Engineered control of enzyme structural dynamics and function

Abstract

Enzymes undergo a range of internal motions from local, active site fluctuations to large‐scale, global conformational changes. These motions are often important for enzyme function, including in ligand binding and dissociation and even preparing the active site for chemical catalysis. Protein engineering efforts have been directed towards manipulating enzyme structural dynamics and conformational changes, including targeting specific amino acid interactions and creation of chimeric enzymes with new regulatory functions. Post‐translational covalent modification can provide an additional level of enzyme control. These studies have not only provided insights into the functional role of protein motions, but they offer opportunities to create stimulus‐responsive enzymes. These enzymes can be engineered to respond to a number of external stimuli, including light, pH, and the presence of novel allosteric modulators. Altogether, the ability to engineer and control enzyme structural dynamics can provide new tools for biotechnology and medicine.

Introduction: The Case for Dynamics

Enzyme catalysis is a very dynamic process: chemical bonds in substrates are broken and formed to yield products, and enzymes often undergo structural changes during these events. In this review, we use the term “structural dynamics” to mean time‐dependent changes in the three‐dimensional protein structure. The amplitudes and timescales for these structural changes can range over many orders of magnitude, from bond vibrations occurring on the femtosecond timescale, to loop motions on the microsecond to millisecond timescales, and to large‐scale domain rearrangements that might occur over seconds or longer;1 all of these events can be important for enzyme function. We recognize that there continues to be controversy about if and how enzyme structural dynamics relates to chemical events and their associated rates.2, 3 Here, we take enzyme catalysis to include all physical and chemical events from initial substrate binding to final product release. Indeed, the overall catalytic rate often depends on events other than chemical catalysis.3 While our understanding of the connections between protein structural dynamics and enzyme function remains poor, even our incomplete understanding has provided a framework to begin engineering protein structural dynamics to change enzyme catalytic activity (e.g., Ref. 4), substrate selectivity (e.g., Ref. 5), the nature of the catalyzed chemical reaction (e.g., Ref. 6), and/or the enzyme's physiochemical properties (e.g., temperature stability7). In this review, we note many of the ways that protein structural dynamics can be altered to modify enzyme function, including site mutations, protein hybrids and other post‐translational chemical modifications. We pay special attention to how these modifications can generate stimulus‐responsive enzymes (e.g., light‐activated), providing unique biotechnology tools. We note that our examples are illustrative, not exhaustive, and that there are other recent reviews related to this area that may highlight alternative cases or approaches.8, 9, 10, 11, 12, 13

Enzyme Structural Dynamics and the Chemical Step

It has been proposed that in some enzymes, there are so‐called rate‐promoting motions that tend to “push” reacting atoms closer together, effectively compressing the reaction coordinate.14, 15, 16, 17, 18 For example, a “network of coupled promoting motions” in Escherichia coli dihydrofolate reductase (DHFR) enzyme has been suggested to sterically push the substrate dihydrofolate (DHF) and the NADPH cofactor closer together, gating the distance between the hydride donor and acceptor atoms19 [Fig. 1(A)]. Amino acid substitutions at these network residues result in changes to the hydride transfer rate and the ability of DHFR to promote hydride quantum tunneling.20, 21, 22, 23, 24 Many of the amino acid changes also lead to changes in DHFR structural dynamics as evidenced by nuclear magnetic resonance (NMR) spectroscopy and molecular dynamics (MD) simulations.25, 26, 27, 28, 29, 30, 31

Figure 1.

