Abstract:

The three-dimensional conformations adopted by a free ligand in solution impact bioactivity and physicochemical properties. Solution 1D NMR spectra inherently contain information on ligand conformational flexibility and three-dimensional shape, as well as the propensity of the free ligand to fully pre-organize into the bioactive conformation. Herein we discuss some key learnings, distilled from our experience developing potent and selective synthetic macrocyclic inhibitors, including Mcl-1 clinical candidate AZD5991. Case studies have been selected from recent oncology research projects, demonstrating how 1D NMR conformational signatures can complement X-ray protein-ligand structural information to guide medicinal chemistry optimization. Learning to extract free ligand conformational information from routinely available 1D NMR signatures has proven to be
fast enough to guide medicinal chemistry decisions within design cycles for compound optimization.

INTRODUCTION

Synthetic macrocycles are an increasingly attractive modality for pharmaceutical drug discovery projects. Macrocyclic scaffolds can enhance potency at their target and selectivity over anti-targets and improve metabolic stability and physicochemical properties compared to their acyclic counterparts, while maintaining oral bioavailability.1-6 In addition, perhaps the greatest advantage of macrocycles in contrast to most acyclic small molecules, is their ability to efficiently bind large, flat, featureless, difficult-to-drug binding sites, such as proteinprotein interaction (PPI) regions. On the other hand, a disadvantage may be the increased synthetic challenge of macrocycles over their acyclic counterparts. Consequently, careful design strategies to focus synthetic efforts and identify optimal linkers to cyclize into a bioactive conformation are desired.

Protein-ligand X-ray crystallography data and associated visualization tools are a vital part of rational drug design, providing insights which help focus molecular design hypotheses towards a goal of high target affinity. However, optimization of ligand potency can still be an empirical process, even when structure-based drug design is enabled through the availability of crystal structures of small molecules bound to target proteins. Protein-ligand X-ray structures generally provide a static snapshot of the final protein-ligand complex, without any information on the journey taken by the ligand to arrive at the bound, bioactive conformation.7 While crystallographic data provides fine detail on inter-atomic interactions, it is not possible to quantify the amount of ligand strain incurred to adopt the observed binding mode without additional information from the conformational preference of the free ligand.8

In our view, information available on a ligand’s conformation in solution is often underutilized or even overlooked, even though it can provide a critical “missing link” in 3D structure-based drug design.9,10 The extent to which the free ligand pre-organizes in solution and whether the bioactive conformation is preferred can influence design hypotheses and subsequent design steps. Accurate and robust predictions of the free and bound conformations of a ligand can be pivotal to successful structure-enabled drug design campaigns.11,12 In addition to optimizing opportunities for ligand-target interactions, binding potency can be maximized by designing free ligands with an innate preference for their bioactive conformation13. Pre-organization of the free ligand reduces free energy penalties due to torsional strain and intra-molecular rearrangement as part of the protein-ligand binding event.11

While the conformational preferences of a small molecule in solution encode an abundance of useful information for medicinal chemistry design, strategies to extract the maximum amount of practical information from the smallest resource investment has been an evolving area of interest. Synthetic chemists typically rely on NMR for structure verification and to check the purity of newly synthesized compounds. These routine 1D NMR spectra implicitly encode 3D information about free ligand conformations, as NMR chemical shifts and J-coupling patterns are exquisitely sensitive to both atomic electronic environment and molecular dynamics.9,14,15 including routine conformational analysis in design through NMR spectral signatures for flexibility and shape can be advantageous. Herein we provide a set of case studies where molecular design has benefited from timely conformational insights obtained from analysis of routine 1D NMR data. The case studies encompass both prospective insight, in which NMR signals were used as markers of success in design iterations, and retrospective analysis in which SAR trends were rationalized based on NMR data. While direct impact cannot be guaranteed in every application of NMR to ligand conformational analysis, we have found that analysis of 1D NMR data enhances understanding of chemical series and is especially valuable for understanding the impact of
macrocyclization on conformational dynamics.

RESULTS AND DISCUSSION

1D NMR Conformational Signatures: Free Ligand Flexibility and Shape

Solution 1D NMR spectra inherently contain information on ligand conformational flexibility and threedimensional shape. To illustrate this principle, Figure 1 compares 1D NMR spectra between piperidine and its 2methylated counterpart. Methylation is a common rigidification strategy in a medicinal chemist’s toolbox.16 In Figure 1a, a separation of signals is observed for the axial and equatorial hydrogen atoms attached to the same carbon atom (whether w, x, y, or z), producing distinct axial and equatorial chemical shift and J-coupling patterns, suggesting conformational pre-organization. The 1D NMR spectral signature in Figure 1a is a hallmark of a chair conformation, with the methyl adopting a preferred equatorial orientation. This observation is familiar to most chemists. By comparison, the non-methylated analog displays fewer resolved signals, suggesting conformational averaging of the local electronic environment, hence a read out of conformational flexibility (Figure 1b), or in other words low populations of preferred conformations.

