How residence time works in allosteric drugs

Introduction

Drug discovery aims to identify optimal protein targets whose inhibition can result in disease treatment. It strives to achieve the desired effect by producing effective drugs that target these proteins. Drug residence time can be considered as the crucial factor to effectively accomplish it, making the understanding of how to lengthen it vital [1,2]. Here we explain that to extend the time the drug stays bound to the target requires considering how the drug works, which differ in orthosteric and allosteric drugs. The distinct working principles breed distinct residence time mechanisms [3].

Orthosteric drugs bind at the active site. The challenge they are facing is side effects from binding to homologous proteins. They need very high affinity to selectively accomplish target-only binding, and long residence time, which is a function of drug and target concentration and thermodynamic and kinetic parameters that depend on drug-active site matching [4]. Allosteric drugs bind elsewhere on the protein surface [5-10]. They work by remodeling the free energy landscape [11-17]. They shift the protein’s conformational distributions, reshaping the active site [18]. Population shift can be effectively accomplished by drug atoms ‘pushing’ or ‘pulling’ protein atoms/groups at the allosteric site [19], triggering perturbation that propagates in the protein structure through preferred pathways, reaching the functional site target. The precise designs of their topography and (steric, electrostatic) nature are critical for optimal allosteric outcome. Affinity is important. More important is the allosteric efficacy; that is, the population shift. Population shift is the mechanism through which allosteric cooperativity is achieved [20]. Cooperativity between the allosteric drug and the active site, through the population shift, can extend the orthosteric drug residence time. This argues that small molecule inhibitor designs should attempt to optimize cooperativity by judicious selection of ‘push’ and ‘pull’ atoms. It also argues that computational methods aiming at scoring allosteric drug binding should consider identifying these atoms and weighing their contributions accordingly. Their location in the allosteric site should further be thought of in terms of the pathways they trigger to the active site.

Multiple factors influence drug residence time and assessing it (e.g., [21-29]). Here we focus on the allosteric effect, and the mechanism through which it exerts the specificity and residence time of orthosteric drugs. At the same time, allostery (cooperativity) is bidirectional [30], implying that it is not only the allosteric drug that influences the orthosteric drug. Orthosteric drugs cooperatively influence the residence time of allosteric drugs. Below, we clarify the differences in the mechanisms of the orthosteric and allosteric drugs and how these are accomplished. We discuss the mechanism of allosteric drugs and its relation to drug specificity, affinity, and residence time, and describe population shift as the means through which cooperativity is exerted and how it influences timescales. Finally, we provide a recent K-Ras^G12C drug targeting both ON and OFF states as an example and conclude with a discussion of what makes an allosteric effector strong.

The mechanism of allosteric drugs: specificity, affinity, and residence time

Drug specificity is crucial [1]. If lacking, the outcome can lead to toxicity, commonly the consequence of interactions with homologous, non-target proteins. Specificity relates to affinity, which is associated with the extent of structural (geometrical) and chemical matching of a drug to the protein receptor. Good pharmacology requires good matching. Poor matching leads to short residence time and limited drug action.

As to affinity, high-affinity binding is highly specific binding [31,32]. However, even though specificity and residence time are crucial for both orthosteric and allosteric drugs, the mechanisms of how they are achieved differ between these drug types. Orthosteric drugs are competitive drugs. They compete with the substrate for the active site, which requires higher affinity (matching). Their binding needs to block the active site pocket, which is not the case for allosteric drugs. Allosteric drugs do not compete with ligands; instead, they can modulate the receptor activity even if an endogenous ligand is bound to another site. Binding away from the active site, they work by leading to a conformational change at the second binding site through multiple allosteric transitions in the population (Fig. 1). That is, they do not actively change the shape of the molecule. Instead, allosteric modulators, like drugs, act by altering the relative stabilities in the conformational ensemble, making the conformer which already has the desired active site shape more stable compared to other conformations through its interactions. This is a vastly important principle to understand. Allosteric drugs mimic allosteric rescue mutations, which rescue native function by stabilizing the functional conformation [33,34]. Both act by cooperative allosteric communication between the allosteric and functional sites. In our allosteric drug case, the interactions of the conformer whose active site shape matches the orthosteric drug, stabilizes it, leading to the allostery-fomented population shift toward this conformation. Design-wise, small molecule allosteric drugs are selected based on restoring orthosteric drug efficacy in drug resistance, making strategies that combine allosteric modulators with orthosteric drugs advantageous. Rescuing the therapeutic effect upon emergence of drug resistance mutations has been emphasized in design [35,36], in addition to high specificity [37,38].