Promoting motions proposed to be important in the compression of the reaction coordinate in enzyme catalyzed reactions. A. Molecular dynamics simulations suggest a network of motions in dihydrofolate reductase that sterically compresses the reaction coordinate and gates the distance between hydride donor (C_D) and acceptor (C_A) atoms. Reprinted with permission from Ref. 19. B. Formation of small pockets near the NAD cofactor of thermophilic alcohol dehydrogenase through the substitution of Val260 (or Leu176) with Ala leads to changes in the catalytic turnover rate (k_cat) and the H/D kinetic isotope effect (KIE) for hydride transfer. The k_cat values at different temperatures for hydrogen and deuterium transfer are shown in black (plotted as a function of the left axis), and red and blue indicate the KIE (=k_cat,H/k_cat,D) highlighting the temperature break in Arrhenius behavior around 30°C (plotted as a function of the right axis). The WT and V260A enzymes have opposing behaviors in the temperature dependence of the KIE. In the V260A variant, hydride tunneling is less important at low temperatures (represented by the blue line) because preorganization is disrupted. At high temperatures (represented by the red line), hydrogen tunneling is more important, but the created cavity increases the need for donor‐acceptor distance sampling (i.e., reorganization is also disrupted). Adapted with permission from Ref. 48. C. Proposed mass‐altered bond vibrational modes in the ¹H, ¹²C, ¹⁴N “light” and ²H, ¹³C, ¹⁵N “heavy” human purine nucleoside phosphorylase enzymes. Green arrows tend to promote transition‐state barrier crossing, and red arrows are orthogonal to the reaction coordinate. In the “heavy” enzyme, local motions are reduced, as indicated by the smaller arrows, including those proposed to be important for compressing the distance between the O5’ and O4’ oxygens of the ribose ring, as indicated by the hashed green line. One key motion involves His257; amino acid changes designed to less sterically constrain His257 lead to mass‐independent barrier crossing rates. Adapted with permission from Ref. 56.

Similar findings have been found in other enzymes, especially those in which quantum tunneling may play a role.32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42 Quantum tunneling behavior is often probed using kinetic isotope effects (KIEs) (see reviews43, 44). In many cases, KIEs are temperature‐independent for the wild‐type (WT) enzyme but become more temperature‐dependent when optimal structural dynamics have been perturbed.22, 32, 33, 45 It has been suggested that these perturbations limit the ability of the enzyme to sample tunneling‐ready active site conformations, requiring more motion at higher temperatures to achieve the required optimal donor‐acceptor distance.46 Results from the thermophilic alcohol dehydrogenase from Bacillus stearothermophilus (htADH) provide even more insight. In WT htADH, the KIE is temperature‐dependent below 30°C, but largely temperature‐independent above 30°C (i.e., nearer its temperature optimum).47, 48 The temperature trend in KIEs is reversed in variants in which small cavities have been introduced near the NAD(H) cofactor by substituting residues Leu176 and Val260 for less bulky residues.48, 49 These results have been interpreted in regards to a model in which both large‐scale global conformational sampling, or “preorganization”, and smaller‐scale local adjustments, or “reorganization”, are important for enzyme function.46, 48 The htADH variants are unable to achieve the necessary conformations for hydrogen tunneling at low temperatures (i.e., preorganization is disrupted), so quantum tunneling is a less important rate factor. At higher temperatures, the overall protein structure is more conducive to hydrogen tunneling, but the created cavities increase the need for donor‐acceptor distance sampling (i.e., reorganization is disrupted) [Fig. 1(B)].

Rate‐promoting vibrations and active site compression may also be important in other types of enzyme reactions. For example, Bruice and colleagues have suggested that the enzymes chorismate mutase and indole‐3‐glycerol phosphate synthase (InGPS) aid their associated chemical reactions by holding their substrates in constrained, more chemically reactive conformations.50, 51 Active site compaction may also be an important consideration in methyl transfer catalyzed by catechol‐O‐methyltransferase.52, 53 In human purine nucleoside phosphorylase (hPNP), it has been proposed that the movement of His257 helps compress the distance between the O5’ and O4’ oxygens of the ribose ring, which promotes re‐shuttling of the electrons into the purine ring and transition‐state barrier crossing.54 Like in previous examples, mutagenesis studies suggest that there is a coupled network of motions, even including residues on the surface of the enzyme.55 Besides mutagenesis, comparisons between light‐ (i.e., ¹H, ¹²C, ¹⁴N) and heavy‐isotope (i.e., ²H, ¹³C, ¹⁵ N) labeled hPNP have been used to gain insight into rate‐promoting vibrations.56, 57, 58 The “heavy” hPNP enzyme had a 9.9% increase in mass compared to the normal “light” enzyme, and the rate constant for the chemical step decreased by 30%.56 Hybrid quantum mechanics/molecular mechanics (QM/MM) calculations indicated that the heavy enzyme had mass‐altered vibrational modes, leading to suboptimal transition‐state barrier crossing58 [Fig. 1(C)]. Consistent with these proposals, the “light” and “heavy” enzymes respond differently to amino acid substitutions designed to affect catalytic site vibrational modes.6 For instance, amino acid substitutions of Glu258 to Asp and Leu261 to Ala were designed to reduce steric contacts with His257 and allow for a greater range of motion, and potential uncoupling of protein and substrate motions. Unlike in the WT enzyme, “light” and “heavy” versions of the E258D/L261A double variant had almost identical rate constants for the chemical step. These results are very analogous to those for the discussed htADH variants discussed above. Studies with other heavy‐isotope labeled enzymes have led to similar ideas.59, 60, 61, 62, 63

