Beyond the Rule of 5: Vastly Expanding Targetable Drugs
By Kyle Proffitt
September 9, 2020 | At the Drug Discovery Chemistry Virtual conference last month, several speakers discussed approaches, techniques, and their results in the quest to expand pharmaceutical reach into previously undruggable territory. Lipinski’s rule of 5—the famous set of guidelines for medicinal chemists to consider in designing small molecules with increased odds of becoming successful, orally-bioavailable drugs—serves its purpose for smaller compounds, but these restrictions may not be universally applicable as researchers pursue higher-molecular weight compounds to access a greater range of targets.
Cameron Pye of Unnatural Products (UNP) presented a compelling case for macrocyclic drug discovery. He began by reminding the audience that roughly 70% of the human genome is considered undruggable, 10% can be targeted by big biologics that can’t access the interior of a cell, and the remaining 20% can be targeted with small molecules typically only because they have well-defined binding pockets such as those in receptors and enzymes. To effectively modulate the vast swath of biological activity involving protein-protein interactions (PPIs) and touch the “undruggable” 70%, he says, we must get large molecules inside cells. “Luckily,” says Pye, “nature has been working on this problem longer than we have.” Nature often uses the macrocycle, a sort of “happy medium” between small molecules and giant antibodies that breaks Lipinski’s rules and yet maintains membrane permeability.
Our ability to rationally synthesize effective macrocycles is still in its infancy, but UNP has performed a systematic characterization and classification of the known macrocyclic scaffolds existing in natural product databases, hoping to identify favored scaffolds with which to work. They found that the vast majority of known macrocycles have unique backbones; nature does not often reuse the same scaffold with a few exceptions such as cyclosporins and vancomycins. However, the passive permeability of several macrocycles still encourages further pursuit.
UNP thus developed a unique pipeline involving iterative design aided by machine learning techniques, parallel synthesis of up to thousands of unique compounds at a time, and direct testing of compounds using in vitro assays. The results are then fed back into the models to improve future rounds of design. Though Pye did not include much information about the types of macrocycles generated and the synthetic routes used, he demonstrated how this strategy works for UNP by starting with 3 previously reported molecules that inhibit MDM2-p53 interaction with decent affinity but poor permeability. Within 8 cycles of the UNP iterative design loop—a kind of bouncing back and forth between efforts to improve potency at the expense of chemical properties and then vice versa—they identified cell-permeable macrocycles with high double-digit-nM potency. The approach was also refined to identify first-in-class passively cell-permeable bispecific MDM2/MDM4 inhibitors. The utility of the bispecific compounds was indicated by their enhanced toxicity toward and ability to upregulate p21 in JEG-3 placental choriocarcinoma cells compared with MDM2-specific inhibitors. Pye said that UNP is working in collaboration with top-tier pharma on undisclosed projects and is interested in developing further associations.
Scott Lokey, from the University of California, Santa Cruz, addressed the similar topic of cyclic peptides as pharmaceuticals in his keynote address. The Lokey lab seeks to predict and control the properties of molecules in this class, and Lokey has also returned to some basic principles, showing that small molecules are very limited in the chemical space they can selectively access, whereas antibodies, “have this magical ability to bind to virtually any protein interface”. However, what he finds interesting is that “the interfaces accessed by these antibodies are really not that much larger than the interfaces that can be engaged by much, much smaller molecules.”
Lokey points out that cyclic peptides and stapled helical peptides often rely on charged residues, tryptophans, and arginines, but, “you rarely see stapled helical peptides that only have leucines on the functional side.” However, he showed that many of the natural product cyclic peptides with good biological activity really don’t have very interesting functionality, relying on hydrogen bonding through backbone carbonyls and amides and van der Waals interactions involving simple leucine-like hydrophobic side chains. He added that simply cyclizing and stapling peptides with charged residues does not translate to good membrane permeability and that we have a lot left to learn about how to design potent and permeable macrocyclic peptides.
