Cracking the allosteric code of NMR chemical shifts

One of the broadest definitions of allostery is in terms of long-range couplings between remote sites within a molecular system (1–4). Allosteric couplings are critical to understand the molecular basis of physiological regulatory mechanisms as well as of pathological deregulation (1–4). Allostery is also opening new opportunities in drug design and diagnostics (1–5). Hence, there is considerable interest in elucidating allosteric mechanisms, which typically involve the modulation of functional conformational equilibria by allosteric effectors. One of the most effective means to monitor allosteric transitions is through NMR chemical shift changes resulting from a library of perturbations designed to interrogate a given allosteric system (6, 7). The perturbation library may include analogs of allosteric ligands and/or mutations designed to modulate allosteric conformational equilibria (7). Through statistical comparative analyses of the NMR chemical shift variations elicited by the selected perturbations it is possible to identify chemical shift patterns that serve as distinctive signatures for specific allosteric mechanisms (3, 7). However, the application of these methods to symmetric homooligomers, which are prototypical allosteric systems, has historically remained challenging due to difficulties in detecting and characterizing the unbound and bound protomers of elusive singly bound intermediates. In PNAS, Falk et al. (2) provide a brilliant solution to this problem by designing mutations that silence binding to a single protomer of a homodimeric enzyme. The method proposed by Falk et al. (2) reveals chemical shift patterns that cannot be fully rationalized through simple two-state exchange models, pointing to the need of more complex mechanisms that go beyond traditional allosteric paradigms.

Allostery plays a central role in physiology, pathology, and pharmacology (1–5). The physiological role of allostery is primarily regulatory and is critical for cell homeostasis and signaling (1–4). In pathology, allostery is essential for the molecular rationalization of disease-related mutations and posttranslational modifications, whereas in pharmacology allosteric processes are of both therapeutic and diagnostic value (1–5). Targeting allosteric sites is therapeutically advantageous because it circumvents competition with endogenous ligands or substrates at orthosteric sites, thus enhancing potency. Furthermore, allosteric loci are typically less evolutionarily conserved than orthosteric sites, thus offering also a selectivity advantage. The diagnostic value of allostery stems from the notion that intrinsic allosteric transitions provide a means to effectively sense protein-bound biomarkers, which often remain elusive to traditional analytical methods originally developed for unbound metabolites (5).

The allosteric phenomenon arises from the modulation of dynamic conformational equilibria by external perturbations (e.g., ligand binding, mutations, or posttranslational modifications) (3). Even when functionally relevant, the redistributions of conformational ensembles underlying allostery are often subtle and may escape detection by traditional structure determination methods (3). Hence, alternative experimental approaches are needed. One of the most effective means to monitor allosteric transitions is through NMR chemical shift variations, because they are exquisitely sensitive to subtle structural and/or dynamical changes. For example, amide ¹⁵N and ¹H chemical shifts are hypersensitive to conformational variations, which poses a major challenge for their computational prediction and for their use in chemical shift-based structure determination but makes them an ideal atomic-resolution probe to sense the effect of allosteric perturbations through comparative NMR analyses (3, 6). The latter include not only conventional apo vs. holo or WT vs. mutant binary chemical shift comparisons, but they extend also to ternary or multiperturbation vectorial, correlative, and statistical analyses (6–13).

One of the primary goals of multiperturbation chemical shift analyses is to effectively identify patterns of chemical shift changes that are diagnostic of specific allosteric models (3). For example, when allosteric transitions are fast on the NMR chemical shift time scale, the observed chemical shifts are population-weighted averages of the values for the pure states (i.e., chemical shifts code for state-specific populations), which, in turn, assuming a Boltzmann distribution, report on the depth of the free energy landscape minima within which allosteric systems function (3, 9). In the simplest case of a two-minima allosteric free-energy landscape, perturbations such as ligand binding or mutations that modulate the allosteric equilibrium typically result in linear “multiplet” spectral patterns with similar cross-peak distributions for different residues, provided that such residues are sufficiently distant from the perturbation sites to experience only negligible nearest-neighbor effects (Fig. 1) (6–16). However, when allostery involves exchange between more than two states, deviations from these linear patterns are observed at least for selected residues (8). Hence, pattern analysis of chemical shift variations provides an effective means to identify subsectors of residues that exhibit concerted responses to allosteric perturbations (7). However, the application of these approaches to map allosteric cross-talk between symmetry-related protomers of symmetric oligomers poses major challenges. These are due to difficulties in isolating singly or partially bound intermediates necessary for dissecting the allosteric mechanism of oligomeric proteins. Such intermediates are often populated only transiently, especially in systems lacking negative interligand cooperativity (17), and may also be subject to fast exchange between apo and holo forms, making the investigation of interprotomer allostery in symmetric homooligomers particularly challenging.

Fig. 1.

Relation between allosteric models and chemical shift patterns observed for the quadruplets proposed by Falk et al. (2). State equilibria for simplified versions of the MWC, KNF, and EAM (18, 19) are shown using the same color code as in the schematic HSQC spectra in the bottom panels. Each protomer is assumed to sample only two conformations (empty squares or circles) exchanging rapidly in the chemical shift NMR time scale. Filled squares denote allosteric ligands; lig_0, lig₁, and lig₂ refer to the apo, singly bound, and doubly bound samples, respectively. The HSQC patterns in the bottom were generated under the assumption that chemical shifts sense primarily the conformation of the protomer for which they are measured, as opposed to ligand binding or the conformation of the adjacent protomer. This figure is based on refs. 2 and 18.