This approach of deducing 3D conformational preferences from routine 1D NMR spectra extends to more complex molecules pre-organized in solution. Figure 3 gives the case of a potent, and selective PPI inhibitor for the oncology target Mcl-1, AZD5991 (1, Figure 3a),17 currently in clinical trials (NCT03218683). The 1D NMR spectrum of free ligand 1 in solution (Figure 3b) reveals molecular conformational pre-organization: distinct chemical shifts for each hydrogen atom correspond to unique electronic environments. This is particularly evident in the aliphatic region with separated methylene signals each with an integral of one. In order to measure what is the preferred conformation and its population, we performed conformational analysis by NMR17 encompassing an exhaustive analysis of multiple NMR constraints, including chemical shifts, through-space distances between pairs of atoms (derived from NOEs and/or dipolar couplings) and torsion angles (derived from J-couplings).9,12,1821 However, recognizing that this process still requires expert interpretations, especially for complex NMR spectra from macrocycles, we also developed a diagnostic NMR conformational approach for rapid conformational verification, which can be used directly by chemists to confirm the presence of expected conformational signatures in new compounds.

To determine the 3D preferred conformation, commercial software provides accessibility to non-experts for quantum mechanical (QM) chemical shift calculations from one or more proposed 3D conformation(s).24,25 Advances in high performance computing have enabled us to routinely use QM models to generate conformational ensembles, as well as calculate NMR parameters, such as 1H chemical shifts and 1H, 1H Jcouplings, providing reliable free ligand 3D structural predictions within a timeframe suitable for synthesis and testing cycles encountered in drug design projects. These conformation dependent shift calculations are compared to an experimental 1D NMR spectrum to validate the proposal. This process of validation of 3D conformations of pre-organized ligands is analogous to routine validation of 2D chemical structures of newly synthesized molecules by predicting carbon chemical shifts values for various 2D structural proposals. Figure 3c shows two different proposed 3D conformations: (1) The calculated global minimum determined with a QM potential (B3LYP/6-31G*), including a polarizable continuum model (PCM) as the continuum solvation method for water, shown in red with thenaphthyl ring “down”. (2) The bioactive conformation, shown in green, derived from the crystal structure of 1 in complex with Mcl-1 (PDB 6FS0), shown on the far right. The 3D conformation dependent QM calculated shifts are tabulated and compared to experimental values (SI Table S1). This comparison is a quick method to determine if the proposed conformation is consistent with the experimental conformation. In this manner, the QM-derived global minimum conformation, with thenaphthyl ring “down”, cannot be regarded as the dominant conformer in solution (labeled and colored red) as the discrepancy with experimental NMR data is too large to support such a conformation (SI Table S1). Notably, the pyrazole CH was calculated to be 4.1 ppm, confirming it is not the bioactive conformation. The dominant free ligand conformation in solution (labeled “bioactive” and colored green) was determined by fitting experimental NMR parameters to the full QM refined conformational ensemble using MSpin.23 The experimental chemical shift of the pyrazole CH, at 4.8 ppm, is identical to the QM calculated shift value for the bioactive conformation.

Although this proton identifies clearly as a sensitive conformational reporter signal, even small changes in ligand conformation correspond to measurable shift differences across the whole 1D NMR Medial proximal tibial angle spectrum, which effectively is a unique fingerprint for a pre-organized conformation. It is worth noting that the free ligand conformational preference of 1 determined in DMSO-d6 is preserved in aqueous buffer at pH=7.4, as established by measuring 1D NMR conformational signatures in buffer (SI Figure S3). A definitive advantage of using 1D NMR conformational signatures to validate 3D conformations is to enable characterization of molecules with submicromolar solubility in any solvent; using high field spectrometers equipped with cryo-probes, such a 1D proton spectrum can be acquired in one to six hours.

To summarize, free AZD5991 (1) in solution is fully pre-organized into its bioactive conformation for Mcl-1 binding and the pyrazole signal was chosen as a bioactivity reporter signal. The recognition that the calculated global QM minimum is not experimentally confirmed by NMR succeeded in de-validating any design hypotheses aimed at changing a putative in silico free ligand conformation to match the bioactive X-ray conformation and confirming the full potential for activity, in terms of protein complementarity and free ligand pre-organization, was likely already achieved. Such timely information to design teams is crucial in saving synthetic effort, by avoiding less promising design ideas, especially when lengthy macrocycle syntheses are involved.

