Sensitivity to temperature helps determine the success of organisms in all habitats, and is caused by the susceptibility of biochemical processes, including enzyme function, to temperature change. A series of studies using two structurally and catalytically related enzymes, A4-lactate dehydrogenase (A4-LDH) and cytosolic malate dehydrogenase (cMDH) have been especially valuable in determining the functional attributes of enzymes most sensitive to temperature, and identifying amino acid substitutions that lead to changes in those attributes. The results of these efforts indicate that ligand binding affinity and catalytic rate are key targets during temperature adaptation: ligand affinity decreases during cold adaptation to allow more rapid catalysis. Structural changes causing these functional shifts often comprise only a single amino acid substitution in an enzyme subunit containing approximately 330 residues; they occur on the surface of the protein in or near regions of the enzyme that move during catalysis, but not in the active site; and they decrease stability in cold-adapted orthologs by altering intra-molecular hydrogen bonding patterns or interactions with the solvent. Despite these structure–function insights, we currently are unable to predict a priori how a particular substitution alters enzyme function in relation to temperature. A predictive ability of this nature might allow a proteome-wide survey of adaptation to temperature and reveal what fraction of the proteome may need to adapt to temperature changes of the order predicted by global warming models. Approaches employing algorithms that calculate changes in protein stability in response to a mutation have the potential to help predict temperature adaptation in enzymes; however, using examples of temperature-adaptive mutations in A4-LDH and cMDH, we find that the algorithms we tested currently lack the sensitivity to detect the small changes in flexibility that are central to enzyme adaptation to temperature.
in which the rate of a reaction, k, increases exponentially with temperature, T (where R is the universal gas constant, A is a reaction-specific pre-exponential constant and Ea represents the Arrhenius activation energy of the reaction). Depending on the value of Ea, rates of most metabolic processes will increase on the order of 2- to 3-fold with a 10°C increase in environmental temperature, giving rise to the familiar ‘Q10’ relationship of thermal physiology. And indeed, when an enzyme is assayed in vitro across a range of temperatures encompassing the normal physiological temperatures of the organism, it usually will show the expected exponential increase in reaction rate, at least until a ‘break point’ is reached and activity begins to decline due to loss of the protein's native structure. However, when metabolic rates of species adapted to different temperatures are compared, the Arrhenius relationship does not hold; that is, a cold-adapted polar fish living at 0°C does not have a metabolic rate 20-times lower than that of a desert lizard living at 40°C. Thus, there must be a compensatory mechanism available, acting on an evolutionary time scale, that allows natural selection to alter rates of metabolic reactions as organisms adapt to new environments. Because most metabolic reactions are universally shared, physiologists recognized that it was likely the catalysts themselves – that is, the enzymes – that would be altered to allow appropriate metabolic rates at different physiological temperatures.
It was not until the 1960s, however, that experimental evidence began to accrue showing that enzyme orthologs [that is, enzymes encoded by a common (homologous) gene from different species] indeed did display modified function correlating with the physiological temperatures the species experienced. These early data consisted mainly of temperatures of maximal enzyme activity and thermal denaturation temperatures (Tm) (e.g. Licht, 1964; Vroman and Brown, 1963; Read, 1964). However, work later in the decade began to show that more functionally relevant kinetic properties, most notably ligand affinity as measured by apparent Michaelis–Menten constants (Km), also were modified through selection to environmental temperatures (e.g. Hochachka and Somero, 1968).
In the time since these first studies correlating enzyme function and adaptation temperature, research on temperature adaptation in enzymes has accelerated and has begun to incorporate a fuller understanding of the protein structural changes that underlie shifts in enzyme function. In this review, we describe studies spanning the past four decades confirming that small differences in habitat temperature can lead to selection for physiologically significant changes in enzyme function. We also describe studies that have pinpointed the location and amount of amino acid change necessary to lead to these functional changes. Substitutions at multiple sites in a sequence are shown to equivalently enable adaptive change: there is more than one way to skin the proverbial cat.
A4-lactate dehydrogenase (skeletal isoform), tetrameric
cytosolic malate dehydrogenase
enzyme turnover number
Michaelis–Menten constant for the cofactor NADH
Michaelis–Menten constant for the substrate pyruvate
lactate dehydrogenase-A (skeletal isoform), monomer
nicotinamide adenine dinucleotide, reduced
protein stability prediction
α-hydroxy acid dehydrogenase
Gibbs free energy of protein folding
difference in ΔGf values between two proteins
However, despite these advances in our understanding of structure–function relationships, we currently remain unable to predict how a particular amino acid change will impact temperature sensitivity of enzyme function. Such an a priori predictive capacity would be extremely valuable for conducting proteome-wide surveys of temperature adaptation. These analyses might reveal how widespread adaptation is among diverse categories of proteins, and would be helpful in predicting how much adaptation to temperature by proteins will be required to allow ectotherms to cope with global climate change. We thus conclude this review with a brief description of current bioinformatics approaches to predict stability changes in enzymes in response to specified amino acid mutations. Some of these algorithms have explained adaptive variation in protein stability among species differing very widely in adaptation temperature (see Gu and Hilser, 2009). However, our results suggest that bioinformatics approaches may not yet be sensitive enough to consistently determine how relatively non-perturbing amino acid changes, such as those involved in adaptation to small differences in habitat temperature, might alter ligand binding and catalytic rate through modifications of flexibility in localized areas of protein molecules.
