Mining electron density for functionally relevant protein polysterism in crystal structures
- First Online:
- Cite this article as:
- Fraser, J.S. & Jackson, C.J. Cell. Mol. Life Sci. (2011) 68: 1829. doi:10.1007/s00018-010-0611-4
This review focuses on conceptual and methodological advances in our understanding and characterization of the conformational heterogeneity of proteins. Focusing on X-ray crystallography, we describe how polysterism, the interconversion of pre-existing conformational substates, has traditionally been analyzed by comparing independent crystal structures or multiple chains within a single crystal asymmetric unit. In contrast, recent studies have focused on mining electron density maps to reveal previously ‘hidden’ minor conformational substates. Functional tests of the importance of minor states suggest that evolutionary selection shapes the entire conformational landscape, including uniquely configured conformational substates, the relative distribution of these substates, and the speed at which the protein can interconvert between them. An increased focus on polysterism may shape the way protein structure and function is studied in the coming years.
KeywordsPolysterism Conformational substates X-ray crystallography NMR Evolution Protein dynamics Catalysis
Nitrogen regulatory protein C
Monod, Wyman and Changeaux (model of allostery)
The idea that proteins fluctuate between many structures of an ensemble is well established . In contrast to the ensemble view, protein X-ray crystallography generally follows a convention, established with myoglobin and hemoglobin in the middle of the last century [2, 3], to describe the model built into the electron density map in the singular form: “we have solved the structure of protein X”. The solved structure generally represents the best fit of a single polypeptide model to the electron density generated by the diffraction pattern of ~1015 molecules in the crystal lattice during seconds to hours of data collection. The protein molecules within a crystal populate a ‘conformational landscape’ dominated by nearly iso-energetic and conformationally similar substates . However, the refinement process, with few exceptions [5, 6, 7], is designed to converge on the single structure.
Here we review how crystallography can be used to reveal conformational substates that interconvert independent of any conformational changes induced by ligand binding or other perturbations. We refer to pre-existing conformational diversity as polysterism (as in [8, 9]), making a semantic distinction between structural diversity (polysterism) and sequence diversity (polymorphism). Early investigations of polysterism that used crystallography and complementary solution methods are highlighted before we focus on several recent studies that have provided new insight into functionally important polysterism, often from a single electron density map. These studies reveal a general theme: the minor substates of the conformational landscape often resemble the structures that play important roles in ligand binding, enzyme catalytic cycles, or evolutionary transitions. Thus, characterizing the polysterism of a single protein can yield insight by identifying minor conformations that may be sampled during evolution or during a reaction cycle.
Crystalline enzymes are often active
Despite the stable lattice arrangement necessary for X-ray diffraction, many enzyme crystals are capable of catalyzing chemical reaction [10, 11]. However, their activity is often reduced compared to the free enzyme in a crowded solution [12, 13]. Because the crowding of a typical protein crystal lattice is similar to the macromolecular crowding in a biological cellular environment [14, 15, 16], the reduction in catalytic efficiency likely originates from specific conformational restrictions of the crystal lattice. These observations provided some of the first indications that the ability to transition between structurally distinct conformational substates may be a functionally important property of proteins.
Crystallography gives endpoints for an induced fit view
The importance of structurally distinct substates in enzyme function was first promoted through Koshland’s  induced fit hypothesis, which proposed that proteins move between discrete structures as a result of interaction with a ligand. Induced fit theory, which was originally based on the necessity for substrate-specific catalytically competent structure of hexokinase to allow activity with glucose and prevent activity with other potential substrates such as water, was generalized to most enzymes and receptors. Koshland’s predictions were later supported by clear structural differences between ligand-bound and unbound crystal structures of carboxypeptidase and hexokinase [18, 19, 20], providing a structural corroboration for the importance of conformational rearrangements during the catalytic cycle of some enzymes. These structural studies established that solving independent crystal structures with different ligands is a powerful tool that can be used to observe the end-points of conformational change. The importance of conformational change in enzyme function was buttressed by solution experiments such as volume changes calculated from centrifugation and small-angle X-ray scattering experiments [18, 20].
