Dan Tawfik’s Lessons for Protein Engineers about Enzymes Adapting to New Substrates

Abstract

Natural evolution has been creating new complex systems for billions of years. The process is spontaneous and requires neither intelligence nor moral purpose but is nevertheless difficult to understand. The late Dan Tawfik spent years studying enzymes as they adapted to recognize new substrates. Much of his work focused on gaining fundamental insights, so the practical utility of his experiments may not be obvious even to accomplished protein engineers. Here we focus on two questions fundamental to any directed evolution experiment. Which proteins are the best starting points for such experiments? Which trait(s) of the chosen parental protein should be evolved to achieve the desired outcome? We summarize Tawfik’s contributions to our understanding of these problems, to honor his memory and encourage those unfamiliar with his ideas to read his publications.

Protein engineers redesign enzymes to improve their utility as biocatalysts, biologic therapies, and diagnostics. When rational design (structure-based site-directed mutagenesis) is challenging, directed evolution (iterated cycles of random mutagenesis and high-throughput screening) techniques are applied to modify the structures and functions of wild-type enzymes. The latter approach has motivated the development of sophisticated techniques for the construction of sequence variant libraries and the evaluation of clones in high-throughput assays. In the absence of an evolutionary model, however, advanced techniques are too often applied through trial and error. What mutation rate is optimal? How many mutant clones should be screened per generation? How many generations will be required? Protein engineers can take the guesswork out of directed evolution by working within the framework established by the late Dan Tawfik, particularly in the first part of his career.

Tawfik was a very prolific writer and speaker, but his inclination toward theory distinguished him in an empirically oriented field. Experimental results generally precede theory in the life sciences. Engineers sometimes suppose that technical advances, such as machine learning algorithms and ultrahigh-throughput screening techniques, will be sufficient to overcome the bottlenecks in evolutionary workflows. One of his early articles, coauthored with Andrew Griffiths, on water-in-oil emulsions¹ remains his most cited because the technology it described proved useful for a wide variety of applications including next-generation DNA sequencing. Tawfik often reminded us, however, that proteins with novel molecular recognition properties evolve very quickly in nature without any need for artificial intelligence or fluorescence activated cell sorting. Here we review his subsequent work that shaped our understanding of the way the substrate specificity of enzymes evolves.

We focus on two important questions that are often overlooked. First, how should the parental protein of a directed evolution experiment be chosen? Some proteins are more evolvable than others for reasons discussed later, so this decision is arguably the most consequential for the design of a successful experiment. Second, what trait(s) should be favored? Everyone who evolves proteins in vitro learns the hard way that you get what you screen for. And yet, proteins must survive multiple selection pressures as they evolve in nature. Tawfik addressed these questions by conducting many experiments, and also by synthesizing the work of others, over two decades. We hope that this brief review of those experiments will guide novices toward more efficient evolutionary schemes.

Tawfik was particularly interested in the evolution of enzymes with novel substrate specificities. Protein engineers have traditionally relied on intuition to choose parent proteins to seed their directed evolution experiments. If an enzyme that catalyzes a particular reaction is desired, it seems sensible to start with one that catalyzes a similar reaction, preferably a member of a versatile protein family that folds properly in Escherichia coli. Although this strategy has worked for numerous investigators, including ourselves,^2,3 we cannot know about unsuccessful, and therefore unpublished, attempts. Moreover, the designation of a single, wild-type protein as an evolutionary starting point is arguably a form of intelligent design. Natural selection as envisioned by Charles Darwin acts upon every population at every level of biological order. Protein engineers oversimplify at their peril.

Tawfik devised a number of practical approaches to choose the starting points for directed evolution experiments. He hypothesized in an influential review article⁴ that “the natural divergence of enzyme families and superfamilies had proceeded through multispecific, or highly promiscuous, progenitors, or ‘node intermediates’”. He noted that “promiscuous functions often appear in one family member but not the others”, which implied that protein engineers should investigate the substrate specificities of more than one family member as he and another pioneer, Shelley Copley, did.⁵⁻⁸ Such studies taught us much about the way substrate evolution evolved in nature. We learned, for example, that the “primary, or native, function of one family member is often identified as a promiscuous activity in other family members”.⁴ But studies of protein families are labor-intensive and impractical for those who strive to evolve enzymes that recognize unnatural substrates.

