Search

Phytoplankton are key components of the oceanic carbon and sulfur cycles1. During bloom events, some species can emit large amounts of the organosulfur volatile dimethyl sulfide (DMS) into the ocean and consequently the atmosphere, where it can modulate aerosol formation and affect climate2,3. In aquatic environments, DMS plays an important role as a chemical signal mediating diverse trophic interactions. Yet, its role in microbial predator–prey interactions remains elusive with contradicting evidence for its role in either algal chemical defence or in the chemo-attraction of grazers to prey cells4,5. Here we investigated the signalling role of DMS during zooplankton–algae interactions by genetic and biochemical manipulation of the algal DMS-generating enzyme dimethylsulfoniopropionate lyase (DL) in the bloom-forming alga Emiliania huxleyi6. We inhibited DL activity in E. huxleyi cells in vivo using the selective DL-inhibitor 2-bromo-3-(dimethylsulfonio)-propionate7 and overexpressed the DL-encoding gene in the model diatom Thalassiosira pseudonana. We showed that algal DL activity did not serve as an anti-grazing chemical defence but paradoxically enhanced predation by the grazer Oxyrrhis marina and other microzooplankton and mesozooplankton, including ciliates and copepods. Consumption of algal prey with induced DL activity also promoted O. marina growth. Overall, our results demonstrate that DMS-mediated grazing may be ecologically important and prevalent during prey–predator dynamics in aquatic ecosystems. The role of algal DMS revealed here, acting as an eat-me signal for grazers, raises fundamental questions regarding the retention of its biosynthetic enzyme through the evolution of dominant bloom-forming phytoplankton in the ocean. Algal production of dimethyl sulfide plays a role in attracting predators and enhancing predation by zooplankton, thus mediating predator–prey relationships in the ocean.

Oligomeric proteins are central to cellular life and the duplication and divergence of their genes is a key driver of evolutionary innovations. The duplication of a gene coding for an oligomeric protein has numerous possible outcomes, which motivates questions on the relationship between structural and functional divergence. How do protein oligomeric states diversify after gene duplication? In the simple case of duplication of a homo-oligomeric protein gene, what properties can influence the fate of descendant paralogs toward forming independent homomers or maintaining their interaction as a complex? Furthermore, how are functional innovations associated with the diversification of oligomeric states? Here, we review recent literature and present specific examples in an attempt to illustrate and answer these questions.

Alteration of antibiotic binding sites through modification of ribosomal RNA (rRNA) is a common form of resistance to ribosome-targeting antibiotics. The rRNA-modifying enzyme Cfr methylates an adenosine nucleotide within the peptidyl transferase center, resulting in the C-8 methylation of A2503 (m8A2503). Acquisition of cfr results in resistance to eight classes of ribosome-targeting antibiotics. Despite the prevalence of this resistance mechanism, it is poorly understood whether and how bacteria modulate Cfr methylation to adapt to antibiotic pressure. Moreover, direct evidence for how m8A2503 alters antibiotic binding sites within the ribosome is lacking. In this study, we performed directed evolution of Cfr under antibiotic selection to generate Cfr variants that confer increased resistance by enhancing methylation of A2503 in cells. Increased rRNA methylation is achieved by improved expression and stability of Cfr through transcriptional and post-transcriptional mechanisms, which may be exploited by pathogens under antibiotic stress as suggested by natural isolates. Using a variant that achieves near-stoichiometric methylation of rRNA, we determined a 2.2 Å cryo-electron microscopy structure of the Cfr-modified ribosome. Our structure reveals the molecular basis for broad resistance to antibiotics and will inform the design of new antibiotics that overcome resistance mediated by Cfr., eLife digest Antibiotics treat or prevent infections by killing bacteria or slowing down their growth. A large proportion of these drugs do this by disrupting an essential piece of cellular machinery called the ribosome which the bacteria need to make proteins. However, over the course of the treatment, some bacteria may gain genetic alterations that allow them to resist the effects of the antibiotic. Antibiotic resistance is a major threat to global health, and understanding how it emerges and spreads is an important area of research. Recent studies have discovered populations of resistant bacteria carrying a gene for a protein named chloramphenicol-florfenicol resistance, or Cfr for short. Cfr inserts a small modification in to the ribosome that prevents antibiotics from inhibiting the production of proteins, making them ineffective against the infection. To date, Cfr has been found to cause resistance to eight different classes of antibiotics. Identifying which mutations enhance its activity and protect bacteria is vital for designing strategies that fight antibiotic resistance. To investigate how the gene for Cfr could mutate and make bacteria more resistant, Tsai et al. performed a laboratory technique called directed evolution, a cyclic process which mimics natural selection. Genetic changes were randomly introduced in the gene for the Cfr protein and bacteria carrying these mutations were treated with tiamulin, an antibiotic rendered ineffective by the modification Cfr introduces into the ribosome. Bacteria that survived were then selected and had more mutations inserted. By repeating this process several times, Tsai et al. identified ‘super’ variants of the Cfr protein that lead to greater resistance. The experiments showed that these variants boosted resistance by increasing the proportion of ribosomes that contained the protective modification. This process was facilitated by mutations that enabled higher levels of Cfr protein to accumulate in the cell. In addition, the current study allowed, for the first time, direct visualization of how the Cfr modification disrupts the effect antibiotics have on the ribosome. These findings will make it easier for clinics to look out for bacteria that carry these ‘super’ resistant mutations. They could also help researchers design a new generation of antibiotics that can overcome resistance caused by the Cfr protein.

