Setting the stage for evolution of a new enzyme

Abstract

Evolution of novel enzymes has fueled the diversification of life on earth for billions of years. Insights into events that set the stage for evolution of a new enzyme can be obtained from ancestral reconstruction and laboratory evolution. Ancestral reconstruction can reveal the emergence of a promiscuous activity in a pre-existing protein and the impact of subsequent mutations that enhance a new activity. Laboratory evolution provides a more holistic view by revealing mutations elsewhere in the genome that indirectly enhance the level of a newly important enzymatic activity. This review will highlight recent studies that probe the early stages of evolution of a new enzyme from these complementary points of view.

Enzymes catalyze chemical reactions by up to 26 orders of magnitude [1]. The evolutionary origins of these magnificent catalysts have fascinated biochemists for decades. This review will highlight two experimental approaches that probe the early stages of enzyme evolution. Ancestral reconstruction [2,3] allows characterization of the level of physiologically irrelevant promiscuous activities in ancestral proteins and the impact of mutations on structure and function as a new enzymatic activity evolves. This approach was proposed by Pauling and Zuckerkandl in 1963 [4], but due to technological limitations, was not brought to fruition by the Benner group for nearly thirty years [5]. Since then, scores of papers have examined the evolution of steroid hormone receptors [6–8], opsins [9–12], transcriptional regulators [13], elongation factors [14], fluorescent proteins [15], globins [16], enzymes [17–20] and metabolic pathways [21].

Since Pauling’s and Zuckerkandl’s seminal article, ever more sophisticated computational methods for ancestral reconstruction have been developed. Recent reviews describe the theory and application of various methods, as well as their strengths and limitations [2,22–24]. In brief, ancestral sequence reconstruction requires a multiple sequence alignment of a set of homologous protein sequences that adequately represents the diversity of extant proteins. Maximum likelihood methods infer ancestral sequences based upon a phylogenetic tree and an evolutionary model that describes the probability of substitution at each aligned position in the protein. Bayesian methods estimate the phylogenetic tree and the parameters of the evolutionary model simultaneously. In either case, the output is a list of the probabilities of each amino acid at each position for each ancestral node. The most likely ancestral sequence has the highest probability amino acid at each position.

A limitation of ancestral reconstruction is that, due to ambiguities in ancestral reconstruction, even the most likely sequence has a low probability of being the real ancestral sequence. For example, a 100-amino acid protein for which each position has been reconstructed with a probability of 95% has a probability of only 0.006 [(0.95)¹⁰⁰]. In practice, many positions are predicted with a probability of <50%, particularly if the extant proteins are highly diverged. Thus, the maximum likelihood ancestor is usually highly unlikely. This problem is typically addressed by reconstructing multiple possible ancestral proteins to ensure that their properties are similar and therefore likely to be representative of the true ancestor [22,23]. Using this approach, Bar-Rogosvky et al. [25] found that the most probable ancestor of mammalian PONs exhibited enzymatic activities with several substrates that were representative of an ensemble of probable sequences that varied at the ambiguous positions most likely to affect function. In contrast, the most probable ancestor of vertebrate PONs, which had more ambiguously reconstructed positions, exhibited only one of many enzymatic phenotypes represented in the ensemble. This work is an important cautionary tale for efforts to reconstruct and characterize very ancient proteins.

A further limitation of ancestral reconstruction is its focus upon a single protein. It cannot address the cellular context in which a new enzyme evolved, or other mechanisms that helped cells respond to a new selective pressure. Laboratory evolution of microbes in which a promiscuous enzyme has been recruited to serve a new function offers the opportunity to observe the early processes in evolution of a new enzyme starting from a known progenitor. Laboratory evolution is carried out by growing an organism in a defined medium for hundreds or thousands of generations. Long-term cultivation is enabled by either serial dilution or use of continuous cultivation devices such as chemostats or turbidostats. Addition of mutagens, which would sprinkle mutations throughout the genome and obscure the identification of adaptive mutations, is not necessary. Mutants with adaptive mutations typically begin to accumulate within tens to hundreds of generations [26–32]. However, evolution of a highly efficient and properly regulated new enzyme may require an impractically long time frame. Thus, these two approaches provide complementary perspectives on the emergence of new enzymes in nature.

