Resistance to Allosteric Inhibitors

Abstract

Allosteric inhibitors have emerged as powerful therapeutic agents capable of overcoming resistance mutations that impair the efficacy of orthosteric inhibitors. However, resistance to allosteric inhibitors can also arise, posing a challenge to their long-term effectiveness. Mechanisms of resistance include altered inhibitor affinity and kinetics, disruption of the allosteric mechanism, changes in receptor recycling and activity, and off-target adaptations such as upregulation of drug efflux pumps or activation of compensatory signaling pathways. Furthermore, the specific mechanism of allosteric regulation induced by inhibitor binding can itself be susceptible to resistance mutations, leading to diminished efficacy. Understanding these diverse resistance mechanisms is crucial for developing strategies to counteract them. One promising approach involves the combination of both allosteric and orthosteric inhibitors, either as separate agents or as linked “bitopic” compounds, to mitigate the impact of resistance mutations. This review explores the molecular basis of resistance to allosteric inhibitors and potential strategies to overcome resistance, offering insights for the development of more resilient therapies.

Introduction

Two major classes of allosteric modulators of proteins have been observed: positive allosteric modulators (PAM) which activate their target, and negative allosteric modulators (NAM) which inhibit their target. These opposing modulators can bind to the same sites on target proteins with minor differences in their binding poses and local interactions. For example, PAMs and NAMs of class A GPCRs, and in particular M2 muscarinic acetylcholine receptors, can bind to the same epitope and differentially regulate signal transduction by virtue of the subtle differences in their interactions [1-6]. M2 allosteric modulators are structurally diverse. PAMs include alcuronium and strychnine, and NAMs include C7/3-phth, dimethyl-W84 and gallamine. Diversity of modulators, and their corresponding effects on the M2 receptor, enable fine tuning of biological signals. A similar behavior is observed in class B GPCRs: GLP-1 receptor agonists [7] (or PAMs) and parathyroid hormone receptor antagonists [8] (NAMs) bind closely neighboring regions at their extracellular peptide/hormone-insertion site, despite their opposite actions. Other protein classes, including kinases, are similarly regulated by cellular ligands. For example, Aurora kinase A (AURKA) is activated by an allosteric interaction with microtubule-associated protein TPX2 [9]. Association of TPX2 with the N-lobe, αC helix, and portions of the AURKA activation segment enforce a hinge-bending of the activation segment through Pro282, which locks AURKA into an active conformation, identifiable by the position of the critical phos-Thr288 in crystal structures of the AURKA-TPX2 complex [9].

Allosteric regulatory mechanisms can be enhanced or disrupted by synthetic allosteric modulators. For example, the M2 receptor synthetic modulator LY2119620 increases the potency of the physiological agonist acetylcholine and the supraphysiological agonist iperoxo towards β-arrestin recruitment while simultaneously inhibiting G-protein activation: LY2119620 therefore acts as a PAM and NAM for these respective downstream signaling pathways [3]. Synthetic allosteric inhibitors enable control over target-specific pathways. Inhibitors designed to bind near the AURKA-TPX2 binding site bar TPX2-mediated kinase activation and relegate the kinase to an inactive conformation [10,11].

Allosteric inhibitors have three major properties that make them attractive alternatives to orthosteric inhibitors. First, they tend to be more selective, binding non-competitively to target-subtype-specific or in some cases even mutantspecific sites that are distinctive from the highly conserved active site where substrate binding occurs [12-16]. Second, rather than abolishing the complete activity of the target proteins (as is the case with orthosteric inhibitors), they interfere with a selected interaction of the target protein and thereby modulate a selected pathway while leaving other (non-targeted) pathways unaffected. This increased selectivity leads to fewer off-target (side) effects and may reduce adverse symptoms that a patient may experience compared to patients taking medications with promiscuous target profiles [17-22]. Third, by binding to alternate sites and leveraging target-specific mechanisms, they often help overcome on-target resistance mutations to orthosteric inhibitors [23-26]. This is particularly evident in EGFR-driven non-small cell lung cancer (NSCLC), where over half of Tyrosine Kinase Inhibitor (TKI)-resistant tumors harbor the T790M gatekeeper mutation [27-29]. The T790M mutation at the orthosteric site renders both first- and second-generation TKIs (gefitinib, erlotinib, afatinib) ineffective by blocking TKI binding [30,31] and further promotes resistance to orthosteric inhibitors by increasing affinity for ATP [32]. Common compound mutations such as L858R/T790M can increase resistance by 300-fold for some orthosteric TKIs [33], and mutations including C797S confer resistance to third-generation TKIs designed to overcome T790M [34]. In response to these challenges, Jia and colleagues developed EAI045, an allosteric inhibitor that specifically targets L858R/T790M and L858R/T790M/C797S EGFR variants [24]. Their success with EAI045 in murine lung cancer models motivated others to adopt allosteric approaches [35,36].

