Dan S. Tawfik is the Nella and Leo Benoziyo Professor in the Department of Biological Chemistry at the Weizmann Institute of Science in Rehovot, Israel.

In 1969, Frank Salisbury, writing in Nature, observed that the fact of protein specificity and the fact of protein evolution are, if not in conflict, then under tension.1 At some moment, argued Salisbury, the prebiotic soup must have disappeared. Early life had to begin fending for itself. Demands were imminent. For each essential chemical that had become either scarce or unavailable, an enzyme had to appear to catalyze its synthesis. Since chemicals had precursors, enzymes would by necessity form pathways and cascades.

Salisbury urged his audience to imagine a mundane protein for such a task, one 300 amino acids in length. Its encoding gene must have been, to use a round figure, one thousand nucleotides.2 Large numbers now accumulate rapidly. Since there are four canonical nucleotides in biology and as many assumed for pre-biology,3 and one thousand positions along its hypothetical chain, there are 41000 ways of arranging this gene. This makes for roughly 10600 possibilities. The number, Salisbury observed, is too large coherently to grasp. Even if trillions of distinct sequences could fulfill a single enzyme function, there is only a 1 in 10500 chance that such an enzyme would arise spontaneously. Neither inflation of resources, bigger oceans, more sloshing chemicals, nor a multiplication of planets would make this probable.

If large numbers were the problem, Salisbury anticipated its resolution. Suppose proteins did not have to be quite so large? Perhaps proteins and their active sites could be only a few amino acids. Thereafter, they could grow bigger. This idea immediately engendered another problem. If active sites were so trivially configured as to be formed by only a few amino acids, Salisbury realized, proteins would be covered with them. Vagrant catalysis and binding potentials would plague every molecule, making sustainable function impossible. This could not be. Life needs proteins of substantial length for most enzymatic catalysis. Life needs sequenced cascades for coherent context.

It did not take long for an orthodox voice to be raised in response.4 Writing in 1970, John Maynard Smith argued that with respect to possibilities, Salisbury was thinking of the wrong ones; and with respect to problems, he had been considering the wrong kind. Do not try to imagine the origin of proteins, Maynard Smith urged. The problem is too difficult. Imagine instead that one protein, or perhaps several, has already been formed; and imagine that some version of the cell is already operational. Let us talk about evolution after that. Rather than considering all possibilities at once, consider only small variations. In a protein one-hundred residues long, each position could, because there are only twenty amino acids in the biological canon, change by single point mutations to one of nineteen other possibilities. Nineteen alternatives at each position × 100 total positions = 1,900 total possibilities for evolution to consider.5 The structural redundancy of the genetic code prohibits some single nucleotide changes; in practice, only about a thousand new alternatives would be accessible from a starting sequence of one hundred amino acids. As long as there is one small modification to which a protein might mutate to gain vitality, or at least not to lose it, evolution could start to diversify a family of proteins. The functional proportion (f) of an ensemble achieved by small mutations (N) has to be greater than or equal to one (fN ≥ 1) for evolution to work. Since there are a thousand possible mutants from any serviceable starting point, evolution requires beneficial and neutral mutations at a rate f ≥ 1/1,000. One can imagine this scenario: After several false starts, one good variant appears; selection preserves it; then another, and another; and the variants begin to vary. A cloud of alternate sequences would cluster around the original sequence as a network of possibilities connected by unit increments.

The idea did not seem unreasonable. In the early 1970s, experiments by Paul Brown and Patricia Clarke et al. demonstrated that enzyme specificity could be altered steadily by single point mutations.6 Functional sequences already exist in a prime corner of the space of all possibilities. They could reach out.

Maynard Smith framed this idea by means of an analogy that inspired the field of protein evolution for many years. Suppose that one draws letters randomly from a box. There is little chance that the sequence will form an English word. But if one begins with a small word, and asks, instead, how much the word can be changed by a modification of its letters: the answer is quite a lot. Single letter switches are sufficient:

WORD → WORE → GORE → GONE → GENE

Endless change is possible.

In all this, a deeper analogy was at work. Life is like language. Proteins are the words; amino acids, the letters; point mutations, the alphabetic changes; and functional chemistry, the meaning of the words. The grandeur of sequence evolution could be grasped by an elementary word game. But as Maurício Carneiro and Daniel Hartl observed, to make this analogy work, one must assume its validity.7 Start the game with “SYZYGY” and it ends immediately. Who could say with authority that the panoply of proteins in the living world represented something as innocent and changeable as the word “WORD” rather than the word, “SYZYGY?”8

Old Guard

In biology, concepts can be close to art. Pure theory invites an aesthetic reaction. Experiments are another matter. In 1890, Emil Fischer introduced organic chemists to the idea that an enzyme fits its substrates as a key fits its lock.9 The metaphor proved durable until 1958, when Daniel Koshland, writing in The Proceedings of the National Academy of Sciences, revised Fischer’s lock-and-key in favor of the induced fit model of enzyme action.10 The new model, and so the new metaphor, suggested that chemistry within active sites was more dynamic than Fisher had supposed. The keys remained keys, but Koshland endowed them with the power to adapt variously to locks. If active sites were more active than previously imagined, they were thus more likely to change than previously supposed.

This, too, seemed plausible given research results about simple enzymes and their path to enhanced specificity. Simple enzymes and their evolution were one thing, but experiments undertaken by Peter Rigby, Bruce Burleigh, and Brian Hartley suggested that evolution, confronted by “a more difficult problem,” would face very significant impediments.11 Various routes from ribitol dehydrogenase to xylitol specificity were tested. No continuous network between them could be found. Single unit steps met with no adaptive dividend. In his argument, Maynard Smith had warned that “... double steps with unfavorable intermediates may occasionally occur, but are probably too rare to be important in evolution.”12 Rigby et al. were prepared to embrace all that Maynard Smith had rejected:

We are forced to conclude that this ribitol dehydrogenase cannot evolve towards a new specificity for xylitol by single step mutations whilst under continuous selective pressure. ... If it is generally true that multiple mutations are required for evolutionary improvement, then it is likely that such mutations will often be incompatible with the tertiary structure [or basic function] of the protein.13

Once a threshold of complexity is reached, enzyme functions do not appear to change seamlessly. There are gaps in the middle of any path. Since selection above all preserves functionality, selection itself will work against those intermediates with no function at all. If evolution were to occur, natural selection would have to lapse.

As one might expect, the story doubles back on itself. So many stories in biology do. Some proteins can, it would seem, follow an adaptive path after all.14 The adaptive routes are not themselves distinct, but blur and tend to overlap. A protein can meander along several adaptive trajectories at once because proteins have, in addition to their primary function, helpful secondary features. If progress drags with respect to one function, selection may switch its focus to another, or to both at once. The result is that something or other is forever under the scrutinizing eye of a ubiquitous Darwinian force; or if not, then it is drifting inexorably to a place where it will be.

The drama of uncertainty and speculation reached its apogee in four questions proposed at the end of Maynard Smith’s Nature commentary.15 The importance of these questions is everywhere overlooked and universally understated. Far from being an afterthought, the issues Maynard Smith raised are the closest equivalent in molecular evolutionary biology to David Hilbert’s famous list of unsolved problems.

Maynard Smith’s list of questions, paraphrased:

  1. Are all proteins on earth evolving along a single unbroken evolutionary network, from a unique point? Two corollaries: (a) Are multiple networks possible? (b) Can there be multiple starting points along the same network?
  2. How frequently does evolution proceed through non-functional intermediates?
  3. What proportion of the evolutionary network has already been explored by life?
  4. What fraction of potentially useful proteins is inaccessible?

To answer these questions with quantitative rigor would be to take hold of molecular biology as Newton took hold of physics. Research in the 1960s and 1970s, though offering a background against which further questions could be raised, offered no solutions. The old guard had reached an impasse in the memorable but receding figure of Maynard Smith.

New Possibilities

In 2003, Dan S. Tawfik and his collaborators showed that the idea of one sequence–one structure–one function was even shakier than expected. The team found that an antibody called SPE7 could, by the rotations of pure chemistry, realign itself so that it alternated in equilibrium between closed and open configurations.16 Different structural isomers arose from the very same sequence. One sequence: two structures: two functions. This was previously unknown and not widely thought possible. The different shapes bind to distinct substrates: one small and the other big, with high efficiency and low efficiency, respectively.

This feature is not peculiar to antibodies. There is a small but growing list of proteins known to be metamorphic in character. They may be thought of as points on a grid. At such points, structures and functions really do grade seamlessly. A given genotype is not used up on a single function. It can transition. It is tempting and intuitive to imagine that historical protein transitions were similar.

While gripping, Tawfik’s experiment stopped short of anything like a decisive conclusion. The type of drastic structural plasticity seen in SPE7 is not unique, but it is rare, perhaps exceedingly rare, among proteins. The SPE7 antibody is anomalous. In biology, there are always exceptions; it is dangerous to move from exceptions to general principles. More frequently, diversity is subtle and occurs when a few amino acids may freely rotate or jiggle around the perimeter of the protein’s active site. Small movements create transient dents, bumps, and grooves in the shape. These indentations harbor evolutionary potential. Slight variants may be able to cling to new substrates, and catalyze new reactions, while performing the original reaction as well.

Something for Nothing

Consider the so-called starting point paradox.17 Stated briefly, evolution will not and cannot work unless a small but selectable function is already present in a population. If a function is there, it can be exaggerated; if not, it can’t. Something cannot evolve from nothing.

If not from nothing, then, of course, from something. There should be hints of new functional starting points, even while the old point is still occupied. Thereafter, evolution has options. In going from A to B, some small activity of function B needs to be initially in A for evolution to have a fighting chance. Activities then change proportions. New functions may be added along the way; they appear by chemical movement and exchange. But they are not primary upon their arrival, and never arise in enzymes that are otherwise functionless. Enzymes, then, must be multifunctional because they must avoid the incredible risk-taking exercise of giving up an entire (and refined) primary catalytic function to gain a weak new one for survival.

This idea seems to challenge the most celebrated property of enzymes: their exquisite specificity. Is it feasible that two or more functions could both appear and be housed on the same enzyme and in the same active site? SPE7 showed that one sequence could assume more than one structure, but equally, that one structure may control more than one function. This, of course, raises a question: Can the new function evolve coherently in relation to the primary one?

In 2005, Dan S. Tawfik et al. analyzed three proteins: serum paraoxonase (PON1), bacterial phosphotriesterase (PTE), and carbonic anhydrase II (CAII).18 The proteins were screened to increase, and so to evolve, their inherent secondary activities. CAII is an enzyme for hydration of carbon dioxide; PTE, for the hydrolysis of paraoxon (a man-made pesticide); and PON1 for the hydrolysis of lactones. These are not their exclusive activities or binding capacities: they can do more. If evolution was able to embellish alternate functions in the wild, it should be so in the lab.

Tawfik’s team used methods involving variation and strong selection to track the laboratory-engineered evolutionary process. Their aim was to enhance what is called promiscuous (or the secondary, or non-primary) activity for new substrates. The good that was found was kept, the bad discarded. The team found that increasing the activity of a secondary function did not (much) disrupt the original function. In terms of efficiency, the average gain-to-loss ratio was 10 to 1. For a 10-fold efficiency gain in secondary activity, there was only a 1-fold loss in the primary. In one unusual instance, the ratio was a startling 106 to 1. This was tested over six different promiscuous substrates.