Protein stability, temperature and catalytic function
To understand the necessity of biochemical adaptation in ectotherms to the thermal habitat each occupies, it is important to realize that a single enzyme cannot efficiently catalyze reactions across the entire range of temperatures found in the biosphere (i.e. −20 to +113°C or more; D'Amico et al., 2003), even if the enzyme should possess adequate stability to allow its native structure to persist across this broad range of thermal conditions. This is because catalysis often requires some amount of motion or rearrangement of protein structure during binding of ligands, formation of the catalytic vacuole (the ‘pocket’ in the enzyme where the actual covalent chemistry occurs) and release of products (Fields and Somero, 1998; Fields, 2001). Although there are many hundreds of stabilizing non-covalent interactions (such as hydrophobic interactions, hydrogen bonds, salt bridges and van der Waals interactions) in a typical enzyme (Jaenicke, 2000), these stabilizing forces are counterbalanced by entropic factors, which favor the unfolding of the secondary and tertiary structures of the enzyme. It is apparent that most enzyme molecules are under selection to maintain a balance between stabilization and flexibility – stability so that the molecule does not unfold and potentially aggregate under normal physiological conditions, and flexibility to allow those motions necessary for catalysis at a rate appropriate to maintain metabolic flux (Fields, 2001; Somero, 2010).
Temperature is a significant complicating factor in the maintenance of this enzyme stability–flexibility balance (Feller, 2010). Rising temperatures tend to weaken the stabilizing interactions responsible for maintaining the native folded state of the enzyme. Consequently, as temperature increases, enzyme reaction rates first tend to accelerate, as stabilizing bonds break and reform more rapidly, and the conformational changes in the protein necessary for catalysis occur at a faster pace. However, as temperature continues to rise, enough stabilizing interactions are disrupted that the enzyme no longer can maintain a conformation with the three-dimensional geometry required to bind ligands (this is observed as a rapid increase in Km), and, ultimately, as secondary and tertiary structure disintegrates, enzyme function is lost. In contrast, if temperature decreases from the range to which an enzyme is adapted, the reduced thermal kinetic energy in the medium leads to over-stabilization of the catalytically mobile regions of the protein, as weak interactions tend to strengthen. The enzyme maintains a binding-competent conformation for a greater proportion of time (seen as a decrease in Km), but because the motions necessary for catalysis are inhibited, the rate of catalysis (kcat) also decreases, ultimately to the point that flux through the pathway can no longer support cellular metabolism. We note that, whereas hydrophobic interactions may weaken in the cold, such destabilization seems insufficient to override the increases in rigidity that arise from cold stabilization of other classes of non-covalent bonds.
Based on the above model of temperature effects on enzyme function, which was developed in part through the studies described below, it is apparent that any one enzyme ortholog will not be able to efficiently catalyze its reaction across the temperature range in which life is found. Furthermore, this model predicts that enzyme adaptation to temperature over evolutionary time should involve amino acid changes that lead to stabilization (in the case of adaptation to higher temperatures) or destabilization (adaptation to colder temperatures) in areas of the molecule whose motions are necessary for catalysis.
A4-lactate dehydrogenase and cytosolic malate dehydrogenase as models for enzyme adaptation to temperature
In the past few decades, physiologists and biochemists have used a wide variety of enzymes from a diversity of taxa to explore the mechanisms of biochemical adaptation to environmental temperature. Proteins from bacteria and Archaea, for example, often have served to elucidate broad patterns in protein adaptation across wide temperature ranges, and have allowed researchers to define general criteria for structural adaptation to temperature (D'Amico et al., 2003; Siddiqui and Cavicchioli, 2006; Feller, 2008, 2010; Gu and Hilser, 2009). In contrast, studies of enzyme orthologs from closely related species of ectothermic metazoans, such as phosphoglucose isomerase in butterflies (Watt et al., 2003), pyruvate kinase in fish (Low and Somero, 1976) and isocitrate dehydrogenase in mussels (Lockwood and Somero, 2012), have the potential to reveal the minimum amount of environmental temperature change necessary to induce modifications in enzyme function and structure, and what types and magnitudes of amino acid substitutions are sufficient to allow adaptation to a new thermal environment. Of these latter types of studies focusing on orthologs from closely related species, among the most fruitful have been a series of studies using A4-lactate dehydrogenase (A4-LDH) and cytosolic malate dehydrogenase (cMDH). Both of these enzymes are α-hydroxy acid dehydrogenases (α-HADHs), which catalyze the reversible reduction or oxidation of the α-carbon of a small, metabolically essential carboxylic acid (Birktoft et al., 1982). A4-LDH, the LDH isoform found in skeletal muscle of vertebrates, is responsible for reducing pyruvate to lactate during bursts of anaerobiosis, with the cofactor NADH (nicotinamide adenine dinucleotide, reduced) providing reducing potential. In comparison, cMDH reduces oxaloacetate to malate in the cytosol (i.e. it is not the MDH isoform involved in the mitochondrial Krebs cycle) in a reaction similar to that of LDH; in fact, the main difference in the reactants is an additional carboxylic acid group in oxaloacetate that is absent in pyruvate (Dunn et al., 1991). Despite the similarity in catalytic chemistry between the reactions mediated by these two enzymes, the proteins themselves are quite dissimilar, at least on the primary structural (amino acid sequence) level. An amino acid alignment indicates only 18% of the amino acid residues are identical between the porcine forms of the two enzymes (Fig. 1). Interestingly, however, despite the lack of congruence in primary structure, the two enzymes share significant overlap in the location of secondary structural features, i.e. β-sheets and α-helices (Fig. 1). Furthermore, the three-dimensional structures of pig A4-LDH and cMDH can be superimposed tightly, including the position of each enzyme's ligands (Fig. 1), indicating that the process of substrate reduction or oxidation is catalyzed through the same mechanism, and with similar conformational rearrangements, in each enzyme.