The vast majority of theoretically possible polypeptide chains do not fold, let alone confer function. Hence, protein evolution from preexisting building blocks has clear potential advantages over ab initio emergence from random sequences. In support of this view, sequence similarities between different proteins is generally indicative of common ancestry, and we collectively refer to such homologous sequences as ‘themes’. At the domain level, sequence homology is routinely detected. However, short themes which are segments, or fragments of intact domains, are particularly interesting because they may provide hints about the emergence of domains, as opposed to divergence of preexisting domains, or their mixing-and-matching to form multi-domain proteins. Here we identified 525 representative short themes, comprising 20-to-80 residues, that are unexpectedly shared between domains considered to have emerged independently. Among these ‘bridging themes’ are ones shared between the most ancient domains, e.g., Rossmann, P-loop NTPase, TIM-barrel, Flavodoxin, and Ferredoxin-like. We elaborate on several particularly interesting cases, where the bridging themes mediate ligand binding. Ligand binding may have contributed to the stability and the plasticity of these building blocks, and to their ability to invade preexisting domains or serve as starting points for completely new domains.

Oligomeric proteins are central to life. Duplication and divergence of their genes is a key evolutionary driver, also because duplications can yield very different outcomes. Given a homomeric ancestor, duplication can yield two paralogs that form two distinct homomeric complexes, or a heteromeric complex comprising both paralogs. Alternatively, one paralog remains a homomer while the other acquires a new partner. However, so far, conflicting trends have been noted with respect to which fate dominates, primarily because different methods and criteria are being used to assign the interaction status of paralogs. Here, we systematically analyzed all Saccharomyces cerevisiae and Escherichia coli oligomeric complexes that include paralogous proteins. We found that the proportions of homo-hetero duplication fates strongly depend on a variety of factors, yet that nonetheless, rigorous filtering gives a consistent picture. In E. coli about 50%, of the paralogous pairs appear to have retained the ancestral homomeric interaction, whereas in S. cerevisiae only ~10% retained a homomeric state. This difference was also observed when unique complexes were counted instead of paralogous gene pairs. We further show that this difference is accounted for by multiple cases of heteromeric yeast complexes that share common ancestry with homomeric bacterial complexes. Our analysis settles contradicting trends and conflicting previous analyses, and provides a systematic and rigorous pipeline for delineating the fate of duplicated oligomers in any organism for which protein-protein interaction data are available., Author summary About half of all proteins assemble as oligomers, either by self-interaction (homomers) or via interaction with another protein (heteromers). The latter can be unrelated, yet, quite commonly, the interacting proteins are paralogs, namely two genes that arose by gene duplication. Indeed, while a homomer is encoded by a single gene, heteromers demand two genes as a minimum. Duplication can therefore yield two discrete homomeric complexes or a single heteromer. Do paralogs tend to retain the ancestral homomeric interaction, or do they mostly diverge into heteromeric complexes? Despite several studies addressing this question, to date, we lack a systematic, rigorous approach for delineating the oligomeric fates of paralogs on an organism scale. To this end, we developed a new pipeline for analysis of molecular interaction databases that includes various filtering steps and unambiguous definitions of all possible oligomeric fates. Applying this method to Escherichia coli and Saccharomyces cerevisiae we noted that paralogous pairs tend to remain homomeric in the former while in the latter heteromeric complexes dominate. We consequently note a systematic trend of homomeric bacterial proteins diverging into heteromeric complexes in eukaryotes.

