved in an IMHB in conformations 3 and 6, whereas conformations 6 and 7 show a hydrogen bond for 12‐NH.

Daclatasvir. As discussed above (cf. Methods), the symmetry of daclatasvir only allows the determination of the conformational ensemble for the asymmetric half of the molecule (Figure 7). The solution ensemble of daclatasvir consists of four major and one minor half‐conformation in DMSO‐d₆ and three major half‐conformations in chloroform (Figure 7A). The pairwise RMSD‐values for the heavy atoms of the peptide backbone in the half‐conformations vary between 0.89 and 2.21 Å in DMSO‐d₆ and between 0.67 and 1.76 Å in chloroform (Table S29). For all heavy atoms the RMSD values ranged between 1.13 and 3.78 Å in DMSO‐d₆ and between 1.23 and 3.98 Å in chloroform (Table S30).

Figure 7.

(A) Solution half‐conformations and their populations (in %) for daclatasvir in DMSO‐d₆ (in blue) and in chloroform (in orange). The two conformations for which the peptide backbone belong to the same conformational class (RMSD<0.5 Å for the heavy atoms) are indicated by the roman numeral I in the tables. (B) The major half‐conformations (≥10 %) of the solution ensembles of daclatasvir and their populations (in %) in DMSO‐d₆ and CDCl₃. All conformations are from the NAMFIS analyses and are given in grey. (C) Overview of intramolecular hydrogen bonds found in the conformational ensembles in DMSO‐d ₆ and in chloroform indicated by blue and red lines, respectively. NHs have been labelled as in Table 2.

One half‐conformation (no. 1) was common to both solution ensembles of daclatasvir, representing 12 % of the ensemble in DMSO‐d₆ and 37 % in chloroform (Figure 7B). In addition, half‐conformations 2 (DMSO‐d₆ ) and 6 (chloroform) are similar (RMSD 0.39 Å) and belong to the same conformational class I. The other half‐conformations of daclatasvir show low similarity, i.e. their pairwise peptide backbone heavy atom RMSD‐values differ by >0.5 Å, mainly due to the flexibility in the orientation of the terminal valine moiety.

Daclatasvir has the opportunity to form intramolecular hydrogen bonds within the peptide backbone. The temperature coefficient of 2’’’‐NH indicates that it may be involved in an IMHB of intermediate strength in both solvents, or partially sterically shielded (Table 2). As none of the half‐conformations contain such an IMHB it is likely that the low temperature coefficient originates from steric shielding from the iso‐propyl side chain of valine. However, all half‐conformations obtained by NAMFIS contain an IMHB between the imidazole 1’‐NH and the carbonyl oxygen atom of the valine moiety both in DMSO‐d₆ and in chloroform (Figure 7C).

Complete conformations of daclatasvir were built from the half‐conformations to allow calculation of the polar surface area and radius of gyration of the solution ensembles (Figure S11 and S12). This was done by combining all half‐conformations within each solution ensemble, for example conformer 1 was combined with conformers 6 and 7 in chloroform, on a template of the central imidazole substituted biphenyl moiety (cf. ensemble optimization in the Experimental section).

Characterization of solution ensembles

The radius of gyration (R_gyr) of a conformation is a measure of its size, and is calculated as the root‐mean‐square distance between its atoms and its centre of mass. ^[83] Large differences in R_gyr between the conformations of an ensemble therefore reveal major difference in size and shape between the conformations. Asunaprevir showed the largest variations in R_gyr, both in DMSO‐d₆ and in chloroform (R_gyr difference 1.22 Å and 1.14 Å, Figure 8, Table S36‐S39). The ensembles of simeprevir and atazanavir showed somewhat smaller variations in size for each solvent (0.89–1.18 Å), whereas the smallest difference was found for daclatasvir (0.95 and 0.72 Å in DMSO‐d₆ and chloroform, respectively). The ensembles were somewhat more compact in an apolar than in a polar environment for each antiviral drug, similar to that found recently for the two macrocyclic antibacterial agents roxithromycin and telithromycin. ^[21]

Figure 8.

