Engineering an Allosteric Control of Protein Function

Abstract

Allosteric regulation in proteins is fundamental to many important biological processes. Allostery has been employed to control protein functions by regulating protein activity. Engineered allosteric regulation allows controlling protein activity in subsecond time scale and has a broad range of applications, from dissecting spatiotemporal dynamics in biochemical cascades to applications in biotechnology and medicine. Here, we review the concept of allostery in proteins and various approaches to identify allosteric sites and pathways. We then provide an overview of strategies and tools used in allosteric protein regulation and their utility in biological applications. We highlight various classes of proteins, where regulation is achieved through allostery. Finally, we analyze the current problems, critical challenges, and future prospective in achieving allosteric regulation in proteins.

Graphical Abstract

INTRODUCTION

Allostery is a common biological phenomenon whereby perturbations at a specific site (the allosteric site) cause a conformational change or dynamics at a distal site in the same protein.^1–3 Allostery is encoded in single molecules, such as nucleic acids and proteins.^4–8 It plays important roles in essential biological processes including cellular signaling,^9–12 antigen binding and antibody activation,¹³ the ubiquitin–proteasome system,¹⁴ and post-translational modifications.¹⁵ Genetic mutations that alter allosteric communications could cause severe disease as the majority of the disease-causing mutations are located outside of the active sites.¹⁶ The molecular mechanism of allostery was initially interpreted by two well-known models including the Monod–Wyman–Changeux (MWC)¹⁷ and the Koshland–Némethy –Filmer (KNF)¹⁸ models in the 1960s. Both models assume a conformational change underlying allostery between two major conformations of proteins, such as the active and inactive states. Two decades later, Cooper and Dryden¹⁹ proposed the dynamics-based model of allostery, in which the structure does not change but the dynamics does. In 1999, the Nussinov group stressed that most proteins exist in an ensemble of states and, therefore, proposed the “conformation selection and population shift” model.^20–22 They proposed that the allosteric effects cause a redistribution of the conformational populations within the ensemble, namely a population shift.^23,24 The theory of population shift recognizes the pre-existence of all the conformational states including the inactive, active, and transition states.^24–31 This theory suggests that the allosteric effect shifts the ensemble from less stable to more stable conformational states rather than induces the formation of new conformational states.^12,32,33 This model expands the concept of allostery from two conformational states to an ensemble of states and therefore has been widely adopted to understand allostery.^34,35

An understanding of allosteric mechanisms has stimulated the application of allostery in a number of fields, such as allosteric materials,³⁶ allosteric drug discovery,³⁷ allosteric polymorphism,³⁸ and protein engineering.² Yan et al.³⁶ performed an in silico evolution experiment to create allosteric materials that can react to a stimulus at a distal site. The allosteric materials evolved could be used to accomplish a specific mechanical task. Drugs interacting with allosteric sites that are more specific and less conserved than the active sites have fewer side effects and allow modulation of protein functions.^5,16,37,39 Inspired by the interprotein propagation of allosteric effects in the signaling network, Nussinov et al. expanded the concept of allosteric drugs to allo-network drugs.^40–42 They suggested that the effects induced by allodrugs can propagate within a protein or across the protein–protein interaction network to improve or block a specific interaction.⁴⁰ Using this concept, authors targeted the longdistance allosteric communication pathways and expanded the scope of drug development.⁴⁰ One of the promising applications of allostery is in understanding the allosteric effects of mutations in disease phenotypes.³⁸ Understanding allosteric polymorphism could play a major role in delineating the latent drivers that are potential sources of expanding the cancer landscape, which in turn has implications in precision medicine.^43,44 Recently, we have seen a growing number of applications of allostery in protein engineering.¹ Engineering allosteric control in proteins allows us to manipulate protein activities in live cells at high spatiotemporal resolution. Moreover, introducing sensor domains to control allosteric communications can be applied to construct complex biological circuits in synthetic biology, where different combinations of inputs could generate specific outputs. In this review, we discuss the engineering of allosteric regulation in proteins, future directions in this area, and how this application contributes to a deep understanding of biology. We discuss the state-of-art methods developed to identify allosteric sites and pathways, a diversity of sensor domains used to control allostery, and major protein families with allostery that could be engineered.