The similarity of catalytic mechanism and higher order structures between A4-LDH and cMDH, contrasted with the significant divergence in amino acid sequence, provides a fascinating opportunity to examine, in two independently evolving proteins, how natural selection modifies enzyme structure to balance structural stability and flexibility and allow catalysis at appropriate rates in differing thermal regimes. In the studies focusing on A4-LDH and cMDH described below, the comparison between these α-HADHs has allowed researchers to address the following general questions regarding enzyme adaptation to relatively slight but physiologically significant changes in temperature: how much change in amino acid sequence is needed to produce a physiologically significant change in function with respect to temperature?; where do amino acid substitutions occur in the three-dimensional structure of the molecule?; do independent instances of temperature adaptation, in both A4-LDH and cMDH, repeatedly exhibit the same types of structural modifications?; and finally, are we able to predict where, how many and what types of amino acid changes are needed to adapt an enzyme to a change in habitat temperature?
Temperature adaptation in teleost A4-LDHs
LDHs, and A4-LDH in particular, have a number of attributes making them attractive models for the study of structure–function relationships in enzymes adapting to new thermal habitats. The enzyme is ubiquitous in vertebrates, and relatively easy to purify from skeletal muscle; its catalytic chemistry is straightforward and, as a result, in vitro assays are uncomplicated; and LDH has been studied extensively by physical biochemists interested in describing the detailed relationships between enzyme structure and catalysis (e.g. Abad-Zapatero et al., 1987; Dunn et al., 1991; Gerstein and Chothia, 1991). As a result, highly resolved crystal structures of vertebrate A4-LDHs are available to allow correlation between types of amino acid substitutions, their locations and their impacts on the catalytic process.
Work in the late 1960s by Hochachka and Somero (1968), which examined orthologs from varied teleost species, provided the first evidence that A4-LDH function could be modified evolutionarily in response to habitat temperature, and furthermore suggested that ligand binding affinity (as quantified by binding affinity for the substrate pyruvate, Km,pyr) was an important and perhaps universal target of temperature adaptation. A decade later, a foundational paper by Yancey and Somero confirmed Km,pyr as a key target of selection, and described the appropriate experimental methods to accurately assess binding affinity in an enzyme across a range of temperatures, using an alpha-stat approach to buffer pH (Yancey and Somero, 1978). Most importantly, the authors examined A4-LDH function in a broad variety of fish from thermal habitats as diverse as the Antarctic Ocean and the Amazon River, showing that Km,pyr remained in a relatively narrow range, from ∼0.15 to 0.35 mmol l−1, when each isoform was assayed within its physiological temperature range. These results demonstrated that maintenance of ligand binding affinity in the face of temperature change is under strong selection, and that orthologs of the A4-LDH enzyme can accommodate structural changes allowing functional compensation, at least across a range of temperatures from −2 to 40°C. However, what, where and how extensive these structural changes might be remained to be elucidated, as did the amount of environmental temperature change necessary to induce selection to modify enzyme function. Subsequent studies on three separate groups of teleost fishes were able to address these questions.
Work by Graves and Somero (1982) was the first to confirm that a relatively small change in habitat (=adaptation) temperature was sufficient to alter enzyme function in A4-LDH orthologs from closely related species. Using members of the genus Sphyraena (barracuda) from the eastern Pacific Ocean, the authors showed that Km,pyr varied in a temperature-adaptive manner; that is, values shifted upward in the more cold-adapted species such that when measured at physiological temperatures, Km,pyr values were comparable (Graves and Somero, 1982). Because the A4-LDH orthologs were from congeners living in habitats whose mid-range temperatures only varied by 3–8°C, the authors inferred that structural differences between the enzymes must be small. However, at the time the study was performed, determining protein sequence was a daunting task, limiting the ability of the authors to associate amino acid changes with the functional changes they had measured.