Radius of gyration (R_gyr) versus solvent accessible 3D polar surface area (SA 3D PSA) of the energy and of the hydrogen bond refined solution ensembles for the four antiviral drugs. The circles for each conformation are labelled with the conformation number, have been scaled according to their population and are placed at the average R_gyr and SA 3D PSA found for a specific conformation. Variation of the R_gyr and SA 3D PSA within a conformation, resulting from refinement of parts lacking sufficient NMR restraints, are indicated by the whisker showing the maximum and minimum values found for each descriptor. The panel at the bottom shows the size for conformations having a 10 % population, while the color coding indicates the solvent of each conformation and whether it originates from an energy or hydrogen bond refinement.

The topological polar surface area (TPSA) provides a useful description of the polarity of Ro5 compliant compounds. ^[28] For compounds in bRo5 space, such as the four antiviral drugs discussed herein, the solvent‐accessible 3D polar surface area (SA 3D PSA) calculated for the conformational ensemble has been found to be a better descriptor of polarity. ^[22] In DMSO‐d₆ , asunaprevir, simeprevir and atazanavir showed large variation in SA 3D PSA within their ensembles (64–73 Ų), whereas daclatasvir displayed a much lower difference (12 Ų) (Figure 8 and Tables S40–S43). The variation in SA 3D PSA in chloroform was similar to that in DMSO‐d ₆ for asunaprevir and atazanavir (67–73 Ų), smaller for simeprevir (44 Ų) and larger for daclatasvir (34 Ų). Just as for the R_gyr, the SA 3D PSA of daclatasvir is associated with a larger degree of uncertainty originating from the lower number of available NMR restraints (Figure 2).

In conclusion, the four antiviral drugs do not show any significant differences in size (R_gyr) and polarity (SA 3D PSA) of the conformational ensembles adopted in apolar (CDCl₃) and polar solutions (DMSO‐d ₆). Their low aqueous solubility prevented the determination of their solution ensembles in water and we can therefore only speculate on the nature and composition of their aqueous ensembles. However, since significant differences were not observed between CDCl₃ and DMSO‐d ₆ we have some confidence that this could also be the case for hypothetical aqueous ensembles.

Subclassification of molecular chameleons

The ability to adopt more compact and less polar conformational ensembles in an apolar than in a polar environment due to a limited but still marked flexibility, is characteristic for molecular chameleons.[ ²⁰ , ²¹ , ²² , ²³ , ²⁵ ] Chameleonicity is believed to be an important property that facilitates for drugs in bRo5 space to display satisfactory solubility, cell permeability and oral absorption. The macrocyclic antibacterial agents roxithromycin, telithromycin and spiramycin all displayed this behavior, i.e. lower R_gyr and SA 3D PSA values in chloroform than in water, in a recent study. ^[21] We propose that such compounds are described as “full molecular chameleons” (Figure 9A). Asunaprevir, simeprevir and atazanavir investigated herein show comparable, or sometimes larger, variation in R_gyr and SA 3D PSA as compared to the three antibacterial agents. However, these three antiviral drugs do not adopt more compact and less polar ensembles in an apolar than in a polar environment (Figure 8), and cannot be described as full molecular chameleons. We suggest that compounds that populate conformational ensembles that display a large variation in size and polarity within, but not between environments of different polarity are termed “partial molecular chameleons” (Figure 9B). The large variation in polarity and size displayed by both types of molecular chameleons should allow them to achieve a favorable balance between aqueous solubility and cell permeability to a greater extent than more rigid drugs. It is, however, important to keep in mind that not all orally absorbed drugs in the bRo5 space behave as molecular chameleons, i.e. even though chameleonicity is considered to be an important property it is not an absolute requirement in this chemical space. Daclatasvir shows only a limited variation in R_gyr and SA 3D PSA and does not behave as a molecular chameleon (Figure 9C). This is also the case for the antibacterial agent rifampicin, which is rigid in the nature of its macrocyclic ring and only carries a small and rigid side chain. ^[21]

Figure 9.

Examples of drugs in bRo5 space that behave as (A) full molecular chameleons, (B) partial molecular chameleons and (C) those that are not molecular chameleons. The three classes differ in the diversity and relative location of their solution ensembles in chemical space defined by size (radius of gyration; R_gyr) and polarity (solvent accessible 3D polar surface area; SA 3D PSA). The circles for each conformation have been scaled according to their population. Panel C shows the size for conformations having a 10 % population and the colour coding indicating the solvent of each conformation. The structure of each drug is given in the corresponding panel.