Evolution of catalytic activity in a non-catalytic protein

The existence of large superfamilies of enzymes [33–38] suggests that most enzymes have evolved from previously existing enzymes by gene duplication and divergence. But how does catalytic activity evolve in the first place? A recent study tracked mutations that conferred enzymatic activity on a non-catalytic protein. Cyclohexadienyl dehydratase (CDT), which converts prephenate to phenylpyruvate (Figure 1a) and L-arogenate to L-phenylalanine, is distantly related to non-catalytic periplasmic amino acid-binding proteins that deliver solutes to inner-membrane ABC transporters [39]. Clifton et al. [40] resurrected ancestral proteins at nodes of a phylogenetic tree connecting CDTs and related solute-binding proteins (Figure 1b). The earliest reconstructed protein, AncCDT1, had high affinity for cationic amino acids. This protein had three residues, Asp29, Asn152 and Thr169, that would later be important in the CDT active site, but no catalytic activity. AncCDT2 had lost affinity for cationic amino acids, but, like its descendant, the extant solute-binding protein Pu1068, had weak affinity for negatively charged ligands. AncCDT2 had also acquired several changes that would later be important for enzymatic activity. Substitution of Glu for Val173 provided the general acid that protonates the departing hydroxyl group of prephenate in CDT (Figure 1c). Two changes (D19T and A20G) altered the position of Trp60 and reshaped the ligand-binding site. Three additional changes introduced residues (Lys100, Asn128 and Asn133) that would later form hydrogen bonds with the departing carboxylate of prephenate. Despite these changes, AncCDT2 lacked catalytic activity. Weak CDT activity (k_cat/K_M = 6 M⁻¹ s⁻¹) appeared in AncCDT3. The only change near the active site between AncCDT2 and AncCDT3 (L198K) does not confer catalytic activity, so unidentified changes at remote sites must also have been important.

Figure 1.

Evolution of CDT activity in a non-catalytic protein. (a) The reaction catalyzed by CDT. (b) A phylogenetic tree connecting extant CDTs and related amino-acid binding proteins. Yellow, affinity for cationic amino acids; grey, affinity for negatively charged ligands; green, CDT activity. (c) Comparison of the ligand binding site of AncCDT1 (purple) and the active site of extant P. aeruginosa CDT (grey). Reprinted by permission from Springer Nature Nature Chemical Biology 14:542–547, Evolution of cyclohexadienyl dehydratase from an ancestral solute-binding protein, Clifton BE, Kaczmarski JA, Carr PD, Gerth ML, Tokuriki N, Jackson CJ.

A crystal structure of P188L AncCDT3, which has a modestly higher k_cat/K_M (155 M⁻¹ s⁻¹), showed that the positions of active site residues are nearly identical to those in Pseudomonas aeruginosa CDT (k_cat/K_M = 9.8 × 10⁵ M⁻¹ s⁻¹). (The beneficial P188L change was identified by shuffling AncCDT3 genes predicted by two different reconstruction methods.) Because simply assembling the catalytic machinery did not suffice to generate high activity, Clifton et al. hypothesized that a change in the dynamics of a hinge that opens and closes the active site might have been required. Indeed, MD simulations showed that AncCDT1 prefers the open conformation, which favors ligand binding and delivery to the ABC transporter, but P. aeruginosa CDT prefers the catalytically competent closed conformation.

Three intriguing findings emerged from this study. First, several residues required for catalysis appeared before the protein acquired any enzymatic activity. Second, an unsuspected novel noncatalytic function evolved between the earliest ancestor, AncCDT1, and CDT. Finally, efficient catalysis required adjustment of protein dynamics.

Exploitation of pre-existing buried aromatic residues enabled more efficient degradation of lignin

Evolution of new catalytic abilities usually exploits a previously existing promiscuous activity that utilizes features of the ancestral active site. Recent studies of the evolution of ligninases [41] revealed that aromatic residues in the hydrophobic core of an ancestral enzyme, rather than its active site, set the stage for a novel electron transfer pathway that changed the efficiency of lignin degradation.

Lignin, a complex polymer formed by radical condensation reactions of phenylpropanoid precursors, provides structural integrity to plant cell walls. Wood-rotting fungi secrete three types of heme-dependent peroxidases that degrade lignin. In each case, the resting enzyme is oxidized by H₂O₂, yielding an Fe(IV)-oxo/porphyrin cation radical complex [42]. Manganese peroxidases (MnPs) oxidize Mn²⁺ at a site near the heme [43,44], releasing Mn³⁺ to diffuse to the surface of lignin and initiate radical depolymerization reactions by oxidizing phenolic residues. In lignin peroxidases (LiPs), a surface tryptophanyl radical formed by transfer of an electron to the porphyrin cation radical can access lignin directly and is powerful enough to oxidize the nonphenolic component of lignin [45]. “Versatile” peroxidases use both strategies [46].