Single bioactive conformational reporter signals of pre-organized ligands in solution can often but not always be identified in a 1D NMR spectrum. Figure 3 reports the free ligand conformation of a second Mcl-1 advanced lead26 and of a pyrazolo[1,5-a]pyrimidine BCL6 PPI inhibitor27 and their unique 1D conformational signatures. The molecules clearly pre-organize in solution, as demonstrated by the complex, yet defined signals corresponding to each hydrogen atom, especially in the aliphatic region. The molecules are also pre-organized into the bioactive conformation in solution, as demonstrated by overlap of their experimental chemical shifts with the values calculated by QM based on the bioactive conformation (as observed in the ligand-protein crystal structures, see SI Tables). A conformational reporter signal can be identified for the tetrahydrobenzazepinone aromatic proton of 2, highlighted in green, which appears at 6.4 ppm (Figure 3a), very different from the predicted shift using empirical 2D based calculations, which is 7.6 ppm (ACD/Labs or 7.4 ppm using MNova). This is not the case for 3, where no single hydrogen atom is flagged by the 2D structural predictions as being anomalous (SI Table S3). In such cases, the entire unique disposition of the 1D NMR chemical shifts and Jcouplings will need to be used over a single reporter signal as an indication of the bioactive conformation. Interestingly, the chemical shift value of the hydroxyl proton for 3 could be compatible with an internal hydrogen bond (IHB). This was confirmed with a spectrum in chloroform (SI Figure S4). The presence of this dynamic IHB effectively decreases the exposed polarity of the molecule in a non-polar environment and can facilitate membrane permeability, demonstrating how the free ligand conformation can affect the molecule’s physicochemical properties.

In conclusion, the examples above show that recognition of 1D NMR conformational signatures can help assess the degree of flexibility/rigidity and bioactivity of a molecule, effectively enhancing 3D structure-based analysis. The inherent simplicity of using routinely available data for every compound synthesized confers a practical advantage of accuracy and speed, to enable the data to influence design decisions within a design cycle of a lead optimization project.

1D NMR Guided Macrocyclization

Designing compounds that pre-organize into a bioactive conformation is a well-known and effective strategy for optimizing potency.28-31 Some strategies include addition of chiral centers,28 rings, hydrophobic collapse,29 intramolecular interactions, etc; with macrocyclization as one of those strategies that is particularly successful with PPI targets. The clinical candidate AZD5991 (1) and the other two advanced leads, 2 and 3 described in the previous section, are fully pre-organized into the bioactive conformation and exhibit very high binding affinities. This correlation between ligand pre-organization into the bioactive state in solution and potency is observed in several SBDD cases, supported by X-ray protein-ligand structures and NMR free ligand conformations.10 This is consistent with achieving the maximum potential for ligand pre-organization and protein complementarity.9,10,31,32 In the following we will show how combining 1D NMR free ligand conformational signatures and x-ray crystallography protein ligand data can effectively guide macrocyclization strategies. Specifically, we will show how to prioritize design hypotheses that exploit protein complementarity, while minimizing free energy penalties and torsional strain of free ligands on binding.

Johannes, et al.26 effectively used NMR-derived information on conformations of a ligand free in solution and bound to its protein of interest, to guide macrocyclization design of potent inhibitors of the Mcl-1 protein. Compound 4 was identified as a low affinity hit from a DNA-encoded library screen.26 shows the methylene hydrogen signals of the NMR spectrum, which indicate some degree of conformational pre-organization (distinct hydrogen atom chemical shifts within a methylene group), while some areas retain higher flexibility (overlapping signals) suggesting absence of a preferred conformation in solution. A mostly flexible conformation, with local rigidity around hydrogen atom x, is preferred by the acyclic free ligand in solution. On the other hand, the X-ray structure of this hit bound to Mcl-1, shows a distinct bioactive shape. Taken together, this structural information on both the free and bound ligands suggests the opportunity for a macrocyclization design, aimed at stabilizing the free ligand conformation into the bioactive state. The X-ray structure provided a clear rationale for linking at the periphery of the binding site to create a macrocycle which retained the observed pharmacophore for binding to the Mcl-1 pocket, while the free ligand NMR conformational flexibility highlighted the potential gain in potency through pursuing this strategy (Figure 4).

Macrocyclization did lead to an increase in potency (Figure 4b). The spectrum of the free ligand 2 confirms the success of the design as the molecule is fully pre-organized into the bioactive state in solution (see previous section). Notably, the large increase in shielding of x” is consistent with ring current effects from the expected bioactive geometry of the dichloro-substituted aromatic ring, as seen in the crystal structure of the protein complex. Three-dimensional shape constraints due to macrocyclization are particularly well suited to NMR signatures. The unique chemical shift of each proton in the methylene pair becomes increasingly distinct as the macrocycle rigidifies their local environments. This example demonstrates that the experimental knowledge of ligand flexibility can be sufficient to trigger ligand rigidification designs.