Cyclosporin A is commonly regarded as the “prototype” molecular chameleon, which switches between one major conformation in an apolar environment such as chloroform and multiple states in aqueous solutions.[ ²⁷ , ⁸⁴ ] This conformational switch is enabled by cis/trans isomerization about an N‐methylated amide bond in the macrocyclic ring and results in the formation of a conformation with low SA 3D PSA in apolar environments, while the aqueous conformations have much higher polar surface areas. It is interesting to note that cis/trans isomerization about the two amide bonds within the macrocyclic ring of simeprevir also results in population of conformations that show large differences in SA 3D PSA and R_gyr, as compared to the conformation in which the amide bonds are trans (Figure S13). In contrast, cis/trans isomerization of the urethane moiety and the prolyl amide in asunaprevir only has minor effects on the two descriptors. Introduction of tertiary amides in macrocycles thus appears to be a particularly powerful strategy for the design of molecular chameleons, as demonstrated recently in a systematic approach to the de novo design of chameleonic cyclic peptides. ^[85]

Qualitative correlations to cell permeability

Drugs are generally assumed to be absorbed from the intestine and to enter target cells by passive permeability across cell membranes. ^[86] Different mechanistic models have been proposed for how passive cell permeability of compounds in the bRo5 space may occur. One postulates that the low‐dielectric conformation of a compound, i.e. the lowest energy conformation adopted in a low dielectric medium, such as the interior of a cell membrane, governs its passive cell permeability, ^[87] while a related model emphasizes the conformation having the minimal SA 3D PSA as the permeating species. ^[22] Yet another model suggests that one or several congruent conformations, i.e. conformations populated both in an apolar and in a polar environment, govern cell permeability. ^[88]

We determined the permeability of the four antiviral drugs across a monolayer of Caco‐2 cells as this model correlates well to oral absorption. ^[86] Cell permeability was dominated by efflux, and efflux ratios ranged from 38 and upwards with that of simeprevir being the largest (Table 3). Attempts to determine the passive, transcellular permeabilities of the drugs in the presence of a cocktail of inhibitors of the three major efflux transporters expressed by Caco‐2 cell did reduce efflux ratios. In spite of that, all but atazanavir showed residual efflux with simeprevir displaying the highest efflux ratio. The residual efflux displayed by three of the drugs prevents attempts to investigate quantitative correlations between passive cell permeability and the conformations in the solution ensembles. Instead, we focused on analysing the ensembles to provide qualitative mechanistic insight into how the drugs may permeate cells.

Table 3.

Caco‐2 permeability data and efflux ratios (ERs) for asunaprevir, simeprevir, atazanavir and daclatasvir.

Compound	Papp AB[a] [×10−6cm/s]	ER[b]	Papp AB+Inhibitors[c] [×10−6cm/s]	ER+Inhibitors
Asunaprevir	3.4±0.72	270	48±9.9	4.1
Simeprevir	0.19±0.065	330	0.50±0.21	13
Atazanavir	11±0.4	40	72±3.6	2.3
Daclatasvir	3.3±0.14	38	15±1.6	3.7

[a] P_app AB: permeability in the apical‐to‐basolateral (AB) direction across Caco‐2 cell monolayers. Values are means±std from three repeats. [b] ER: efflux ratio (Papp BA/Papp AB). [c] P_app AB+Inhibitors: permeability in the apical‐to‐basolateral (AB) direction across Caco‐2 cell monolayers, determined in the presence of a cocktail of three inhibitors that target the three major efflux transporters. Values are means±std from three repeats.