Falk et al. (2) propose an elegant solution to this long-standing problem by stabilizing the singly bound dimers of the homodimeric enzyme thymidylate synthase (TS) through R-to-E mutations that silence binding of dUMP to only a single protomer (i.e., the WT protomer). The WT:mutant mixed dimers were then separated from the nonmixed dimers by ion-exchange chromatography and investigated by NMR chemical shift mapping, taking advantage of the fact that the TS dimers are sufficiently tight to avoid disproportionation during the chromatographic separation and the NMR data acquisition. Because WT and mutant protomers of the mixed dimers can be selectively labeled, chemical shifts could be measured for both apo and holo protomers of the singly bound WT:mutant mixed dimers and compared with those of the fully apo and doubly bound dimers, resulting in four observable peaks per residue (i.e., a “quadruplet”; Fig. 1). Furthermore, Falk et al. (2) devised a vectorial scheme to correct for chemical shift changes introduced by the mutations required to trap the singly bound dimers, thus minimizing “chemical shift noise” in the mutants and making the corrected quadruplet patterns a truly diagnostic signature that genuinely reflects allosterically relevant conformational equilibria.

Through this ingenious experimental design, Falk et al. (2) discovered that the allosteric responses of the apo and holo protomers in singly bound dimers are markedly distinct, with dUMP binding perturbing primarily just the bound protomers. However, when a second molecule of dUMP binds, the allosteric perturbations propagate to both protomers, suggesting that full interprotomer cross-talk requires both bound ligands (2). Hence, the first ligand primes

The novelty of the contribution by Falk et al. (2) lies both in the elegant use of mutations to break spectroscopic and binding degeneracies of symmetric dimers and in the unprecedented insight into how ligand binding controls allosteric interprotomer cross-talk.

the dimer for the allosteric effect elicited by the second ligand, an unprecedented result that could not have been easily anticipated without the clever method proposed by Falk et al. (2).

Another remarkable result by Falk et al. (2) is that for several residues well separated from the binding sites of TS the chemical shifts of the singly bound dimers are not simple linear interpolations between those of the apo or doubly bound dimers (Fig. 1). The deviations from the patterns expected based on two-state models (Fig. 1) are denoted by Falk et al. (2) as “super-shifting” or “orthogonal shifting” and point to the presence of more than two functional states. This conclusion is particularly ground-breaking, because the two main classical theories for allostery in homooligomers [i.e., the Monod–Wyman–Changeux (MWC) and Koshland–Nemethy–Filmer (KNF) models], as well as the simplest form of the ensemble allosteric model (EAM) (18, 19), although differing in terms of symmetry conservation, all posit that each protomer samples only two states (Fig. 1). Falk et al. (2) show that not all residues of TS exhibit chemical shift signatures consistent with the MWC, KNF, or the simplest EAM models, suggesting the need for more complex schemes that go beyond these allosteric paradigms (20). In this respect, we anticipate that the work by Falk et al. (2) may inform further theoretical development, especially in the context of more general EAMs developed by Hilser and coworkers (19).

The EAM framework is able to capture the hierarchy of conformational states accessible within free-energy landscapes that are more complex than those underlying either the MWC or KNF models, while at the same time maintaining parameter parsimony (19). Hence, it is possible that the method proposed by Falk et al. (2) will enable the measurement of EAM parameters, including the free energy of interprotomer coupling in symmetric holodimers, which is a descriptor critical for quantitatively assessing the degree of intersubunit cooperativity. Another interesting development of the work by Falk et al. (2) is that the samples defining the “quartet” (i.e., the apo WT, the two singly bound mutants, and the doubly bound WT), and possibly more samples engineered to result in modulated allosteric responses, may serve as a library for the implementation of the CHEmical Shift Covariance Analysis, or CHESCA (7), a methodology that will facilitate the systematic classification of diagnostic chemical shift patterns and signatures.

The novelty of the contribution by Falk et al. (2) lies both in the elegant use of mutations to break spectroscopic and binding degeneracies of symmetric dimers and in the unprecedented insight into how ligand binding controls allosteric interprotomer cross-talk, a process that had remained largely elusive until recently. In addition, the conclusions reported by Falk et al. (2) are relevant not only for TS but possibly for several other homooligomeric allosteric systems, including growth factors, cytokines, kinases, G protein-coupled receptors, and transcription factors. Revisiting these allosteric systems in light of the method proposed by Falk et al. (2) will undoubtedly further advance our understanding of allostery. For example, the chemical shift patterns identified in PNAS by Falk et al. (2) will provide a clear signature to differentiate KNF vs. MWC vs. other more complex multistate models of allostery, such as the general EAM (19). These exciting developments will take us a step closer toward better understanding allosteric regulation and deregulation in physiology and pathology, respectively, and toward fully exploiting the therapeutic and diagnostic opportunities opened by allostery.