Many conformational substates are present in a protein crystal
Despite an initial emphasis on the single-structure view of proteins, polysterism has been widely appreciated from the beginnings of structure–function studies. For instance, polysterism was a fundamental aspect of the allosteric model proposed by Monod, Wyman and Changeaux (MWC)  in 1965, in which allosteric effectors were proposed to function through changing the pre-existing distribution of active or inactive conformational substates within a population of protein molecules. Polysterism was also an important part of statistical mechanical descriptions of protein conformation and function [22, 23], while solvent interaction was predicted to introduce a heterogeneity in proteins at an early stage .
Although a role for intrinsic conformational diversity was recognized conceptually, the structural evidence to support it was initially slight. In the early stages of the field of protein crystallography, when high-resolution diffraction was relatively rare, polysterism was often not observable, even if it was present. More importantly, crystallization, by its very nature, favors ordered aggregation of molecules that are homogeneous in composition and conformation, establishing the direct relationship between crystalline order and diffraction. Indeed, large regions of disordered solvent molecules were recognized as a likely origin of X-ray diffraction resolution limits . These observations introduced a paradox—in order to resolve functional conformational heterogeneity, we need high-resolution X-ray diffraction, but intrinsic heterogeneity limits diffraction.
To resolve this apparent contradiction, the pioneering studies of Frauenfelder  used an indirect readout of heterogeneity at moderate resolutions: atomic B-factors (also called Debye–Waller factors, temperature factors, atomic displacement parameters, and thermal parameters). The B-factor fits the decay of electron density surrounding the modeled mean position of an atom. Using atomic B-factors, they calculated the magnitudes of the spatial distribution of protein atomic movements without resolving independent discrete conformations. Frauenfelder and Petsko used crystals of myoglobin and variable diffraction temperatures to discover differences between the mobility of core and surface atoms. They observed that the magnitude of motion varied according to the distance from the central haem-group. Building upon this observation, they produced a model of a rigid core surrounded by semi-liquid regions to the outside, and suggested a possible ‘dynamic’ pathway for substrate and product diffusion. This analysis suggested that conformational substates, particularly for side chains, exist within the crystal and that temperature can affect the magnitude of motion within these substates. Concurrently, Phillips and coworkers  demonstrated a conservation of the distribution of B-factors between homologous proteins crystallized in different space groups. Since the homologous proteins crystallized in different lattice arrangements, their results suggested that B-factor distributions are, at least partially, independent of the crystallization process. Additionally, the conservation of high B-factors at the entrance to the active site suggested that the protein mobility might have been maintained throughout evolution and play a significant part in the enzymatic activity. These groundbreaking papers provided structural support for the role of polysterism and heralded a new view within protein crystallography: there are interconverting structures within the crystal, revealed by deeply considering experimental data, beyond the singular model of the protein.
Evidence for conformational substates from solution
While crystallography has remained popular for independently revealing interconverting structures, substantial evidence for the importance of polysterism has come from solution methods. In particular, nuclear magnetic resonance (NMR) relaxation-based methods have opened up the field of protein dynamics at a wide range of time scales . Order parameters of fast (ps–ns) time scale dynamics can lead to degenerate models of motion . These measurements are limited to backbone amides and specific side-chain residues and do not provide any information about coupling or correlated motion between sites . Intermediate (μs–ms) time scale motions revealed by CPMG (Carr Purcell Meiboom Gill, R2) or R1ρ relaxation dispersion methods yield information about two- or three-state differences in chemical shift, population, and exchange rates . However, since distinct states are inferred due to differences in chemical shift, many types of motion can be invoked to explain the signal. Further, movement of surrounding residues, rather than the residue being probed, can drive the chemical shift differences. The simultaneous fitting of relative population, exchange rates, and chemical shifts is generally underdetermined for an individual residue. This is normally overcome by implying a collective motion for a group of residues and jointly constraining their relative populations and interconversion rates . NMR relaxation methods can identify groups of residues that likely experience exchange between states, but these methods do not provide a reference to a causal structural mechanism of the dynamic signal.
The power of these techniques was demonstrated in experiments that probed the allosteric behavior of the signaling protein nitrogen regulatory protein C (NtrC). Relaxation methods demonstrated that phosphorylation affects a population shift between two pre-existing conformations, providing strong validation of the MWC model of allostery . Mutational perturbations to the structural transition pathway predicted by molecular dynamics simulations confirmed that phosphorylation stabilizes the minor conformation of the unphosphorylated state . Later in this review, we describe three enzymes where different classes of structural rearrangements have been probed by relaxation methods and how crystallography can be leveraged to provide a structural interpretation of the dynamic signal.