Formation of intramolecular hydrogen bonds (IMHBs) is an effective way for a compound to reduce its SA 3D PSA in an apolar environment. Ample evidence supports the importance of IMHBs for the cell permeability of compounds in the bRo5 space.[ ²² , ³¹ , ⁸⁹ ] As summarized above, the ensembles of all four antivirals contain conformations that have varying types and numbers of IMHBs (Figures 4, 5, 6, 7). However, none of the compounds display significant differences in IMHBs between an apolar and a polar environment, which agrees well with the fact that their solution ensembles in the two environments cover essentially the same SA 3D PSA ranges (Figure 8). Interestingly, comparison of the major (≥10 %) conformation of each drug that has the highest SA 3D PSA in DMSO‐d ₆ to the major one having the lowest SA 3D PSA in chloroform provided additional insight into how intramolecular interactions may contribute to cell permeability (Figure 10). For these conformations, both asunaprevir and atazanavir displayed an additional IMHB in chloroform as compared to DMSO‐d ₆. However, the most important difference between the major conformation in DMSO‐d ₆ and in chloroform was the overall fold. The flexible termini of all four compounds were folded towards other parts of the molecule in chloroform as compared to being extended in DMSO‐d₆ . Folding resulted in some reduction in the R_gyr, but the SA 3D PSA underwent a major reduction (52–74 Ų) for asunaprevir, simeprevir and atazanavir. It is interesting to note that the large reduction in SA 3D PSA for simeprevir (62 Ų) was accomplished without formation of any additional IMHB in chloroform, i.e. that large reductions in SA 3D PSA can be obtained also by other types of intramolecular interactions. ^[22]

Figure 10.

The major conformations (≥10 %) for (A) asunaprevir, (B) simeprevir, (C) atazanavir and (D) daclatasvir with the highest SA 3D PSA found in DMSO‐d ₆ (blue background) and the lowest SA 3D PSA found in chloroform (orange background). The conformation number, its population in %, R_gyr and SA 3D PSA are given for each conformation. Hydrogen bonds are indicated by green dotted lines and the termini that fold over in an apolar environment are encircles in the DMSO‐d₆ conformation for each drug.

It is tempting to speculate that partial molecular chameleons may permeate cells in partially folded, low SA 3D PSA conformations that are well populated in an apolar environment (e.g. the conformations illustrated for asunaprevir, simeprevir and atazanavir in Figure 10), and that reside within chemical space also populated in a polar environment. Such a model has many similarities to that emphasizing congruent conformations as governing cell permeability. ^[88] In contrast, full molecular chameleons such as spiramycin and telithromycin may rely on conformations having low SA 3D PSA and/or low energy that are well populated in an apolar environment for cell permeability.[ ²² , ²⁵ ] Future experimental and computational studies are required to provide deeper insight into the importance of different mechanism for the cell permeability of molecular chameleons.

Conclusions

We have used the NAMFIS algorithm ^[57] to determine the conformations of four antiviral drugs that reside in the chemical space beyond the rule of 5 (bRo5)[ ¹⁷ , ¹⁸ , ¹⁹ ] in polar and apolar solutions to understand whether they behave as molecular chameleons, and if this might be of importance for their cell permeability and oral administration. In some cases, the termini and side chains of the four drugs were not described sufficiently well by the NMR data to allow their reliable definition using NAMFIS. Those parts were instead described by sampling of their conformational space, followed by selection of refined conformational sets based on energy or IMHB criteria.

Asunaprevir, simeprevir and atazanavir showed a large degree of flexibility as revealed by the number of populated conformations and by their structural diversity in both environments. The solution ensembles of the three drugs consisted of five to nine conformations, with the two most different conformations of each ensemble displaying a high all heavy atom RMSD value (4.20–5.31 Å). As a consequence, the property space populated by the drugs was characterized by a large variation in descriptors that correlate to cell permeability. In particular, the solvent accessible 3D polar surface area (SA 3D PSA) varied by up to 73 Ų in both environments, while the radius of gyration (R_gyr) varied by up to 1.22 Å. In contrast, the solution ensembles of daclatasvir displayed a lower degree of flexibility and variation in SA 3D PSA and R_gyr due to the rigidity of its aromatic central part. The low solubility prevented the determination of the solution ensembles for the four antiviral drugs in aqueous solution using NMR spectroscopy. As no significant differences in R_gyr and SA 3D PSA were observed between the ensembles of each drug in DMSO‐d ₆ and chloroform, no significant difference can be expected for their aqueous ensembles.