Resistance mechanisms common to allosteric and orthosteric modulators

Despite the beneficial properties of allosteric inhibitors, on-target resistance can still develop against these inhibitors. Six major interrelated mechanisms underlie on-target drug resistance to allosteric and orthosteric modulation (Figure 1). First, mutations can remove attractive interactions or sterically clash with the drug, thereby weakening the affinity of a drug and decreasing the fractional occupancy of the receptor at therapeutic doses [37-39] (Figure 1B). Second, mutations can increase the off-rate of the drug, decreasing on-target residence time and consequently also extending the lifetime of the undrugged target: on-target “kinetic” resistance [40] (Figure 1C). Third, mutations can stabilize a drug binding-incompetent conformation of the target protein or destabilize the drug binding-competent protein conformation (Figure 1F-G), which may lead to a weaker affinity for the drug [41,42]. This in turn may strengthen the affinity of the target for endogenous substrates that could compete with the inhibitor, such as ATP in the case of kinase, leading to an increase in intrinsic protein activity (Figure 1D). This effect could be circumvented with drug(s) adopting orthosteric inhibitors or bitopic inhibitors with mixed orthosteric and allosteric binding modes. For example, if an allosteric inhibitor binds to an inactive conformation of a protein but a mutation stabilizes an active conformation, the affinity of the inhibitor may decrease and the affinity of endogenous substrate that binds to the active conformation may increase. Fourth, mutations can increase the overall population of the target protein in a cell, either by increasing the thermodynamic stability of the protein or by decreasing the rate of degradation by, for example, reducing interactions with ubiquitin-pathway intermediates [43-45] (Figure 1E). Fifth, mutations can increase the pathophysiological behavior of a drug target, such as activating mutations in the case of an oncogenic signaling target [46,47] (Figure 1D). In these cases, although drugs might be effective at reducing signaling by that target, subpopulations of unbound proteins may still drive pathological signaling programs via various mechanisms. Sixth, allosteric mutations, described by Nussinov et al., may promote activation by altering energy levels of functional states or the transition barrier between states [48]. This disruption of the allosteric mechanism can either stabilize/destabilize the active or inactive states or facilitate/hamper the transition towards the active or inactive states [49], similar to the stabilization/destabilization of alternative conformers depicted in Figure 1F-G. Indirect evidence for this mechanism in the context of resistance to the BCR::ABL1 allosteric inhibitor asciminib is discussed in further detail below [50,51]

Figure 1. General mechanisms of on-target drug resistance.

(A) A model allosteric inhibitor (purple) binds to an allosteric site distal from the model protein receptor active site (green); this model does not represent a known system and is purely illustrative. Mutations (orange) (B) decrease the affinity of the inhibitor for the receptor or (C) decrease the residence time of the inhibitor on the receptor. (D) Mutations (maroon) increase intrinsic protein activity. (E) Mutations (pink) decrease receptor recycling. Mutations (yellow) bias the receptor (F) towards a conformation incompatible with the binding mode of the inhibitor, or (G) destabilize the conformation of the receptor compatible with the binding mode of the inhibitor. Panels F and G could alternatively refer to the stabilization or destabilization of active or inactive states if the two endpoints refer to active and inactive conformers. The curves on top of the transition arrow schematically describe the change in the energy profile between the two conformers.

These general mechanisms are non-exclusive; even a single missense mutation can exert resistance through multiple mechanisms. For example, a missense mutation such as H396P in ABL1 kinase that biases the protein towards an active signaling conformation may also reduce the affinity of drugs such as imatinib that binds to inactive conformations of that receptor.