Modulation of Loop Dynamics

Loops are among the most structurally dynamic features of proteins.64, 65 They always link more ordered secondary structures (i.e., α‐helices and β‐strands), and can range from well‐defined turns to random‐coil‐like structures. Many of these loops are “invisible” to structure determining methods owing to their conformationally dynamic nature.66 Loops can be important for a range of functions, including interactions with ligands, other parts of the protein, or other macromolecular binding partners.64, 65 Loop motions can also play direct roles in chemical catalysis through positioning catalytically important amino acid residues (e.g., Ref. 67).

While the length, amino acid composition, and the presence of hydrogen‐bonds can all affect loop motions,68 loops often interact with each other in order to gate their motions.69, 70 An illustrative example can be found in E. coli DHFR. There are three major loops in DHFR, those being the Met20, FG, and GH loops71 [Fig. 2(A)]. The Met20 loop folds over the active site and changes conformation between the open, closed, and occluded conformations.71, 72 In the closed conformation, the FG loop contacts the Met20 loop through interactions between the Asp122 side chain and the Gly17 backbone. After the chemical step, the Met20 loop assumes the occluded conformation that blocks the nicotinamide ring of the NADP⁺ cofactor from entering the active site. In the occluded conformation, the FG‐Met20 loop interactions are broken, and new interactions form between Ser148 of the GH loops and the Asn23 backbone. Amino acid changes to alter the FG‐Met20 or Met20‐GH loop interactions results in changes to the catalytic cycle rate constants.73, 74 For example, the Gly121Val substitution on the FG loop, which perturbs the closed conformation, leads to a 40‐fold decrease in NADPH binding and a 200‐fold decrease in hydride transfer rate.73 The substitution results in changes to the loop motions, and motions elsewhere in the protein, on the picosecond‐nanosecond and microsecond‐millisecond timescales, according to NMR relaxation analyses.25, 27 Amino acid changes to position 148 on the GH loop also result in changes to ligand affinities and ligand off‐rates.74

Figure 2.

Gated loop motions important in enzyme function. A. In E. coli dihydrofolate reductase, the FG and GH loops compete for hydrogen‐bonding interactions with the active site Met20 loop. When interactions between the FG and Met20 loops are disrupted, the occluded conformation is favored and the hydride transfer rate decreases. When interactions between the GH and Met20 loops are disrupted, the closed conformation is favored and substrate and cofactor affinities are altered. B. In S. solfataricus indole‐3‐glycerol phosphate synthase, there are competing interactions for the active site β1α1 loop. When either set of interactions is lost, the second chemical step (i.e., dehydration) is rate limiting, but if both sets of interactions are disrupted, the first chemical step (i.e., ring closure) is rate limiting similar to the WT enzyme. Results suggest that the competing interactions keep the β1α1 loop appropriately dynamic. C. In ribonuclease A, the protonation state of His48 helps to control μs‐ms timescale motions. His48 makes interactions with loop 1 and β‐strands housing functionally important residues. The μs‐ms timescale motions can be evaluated using NMR relaxation dispersion methods (see182). When the time between successive 180° refocusing pulses (τ_cp) is increased in the NMR relaxation dispersion experiments, conformational exchange can contribute to a higher apparent R₂ transverse relaxation rate. Contributions to the R₂ from conformational exchange are suppressed when τ_cp is decreased. Some motions are repressed when His48 is substituted with Cys (yellow curves, left) or Ala (black curves, left), but these motions are essentially restored by re‐introducing the imidazole moiety through reaction of the H48C variant with 4‐chloromethyl imidazole (pink curves, right). Some motions in the WT enzyme are also pH dependent (pH 7, brown curves; pH 6.4, blue curves). The motions of Asn71 are not perturbed by His48 substitution or by a change in pH. These studies indicate that important motions can be pH modulated and changed through covalent modification. Reprinted from Ref. 90 with permission from Biophysical Society.