Evolution was, in test cases, highly coherent. These proteins did not have to forfeit their original function before acquiring a new one. Furthermore, specialization occurred spontaneously. As one promiscuous secondary activity was gained and the original activity maintained, alternative activities diminished. The experiment targeted an increase in one promiscuous activity (esterase), but changes that made an increase possible led to a decrease in another promiscuous activity (hydrolase) in the same enzyme, both by a factor of one hundred—hence specialization. Tawfik multiplied the activity increase in one secondary feature (for example, one hundred times greater esterase efficiency) by the activity decrease in the other (e.g. 102 in hydrolase) to gain some sense of the activity reshaping, or respecialization, in the evolved enzyme. In the best and most extreme cases, there was a 104 shift in the secondary activity proportions. This is notable because specialization, along with efficiency, constitutes the chief concept of specificity: an integral, if not preeminent, feature of biology.

Promiscuity did not require drastic rearrangements; it proceeded by tweaks. Changes did not occur in the proteins’ crucial infrastructures, nor did they occur in the key catalytic machinery of the active site. Those sections contain critical residues that are difficult to vary without killing the protein. Change occurred in the peripheral loops on the protein’s surface, and so along the perimeter and walls of its active sites. Surface loops form and align the binding pocket of the protein. If the binding pocket changes, a new substrate can better fit the active site.19

There is a temptation to think that this is something of a chemical fluke. Are coherence and specialization that easy to purchase? If the general vicinity of an active site is filled with different sorts of atoms, there must generically be combinations that are sticky enough to cling to something or other. It would be difficult to induce them to move beyond this point. High-grade catalysis is, in contrast, exceedingly specific.

Tawfik and his team discovered that, at least in the case of SPE7, promiscuous binding need not be generic.20 They observed no kind of general hydrophobic stickiness. There were specific hydrogen bonding patterns between substrate and protein. Interaction of the protein with a new substance could involve different electronic contributions from the same group of atoms, or even different configurations of structure. If specific, these hydrogen-bonding patterns were obviously more biologically relevant.

This, too, was previously unknown.

To connect this phenomenon with reality, a level of respect must be given to practical complications. Much variation in biology is not the result of an omnipresent, directional selection. Variations occur and are maintained below, or even without, a positive selection threshold. Variations are sometimes resolute for drift. The promiscuous functions that Tawfik discerned did, of course, do something, and so they provided something. In life, these could perhaps be starting points for evolution. But from the experimental perspective, selection levels were unrealistic; and survival was based on designer protocols that were incommensurate with selective advantages.

In the wild, who knows?

Surfing Networks

Tawfik and his collaborators attempted to address this question. Selection could be directional. No one discounted the idea. What if it were not? What if selection were stabilizing, or purifying? Under this scenario, nature isn’t eagerly awaiting a fortunate novelty, nor is it all that concerned with amplifying it once the first trace has appeared. Perhaps a mutation provides an extra ability. Suppose an organism can, in its local environment, get along just as well without it, especially in its rudimentary state? At that point, a novelty may be something extra without being something helpful.

What then?

Gil Amitai, Rinkoo Devi Gupta, and Dan S. Tawfik designed an experiment to simulate purifying selection on serum paraoxonase, PON1.21 The protocol was straightforward. Vary gene copies by mutation, then select the expressed proteins for their primary function. Preserving the original function was the only goal. As long as this condition was met, all else was permitted. Selection would neither purge nor promote various background changes. They were lateral steps, neutral in their expression.

Still, what is neutral is relative. A change that does not perturb one feature may be significant with respect to another. In this case, the primary function is the target of selection screens. By design, changes diminishing it are rejected, the function itself remaining unaffected. Secondary functions, however, are otherwise. This can be imagined in the following way. A mutation introduces some small alteration of structure. The alteration does nothing to the primary function, but confers some chemical benefits on the secondary function. By an accumulation of neutral sequence mutations, the sequence pool expands toward transition points and new abilities. A small network is formed. Under stable conditions, small features offer no great advantage. However, a firm environmental shake (these happen from time to time) could expose a growing promiscuous function. Natural selection may favor it, especially in its exaggerated condition, though it was previously irrelevant. Life may derive benefits from a changing context.

The cloud of neutral sequences will be diverse, forming what Maynard Smith referred to as a neutral network. In surfing these networks, function is important, but so is position. Two organisms, for example, may have identical features and functions, and be selectively indistinct. Yet one organism may be five mutations away from a new useful feature, and the other organism, after collecting neutral changes, only one. In adaptive hardship, it will be in a much better position to evolve. Neutral evolution and Darwinian evolution, instead of being exclusive, can operate in symbiosis. A molecule that is in one sense targeted for survival accumulates additional features that are neutral to survival. Later, these features prove useful.

That was the idea. How did it fare? Quite well. PON1 is a lipolactonase. Tawfik’s team created a small library of 311 variants of PON1 and tracked the activity (the rate of formation of product molecules) of each variant, as well as its expression levels (or protein concentration). A neutral drift of up to seven amino acid exchanges was created. About half had altered promiscuous affinities. Since PON1 was also used in the positive selection study, a direct comparison of changes was possible. Several of the very same beneficial mutations appeared in the neutral study as in the directional study. These were, again, changes to binding pockets, outside of the key active residues. The implication was that favorable changes were consistent with both positive and purifying selection regimes.

This experiment scored two notable successes. It marked the first rigorous characterization of a neutral network appearing in the literature. Until then, Maynard Smith’s ideas had remained on paper or computers, and they were diffused among various non-systematic experimental results. Tawfik’s laboratory changed this. Parameters of fitness were monitored continuously and associated with precise sequence identity. Libraries did not have to be vast in order to exhibit functional gain.

Rewiring the Scaffold

Promiscuous functions are real, and their existence is intriguing, but they are not powerful. Promiscuous functions are typically three to five orders of magnitude less potent than the primary function of the enzyme on which both reside. An experiment demonstrating a clean transition in which secondary activity becomes primary, and primary activity secondary, would be a formidable demonstration of evolutionary principles.

The enzyme studied by Tawfik’s team was phosphotriesterase (PTE), already known to be something of an evolutionary superstar. Having evolved on the scale of decades instead of millions of generations, it acted with extraordinary efficiency on a man-made pesticide, paraoxon. A secondary activity was known. PTE could act on 2-naphthyl hexanoate or 2NH. The experiment used mutation, selection, and recombination to fully exchange primary and secondary activities.22 Both the evolved and original enzymes made use of the same key step, which involved the substrate hydrolysis. The difference between the evolved and the original enzyme resided in the shapes of the molecules (2NH versus paraoxon), their relative transition states, their binding orientation, and the atomic bond under attack (P—O versus C—O).

A smooth gain of 2NH activity was beautifully documented, and it was coordinated with a loss of PTE activity. The gain in secondary function left the original function largely intact. But only to a point. Halfway through the transition sequence, the magnitude of loss in original activity started to overtake the magnitude of gain in new activity. Whereas the functional gain per mutation was 4.4-fold in the initial rounds, in the final rounds it slipped to 0.8. Diminishing returns were at work. Still, a full exchange was seen. The loss of original activity was about 105 kcat/Km (or rate of product formation) and the gain about 104 kcat/Km. Selectivity changed by an astonishing nine orders of magnitude. Previous experiments had documented similarly large transitions, but they used specifically designed mutations and some even switched substrates through their trajectories.23 This experiment did not employ these heavy-handed techniques. It was truer to the parameters of natural evolution.

Even bigger changes were possible. The scaffold, or body, of an enzyme has an intricate geometrical structure. Enzyme families can have beautiful but remarkably different three-dimensional turns, twists, and shapes. It is one thing to wiggle a loop on the surface of a protein to promote activity. It is another to rewire its geometry. How can evolution generate large-scale structural (as opposed to functional) diversity?

Tawfik’s laboratory tackled the issue in three studies using the enzyme tachylectin-2, a protein that works in the immune response of certain species to bind specific molecules and clump cells.24 Its structure is symmetrical. Five protruding blades surround a core, and five sugar-binding sites are located between the blades. Tachylectin-2 is part of the beta-propeller group, which is characterized by blades in different numbers arranged symmetrically.

The structure of tachylectin contains an overt pattern. A popular, indeed prevailing, hypothesis is that large proteins must have evolved from smaller functional segments. These were fused and then shuffled by evolution. The process would leave a mark by virtue of the resulting internal symmetry. And this symmetry is characteristic of tachylectin-2, along with half of all known proteins.

Previous laboratory studies had demonstrated that small molecular modules could, indeed, be fused.25 They would form coherent shapes. Yet in all such cases, the fused molecules did nothing very useful. Tawfik and his colleague Itamar Yadid asked how small these units could be. His group sought to answer this question by cutting off the ends of a wild-type tachylectin, originally 236 amino acids in length, and screening to test how much it could be functionally reduced. They found a minimal working module, approximately 100 amino acids in length. The resulting paper, entitled “Reconstruction of Functional β-propeller Lectins via Homo-oligomeric Assembly of Shorter Fragments,”26 is a bit ambiguous on this point, but it appears that to achieve function,27 this fragment spontaneously bound four other identical modules, forming a pentameric structure through non-covalent interactions. For the first time, a function—in this case, sugar binding—remained intact through fusion. One of the most intriguing discoveries came when the structure of this pentamer was solved, revealing a dramatic degree of pliancy. Of the five conjoined fragments, two pairs on either side had fused; the remaining fragment formed a malleable bridge. The spine of this module unexpectedly twisted into a new formation; yet in context, the entire molecule retained its activity.

Large structural elements could move, and even rearrange themselves. This was a spectacular, if retrospective, corroboration of evolutionary principles. Could these principles be authentically predictive? In addressing this question, Tawfik’s group included Sergio Peisajovich and Liat Rockah. “New protein folds have emerged throughout evolution,” the team asserted, “but it remains unclear how a protein fold can evolve while [emphasis added] maintaining its function, particularly when fold changes require several sequential gene rearrangements.”28 That small fragments may fuse and diversify, as Tawfik and Yadid noted, was a provocative idea. In the lab, it worked under certain conditions. But could it work in the wild? One version was called the permutation-by-duplication model. A gene duplicates and is then fused in its chromosomal frame, adjacent to its parent. Thereafter, mutations degrade either end of the doubled sequence, forming a protein with a new context for shape. Diversity in several gene families is conjectured to have arisen through permutation by duplication. Researchers have been hesitant to accept this notion, because some thought such sloppy cut-and-paste mutational processes would ruin the sensitive protein by exposing the wrong residues.

Tawfik’s team determined to test this.29 They used DNA methyltransferase, a protein family whose diversity is thought to have arisen by such means. This family is segregated into seven groups that differ in the linear order of nine sequence motifs. The team created two libraries of fused duplicates. In one, they truncated the head of the first sequence; in the other, they truncated the tail of the last. This simulated intermediate steps in complex evolutionary rearrangements. To generate natural sequence diversities, evolution would have to move one step at a time through something like this series of changes, given duplication and fusion. As part of the protocol, after one end of the sequence was cut, the other followed.

The results were remarkable. In terms of mutations, early stops and late starts appeared at various places throughout the sequence frame, yielding a variety of intermediates, and in the next step, full permutants whose motifs had a distinct linear order. Significantly, several intermediates had function; the permutants resembled three known natural families of DNA methyltransferase. Evolution had been reconstructed. Some laboratory variants, however, did not match any known family, forming instead their own curious class with a distinctive feature: a part of one of the motifs was, by the knife of mutation, split on either end of the sequence. The team utilized this feature in a homology search. As it happened, this led to the discovery of a previously unknown natural class of methyltransferases in bacterial genomes. Under artificial laboratory conditions, evolutionary theory could be used successfully to predict natural forms.