While numerous mechanisms of clinical resistance to allosteric inhibitors have been observed, in principle, most of these concepts can be applied to allosteric activators. Mutations could decrease affinity, residence time, or preference for the drug-binding competent conformation. Mutations could decrease protein stability or increase turnover, inactivate the target by replacing critical catalytic residues, or indirectly inactivate the target by altering the transition barrier between the active and inactive states. However, the pathophysiology of a disease must enable such on-target mutations to occur; loss-of-function targets may be able to gain resistance to allosteric activators in the context of human cancers but perhaps not for targets in other diseases without selection pressures. For example, allosteric ABL1 activators [52] have been proposed as possible candidates to block mammary tumor growth in mice on the basis of ABL1 function [53]. It is possible that the selection pressure of a tumor microenvironment might enable ABL1 on-target resistance mutations against such activators to emerge.

Off-target mechanisms, such as upregulation of drug efflux pumps or compensatory signaling pathways, can facilitate disease progression [54-59]. In GPCRs, prolonged activation with PAMs can lead to receptor desensitization as an activation-dependent resistance mechanism [60]. In contrast, cellular survival may rely on a basal level of activity of certain kinases, thus kinase suppression may trigger adaptive responses [61]. For example, treatment with the AKT1 allosteric inhibitor MK-2206 can lead to compensatory upregulation of AKT3 [62]. Off-target mechanisms may be unrelated to the specific mechanism of a small molecule drug, and as such are not reviewed in depth here. Understanding the specific mechanism(s) of resistance mutations furthers pharmacological design of more effective therapeutics and may inform treatment selection in clinical cohorts presenting different on-target resistance mutations.

Resistance mutations at allosteric sites may have additional subtle or complex effects

Resistance mutations towards orthosteric inhibitors face an evolutionary challenge: they need to reduce the effect of the inhibitor while retaining the enzymatic activity of the target. For example, many mutations in the ATP-binding site of tyrosine kinases would clash with imatinib binding, but many of these would also weaken ATP binding or disrupt kinase activity. In contrast, allosteric inhibitors bind to distal regions of the same protein domain or, in some cases, entirely different domains [63]. The allosteric binding site and the amino acid networks that facilitate allosteric regulation may not be essential for the activity of the target protein. Therefore, mutations in allosteric sites are more tolerable and more likely to accumulate than mutations in orthosteric sites that are prone to disrupt essential functions of the protein [15,64,65]. Below we present several cases where mutations at allosteric sites induce a broad spectrum of responses, ranging from rather subtle/localized effects to complex/cooperative effects on allosteric communication.

Resistance to ceftaroline, an allosteric PBP2a inhibitor

Ceftaroline (CFT), a cephalosporin antibiotic used to treat methicillin-resistant Staphylococcus aureus (MRSA) infections, inhibits the penicillin-binding protein 2a (PBP2a) expressed by MRSA by binding to an allosteric site [66-68]. This first binding event allosterically opens the PBP2a active site which then allows a second ceftaroline molecule to bind to the active site and inhibit PBP2a [67]. Detailed structural analyses of PBP2a-ceftaroline interactions rationalized clinically observed ceftaroline resistance mutations distal from the PBP2a active site. For example, the PBP2a mutation E150K that did not alter the binding affinity of ceftaroline to the allosteric site did alter the protein backbone in adjacent domains, and this alteration, in turn, impaired the allosteric communication to the active site [68]. This illustrates the complexity of identifying the mechanism of resistance mutations to allosteric drugs; a native PBP2a allosteric regulatory mechanism was pharmacologically targeted via ceftaroline and later yielded resistance mutations that disrupted allosteric crosstalk between the ceftaroline binding site and the PBP2a active site.

Resistance to allosteric AKT inhibitors

Resistance to allosteric drugs extends to critical oncotargets such as AKT kinase. As a central component of the PI3K/AKT/mTOR signaling axis, AKT kinase plays a fundamental role in regulating cell growth, survival, and proliferation [69]. Dysregulated AKT signaling is frequently observed in multiple cancer types, including ovarian, lung, and pancreatic cancers, making it a crucial therapeutic target [70]. The regulation of AKT kinase activity is governed by an interaction between its N-terminal pleckstrin homology (PH) domain and the catalytic kinase domain [71]. This interaction defines two distinct conformational states: a closed, inactive conformation (PH-in) targeted by the allosteric inhibitor MK-2206 [72], and an open, active conformation (PH-out) targeted by the ATP-competitive inhibitor ipatasertib. Growth assays across diverse cell lines have identified on-target mutations, including W80C and W80R, as a strategy adapted to resist treatment with MK-2206 [62]. Consistent with prior molecular dynamics simulations [73], W80 plays a crucial role in maintaining the structural integrity of the MK-2206-binding pocket of the PH-in conformation. Notably, mutations at W80 do not affect sensitivity to the orthosteric inhibitor ipatasertib. The detection of W80 mutations across uterine, colon and breast cancer in patients has important clinical implications, suggesting that mutation screening could be valuable for optimizing treatment strategies.