A subsequent study, however, extended the work of Graves and Somero on barracuda A4-LDHs, and for the first time was able to associate a specific amino acid substitution with a measurable, significant difference in a temperature-associated functional attribute, Km,pyr (Holland et al., 1997). In this study, A4-LDHs of five Sphyraena congeners from differing thermal habitats were examined, but two orthologs produced especially noteworthy results. The A4-LDHs of Sphyraena lucasana and Sphyraena idiastes showed a temperature-adaptive divergence in Km,pyr (i.e. higher Km,pyr values from the ortholog of S. idiastes, which occupies the colder habitat) (Fig. 2A), and only a single amino acid difference, N8D, was found between the orthologs. In a result that would be repeated in all subsequent studies, the authors found that the substituted residue was not a component of the ligand-binding pocket (active site) of the enzyme. At the time, full interpretation of the effects of this substitution on enzyme function could not be made because of limited information on A4-LDH's structural alterations during binding and catalysis.
Soon after the study by Holland and colleagues, Fields and Somero, using notothenioid fishes, further described how A4-LDH temperature sensitivity could be modified by specific amino acid substitutions (Fields and Somero, 1998). The notothenioids are a suborder of teleosts (order Perciformes), most of which have adapted to the cold and thermally stable waters surrounding Antarctica. The Km,pyr values of Antarctic notothenioid orthologs follow the expected pattern of temperature adaptation, showing higher values than temperate teleost A4-LDHs at any measurement temperature (Fig. 2B), which allows these orthologs to maintain appropriate affinity at the very low temperatures the notothenioids experience. Additionally, the authors showed that turnover numbers (kcat) for notothenioid A4-LDHs were higher than those measured for orthologs from more warm-adapted vertebrates, indicating that the enzymes had evolved increased flexibility to maintain metabolic flux at colder temperatures (Fields and Somero, 1998).
Non-conservative amino acid substitutions occurred at eight positions in the Antarctic notothenioid LDH-A consensus sequence relative to a consensus of temperate teleost LDH-A, and subsequent work using expressed recombinant enzymes and site-directed mutagenesis was able to show that only two of these substitutions, E233M and Q317V, were responsible for the cold-adapted functional attributes of notothenioid A4-LDHs (Fields and Houseman, 2004). Furthermore, through molecular modeling techniques, the notothenioid consensus sequence could be threaded onto a high-resolution three-dimensional LDH-A structure from dogfish (Protein Data Bank file PDB 6LDH; Abad-Zapatero et al., 1987); this structural analysis revealed that these two substitutions were associated with secondary structural elements that move during the catalytic cycle (Gerstein and Chothia, 1991). Specifically, two relatively rigid α-helices that bracket the active site, αG and αH (Fig. 3), must move laterally to accommodate entry of cofactor and substrate and exit of product; these movements are rate limiting for the LDH reaction (Gerstein and Chothia, 1991). From these observations, the authors posited that temperature-adaptive substitutions were likely to target regions of the molecule associated with structures surrounding the active site, which in A4-LDH includes helices αG and αH as well as an additional ‘major mover’, the catalytic loop (Fig. 3), and whose motion was necessary for catalysis (Fields and Somero, 1998; Fields and Houseman, 2004). Thus, by increasing the flexibility of these regions the kcat of the enzyme could be increased at low temperatures, but at the cost of decreased ligand affinity.
These conclusions were strengthened and extended in a final study on teleost A4-LDHs, focusing on a congeneric group of damselfish (genus Chromis) (Johns and Somero, 2004). The authors examined orthologs from two tropically distributed species, Chromis xanthochira and Chromis caudalis, and one temperate species adapted to significantly cooler temperatures, Chromis punctipinnis. Again, the expected temperature-adaptive differences in substrate affinity were found (Fig. 2C), with the cold-adapted C. punctipinnis A4-LDH having higher Km,pyr values than the more warm-adapted orthologs; kcat was higher at each measurement temperature for the C. punctipinnis ortholog as well. Johns and Somero found three non-conservative substitutions between the tropical (C. xanthochira) and temperate orthologs, and through site-directed mutagenesis were able to show that only one of these, T219A, was necessary and sufficient to alter pyruvate affinity from the cold-adapted to the warm-adapted state. Interestingly, when the authors used molecular modeling techniques to describe the location of this single substitution within the three-dimensional structure of the LDH-A monomer, they found that it was associated with a loop region adjacent to helix αG, one of the two structures found to be important in temperature adaptation of A4-LDH function in notothenioids (Fig. 3) (Johns and Somero, 2004). The combined results of the notothenioid and damselfish A4-LDH studies suggested that there might be a general pattern to temperature-adaptive substitutions, at least in this particular enzyme: substitutions would be solvent exposed, they would occur in regions of the molecule whose mobility was necessary to allow catalysis to occur, and, in the direction of cold-to-warm adaptation, substitutions would replace charged or polar residues with non-polar residues (i.e. Q317V and E233M in notothenioid A4-LDHs; T219A in the Chromis ortholog).