Macrocyclization that led to AZD5991 (1) also showed progressive pre-organization of the bioactive conformation in solution associated with increases in potency. Figure 5 shows acyclic precursors of 1 (Figure 3). A previously reported33 acyclic indole-2 carboxylic acid 5 that binds to the BH3-binding domain of Mcl-1, shows conformational flexibility, as determined by the NMR aliphatic signature (Figure 5a). The chemical shift values of the methylene protons in 5 are similar, indicating motional averaging and flexibility. Addition of a thioether, 6, leads to an increase in potency and broad, separated signals on the 1D NMR spectrum for one pair of methylene hydrogen atoms proximal to the sulfur atom indicating partial rigidification of this region (w′ and w″, b). An Xray structure shows the bioactive conformation of 5 adopts a U-shape and highlights the proximity of regions which would be widely separated in an open conformation (pyrazole to naphthyl). The bioactive molecular disposition could explain the discrepancy between the observed pyrazole chemical shift (5.6 ppm) and the value of 6.2 ppm from empirical 2D based shift predictions (SI Figure S5). However, the remaining methylene groups still show clear flexibility. This example emphasizes the molecular details provided by simple 1D NMR spectra: such atom-specific information is not available by other methods. Macrocyclization led to a 10-fold increase in potency, and a further boost in potency was achieved by rigidification through atropisomerization, i.e. methylation of the indazole nitrogen and addition of a chlorine atom.17 The clinical candidate AZD5991 (1), a sub-nanomolar inhibitor of Mcl-1, is the active atropisomer and it is fully pre-organized into the bioactive conformation as confirmed by its 1D NMR conformational signature (Figure 3).

The same rigid indole core with varying degrees of aliphatic conformational flexibility (1H, DMSO-d6, 27 ºC, 500 MHz). The ensemble of conformers (grey) is shown superposed with the closest minimized conformation of the bound ligand (green, PDB 6FS1, 3.6 Å inter-atomic distance between the naphthyl 3-carbon atom and the pyrazole 5-methyl group) indicates the bioactive conformation is not highly populated by the free ligand in solution. The conformational reporter signal, highlighted in red appears by NMR at 5.6 ppm, highly deshielded compared to the QM calculated value of 4.3 ppm for the bioactive conformation (SI FigureS5).5

In a third example, McCoull, et al.27 developed a potent and specific probe molecule to evaluate BCL6 as an oncogenic drug target. The proton spectral signature of 7 (Figure 6) shows conformational flexibility. Of note is the broad line-width of a strongly shielded aromatic CH proton (5.6 ppm, empirical 2D based shift predictions (ACD/Labs and MNova)>6.1 ppm), indicative of conformational dynamics on the millisecond time scale consistent with restricted rotation around the bridging NH rotatable bonds. The 3D bioactive conformation (based on the crystal structure of a structural analog, PDB 5N21, with a carboxylic acid substituted pyrrolidine, in lieu of the hydroxyl substitution, see SI Table S4), has a proximal 7 Å inter-atomic distance, highlighted in Figure 6a, which suggested an opportunity for macrocyclization to increase potency. Figure 6b shows the potent macrocycle 3 adopts preferred molecular conformation in solution. Pre-organization is demonstrated by the loss of line broadening of the aromatic CH hydrogen atom (x*) and the clear separation of most of the methylene hydrogen atoms, apart from the pair labeled z, only beginning to show differentiation (Figure 6, inset). The fine features in the inset reflect, relative to the other well differentiated methylene pairs, changes in localized dynamics and/or extent of difference in electronic shielding between the two distinct chemical environments of the hydrogen atoms of interest. analog, with a carboxylic acid substituted pyrrolidine in lieu of the hydroxyl substitution (SI Table S4). (b) 1D NMR (1H, DMSO-d6, 27 ºC, 600 MHz) of the macrocycle reveals free ligand pre-organization and correlates with improved potency (SI Table S3).6

1D NMR Signatures to Enhance SAR for Linker Optimization

The correlation between ligand affinity and the degree of ligand bioactivity as captured by the 1D NMR conformational signatures is well suited for a project series of analogs, where a bioactive reporter signal can be identified and used repeatedly across multiple compounds to support SAR and design hypothesis generation. By investing in discernment of a bioactive 1D NMR conformational marker for a representative chemical series in solution, those learnings can be applied to other similar scaffolds in a series directly from routine 1D NMR
spectra.

Johannes, et al.26 effectively utilized the knowledge of free ligand conformation in solution to explain unexpected SAR findings upon amide N-methylation. In solution, the non-natural peptidic macrocyclic Mcl-1 inhibitor 2, with an N-methylated amide, adopts an E configuration at the amide bond (Figure 7a), measured using solution NMR NOEs and conformational analysis (see previously published NMR Supplementary Information26). The E configuration adopted in solution is in accordance with the bioactive conformation seen in the crystal structure of the protein-bound complex (Figure 4b). The bioactive conformation was similarly quickly deduced by the presence of the bioactive reporter signal, highlighted in Figure 7. The chemical shift of this hydrogen atom, being sensitive to the relative geometry of the dichloro-substituted aromatic ring, provides a quick method to assess that the bioactive conformation is highly populated in solution. In the bioactive conformation, the ring currents shift the observed proton peak from an expected 7.6 ppm, predicted by 2D structure based empirical methods (ACD/Labs and MNova), to an observed 6.4 ppm, in agreement with 3D shape dependent QM calculations performed on the bioactive conformation (SI Table S2 and Figure 3a).