Trade-offs

That things can change, and change significantly, was already known. What does it take to evolve? With respect to this question, life faces a peculiar conflict of interest. Life must be able to adapt. Proteins must be keen to maximize the utility of good mutations. At the same time, proteins should be robust, in the sense that they suppress and minimize the harmful effect of bad mutations. The problem is that most beneficial mutations are also harmful. Gaining function often comes at the expense of destabilizing protein structure, causing misfolding and aggregation. This means that even if a good mutation speeds up a reaction, and even if there are favorable trade-offs with the original function, protein fitness could nevertheless decline, especially with a steady accumulation of destabilizing changes. Should cells follow a strategy of suppressing mutational effects at the risk of becoming an evolutionary dead end? Or, should they allow their chemistry to be sensitive to them, but risk being perpetually dragged down under the weight of inevitable side effects? In either case, selection in the end could become confused.

Tawfik and a talented researcher and student, Nobuhiko Tokuriki, explored this issue.30 They observed what was already familiar in the literature: just as some mutations decreased stability, others could increase it. Good changes could compensate for bad ones. Inevitably, evolution would be slow and adaptive pathways more limited. But it is much better than the alternative. Without compensating margins of stability, the accumulation of a mere five mutations could drop fitness by twenty percent. Tawfik and Tokuriki saw the obvious. If proteins were, by mutation, becoming unstable and therefore misfolded, there were molecules in the cell to help that: chaperones, otherwise known as heat shock proteins, specialize in helping wayward proteins pack tightly and fold correctly. If these were overexpressed, then more beneficial but destabilizing mutations could be tolerated; there would be more chaperones to pack them in a tolerable way. The Tawfik laboratory demonstrated this experimentally. A chaperone complex called GroEL/ES was overexpressed during rounds of purifying selection. With an increased concentration of the chaperone, twice the number of neutral mutations could accumulate as without. The extra mutations would have been disruptive if not carefully packed. Yet these disruptive (and now tolerated) changes were also much more likely to offer significant benefits, imparting more than ten times the activity of mutations appearing without overexpression.

If compensating mutations and increased chaperone concentrations increased the rate of evolution, it was natural to ask whether still other structural features affected its pace, path, and mode. Sometimes the selective effects of individual mutations do not quite add as one might expect them to. The whole can be more, or less, than the sum of its parts. Such is epistasis. It may work to increase or speed adaptations in some instances, but to slow them down to a virtual halt in others. Tawfik’s team found that, in cases of antibiotic resistance in the protein TEM-1 beta-lactamase, this phenomenon was quite pervasive.31 There were a limited number of mutational constellations that could solve a given adaptive problem or crisis; whether certain mutations were good or bad in these constellations depended not on their intrinsic chemical makeup, but on their genetic background or context. The very same chemical change could be helpful or harmful depending on its position in a mutational sequence. Adaptive trajectories may, to a significant degree, be determined by the initial mutations in a series.

They set the context for everything else.

Touch of the Maverick

When speaking professionally, Dan S. Tawfik is clear, humble, patient, engaging, and prudent. In the professional literature, he goes after big game, or what comes to the same thing, big names. Consider his criticisms of Susumu Ohno’s model of duplication and divergence.32 Ohno argues that adaptive trade-offs are so strong, and mutational entanglements so dense, that evolution can’t plausibly go anywhere while natural selection is turned on. The solution is obviously to turn it off. If there are two genes, one may continue the function of the original, and so satisfy a type of stable pruning selection. The copy, or duplicate, is free to drift. So long as its parent has assumed needed physiological functions, the daughter gene may try various mutations. Many won’t work, and so dead genes may accumulate in the genome, but some will, and in the appropriate environmental context, selection will find some useful feature that has emerged from this random walk.

The Ohno model enjoys widespread acceptance, but as Tawfik noted, it is not the whole story, and it is likely not the major story or summary, either. The unfavorable trade-offs that Ohno assumes may not be as imposing as once thought. Tawfik’s laboratory had already found that new functions could arise without old ones being discarded. Chaperones and compensating mutations add stability to the process. There are more potent reasons, Tawfik argued, for rejecting Ohno’s model:33

  1. It appears natural selection is, in fact, aware of most duplicate genes, and is involved in preserving them as they accumulate mutations.
  2. Functionless, redundant gene copies cost the cell energy and resources. If they provide no immediate benefit, selection will provide no immediate free pass.
  3. Duplication-and-divergence assumes duplication is neutral. On the contrary. Duplication can function to boost the total activity of a gene by increasing protein copies, each with activity levels that add up. This level of increase may be enforced by positive selection, but not drift.
  4. Perhaps most crucially, thirty-three to forty percent of all mutations are deleterious, whereas only a tenth of a percent create new functions. Where there is no selection, deleterious mutations would quickly and irreversibly destabilize a gene. The birth of new, unspoiled proteins with impeccable function would not just be rare, but exceedingly rare.
  5. A duplicate gene may experience gain of function at a rate of 10-9 per site, yet if an adaptive trajectory is dominated by epistasis, and two independent mutations must co-occur to confer a new phenotype, the probability is the square of this frequency, or 10-18. This, Tawfik’s references, Jonathan Gressel and Avraham Levy, observe, is “probably in a range where the odds for such mutations in any mammalian species are almost nil.”34

Collectively, these observations comprise a powerful argument against the Ohno model. The alternative favored by Tawfik is one of multifunctional intermediate links. Divergence is happening, in this case, before duplication has anything to do with it. Every step is smooth, incremental, and functional, just as Maynard Smith had imagined.

A Hitch, and an Itch

Tawfik’s work animates the sketch drawn by Maynard Smith with unforgettable vitality. The unfolding drama is startling:

  • Proteins have alternate functions;
  • the alternates are of specific chemistry;
  • they are, at times, mediated by drastic structural changes, once thought impossible;
  • they can evolve coherently;
  • multiple scenarios of evolution (neutral and directional) produce useful changes;
  • strategies of buffering and packing allow the bad qualities of changes to be offset, and the good to be utilized;
  • function can be established by uniting small elements of structure;
  • the small elements themselves are structurally malleable;
  • a small evolutionary network has for the first time been rigorously characterized;
  • a clean transition through this network has been noted; and finally
  • laboratory evolution has predicted natural forms.

What is left for Maynard Smith but total vindication?

However, there is among these results an itch, a feeling of something unsettled. If Salisbury’s argument has been rebutted, it has not been expunged. Salisbury’s paper raises several questions, but it is most concerned with the origin of the first protein. Bound up with this are all the urgent issues: the origin of a first pathway, a first apparatus to produce the proteins genetically, a first system in general. For all anyone knew in the late 1960s and early 1970s, Salisbury may have been correct, if not in detail, then in essence. The specificity of proteins and their evolution are under tension.

Tawfik admits the issue of a first protein is “a complete mystery”35 because it reveals a paradox: enzymatic function depends upon the well-defined, three-dimensional structure of a protein scaffold, yet the 3D structure is too complex, too intricate, and too coordinated to arise without simpler precursors and intermediates.36 For nature to have favored further evolution, the precursors would have needed a function, but they had no capacity to acquire any function because they were too simple to form valid structures. If in tracing life’s history backwards, evolution proceeded step by step, at the first step it had to take a precarious and improbable leap. It is “an unlikely scenario,” writes Tawfik, that “random sampling must simultaneously produce a sequence with a discrete 3D structure and a selectable function.”37 The problem itself becomes a vicious circle. To begin evolution, you must have already evolved.

There are ideas that may serve to break this circle. Protein sequences, recall, do not have to be exhausted by coiling up into one shape. Sometimes the same sequence can coil up into different structures and geometries. If taken to conceptual extremes, the concept may find some use. Consider the large class of modern proteins called IDPs, or Intrinsically Disordered Proteins. By itself, an IDP is disorganized, with little order in its chain. Yet upon binding a particular chemical, a folding pattern is triggered, and the IDP becomes well-organized. Suppose this were the case for the first protein. If so, a nice way out. Viable protein shapes, perhaps twenty or thirty amino acids in length, floated free of physiological constraint. They had no durable intrinsic structure; the more unstructured they were, the more leeway their chemical bonds had to contort into new and various shapes. If many sequences were available, this would intuitively seem to increase the odds of finding some workable structure. Finally, upon meeting the right ligand, some of the shifting conformations fit well, and with binding, a folding pattern was triggered. The right shape emerged spontaneously.38

While superficially plausible, there are significant problems with this idea. Salisbury was asking a specific question: How were those first proteins built? He meant, of course, the first necessary proteins. IDPs have their own special sequence properties.39 It is not clear that appealing to them is appealing to a generalizable biological characteristic: a way in which necessary proteins appeared and then functioned collectively. Consider the vital functions of life: nucleotide metabolism, nucleotide synthesis, translation, glycolysis, error-correction, or the Krebs or Calvin cycles. These systems utilize cascades and cycles of proteins. Drastic disorganization and malleability of shape is not the key feature of these systems. Where IDPs do appear in the vital and most basic systems—ribosomes, for example—the parts that fluctuate are under the control of their sequence identities. These identities are, in the case of the ribosome, constrained, conserved, and ordered, more so, even, than regular, non-IDP counterparts.40 Control and precision are not evaded: disorder gains purchase by order at a more basic level.

Specific, improbable sequences—such as those imagined by Salisbury—should then be expected, whether an obliging protein shape is more movable or less. Moreover, it is unlikely, if the first proteins were indeed like IDPs, that they would be reliable enough to perform their tasks. IDPs are implicated in many diseases because of their tendency to aggregate, misfold, and generally cause problems.41 It is difficult to see how an emerging system, dominated by disordered proteins and with only rudimentary control and modification apparatuses, could manage very much for very long.

This issue has other faces, some of them numerical. Salisbury raised his objection based upon a calculation, however rough. Scientists knew that the space of possibilities was immense, but no one really knew how rare or common anything was. What was known was only that, generally speaking, good proteins are rare. The state of uncertainty has for the most part remained unchanged, but there are now interesting hints. Experimental results concerning the rarity of proteins range greatly. Differences depend upon select example systems, and some extrapolation is involved. Nonetheless, one of the most favorable and liberal estimates is by Jack Szostak: 1 in 1011.42 He ascertained this figure by looking to see how random sequences—about eighty amino acids in length, long enough to fold—could cling to the biologically crucial molecule adenosine triphosphate, or ATP.

At first glance, this is an improvement over Salisbury’s calculations by 489 powers of ten. But while an issue has been addressed, the problem has only been deferred. Despite persistent hopes to the contrary, and despite vague popularizations, it is completely unthinkable that the operations of a cell could be crammed into a single molecule. No one protein is superb enough to control its own synthesis, metabolism, protection, and translation. The best experiments will inevitably have an experimenter standing in for the missing operations; feeding the cells energized molecules; and providing a controlled environment of salts, pH, metals, and ions; these procedures are governed by expression and sequencing techniques and perfect selection. Only with this in place is it observed that an emergent protein can tether or bind a molecule. A cell of hundreds of proteins has been profoundly assumed. If a primordial protein bound ATP, or some other valuable small molecule, so what? It could not reproduce a sequence; it could not act reflexively on its own sequence to control heredity; it could not protect itself; it could not energize itself. It would be destined to vanish, either by decaying or succumbing to various unhelpful linkages or bond-breaking molecules.