Some, but not all of these conclusions are supported in a retrospective examination of the location of the amino acid substitution found between A4-LDH orthologs of S. idiastes and S. lucasana described above (Holland et al., 1997). Position 8 is close to the N-terminus, on a structure that seems to extend away from the molecule and shows no intra-molecular interactions when the LDH-A monomer is examined (Fig. 3A). However, LDH functions as a tetramer, and when the N8D substitution is visualized in the context of the multimeric enzyme, it becomes clear that the N-terminus nestles between two neighboring monomers. In fact, residue 8 is positioned in the homotetramer such that it can hydrogen bond with residues on a β-sheet from a second monomer (β-sheet III; Birktoft et al., 1989). β-sheet III anchors an extended strand leading to helix αH, one of the two ‘major-mover’ regions shown to be a target of temperature adaptation in the study of notothenioid A4-LDHs (Gerstein and Chothia, 1991) (Fig. 3B). Thus, the results of a structural analysis of the barracuda substitution support the hypotheses that structural changes lead to alterations in the mobility of key regions of the enzyme, and that these substitutions occur on the surface rather than in the core of the enzyme. However, findings from the Sphyraena A4-LDH study fail to support the hypothesis that temperature adaptation, in the cold→warm direction, necessarily involves polar/charged→non-polar substitutions.
Temperature adaptation in mollusk cMDHs
As research on temperature adaptation in teleost A4-LDHs progressed, equally informative work was being conducted on cMDH orthologs in mollusks. Starting in the early 1990s, evidence for functional adaptation of the enzyme, in the form of changes to Km for the cofactor NADH (Km,NADH), was collected using orthologs from the gastropod mollusk genus Haliotis (abalone) (Dahlhoff and Somero, 1993). The results of this research were comparable to the early work on barracuda A4-LDHs (Graves and Somero, 1982): cMDH Km,NADH values of Haliotis species from higher latitude (colder) habitats were higher at any assay temperature, but showed compensation such that Km,NADH values were similar when compared within each species' physiological temperature range (Fig. 4A) (Dahlhoff and Somero, 1993). At the time of the study, however, it was not possible to determine the amino acid sequences of the five cMDH orthologs examined, and so the structural changes to the proteins that underlay the adaptive functional differences were not determined. We still do not know the amino acid substitutions responsible for temperature adaptation in Haliotis cMDHs shown in this early study; this would be a valuable area for future research. Nevertheless, the work by Dahlhoff and Somero confirmed that enzyme adaptation to temperature extended beyond the A4-LDHs already examined in teleosts, that similar patterns of functional change in enzymes in response to different thermal habitats were present in highly divergent taxa, and that changes in ligand binding affinity continued to be a hallmark of biochemical adaptation.
Over a decade later, temperature adaptation in cMDH was revisited, using a bivalve genus, Mytilus (Fields et al., 2006). Mytilus trossulus is the blue mussel species native to the eastern North Pacific Ocean, but over the last century it has begun to be displaced in the southern portion of its range by Mytilus galloprovincialis, an invasive species from the Mediterranean Sea (Geller, 1999). An abundance of evidence, including evidence on the biochemical and physiological levels, indicates that M. galloprovincialis is more heat tolerant than M. trossulus (Braby and Somero, 2006; Tomanek and Zuzow, 2010; Lockwood and Somero, 2011; Fields et al., 2012), helping to explain the former species' invasive success in the warm southern section of the latter's range. Fields and colleagues compared cMDHs from mantle tissue of each congener across a range of temperatures, and found that the M. trossulus ortholog had higher Km,NADH values, especially at the highest assay temperatures (Fig. 4B) (Fields et al., 2006). Furthermore, the authors deduced the amino acid sequence for each ortholog, and found only a single non-conservative amino acid difference between the two, V114N (M. trossulus→M. galloprovincialis). Through site-directed mutagenesis, they were able to show that this single amino acid substitution was sufficient to shift the Km,NADH–temperature relationship of the cold-adapted M. trossulus ortholog to that of the warm-adapted M. galloprovincialis (Fig. 4B). Notably, unlike some of the substitutions found in A4-LDH that shifted from polar/charged→non-polar in the cold→warm direction, V114N in Mytilus cMDH represents a non-polar→polar change.
When the location of the mutation at residue 114 was visualized within the three-dimensional structure of cMDH via molecular modeling, it was found to associate with neither helix αG nor αH, as had been described for temperature-sensitive mutations in A4-LDH; instead, residue 114 is located at the margin of the region that moves most during catalysis, the catalytic loop (Fig. 5) (Fields et al., 2006). Thus, in some aspects, the results of the functional and structural analysis of Mytilus cMDH confirmed patterns of temperature adaptation first seen in teleost A4-LDHs – a single amino acid substitution on the surface of the protein is sufficient to adapt an ortholog to a new temperature regime, and the substitution is at the edge or ‘hinge’ of a secondary structural element that must move significantly during the catalytic cycle. However, the association of the Mytilus cMDH substitution with the catalytic loop rather than either of the two helices, αG or αH, as had been found in A4-LDHs, indicated that there were likely to be many sites within the enzyme molecule where substitutions could lead to adaptively important changes in local mobility.
