Kinases include a wide variety of valuable drug targets, but full therapeutic exploitation requires a high degree of selectivity. A promising avenue to engineer the desired kinase selectivity relies on allosteric sites. Here, we provide a focused minireview of recent progress in allosteric modulation of cyclic nucleotide-dependent kinases, including protein kinases A and G. We show how apparently diverse emerging concepts, such as allosteric pluripotency, allosteric nonadditive binding, and uncompetitive allosteric inhibition, are all manifestations of complex conformational ensembles. Such ensembles include not only the typical apo-inactive and effector-bound-active states, but also mixed intermediates that feature attributes of the former states within a single molecule. We also discuss how allosteric responses are amplified by aggregation processes, thus establishing a novel interface between the signaling and amyloid fields. Finally, we critically evaluate the challenges and opportunities for clinical translation opened by these emerging allosteric concepts.
Les kinases englobent une grande variété de cibles médicamenteuses de grand intérêt, mais il faut être en mesure d’atteindre un degré élevé de sélectivité pour pouvoir les exploiter pleinement en thérapeutique. Une voie prometteuse pour parvenir à une sélectivité pour les kinases visées repose sur les sites allostériques. Dans le présent article, nous proposons une minisynthèse ciblée des récentes avancées en modulation allostérique des kinases dépendantes de nucléotides cycliques, notamment les protéines kinases A et G. Nous montrons comment des concepts émergents en apparence divers, tels que la pluripotence allostérique, la liaison allostérique non additive et l’inhibition allostérique non compétitive, sont tous en réalité des manifestations d’ensembles conformationnels complexes. Ces ensembles comprennent non seulement les formes apo (inactives) typiques et les formes liées à l’effecteur (actives), mais aussi des intermédiaires mixtes qui présentent les attributs de ces dernières formes au sein d’une seule molécule. Nous discutons également de l’amplification des réponses allostériques par les processus d’agrégation, établissant ainsi une nouvelle interface entre les champs de la signalisation et de l’amyloïde. Enfin, nous évaluons de manière critique les difficultés et les possibilités de transposition en clinique révélées par ces concepts allostériques émergents. [Traduit par la Rédaction]
The kinase superfamily provides a multitude of drug targets for a variety of pathological conditions related to cancer, neurodegeneration, and infections, e.g., leukemia, Huntington's disease, as well as malaria and antibiotic resistance.1–4 However, with these therapeutic opportunities also comes a major selectivity challenge. It is often challenging to selectively inhibit one specific kinase but not another, because their active sites, where MgATP and substrates dock, are highly conserved (Fig. 1).5 The conservation of kinase active sites is a response to evolutionary pressure to efficiently catalyze phosphoryl transfer from ATP to downstream protein substrates. As a result, kinase inhibitors that directly target active sites, for example by mimicking ATP, are often poorly selective and prone to induce unwanted side effects.
A viable means to overcome the kinase selectivity challenge relies on allosteric modulators. Ligands that target kinase allosteric sites are more likely to be sufficiently selective because allosteric sites are not subject to the same degree of evolutionary pressure for conservation as active (or orthosteric) sites (Fig. 1). Allosteric loci are not only more diverse than their orthosteric counterparts, but they also offer the added advantage that they are less likely to bind endogenous ligands, such as active site substrates. Hence, allosteric modulators may also exhibit higher effective potency than orthosteric inhibitors. Given the potentially higher degree of selectivity and potency of allosteric vs. orthosteric ligands, the former provide an appealing avenue to fully exploit the therapeutic potential of protein kinases.6–9
Once an allosteric ligand for a given allosteric pocket is identified as a potent and selective inhibitor of a kinase target, another challenge remains due to allosteric pluripotency.10–14 The response of a kinase to an allosteric modulator is pluripotent when it depends on the specific metabolic and proteomic environment (Fig. 2). For example, a single allosteric ligand may act as a partial agonist for a kinase within a given subcellular compartment but as an antagonist or inverse agonist for the same kinase with a different subcellular localization (Fig. 2). This type of allosterically pluripotent response has been predicted on theoretical thermodynamic grounds based on the ensemble allosteric model (EAM),15 and it is also a clinical reality.16 For example, tamoxifen functions as an inhibitor for the estrogen receptor in breast tissues but as an activator for the same receptor in uterine tissues.16
Protein Kinase A (PKA)
Another notable example of allosteric pluripotency is provided by Protein Kinase A (PKA). PKA is a major signaling hub17 and has historically served as a prototype for the whole protein kinase superfamily.18,19 In addition, PKA hosts multiple inherited mutations related to several diseases, from Cushing's syndrome to the Carney complex and acrodysostosis.20–25 Under resting conditions, PKA exists as an inactive complex in which a dimeric regulator subunit (R) binds and inhibits two equivalents of the catalytic subunit (C) (Fig. 3).10,11,14,26–29 The critical segment of R needed to bind C spans an inhibitory site (IS) and two tandem homologous cyclic nucleotide-binding domains (CNBD-A and B; Fig. 3).26,28 The CNBD-A is connected to the IS through an N-terminal linker (Fig. 3).