In contrast, the matched pair des-methyl macrocycle 8 in Figure 7b adopts a non-bioactive conformation in solution. This is quickly ascertained from inspection of the 1D NMR spectrum. The bioactive reporter signal is no longer shifted to lower ppm values by the ring current effect. A conformational analysis, using NMR NOE distance measurements26 (no crystal structure available), reveals that 8 is pre-organized preferentially into the trans or Z configuration (Figure 7b). The difference in torsion angles affects the position of the functional group substitutions off the main macrocycle. The result is an alteration in the location of the reporter hydrogen relative to the dichloro substituted aromatic ring, changing the electronic environment such that the signal appears at a higher ppm value, and providing a distinctive conformational signature of the non-bioactive state (Figure 7b). The observed chemical shift for the non-bioactive conformation is at 7.1 ppm, approaching the higher empirical predictions of 7.6 ppm (ACD/Labs) or 7.4 ppm (MNova). Due to molecular pre-organization within https://www.selleckchem.com/products/bpv-hopic.html this series of inhibitors, the signal of this reporter hydrogen serves as a quick and useful marker of the bioactive free ligand solution conformation: ~6.4 ppm, the bioactive conformation is highly populated, compared with ~7.1 ppm, when less than 1% of the macrocycle adopts the desired bioactive conformation. This is consistent with the ~30-fold poorer potency of the non-bioactive conformation 8 relative to 2, which preferentially adopts the correct bioactive conformation. The +0.6-fold increase in logD is insufficient to account for the ~30-fold change in potency, further suggesting that the differences in potency are driven medicare current beneficiaries survey by locking 8 into a non-bioactive state.

Consequently, the greater understanding of conformation can significantly affect design hypotheses. sulfone in place of the chlorine atom of the mono-chlorophenyl moiety. The E configuration adopted around the amide bond is also observed in the free ligand in solution, as determined by proton NOEs (see SI of Johannes, et al.26). (b) In contrast, the matched pair NH amide, also conformationally rigid, adopts a non-bioactive trans or Z configuration, corresponding with a global change in conformation of the solution macrocycle (no crystal structure available; conformational analysis determined by experimental NOEs, spectra published previously in SI NMR).267

Within this macrocycle series, the bioactive reporter signal provided a handle to enhance SAR for linker optimization, by using 1D NMR guided SAR to understand the effect of a methyl scan on the linker. Potency will be affected by entropic penalties to binding (conformational flexibility), energy penalties for free ligand conformational rearrangements required for binding (wrong conformation), ligand-protein complementarity (sterics and/or loss or gain of polar interactions), and solvation/desolvation contributions. The individual contributions from each of these different factors are not apparent from the measured enzyme potency values. Separating free ligand preferences for the bioactive conformation from bound ligand complementarity to protein enhances understanding of linker SAR. Figure 8 demonstrates how to deconvolute potency contributions to help focus design strategies aimed to maximize bioactivity. The 1D NMR conformational signature of each molecule shown in this example has a typical difference in chemical shift between bioactive and non-bioactive conformations of ~1 ppm due to differences arising from through-space ring current effects, as described above for this series (Figure 7). In this example, four different linkers which vary in the methyl group position all result in rigid molecules by 1D NMR (complex spectra in the aliphatic region, Figure 8). Three of the compounds are highly populated with the bioactive conformation as demonstrated by the presence of the bioactive reporter signal. An enhanced SAR analysis using free ligand 3D structural information can be achieved with routine 1D NMR spectra acquired for the initial purpose of compound registration. Comparisons across ligands are best made with the same solvent under similar conditions such as temperature, hydration and protonation states, as these can influence the extent of pre-organization and the preferred conformation. This type of standardization works well with automated NMR spectrometers.

The ability to experimentally observe conformational flexibility in solution with atomic-level resolution provides a more complete diagnosis of inadequate bioactivity and is not available by any other methodology. Two linker designs with similarly poor potency may need different strategies for improvement of potency. These macrocycles demonstrate the ability to quickly access atomic resolution experimental data on free ligand flexibility and how to use it to enhance SAR understanding during linker optimization. The combination of NMR
signatures and binding affinities, through comparison of matched pairs or more broadly with compounds in a series, may inspire alternative design hypotheses ultimately focusing synthetic efforts on designs with higher success rates.