Life needs more than one molecule. Salisbury’s probabilities now accumulate with force. If a pathway is coherent, not just any binding sequence will do. Particular reactions, in a particular sequence, must be localized and coordinated. Szostak’s experiment had not measured the probability of a particular sequence arising. He had measured the frequency, or probability, with which a particular function arises. The probabilities multiply if a life-necessary pathway, say nucleotide synthesis, requires several steps. If five enzyme functions were needed (ten are needed in modern adenine synthesis),43 then the probability would be 1 in (1011)5, or 1 in 1055. If all the operations needed for a small autonomous biology were ten functions—this is before evolution can even start to help—the probability is 1 in (1011)10, or 1 in 10110. This is more than the number of seconds since the Big Bang, more protons than there are in the universe. In considering a similar figure derived in a different context, Tawfik concedes that if true, this would make “the emergence of sequences with function a highly improbable event, despite considerable redundancy (many sequences giving the same structure and function).”44 In other words, these odds are impossible.

And it gets worse. The figure of 1 in 1011 is observed in studies of protein molecules that can cling to ATP. But ATP production itself is not a spontaneous act; it is controlled by many enzymes, a notable one being phosphofructokinase. This enzyme has the ability to bind ATP and use it to catalyze a specific transfer reaction, while simultaneously being sensitive to environmental cues for its own regulation. Szostak procured molecules to bind ATP, but without coordinating that binding (and the energy it affords) to something useful. No ability for regulation—switching on or off at crucial times—was reported. It would be much harder, more rare and improbable, to get a molecule that could not only bind the biologically necessary ATP, but use it effectively or act to produce it. Even though Szostak’s molecules can bind ATP, they only do so unreliably. It is questionable whether prebiotic molecules like these could be physiologically relevant. After eight rounds of selection, evolved specimens were sampled and then cloned. Only 10 percent of identical sequences worked. The vast majority of clones could not, most likely because their conformations were too destabilized and unwieldy. They could not coil effectively. One sequence could not be relied upon to fold. The pattern improved but was nonetheless preserved, even to the experimental end. The most robust specimens were seen after eighteen rounds of evolution; even then, at their very best, 60 percent of the copies did not work. No living cell could tolerate this failure rate, especially early cells with only primitive repair and control systems; reactions would be gummed up and physiology crippled.

Tawfik soberly recognizes the problem. The appearance of early protein families, he has remarked, is “something like close to a miracle.”45

If the argument Salisbury proposed can be restated, so can its rebuttal. If functioning proteins are rare when big, perhaps they may be common if small. Salisbury had only given an intuitive argument against this. It was by no means rigorous.

Some studies do, indeed, show that some elements of structure can be formed from smaller combinations of amino acids: as few as ten, in one case.46 But such studies, while persuasive to protein engineers, are an irrelevance to evolutionary biologists. These structures are transient, and most wash away. Moreover, they have no demonstrable catalysis.47 Tawfik acknowledges this point. In his study of tachylectin, he could not reduce the functional structure from 100 amino acids to, say, 40 or 50 amino acids. What he could do was reduce the original protein to a large, 100-amino-acid fragment. Even then, however, that one, single, 100-amino-acid fragment was not enough: It bound to several cloned modules of itself, perhaps to establish a coherent geometrical frame.48 This reduction in size, from the original protein to the fragment, also caused major problems: more than 90 percent of the sequences aggregated and misfolded. In any working cell, primordial or not, this might mean extinction.

Brian Matthews et al. examined this issue in an important paper published in 1994, noting that the smallest known natural enzyme is composed of eighty amino acids.49 This is not small. The function afforded by this smallest enzyme is for restricted modes of catalysis. For autonomous life functions, it may well need to be bigger. Bigger enzymes, like bigger words, are more infrequent and improbable. Hence the force of Salisbury’s argument. Matthews observed that an enzyme needs to be big to have an adequate framework to substantially accelerate reactions, catalytic groups to forge bonds, a binding site, and structural elements for context. Those having fewer than forty amino acids tend to have too many degrees of freedom to form a well-ordered fold and too little folding energy to maintain structure.

So the paradox still exists: evolution cannot occur without life; life cannot occur without enzymes. The first enzymes had to achieve their complexity without help, either from other cells or from a guiding evolutionary process.

Various workarounds have been suggested, most notably, the RNA world hypothesis. This idea is conspicuous by its absence in Tawfik’s writings. The RNA world hypothesis assumes that if proteins faced mounting difficulties (such as size and hereditary control), perhaps another molecule, RNA, might do better. Because it can fold into complex shapes, RNA is the only cellular molecule beyond the proteins themselves that can carry out anything like catalysis.50 Evolution could conceivably have traveled from nucleic acids to proteins.

Many origin-of-life theorists conjecture that RNA came before protein. An era of RNA (or ribozyme) evolution served as an elaborate scaffold to bring proteins into existence. In a conjectured ocean, ribozymes floated freely; some were resourceful enough to use single or double amino acids for their own purposes, perhaps to increase chemical diversity, or to mark important points of replication, or to help the RNA fold. Longer amino acid chains meant more diversity. Various chemical affinities could be in play. In the end, natural selection began to favor the function of the proteins more than the ribozymes connected to them. Life emerged.

This hypothesis Tawfik does not assume. He seems to imagine a classic approach: a free-floating protein exists in a disorganized state of alternating conformations, one of them useful to a mysterious evolving biology. The protein is from a random repertoire, apparently unconnected to and unconcerned with RNA function. Though a portion of the technical community demurs, an implicit disregard for the RNA world scenario is surely rational. NASA’s Andrew Pohorille and Michael New supply three reasons to doubt an RNA world:51

  1. Protein building blocks are more easily formed than RNA building blocks, which are notoriously difficult to form.
  2. RNA cannot easily achieve the crucial aspects of metabolism, such as energy capture and transport. This may have been necessary from the beginning.52 No ribozymes have ever been observed to do something as crucial as establishing a proton gradient to produce ATP energy or synthesizing nucleotides for new RNA strands.53
  3. An RNA world would seem to offer a great deal of confusion to natural selection. “[S]ince there is no relationship,” Pohorille and New write, “between the function of a catalytic RNA and the function, if any, of the protein for which it can code, there is no clear path from the ‘RNA world’ to today’s world of protein catalysis and nucleic acid information storage.”54

Further serious problems remain. RNA enzymes emerge at about the frequency suggested by Szostak for proteins, so no relief of probabilities is offered. Additionally, RNA enzymes that break bonds or perform irrelevant tasks may more frequently emerge than the few that would do a cell much good. Any good gained could be lost in this mix. Finally, the backbone structure of RNA suggests that it was impossibly fragile in early watery conditions.55 There are many possible scenarios, some creative enough to avoid hurdles. It is a realm of speculation and uncertainty. It must not be forgotten that aspects of modern life remain elusive and mysterious as well: a third of proteins essential to life have unknown functions.56 If we cannot say with certainty what modern life needs, how can we judge speculations about the nature of the earliest life? As it stands, the appearance of the first proteins is an enigma.

Generalist to Specialist

Throughout his work, Dan S. Tawfik appeals to a universal biological motion governing the concept of an evolutionary network: the movement from generalist to specialist, and then back again. The specialists are proteins that do one job very well. The generalists are structures serving to connect specialist end points in an evolutionary sequence.

The motion is cyclical. Proteins, or at least some of them, have a main function plus several extra functions. With mutations, these can change prominence. In changing roles, the specificities must first untie before they can retie in a different way. With respect to catalysis, a protein under evolution is a great master of one trade, then a jack of several trades, then a master once again.

The alluring idea of something for nothing works, at best, for a while, but cannot last. Eventually, it fizzles out. The relationship demands that something is gained only if something else is lost. The loss in biology is a zone of transition in which a link in an evolutionary chain is just a little too promiscuous for its own good. It can catalyze the original reaction, a new reaction, or other unwanted or even catastrophic reactions.

This observation is a substantial drawback for the entire notion of promiscuous reactions and generalist transition points. In the cell, proteins exist with multiple potentials. But these are under delicate control. Timing, expression, and regulation have everything to do with viability. A sensitive apparatus is in place; without it, adaptive potential wanes.57 With this in mind, it is worthwhile to recall the experiment by Rigby et al. The researchers, in contrast to other early experimenters, considered a hard evolutionary problem, one involving multiple mutations and xylitol specificity. They observed that cells, instead of refining an ability, crashed their own regulation systems and allowed a critical functional threshold to be reached merely by multiplying the number of promiscuous enzymes operating at low-grade levels. Multiplying promiscuous enzymes without regulation solves one problem only by creating another. “In a living cell,” Tawfik writes, “the toll [emphasis added] of a generalist on fitness might be too high, and the driving force for specialization is likely to be stronger than observed in vitro.”58 Tawfik provides a brief but interesting remark about the looming aspect of undesirable promiscuous interactions. “[A]lthough increased protein doses can make a weak, promiscuous activity come into action and thereby provide an evolutionary starting point, these increased doses may also become deleterious [emphasis added] owing to the very same effect.”59

When one arrives at the origin of life, the matter gets complicated beyond belief. At that point, there is no resilient context of the cell to buffer certain bad effects. The earliest and most rudimentary enzymes must have been the most general in evolutionary history. Their proportion in the primordial cells would have been high. In an unchecked, unregulated early cell, where promiscuous reactions would be rampant, how could a cell survive? How would early, simple cells have gotten by without a level of quality control?

Distinction and Scruples

To the extent that Tawfik’s selection experiments were successful, it is because mutations were localized and contextualized. Mutation had a key but confined role. If evolution proceeded, the prevailing architecture of the active sites and protein shapes nonetheless remains intact. Changes were not to central structures, but to peripheral loops. A great deal of flexibility was discovered. Still, it is hard to see how any of this could build proteins—that is, in the sense of building their fundamental shapes, or scaffolds; and build proteins in terms of explaining the key catalytic strategies of each active site. Even in the impressive demonstration of a transition through nine orders of magnitude, in which a full exchange of a promiscuous activity for the primary activity was seen, the overall geometry of the protein was unchanged, and, although substrates had changed, the fundamental active site strategy stayed the same. It is much easier to explain how this process allows diversification within protein folds once the folds themselves are established than to explain how the folds were first established. “Modern neo-Darwinism and neutral evolutionary treatments,” remark Leonard Bogarad and Michael Deem, “fail to explain satisfactorily the generation of the diversity of life found on our planet.” It is not that they did not evolve, they say, but that “... most theoretical treatments of evolution consider only the limited point-mutation events that form the basis of these theories.” Their sober conclusion is that “point mutation alone is incapable of evolving systems with substantially new protein folds.”60

Since point mutation is responsible for the novelties seen in several of the Tawfik team’s experiments, these novelties become of questionable relevance to the issue of universal origins. The central question remains. With this, Tawfik would seem to agree: “There is little doubt ... that major switches in function demand [emphasis added] major sequence rearrangements” or larger changes, “including insertion, deletion and recombinations.”61 Yet experiments showing the relevant larger changes, ones involving “insertion and/or deletion of entire polypeptide segments,” Tawfik adds, have been “scarcely exercised.”62 By “exercised,” Tawfik means requisite changes are not often seen; they are rare. In other words, there aren’t all that many experiments documenting necessary big changes to active sites. This must be so, because the changes they demand are bold and difficult alterations of chemistry.

Generally speaking, the relatively scarce experiments that do succeed in large-scale rearrangements do so by non-Darwinian means, or else by means of dubious relevance to natural evolution. Tawfik describes an experiment by Hee-Sung Park et al.63 Entire loops on enzymes were exchanged. The trade depends on residues that anchor the loops, permitting the switch. Yet all the transitions in Park’s study went through stages of completely abolished function. While good for laboratory purposes, this would make any real-world Darwinian progression prohibitive. A similar phenomenon was seen in some of Tawfik’s best, most provocative studies. Consider the experiments involving drastic structural plasticity, such as those studying tachylectin. When the initial enzymes were truncated to test activity in fragments of reduced size, there was marked loss of activity and stability. In other words, 92 percent of the library lost function.64 There was instead a high tendency for sequences to clump and do nothing of value. The problem was in the sequence itself. It could not fold well. This condition was improved later through generous selective protocols. But the wild world does not have generous selective protocols. In an evolutionary sequence, if early stages were ever uncontrolled and folding unreliable, it is not easy to see how cells could have survived long enough to evolve a solution.