Resistance to the allosteric BCR::ABL1 inhibitor asciminib

The allosteric BCR::ABL1 inhibitor asciminib is an excellent model drug for describing on-target resistance because of the well-understood allosteric regulation of ABL1. A substantial body of clinical and experimental evidence has emerged since before its initial FDA approval of asciminib in 2021. Asciminib addresses the clinical need to safely overcome most ABL1 resistance mutations to first and second-generation orthosteric BCR:: ABL1 inhibitors [74]. GNF-2 was developed by targeting the ABL1 myristoyl binding pocket [75]. Subsequent work by Wylie et al. employed fragment-based NMR screens and in silico docking and obtained asciminib (ABL001), a 100-fold more potent allosteric compound than GNF-2 [13].

Asciminib acts by stabilizing the assembled and autoinhibited complex of the ABL1 kinase, SH2, and SH3 domains (Figure 2) [13]. Asciminib binds to the same ABL1 pocket that is otherwise occupied by a myristoyl group in the autoinhibited ABL1 kinase. Since the BCR::ABL1 fusion does not include ABL1 residues containing the myristylation site, asciminib binding mimics and replaces the native autoregulatory mechanism to lock ABL1 in the autoinhibited and assembled complex. The mechanism of action of asciminib is not unique; other myristoylation pocket binders including GNF-2 exert inhibitory effects through similar mechanisms [26]. It is also important to reiterate that the native ABL1 autoregulatory mechanism requires an intact system; for example, deletion of the SH2 or SH3 domains promotes constitutive activation of ABL1 kinase [76]. As we will elaborate below, the disassembly of the three domains of ABL1 kinase – kinase, SH2 and SH3 domains, perturbs its allosteric regulation and can lead to clinical resistance to asciminib.

Figure 2.

Mechanism of inhibition of BCR::ABL1 by the allosteric drug asciminib. (A) The ABL1 αI helix adopts a structured kinked conformation in the presence of a myristoylated peptide. Tan: PDB 1M52 [119], no myristate. Blue: PDB 1OPJ [120], contains myristate. Some helical residues observed in the absence of myristoylated peptide are unstructured in the peptide-bound form. (B) Asciminib binds to the same pocket as the myristoylated peptide, inducing the same local conformation of the ABL1 kinase domain. Blue: PDB 1OPJ. Grey: PDB 5MO4 [13]. (C) Asciminib binding enables docking of the SH2 domain (green) onto the kinase domain C-lobe (grey), mediated by a hydrogen bond network and an interaction between E157 and the αI’ helix dipole. PDB 5MO4. (D) SH2 (green) binding recruits the SH3 (pink) domain, which locks the kinase domain in an assembled, inactive conformation, accompanied by outwards rotation of the αC helix (rotation not pictured). PDB 5MO4. Unstructured residues and 5MO4 ligand nilotinib not pictured. Disruption of any part of this mechanism may contribute towards resistance to asciminib.

Clinical asciminib resistance mutations have been observed in the ABL1 myristate binding pocket where asciminib binds to ABL1 [77]. Of the six on-target resistance frameworks outlined above, a likely mechanism is that they reduce asciminib binding affinity. Such mutations include, for example, ABL1(A337V), ABL1(V468F), and ABL1 (P465S) which are resistant to asciminib and its precursor compound GNF-5 in cellular assays [77]. These mutations exhibit a substantially weakened affinity for asciminib which may be the primary mechanism of resistance. In cases where mutations do not block the pocket completely, it may be possible to develop allosteric compounds that avoid the mutations in analogy to the development of successive generations of active site ABL1 inhibitors.