To understand how affinity and residence time work in allosteric drugs as compared to orthosteric drugs, consider the tenets of allosteric drugs. The most important among them is that allosteric drugs are based on the energy landscape, which describes proteins statistically [39]. Different than the classical allosteric model of Monod-Wyman-Changeux (MWC) [40], which depicted allosteric conformational change as resulting from a transition between two preexisting states, we proposed that there are many preexisting—interconverting—states. And different than the Koshland-Némethy-Filmer (KNF) model [41], who suggested a sequential, induced conformational change, we suggested a shift in the conformational landscape, resulting from the many transitions—among multiple preexisting—conformations [42]. Both the MWC and the KNF models consider proteins rigid, crystallographic states, not ensembles of interconverting states around the native minima with different energies separated by barriers.

Allosteric drugs displace water molecules. Drug atoms push (pull), those of the protein with which they are in contact. Newly created interactions are optimized, necessitating movements of residues. The resulting perturbation of the protein structure at the allosteric site creates strain energy, whose dissipation propagates, altering the relative stabilities of the conformations. The propagation—through multiple pathways—drives the population shift of the preexisting conformations, with strain release being the driving force of the transitions. Larger strain energy may promote a larger conformational and dynamic change. Such a view of the cooperativity through the multiple transitions of the populations could not be scripted by the MWC or the KNF models, which are based on crystallographic snapshots, a single pathway, and a conformational change of the active site. In contrast, our landscape view assumes multiple pathways, with some more favored than others, multiple preexisting conformations, some already with changed active site. Our allosteric landscape view explains that the population shift involves stabilization of some conformers, that is, an increase in their propensities. The population shift also influences dynamics, altogether making the orthosteric drug binding more (or less) favorable by cooperatively decreasing or extending the orthosteric drug residence time.

Cooperativity is thus the major determinant [20,43,44]; cooperativity rather than affinity. That is, allosteric drugs determine the orthosteric drug residence time by the nature and extent of the population shift that they promote, thereby modulating the active site conformation. At the same time, considering their cooperative nature, the effect is bidirectional; orthosteric drugs can increase (decrease) residence time of allosteric drugs. This cooperativity can help in predicting preferred allosteric sites [45].

Population shift: cooperativity and timescales

Cooperativity can be dominated by enthalpy and/or entropy [42]. Enthalpy-dominated, positive cooperativity indicates tightening of the drug-protein interactions. It is coupled with an entropic penalty due to loss of disorder. The outcome is a conformational change. Entropy-dominated positive cooperativity pays the cost of the orthosteric and allosteric binding, which is less than their sum. It may or may not be accompanied by conformational change. Negative cooperativity commonly has unfavorable entropy and a conformational change.

Timescales matter [46]. Drugs select the most compatible protein conformation. Binding may induce minor optimization of the protein-drug interaction. The allosteric perturbation signal propagates via multiple pathways. The preferred pathway reaches the active site. It cooperatively tightens the orthosteric drug interactions, and (or) alters the conformational dynamics, shifting the population toward the stabilized orthosteric drug-compatible conformation (Fig. 2). This scenario establishes that the heights of the barriers separating the on-path conformational states can indicate the timescales of the allosteric transitions. Since the allosteric signal involves multiple barrier-crossing transitions, this cooperative process takes time. In contrast, in the KNF induced fit model, the allosteric drug induces the protein to adapt its active site shape to the bound orthosteric drug, which is a faster process. Which scenario takes place depends on the drug dose [47]: if very high—normally beyond the drug prescription levels due to side effects—induced fit is expected. The reason why induced fit prevails under high (but not low) concentration is that in solution, the population of the conformation that fits the drug shape is likely to be low. However, in the KNF scenario, the drug can bind to any protein shape. The different kinetics clarify why experiments and simulations may not capture the selection of the sparse conformer, despite its being the key mechanism [48,49].

Drugs that bind and effectively perturb residues on major propagation pathways, can promote effective population shift [46], which can increase positive cooperativity, enhance the affinity, and most importantly, residence time. With a limited increase in drug concentration, allosteric activation shifts the population toward the active state, overcoming sizeable but surmountable free energy barriers between preexisting conformations, restoring the dynamic equilibrium depleted by the open conformation. Hence, identification of major pathways in the ensemble is an important goal.

Focus on cooperativity

The allosteric effect is determined by the allosteric efficacy, not the absolute (overall) ligand affinity [30,42]. Affinity relates to the strength of the binding, commonly measured by the equilibrium dissociation constant (K_d). Residence time is inversely related to the dissociation rate (K_off). Both affinity and residence time are crucial for drug efficacy. However, residence time is more important. The lifetime of the drug-target complex, not the affinity, is the critical element of the therapeutic outcome. It is also more important for allosteric drugs than for orthosteric drugs. Residence time, a non-equilibrium variable, quantifies the lifetime of the drug-target complex. For allosteric drugs, it relates to the time it influences the target protein by shifting the equilibrium of its conformational distributions. The extent of the shift from the active to the inactive state (vice versa for repressors) is the key determinant of the duration and strength of the inhibitor (modulator). Effective cooperativity can extend drug residence time (Fig. 2). Understanding how allosteric population shifts influence drug residence time is crucial for designing more effective and selective therapeutics. Cooperativity is bidirectional; binding at the functional site will increase (decrease) binding affinity at the allosteric site.