The Martinez group used ancestral reconstruction to investigate the evolution of these three classes of peroxidases in Polyporales fungi [41]. The earliest reconstructed enzyme was a MnP. A surface tryptophan was acquired independently in Clades B and D (Figure 2). The ability to oxidize lignosulfonate (a soluble form of lignin produced by treating wood pulp with sulfite) increased at each stage of the Clade D lineage with the exception of the ancestor of lignin peroxidases (ALiP) [47]. This anomaly might truly reflect a decrease in activity, but might also be due to a peculiarity of the sequence reconstructed for either the ancestral VP in clade D (AVPd) or the ALiP. Only one sequence was resurrected at each node, so one or both might not be representative of the true ancestral enzyme. Notably, the enzymes became progressively stronger oxidants along the evolutionary trajectory [48]. ¹H-NMR data suggest that the bond between His177 and the Fe atom became progressively weaker, destabilizing the Fe(IV)-oxo species and making it a stronger oxidant.

Figure 2.

Evolution of lignin peroxidases. (a) Phylogenetic tree of fungal lignin peroxidases. CaPo, common ancestor of Polyporales peroxidases; CaD, common ancestor of Clade D peroxidases; AVPd, ancestral VP in Clade D; ALiP, ancestral lignin peroxidase; PCLiPA, P. chrysosporium lignin peroxidase A. Reprinted from ref. [47] under Creative Commons license 4.0 CC BY-NC-ND. (b) Homology models of ancestral peroxidases. Dashed circles on the CaPo structure show the positions that will be mutated as evolution toward the extant lignin peroxidases proceeds. Reprinted with modifications from ref. [41] under Creative Commons license 4.0 CC BY-NC-ND.

An intriguing aspect of this study was the observation that an incipient path for electron transfer from the heme to the enzyme surface was already present in the ancestor. Trp252 and Phe205 formed part of the hydrophobic core of the ancestral protein, but played no role in catalysis. However, their fortuitous location between the heme and the surface of the enzyme set the stage for emergence of a new catalytic strategy—transfer of electrons between the active site and the surface—as a result of a change of Ala172 to Trp in Clade D, and of Ala172 to Asp and then to Trp in Clade B.

Evolution of enzymes that degrade anthropogenic pollutants

In response to novel selective pressures presented by large-scale use of anthropogenic pesticides since the 1940s, enzymes that detoxify or degrade compounds such as DDT [49], atrazine [50], pentachlorophenol [51] and organophosphate insecticides [52] have evolved in an evolutionary blink of an eye. Yang et al. [53] examined the evolutionary origin of methyl parathion hydrolase (MPH). MPH was first identified in Pseudomonas sp. WBC-3, which was isolated from contaminated soil near a plant that manufactured methyl parathion [54]. MPH is a member of the metallo-β-lactamase superfamily, and is most closely related to bacterial dihydrocoumarin hydrolases (DHCHs). Yang et al. reconstructed enzymes at ancestral nodes separating extant MPHs and DHCHs. The ancestral DHCHs have high efficiency for hydrolysis of DHC (Figure 3). Each also has an inefficient ability to hydrolyze methyl parathion and the related organophosphates ethyl parathion, methyl paraoxon and ethyl paraoxon, substrates that appeared in the environment only in the 20^th century.

Figure 3.

Evolution of MPH from a DHCH. (a) Substrates for ancestral and extant enzymes. (b) Phylogenetic tree connecting MPHs and DHCHs. Catalytic efficiencies are shown for reconstructed ancestral proteins and extant DHCHs from Serratia marcescens (SmDHCH), Janthinobacterium sp. HH01 (JsDHCH), Pseudomonas sp. WBC-3 MPH and Burkholderiales bacterium JOSHI_001 (BbDHCH). Bar colors correspond to substrates shown in (a). (c) Crystal structure of Pseudomonas sp. WBC-3 MPH (PDB 1P9E). Five residues that improved MPH activity by 970-fold when introduced into AncDHCH1 are highlighted in orange. (d) Comparison of the active sites of AncCDT1 and MPH, showing that remodeling of the active site allowed binding of methyl parathion. (e) The fitness landscape for evolution of MPH from AncCDT1. Each node is designated by the absence (0) or presence (1) of sequence changes at five positions in the following order: 72, 193, 258, 271 and 273. Fold-changes in MPH activity are indicated in each node. Parts a, b, d and e reprinted with permission from Springer Nature Nature Chemical Biology 15:1120–1128, Higher-order epistasis shapes the fitness landscape of a xenobiotic-degrading enzyme, Yang G, Anderson DW, Baier F, Dohmen E, Hong N, Carr PD, Kamerlin SCL, Jackson CJ, Bornberg-Bauer E, Tokuriki N.