The principle that great transitions work only under non-Darwinian (or non-Smithian) protocols reappears in the above-mentioned study of permutation by duplication. Despite the shocking predictive power of the experiment, there were sobering facts. Change was, again, contextualized and localized. Clusters of mutations occurred at restricted places in sequence, none of them in control of the monumental aspects of structure and chemistry. This limits adaptive options. Additionally, natural selection proved thirty times more likely to halt new evolution and revert to the original sequence than allow an evolutionary rearrangement in motifs.65 Apparently, the original sequence was much fitter than most intermediates. If selection operates like this for most proteins in the wild, it is difficult to imagine how new mutants could have taken over a population. For evolution to proceed, selection had to be relaxed. Yet if constraints are stringent and selection pressures relaxed for this type of change, what would drive greater challenges? If natural selection must be removed here, it must be removed when considering the difficulties involved in accounting for the origin of life, folding patterns, and active site architecture. What then is left in the field of explanation besides raw chance? Old and forbidden concepts now reappear. It is in no sense obvious that the simple mechanisms underlying diversification within protein superfamilies are adequate to account for the major differences of form and active strategy between superfamilies. Just how did those differences get there? Writing in Structure in June 2014, Tawfik and Ágnes Tóth-Petróczy ask, “[C]an small gradual changes explain major evolutionary transitions and, foremost, the gaps [emphasis added] between micro and macroevolution?”66

Maynard Smith and loyalists of the Old Guard would answer yes. Tawfik and Tóth-Petróczy’s answer seems to be no. They appeal instead to what Maynard Smith instinctively disallowed: macromutations, “hopeful monsters,” rare or improbable changes of form, and daring leaps in fitness.

If a gain of some sort is apparent in the new perspective, it resides largely in speculation, not empirics. As previously emphasized, few experiments document anything like that scale of viable change. Bigger changes lead to more catastrophic effects. Most will be unusable. In those experiments that are successful, evolutionary relevance is always in question. Where drastic changes are actually tolerated, as seen recently in green fluorescent protein,67 they nonetheless do not control fundamental form or active strategy.68 “In fact, to our knowledge,” Tawfik and Tóth-Petróczy write, “no macromutations ... that gave birth to novel proteins have yet been identified.”69

The emerging picture, once luminous, has settled to gray. It is not clear how natural selection can operate in the origin of folds or active site architecture. It is equally unclear how either micromutations or macromutations could repeatedly and reliably lead to large evolutionary transitions. What remains is a deep, tantalizing, perhaps immovable mystery.

Behind the Veil

Perhaps more than any current researchers, Tawfik and his laboratory have advanced experiments against the four questions of enduring importance articulated by Maynard Smith. The experiments Tawfik has undertaken are remarkable demonstrations; they are more resourceful, more thorough, and more solid than almost any in his field, either currently or historically. Yet they seem to revolve around a certain nexus, a void in knowledge, an unavoidable place of wilderness. That a protein space—the space of every combinatorial possibility—may be somehow navigated so that sure answers can be gleaned from deep, well-posed questions, like those of Maynard Smith, may in fact be impossible. Characterizing the most important properties of a global protein evolutionary network may be, now and perhaps forever, beyond our grasp. This is so if only because the network is too vast and there are too many exceptions, caveats, and confusions.

In the meantime, short-range questions have urgency. These can be addressed. For the moment, this is the preoccupation of evolutionary protein science. If smaller and perhaps less rich questions do not give final answers, they do provide a magnificent corpus of tantalizing and, at times, frustrating hints.

Publications Guide

Slow Protein Evolutionary Rates are Dictated by Surface-Core Association

Proceedings of the National Academy of Sciences of the United States of America (July 2011)70

A paper concerned with the statistical analysis of a large protein dataset. Proteins are composed of distinct 3D regions: the surface, the core, and the active site. Amino acids at each position in each region face different functional demands, and therefore have a different sensitivity to change. The experimenters ask: what determines the rate of change? They answer: expression levels, functional essentiality (both of the protein, and each position’s contribution to overall function), the number and nature of binding interfaces, the number of protein binding partners, and the need to maintain structural stability.

Stability Effects of Mutations and Protein Evolvability

Current Opinion in Structural Biology (2009)71

This review describes recent changes in our understanding of protein mutations. Until around ten years ago, it was thought that proteins could tolerate many mutations at no cost. This is now known to be false: a significant fraction of mutations are deleterious. They can be buffered for a short time by an excess of stability, but this reserve dwindles as mutations accumulate. As a result, fitness declines exponentially. Mutations, even if they afford new functions, can actually reduce fitness because they impose disruptive effects on protein structure, which can lead to misfolding and aggregation (which, in turn, leads to diseases, reduced fitness, and negative complications).

The authors propose that the order and nature of arriving mutations would probably, then, be constrained. Evolution will be slower and much more difficult if compensating mutations arrive after the stability of a protein has already been compromised. Ideally, compensation will arrive before the critical activity-altering mutations. This would increase stability margins so that further change can be selectively favored. This is the easiest evolutionary path. However, it is not discussed how such mutations, if arriving first, could selectively sweep a population, since they seem to be useful only in context of future advantages. The authors hypothesize that compensating mutations could originate and be maintained in a population as part of neutral drift; later, the scenario goes, the two mutations could meet in the same organism, creating a selective advantage.

Compensating mutations appearing in lab studies have an interesting character: they resemble the phylogenetically-derived ancestral sequence of the protein family to which they belong. If the purported ancestor were a more inclusive (but less refined) representative of various features of the whole family, then diversification within the family might be owed to a pattern of generalization and respecification. Finally, the authors propose that the action of chaperone proteins may help pack deleterious mutations in such a way as to make them tolerable, while exploiting any favorable properties.

The ‘Evolvability’ of Promiscuous Protein Functions

Nature Genetics (January 2005)72

Proteins can have, in addition to their primary function, other alternative, weaker, secondary functions. In a changing environmental context, these could be useful. The team tested how the rates of the primary and secondary activities trade-off with one another. The regime was mutation and selection (the team did not measure the stability, folding, and aggregation of the protein, just raw rates of conversion to product). The experimental system was devised around three proteins: CAII, PTE, and PON1. It was found that, on average, mutational change could increase the secondary activity at little expense to the primary activity. The ratio was 10-fold gain to 1-fold loss. It became possible to imagine evolution proceeding through multispecific intermediate proteins. The experiment did not account for the protein folding and stabilization effects of mutations. The paper also asked whether evolvability is itself an evolvable trait.

Latent Evolutionary Potentials Under the Neutral Mutational Drift of an Enzyme

HFSP Journal (May 2007)73

Tawfik’s team tested the protein PON1 under purifying selection, measuring the activity parameter for both primary and secondary functions, as well as expression levels. The team did this at each stage for three rounds of mutation and selection, rigorously characterizing a small neutral network. The experiment is also notable because the library size of the protein variants was small (311 variants). Despite the small size of the library, about half the variants accumulated mutations that increased a secondary function. This supports the hypothesis that a population may acquire mutations that permit generalist, multispecific evolutionary intermediates.

Diminishing Returns and Tradeoffs Constrain the Laboratory Optimization of an Enzyme

Nature Communications (December 2012)74

In this experiment, an old primary function was decreased by about 4.5 orders of magnitude, and the new function increased by 4.5 orders of magnitude, for a total nine orders of magnitude change (109). The effect of weak (favorable) trade-offs was shown to be real, but only in the initial stages. As evolution progressed, trade-offs between the old and new function became more severe. The experiment demonstrated the reality of a multifunctional intermediate. Still, the derived enzyme had the same catalytic strategy as its ancestor, although the particular atoms of the substrate are different and the geometries of the target molecules are distinct. Moreover, to get such a large span (109) of total activity change, the team used techniques to push evolution along. For example, at later stages in the experiment the selection threshold was relaxed. Furthermore, at various stages the team adjusted both temperature, to increase the number of mutations, and chaperone expression, to help proteins tolerate additional mutations. Relative fitness levels did not determine the fixation rate of mutants. Mutations after round five of this experiment provided relatively small gains; in the wild, they may never sweep a population. The team was able to provide a proof of concept experimentally. While this can be done in a laboratory, as the authors acknowledge, it is difficult to switch off a protein’s old function and optimize its new one; late mutations don’t make enough of a difference to survival for natural selection to promote them across an entire population. Positive mutational effects exhibit diminishing returns: the best mutations carrying the biggest impact come early, and later mutations (which help stabilize and reinforce earlier ones) have less advantage or fitness than those that came before.

Reconstruction of Functional β-Propeller Lectins via Homo-oligomeric Assembly of Shorter Fragments

Journal of Molecular Biology (January 2007)75

An extremely important paper investigating the question of how the earliest proteins appeared and functioned. One idea is that small, structural, working pieces fused or bound together chemically, and from this fusion a working protein formed. This experiment was the first demonstration that small segments, fused to form a stable structure, also retained function in a laboratory setting. The Tawfik team took a protein, truncated it to get the smallest functional fragments, and observed that these fragments existed in a state in which they were bound to each other in clusters of five (meaning they were pentameric).

Enzyme Promiscuity: A Mechanistic and Evolutionary Perspective

Annual Review of Biochemistry (March 2010)76

One of the most important survey papers from Tawfik’s laboratory. It contains an up-to-date review of research about enzyme promiscuity, or the ability of (many) proteins to catalyze alternative reactions in addition to their primary reaction.

Metamorphic Proteins Mediate Evolutionary Transitions of Structure

Proceedings of the National Academy of Sciences of the United States of America (April 2010)77

Protein evolution is subject to a general constraint: significant structural change during evolution will perturb the protein core, causing fitness decline. This effectively prohibits positive evolution. If the structural scaffold, or overarching framework, of the protein evolved, proteins would have to be able to assume more than one 3D structure. This experiment worked with the protein tachylectin. Independent fragments, though identical in sequence, in concert with each other adopted two or even three separate shapes. This change only occurred in bridging subunits. The evolved pentameric protein resembles the type found in the wild: Two small fragments interact to form the same basic shape as the original, and this on two sides; the fifth fragment forms a bridge between them. The bridge has a different structure, even though it is the same sequence. This type of flexibility is important in imagining early transitional protein structures for the evolution not merely of new functions, but of entirely new protein structural scaffolds and architectures.

Functional β-Propeller Lectins by Tandem Duplications of Repetitive Units

Protein Engineering, Design & Selection (January 2011)78

This experiment is similar to others that have been conducted concerning the assembly and fusion of small modules to form a working protein. The model protein was again tachylectin-2. Instead of using the previously derived 100-amino-acid long fragments as the basic structural unit to form a coherent structure, this experiment used designer sequence elements based upon repeats in tachylectin’s natural structure: 47-amino-acid modules. The modules were reengineered to possess the same sequence core but diversified ends, which were more conducive for binding to form structure. The experiment did not attempt to demonstrate that any single module possesses sustainable, independent lectin function. When fused in groups of five, however, functional activity was exhibited. Though activity was exhibited in a minority of the protein library, as many as 99 percent of the modules formed, in the worst cases, “insoluble inclusion bodies,” or non-functional clumps. By targeting connecting regions for segment exchange (and leaving alone the critical sequence core), greater folding and solubility were seen, though still with perhaps 80 percent non-functionality (in the best case). Fusing between 40 and 60 percent of the original tachylectin structure to the diversified designer modules created the most robust protein specimens.