ABL1 kinase N-lobe and C-lobe point mutations have been observed in patients with clinical resistance to asciminib. These mutations are situated distally from where asciminib binds to ABL1, an initially mysterious observation since it had been reasonably anticipated that resistance mutations would cluster in the myristate binding pocket. Such mutations include, for example, ABL1(M244V), ABL1(L248V), ABL1(Y253F), ABL1(F317L/V),ABL1(E355G), and ABL1 (F359V), which are distributed widely throughout the kinase domain. Of these, ABL1(M244V) is the best described; it promotes cellular proliferation in the presence of asciminib and binds with near wild-type affinity in isothermal titration calorimetry studies [51]. Molecular dynamics simulations suggest that ABL1(M244V) biases the kinase towards an active conformation and promotes disassembly of the SH3/SH2/kinase domain complex. To summarize, asciminib binds to ABL1(M244V) with near wild-type affinity, but it is hypothesized that M244V is unable to induce autoinhibition of the kinase.

In addition to point mutations, alternate BCR:: ABL1 chromosome translocation events in some patients can fuse BCR to ABL1 3′ of ABL1 exon 2, excluding ABL1 residues containing part of the SH3 domain (BCR::ABL1 b2a3/b3a3). Leukemic cells harboring the b2a3/b3a3 variants are highly resistant to asciminib therapy [50,78,79]. Notably, this resistance occurs in the absence of any kinase domain point mutations. While still able to potently bind, asciminib is unable to transduce its autoregulatory effect in b3a3 constructs (Figure 3) [50,78,80]. As demonstrated by Leyte-Vidal et al in the context of the full-length BCR::ABL1 protein, asciminib is able to displace an orthosteric tracer compound that reports on the active αC-helix “in” conformation. This occurs because asciminib allosterically induces the autoinhibited αC-helix “out” conformation via engagement of the SH3/SH domains. Similarly, asciminib is unable to displace the orthosteric tracer compound in the b3a3 variant, despite still being capable of displacing an asciminib-based tracer compound with low nanomolar potency (Figure 3). This work emphasized the specific mechanism of resistance of BCR::ABL1 variants lacking exon 2; by destabilizing and unfolding the SH3 domain, BCR::ABL1 is no longer susceptible to autoinhibition induced by asciminib even though asciminib still potently binds to the allosteric site.

Figure 3.

An allosteric mechanism of on-target resistance to the allosteric BCR::ABL1 inhibitor asciminib. The BCR::ABL1 Δexon2 deletion (b3a3) removes SH3 domain residues essential for maintaining the closed autoinhibited ABL1 conformation. NanoBRET figure artwork adapted from Leyte-Vidal et al (2024) [50]; larger bioluminescence resonance energy transfer (BRET) ratio signal is observed given higher fractional occupancy of bound tracer compound. (A) Dasatinib, an orthosteric BCR::ABL1 inhibitor, displaces a NanoBRET tracer compound derived from dasatinib in both BCR::ABL1 p210 (b3a2) which contains a complex SH3 domain, as well as in BCR::ABL1 p210 Δexon2 (b3a3). (B) Asciminib displaces an asciminib-derived NanoBRET tracer compound with equal affinity in both the BCR::ABL1 p210 (b3a2) and BCR::ABl1 p210 Δexon2 (b3a3) constructs. (C) However, asciminib cannot displace a dasatinib-derived tracer compound that binds at the orthosteric site in the context of the BCR::ABl1 p210 Δexon2 (b3a3) construct, unlike BCR::ABL1 p210 (b3a2) which contains a complex SH3 domain; formation of the assembled complex displaces das-tracer from the orthosteric site.

High-throughput screening experiments help identify resistance mutation sites

Most resistance mutations are detected in clinical samples of patients whose disease no longer responds to treatment with an allosteric inhibitor. Alternatively, screening campaigns can predict resistance mutations to allosteric inhibitors before they are detected in patients. For instance, large CRISPR-based mutagenesis screens combined with single-cell transcriptomics can distinguish between cancer drug resistance mechanisms [81]. Coelho et al. performed large-scale mutagenesis screens in HT-29 cells on targets, including MEK1/2, in the presence or absence of drugs including trametinib, an allosteric MEK1/2 inhibitor. By comparing the difference in proliferation in the presence or absence of a drug, hypotheses towards “canonical drug resistance variants” or “driver variants” could be assigned to individual mutations depending upon whether they respectively provided a proliferation advantage in, or additionally in the absence of, each drug tested. Such genetic screens enable large-scale hypothesis generation for the mechanism of resistance mutations, and in particular for the identification of potential driver mutations.