The BCL6 series provides a further example of using methylation to introduce rigidity into a linker. In Figure 9, the SAR analysis of matched pair BCL6 macrocycles are reported. The molecule 13 has 7.6-fold lower affinity compared to the methylated analog 14. The increase in potency is not solely attributable to the +0.4 unit increase in logD. Additionally, the racemate of 14 is a weak binder (BCL6 TR-FRET IC50 >10 μM). This supports our thesis that affinity is driven by an increased population of the bioactive state of the free ligand. This can be assessed directly by inspecting the aliphatic region as a marker of pre-organization, without the need of a full NMR conformational analysis.

Free Ligand Conformations and Structure Kinetics Relationships

Biophysical and biochemical methods can provide high quality data on the kinetics of protein-ligand interactions.34 One challenge, though, is how to incorporate this data into the design of new and improved compounds. As conformation, desolvation and binding site complementarity influence onand off-rate constants,35,36 combining NMR conformational signatures and X-ray structures of bound ligands with experimental onand off-rate constants can provide insights suitable to guide design. Combined structural and biophysical data can focus design towards those specific molecular determinants of drug-target binding with the most to gain from optimization efforts.

The Mcl-1 binders encountered in Figure 7 provide an interesting case study. This matched pair exhibits dramatic conformational differences upon small structural changes (N-methylation). Biophysical SPR data is available for comparison to tease out the contributions of ligand pre-organization vs. protein complementarity on binding potency. The two compounds are both rigidly pre-organized in solution. Only the N-methylated amide 2 adopts the bioactive conformation; the NH amide 8 is highly populated in a non-bioactive conformation. Only the bioactive conformations of the free ligand will bind the target, hence once bound, protein complementarity for the two compounds is nearly the same, as evaluated by the similar experimental off-rate constants (Table 1).26 The similar off-rates mean that the ~10-fold difference in binding affinity is driven entirely by a ~10-fold change in on-rate kinetics.

The larger kon of 2 correlates with the higher population of the bioactive conformation in solution. Preorganization into the bioactive conformation is expected to translate to a faster on-rate arising from an increased effective concentration of the free ligand bioactive conformation. The free energy penalty on the trans NH isomer originates from the need to rearrange into the bioactive conformation prior to binding. Consequently, the trans NH free ligand 8 disfavors effective interactions with the protein (unlike the cis N-CH3 isomer) and this results in poorer on-rate kinetics. This shows the power of combining experimental free ligand conformational information with SPR data to validate hypotheses.36-39

The correlation of kon with free ligand pre-organization implies 1D NMR signatures can augment structure-kinetic relationships; when kon is optimized by highly populated bioactive states, potency will be driven by koff. Disentangling ligand pre-organization from protein-ligand complementarity provides a method to evaluate the effectiveness of putative engagement of exposed hydrophobic and polar functional groups to increase residence times. In Figure 10, two structurally different Mcl-1 inhibitors are compared. Both molecules are pre-organized in solution into the preferred binding mode, and consequently, on-rate kinetic constants are similar (Table 1). The 100-fold difference in affinity correlates with the difference in the off-rate, which must reflect a greater protein complementarity / an ability to form further specific hydrophilic and hydrophobic interactions with 1, specifically with the Arg-123 residue.17 and 15) have similar on-rate kinetic constants, as expected from ligand pre-organization into the bioactive state, so the 100-fold difference in affinity is reflected in the 100-fold difference in off-rate kinetics. 10

McCoull, et al.27 used this strategy to increase affinity of BCL6 macrocyclic inhibitors. Three BCL6 inhibitors are compared in Figure 11. The flexible acyclic molecule 16 has the lowest on-rate kinetics, inline with a low population of the bioactive state. In contrast, the macrocyclic analogs 17 and 3 (Figure 11b and c) are fully preorganized into the bioactive conformation, as can be inferred from the 1D NMR signature for a rigid conformation and the strong binding affinity as measured by the dissociation constant (pre-organization into the wrong conformation results in poor binding affinity). The on-rate constants for 17 and 3 are of the same order of magnitude (5.5·106 M-1s-1 vs. 1.6·106 M-1s-1). The difference in binding affinity between 17 and 3 is dictated by a 17-fold decrease in off-rate (0.066 s-1 vs. 0.0006s-1). A docked pose of the macrocycle in b suggested addition of a polar group to the pyrrolidine on the ligand could pick up an extra protein interaction with the proximal Arg-28 residue. Off-rate kinetics were reduced by incorporating an (S)-methanol group to interact favorably with a stable conserved water close to the Arg residue. This decrease in off-rate and improvement in potency is thus derived from increased protein complementarity.27 ºC, 400 MHz) with sub-optimal on-rate kinetics, protein-ligand complex crystal structure is PDB 5N21. (b) A cyclized analog with 50-fold higher affinity significantly assisted by pre-organization into the bioactive conformation (1H, DMSO-d6, 30 ºC, 400 MHz), with 17 docked into PDB 5N21 (see Si Table S4). (c) A further 3.3-fold improvement in affinity is dominated by off-rate kinetics by increasing ligand-protein interactions (1H, DMSO-d6, 27 ºC, 600 MHz), protein-ligand complex crystal structure is PDB 5N1Z. An increase in protein complementarity arises from the engagement of ligand to protein Arg-28 polar interactions, leading to increased residence time. 11