Upon cellular stimulation by extracellular signals (e.g., hormones targeting G-protein coupled receptors), levels of cyclic AMP (cAMP) increase above the dissociation constant for the C-bound CNBDs so that cAMP binds to the R-subunits of PKA.18,26,28,30 When cAMP binds to the tandem CNBDs, they undergo structural changes to conformations incompetent to inhibit the C subunit. In addition, cAMP binding causes a transition of the relative domain orientation from an open to a closed topology (Fig. 3).18,26,28 The combined effect of such intra- and inter-domain conformational changes leads to the release of uninhibited C subunit, which starts to catalyze the phosphorylation of downstream protein substrates.26,28 The allosteric conformational changes induced by cAMP in the R-subunit are responsible for the cAMP-dependence of PKA's kinase activity.
Allosteric modulation of PKA
Given the central role of cAMP in the allosteric control of PKA, hundreds of cAMP analogs have been synthesized and screened as potential allosteric inhibitors of PKA. Despite intense efforts in this direction, to date, most cAMP-like antagonists are derivatives of a single analog, i.e., the phosphorothioate Rp-cAMPS (or Rp in short), in which the exocyclic equatorial phosphate oxygen atom is isolobally replaced with a bulkier and more polarizable sulfur atom (Fig. 4).10,11,31–35 While this oxygen-to-sulfur substitution may appear relatively subtle, the functional implications are quite dramatic. In the presence of excess MgATP, Rp acts as a very effective antagonist even at mM Rp concentrations.34 However, as the MgATP concentration decreases, Rp starts to function as a partial agonist.34 The marked MgATP dependence of the Rp-induced agonism with respect to PKA activation means that Rp is an allosterically pluripotent ligand. Such allosteric pluripotency is quite unique to Rp, as cAMP and the diastereoisomer Sp, in which the sulfur is in the axial position, consistently act as agonists irrespective of the MgATP concentration (Fig. 4).34
While the original observation of allosteric pluripotency for Rp dates back around three decades,34 an explanation of the underlying molecular mechanism of action has become available only recently.10,11 A first important step in this direction has been the crystal structure of a two-CNBD construct of the R-subunit in complex with two equivalents of Rp (R:Rp2).28,36 Interestingly, the conformation of this ternary complex was found to be highly similar to its cognate complex with cAMP (R:cAMP2; root mean square deviation ∼0.5 Å; Fig. 3),28,36 thus potentially accounting for the Rp agonism but not its antagonism or the switch between the two. Hence, we took a different approach aimed at probing the structural dynamics of the R-subunit and relating it to function. We refer to such experimental design as ThermoDynamics, where the capital D emphasizes the critical role played by flexibility (Fig. 5).