Lokey highlighted the chameleonicity of cyclic peptides—their solvent-dependent structural perturbation—as a parameter that needs to be better understood. Cyclosporin is an example of a chameleonic compound that displays a much more open conformation when bound to its target but appears more compact with intramolecular hydrogen bonds in NMR and crystal structures. However, other molecules such as argyrin show nearly identical structures in bound and free states.
To help understand how chameleonicity affects compound properties, Lokey then told the story of modifying cyclic peptides by creating peptoids with attachment through primary amines to create peptomers. “We began to explore chameleonicity sort of by accident,” he said. Upon creating two different libraries of cyclic decapeptides based on a scaffold developed at Novartis, each differing by which opposing pair from the 4 available backbone N-methyl groups were substituted with different peptoids of varying lipophilicity, his group found divergent behavior in terms of how quickly permeability decreases as AlogP increases. This “really allowed us to begin looking at chameleonicity as a tunable property,” he said, and to ask, “’What is the effect of chameleonicity on ADME properties?’ in a well-defined synthetic system.” He hypothesized that solvent-dependent behavior was responsible. Variable-temperature NMR experiments confirmed that one series of compounds demonstrated solvent-exposed amides as solvent polarity increases—chameleonic behavior—whereas the other series maintained a more rigid structure independent of solvent. Molecular dynamics studies further corroborated these findings.
Additional analysis included assessment of permeability in PAMPA and MDCK assays, P-gp efflux ratios, microsomal stability, and serum protein binding, thus providing a good survey of how chameleonicity affects these compounds’ characteristics. The group also assessed oral bioavailability, which was generally poor but could be improved greatly by concurrent treatment with CYP and P-gp inhibitors. According to Lokey, “These data suggest that you can have high permeability and high solubility with both chameleonic and rigid molecules, but you have to tune the overall lipophilicity of those compounds.” Again, he summarized by saying “Both chameleonicity and rigidity can afford interesting properties, interesting molecules, as long as the overall properties, especially overall lipophilicity, are taken into account.”
Adrian Whitty of Boston University reported his group’s progress in developing a cyclic peptide to inhibit Nrf2/KEAP1 interaction, a target of interest in various inflammation-mediated processes and diseases. They started with resynthesis of a known 7-mer from Nrf2 in both linear and cyclized forms and assessed binding affinity, revealing significantly reduced affinity in the linear form and reduced affinity from a slightly longer 9-mer. They were surprised by this result because the extra residues in the 9-mer appear to have no significant interaction with KEAP1 in the crystal structure. Overlay of the 9-mer and 7-mer structures bound to KEAP1 revealed strong overlap, and thus they inferred that the reduced affinity of the shorter species resulted from some added strain necessary to adopt the proper pose. Cyclization apparently helped reduce this strain but not enough to match the affinity of the 9-mer.
Whitty emphasized here that compound cyclization introduces constraints in both the unbound and the bound states, and thus there is a delicate balance. Essentially, you want your compound to have limited conformational flexibility in an unbound state, so there’s a small entropic penalty in adopting the correct structure, but at the same time, “you don’t want to rigidify the molecule in any state that’s even slightly different from the bound state,” Whitty said, and suffer an energetic cost from distorting the ligand into the binding state.
Whitty’s group also uses an iterative, machine-learning guided approach to identify macrocycles with Goldilocks-level, just-right flexibility. After design and synthesis, compound affinity and bound and unbound structures are analyzed and used for optimized design, and this approach has successfully produced cyclic peptides with improved binding. Additional principal component analysis of dihedral angles helped reveal how divergence from the orientation seen with the linear 9-mer decreased affinity by requiring the molecules to undergo strain to adopt optimal conformations, and the cyclic peptide with the best affinity closely matched the dihedral angles seen for the 9-mer. Whitty says that they are now working to optimize permeability through charge elimination and exploring new binding energy hot spots.
Editor’s Note: Did you miss the 2020 Drug Discovery Chemistry conference? Because the event was virtual, you can still access the event including all of the recorded sessions, presentations, and materials. Register for PREMIUM POST-EVENT ON-DEMAND.