Pseudomonas sp. WBC-3 MPH is 89% identical to the most recent common ancestor of DHCHs and MPHs, differing at 32 positions (Figure 3b). Introduction of just 5 changes near the active site (Figure 3c) into the ancestral enzyme improved MPH activity by 970-fold and yielded an enzyme with kinetic properties nearly identical to those of extant MPH. Four of these changes enlarge the substrate binding site, allowing methyl parathion to bind (Figure 3d). Deletion of Ser193 in a nearby loop alters the conformation of the loop.

The recent emergence of MPH made it possible to identify five mutations that enhanced MPH activity in an ancestral DHCH, but the order in which the mutations occurred cannot be discerned based upon the ancestral reconstruction. To explore this question, the authors characterized 32 variants with all possible combinations of the five critical sequence changes. Only 19 of the 120 trajectories for accumulation of the five mutations were feasible (Figure 3e). The remainder involved at least one step where an additional mutation decreased the evolving MPH activity. Productive trajectories toward improved MPH activity could have begun with four of the five changes. Whereas one trajectory (broad arrow in Figure 3e) was most propitious, the availability of 19 others increases the probability of a successful outcome in a relatively short period of time.

The early stages of recruitment of a promiscuous enzyme to serve a new function

Ancestral reconstruction provides the opportunity to trace the evolution of improved function as an enzyme evolves. However, it cannot answer many questions about the process of evolving a new enzyme. For example, what promiscuous activities were available in the organism in which a new enzyme evolved? If there was more than one suitable starting point, why was one “chosen”? What changes in the environment or the genome enabled recruitment of a promiscuous enzyme to serve a new function? Was gene duplication/amplification involved? Laboratory evolution experiments allow many of these questions to be addressed. The environment and the genome of the parental strain are known, and genomes of intermediate strains can be sequenced to identify mutations acquired during adaptive evolution. Transcriptomic, proteomic and metabolomic analyses can be employed to identify physiological changes in the system as a whole, rather than in the protein alone, as evolution of a new enzyme proceeds.

Three recent studies have shown that evolution of a new enzyme begins with recruitment of a promiscuous enzyme to serve a new function, as expected, but that the earliest mutations increase the cellular activity of the weak-link enzyme indirectly rather than modifying the enzyme itself. Kim et al. [28] identified a four-step bypass pathway (Figure 4a) that emerges within 150 generations when pyridoxal 5’-phosphate (PLP) synthesis in E. coli is blocked by deletion of pdxB. This pathway depends critically on a promiscuous erythronate dehydrogenase activity of SerA, which normally oxidizes 3-phosphoglycerate in the first step of serine biosynthesis. The erythronate dehydrogenase activity of SerA is compromised by competition from the native substrate, 3-phosphoglycerate, and also by feedback inhibition of the enzyme by serine. Mutations abolished feedback inhibition by serine in three of 10 evolved strains, allowing PLP synthesis to continue even when adequate serine was present. However, most of the evolved strains had no mutations in serA. In one of these strains, mutations in genes encoding enzymes in glycolysis and the pentose phosphate pathway led to diversion of flux away from glycolysis, resulting in low levels of 3-phosphoglycerate, a glycolytic intermediate, and its downstream product, serine. Elevating the productivity of the promiscuous reaction by minimizing competition from the native substrate and feedback inhibition clearly provided the most accessible way to immediately improve PLP synthesis.

Figure 4.

Promiscuous enzyme activities recruited to serve new functions in laboratory evolution experiments in E. coli. (a) A four-step bypass pathway restores synthesis of PLP after deletion of pdxB [28]. (b) Recruitment of MetB to synthesize 2-ketobutyrate restores isoleucine synthesis after deletion of ilvA and tdcB [55]. Dashed lines indicate promiscuous reactions.