The most recent research examining temperature adaptation in molluscan cMDHs continues to demonstrate the ubiquity of enzyme adaptation to even small changes in environmental temperature, at least in α-HADHs, while expanding the number of regions within the protein that may be targets of temperature-adaptive selection. Dong and Somero examined cMDH from foot muscle of six congeners of the limpet Lottia (class Gastropoda), which occupy different latitudinal and vertical intertidal ranges along the west coast of North America (Dong and Somero, 2009). Of particular interest was one species pair, Lottia digitalis and Lottia austrodigitalis, which are morphologically very similar [until the 1970s, they were considered a single species (Murphy, 1978)], but which have relatively discrete biogeographic ranges. Lottia digitalis, the northern form, is found from the Aleutian Islands south to Point Conception, while L. austrodigitalis extends from the Monterey Peninsula to Baja California (Murphy, 1978; Morris et al., 1980). When comparing cMDH Km,NADH values from foot muscle of the two species, the authors found the expected signature of temperature adaptation – lower ligand affinity (higher Km values) at each assay temperature for the cold-adapted enzyme relative to the warm-adapted one (Fig. 4C). Not surprisingly, given the close relatedness of the two species, this functional difference was the result of only a single amino acid difference between the two orthologs, G291S (L. digitalis→L. austrodigitalis) (Dong and Somero, 2009).
When the location of this substitution was visualized via molecular modeling (Fig. 5), it was found to be far from any of the major-mover structures (helices αG or αH or the catalytic loop) that had appeared to be targets for temperature adaptation in earlier studies, whether of A4-LDH or cMDH. However, a close analysis of the effects of the replacement of a glycyl with a seryl residue at this location provides a mechanism for the altered cofactor affinity that the authors found. Position 291 sits within β-sheet III, adjacent to a β-turn. In this area, the seryl residue allows the formation of five hydrogen bonds, which help stabilize the β-sheet. In contrast, when a glycyl residue is inserted at position 291, as found in the more cold-adapted L. digitalis cMDH, only two hydrogen bonds are formed, leading to destabilization of the structure (Dong and Somero, 2009). As noted above in the context of Sphyraena A4-LDH, this β-sheet is attached via a short extended strand to the catalytically mobile region helix αH (Fig. 5); the replacement of a seryl by a glycyl residue thus has the potential to increase the conformational freedom of αH, reducing ligand affinity but increasing catalytic rate. Thus, despite its physical distance from the major mover structures within Lottia cMDHs, residue 291 may still be appropriately positioned to affect the thermodynamics of catalytic motions in a temperature-adaptive manner. Remarkably, despite the significantly different locations of A4-LDH residue 8 and cMDH residue 291 in each monomer, and despite the very different amino acid compositions between the two α-HADHs in those regions, the similarity of the locations of these two substitutions in the multimeric enzymes strengthens the conclusion that the small β-sheet N-terminal to helix αH can be a target of temperature adaptation.
Comparing these studies of temperature adaptation in molluscan cMDH orthologs with research focusing on teleost A4-LDHs allows us to draw a number of conclusions. First, a small change in habitat temperature, of the order of a few degrees, repeatedly has been sufficient to elicit selection for functional changes in the enzymes. Second, ligand affinity appears to be a universal target of temperature adaptation in the α-HADHs, with more cold-adapted enzymes exhibiting lower ligand affinities (higher Km values) at a common temperature of measurement, but similar affinities to those of more warm-adapted orthologs when comparisons are made at physiologically relevant temperatures. We interpret the lower affinity of a cold-adapted ortholog seen at a common temperature of measurement to be a consequence of an increase in conformational flexibility; this higher flexibility leads to a smaller fraction of the population of the cold-adapted enzyme molecules possessing a binding-competent conformation. However, this higher flexibility allows faster catalysis at lower temperatures. Third, the structural changes necessary to elicit these functional differences are small, usually a single amino acid. And fourth, amino acid substitutions always occur on the surface of the enzyme, associated directly or indirectly (via a network of hydrogen bonds) with structures that must move during catalysis. Nevertheless, despite elucidating these general rules of enzyme temperature adaptation, at least in α-HADHs, the studies described above on A4-LDHs and cMDHs have not provided us with the ability to predict, a priori, whether an amino acid substitution at a particular site will lead to a level of change in conformational flexibility appropriate to a specific level of environmental temperature change.
Can we predict temperature-sensitive changes in enzyme function from sequence alone?