The differences in flexibility, size and polarity descriptors displayed by asunaprevir, simeprevir and atazanavir are similar to those reported for the natural product derived antibacterial agents roxithromycin, telithromycin and spiramycin. ^[21] However, no significant differences were found between the ensembles adopted in the polar and apolar environment by each of the three antivirals, in contrast to the three antibacterial agents which adopted more compact and less polar conformational ensembles in chloroform than in water. We propose that compounds behaving as the three antibacterials are described as “full molecular chameleons”, while compounds behaving as asunaprevir, simeprevir and atazanavir are characterized as “partial molecular chameleons”. The difference between the two classes of drugs may originate from the fact that the three antibacterials reside somewhat further into the bRo5 chemical space than the antivirals (cf. Figure 2).

The folding differed significantly between the most polar major conformer in DMSO‐d₆ and the least polar one in chloroform for each antiviral. For the conformations adopted in chloroform the flexible termini of each drug were folded over adjacent parts of the molecule in chloroform, but was extended in DMSO‐d₆ . This resulted in major reductions in SA 3D PSA (≤74 Ų) between the pairs of conformations for asunaprevir, simeprevir and atazanavir. Previous studies have estimated that reductions in SA 3D PSA of approximately 75 Ų correspond to a close to 200‐fold increase in cell permeability.[ ²² , ⁹⁰ ] It is therefore tempting to hypothesize that the high flexibility, which allows the antivirals to adopt conformations with low SA 3D PSA, improves their cell permeability and contributes to allowing them to be administered orally.

Experimental Section

NMR spectroscopy: Asunaprevir, simeprevir, atazanavir and daclatasvir were purchased from Selleckchem in >95 % purity. For the assignment and the NAMFIS analyses of the two HCV NS3/NS4A protease inhibitors, ¹H and ¹³C NMR, COSY, TOCSY, HSQC, HMBC and NOESY spectra were recorded at 25 °C on a 900 MHz Bruker Avance III HD NMR spectrometer equipped with a 5 mm TCI cryogenic probe. For the remaining two compounds the same series of spectra were recorded at 25 °C on an 800 MHz Bruker Avance III HD NMR using a 5 mm TXO cryogenic probe. Dihedral angles were derived from ³ J _HH scalar coupling constants measured by ¹H NMR spectra using the Karplus equation.[ ⁷¹ , ⁹¹ ] Interproton distances were derived from the NOE build‐up rate through the initial rate approximation, and used for the NAMFIS analyses. For every sample seven NOESY spectra with different mixing times (100–700 ms with 100 ms interval) were recorded without solvent suppression. Every NOESY spectrum was recorded with a relaxation delay of 2.5 s, 16 scans, 4096 points in the direct dimension and 512 points in the indirect dimension (spectral window of 11–12 ppm). The interproton distances were calculated using either two geminal methylene protons (1.78 Å) or two ortho protons (2.51 Å) as a reference. Other reference proton pairs within the molecule served as a quality check for the calculated distances. Normalized intensities of cross‐peaks by the diagonal‐peaks were used for distance calculation. Normalization was done according to (cross‐peak_a,b×cross‐peak_b,a)/(diagonal‐peak_a×diagonal‐peak_b). Only proton‐pairs with at least four mixing times that gave a linear fit to the NOE build‐up curves (R²>0.94, typically>0.97) were used for distance calculation. Distances were calculated according r_ab=r_ref(σ_ref/σ_ab)^(1/6), where r_ab is the distance between protons H_a and H_b in Ångström, r_ref is 1.78 Å or 2.51 Å, and σ_ref and σ_ab is the slope of the linear part of the build‐up curve for the reference and the a‐b proton pair, respectively. Full ¹H‐assignements, distances and coupling constants are provided in the Supporting Information (Table S2–S14).

The amide temperature coefficients were determined by variable temperature NMR studies (Tables S32‐S35). All ¹H spectra were recorded on a 500 MHz Bruker Avance NEO equipped with a 5 mm cryogenic TXO probe using 16 scans, a relaxation delay of 0.7 s and 32768 number of points (spectral window of 14.5 ppm). For the antiviral in DMSO‐d₆ a series of ¹H spectra were recorded ranging from 25 to 65 or 20 to 60 °C with increments of 10 °C. Temperature coefficients in CDCl₃ were recorded at temperature ranging from 10 to 50 °C, with steps of 10 °C. The coefficients were calculated according: (δ_high−δ_low)/(T_high−T_low). ^[75]