Cotton et al. [55] carried out a similar experiment in E. coli by deleting genes encoding two threonine deaminases that produce 2-ketobutyrate in the isoleucine synthesis pathway. Evolution of this strain on glucose resulted in recruitment of MetB (cystathionine γ-synthase) to convert O-succinylhomoserine to 2-ketobutyrate (Figure 4b). MetB catalyzes this reaction quite efficiently (k_cat/K_M = 1.6 × 10⁴ M⁻¹s⁻¹) in the absence of the usual co-substrate, cysteine. However, in the presence of cysteine, O-succinylhomoserine is converted to L-cystathionine. Although one might expect that a mutation in metB could increase the efficiency of O-succinylhomoserine conversion to 2-ketobutyrate, none of the evolved strains acquired a mutation in metB. Rather, the evolved strains appear to have favored the cleavage reaction by altering the ratio of O-succinylhomoserine to cysteine.

Six of nine evolved strains increased the level of O-succinylhomoserine by deleting metC. MetC is one of two enzymes in E. coli that convert L-cystathionine to L-homocysteine in the methionine synthesis pathway. Deletion of metC in a “Δ5” strain lacking the two threonine deaminases and also three serine deaminases that might have substituted for the missing threonine deaminases (but didn’t) caused a partial block in methionine synthesis and consequently a 3-fold increase in the level of O-succinylhomoserine upstream of the block.

One evolved strain decreased the level of cysteine. A mutation in cysE, which encodes serine acetyl transferase, decreased k_cat/K_{M,acetyl CoA} by 17-fold. Introduction of this mutation into the Δ5 strain resulted in a 2-fold decrease in the level of cysteine. Further, the decreased availability of cysteine impaired methionine synthesis, leading to 5-fold upregulation of metB, providing a second mechanism for increasing 2-ketobutyrate production.

An increase in fitness due to mutations elsewhere in the genome was also observed after evolution of a strain of E. coli in which a promiscuous activity of E383A ProA (γ-glutamyl phosphate reductase)(ProA*) had been recruited to substitute for ArgC (N-acetylglutamyl phosphate reductase) in arginine synthesis [27]. proA* amplified within 200 generations of growth of the ΔargC proA* strain on glucose + proline to select for mutations that improved arginine synthesis. Mutations that indirectly enhanced the neo-ArgC activity of ProA*occurred in all eight replicate populations. Some increased expression of the enzyme upstream of the weak-link ProA*, likely pushing material through the compromised pathway. Others abolished feedback inhibition of an enzyme that produces a co-substrate for a later enzyme, likely pulling material through the pathway. In just one population, a mutation that changed Phe372 to Ala boosted the inefficient neo-ArgC activity of ProA*.

A common feature of these cases is that the effectiveness of a newly important activity was increased indirectly by mutations elsewhere in the genome, rather than mutation of the gene encoding the enzyme itself. The rarity of mutations that enhance the catalytic efficiency of newly recruited enzymes in the short term is likely due to differences in mutational target size. Most of the beneficial mutations in these three experiments altered the metabolic network due to total or partial loss of function of an enzyme. Loss of function is easily achieved by insertions, deletions, or point mutations at many places in a gene. In contrast, increasing the efficiency of a new activity may require surgically precise modification of specific active site residues.

Summary

Ancestral reconstruction experiments have provided a fascinating glimpse into how mutations set the stage for emergence of new activities as well as the structural basis for improvements in a new activity. Experimental evolution studies place the process of evolving a new enzyme in a cellular context, and have revealed unexpected complexities surrounding the initial recruitment of a promiscuous enzyme to serve a new function. Both approaches contribute to our expanding understanding of enzyme evolution and could potentially be used in conjunction by carrying out experimental evolution under selective pressure for improvement of a predicted ancestral enzyme. Major gaps and challenges still remain, however. Evolution of a new enzyme requires not only improvements in catalytic activity, but also evolution of proper regulation. In particular, we have little insight into the evolution of allostery, a major mechanism for post-translational regulation of enzyme activity. One recent study used ancestral reconstruction to investigate the evolution of allostery [56], but the findings are controversial [57,58]. The prevalence of loss-of-function mutations in laboratory evolution experiments is well-recognized [59–62], but how mutations that indirectly improve the efficiency of newly recruited enzymes at a cost to previously well-evolved functions are repaired after evolution of a new enzyme has not yet been addressed.