In the absence of allosteric drugs, the residence time of orthosteric drugs is influenced by e.g., binding kinetics (association and dissociation rates), drug metabolism, and target vulnerability [50].

Implications for design: the K-Ras^G12C and Hsp90 family N-glycosylation examples

Understanding how allosteric population shifts influence drug residence time is crucial for designing effective therapeutics [46]. A new example clarifies the importance of allosteric population shifts in drug discovery [51]. Recently, a first-in-class small-molecule covalent K-Ras^G12C inhibitor, BBO-8520, was developed [52] (Fig. 3). Inhibition of K-Ras mutants has been a long-sought aim [53]. Kevan Shokat pioneered a breakthrough, a covalent drug against K-Ras^G12C that binds to the inactive (OFF) state. The covalent bond created by the drug with Cys12 of K-Ras^G12C mutant, shifted the conformational distribution, exposing a hitherto unobserved sparse cryptic pocket conformation, which was not captured earlier by crystallography. Cryptic pockets are allosteric pockets, common in all dynamic proteins. The question is whether they are sufficiently ‘good’ to stably hold a drug. In the case of K-Ras, it is not, which is why the Shokat team used a covalent drug. The drug-bound conformation hampered the exchange of GDP by GTP, required for Ras signaling. Subsequent K-Ras^G12C innovations harnessed companion drugs to upstream regulators or downstream kinases. However, acquired mutations, and the pre-existing mutation load, seeded growth of insensitive cell clones, eventually leading to drug resistance and cell proliferation. The new K-Ras^G12C covalent BBO-8520, stabilizes the conformation with the pocket, making it spend more time in this state, which the drug matches well. BBO-8520 binding leads to subtle pocket conformational changes, which optimize the interactions of the drug with both the inactive and the active states. Especially noteworthy are the improved timescales—now down to 30 minutes [51]. To understand this improvement at the conformational level, we again consider the energy landscape. Binding of the first-generation K-Ras^G12C inhibitors sotorasib (Lumakras) and adagrasib (Krazati) to the sparse inactive state required a shift of the ensemble from the active state population toward the inactive state, crossing barriers. As we discussed above, this takes time. Here, BBO-8520 fits the cryptic pocket, requiring only minor modifications, resulting in a significantly faster event. Thus, the drug achieves a more rapid docking, culminating in effective inhibition of both active (ON) and inactive (OFF) states of K-Ras^G12C by well-matched states. Since in our case the drug is covalent, the residence time is expected to be long.

Post-translational modifications (PTMs) are allosteric effectors [54]. Learning their allosteric mechanisms was elegantly offered as a strategy to accomplish longer residence times of ligands (drugs) in the active site. N-glycosylation at Gly 62 induces Grp94, a member of the Hsp90 family, pathological protein conformations. A recent work, exemplified by Grp94, suggested how the pathological N-glycosylation can be usefully employed [55]. The authors hypothesized that drugs that target conformations induced by this cancer specific glycosylation are likely to have longer residence times at the active site. To test this, first, their molecular dynamics simulations identified specific Grp94 conformations induced by the N-glycosylation. These were integrated with dynamic ligand unbinding data, structure activity relationship (SAR) data, conformational properties of the binding site, and computed the N-glycosylation-induced modifications of the interaction networks of Grp94. The pathological conformations were shown to advantageously guide selection of protein ensembles whose targeting can act as a functional switch.

Conclusions: What makes an allosteric effector strong

An allosteric effector is ‘strong’ if the population shift that it promotes is productive, propagating through optimal pathways to the active site, competently modulating its detailed shape, thus drug residence time. This suggests that computational approaches to predict the allosteric drug binding sites and allosteric drug design should consider not only the shape and chemistry of the site, but also ‘push’ and ‘pull’ residues that the drug could contact [19], and the allosteric communication pathways. Also different from orthosteric drugs, allosteric drug discovery design requires an effective allosteric population shift, producing a system where a small molecule can significantly alter the relative propensities of different conformational states of a protein. This can be accomplished by stabilizing the active state and/or destabilizing the inactive state. It also requires producing interactions that can produce an adequate conformational change. Protein function (dysfunction) relates to the relative propensities of its conformational ensembles [56,57]. Molecular dynamics simulations, Markov state and free energy perturbation (FEP) models can help predict and assess the system’s dynamics and conformational changes. Machine learning approaches, cryptic pocket predictions, and ensemble-based mutational scanning [58], can help determine communication pathways. NMR spectroscopy and cryo-electron microscopy can help verify conformational changes. Altogether, protein allostery is powerful in rational anticancer and broadly drug design [59].

Supplementary Material