Connecting polysterism to enzyme function
Much of the effort in protein structure–function studies over the past decades has focused on defining structural folds and static active site chemistries. Considerably less energy has been devoted to defining the role of polysterism and conformational change. This allocation of effort is understandable since studying the full conformational landscape of an enzyme before understanding its catalytic mechanism and structural fold would not provide particularly useful information. Another reason for the comparative neglect of polysterism in the literature is that experimentally characterizing scarcely populated states of the ensemble remains difficult and that linking these minor conformations to function is even more challenging [32, 33].
However, several recent developments have lead to the increased study of the role of polysterism in enzyme function. Our understanding of active site chemistry has now developed to the level where functional active sites can be designed de novo [34, 35]. Concurrently, many technical advances in crystalline and solution structural studies have made studying dynamics or polysterism feasible for a majority of proteins . This has meant that simply describing a structure and putative catalytic mechanism is now only the beginning of many structure–function studies. This review examines ‘protein structural dynamics’ by focusing on the detection of protein polysterism and its relevance to the equilibrium steps of protein function and evolution. The dynamic, or time-resolved, aspect and the role of conformational change in the chemical step of catalysis are beyond the scope of the review and have recently been reviewed elsewhere [32, 33]. The challenge over the coming years will involve understanding how conformational substates, and the dynamic transitions between them, function during all steps of the catalytic cycle of an enzyme. We will now discuss six recent examples where this challenge has been answered.
Dihydrofolate reductase; polysterism through a catalytic cycle
The dynamic properties of DHFR have lead to new therapeutic and protein engineering strategies. A small molecule that breaks the coupling of dynamic features acts as an efficient inhibitor of the enzyme . The fusion of a PAS domain to a specific surface site of DHFR allosterically couples dynamic loop regions and a network of internal residues that promote catalysis . This strategy generated a novel allosteric circuit that allowed for light/dark-control of DHFR enzymatic activity . Domain fusions that bias the conformations of flexible active site loops may represent a general strategy for engineering allosteric control of enzyme function. For other enzymes, the challenge is to define polysteric regions and evolutionarily conserved sites that can be used to mechanically couple allostery, conformational flexibility, and enzyme function.
Antibodies; polysterism leads to binding plasticity
The work on antibody specificity by Tawfik and colleagues  has provided clear evidence of functionally relevant polysterism, through the use of crystallography and fast-kinetic analysis of pre-equilibrium states. Crystallographic analysis of catalytic antibodies raised against a transition-state analogue of carboxylesters indicated that the structure of the ligand-free form of the antibody was essentially identical to the structure of the antibody-TS analogue complex. These results led to the conclusion that catalytic antibodies most likely follow a simple lock-and-key mechanism in ligand binding [46, 47]. However, pre-steady-state kinetic experiments revealed that the antibodies exist in a pre-equilibrium between distinct conformational substates and that ligand binding induces an equilibrium shift to the bound state conformation . Moreover, the crystallization conditions were found to bias the conformational distribution of the apo-enzyme towards the bound-like state, explaining why only this configuration was observed by X-ray diffraction.
Acetylcholinesterase; polysterism from simulations and crystallography
Colletier et al.  studied the migration of acetylcholine into the active site using substrate analogues and exploiting substrate inhibition. They observed that in native crystal structures, the cross section between F330 and Y121, what is known as the active site ‘gate’ is only 5 Å; less than the diameter of the quaternary group of the substrate acetylcholine. A series of crystal structures were acquired in which different rotameric states of F330 were observed to drastically change the size of the active site gate. These conformational changes were found both to control access to the active site by fluctuating between open and closed conformations and to provide cationic interactions with the substrate during the chemical step in the catalytic cycle. Different rotameric states of W279, located in the peripheral anionic site, have also been observed as a result of ligand binding, in particular with bi-functional inhibitors that bind at both the peripheral anionic site and the active site . This polysterism and its role in ligand binding has been referred to as “dynamic selectivity” by McCammon and colleagues  who observed its effects through molecular dynamics (MD) simulations. Finally, the active site residue W84 may fluctuate through rotameric states that allow product release through a “backdoor”, which would eliminate competition with incoming substrate for entrance to the active site through the narrow gorge . While structural comparisons have revealed heterogeneity in this region through static [55, 63], and kinetic , crystallography experiments and molecular dynamics simulations , mutagenesis studies have been inconclusive [65, 66].