IDENTIFICATION OF ALLOSTERIC SITES AND PATHWAYS IN PROTEINS

To engineer allosteric control over protein functions, the first step is to identify allosteric sites for the insertion of regulatory domains. Both experimental and computational approaches have been developed to screen allosteric sites and understand how perturbations introduced at these sites transmit to the active site and hence affect its functions. Site-directed mutagenesis is one of the commonly used experimental methods to discover novel allosteric sites. Huang et al.⁴⁵ mutated seven cysteine residues to serine and identified a functional cysteine site that affects the catalytic activity of fructose-1,6-bisphosphatase. Mutagenesis combined with molecular dynamics (MD) simulations not only identified an extracellular allosteric modulator site of human glycine transporter GlyT2 but also provided information about how the bioactive lipids interact with this site and inhibit protein functions.⁴⁶ Another experimental method used to identify druggable allosteric sites of a protein is to design a high-throughput biochemical screening assay against a compound library. The allosteric binding site could be visualized by the crystallographic studies of the compound and protein complex.⁴⁷ Recently, Fogha et al.⁴⁸ proposed a new method to identify allosteric sites by analyzing the density and clustering of crystallization additives. They found that these additives used to stabilize proteins during crystallization tend to distribute near allosteric sites, which is an efficient experimental mean for allosteric site prediction. Experimental methods used to identify allosteric sites are time-consuming and not cost-effective. Therefore, several computational methods have been developed to predict allosteric sites and analyze allosteric communication pathways efficiently.

Two different groups of computational methods have been developed. One of them is based on molecular biophysics or statistical mechanical theories. In 1999, Lockless et al.⁴⁹ used evolutionary data of PDZ proteins to measure statistical interactions between residues and identified energy conduction pathways connected by interacting residues within this protein family. More recently, Guarnera et al.^50,51 developed the structure-based statistical mechanical model and implemented it in the AlloSigMA server to explore the causality and energetics of allosteric pathways between regulatory and functional sites. In light of the structure-based statistical mechanical model, Tee et al.⁵² proposed a reverse perturbation approach to predict allosteric sites. As a result of the perturbation applied at the functional sites, the allosteric sites as well as the extended protein regions including regulatory exosites cound be identified. On the basis of elastic network models, Chennubhotla et al.⁵³ investigated the stochastics of a discrete-state discrete-time Markov process of information transfer. They found that functional sites possess increased communication propensities and secondary structures are quite efficient in mediating allosteric communications. Recently, Ayyildiz et al.⁵⁴ incorporated elastic network modeling with solvent mapping and sequential and structural alignments to identify species-specific allosteric sites in glycolytic enzymes of bacteria, parasites, and humans. Ota et al.⁵⁵ used anisotropic thermal diffusion, a nonequilibrium MD simulation method,^56,57 to study the intramolecular signaling pathways in PDZ domain proteins. This method requires neither homologous proteins nor long simulation times to observe complete communication pathways within proteins. McClendon et al.⁵⁸ developed “MutInf” to identify novel allosteric sites by studying the significantly correlated motions from equilibrium molecular dynamics simulations. Goncearenco et al.⁵⁹ developed the SPACER server based on geometric features of the static structure and protein dynamics. It offers an interactive framework that allows users to predict allosteric sites and quantify allosteric communication between the regulatory and functional sites. Rocha et al.⁶⁰ combined MD simulations with network theory analyses to predict potential allosteric sites based on the efficiency of intramolecular communication.

The other computational method used to identify allosteric sites is derived from information theory or spectral graph methods. An increasing number of studies based on this method have been conducted in recent years. In this method, proteins are treated as networks of interacting residues where nodes are the amino acids and edges indicate the interaction between residues. The network structure connecting all the residues with added energy weights could be used to identify distal sites coupled with the active site and illustrate the communication pathways in response to perturbations. Chennubhotla and Bahar⁶¹ used Markov propagation of information to partition the interaction network of a protein into soft clusters. The amino acids shared by adjacent clusters act as messengers that transmit information across clusters, while the residues located in a single cluster serve as hubs. Using this approach, they discovered two potential communication pathways between the ATP- and cochaperoninbinding sites. Atilgan et al.⁶² assigned weights to nonbonded inter-residue contacts using knowledge-based potentials. From the weighted network, they were able to obtain the optimal paths mediating robust residue communication in proteins.⁶² Later on, they developed a new tool termed perturbation-response scanning that systematically uses computational perturbation/response techniques based on linear response theory.⁶³ This tool was able to efficiently characterize the response of residues in a protein to a given perturbation.⁶³ Vijayabaskar and Vishveshwara⁶⁴ constructed the protein structure network based on noncovalent interaction energies between residues. Analysis of this network revealed the communication paths between ditsal sites in the protein. Amor et al.⁶⁵ proposed a graph-theoretical framework that uses an edge-to-edge transfer function to calculate the bond propensity, which predicts the allosteric interactions (sites and pathways coupled to the active site). They correctly predicted the key allosteric sites and communication pathways for three proteins including caspase-1, CheY, and h-Ras.⁶⁵ More recently, the Dokholyan group developed a conceptually different approach for allosteric analysis as it considers the protein disordered media and uses physics and the perturbation propagation algorithm to delineate pathways. They created a comprehensive platform, called Ohm, to facilitate the application of this method.³ Based on the three-dimensional structure of the target protein, Ohm can identify the allosteric sites coupled to the active site, analyze the allosteric communication pathways, predict the critical residues in the communication pathways, and calculate the allosteric correlations between residue pairs (Figure 1). The performance of Ohm has been validated by successfully mapping the allosteric networks of 20 proteins from a variety of protein families with distinct structures.