Protein function-associated screens have also been adopted to generate hypotheses for resistance mutations [82]. Through a large mixed protein abundance and binding partner interaction screen of over 26,000 KRAS mutations, Weng et al identified the allosteric determinants of KRAS specificity for different binding partners, and the effects of individual mutations on KRAS interactions with those binding partners. The effect of each mutation on KRAS folding and binding to six other proteins were determined from two experimental formats [82,83]. The first format, BindingPCA, reported on protein–protein interactions following fusion of each pair of interest to a fragment of a reporter enzyme. The second format involved fusing a single target to part of the reporter enzyme, expressed in an overexpression background of the other part of the reporter enzyme. These two formats allowed Weng et al to decouple the effects of a mutation on protein abundance and binding interactions, identifying mutations that allosterically promoted or inhibited KRAS interactions with the six tested binding partners. The authors identified a group of 71 mutations in residues that line the sotorasib binding pocket, an allosteric inhibitor of pathogenic KRAS(G12C) that inhibit the activity of KRAS through reduced binding to RAF1. These 71 mutations mimic the allosteric inhibitory effects of sotorasib. Conversely, mutations identified in these screens that increase KRAS interactions with RAF1, either through the same sotorasib-associated allosteric network or alternate networks identified in the work, could be candidate sotorasib resistance mutations.

Computational methods help identify potential allosteric sites and effect of mutations at those sites

To overcome allosteric drug resistance, computational tools can help locate alternative potential allosteric sites [84-86]. Identifying allosteric sites requires structural and dynamic information that defines its role in regulating protein functions. As reviewed by Nerin-Fonz and Cournia [87], several machine-learning (ML) based methods account for either or both types of information. Additionally, ML methods can incorporate physicochemical features, including “druggability”, hydrophobicity and solvent-accessibility, enabling the pharmaceutical search for the potential allosteric pockets. Amongst the current ML methods, AlloSite [88], Protein Allosteric Sites Server (PASSer) [89], and PASSerRank [90] achieve >95% of accuracy against their respective test sets.

Despite their success, ML models can be limited by the representation of known allosteric sites in databases and predictions from single structures. Detecting transiently inaccessible sites in apo structures, such as cryptic pockets, can be challenging under conditions of data sparsity. Some limitations can be overcome with conformer generation using the elastic network model (ENM)-based methods [91-94]. When performing essential site scanning analysis (ESSA) on the ClustENM-generated conformers, ESSA outperforms PARS [95] and AllositePro [96] capable of detecting cryptic sites [92]. Additional structural ensembles, or even distinct conformational ensembles, can be obtained using generative methods including AlphaFold 3.0 [97]. Physics-based tools such as enhanced sampling in molecular dynamics simulations [94,98] and ENM-based generation of ensemble of conformers using ProDy [99] enable sampling of a diverse spectrum of conformations and their dynamics. These approaches shed light to ensembles of conformations accessible via allosteric transitions under native state conditions, rather than the limited set of structures for a target that may, or may not, be available from the PDB. Comprehensive evaluation of alternative allosteric sites and their mutations could benefit from accessible in silico methods.

Building on these advancements, recent development in ML tools for evaluating the effect of single amino acid variants (SAVs) have highlighted the role of structural dynamics as a determinant of pathogenicity, and further pointed to the functional significance of residues that act as sensors and effectors of allosteric signals [104,105]. These studies have shown how selected mutations at those sites could impact allosteric communication, thus providing information on specific mutations that may interfere with allosteric modulation of target proteins.

Moreover, additional identification of allosteric sites is enabled through structure-based statistical mechanical models of allostery allowing rationalized drug design [100,101]. Potential allosteric sites can be identified through probing the effect of ligand binding on a functional site and measuring the subsequent perturbation on other residues in the same protein. Through the principle of reversibility of allosteric communication, spatially neighboring residues that are affected by probe binding at the functional site might represent candidate allosteric sites. Adopting molecular docking [102] and drug repurposing tools [103] further facilitate screening and fragment-based optimization at these candidate allosteric sites. These computational methods, combined with generative ML models and enhanced sampling techniques, offer a robust pipeline for the discovery and design of allosteric drugs.