The polysterism discussed above was observed through analysis of many separate crystal structures; it was not known whether these conformations existed in the absence of ligands. To address this question, Xu et al. [67, 68] performed a series of molecular dynamics simulations and compared the sampled geometries with the many apo-enzyme and ligand bound structures of AChE available in the literature. They found that the apo-enzyme continuously samples almost all of the side-chain conformations observed to result from ligand binding. These results suggest that, in AChE at least, conformational fluctuations provide access to a polysteric landscape of distinct conformational substates that possess unique catalytic properties tailored to specific steps along the catalytic cycle and have different ligand affinities.
Adenylate kinase; subdomain polysterism from a single crystal
Adenylate kinase (AdK) maintains the balance of cellular energy by catalyzing the interconversion of adenine nucleotides. Alignment of the apo and nucleotide-bound crystal structures of AdK revealed a very large conformational transition . The apo-structure resembles an open hand with the finger and thumb lids (Fig. 1b). In structures of the enzyme bound to its nucleotide substrates, the two lids close down around the active site providing a favorable electrostatic environment for phospho-transfer. NMR studies showed that the conformational lid rearrangement, rather than the chemical phospho-transfer step, limits the rate of the catalytic cycle in homologous AdKs . Given the importance of lid flexibility, several studies have examined how motion is induced by substrate.
Surprisingly, convincing evidence that lid rearrangement is pre-existing in the apo-enzyme came from comparisons of independent chains within a single crystal (Fig. 1b) . In crystals obtained by Henzler-Wildman et al., three independent copies of apo-AdK each experience a different crystal-packing environment. The changes in crystal contacts allowed the two lids to rotate about several hinges. The phenomenon of hinge-bending in multiple chains from a single crystal was first observed in a mutant T4 lysozyme . For AdK, the alignment of the independent copies suggested a concerted movement of the lids towards the closed conformations even in the absence of substrate. Adk represents an example of polysterism observed through multiple chains within a single crystal. The conformational heterogeneity observed in the crystal resembles the solution trajectory predicted by computational analysis of crystal structures of the Adk in different nucleotide-bound states.
A comprehensive suite of single-molecule FRET, NMR, and computational studies further confirmed the relevance of crystalline polysterism in AdK. These studies concluded that the enzyme could harvest the directionality of pre-existing motions during a catalytic cycle [71, 73]. This result builds upon the theme that substrate binding does not induce entirely new structures; rather, binding shifts the populations of substates that pre-exist in the ensemble. Although this work defined the apo-state energy landscape, many detailed mechanistic questions remain about how the lid movements are controlled. Particularly, questions remain as to whether lid closing involves local unfolding  and the ratio of populations of the open and closed forms in different nucleotide states . AdK remains an excellent system for examining the relevance of enzyme conformational heterogeneity to the catalytic cycle. Polysterism, from multiple independent chains in the same crystal, was key to unraveling this relationship in AdK. However, molecular dynamics simulations and NMR experiments provide the necessary time-scale resolution to augment the functional importance of the pre-existing conformational substates.
Cyclophilin A; polysterism in electron density at different temperatures
Cyclophilin A (CypA), catalyzes the isomerization of proline residues in proteins, which can regulate protein folding, signaling, and molecular assembly. CypA, like AdK and DHFR, undergoes motions detectable by NMR during catalysis that correspond to the catalytic turnover rate . Furthermore, motions in CypA are pre-existing in the active site, hydrophobic core, and loop regions whether or not substrate is bound . However, unlike AdK or DHFR, comparison of X-ray structures did not suggest a mechanism for the conformational diversity in the active site and core regions detected by NMR. In other systems, large backbone movements of loops or hinge-based subdomain reorientations provide the chemical shift differences detected by relaxation dispersion experiments. Although dynamics of the loop regions of CypA arise because of this conventional motion, the active site and core motions are surprisingly caused by a coupled network of interconverting side-chain substates  (Fig. 1c).
The minor state side-chain conformations of apo-CypA were discovered by using the program Ringer  to sample low electron density levels in a map obtained from data collected at room temperature. Interestingly, conventionally frozen crystals only displayed evidence of the major state. The observation of these substates confirms a link between crystal data collection temperature and substate population . Due to the limited number of enzymes where structures are available at multiple temperatures, it is unclear whether room temperature data collection universally leads to a more accurate description of conformational substates. For example, in alpha-lytic protease, higher resolution cryogenic data collection enabled a more complete description of the conformational landscape than a room temperature structure .