Figure 1.

Allosteric interactions predicted by Ohm.³ The three-dimensional structure of a protein is used as the input for Ohm. If the active site is known, Ohm predicts the allosteric hot spots coupled to the active site. If both the active site and the allosteric site are given, Ohm identifies the allosteric communication pathways and the critical residues in each pathway. If neither the allosteric site nor the active site is provided, Ohm calculates the allosteric correlations for each residue pair.

OPTOGENETIC AND CHEMOGENETIC APPROACHES FOR ALLOSTERIC CONTROL OF PROTEINS

Chemogenetics and optogenetics have proven to be powerful techniques in constructing the allosterically regulated protein systems.⁶⁶ In chemogenetics and optogenetics, allosteric protein regulation is achieved by fusing a regulatory sensor domain to a host protein to create a chimeric protein via molecular cloning techniques.⁶⁷ Several sensor domains have been engineered for the allosteric regulation of protein functions. Upon specific signal reception (e.g., ligands, light, or pH), sensor domains usually undergo large conformational change, which results in substantial change in the host protein activity. Optogenetic tools offer noninvasive control of protein function with unprecedented spatiotemporal resolution. Optogenetic tools work by light energy absorption which leads to changes in intra- and intermolecular contacts and therefore causes a large change in protein conformation. Optogenetics provides precise spatiotemporal control of protein functions with light, while chemogenetics offers the capability to penetrate deep tissues and control protein activities for hours with the single addition of a compound. Diverse optogenetic and chemogenetic tools have been reported by a number of research groups. Some of the well-established tools include LOV domains,^68,69 BLUF domains,⁷⁰ the phytochrome B (PhyB) and PIF protein pair,⁷¹ the cryptochrome 2 (CRY2) and CIB1/CIBN protein pair,⁷² bacterial phytochrome BphP1 and its binding partner PpsR2,⁷³ UV resistance locus 8 (UVR8),⁷⁴ Vivid (VVD),⁷⁵ Dronpa,⁷⁶ uniRapR, ER-LBD, and cpDHFR. Different approaches have been employed to control protein functions by using the aforementioned tools.