The loop interactions in DHFR are also likely important for catalytic flux, which may depend on the cellular environment and NADP⁺/NADPH concentrations.75, 76 For example, the occluded conformation is disfavored by a polyproline sequence at the C‐terminal end of the Met20 loop in human DHFR. When this sequence was introduced in the E. coli homolog (i.e., N23PP), the µs‐ms timescale dynamics were ablated, although the global protein structure did not change substantially.77 The hydride transfer step was also severely impaired, which was put forth as evidence that the structural dynamics of these loops was essential for chemical catalysis in this enzyme,77 although there exists skepticism regarding the direct connection between protein motions and hydride transfer.78 The loop dynamics may also be important for biological function, considering that human DHFR is unable to rescue DHFR‐deficient E. coli cells.76

Similar loop‐loop interactions are important in other enzymes,69, 70 including for InGPS. As with other (β/α)₈ barrel enzymes, the residues directly involved in chemical catalysis lie on the β‐strands and on loops connecting the β‐strands to the proceeding α‐helices.79 InGPS catalyzes the ring closure and dehydration of 1‐(o‐carboxyphenylamino)‐1‐deoxyribulose‐5‐phosphate (CdRP) to form indole‐3‐glycerol phosphate (IGP) in two discrete steps. In the Sulfolobus solfataricus enzyme (ssInGPS), the rate‐limiting step changes based on the temperature, suggesting a connection between enzyme structural dynamics and the kinetic mechanism.80 It has been suggested that substrate binding, product release and intermediate reorientation are coordinated by the β1α1 loop dynamics.51, 81, 82, 83 The β1α1 loop interacts with the β2α2 loop through a hydrogen bonding interaction between Arg54 and Asn90 [Fig. 2(B)]. The N90A substitution (on the β2α2 loop) induces small changes to the InGPS kinetic parameters, changes the identity of the rate‐limiting step to the dehydration step, and leads to a surprising decrease in β1α1 loop flexibility.84 The β1α1 loop also interacts with residues on the adjacent α1 and α8’ helices. Removal of these interactions through the R64A/D65A substitutions also results in the dehydration step becoming rate‐limiting and a decrease in the β1α1 loop flexibility.85 However, severing the hydrogen bonds on both the N‐ and C‐terminal sides of the loop as in the R64A/D65A/N90A variant restores some flexibility to the β1α1 loop and the ring closure step becomes rate‐limiting again.85 We have suggested that this “tug‐of‐war” between hydrogen‐bonding partners on either side of the β1α1 loop facilitates the motions needed to properly catalyze the ssInGPS reaction.85 This type of “frustration” (i.e., the inability to fulfill all potential interactions simultaneously) may in general be important for controlling protein structural dynamics.86

Changing loop motions offers a means to control enzyme structural dynamics in a stimulus‐dependent manner. For example, hydrogen bonding interactions from the side chain of His48 helps to control loop dynamics and function in ribonuclease A (RNase A).87, 88, 89, 90 As also observed for E. coli DHFR,91 there are both pH‐sensitive and pH‐insensitive motions in RNase.90 Interestingly, the pH‐dependence of the motions closely resembles the pH‐dependence of the RNase A catalytic activity,90 which might both be dependent on the protonation state of His48. The H48C substitution abrogates the pH‐dependent motions in loop 1 and connected regions, but covalent attachment of 4‐methylimidazole at Cys48 restores these motions90 [Fig. 2(C)]. These studies elegantly demonstrate that functionally important motions can be pH‐modulated, and these motions can be activated (or potentially deactivated) through covalent modification at critical residues. Chemical rescue of perturbed noncovalent interactions may be a general way to allosterically control enzyme catalytic activity as also shown in indole rescue of Trp to Gly variants of β‐glycosidase and β‐glucuronidase enzymes.92