A remote mutation designed to test the importance of the side-chain substates in CypA not only flipped the population of the substates, but also caused a parallel decrease in the rate of conformational interconversions measured by NMR and the rate of catalysis . This and other mutations tested with the HIV capsid as a substrate  suggest that the catalytic cycle of CypA is limited by conformational transitions that are populated in the apo-state. Interestingly, evolution has harvested CypA polysterism as a mechanism for species restriction in HIV. Elegant structural and thermodynamic studies revealed the mechanism that targets a retrotransposed mutant CypA in rhesus macaque monkeys to bind specifically to HIV-2 capsids . Two mutations, located in the loop regions identified as polysteric by NMR relaxation dispersion analysis, dramatically reshape the dominant conformation of the binding pocket. This study further showed a correspondence between loop conformation, binding selectivity, and sensitivity to viral infection. Since wild-type CypA binds the capsid of both HIV-1 and HIV-2, the conformation stabilized by the mutations is likely present in the wild-type ensemble. Thus, CypA represent an example of how polysterism, revealed through comparison of multiple crystal structures, can be used in the evolutionary arms race between host and pathogen by sculpting loop conformations to determine binding specificity. The binding selectivity imparted by loop motion complements the network of residues extending from the core to the active site that undergo side-chain rearrangements that are necessary for turnover .
Bacterial phosphotriesterase; polysterism promotes evolution
The bacterial phosphotriesterase (PTE) has attracted considerable attention as a result of its rapid evolution of extremely high catalytic efficiency for the breakdown of human-made pesticides such as paraoxon. Raushel and coworkers [83, 84, 85] have established that the catalytic cycle is comprised of separate steps involving formation of the Michaelis complex, the chemical step and product release and that for reactive substrates, such as paraoxon, the slowest rate is product release or conformational change. Recently different states along the reaction coordinate have been captured in crystallo, providing new structural insight [11, 86].
A similar story is seen in the role of polysterism in promiscuous, or non-native, PTE activity. Previous work has shown that the slow turnover of the pesticide chlorfenvinphos was principally a result of the active site gorge of the enzyme being too narrow to accommodate the substrate . Mutation of a phenylalanine to an alanine opened the active site gorge and resulted in a 500-fold increase in catalytic efficiency. The question remained: how does the enzyme catalyze hydrolysis of the chlorfenvinphos at all when there is no possible binding mode? The description of the multi-state equilibrium model of the phosphotriesterase provides an explanation : the primary residue that blocks chlorfenvinphos binding (F132) is observed to adopt a minor conformation in which its ring is flipped out of the active site in such a way as to allow productive chlorfenvinphos binding. This therefore provides a straightforward example of the hypothesis put forward by Tokuriki and Tawfik , which posited that conformational diversity can provide important starting points for evolution. This also demonstrates that polysterism can allow promiscuous activities to be catalyzed by minor conformational substates that may not be used for the analogous step in the native catalytic cycle.
Glucocorticoid receptor; polysterism in allosteric signaling
Technical challenges and recent advances
Given the importance of conformational substates in evolution and the catalytic cycles of enzymes, the question remains: what are the best ways to reveal these critical minor states? Clearly the best description of the native ensemble comes from a combination of methods, often including NMR relaxation techniques, and experimental tests of the importance of the observed polysterism. Here, we have reviewed several ways that protein crystallography can provide insight into spatial distribution of conformational substates. For example, structural changes can be revealed through independent crystal structures (DHFR, Antibodies, AChE), comparisons of multiple chains in the same crystal (AdK), or by examining the electron density for minor conformations (CypA, PTE, GR).
Recently developed crystallographic refinement methods aim to describe the ensemble populated in the crystal. These techniques complement the traditional B-factor calculation, which gives an indication of the small-amplitude flexibility around a ground state. Although there is often a concordance between B-factors and fast small-amplitude harmonic motions in solution , several alternative methods have been proposed to explore more dramatic motions that can be difficult to interpret in “messy” electron density. For example, automated rebuilding and refinement techniques can create an “ensemble of independent models” [91, 92]. This family of models is generally a fixed number of independent models that do not simultaneously contribute to the refinement process. These methods have been used to suggest that many models can equally explain the diffraction data . A systematic investigation of an “ensemble of models” created with synthetic data revealed that independently refined models represent the uncertainty inherent in a single-model description of the possible conformational space populated in the crystal .