These three examples show that structure kinetic relationships can be rationalized if structural information from both the free ligand and the protein-bound ligand are available. Improvement of ligand design using rigidification strategies that increase the population of free ligand preferentially pre-organized into the bioactive conformation increases ligand affinity (Table 1). Three flexible acyclic ligands, with interconversion rates of milliseconds (broad 1D NMR line widths) and nanoseconds (narrow 1D NMR line widths), are an order of magnitude smaller in kon compared to the potent bioactive pre-organized macrocycles. Notably, optimization of the free ligand conformation into the bioactive conformation drives on-rate binding kinetics, while optimization of protein ligand complementarity most strongly influences off-rate kinetics. Interestingly, the three structurally unrelated potent macrocyclic PPI inhibitors of Mcl-1 and BCL6 are bioactive molecules and display on-rate kinetics of the same order of magnitude ~ 106 M-1 s-1 .

Prediction Methods for Linker Design Based on Free Ligand Conformation

The examples herein demonstrate that preferred free ligand 3D conformations determined from NMR signatures are a valuable means of relating the pre-organization effect to observed affinity changes. Having established this relationship, it would be beneficial to be able to predict the linkers that stabilize the free ligand conformation prior to synthesis, enabling chemists to prioritize only macrocyclic compounds with the correct bioactive conformation. Though a single major conformation may dominate, manifesting as distinct NMR signatures, the major conformation of a ligand is still dynamic and exists as an ensemble of similar conformations in rapid equilibrium. Accurate representation of this ensemble and the relative conformation populations therein is the goal of computational conformation prediction methods for molecules in design.

In general, there are two classes of techniques used to predict free ligand conformation ensembles: rule-based techniques,40-42 whereby the dihedral angle of each bond is systematically rotated through pre-defined energy minima, and simulation-based techniques43,44 where the macrocycle is allowed to move and explore conformational space defined by a molecular mechanics force field. The former technique is much faster (typically seconds per molecule) but is hampered by poorly defined internal strain energy of each conformation. Without an accurate energy, it is impossible to determine relative populations of each conformation. Furthermore, these rule-based methods have been developed primarily for acyclic molecules and perform poorly for macrocyclic ligands.45,46 The latter technique has the potential to be much more accurate, directly quantifying conformation populations as the summation of the frames of a molecular dynamics simulation and being applicable to both acyclic and macrocyclic ligands. However, to ensure complete sampling of the conformational space, particularly in the case of conformationally restricted macrocycles, long simulations (typically hours per molecule) are required.

McCoull, et al.27 used an accelerated simulation method to determine populations accurately in less time.47 The mixed Monte Carlo / Molecular Dynamics (MC/MD) method uses MC for the initial randomized atom positions, followed by a short MD run and minimization to define each conformation. This was repeated 10,000 times for each molecule to ensure adequate sampling and that an equilibrium conformational population had been obtained. Rather than cluster conformations as representative averages, calculations were performed on the entire ensemble. The RMSD was calculated for all heavy atoms in each conformation (except those in the variable linker) compared to the known bioactive conformation from X-ray crystal structures. It was observed that the proportion of the ensemble with low RMSD was large when affinity of the ligands for BCL6 was improved and was consistent with NMR signals showing pre-organization of the ligand for binding. Though an imperfect correlation, this model was subsequently used to prioritize linker designs for synthesis and to optimize ligands for potency. It is interesting to note that compound 3 is an outlier to the trend. Most likely this is a consequence of an additional protein-ligand interaction with the proximalArg-28 residue, as stated above.

In summary, we have shown that free ligand conformational preferences can be derived from routinely available 1D NMR spectra, by using a validation approach relying on QM calculation of conformational proposals and NMR parameters. We have presented three macrocyclic structure-based drug design cases enhanced by the ready knowledge of free ligand conformations. The opportunity for macrocyclization as a ligand rigidification strategy was highlighted by the experimental knowledge that the acyclic inhibitors were highly flexible in solution, while the linker design was informed by the availability of the protein-ligand crystal structure. Effective real time lead optimization with 3D-based structure-activity and structure-kinetics relationships was obtained for all synthesized compounds by using routinely available 1D NMR spectra as markers of conformational preference. The resulting optimized macrocyclic inhibitors, including the clinical candidate AZD5991 (1) are highly pre-organized in solution and display exquisite potency and selectivity as PPI inhibitors, suggesting that ligand rigidity achieved by macrocyclizaton is an important component of their activity. Interestingly, inspection of ligand kinetics reveals that structurally different pre-organized bioactive macrocyclic PPI inhibitors have on-rate kinetics within the same order of magnitude, suggesting that ligand affinity of pre-organized macrocycles is dictated by off-rate kinetics. Notably for the clinical candidate 1 and the DEL-derived series 2 the 100-fold difference in affinity and off-rate, reflects a greater protein complementarity and an ability to form further specific hydrophilic and hydrophobic interactions with 1, specifically with the Arg-123 residue. Conversely, locking a macrocycle into a non-bioactive conformation is detrimental to affinity and more specifically to on-rate kinetics, as demonstrated by the methylated amide matched pair 2 and 8.