Evolution of New Protein Topologies Through Multistep Gene Rearrangements

Nature Genetics (February 2006)79

In this experiment, Tawfik and his team perform a remarkably thorough and well-conceived experiment concerning permutation by duplication. Some proteins appear as if their internal structural motifs have been rearranged. One hypothesis for the origin of this phenomenon is that after a gene duplicates, it may then be located adjacent to its parent gene on a chromosome. The ends of each respective gene are degraded by mutation; this results in fusion among genes. The new gene can display structural motifs that are unlike the duplicate or the parent. This experiment included the discovery of a natural class of DNA methyltransferase whose topological arrangements had previously been unknown.

Protein Dynamism and Evolvability

Science (April 2009)80

A review of the new view of protein evolution.

What Makes a Protein Fold Amendable to Functional Innovation? Fold Polarity and Stability Trade-offs

Journal of Molecular Biology (March 2013)81

An account of, or hypothesis about, the features of proteins that make them more or less subject to functional innovation by evolution. Proteins are subject to conflicting demands: a certain level of insensitivity to mutation is needed to buffer the effects of bad mutations, but at the same time, a level of sensitivity to mutation is also necessary to allow for the possibility of evolution itself. The way a protein balances stability and evolvability depends on its structural arrangement. If these features are assigned to different regions within the same protein, they are not in conflict. Thus if the scaffold or body of a protein is largely responsible for stability, then the scaffold can have many bonds mediating high stability. At the same time, if the active site is responsible for plasticity, and it is free to rotate, move, and wiggle, this makes for better adaptability.

Conformational Diversity and Protein Evolution – a 60-year-old Hypothesis Revisited

Trends in Biochemical Sciences (July 2003)82

This paper is notable for its section on the origin of the first proteins. The authors state that the problem is “a complete mystery” (and one with a chicken-egg aspect) in that for enzymatic function to be sustainable, a robust protein structure is needed; but for a complex protein structure to exist, it has to have evolved from prior functional precursors. The authors conjecture that early proteins coevolved properties of fold and function.

Do Viral Proteins Possess Unique Biophysical Features?

Trends in Biochemical Sciences (February 2009)83

Viral proteins have structures reflecting survival strategies unlike those of highly thermostable proteins (from eukaryotes and prokaryotes). For example, viral proteins are subject to higher mutation rates than prokaryotic or eukaryotic proteins. To buffer this fact of life, viral proteins seem to employ the strategy that having little means there is little to lose. They tend to have disordered segments and fewer bonding contacts per amino acid. These properties may reduce the destabilizing effect of mutations.

Mutational Effects and the Evolution of New Protein Functions

Nature Reviews Genetics (August 2010)84

According to its authors, the review “describes our current knowledge of the effects of mutations on the structural integrity and activity of proteins.” The paper includes a very good section about the untenability of the Ohno model of duplication and divergence.

TRINS: A Method for Gene Modification by Randomized Tandem Repeat Insertions

Protein Engineering, Design & Selection (June 2012)85

Typically, directed evolution experiments are limited in their methodology: diversity is principally generated from the technique of random mutagenesis, yielding a few mutations at a time. The authors developed a technique known as TRINS (Tandem Repeat Insertion). This method produces and inserts chunks of repetitive DNA throughout a gene, with length and frequency functioning as controlled parameters. Though it does not allow control over the location of insertion and does not permit truly random sequence insertions, it still comprises a model that could be useful for exploring sequence space.

Role of Chemistry Versus Substrate Binding in Recruiting Promiscuous Enzyme Functions

Biochemistry (April 2011)86

This paper explores the nature of promiscuous enzymatic reactions. Do promiscuous reactions represent an old chemical reaction performed on a new substrate, or a new catalytic reaction performed on an old substrate (i.e., something chemically similar to the original)? The experiment by its nature was limited, but the results indicated that “substrate ambiguity” (same reaction, new substrate) is four times more common than “catalytic ambiguity” (different reaction, chemically similar substrate).

Loop Grafting and the Origins of Enzyme Species

Science (January 2006)87

A short but important review of an experiment undertaken by Hee-Sung Park et al. in 2006.

Initial Mutations Direct Alternative Pathways of Protein Evolution

PLoS Genetics (March 2011)88

This paper was yet another study on the effects of epistasis, in this case, to determine the existence and coherence of multiple (as opposed to a single) adaptive pathways to a given adaptive endpoint. The researchers found that for the model they tested, there was more than one single adaptive pathway, but the fitness terrain was rugged. Which path is taken depends on the historical contingency of the first mutation in a series; thereafter, there is a “deterministic structure” to the pathways. The paper suggests that epistasis determines the fitness terrain and the contingency of trajectories in evolution.

Protein Engineers Turned Evolutionists

Nature Methods (December 2007)89

A short review of protein engineering in relation to evolution.

  1. Frank Salisbury, “Natural Selection and the Complexity of the Gene,” Nature 224, no. 5,217 (1969): 342–43. 
  2. Salisbury’s proposal takes for granted that a molecular system of translation (ribosomes, tRNAs, etc.) is operational at, or before, the origin of the first proteins. His figures can be re-expressed without this assumption. It is notable that autonomous life has never been observed without such a translation system. 
  3. This figure should not be taken very seriously except for the purposes of the thought experiment. This aspect of the situation could be worse. Full nucleotides are notoriously difficult to synthesize. They are hard or impossible to find in average or exotic conditions, on earth or in meteorites. Where certain facets of prebiotic nucleotide synthesis have proven possible, numerous unhelpful isomers, linkages, and alternative structures typically mar any simple, happy picture. There has been recent progress in the field, most notably in the Sutherland group’s experiments, but as Joyce and Robertson intimate:
    [T]he story remains incomplete because these syntheses [by Sutherland and others] still require temporally separated reactions using high concentrations of just the right reactants, and would be disrupted by the presence of other closely related compounds.
    Michael Robertson and Gerald Joyce, “The Origins of the RNA World,” Cold Spring Harbor Perspectives in Biology (2010), doi: 10.1101/cshperspect.a003608. 
  4. John Maynard Smith, “Natural Selection and the Concept of a Protein Space,” Nature 255, no. 5232 (1970): 563–64. 
  5. There are other mutational events besides exchange: additions, deletions, and frameshifts. Maynard Smith restricted his consideration of mutation to exchange because it was illustrative of small incremental change. He doubted the relevance of frameshifts and inversions to the origin of biological novelty: they interrupted the biological process too drastically to be routinely used by evolution. 
  6. Joan Betz et al., “Evolution in Action,” Nature 247, no. 5439 (1974): 261–64; Jane Brown, P.R. Brown, and Patricia Clarke, “Butyramide-utilizing Mutants of Pseudomonas aeruginosa 8602 which Produce an Amidase with Altered Substrate Specificity,” Microbiology 57, no. 2 (1969): 273–85. 
  7. Maurício Carneiro and Daniel Hartl, “Adaptive Landscapes and Protein Evolution,” Proceedings of the National Academy of Sciences of the United States of America 107, no. 1 (2010): 1,747–51. 
  8. “In this respect,” Maynard Smith observed, “functional proteins resemble four-letter words in the English language, rather than eight-letter words, for the latter form a series of small isolated islands in a sea of nonsense sequences.” In other words, it is assumed so because it must be so. In response to the question of how one may know what must be, he answered by allusion: “The justification for making these assumptions,” he wrote, “is that no sensible alternatives have been suggested and that no evidence exists at the moment to invalidate them.” John Maynard Smith, “Natural Selection and the Concept of a Protein Space,” Nature 255, no. 5232 (1970): 564. 
  9. Emil Fischer, “Einfluß der Konfiguration auf die Wirkung der Enzyme (Influence of Configuration on the Action of Enzymes),” Berichte der Deutschen chemischen Gesellschaft zu Berlin 27, no. 3 (1894): 2,985–93. 
  10. Daniel Koshland, “Application of a Theory of Enzyme Specificity to Protein Synthesis,” Proceedings of the National Academy of Sciences of the United States of America 44, no. 2 (1958): 98–104; Daniel Koshland, “The Key-Lock Theory and the Induced Fit Theory,” Angewandte Chemie International Edition in English 33, no. 23-24 (1994): 2,375–78. 
  11. Peter Rigby, Bruce Burleigh, and Brian Hartley, “Gene Duplication in Experimental Enzyme Evolution,” Nature 251, no. 5,472 (1974): 204. 
  12. John Maynard Smith, “Natural Selection and the Concept of a Protein Space,” Nature 255, no. 5,232 (1970): 564. 
  13. Peter Rigby, Bruce Burleigh, and Brian Hartley, “Gene Duplication in Experimental Enzyme Evolution,” Nature 251, no. 5,472 (1974): 203–204. 
  14. Roy Jensen, “Enzyme Recruitment in the Evolution of New Function,” Annual Review of Microbiology 30 (1976): 409–25. 
  15. John Maynard Smith, “Natural Selection and the Concept of a Protein Space,” Nature 255, no. 5,232 (1970): 564. 
  16. Leo James, Pietro Roversi, and Dan S. Tawfik, “Antibody Multispecificity Mediated by Conformational Diversity,” Science 299, no. 5,611 (2003): 1,362–67. 
  17. Sergio Peisajovich and Dan S. Tawfik, “Protein Engineers Turned Evolutionists,” Nature Methods 4, no. 12 (2007): 991–94. 
  18. Dan S. Tawfik et al., “The ‘Evolvability’ of Promiscuous Protein Functions,” Nature Genetics 37, no. 1 (2005): 73–76. 
  19. This sort of structural alteration is probably how antibiotics lose effectiveness: Bacterial proteins change to fit only their original substrate (if less efficiently) without fitting a fatal antibiotic. 
  20. Leo James and Dan S. Tawfik, “The Specificity of Cross-Reactivity: Promiscuous Antibody Binding Involves Specific Hydrogen Bonds Rather than Nonspecific Hydrophobic Stickiness,” Protein Science 12, no. 10 (2003): 2,183–93. 
  21. Gil Amitai, Rinkoo Devi Gupta, and Dan S. Tawfik, “Latent Evolutionary Potentials under the Neutral Mutational Drift of an Enzyme,” HFSP Journal 1, no. 1 (2007): 67–78. 
  22. Dan S. Tawfik et al., “Diminishing Returns and Tradeoffs Constrain the Laboratory Optimization of an Enzyme,” Nature Communications 3, no. 1,257 (2012), doi:10.1038/ncomms2246. 
  23. Paul Dalby, “Strategy and Success for the Directed Evolution of Enzymes,” Current Opinion in Structural Biology 4, no. 21 (2011): 473–80. See also Rudi Fasan et al., “Evolutionary History of a Specialized P450 Propane Monooxygenase,” Journal of Molecular Biology 383, no. 5 (2008): 1069–80. 
  24. Itamar Yadid and Dan S. Tawfik, “Reconstruction of Functional β-Propeller Lectins via Homo-oligomeric Assembly of Shorter Fragments,” Journal of Molecular Biology 365, no. 1 (2007): 10–17; Dan S. Tawfik et al., “Metamorphic Proteins Mediate Evolutionary Transitions of Structure,” Proceedings of the National Academy of Sciences of the United States of America 107, no. 16 (2010): 7,287–92; Itamar Yadid and Dan S. Tawfik, “Functional β-Propeller Lectins by Tandem Duplications of Repetitive Units,” Protein Engineering, Design & Selection 24, no. 1 (2011): 185–95. 
  25. Marc Beauregard et al., “Spectroscopic Investigation of Structure in Octarellin (a de novo Protein Designed to Adopt the α / β-Barrel Packing),” Protein Engineering 4, no. 7 (1991): 745–49; Birte Höcker, Jörg Claren, and Reinhard Sterner, “Mimicking Enzyme Evolution by Generating New (βα)8-Barrels from (βα)4-Half-barrels,” Proceedings of the National Academy of Sciences of the United States of America 101, no. 47 (2004) 16,448–53; Maryam Nikkhah et al., “Engineering of β-Propeller Protein Scaffolds by Multiple Gene Duplication and Fusion of an Idealized WD Repeat,” Biomolecular Engineering 23, no. 4 (2006): 185–94. 
  26. Itamar Yadid and Dan S. Tawfik, “Reconstruction of Functional β-Propeller Lectins via Homo-oligomeric Assembly of Shorter Fragments,” Journal of Molecular Biology 365, no. 1 (2007): 10–17. 
  27. See endnote 48
  28. Sergio G. Peisajovich, Liat Rockah, and Dan S. Tawfik, “Evolution of New Protein Topologies Through Multistep Gene Rearrangements,” Nature Genetics 38, no. 2 (2006): 168. 
  29. Sergio G. Peisajovich, Liat Rockah, and Dan S. Tawfik, “Evolution of New Protein Topologies Through Multistep Gene Rearrangements,” Nature Genetics 38, no. 2 (2006): 168–74. 
  30. Nobuhiko Tokuriki and Dan S. Tawfik, “Stability Effects of Mutations and Protein Evolvability,” Current Opinion in Structural Biology 19, no. 5 (2009): 596–604. 
  31. Dan S. Tawfik et al., “Initial Mutations Direct Alternative Pathways of Protein Evolution,” PLoS Genetics 7, no. 3 (2011), doi:10.1371/journal.pgen.1001321. 
  32. Shimon Bershtein and Dan S. Tawfik, “Ohno’s Model Revisited: Measuring the Frequency of Potentially Adaptive Mutations under Various Mutational Drifts,” Molecular Biology and Evolution 25, no. 11 (2008): 2,311–18. 
  33. Shimon Bershtein and Dan S. Tawfik, “Ohno’s Model Revisited: Measuring the Frequency of Potentially Adaptive Mutations under Various Mutational Drifts,” Molecular Biology and Evolution 25, no. 11 (2008): 2,311–18; Olga Khersonsky and Dan S. Tawfik, “Enzyme Promiscuity: A Mechanistic and Evolutionary Perspective,” Annual Review of Biochemistry 79 (2010): 471–505. 
  34. Jonathan Gressel and Avraham Levy, “Agriculture: The Selector of Improbable Mutations,” Proceedings of the National Academy of Sciences of the United States of America 103, no. 33 (2006): 12,215. 
  35. Leo James and Dan S. Tawfik, “Conformational Diversity and Protein Evolution – a 60-Year-Old Hypothesis Revisited,” Trends in Biochemical Sciences 28, no. 7 (2003): 366. 
  36. Leo James and Dan S. Tawfik, “Conformational Diversity and Protein Evolution – a 60-Year-Old Hypothesis Revisited,” Trends in Biochemical Sciences 28, no. 7 (2003): 361–68. 
  37. Leo James and Dan S. Tawfik, “Conformational Diversity and Protein Evolution – a 60-Year-Old Hypothesis Revisited,” Trends in Biochemical Sciences 28, no. 7 (2003): 366. 
  38. A ligand is the target molecule to which a protein binds and upon which it acts. 
  39. Nobuhiko Tokuriki and Dan S. Tawfik, “Protein Dynamism and Evolvability,” Science 324, no. 5924 (2009): 203–07. The authors write: “IDPs are considered as a separate class of proteins with specific sequence compositions and functions that correlate with their unusual properties.” 205. 
  40. Zhenling Peng et al., “A Creature With a Hundred Waggly Tails: Intrinsically Disordered Proteins in the Ribosome,” Cellular and Molecular Life Sciences 71, no. 8 (2014), doi:10.1007/s00018-013-1446-6. The authors state:
    [W]e investigate evolutionary conservation of intrinsic disorder in ribosomal proteins. ... [O]ur results show that disordered residues are more conserved [emphasis added] when compared with the structured parts of the ribosomal proteins. ... Moreover, we show that residues located in long disordered segments of ribosomal proteins are more conserved than the overall population of both disordered and ordered amino acids across all three domains of life.
     