To more directly model the multiple structures populated in the crystal, many “multi-copy ensemble” refinement methods have been developed . In “multi-copy ensemble” procedures, a fixed number of interdependent models are simultaneously refined . For certain anharmonic polysteric regions, this procedure can lead to a much better description of the electron density. However, given the limited number of experimental observations in crystallographic data, for many residues it may not be necessary to introduce the added parameters needed to refine locally similar regions for the multiple copies of the ensemble.
At high resolution, it may be more advantageous for many residues to be modeled as in a single conformation with anisotropic B-factors rather than as a part of a “multi-copy ensemble”. For other residues, multiple alternative conformations may be needed to fit the electron density. This is traditionally accomplished through manual inspection of electron density maps, which often leads to an under-estimation of the extent of polysterism . An algorithm, qFit, that implements such a strategy in an automated fashion has recently been implemented . qFit generates a “multi-conformer model” that is generally a single model. However, for select residues, alternative backbone and side-chain conformations are modeled into the electron density. This method gives qualitatively similar results to iterative electron density sampling and manual rebuilding and can capture complex coupled motions that extend over many residues .
The automation provided by qFit augments existing strategies for discovering and modeling alternative conformations. For example, electron density sampling as implemented in Ringer , reveals that many alternative conformations exist at electron density levels that have been traditionally considered noise by human model builders. Ringer samples electron density in dihedral space and discovers peaks indicative of unmodeled alternative conformations. These alternative conformations can be built after manual inspection of the electron density map and further refined for occupancy by phenix.refine . As seen in PTE and CypA, alternative conformations, revealed at low levels of electron density (generally 0.3–1σ), can be critical for enzyme function.
Another means by which crystallography can access protein polysterism is the analysis of thermal diffuse scattering [95, 96]. Diffuse scattering is the remaining X-ray intensity that is recorded in diffraction images once the Bragg reflections are subtracted. While analysis of the Bragg reflections can give accurate estimates of the individual atomic mean squared displacements and alternate conformations, the diffuse scattering can provide valuable insight into isotropic liquid like displacements  and strongly correlated rigid body movements of groups of atoms . This can reveal how particular groups of atoms are displaced together as rigid bodies. Techniques that predict low frequency correlated rigid body movements, such as principal component analysis of molecular dynamics simulations or normal mode analysis of elastic network models, can be validated by analysis of diffuse scatter [99, 100]. This theoretical–experimental cross validation approach can also serve to support or exclude other models of correlated motion used in refinement, such as translation-libation-screw (TLS) refinement [101, 102]. Whether analyzing diffuse or Bragg scatter, access to primary image data would greatly increase the ability to re-examine biological conclusions. Although deposition of scaled reflection data, which permits re-calculation of electron density maps for analysis by Ringer and other software, is now standard for almost all journals, image data are rarely deposited . A major limitation to deposition of image data is storage space, but several institutions, structural genomics consortia, and synchrotrons have started voluntary image deposition. As interpretation of atomic displacements becomes more important to our understanding of protein function, our reliance on the re-analysis of primary data is likely to increase in coming years.
Polysterism has contributed to our understanding of how new functions evolve from existing functions. Many studies have now demonstrated that an ensemble of interconverting structures, rather than a single structure, exerts the function of the protein. Thus, evolutionary selection does not optimize only an active site or the stability of a single structure. Rather, it shapes an entire conformational landscape comprising many sub-structures, their relative populations, and their rates of interconversion. Because mutations may affect the dominant structure modeled in a typical crystal structure or “hidden” minor conformations, polysterism must be carefully considered when considering the role of remote mutations on both catalysis and allostery . If we can accurately describe the conformational landscape of proteins, we might be able to better predict the nature and conformations of transitions between the dominant structures as a protein is perturbed. Realizing this dream has implications for understanding protein evolution and designing better enzymes and inhibitors .
JSF was funded by an EMBO Short-term Fellowship while this article was conceived. CJJ is supported by a Marie Curie International Incoming Fellowship. We acknowledge Dan Tawfik for providing a stimulating environment. We also thank Tom Alber, David Wemmer, and Martin Weik for discussions. All figures in this review were generated with PyMol (http://pymol.org); we dedicate this article to the legacy of PyMol’s creator Warren DeLano.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.