A method to regulate protein function is by perturbing allosteric site, for example, using order–disorder control.^77,78 Optogenetic switches that rely on direct perturbation have mainly been engineered on the basis of the light-oxygen-voltage 2 (LOV2) domain from Avena sativa (Figure 2a, top panel). LOV2 is a member of the PAS superfamily. LOV2 consists of a cofactor flavin mononucleotide (FMN), a PAS fold, and a large α-helical region (Jα helix) at the C-terminus of the fold. Upon blue-light exposure, a covalent adduct is formed between the FMN and a cysteine in the PAS fold, resulting in a large conformational change leading to unfolding of the Jα helix.^79,80 When irradiation ceases, LOV2 returns to its original conformation within a half-time of ~30 s. The rapid photocycle and the strong conformational change upon photoexcitation make LOV2 particularly a potent tool.⁸¹ LOV2-based regulations are typically achieved by fusing the C-terminus of Jα to a desired target protein. LOV2-based allosteric regulations have been achieved in a wide range of proteins such as Rac1-GTPase, Rho GTPase, GEFs, Src kinase,^66,82,83 histidine kinases,⁸⁴ caspase,⁸⁵ protein tyrosine phosphatases,⁸⁶ dihydrofolate reductase,⁶⁸ Trp repressor proteins,⁸⁷ inteins,⁸⁸ and transcription factors.⁸⁹ An example of a chemogenetic tool which works on the principle of order–disorderness is the circularly permutated bacterial dihydrofolate reductase (cpDHFR). cpDHFR adopts a well-defined three-dimensional structure when binding to the cofactor nicotinamide adenine dinucleotide phosphate (NADPH) and its inhibitor trimethoprim (TMP).⁹⁰ In the absence of these ligands, however, cpDHFR is partially unfolded. In light of this conformational change, Farrants et al.⁹¹ designed the ligandmodulated antibody fragments by inserting cpDHFR in the complementary-determining region 3 of the nanobody. The binding affinity of the engineered nanobody to its target was successfully controlled by NADPH and TMP. Another example of a chemogenetic tool which works by direct perturbation is uniRapR (Figure 2a, bottom panel). The FK506 binding protein (FKBP12) and FKBP12–rapamycin binding protein (FRB) undergo conformational changes upon the interaction with rapamycin. To use this protein complex as a sensor domain, Dagliyan et al.⁹² rewired the protein complex and built a single polypeptide chain that senses rapamycin, which is named the uniRapR domain. Insertion of the uniRapR sensor domain into the kinases successfully controls the kinase function in vivo.^92,93

Figure 2.

Optogenetic and chemogenetic strategies for allosteric control of protein activity. (a) In light-mediated optogenetic control, the LOV2 domain inserted at the surface exposed, allosteric region on the protein of interest (POI) allosterically inactivates the host protein in response to light (top). In rapamycin-mediated chemogenetic control, the inserted uniRapR domain on POI allosterically activates the host protein in response to rapamycin (bottom). The activated protein further interacts with downstream targets (T). (b) Inducible dimerization systems. Light-induced CRY2–CIB1 dimerization system (top); rapamycin-induced FRB and FKBP dimerization system (bottom). (c) Split protein-based dimerization system for protein regulation. POI is split into N- and C-lobes. Assembly of two lobes can be mediated by chemical (e.g., FKBP, FRB, and rapamycin) or light (e.g., CRY2–CIB1)-induced systems. (d) Rapamycin-induced trimerization system. FRB is split into two parts (N-FRB and C-FRB). Trimerization is mediated by the rapamycin interaction.

In contrast to direct perturbation, induced dimerization approaches achieve regulation via controlling proximity of molecules in biological systems.^94,95 In optogenetic dimerization systems, photosensitive proteins undergo a conformational change upon illumination and induce protein–protein interaction (Figure 2b, top panel). Photoinducible dimerization modules such as fluorescent protein Dronpa, Cry2-CIBN, VVDs, and PhyB-PIF have been successfully used to control the activity of GTPases,^71,96 GEFs,⁹⁷ Abl tyrosine kinases,⁹⁸ bRaf kinases,⁹⁸ Cre recombinase,⁹⁸ and mitogen-activated protein kinase (MAPK) signaling.⁹⁹ LOV2-based, engineered light-induced dimerization systems include, tunable light-inducible dimerization tag (TULIP¹⁰⁰), improved light-induced dimer (iLID¹⁰¹), and LOV2 trap and release of protein (LOVTRAP¹⁰²). An example of a chemically induced dimerization (CID) system is FKBP and FRB, where dimerization is induced by rapamycin¹⁰³ (Figure 2b, bottom panel). Another variant of the induced dimerization approach is the split strategy.¹⁰⁴ In this approach, a protein is split into two halves, and then individual pieces are fused to form a dimer. Dimerization is induced either by light or a small molecule (Figure 2c). To computationally identify split sites on the basis of the “split energy” of the protein, Dagliyan et al. developed an approach called “split proteins regulated by a ligand or by light” (SPELL¹⁰⁴). Employing SPELL, they used their approach for regulating proteins such as tyrosine kinase, TEV proteinase, and guanine exchange factor. Another example for split strategy is rapamycin inducible Cas9 reported by Zetsche et al. Authors split Cas9 into two fragments and rendered chemically inducible by rapamycin-sensitive dimerization domains (FKBP and FRB) for controlled reassembly to mediate genome editing and transcription modulation.¹⁰⁵