In conclusion, we have shown that the ability to read 1D NMR signatures of the free macrocyclic ligands in solution provides medicinal chemists with a quick analytical marker of conformational flexibility and molecular shape and can guide medicinal chemistry design decisions during lead optimization, while focusing synthetic efforts on productive design hypotheses. 1D NMR conformational reporter signals and/or broader signal fingerprints across a range of compounds within a series can be used to assess whether conformational bioactivity has been retained or successfully imparted to newly designed molecules, or conversely, if the ligand has adopted a non-bioactive conformation. The inherent simplicity of using routinely available data for every synthesized compound confers a practical advantage of accuracy and speed, to enable NMR data to influence design decisions between design cycles of a lead optimization project. In our experience, the ability to infer free ligand conformation from routine 1D NMR spectra is applicable to a broad range of drug modalities, whether the ligands are small molecules, or peptides, or more recently established modalities, such as PROTACs48. This approach is compatible with the analysis of compound libraries and opens the door to automation with machine learning approaches, with notable NMR-derived conformational signatures identified without human intervention.

EXPERIMENTAL SECTION

General. Proton magnetic resonance spectra were acquired on Bruker spectrometers, ranging from 300 600 MHz proton resonance frequency, with standard Bruker pulse sequences. Unless stated otherwise, spectra were recorded in DMSO-d6 at 298 300 K, with shifts referenced to the residual proton solvent signal at 2.5 ppm. ACD/Labs22 and MestreLabs23 Research software packages were used for analysis, plotting, and 2D empirical predictions (HOSE Code Algorithms accessed from ACD/Labs). Piperidine (Aldrich, 99.5%) and 2-methyl piperidine (Acros, 99%), solutions used 0.05 mL compound as received into 0.75 mL DMSO-d6 and spectra acquired with a Bruker AvanceIII 600. AstraZeneca macrocycles were obtained internally, and details of the syntheses can be found in the cited literature. 1 atropisomers were separated by chiral SFC as follows: column, Chiralpak ID (4.6 x 100 mm); mobile phase, CO2-EtOH (60:40) at 5.0 mL/min. Kinetics data was obtained using a Biacore direct binding assay and immobilization of target via nickel capture and NHS/EDC amine coupling to a NTA Biacore sensor chip; analyses were performed in the Biacore T200 BIAevalutation software. Images of ligand-target complexes were generated using MOE.24 logD values were determined at pH 7.4 using shake flask methodologies. Alternatively, cLogD values (2 and 8) were calculated using a conformal SVM signature model based on measured data of compounds within and beyond the same series.49

Conformational searches to visualize flexibility by NMR signatures used a lowModeMD search on acyclic compounds using MOE with default values (7kcal/mol energy window). Conformational searches on macrocycles were performed in Schrodinger/Maestro50 or MOE.24 Full conformational analysis of free ligands in solution used ROESY distance constraints acquired with Bruker spectrometers and analyzed using either ACD/Labs22or MNova23. Conversions from NOEs to distances was done in an Excel spreadsheet. MSpin23 was used to select the best match between conformer sets and experimental NMR distances. Quantum mechanical calculations were carried out using Gaussian 16.25 Geometry optimization and NMR chemical shift tensors employed the GIAO DFT method at the B3LYP/6-31G(d,p) level with PCM solvent modelling in DMSO (same as experimental NMR). Chemical shifts were referenced to TMS. Pre-optimized structures were extracted from the X-ray structures of the protein-bound ligands. Bound ligands shown in figures as structural analogs were docked into the given crystal structure using MOE24 template docking.

Chemistry. Unless otherwise stated, commercially available reagents were used as supplied. All reactions requiring anhydrous conditions were conducted in dried apparatus under an atmosphere of nitrogen. 1H NMR spectra were recorded using a Bruker AVIII 300 or AVIIIHD 400 NMR. Chemical shifts δ are reported in ppm, and multiplicity of signals are denoted s=singlet, d=doublet, t=triplet and m =multiplet, respectively. MS were recorded using a Shimadzu LCMS-2020 instrument (ESI+).