ThermoDynamics is a three-pronged strategy based on NMR, EAM, and enzymatic assays (Fig. 5). NMR experiments such as paramagnetic relaxation enhancements (PREs) are ideally suited to gauge interdomain orientations, while chemical shift-based analyses, such as the CHEmical Shift Covariance Analysis (CHESCA) and CHEmical Shift Projections Analysis (CHESPA), probe intradomain conformational transitions in the fast-exchange regime.29,37–39 Briefly, PREs rely on the introduction of paramagnetic spin labels causing enhanced relaxation of the spins in their proximity. This technique offers long-range (up to ∼20 Å) structural information.29 CHESPA and CHESCA are methods to analyze chemical shifts and map allosteric networks. Both CHESPA and CHESCA assume that the dynamic conformational equilibria of allosteric couplings occur in the fast chemical-exchange regime, in which case the observed chemical shifts form linear patterns representing population-weighted averages of conformer-specific values.40
The use of PREs, CHESCA, and CHESPA makes it possible to identify relevant conformational states sampled by the R subunit as well as their relative populations. Such states are the unbound apo-state, the cAMP-bound state (active), or C-bound state (inactive). The latter are then converted into normalized Boltzmann factors to recapitulate the free-energy hierarchy of the underlying EAM. The EAM is parametrized through further ad hoc NMR experiments, such as ligand titrations and H/D exchange.11,12 The parametrized EAM model allows for quantitative predictions of kinase activation, which are tested through enzymatic assays.29,37–39 The functional validation of the EAM model informs the design of new protein mutants or ligand analogs to be investigated through a new iteration of the three-pronged ThermoDynamic cycle (Fig. 5).29,37–39
For the sake of illustrating the ThermoDynamic approach, we will focus on a single well-resolved TROSY cross-peak, i.e., arising from L221 in CNBD-A, which is sufficiently far from the cAMP- and C-binding interface to report primarily on the conformational (off vs on) equilibria of CNBD-A.29 When TROSY spectra are acquired for a two-domain construct of R either unbound (apo; black) or bound to C (inactive; red) or cAMP (active; green), a linear pattern is observed (Fig. 6) pointing to a fast exchange between active and inactive conformations.29 The intermediate position of the apo cross-peak relative to cAMP-bound (active) and C-bound (inactive) cross-peak positions is a weighted average of the active and inactive chemical shift ppm values, which report directly on the relative active vs. inactive populations. Hence, the central position of the apo cross-peak (Fig. 6) is indicative of a relative degenerate free-energy landscape for CNBD-A (Fig. 6). A similar scenario is observed for the CNBD-B in apo R.
The degeneracy of the free-energy landscape of apo R is compromised by Rp binding.10 The analysis of the L221 cross-peak patterns for the R:Rp2 ternary complex reveals the presence of two sets of peaks in slow exchange (Fig. 7, blue): a major set at a position quite similar to the active R:cAMP2 ternary complex, and a minor set at a position quite similar to the apo R (Fig. 7, blue).10 Similar patterns were observed for several other CNBD-A residues of R:Rp2, while for CNBD-B only one set of peaks is observed at a position similar to the active R:cAMP2, in spite of the sequence and structural homology between the two CNBD domains. One of the simplest models accounting for these observation posits that the conformational ensemble sampled by R:Rp2 includes three main conformers (Fig. 7).10
The ground state adopts a closed-topology similar to R:cAMP2 (Fig. 7), also explaining the X-ray results. The other two states are less populated and feature X open topology with CNBD-A switching between C-inhibition competent and incompetent conformations (Fig. 7).10 Notably, when similar analyses were repeated for the quaternary C:R:Rp2 complex, the two states corresponding to the minor set in R:Rp2 have now become the ground states, indicating that the excited states of R:Rp2 are both C-inhibition competent.10 Based on these observations, we could recapitulate the free-energy diagram of R:Rp2 both in the absence and presence of C and build an EAM for allosteric pluripotency of Rp (Fig. 8).10