Conformational sampling: In total nine conformational searches were performed for each of the two HCV NS3/NS4A protease inhibitors and ten for the remaining two compounds, largely consisting of Monte Carlo Molecular Modelling (MCMM) calculations to create the theoretical input ensemble. All conformational searches were set‐up using MacroModel as implemented in the Schrödinger suite. ^[92] A combination of force‐fields and solvation models as well as a high energy cut‐off was used to ensure sampling of the entire conformational space. For every molecule, eight or ten different MCMM conformational searches were performed: four or five different force‐fields (OPLS, OPLS 2005, OPLS3 or OPLS4, AMBER* and MMFF) combined with two different GB/SA continuum solvation models ^[93] (water and chloroform). The MCMM were performed using 50000 Monte Carlo steps, the energy window for saving structures was set at 10 kcal/mol, an RMSD‐value varying between 0.5 and 2.0 Å and the Polak‐Ribiere type conjugate gradient (PRCG) with a maximum of 5000 iteration steps for the energy minimization. An additional simulation employed macrocycle conformational sampling (MCS), which is a hybrid of large‐scale low‐mode (LLMOD) sampling with simulated annealing (molecular dynamics). ^[94] The MCS was performed using OPLS‐2005 with GB/SA (water) as electrostatic treatment. The MCS was set‐up with 5000 simulation cycles, 5000 LLMOD search steps, and the energy window for saving structures set at 10 kcal/mol with an RMSD of 2.0 Å. For each drug all conformational searches were combined and followed by redundant conformer elimination, which was done by comparison of all heavy atom positions including OH and SH, and using an RMSD cut‐off between 1.0 and 2.5 Å. To the combined and reduced ensemble, all available PDB and CSD X‐ray structures were added, yielding the final input ensemble used for the NAMFIS analyses. All conformational searches fulfilled the equation 1‐(1‐(1/N))M, where N is the number of conformers and M the number of search steps, which is an indication that the conformational space is fully covered. ^[68] The Supporting Information contains more information about the input ensemble used for the NAMFIS analyses (Table S15 and S16).

NAMFIS analyses: The NAMFIS algorithm aims to find the combination of conformers, and their populations, from the theoretically generated input ensemble that has the best fit solution for the population‐weighted back‐calculated distances and coupling constants to the experimentally determined data. The experimental data consist of distances between single protons, a single proton and an averaged methyl group and coupling constants. The theoretical distances involving a methyl group are averaged according to d=(((d_a ⁻⁶)+(d_b ⁻⁶)+(d_c ⁻⁶))/3)^−1/6. Validation of the obtained populations was done by removal of single data points, by addition or subtraction of up to 10 % random noise to the experimental distances, and by comparison of the experimental data to the by NAMFIS calculated values. Only conformers selected for more than 1 % were considered significant and population changes up to 10 % were considered part of a validated set of experimental data. The termini of asunaprevir were too flexible to give reliable solution ensembles and their conformations should be interpreted with care. To a set of validated distances, additional distances were given to get an impression of the general orientation. However, these distances could not be validated (indicated with purple lines in Figure 3). The results of the NAMFIS analyses for the individual molecules can be found in the Supporting Information (Table. S17–22, Figure S5–12).

Ensemble optimization: The parts of the NAMFIS selected conformations that were not described by experimental data for asunaprevir, simeprevir and atazanavir were further refined by coordinate scans using MacroModel as implemented in the Schrödinger suite. ^[92] Figure 3 shows the experimental data‐coverage and the bonds that were refined for each solution ensemble are highlighted with black arrows. Coordinate scan were performed using the OPLS4 force‐field, fixing the torsion angles for which NMR data was available, and using the appropriate solvation model (CHCl₃, or a dielectric constant of 46.7 for DMSO) without any further energy minimization. Conformations that involved a single coordinate scan, i.e. single torsional angle, were sampled every single degree, and those involving two coordinate scans were sampled every 10 degrees. All conformations within a 5 kcal/mol energy window from the minimum energy conformation of the coordinate scan were selected as the first set of likely conformations. Conformations displaying IMHBs in line with the variable temperature experiments for the coordinate scan were selected as a second set of likely conformations. Such sets of conformations were selected within a 5 kcal/mol window from the lowest energy conformation displaying the particular IMHB, and were found only for asunaprevir (conformations 1, 2, 3 and 6) and atazanavir (conformation 16). The presence of an IMHB was defined by a hydrogen‐bond acceptor angle of 120°+/−30°, a hydrogen‐bond donor angle of 150 °+/‐50 ° and a heteroatom distance below 3.3 Å. ^[95] Several low‐energy conformations were found around an angle of 0° and 180° for the coordinate scan of the thiazole moiety of simeprevir. Only the lowest energy conformation for each of the two angles was selected to be used in the further analysis.