Enzymes as Amino Acid Interaction Networks

Globular proteins can be viewed as three‐dimensional interaction networks.9, 12, 93, 94 Many structure‐ and sequence‐based methods have been developed to identify amino acid interaction networks in proteins.95 For example, networks have been proposed for E. coli DHFR based on amino acid covariation in multiple sequence alignments,96, 97 molecular dynamics simulations19 and X‐ray crystallography analysis.98, 99 The CONTACT (Contact Networks Through Alternate Conformation Transitions) algorithm uses high resolution X‐ray data to identify alternative side chain conformations that may propagate motions through steric interactions98, 100 [Fig. 3(A)]. This method provided atomic‐level insight into how distal amino acid substitutions, like G121V and S148A, can propagate their effects throughout the enzyme. Similar types of network analyses have contributed to our understanding in many other enzyme systems.101, 102, 103, 104, 105, 106, 107

Figure 3.

Dynamic amino acid interaction networks in the function and allosteric regulation of enzymes. A. The CONTACT approach, based on fitting alternative side‐chain conformations in X‐ray diffraction maps, was used to identify amino acid interaction networks in E. coli dihydrofolate reductase. This approach provides experimental insight into the connections between distant parts of the enzyme, including those regions identified as being part of a network of coupled promoting motions [see Fig. 1(A)]. Reprinted by permission from Macmillan Publishers Ltd.98 B. The CHESCA approach,111 based on analyzing NMR chemical shift changes induced by protein perturbations, was used to identify amino acid interactions networks in the alpha subunit of tryptophan synthase. The orange and blue spheres indicate two different networks. The catalytic residue Glu49 has a different network association in the apo resting state and in the catalytically active working state. Network residues also include those at the alpha/beta interface. C. Amino acid interaction network communities (represented by different colors) in imidazole glycerol phosphate synthase from the histidine biosynthetic pathway, which includes both HisH and HisF subunits. Ammonia is passed from the HisH to the HisF subunits. Binding of PRFAR (N’‐[(5’‐phosphoribulosyl)formimino]‐5‐aminoimidazole‐4‐carboxamide ribonucleotide) changes community interactions. Reprinted with permission from Ref. 116. D. Statistical coupling analysis, based on amino acid covariation in a multiple sequence alignment,166 was used to identify amino acid interaction networks (or sectors) in indole‐3‐glycerol phosphate synthase. Residues highlighted in red are those at the active site. Amino acid substitutions at many of these positions lead to changes in competitive effectiveness of yeast strains carrying these variants, according to high throughput mutational analyses. Reprinted with permission from Ref. 132.

These long‐range networks may also offer means through which enzymes communicate in multi‐enzyme complexes.97 For example, we used NMR chemical shift covariance analysis108, 109among a series of site‐directed protein variants to identify amino acid interaction networks in the alpha subunit of E. coli tryptophan synthase (αTS).110, 111 Our analysis revealed that there are two networks of residues [colored orange and blue in Fig. 3(B)], and the residues involved in these networks change depending on what is bound to the protein.110, 111, 112 Intriguingly, Glu49, a residue directly involved in catalysis, belongs to different networks when comparing the apo resting state and when the enzyme is actively turning over substrate/products (i.e., the working state).111 Molecular dynamics (MD) simulations also indicated that the motions of Glu49 are correlated to different sets of residues in these states.111, 113, 114 Amino acid substitutions at surface‐exposed network positions led to modest changes in αTS kinetic parameters.111 These networks may also be important for how αTS communicates and coordinates channeling of its product indole into the active site of the beta subunit of tryptophan synthase (βTS). An amino acid substitution known to disrupt direct channeling of indole from the α‐ to β‐active sites also disrupts networks that might transmit structural dynamic information from the α‐active site to the α/β binding interface.112