Another way to choose an ancestral population is to evolve one from a semirationally chosen wild-type gene. The groups of Tawfik and Frances Arnold, working separately, showed that neutral drift breeds populations that are more evolvable than the progenitor clone.⁹⁻¹¹ Strictly speaking, population biologists define genetic drift as the change in allele frequency in the absence of selection. It therefore underpins the neutral theory of evolution.¹² On the other hand, and somewhat confusingly, protein engineers have come to define neutral drift as iterated rounds of random mutagenesis with retention of the clones with wild-type activity or specificity. Why are the populations that survive a neutral drift experiment more evolvable than a phenotypically similar clone? In practice, it is very difficult to direct the evolution of enzyme variants that react with a novel substrate unless that reaction can be detected by a high-throughput assay. Such promiscuous reactions are sometimes shared by different members of a protein family but can vary greatly in magnitude.⁴ Promiscuity by definition does not affect fitness in nature, so the strength and evolvability of any particular promiscuous activity is partly a function of contingency.¹³ The probability that any particular wild-type enzyme will possess a detectable secondary catalytic activity is much lower than that of a population of sequence variants derived from that progenitor. Tawfik set out to quantify this by analyzing 311 variants from a neutral drift experiment on the PON1 paraoxonase enzyme. Overall, 29 of the variants showed improvements in specific activity toward at least one promiscuous substrate, and 37 showed changes in substrate selectivity of ≥5-fold compared to wild-type PON1.⁹ It is clear that a subset of variants exhibiting an activity is more likely than any single clone to include unique amino acid residues that are neutral in isolation but synergistic in combination with some subsequent beneficial mutation(s).^9,11

The benefits of neutral drift extend beyond the formation of “neutral networks” of sequence variants. A significant amount of purifying selection occurs during such experiments; that is, neutral drift experiments apply a selection pressure for removing variants that are deleterious to the fitness of the host cell. Random mutagenesis introduces mutations that can affect fitness in different ways.¹⁴ Most proteins are marginally stable, so robustness to mutation is correlated with conformational thermostability.¹⁶ Most random mutations are destabilizing,¹⁷ but only those deleterious enough to cause a protein to unfold impart any fitness cost. Deep mutational scanning experiments (in which most or all possible single-amino-acid substitutions are analyzed) show that most mutations reduce the wild-type activity of a protein; for example, ∼75% of mutations in green fluorescent protein reduce its fluorescence.¹⁸ This ensures mutations that are selectively neutral or beneficial to any extent will accumulate over multiple rounds of mutagenesis and high-throughput screening for the wild-type activity. To summarize, the population of variants that survives neutral drift is more evolvable because some individual proteins exhibit higher levels of a desired promiscuous activity and because many are better able to accommodate mutations that simultaneously alter substrate specificity and destabilize the active conformation. Overall, this process is likely to improve the mean thermostability of the evolving population and, therefore, its robustness to subsequent mutations.^10,19

If the benefits of neutral drift are understood, can they be captured more expeditiously? One way to try is to align the amino acid sequences of known orthologues, determine a consensus, synthesize that consensus sequence (in accordance with the codon bias of the host species), and employ it as the progenitor for subsequent directed evolution.²⁰ Consensus sequences tend to be thermostable and, therefore, robust to mutation.^10,21 Family shuffling attempts to recapitulate neutral networks, so that variation which survived natural selection is overrepresented in libraries of mutants.²² Advances in deep sequencing, however, are making it clear that even the most comprehensive databases capture only a tiny fraction of all enzyme sequence diversity in the biosphere,²³⁻²⁶ and some groups are beginning to reveal the diversity in wild-type function across ever-larger phylogenetic distances.²⁷ The next step will be to include vastly more of these potentially informative and useful sequences into the starting populations for directed evolution experiments.

The choice of an evolvable progenitor, one that is robust and exhibits detectable promiscuous activity, is consequential but also potentially time-consuming. Tawfik offered labor-saving suggestions to those reluctant to invest the time required to evolve a progenitor. One approach is to screen an Open Reading Frame (ORF) collection made by others,²⁸ which requires no design or cloning at all. Another alternative is to start with a rapidly evolving viral protein because those are products of high mutation rates and neutral selection.²⁹ Tawfik also argued that enzymes catalyzing the biosynthesis of natural products are more evolvable than their counterparts in central metabolism. Secondary metabolic enzymes constantly adapt to changes in the chemical environment.²⁹ Secondary pathways evolved more recently than did enzymes of core metabolism and are consequently broader in specificity and on average 30-fold lower in their turnover number, k_cat.³⁰ In other words, most enzymes that catalyze core metabolic reactions are specialists, while those that catalyze secondary reactions are more likely to be promiscuous, robust to mutations, and therefore evolvable.

Once the progenitor protein(s) are chosen, random mutations are introduced at moderate rates. For whole gene mutagenesis, ∼2 mutations/gene is appropriate for screens with throughput high enough to assess all likely single mutants. Herman and Tawfik devised a way to incorporate synthetic nucleotides into assembly PCRs so that mutations occur in moderately conserved residues.³¹ Other ways to mutate particular residues via rational and computer-guided design³²⁻³⁸ can be more efficient than whole gene mutagenesis. These approaches, however, fall outside of the scope of this review. We focus instead on the traits that ought to be optimized via directed evolution and the outcomes that users can reasonably expect. The strategy of neutral drift was already described earlier. It is also possible to screen for hypermorphic variants, with increased native activity due to improvement in expression, folding, or thermostability.