Assembly of the complete conformations of daclatasvir was done so that the biphenyl ring adopted a 32° twisted conformation, i.e. an average of the angle found in many crystal structures (e.g. PDB IDs 6 U67, ^[96] 2XRX, ^[97] 2GGX ^[98] and 5AEW ^[99] ). Similarly, the imidazole moieties were positioned to adopt a 26° twisted conformation with regards to the adjacent phenyl ring. ^[100] Then all possible combinations of the experimentally determined half‐conformations were modelled onto the central bis‐imidazole substituted biphenyl system, including orienting the peptidic termini so that they were cis and trans oriented with respect to the central biphenyl moiety. The IMHB between the imidazole NH and the valine carbonyl oxygen atom found in all half‐conformations from NAMFIS was maintained in all complete conformations. This provided a total of 28 conformations in DMSO and 12 conformations in CHCl₃. Following that, the geometry of the conformations was optimized (with an RMSD tolerance set to 0.5 Å) using the PM7 method, as implemented in the Schrödinger package‘s MOPAC2012 module. The energies of the resulting conformations were calculated using the current energy function with the OPLS3e force‐field and the solvation models using the same setting as for the optimization of the ensembles of asunaprevir, simeprevir and atazanavir. All conformers within a 5 kcal/mol energy window from the minimum energy conformer were selected.

Structural analysis of ensembles: Conformations were grouped in classes based on the macrocyclic core for simeprevir and the peptide backbone for the other three inhibitors. Grouping into classes was based on the minimum energy conformation of each of the sets of energy‐ and IMHB‐refined conformations (cf. refinement of conformations, above). Conformations were considered to belong to the same conformational class when they had root‐mean‐square deviation (RMSD) <0.5 Å for all heavy atoms of the peptide backbone or <0.4 Å for the more rigid macrocyclic core of simeprevir, respectively (cf. Figure 1). Comparison of overall conformations was done in an identical manner, but using all heavy atoms.

The presence of IMHBs in all conformations was analyzed using the IMHB criteria listed above (cf. Ensemble optimization). Conformational sets obtained by the energy‐ and IMHB‐based refinement were considered to possess a specific IMHB if at least one of the conformations in the set displayed an IMHB that met the criteria. Usually, most conformations in such a set would display that specific IMHB.

Molecular descriptor calculations: For each conformation the molecular radius of gyration (R_gyr) was calculated using MOE (v2015.10), while the solvent accessible three‐dimensional polar surface area (SA 3D PSA) was calculated with PyMol v1.7.4. ^[101] For the SA 3D PSA, absolute partial charges were first calculated with the B3LYP/6‐31G** method in the Jaguar tool (available in the Schrödinger suite). ^[102] Then the solvent accessible surface area was defined in PyMol using a solvent probe with a radius of 1.4 Å, and partial charges greater than 1 or smaller than −1 were included in the calculation of SA 3D PSA. Further details on 3D PSA calculations have been reported elsewhere.[ ²¹ , ²² ]

Aqueous solubility, LogD and cell permeability: The thermodynamic aqueous solubility, LogD and cell permeability across Caco‐2 cell monolayers were determined as reported previously. ^[22]

Conflict of interest

The authors declare no conflict of interest.

Supporting information

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

Supporting Information

Wieske L. H. E., Atilaw Y., Poongavanam V., Erdélyi M., Kihlberg J., Chem. Eur. J. 2023, 29, e202202798.

Data Availability Statement

Original NMR data (FIDs) for the NOESY buildups and for the variable temperature NMR experiments, as well as output conformational ensembles, are available free of charge at Zenodo as DOI: 10.5281/zenodo.7231862.