The functional coordination of different active sites through amino acid interaction networks is exemplified by the histidine biosynthetic enzyme imidazole glycerol phosphate synthase (ImGPS).115, 116, 117, 118 ImGPS is a heterodimeric enzyme consisting of the HisH subunit, which catalyzes the hydrolysis of glutamine to produce ammonia, and the HisF subunit, which receives the ammonia and catalyzes its reaction with PRFAR (N’‐[(5'phosphoribusyl}formimino]‐5]aminoimidazle‐4‐carboxamide ribonucleotide).119 The glutaminase activity of the HisH subunit is dependent on binding of PRFAR to the HisH subunit.120 The Batista and Loria groups have used MD simulations to identify correlated motions in ImGPS and community analysis to identify allosteric pathways116 [Fig. 3(C)]. Amino acid substitutions at network positions in the HisH subunit led to changes in the microsecond‐millisecond timescale dynamics according to NMR relaxation dispersion spectroscopy, and decreased PRFAR‐enhancement of the HisF catalytic activity.117 Intriguingly, these groups have also found a small molecule inhibitor that binds at the HisH‐HisF interface also disrupts ImGPS motions coordinated by the proposed network.121 These results and many others (e.g., Refs. 10, 122, 123, 124, 125, 126, 127) have suggested that surface‐exposed network residues may serve as druggable interfaces for allosteric modulators.

High throughput mutational analyses128, 129, 130, 131 may allow for more comprehensive interrogation and validation of these amino acid interaction networks. For example, the Matthews groups conducted a comprehensive mutational analysis of three InGPS enzymes (i.e., >5000 mutations) expressed in yeast cells in which the endogenous IGPS gene was deleted,132 following the deep mutational scan approach developed by the Bolon group.133 One of their most intriguing findings was the discovery of a long‐range allosteric pathway connecting the outer α‐helical shell to the active site, as elucidated by the analysis of beneficial mutations in their high throughput analysis, and consistent with amino acid covariation in the naturally occurring sequences of InGPS enzymes132 [Fig. 3(D)]. The connection between surface‐exposed sites and the active site is very reminiscent of the NMR‐derived networks we elucidated for the structurally and functionally related αTS enzyme.111

Controlling Enzyme Structural Dynamics through Covalent Constraints

In most of the above discussion, the goal was to modify noncovalent interactions to affect enzyme function and structural dynamics. Motions might also be altered through the introduction of covalent constraints, or “molecular stapling.” Cysteine mutagenesis has been commonly done to introduce disulfide bonds to measure their impact on protein structure and stability,134, 135, 136 including classic enzyme examples.137, 138, 139, 140 For example, Ma et al. introduced disulfide bonds to target the fluctuations of mobile elements in T7 RNA polymerase139 [Fig. 4(A)]. Through these studies, they were able to identify conformations and their associated transitions most important for different stages of RNA replication, including initiation and elongation. There is also the potential to control the catalytic function of enzymes with introduced disulfide bonds through their redox states,138 or even potentially through microwave irradiation.141 Other types of crosslinking technologies are also being developed,142, 143, 144 including those that can be genetically encoded.145, 146

Figure 4.

Stimulus‐responsive covalent constraints can control enzyme conformational changes and function. A. Disulfide bridges were engineered to capture the initiation (IC; C147‐C292) and elongation (EC; C179‐C750) complexes of T7 RNA polymerase. Reprinted with permission from Ref. 139. B. Light‐dependent enzyme activity through engineering azobenzene‐crosslinked PvuII restriction endonuclease. Photoconversion of the cross‐linked azobenzene moiety to the cis form results in greater restriction endonuclease activity. Reprinted with permission from Ref. 152.

Some of the most interesting cross‐linkers are those sensitive to environmental conditions. For example, azobenzene and its derivatives have proven to be useful photoswitches, as they will undergo a cis/trans isomerization which can lead to structural changes to their attached biomolecule.147, 148, 149 An azobenzene photoswitch was covalently linked across a binding groove of a PDZ domain, resulting in conformational changes that mimic those involved in ligand binding.150 MD simulations also supported the notion that the photoswitch mimicked the transitions between ligand free and bound forms of the protein.151 In another example, multiple azobenzene‐crosslinked derivatives of the restriction enzyme PvuII were generated,152 guided by the sGAL program that helps to find suitable surface‐exposed sites for Cys‐mediated crosslinks153 [Fig. 4(B)]. Those variants with double crosslinks, especially those that included crosslinks closer to the active site, gave the greatest enzyme catalytic activity difference between the cis and trans configuration of the azobenzene photoswitches.152 It should be noted that in some cases the azobenzene does not need to act as part of a crosslink to affect an enzyme's catalytic activity. Azobenzene‐maleimide probes were attached through a single Cys substitution on an important loop in bacterial histone deacetylase amidohydrolase (HDAH) to create a light‐switchable enzyme.154 The modified HDAH enzyme was more active when the azobenzene photoswitch was in the trans‐ compared to the cis‐configuration, potentially because in the cis‐configuration the moiety interacts with the enzyme surface.155