Recently, Wu et al. rationally designed a rapamycin inducible trimerization system (CIT), comprising split FRB and FKBP proteins (Figure 2d). The CIT system operates on a time scale of seconds to minutes. Using CIT, authors induced triorganellar junctions and perturbed intended membrane lipids specifically at select membrane contact sites.¹⁰⁶

ENGINEERING ALLOSTERIC CONTROL OF PROTEIN FUNCTIONS IN DIVERSE PROTEIN FAMILIES

In the past decade, a wide range of proteins have been regulated via allosteric control. Here we discuss several examples to show the broad applicability of allostery-based protein regulations. Protein kinases regulate key physiological events of the cells and are therefore important therapeutic targets. Several kinases are allosterically engineered with spatiotemporal precision to uncover their functions. A chemogenetic approach was used for the regulation of kinase function of focal adhesion kinase (FAK).⁹⁵ In this approach, Karginov et al. used a rapamycin binding domain iFKBP and its partner FKBP12-rapamycin binding (FRB) domain. When iFKBP was allosterically inserted, the resulting chimeric FAK was inactive in the absence of rapamycin, and the function was restored upon the addition of rapamycin. A similar approach was used to create rapamycin-induced Src kinase by allosterically introducing the rapamycin binding uniRapR domain.^83,92 The same insertion site on Src was used to create photoinhibitable Src kinase (PI-Src) by using the LOV2 domain. Blue-light exposure induces the unfolding and undocking of the Jα helix from the LOV core resulting in inhibition of Src kinase, whereas in dark conditions the kinase function was uninhibited. Zhou et al. created photoswitchable psRaf1, psMEK1, psMEK2, and psCDK5 kinases using engineered Dronpa protein.⁹⁷ The input signal specificity in histidine kinase was reprogrammed by replacing its chemo-sensor domain by a light-oxygen-voltage (LOV) photosensor domain to create light-regulated. histidine kinase.⁸⁴ An allosterically engineered, photoswitchable pyruvate kinase M2 (PKM2) variant created by the LOV2 insertion showed reversible regulation of the enzyme function.¹⁰⁷ Kinase function regulation was achieved in Abl tyrosine kinase, Src kinase, and bRaf kinases by the allosteric insertion of a photodimerizable system VVD.⁹⁸

Apart from kinases, several other groups of proteins and enzymes are allosterically engineered to regulate specific functions. Lee et al. identified an important allosteric region on E. coli dihydrofolate reductase (DHFR) by statistical coupling analysis and inserted AsLOV2 domain into the identified surface-exposed region. AsLOV2 insertion at this region resulted in a blue-light-dependent increase of DHFR activity caused by the increased flexibility of the targeted region upon Jα helix transition.⁶⁸ Hahn and co-workers engineered a genetically encoded photoactivatable Rac1 (PA-Rac1) that facilitates specific spatial and temporal control of Rac activity in living cells.¹⁰⁸ Our group has created a set of blue-light-inactivated proteins, small GTPases, and guanine nucleotide exchange factors (GEFs) by inserting AsLOV2 into the allosteric region of proteins.⁸³ LOV2-based allosteric photocontrol of tyrosine phosphatase 1B (PTP1B) was achieved by Hongdusit et al.⁸⁶ Strickland and co-workers rationally designed an “allosteric lever arm” by the fusions between LOV2 and the E. coli trp repressor. This fusion protein, upon light exposure, selectively bound to operator DNA and protected it from nuclease digestion.⁸⁷ Ligand-dependent activation was shown in an allosterically regulated Cas9, which was created by the insertion of the estrogen receptor α ligand binding domain into Cas9.¹⁰⁹ Myosin and kinesin speed/directionality regulation was attained by designing LOV2-motor protein chimeric fusions. Photoactivation was achieved through conformation changes in the engineered “lever arm”.¹¹⁰ Wong et al. used AsLOV2 as a switch to create photoactivatable intein (LOVInC) to modulate the splicing activity of the split DnaE intein. Periodic blue light illumination of LOVInC resulted in protein splicing in mammalian cells.⁸⁸ Very recently, Toettcher’s group created the light-switchable nanobodies (OptoNBs), whose binding to proteins of interest (range of target epitopes) could be enhanced or inhibited upon blue light illumination.¹¹¹ Farrants et al. engineered chemogenetic nanobodies, whose affinity for green fluorescent protein (GFP) could be switched “on” and “ off” by using small molecules.¹¹²

CONCLUSIONS AND PROSPECTS

In proteins, allosteric control is a critical mechanism that allows the regulation of a myriad of biological processes such as metabolism, signal transduction, enzyme activity, and gene regulation.¹¹³ Engineering allosteric regulation of proteins facilitates the integration of one or more external signals to regulate the functional response of the protein. Until now, we have seen a great progress in engineering allosteric regulation in proteins. However, we still face a few challenges.