To appreciate the implication of the EAM for the Rp allosteric pluripotency, it is sufficient to focus on two critical free-energy differences, i.e., the ΔGgap and the ΔGbinding (Fig. 8).10 The former is defined as the free-energy difference between the ground noninhibitory state and the lowest energy excited inhibitory state of R:Rp2 (Fig. 8). The latter refers to the effective free energy of C-binding to R:Rp2 (Fig. 8). The relative magnitude of ΔGgap and ΔGbinding dictates the allosteric pluripotency observed for Rp.10 For example, at high MgATP concentrations, the R:C affinity is high as MgATP is a cofactor stabilizing the interactions between R and C. As a result, |ΔGbinding| > |ΔGgap|, which means that in the presence of C a stable C:R:Rp2 inhibitory complex forms, i.e., Rp acts as an antagonist. However, as the MgATP levels decrease, the R:C affinity is lowered until |ΔGbinding| < |ΔGgap|, in which case a stable C:R:Rp2 inhibitory complex cannot form, i.e., Rp acts as an agonist. Therefore, the EAM fully accounts for the MgATP-dependent allosteric pluripotency previously observed for Rp. Furthermore, the NMR-based EAM also unveils new previously unanticipated drivers for allosteric pluripotency.10
Another critical determinant of pluripotency is the concentration and affinity of PKA substrates.10 Since the R subunit acts as a competitive inhibitor of the C subunit, high concentrations of substrates with low Km values reduced the effective R:C affinity, leading to a scenario in which |ΔGbinding| < |ΔGgap|, which, in turn, favors agonism. In contrast, low concentrations of substrates with high Km values are expected to reduce the degree of agonism. These EAM predictions have been quantitatively confirmed by kinase assays, thus illustrating the effectiveness of the combined NMR/EAM approach in dissecting the drivers of allosteric pluripotency and understanding how local metabolomic and proteomic context affects the allosteric response of PKA to Rp.
In general terms, the thermoDynamics approach has revealed a simple but effective explanation for allosteric pluripotency in terms of a threshold-dependent functional response to Rp (Fig. 8).10,11 When the |ΔGbinding| value is comparable to the critical threshold defined by |ΔGgap|, the dependence on the MgATP metabolite and substrate proteins is particularly acute, and phenomenological manifestations of allosteric pluripotency are most evident, as discussed above. When |ΔGbinding| >> |ΔGgap|, consistent antagonism is expected, while when |ΔGbinding| << |ΔGgap|, consistent agonism is predicted.11 This is the case of cAMP and Sp-cAMPS that selectively stabilize the ground noninhibitory state relative to the excited inhibitory state (Fig. 8), as they favor the active conformation of CNBD-A. This is also the case of the R209K mutant of PKA R, which effectively removes the state-selective frustration, another key driver of allosteric pluripotency in R:Rp2, as it would otherwise destabilize the ground noninhibitory state relative to the most stable excited inhibitory state (Fig. 8).11
The |ΔGbinding| << |ΔGgap| scenario, for which consistent agonism is anticipated, applies also to Cushing's syndrome C-subunit mutants that weaken the affinity for the R-subunit.41,42 In this case, Rp is expected to function as an agonist, thus aggravating the overactivation of PKA C typical of this pathological condition. A possible solution to this problem may come from the use of drug combinations. For example, Rp could be paired with an additional ligand targeting the CNBD-A/B interdomain interface with the goal of destabilizing the ground noninhibitory state relative to the most stable excited inhibitory state and thus decreasing |ΔGgap| (Fig. 8). Hence, models of allosteric pluripotency may serve as a guide in the choice of multi-drug combinations, which are notoriously challenging to screen systematically.