A great deal of effort is necessary to accomplish the type of research described above associating functional changes in enzymes with their underlying structural modifications. Typically, the workflow involves measurement of enzyme kinetics across a range of temperatures from purified enzyme or tissue homogenates, sequencing of cDNA of each ortholog to identify differences in amino acid sequence, site-directed mutagenesis and expression of recombinant protein, and further kinetics assays to confirm that candidate amino acid substitutions indeed are responsible for the differences in function that have been ascribed to them. The process can be time consuming, technically challenging and expensive. Thus, a valuable addition to research on enzyme adaptation to temperature would be an in silico approach that would allow us to predict whether amino acid substitutions between enzyme orthologs of two species adapted to different temperatures lead to the expected direction and magnitude of change in a functionally important attribute like ligand binding affinity. Such a predictive capacity might allow in silico screening for adaptive change across the proteome.
In the past 10 years, a number of algorithms have been developed that may have the potential to allow these types of predictions. These algorithms are designed to assess how specified amino acid substitutions will affect global protein stability, specifically by estimating the Gibbs free energy change as the protein transitions from the unfolded to the folded state (ΔGf) – a process that should be energetically favorable (because the folded state is more stable than the unfolded state under physiological conditions), leading to a negative ΔGf value. These calculations are based on a set of rules determining thermodynamic changes expected as a specified amino acid sequence folds, and can use sequence data alone or incorporate information on the three-dimensional structure of the protein. When a protein with a particular amino acid at position X is compared with the same protein with a new amino acid at that position, two ΔGf values can be estimated, one for the ‘wild-type’ and one for the ‘mutant’ protein, and the difference, ΔΔG, will indicate whether the amino acid substitution is predicted to stabilize or destabilize the folded protein. When the mutant is stabilized relative to the wild-type, the mutant ΔGf will have a negative value of greater magnitude than the ΔGf of the wild-type, resulting in a positive ΔΔG value.
The studies described above focusing on temperature adaptation in A4-LDHs and cMDHs provide an excellent opportunity to test the usefulness of these protein stability prediction (PSP) algorithms in assessing enzyme adaptation to small differences in habitat temperature. Thus, we can compare six independent examples of one-amino-acid changes associated with small but physiologically significant changes in enzyme function (ligand affinity): in A4-LDH the substitutions (in the cold→warm direction) are N8D (Sphyraena), E233M and Q317V (notothenioids) and T219A (Chromis); in cMDH the substitutions are V114N (Mytilus) and G291S (Lottia).
To test the utility of PSP algorithms for prediction of enzyme adaptation to small changes in environmental temperature, we used the six substitutions listed above in four of the many algorithms available, which use a variety of approaches to predict ΔΔG. Briefly, algorithms may compute potential energy functions of the proteins to calculate ΔΔG (for example, physical potentials or statistically derived potentials), or may utilize machine-learning methods, in which they are trained using large datasets of proteins for which mutation–stability data are already available, and then apply the patterns of stabilization ‘learned’ from these test proteins to novel mutations. Algorithms can also be divided into those that use amino acid sequence directly to compute ΔΔG (i.e. not incorporating any information about the three-dimensional structure of the protein), versus those that utilize three-dimensional protein models (PDB files) that can be generated automatically from amino acid sequences (Fig. 6). There are a number of reviews of PSP algorithms and their relative performance that provide greater detail on the variety of algorithms available, and the methods they use to calculate ΔΔG (e.g. Thusberg and Vihinen, 2009; Khan and Vihinen, 2010).
Our goal was to determine whether researchers who were not expert in the biophysics and thermodynamics of protein folding could use available PSP algorithms to determine whether a particular mutation of interest would be temperature adaptive in a given environmental setting. From the algorithms available, we chose the four used here because they were web based (no software needed to be downloaded), and provided both a direction and magnitude of ΔΔG, expressed in kcal mol−1 (where 1 kcal=4184 J). The algorithms are I-Mutant 3.0 (sequence based) and I-Mutant 3.0 (structure based), which are support-vector-based (i.e. machine learning) algorithms (Capriotti et al., 2005, 2008; http://gpcr2.biocomp.unibo.it/cgi/predictors/I-Mutant3.0/), Eris (structure based), which uses physical potentials combined with modeling of changes in backbone and side-chain conformations (Yin et al., 2007; http://troll.med.unc.edu/eris/) and PoPMuSiC (structure based), which employs a neural network trained with a database of protein mutants for which stability changes are known (Dehouck et al., 2009; http://dezyme.com/). For those algorithms that were able to utilize structural information, we created molecular models of each of our six protein/mutant pairs using the automated Phyre server (Kelley and Sternberg, 2009) under default settings; Phyre accepts amino acid sequences and automatically searches for appropriate templates from the Protein Data Bank (http://rcsb.org) upon which to build a three-dimensional model; the new model is returned to the user as a PDB file.