One of the critical challenges has been the prediction of allosteric sites and communication network in proteins, which is addressed by many experimental and computational methods summarized above. Recently, deep learning revolutionizes the field of protein structure prediction. The artificial intelligence network, named AlphaFold, created high-accuracy structures even for gene sequences with fewer homologous templates in the Critical Assessment of Protein Structure Prediction (CASP13).¹¹⁴ Encouraged by this innovation, we anticipate that the identification of allosteric sites and communication pathways will be significantly advanced by the application of machine learning methods. Currently, a few machine learning methods have been used for allosteric network prediction. The support the vector machine method used to distinguish allosteric hot spots from the nonhot spots only achieved a recall of 73–81% and a precision of 64–71%.¹¹⁵ Machine learning is still underexplored in the field of allostery. With the increasing database of allosteric and nonallosteric sites, we expect that allosteric site prediction will benefit from machine learning in the near future.

Another challenge in engineering allosteric communication in proteins is the limited number of orthogonal switches. Chemogenetic and optogenetic tools are the most widely used for modulating protein functions. Rational design of novel protein regulatory tools responding to diverse molecular cues, such as temperature, pH, and pressure, would expand the regulatory toolbox and facilitate the creation of higher order allosterically regulated protein systems. Some of these tools would find direct applications. For example, pH-regulated systems would find a lot of application in biomedical sciences because many of the disease onset and progression associate with a change in pH at the cellular and subcellular level. However, creating such novel tools is challenging because of the limited understanding of the protein dynamics. In addition, it requires an extensive characterization and optimization process.

So far, a multitude of proteins have been allosterically engineered to regulate specific activities. The same strategies could be applied to other protein families to explore their biological functions or construct complex biological circuits. G protein-coupled receptors (GPCRs) are one of the important targets, whose function could be allosterically regulated. The GPCR superfamily is encoded by over 800 human genes forming the largest class of superfamily of receptors.¹¹⁶ Currently, the understanding of GPCRs functional mechanism is relatively limited. Typically, a combination of native and synthetic pharmacology approach has been used to manipulate GPCRs. A number of agonists, antagonists, and modulators (including allosteric modulators) have been used for different subtypes of GPCRs.¹¹⁷ The main limitation associated with regulation using chemical ligands is the specificity. To overcome this obstacle, Roth and colleagues developed a chemogenetic approach termed DREADDs (Designer Receptors Exclusively Activated by Designer Drugs), which improved cell-type precision of GPCR activation.¹¹⁸ However, the fine details of specific GPCR-mediated signaling mechanisms are limited in this approach. As an alternative to DREADDs, opsin-based optical control of GPCR signaling has been attempted, but these methods are limited by the intrinsically slow off-kinetics of opsins.¹¹⁹ To solve these issues, chemo and opto toolkits could be used for allosteric regulation of GPCRs at the physiological setting with a higher spatiotemporal precision. Such regulation would allow a detailed mechanistic analysis and better understanding of GPCRs functionality, which will also have therapeutic applications.

Two-component systems typically comprise a transmembrane histidine kinase and a response regulator. Histidine kinase senses specific environmental signals, which allosterically changes intracellular conformation and activates auto-phosphorylation. The phosphoryl group is further transferred to response regulator and then induces downstream gene expression. Compared to the two-component system, the one-component system is much simpler containing both input and output domains in a single protein, such as the LacI lactose operon repressor. Taylor et al.¹²⁰ designed new variants of LacI that respond to four new inducer molecules based on computational protein design, saturation or random mutagenesis, and multiplex assembly. Similar strategies could be applied to two-component system to expand the sensor specificity and create novel switches.

Overall, allosteric regulation in proteins is still at its early stages of expansion to the application domain. The development of broadly applicable tools and strategies is highly desirable to achieve regulation in a wide range of proteins. The increase in the understanding of structure–function relationship in proteins and advances in the computational modeling and machine learning methods should pave the way for generalized approaches. In the coming days, we anticipate advanced strategies for rational design of allosteric protein switches and, in turn, protein regulation, benefiting a plethora of applications in various fields.