Surface‐exposed Cys residues may in general serve as attachment points for novel, allosteric modulators. Recently, Wu and coworkers have created the “cysteinome”—a comprehensive database of proteins with targetable cysteines along with any known covalent inhibitors.156 Indeed, Backus and coworkers performed a quantitative analysis of small molecule fragments reactive towards cysteine.157 Covalent allosteric drugs were previously known for several high priority drug targets, including GTPases, kinases, GPCRs, heat‐shock proteins and ribonucleases.158, 159 Natural post‐translation modifications may likewise affect protein function through allosteric mechanisms.160 All of these effects may depend in part on how covalent modification alters protein structural dynamics.

Grafting Proteins Domains to Control Enzyme Function

Larger global changes to protein structural dynamics can be accomplished by creating hybrid or fusion proteins, in which existing regulatory domains are engineered onto another protein.161, 162, 163, 164, 165 The hybrid enzymes can then respond to the presence of different ligands or even light, likely through induced conformational changes.162 For example, the Ranganathan and Benkovic groups created a light‐dependent DHFR enzyme by inserting the light‐sensitive LOV2 domain into the FG loop of DHFR based on network sites derived from their amino acid covariation analysis known as statistical coupling analysis96, 166, 167, 168 [Fig. 5(A)]. Other network methods169 and knowledge of the structural dynamics of the regulatory and catalytic domains can inform on the appropriate attachment points. Hahn and colleagues created a number of ligand‐ and light‐dependent hybrid enzymes by inserting regulatory domains into loops that were determined by MD simulations to be mechanically coupled to the active site170, 171 [Fig. 5(B)]. These studies also exemplify the vast practical utility of these biotechnology tools to explore biological interactions and function.170, 172

Figure 5.

Chimeric enzymes to control structural dynamics and catalytic function according to external stimuli. A. Statistical coupling analysis was used to identify surface‐exposed network residues in E. coli dihydrofolate reductase and the light‐responsive LOV domain. Engineering of chimeric enzymes to join these positions results in light‐dependent catalytic conversion of dihydrofolate (DHF) to tetrahydrofolate (THF); use of other non‐network, surface‐exposed sites does not lead to light‐dependent enzyme activity. Reprinted from Ref. 97 with permission from Cell Press. B. Engineering of chimeric β‐lactamase‐maltose binding proteins. The c4 fusion protein as shown in the ribbon diagram is not responsive to the presence of maltose, but the introduction of additional glycine residues into the linker, which presumably increases molecular flexibility, led to maltose‐ and temperature‐dependent catalytic activity and biological response (according to ampicillin minimum inhibitory concentration, MIC). Reprinted with permission from Ref. 163.

Future Outlook

There have been tremendous advancements over the past decade in our understanding of protein structure and design173 and enzyme engineering,174 along with a continued acknowledgement of the importance of protein structural dynamics to enzyme function and regulation. We may be approaching a time when not only de novo protein design yields desired structures, but fluctuations between two or more protein conformations on desired timescales can be more aptly engineered.175, 176, 177 For example, the Goto and Chica labs used multistate design to induce Streptococcal protein G domain β1 protein to fluctuate between two designed conformations.177 Similar methods might be used to more rationally engineer enzyme structural dynamics. Studies presented in this review and elsewhere have shown that enzyme structural dynamics are naturally, or can be engineered to be, sensitive to external stimuli, including changes in pH, light, pressure and covalent modification. This ability to control enzyme structural dynamics may provide many powerful biotechnology tools. It should also be pointed out that issues pertaining to molecular structural dynamics extend to non‐biological catalysts (e.g., Refs. 178, 179, 180). As these molecular machines181 become larger and more complex, there will be an increased need to control the magnitude, trajectories, and timescales of their moving parts. Lessons learned from enzymes and other proteins will likely be applicable for resolving these issues.