The results of our analysis suggest that for the six mutations in α-HADHs for which we have temperature-adaptive functional data, the PSP algorithms do not uniformly predict stability changes of the expected direction or magnitude. Fig. 7 shows the ΔΔG values predicted by the four algorithms for each of the six mutations. In each case, the mutation is from the cold to the warm ortholog, which we expected would lead to stabilization and thus a positive ΔΔG. In most cases, however, ΔΔG was predicted to be negative (i.e. destabilizing), or very close to zero. Furthermore, in all cases the magnitude of change was small. To provide context, a single hydrogen bond is predicted to provide 3–9 kcal mol−1 of stabilization to a protein (Fersht, 1985), depending on the types of residues and the atoms involved. Thus, all of the predicted stability changes fall within or below the stabilization or destabilization energy changes expected from the loss or gain of a single hydrogen bond (Fig. 7).
Based on the variable results obtained from these PSP algorithms, and the relatively small ΔΔG values calculated, we conclude that the algorithms may not be appropriate tools for the analysis of amino acid substitutions that are involved in adaptation to relatively small changes in environmental temperature. The conclusion is supported by a series of observations. First, the set of six amino acid substitutions tested here may represent relatively non-perturbing changes to protein structure, as the resulting changes in enzyme function are relatively minor – they modify catalytic rate and ligand affinity in response to shifts in habitat temperature of only a few degrees, corresponding to about a 1% change in thermal kinetic energy in the medium (Hochachka and Somero, 2002). Second, the position of each substitution on the surface of the protein, rather than in the core, supports the claim that the mutations will not have a large effect on overall (‘global’) protein stability, as it has been shown that substitutions in the core are more perturbing than those on the surface. Finally, many of the PSP algorithms are trained or tested against a database of protein mutation versus stability that incorporates examples from a wide range of habitat temperatures, including Archaeal proteins that may be stabilized at temperatures approaching 100°C (i.e. the ProTherm database; http://www.abren.net/protherm). Thus, we may be asking the PSP algorithms to detect changes in protein stability at a resolution that exceeds their sensitivity. In any case, our data suggest that current PSP algorithms may not be suitable for the analysis of stability changes in enzymes adapted to thermal habitats varying by only a few degrees. Instead, we must continue to rely on laboratory-based approaches to assess functional differences in orthologs, and associate them with specific amino acid substitutions.
Studies of LDH-A and cMDH have elucidated several key aspects of enzyme adaptation to temperature, including the traits most sensitive to selection, notably ligand binding and catalytic rate, and the structural alterations required to effect these adaptations in function. When we place the comparative studies of enzyme evolution into the broader context of the entire proteome and ask, ‘How much protein evolution to temperature is likely to be required for ectotherms to cope adequately with relatively small changes in temperature, such as those expected to occur with on-going global warming?’ we face the need to examine proteome-wide differences in thermal properties of proteins. Whereas we currently cannot determine what proportion of proteins in the proteome must adapt to increases in temperature of a few degrees Celsius, there are some data suggesting that not all proteins may be as sensitive to temperature as LDH-A and cMDH. For example, in comparisons of orthologous variants of six enzymes in ATP-generating pathways of the congeneric blue mussels M. galloprovincialis and M. trossulus, only two enzymes, cMDH (Fields et al., 2006) and isocitrate dehydrogenase (IDH) (Lockwood and Somero, 2012), showed adaptive variation. No differences in temperature versus Km responses were seen for phosphoglucomutase, phosphoglucose isomerase, phosphoenolpyruvate carboxykinase or pyruvate kinase (Lockwood and Somero, 2012). For IDH, two amino acid substitutions that affected flexibility were found, consistent with work on LDH-A and cMDH that demonstrated that adaptation may require minimal change in sequence.
The finding that not all enzymes appear to be equally likely to undergo adaptation to slight differences in temperature makes proteome-wide comparisons a priority, not only to provide a more complete account of past evolutionary events but also to develop predictive models of protein evolution in a warming world. As sequenced genomes appear in greater numbers, the ability to deduce sequence differences among orthologous proteins of differently thermally adapted organisms will concomitantly grow. If researchers can conduct proteome-wide analyses of evolution of thermal sensitivities using algorithms employing sequence data, perhaps in conjunction with models of three-dimensional structures of the relevant proteins, we will gain the ability to quickly predict how proteomes must change in response to the rising environmental temperatures predicted by models of global change.
We acknowledge the contributions of other members of ‘Team Dehydrogenase’ – Elizabeth Dahlhoff, John Graves, George Greaney, Peter Hochachka, Linda Holland, Glenn Johns, Margaret McFall-Ngai, Joseph Siebenaller and Paul Yancey – in developing the perspective on protein adaptation presented in this review. We also acknowledge the intellectual guidance of Bruce Sidell and the opportunity he provided to study Antarctic notothenioids at the Palmer Research Station.
All authors contributed to the conceptual organization of the paper; P.A.F. drafted the manuscript and G.N.S., Y.D. and X.M. helped develop the final draft; P.A.F. performed the bioinformatics analyses.
Portions of this work were supported by grants from the National Science Foundation (IOS-0718734 to G.N.S. and MCB02-35686 to P.A.F.) and the Partnership for the Interdisciplinary Study of the Coastal Oceans (PISCO) sponsored by the David and Lucile Packard Foundation and the Gordon and Betty Moore Foundation.
The authors declare no competing or financial interests.