Another notable feature revealed by the Rp allosteric pluripotency model is the central role played by mixed intermediates, such as the most stable excited inhibitory state (Fig. 8). Such intermediate states are called “mixed” because they feature within a single-molecule attributes of both active and inactive states. For example, the most stable excited inhibitory state of R:Rp2 unexpectedly adopts the inactive state for CNBD-A but the active state for the homologous CNBD-B domain. Such mixed states may occur also within a single domain, as in the case of Plasmodium falciparum Protein Kinase G (PfPKG), which is a malaria drug target,43 and of the exchange protein directly activated by cAMP (EPAC), which is a cardiovascular drug target.13,44–49
Allosteric modulation in other cyclic nucleotide-dependent systems
In the case of PfPKG, the allosteric partial agonist 8-NBD-cGMP acts by stabilizing a mixed state within the CNBD that controls the closed-to-open topology transition at the basis of the cGMP-dependent PfPKG activation.10,43 Such mixed intermediate achieves inactivation of this kinase by disengaging only the lid region capping the cyclic nucleotide base.10,43 The advantage of reaching nearly complete inhibition with minimal disengagement of the structural elements that contact the cNMP ligand is that high potency is preserved, comparable to the endogenous cGMP effector.10,43 If PfPKG inhibition were obtained without sampling mixed intermediates and simply reverting the two-state inactive–active equilibrium, a nearly full designment of the cNMP recognition elements would be necessary, which typically results in significant affinity losses. Hence, mixed intermediates are pivotal in reconciling high inhibitory efficacy with high inhibitory potency by allosteric modulators.10,43
In the case of EPAC, mixed intermediates within the CNBD that controls autoinhibition play a central role in explaining partial agonism and uncompetitive allosteric inhibition by noncyclic nucleotide ligands, such as I942 and CE3F4R, respectively.13,40,44,48,50–54 The recurrence of such mixed states is in retrospect not a surprise if we consider that allosteric modulators often include multiple allosteric drivers within the same ligand.55 If different drivers exhibit different active vs. inactive state selectivity, they will drive adjacent domain moieties to adopt different states, thus giving rise to mixed intermediates.12
The realization that mixed states simply reflect the mixed drivers present in several allosteric ligands led us to investigate more systematically the couplings between two allosteric drivers within the same ligand.56 Allosteric couplings between two functional groups within the same ligand are particularly relevant when they are sufficiently far from each other to assume only negligible reciprocal covalent influences. One example of such allosteric drivers is provided by the equatorial sulfur in Rp and the guanine base in cGMP. To systematically and quantitatively evaluate the allosteric couplings between the two drivers, we relied on a double-ligand cycle, conceptually analogous to the double-mutant cycles originally proposed by Horovitz and Fersht, in which thermodynamics measurements of the wt protein are compared to similar observables for the two single mutants, as well as the double mutant.57 Similarly, a double-ligand cycle is composed of the complex between the allosteric protein and the unmodified ligand (e.g., cAMP), as well as the two complexes with the singly substituted ligands (e.g., Rp-cAMPS and cGMP) and the complex with the double substituted ligand including both allosteric drivers (e.g., Rp-cGMPS; Fig. 9).56
By measuring the affinities (Kd values) for the four protein:ligand complexes in the double-ligand cycle (Fig. 9D), it is possible to quantify the free energy of coupling between the two allosteric drivers as −RT ln γ, where γ is the nonadditivity factor defined as the ratio of the product between the Kd values of the two singly substituted ligands and the product between the Kd values of the doubly substituted and unsubstituted ligands (Fig. 10).56 In addition, if it is possible to measure the populations of active and inactive states in each of the four complexes within the cycle, it is also possible to dissect the experimental nonadditivity factor γ into three contributions arising from the concerted active–inactive conformational transition, as well as two state-specific contributions (Fig. 10).56
For the purpose of dissecting the observed γ factor into transition- and state-specific contributions, state population can be measured through methods such as Chemical Exchange Saturation Transfer (CEST), NMR dispersion (NMRD), or CHESCA Singular Value Decomposition (SVD) suitable for systems in which the active–inactive exchange occurs in the slow, intermediate, and fast regimes, respectively.29,37,40,51,58–61 The application of such approaches to the Rp-cGMP allosteric ligand of the hyperpolarization and cNMP-modulated ion channels (HCN) revealed that the most significant contributions to the coupling between the equatorial sulfur and guanine base allosteric drivers arise from the active state, most likely due to its compact nature, with only negligible contributions from the less compact and more open inactive state 1 (Fig. 10).56