Tawfik and his colleagues showed that, as a general rule, detectable promiscuous activities of enzymes can increase 1000-fold or more from the accumulation of small numbers of missense mutations over three to five rounds of directed evolution.³⁹ Native enzyme activities are, in general, apparently more robust than promiscuous ones, so directed evolution generally results in variants with both the original and novel activities. In his highly influential 2005 paper,³⁹ Tawfik observed that the robustness of native activities, and the evolvability of promiscuous ones, reflected an important structural feature. Mutations that enhanced promiscuous activities were on mobile loops at the perimeters of enzyme active sites—not in the catalytic machinery itself, nor in the core scaffold responsible for the stability of the overall fold. Subsequent work built on this powerful idea, emphasizing that highly evolvable folds were characterized by showing this “division of labor” between a structurally rigid scaffold, key catalytic residues, and the mutable regions responsible for substrate binding.⁴⁰ Well-known examples include TIM barrels and Rossmann folds.⁴¹ On the other hand, proteins in which the evolution of surface residues is constrained (such as when they are involved in protein–protein interactions), “freezes” evolution of the core as well.⁴² These evolutionary insights made it clear why folds such as the TIM barrel are also the best starting points for protein design.⁴³

One aspect of this robustness model was critically examined in 2016 by Nobuhiko Tokuriki and his colleagues.⁴⁴ They showed that the 26 mutations previously discovered to adapt a model enzyme, phosphotriesterase, were individually more deleterious to the native activity than to the promiscuous arylesterase activity, contrary to the predictions of Tawfik’s model from a decade earlier (i.e., a robust native activity and evolvable promiscuous ones). The mutations were epistatic, so that earlier ones increased the promiscuous activity greatly without reducing the native activity much. Later mutations, in the context of the earlier ones, greatly diminished the native activity but only modestly improved the promiscuous one. The observed mutational robustness of native enzyme activities is thus a function of epistasis rather than any intrinsic structural difference between native and novel activities.

The observed epistasis is broadly analogous to a person changing clothes—the first and last steps (buttons and zippers) are essential to the process but do not make as much difference as those in the middle (removal of one shirt and donning of another). The authors also showed that 400 random mutations of the wild-type PTE were mostly either deleterious to the native activity, or to the promiscuous activity in roughly equal proportions, or to both. Another small fraction was neutral, and an even smaller fraction exhibited improvement in the promiscuous arylesterase activity. These results confirm that the native activity is not uniquely robust. Finally, adaptive mutations that occurred in the natural evolutionary transition between the melanine deaminase, TriA, and atrazine dechlorinase, AtzA, exhibit the same epistatic pattern observed during the laboratory evolution of PTE into an arylesterase. These results confirm that the effect is not unique to phosphotriesterase.

The findings of Kaltenbach et al.⁴⁴ offer guidance to protein engineers who wish to evolve enzymes so that their promiscuous activity becomes as efficient and specific as the original native activity. Positive selection of shuffled (recombined) mutants with improved promiscuous activity will lead to the evolution of broad specificity variants that retain considerable levels of native activity. Additional adaptation will improve the desired promiscuous activity more modestly, while precipitously reducing the original native activity. High-throughput assays for positive selection would have to be broad in dynamic range and precise (e.g., little phenotypic variation between genetically identical clonal cultures) to detect late-stage adaptive mutants with any confidence. On the other hand, a negative selection (for weakened native activity) followed by positive selection for those with promiscuous activities similar or better than those isolated in the previous round might prove more efficacious.

Box 1. Glossary of key terms for practitioners of directed enzyme evolution.

• Contingency: the sensitivity of evolutionary outcomes to circumstances particular to the history of the ancestor.

• Consensus sequence: a synthetic protein sequence designed by aligning the sequences of available homologues and identifying the most common amino acid residue at each site.

• Directed evolution: iterated cycles of random mutagenesis and high-throughput screening.

• Epistasis: amino acid changes that are nonadditive in effect.

• Evolvability: the capacity of a protein to change in structure and function over time.

• Generalist: an enzyme that reacts with multiple substrates, particularly in instances in which none of the reactions are catalyzed very efficiently.

• Hypermorphic variant: a mutant that exhibits the same activity as its wild-type ancestor but at greater magnitude.

• Library: a population of alleles.

• Neutral drift: iterated rounds of random mutagenesis and identification of clones that retain wild-type activity.

• Neutral network: population of alleles that are phenotypically identical but vary in their evolvability.

• Promiscuity: activity toward substrates or reactions that are not the native ones. Critically, and to quote Tawfik himself, promiscuous activities “are purely accidental, and emerged under no selection or physiological relevance”.⁴⁵ More precisely, promiscuity can be subdivided into substrate ambiguity (catalyzing the same chemistry on a molecule resembling the native substrate) and catalytic promiscuity (catalyzing a different chemical transformation via a different mechanism).⁴⁶

• Rational design: structure-based site-directed mutagenesis.

• Robustness: resistance to changes in the sequence or the (bio)chemical environment without change in structure or function.

W.M.P. was supported by Grant 18-VUW-050 from the New Zealand Marsden Fund.

The authors declare no competing financial interest.

Author Status

I.M. is an emeritus (retired) professor. He retains his email address (imatsum@emory.edu).