  41. M. Madan Babu et al., “Intrinsically Disordered Proteins: Regulation and Disease,” Current Opinion in Structural Biology 21, no. 3 (2011), doi:10.1016/j.sbi.2011.03.011. The authors write:
    Because of their unusual structural features and important functional properties, the presence of IDPs in a cell needs to be carefully monitored. ... [I]t is not surprising that altered availability (i.e., abundance and half-life) of IDPs is associated with several disease conditions ... These examples suggest that maintaining the right endogenous levels of IDPs and the integrity of the associated regulatory mechanism is crucial for normal cellular function. (1–2)
     
  42. Anthony Keefe and Jack Szostak, “Functional Proteins from a Random-Sequence Library,” Nature 410 (2001): 715–18. 
  43. Purine Nucleotide Biosynthesis,” The Medical Biochemistry Page. 
  44. Nobuhiko Tokuriki and Dan S. Tawfik, “Protein Dynamism and Evolvability,” Science 324, no. 5,924 (2009): 206. The authors note:
    The size of sequence space of biopolymers ... can be daunting (20100 ≈ 10130 sequence permutations for a 100-amino acid polypeptide). It makes the emergence of sequences with function a highly improbable event, despite considerable redundancy (many sequences giving the same structure and function).
     
  45. Rajendrani Mukhopadhyay, “‘Close to a Miracle:’ Researchers are Debating the Origins of Proteins,” American Society for Biochemistry and Molecular Biology 12, no. 9 (2013): 13. 
  46. Shinya Honda et al., “10 Residue Folded Peptide Designed by Segment Statistics,” Structure 12, no. 8 (2004): 1,507–18. 
  47. Shinya Honda et al., “10 Residue Folded Peptide Designed by Segment Statistics,” Structure 12, no. 8 (2004): 1507–18. The experiment describes the design of a protein named chignolin, the smallest folded “protein” to date. Chignolin is termed a protein because it exhibits a shape and folds cooperatively. However, this description can be somewhat deceptive, because the shape is a β-hairpin, and as one recent review notes, “β-hairpin peptides ... are much less complex [emphasis added] in sequence and structure than full proteins.” See Søren Enemark and Raj Rajagopalan, “Turn-Directed Folding Dynamics of β-Hairpin-Forming de novo Decapeptide Chignolin,” Physical Chemistry Chemical Physics 14 (2012): 12,442–50. Hairpins are basic elements of anti-parallel β-sheets, which are subunits of the larger protein shape. Moreover, chignolin exists in equilibrium, not between two stable states, but between one small stable state and a random coil with no shape. Chignolin does not have catalysis. It is not clearly useful for anything on its own, except in the context of a larger structure. Finally, chignolin is, if anything, an exception, as Honda et al. write on page 1,514: “... it is also true that most fragments of this size cannot keep their original structure.” 
  48. It is relevant that both in the initial and subsequent analyses, the homopentameric form of the fragments was not the only one exhibited. Fragments were found to exist also as monomers, dimers, trimers, tetramers, and larger inclusion bodies. A fraction of these forms could be, or were, active, although the authors do not make clear what fraction of each form was functional. It is not clear, however, the extent to which these variants were stable. Could a single 100-amino-acid fragment function alone? Though the monomer form (a single fragment) does make an appearance in the experiment, this form is probably much more prone to structural destabilization. Its relevance to evolution, outside a lab, would be suspect. The authors generally emphasize the pentamer, not the individual fragment’s ability. When challenged to stabilize structure, for example, in vitro evolution seems to have favored the pentameric form, as exhibited in the solved crystal structure in the 2010 study. See Dan S. Tawfik et al., “Metamorphic Proteins Mediate Evolutionary Transitions of Structure,” Proceedings of the National Academy of Sciences of the United States of America 107, no. 16 (2010): 7,287–92. The authors note that the pentameric form was the main oligomeric state throughout its evolution—probably because it was the most stable solution. It is not clear that a single fragment, acting alone, can function to appreciable physiological levels without being hindered by severe aggregation and structural destabilization. In 2010 the authors emphasized the importance of oligomeric assembly (that is, more than one fragment) for context:
    The evolution of β-propellers is likely to have proceeded through stepwise additions of smaller elements. Our results indicate that such elements could be functional via oligomeric assembly [emphasis added] and that backbone plasticity can mediate the assembly of such small elements to give functional oligomers. (7,291)
  49. Brian Matthews, Charles Craik, and Hans Neurath, “Can Small Cyclic Peptides Have the Activity and Specificity of Proteolytic Enzymes?” Proceedings of the National Academy of Sciences of the United States of America 91, no. 10 (1994): 4,103–05. 
  50. Deoxyribozymes are a class of enzymes based on the structure of DNA. These molecules are more limited than ribozymes, and no one in the field currently expects that these are highly relevant to the origin of life. 
  51. Andrew Pohorille and Michael New, “Models of Protocellular Structures, Functions and Evolution,” Preprint of conference proceedings, “Rencontres de Blois Frontiers of Life [sic], June 15, 2000-July 1, 2000,” Chateau de Blois, France, NASA Technical Reports Server, October 4, 2000. 
  52. For an enlightening pair of articles, see Michael Robertson and Gerald Joyce, “The Origins of the RNA World,” Cold Spring Harbor Perspectives in Biology (2010), doi: 10.1101/cshperspect.a003608; and Steven Benner, Andrew Ellington and Andreas Tauer, “Modern Metabolism as a Palimpsest of the RNA World,” Proceedings of the National Academy of Sciences of the United States of America 86, no. 18 (1989): 7,054–8. With respect to the essentiality of nucleotide synthesis, Robertson and Joyce note:
    RNA replicase activity is probably not [emphasis added] the only catalytic behavior that was essential for the existence of the RNA world. Maintaining an adequate supply of the four activated nucleotides would have been a top priority. Even if the prebiotic environment contained a large reservoir of these compounds, the reservoir would eventually become depleted, and some capacity for nucleotide biosynthesis would have been required. (11)
    Steven Benner, Andrew Ellington, and Andreas Tauer state, of metabolism:
    [P]hotosynthesis (of some sort) is expected to have emerged in the RNA world ... a complex metabolism would rapidly exhaust the high-energy molecules created abiotically [emphasis added] ... (7,056–57) it is difficult to imagine the evolution of ... translation without having, for example, a renewable source of high-energy phosphates [metabolism] or biosynthetic routes to sugars [nucleotides]. (7,057)
    None of these the authors say exactly when metabolism and nucleotide biosynthesis must have appeared. But considering uncontrollable environmental circumstances and the intrinsic demands of life, one might say extremely early. For all we know, perhaps simultaneously. 
  53. Though several notable elements are missing (such as a proton pump and gradient), ribozymes have been observed that can create bonds that appear to be several steps away from metabolism and biosynthesis. However, note three points. First, the number of reactions and critical steps for metabolism and biosynthesis do not diminish. This means a suite of ribozymes (to control each step) appears to be necessary very early in the process, perhaps from the beginning. Secondly, to the author’s knowledge, these ribozymes have never been put together coherently or partially coherently in a pathway, let alone those fully constructing oxidative phosphorylation, substrate level phosphorylation, or methanogenesis. Thirdly, we must account for the nature of those reactions that do parallel natural reactions, such as nucleotide biosynthesis:
    None of these four [examples of] RNA-catalyzed reactions [have] precisely the right format for the corresponding reaction in a hypothetical nucleotide biosynthesis pathway in the RNA world [the same might be true of metabolism]. However, they show that RNA is capable of performing the relevant chemistry with substantial catalytic rate enhancement. It remains to be seen whether ribozymes can be developed that catalyze the formation of the fundamental building blocks of RNA, D-ribose and the four nucleotide bases, using starting materials that would have been abundant on the primitive Earth.
    Michael Robertson and Gerald Joyce, “The Origins of the RNA World,” Cold Spring Harbor Perspectives in Biology (2010): 12, doi: 10.1101/cshperspect.a003608. 
  54. Andrew Pohorille and Michael New, “Models of Protocellular Structures, Functions and Evolution,” Preprint of conference proceedings, “Rencontres de Blois Frontiers of Life [sic], June 15, 2000-July 1, 2000,” Chateau de Blois, France, NASA Technical Reports Server, October 4, 2000. 7. 
  55. Committee on the Limits of Organic Life in Planetary Systems, Committee on the Origins and Evolution of Life, National Research Council, The Limits of Organic Life in Planetary Systems (Washington, DC: The National Academies Press, 2007), 28:
    This conundrum underlies the most significant paradox in the structure of genetic matter with respect to the origin of life. On the one hand, the repeating charge in the backbone suggests that the molecule worked in a hydrophilic solvent such as water. On the other hand, the abundance of easily hydrolyzable bonds suggests that RNA could not have been easily assembled in water.
     