Combination targeting with orthosteric and allosteric compounds

Dual-targeting with both an orthosteric drug and an allosteric drug is a tempting strategy that is being widely evaluated, particularly in the setting of prior acquired on-target resistance, or as a strategy to preemptively reduce the possibility of on-target resistance [26,106,107]. Allosteric and orthosteric drugs are more likely to have non-overlapping profiles of resistance mutations that decrease drug affinity. Conceptually, this is a similar rationale to combining orthosteric drugs with different binding modes, since such drugs have also been observed to have non-overlapping resistance profiles [108]. The major advantage of using an allosteric drug over a second orthosteric compound is that orthosteric drugs that bind the same target frequently also have overlapping off-target profiles [109-111]; this may limit clinical use due to side effects from increased off-target inhibition as a consequence of co-administration of multiple orthosteric drugs. While methods have been developed to address this issue and optimally co-dose orthosteric compound combinations for selectivity given their off-target profiles [112], no such methods have been considered necessary for allosteric compounds due to their frequently exquisite selectivity.

A protein that is dual-targeted by two inhibitors, one orthosteric and the other allosteric, may be unable to accommodate both compounds at the same time depending upon the conformational changes each inhibitor induces on the target. However, this does not mean that a combination cannot be efficacious. While binding of the allosteric BCR::ABL1 inhibitor asciminib and orthosteric inhibitors (i.e., imatinib, dasatinib, nilotinib, ponatinib) is antagonistic, [113], combination of a orthosteric inhibitor with asciminib can target resistance mutations. In particular, ponatinib and asciminib synergize in cellular tests and in mouse models and this combination is therefore a strong candidate to overcome challenging BCR:: ABL1 compound resistance mutations including E255V/T315I [106,107,114]. Compounds that are conformationally synergistic, that is, they preferentially bind to the same conformation, can both be affected by resistance mutations that bias the receptor away from that single conformation. For example, asciminib is conformationally synergistic with the compound DAS-CHO-II; both compounds promote the αC-helix “out” closed and assembled ABL1 SH3/SH2/kinase domain complex [113]. Any single mutation that sufficiently biases the kinase away from this conformation could be resistant to both inhibitors, emphasizing that conformational diversity may be a preferred strategy in cases where on-target resistance can emerge.

Linked dual-targeted compounds (“bitopic inhibitors”) may have better selectivity than mixtures of the component orthosteric and allosteric inhibitors. Early efforts to develop bivalent inhibitors of protein kinases used linked peptide analogs of protein target specificrecognition sequences as reviewed in [115]; current designs can employ small-molecule combinations of known allosteric and active site binders [116,117]. For example, RapaLink-1 and DasatiLink-1, bitopic compounds containing, respectively, rapamycin-like and sapanisertib-like, and asciminib-like and dasatinib-like groups, exhibit enhanced selectivity [118]. RapaLink-1 is selective for mTOR complex 1 over mTOR complex 2, and DasatiLink-1 has minimal off-target occupancy against non-BCR::ABL1 kinases that were occupied by the analogous 1:1 mixture of dasatinib and asciminib. Notably, bitopic inhibitors violate medicinal chemistry rules for conventional inhibitors and they require distinct transport mechanisms, such as interferon-induced transmembrane protein assisted for uptake into cells [118]. As for combinations of orthosteric and allosteric compounds, whether or not both warheads of a bitopic compound can bind at once depends upon the conformational demands of each group on the target protein. Unlike compound combinations, however, bitopic compounds cause a high effective local concentration of the second warhead when the first is bound, and vice-versa. Therefore, the linked compounds will bind more readily than the individual compounds even if the individual compounds bound with modest antagonism. Ultimately, whether or not both sites are occupied depends upon the conformational cooperativity (positive or negative) between the binding modes of the two groups and the design of the bitopic compound, including features such as the linker length, composition, and placement.

Perspectives

In summary, resistance to allosteric drugs represents a diverse set of mechanisms including those pertinent to orthosteric drugs and those related to the specific mode of regulation accessed in each drug-target pair. Defining and categorizing resistance mechanisms enables the correct drug to be selected for each case, or, as is becoming more common, the correct combination of allosteric and orthosteric drugs. Such categorization is enabled by direct measurements of drug affinity, kinetics, and target conformational readouts as well as computational assessments of target flexibility and allosteric coupling between drug-binding and active sites. Further work in this exciting field consists of the design and evaluation of allosteric compounds as monotherapies, evaluation within combination therapies, and as the chemical basis for the design of linked compounds.