More recently, we extended similar approaches to understand the cGMP vs. cAMP selectivity of human PKG (hPKG) by focusing on one of CNBDs that controls the hPKG autoinhibitory open-to-closed transition.62 We found that the cGMP vs. cAMP affinity differential arises mainly from nonadditive binding contributions of the 6-oxo and 2-amino guanine base substituents.62 A key contributor to such nonadditivity is mutual protein–ligand conformational selection whereby not only the ligand select a given protein conformation, but also the protein preferentially binds one ligand conformation.62 Mutual conformational selection is distinct from double conformational selection, whereby a single ligand selects conformational states for two tandem-coupled domains.29
The case studies discussed above illustrate how mechanisms involving equilibria between conformational states within a single phase (solution) explain a diverse range of allostery-related phenomena, from allosteric pluripotency to nonadditive binding. However, we recently encountered an instance in which similar allosteric mechanisms were not sufficient to explain the observed functional phenotype. This is the case of the A211D Carney complex-related mutation in the R subunit of PKA isoform 1a. Although this mutant exhibits significantly reduced affinity for the endogenous allosteric activator (i.e., cAMP) compared to wt, it induced overactivation of PKA, which, in turn, results in a generalized tumor predisposition typical of the Carney complex.63
A viable hypothesis to explain this apparent paradox of PKA overactivation in the A211D PKA R mutant with low cAMP-affinity is that PKA is subject to aggregation-induced noncanonical activation.63 While the canonical activation of PKA is cAMP-dependent, as discussed above (Fig. 3), the noncanonical activation of PKA is caused by the self-association of apo R subunits into oligomers that are not competent to inhibit the C subunit.63 The reduced cAMP-affinity of the A211D R subunit mutant increases the amount of apo R subunit, which is partially unfolded and more aggregation-prone than cAMP-bound R subunit. A similar mechanism was also discovered for another Carney-related mutant (G287W PKA R subunit) and aggregation-induced losses of function are also known for mutants of another tumor suppressor different from PKA R, such as p53.64,65 In this respect, it is possible that aggregation-prone intermediates provide an effective means to amplify allosteric responses to disease-related mutations.
In conclusion, we have shown how by combining NMR, EAM, and enzymatic assays, it is possible to map the mechanism of action for several diverse allosteric inhibitors targeting cyclic nucleotide-dependent systems. Despite the diversity of such allosteric modulators, the underlying mechanisms share a common feature—the pivotal role of mixed intermediates in which different moieties of the same molecule adopt different active or inactive-like local conformations. Mixed intermediates explain a wide range of emerging concepts, from the concurrent maximization of efficacy and potency to uncompetitive allosteric inhibition, nonadditive binding, and allosteric pluripotency. The latter calls for a redefinition of allosteric drug targets beyond the isolated protein to include also its local metabolic and proteomic environment. Hence, allosteric pluripotency implies that subcellular localization and possibly also liquid–liquid phase separation (LLPS) into membrane-less compartments may be key determinants of allosteric responses. The high local protein concentration typical of LLPS may also promote transitions toward fibrillar aggregates, which lead to losses of inhibitory function, thus amplifying the allosteric effect of disease-related mutations. Furthermore, models of allosteric pluripotency may also guide the selection of effective multi-drug combinations, which are notoriously difficult to screen systematically due to combinatorial complexity.
We thank Dr. Rashik Ahmed, Dr. Stephen Boulton, Dr. Naeimeh Jafari, Katherine Van, and Hongzhao Shao for helpful discussions. Financial support provided by the Canadian Institutes of Health Research Grant 389522 (to G. M.) and the Natural Sciences and Engineering Research Council of Canada Grant RGPIN-2019-05990 (to G. M.).
Atwal R. S.; Desmond C. R.; Caron N.; Maiuri T.; Xia J.; Sipione S.; Truant R. Nat. Chem. Biol. 2011, 7(7), 453.
Karla Martinez Pomier, Madoka Akimoto, Jung Ah Byun, Mariia Khamina, and Giuseppe Melacini. Allosteric regulation of cyclic nucleotide-dependent protein kinases. Canadian Journal of Chemistry.
100(9): 649-659. https://doi.org/10.1139/cjc-2021-0359
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.