  56. Anthony Forster and George Church write:
    An additional challenge is that about a third of the essential genes have unknown functions. ... [t]he functions of about a third of the genes would remain unknown. Building a machine from mysterious parts can only create a mysterious machine. What is needed is some way of defining a near-minimal genome and then a strategy that will lead inexorably to an understanding of all of its parts. (1,070)
    The authors hint at what remains to be discovered:
    Along the way to synthesizing a cell, we might discover new biochemical functions essential for replication, unsuspected macromolecular modifications or previously unrecognized patterns of coordinated expression. (1,069)
    Anthony Forster and George Church, “Towards Synthesis of a Minimal Cell,” Molecular Systems Biology 2, no. 45 (2006): 1,069–80, doi: 10.1038/msb4100090. 
  57. The promiscuous HAD-like hydrolases in E. coli, for example, are employed in the cell, but survive only by a severe regulation process, which “prevents many of the undesirable outcomes of promiscuity,” because only a fraction of the activity “is accessible and active at a given time and cellular location.” The danger of promiscuity can be similarly noted in kinases, in which incorrect phosphorylation of wrong molecules can lead to a host of complications, most notably, death. 
  58. Olga Khersonsky and Dan S. Tawfik, “Enzyme Promiscuity: A Mechanistic and Evolutionary Perspective,” Annual Review of Biochemistry 79 (2010): 493. 
  59. Misha Soskine and Dan S. Tawfik, “Mutational Effects and the Evolution of New Protein Functions,” Nature Reviews Genetics 11 (2010): 577. 
  60. Leonard Bogarad and Michael Deem, “A Hierarchical Approach to Molecular Evolution,” Proceedings of the National Academy of Sciences of the United States of America 96, no. 6 (1999): 2,591. 
  61. Dan S. Tawfik, “Loop Grafting and the Origins of Enzyme Species,” Science 311 (2006): 475. 
  62. Dan S. Tawfik, “Loop Grafting and the Origins of Enzyme Species,” Science 311 (2006): 475. 
  63. Dan S. Tawfik, “Loop Grafting and the Origins of Enzyme Species,” Science 311 (2006): 475. 
  64. Itamar Yadid and Dan S. Tawfik, “Reconstruction of Functional β-Propeller Lectins via Homo-oligomeric Assembly of Shorter Fragments,” Journal of Molecular Biology 365, no. 1 (2007): 11. The authors write:
    The screen yielded an unexpectedly high proportion of active variants (∼7%) that varied in length from 96 to 116 amino acid residues. ... The subsequent screen of 344 individual clones yielded a similar result (7.5% positive clones), with active variants ranging from 93 to 136 amino acid residues.
    This implies that the inactive variants were between 92.5 percent and 93 percent, roughly, of the population. 
  65. To illustrate: one protein, ABC, was duplicated and fused, resulting in ABCABC. When mutated, the majority of sequences passing selection screens had truncated the copy. They reverted to the original (and fitter) sequence ABC instead of cutting both ends, and creating new rearrangements: BCA or CAB. The ratio of reversion was 30 to 1. 
  66. Ágnes Tóth-Petróczy and Dan S. Tawfik, “Hopeful (Protein InDel) Monsters?” Structure 22 (2014): 803. 
  67. James Arpino et al., “Random Single Amino Acid Deletion Sampling Unveils Structural Tolerance and the Benefits of Helical Registry Shift on GFP Folding and Structure,” Structure 22 (2014): 889–98. 
  68. Tawfik notes one rare, documented, and awe-striking change that controlled both fundamental form and function: a shift between two small domains composing Protein G, named GA and GB. See Alon Wellner, Maria Raitses Gurevich, and Dan S. Tawfik, “Mechanisms of Sequence Divergence and Incompatibility,” PLoS Genetics 9, no. 7 (2013), doi:10.1371/journal.pgen.1003665.

    A team that included Patrick Alexander, Yanan He and others was able to induce a switch in form and function between GA and GB, in which the intermediate sequences were viable, though less stable than the originals. See Patrick Alexander et al., “The Design and Characterization of Two Proteins with 88% Sequence Identity but Different Structure and Function,” Proceedings of the National Academy of Sciences of the United States of America 104, no. 29 (2007): 11,963–68. This demonstrated the possibility that a large percentage of amino acids in a protein (between 88 percent and 98 percent) can be compatible with multiple folds, even when the domains were not previously known to be metamorphic. Context dependence is in effect again. The catch? Evolution had almost nothing to do with it. It was a supreme effort of engineering. The experiment did not use an evolutionary library, whether neutral or directional, to obtain the transitional sequences. Instead, the team deliberately fixed a complete, but dormant, binding motif of three mutations in one of the domains, and a motif of five mutations in the other, to ensure that any structural transition would also have a new accompanying function. After this, the team progressively engineered further sequence changes to draw the two sequences to near unity while drawing out the dormant binding potential they had originally inserted. In the meantime, they largely ignored each change’s immediate selective context or value. Once a suitable background was reached, the dormant binding motif was activated and new function emerged. This protocol, of course, does not translate to the real world. If a function is not there to be selected initially, it must be obtained by drift, or in other words, by chance. Yet the collection, by chance, of a specific combination of three to five amino acids for the binding site in its dormant mode is improbable, to say nothing of the further, necessary sequence changes to expose full functional potential for the full transition.

    One paper, in addressing a different problem, nonetheless captures the essential criticism: there has not been enough time to create the combination of amino acids at other sites to allow them to evolve. See Yanan He et al., “Mutational Tipping Points for Switching Protein Folds and Functions,” Structure 20, no. 2 (2012): 283–91.

    Finally, the results of the abovementioned paper by Alexander et al., however dramatic, may not be widely applicable: antibody-binding proteins such as Protein G simply bind molecules; they do not catalyze reactions. They do not operate under the same strict constraints as enzymes and their catalytic groups. It is not clear whether enzymes could tolerate the same functional and structural rearrangements. This question, the crucial one, remains open. 
  69. Ágnes Tóth-Petróczy and Dan S. Tawfik, “Hopeful (Protein InDel) Monsters?” Structure 22 (2014): 803. 
  70. Ágnes Tóth-Petróczy and Dan S. Tawfik, “Slow Protein Evolutionary Rates are Dictated by Surface–Core Association,” Proceedings of the National Academy of Sciences of the United States of America 108, no. 27 (2011): 11,151–56. 
  71. Nobuhiko Tokuriki and Dan S. Tawfik, “Stability Effects of Mutations and Protein Evolvability,” Current Opinion in Structural Biology 19, no. 5 (2009): 596–604. 
  72. Dan S. Tawfik et al., “The ‘Evolvability’ of Promiscuous Protein Functions,” Nature Genetics 37, no. 1 (2005): 73–76. 
  73. Gil Amitai, Rinkoo Devi Gupta, and Dan S. Tawfik, “Latent Evolutionary Potentials under the Neutral Mutational Drift of an Enzyme,” HFSP Journal 1, no. 1 (2007): 67–78. 
  74. Dan S. Tawfik et al., “Diminishing Returns and Tradeoffs Constrain the Laboratory Optimization of an Enzyme,” Nature Communications 3, no. 1257 (2012), doi:10.1038/ncomms2246. 
  75. Itamar Yadid and Dan S. Tawfik, “Reconstruction of Functional β-Propeller Lectins via Homo-oligomeric Assembly of Shorter Fragments,” Journal of Molecular Biology 365, no. 1 (2007): 10–17. 
  76. Olga Khersonsky and Dan S. Tawfik, “Enzyme Promiscuity: A Mechanistic and Evolutionary Perspective,” Annual Review of Biochemistry 79 (2010): 471–505. 
  77. Dan S. Tawfik et al., “Metamorphic Proteins Mediate Evolutionary Transitions of Structure,” Proceedings of the National Academy of Sciences of the United States of America 107, no. 16 (2010): 7,287–92. 
  78. Itamar Yadid and Dan S. Tawfik, “Functional β-Propeller Lectins by Tandem Duplications of Repetitive Units,” Protein Engineering, Design & Selection 24, no. 1 (2011): 185–95. 
  79. Sergio G. Peisajovich, Liat Rockah, and Dan S. Tawfik, “Evolution of New Protein Topologies Through Multistep Gene Rearrangements,” Nature Genetics 38, no. 2 (2006): 168–74. 
  80. Nobuhiko Tokuriki and Dan S. Tawfik, “Protein Dynamism and Evolvability,” Science 324, no. 5924 (2009): 203–207. 
  81. Dan S. Tawfik et al., “What Makes a Protein Fold Amenable to Functional Innovation? Fold Polarity and Stability Trade-offs,” Journal of Molecular Biology 425 (2013): 2,609–21. 
  82. Leo James and Dan S. Tawfik, “Conformational Diversity and Protein Evolution – a 60-Year-Old Hypothesis Revisited,” Trends in Biochemical Sciences 28, no. 7 (2003): 361–68. 
  83. Dan S. Tawfik et al., “Do Viral Proteins Possess Unique Biophysical Features?” Trends in Biochemical Sciences 34, no. 2 (2009): 53–59. 
  84. Misha Soskine and Dan S. Tawfik, “Mutational Effects and the Evolution of New Protein Functions,” Nature Reviews Genetics 11 (2010): 579–82. 
  85. Yakov Kipnis, Eynat Dellus-Gur, and Dan S. Tawfik, “TRINS: A Method for Gene Modification by Randomized Tandem Repeat Insertions,” Protein Engineering, Design & Selection 25, no. 9 (2012): 437–44. 
  86. Dan S. Tawfik et al., “Role of Chemistry Versus Substrate Binding in Recruiting Promiscuous Enzyme Functions,” Biochemistry 50 (2011): 2,683–90. 
  87. Dan S. Tawfik, “Loop Grafting and the Origins of Enzyme Species,” Science 311 (2006): 475–76. 
  88. Dan S. Tawfik et al., “Initial Mutations Direct Alternative Pathways of Protein Evolution,” PLoS Genetics 7, no. 3 (2011), doi:10.1371/journal.pgen.1001321. 
  89. Sergio G. Peisajovich and Dan S. Tawfik, “Protein Engineers Turned Evolutionists,” Nature Methods 4, no. 12 (2007): 991–94.