Evolutionary trajectories are deemed largely irreversible. In a newly diverged protein, reversion of mutations that led to the functional switch typically results in loss of both the new and the ancestral functions. Nonetheless, evolutionary transitions where reversions are viable have also been described. The structural and mechanistic causes of reversion compatibility versus incompatibility therefore remain unclear. We examined two laboratory evolution trajectories of mammalian paraoxonase-1, a lactonase with promiscuous organophosphate hydrolase (OPH) activity. Both trajectories began with the same active-site mutant, His115Trp, which lost the native lactonase activity and acquired higher OPH activity. A neo-functionalization trajectory amplified the promiscuous OPH activity, whereas the re-functionalization trajectory restored the native activity, thus generating a new lactonase that lacks His115. The His115 revertants of these trajectories indicated opposite trends. Revertants of the neo-functionalization trajectory lost both the evolved OPH and the original lactonase activity. Revertants of the trajectory that restored the original lactonase function were, however, fully active. Crystal structures and molecular simulations show that in the newly diverged OPH, the reverted His115 and other catalytic residues are displaced, thus causing loss of both the original and the new activity. In contrast, in the re-functionalization trajectory, reversion compatibility of the original lactonase activity derives from mechanistic versatility whereby multiple residues can fulfill the same task. This versatility enables unique sequence-reversible compositions that are inaccessible when the active site was repurposed toward a new function.

The vast majority of theoretically possible polypeptide chains do not fold, let alone confer function. Hence, protein evolution from preexisting building blocks has clear potential advantages over ab initio emergence from random sequences. In support of this view, sequence similarities between different proteins is generally indicative of common ancestry, and we collectively refer to such homologous sequences as "themes." At the domain level, sequence homology is routinely detected. However, short themes which are segments, or fragments of intact domains, are particularly interesting because they may provide hints about the emergence of domains, as opposed to divergence of preexisting domains, or their mixing-and-matching to form multi-domain proteins. Here we identified 525 representative short themes, comprising 20-80 residues that are unexpectedly shared between domains considered to have emerged independently. Among these "bridging themes" are ones shared between the most ancient domains, for example, Rossmann, P-loop NTPase, TIM-barrel, flavodoxin, and ferredoxin-like. We elaborate on several particularly interesting cases, where the bridging themes mediate ligand binding. Ligand binding may have contributed to the stability and the plasticity of these building blocks, and to their ability to invade preexisting domains or serve as starting points for completely new domains.

The linkage between regulatory elements of transcription, such as promoters, and their protein products is central to gene function. Promoter-protein coevolution is therefore expected, but rarely observed, and the manner by which these two regulatory levels are linked remains largely unknown. We study glutamate dehydrogenase-a hub of carbon and nitrogen metabolism. In Bacillus subtilis, two paralogues exist: GudB is constitutively transcribed whereas RocG is tightly regulated. In their active, oligomeric states, both enzymes show similar enzymatic rates. However, swaps of enzymes and promoters cause severe fitness losses, thus indicating promoter-enzyme coevolution. Characterization of the proteins shows that, compared to RocG, GudB's enzymatic activity is highly dependent on glutamate and pH Promoter-enzyme swaps therefore result in excessive glutamate degradation when expressing a constitutive enzyme under a constitutive promoter, or insufficient activity when both the enzyme and its promoter are tightly regulated. Coevolution of transcriptional and enzymatic regulation therefore underlies paralogue-specific spatio-temporal control, especially under diverse growth conditions.

Ancestral reconstruction provides instrumental insights regarding the biochemical and biophysical characteristics of past proteins. A striking observation relates to the remarkably high thermostability of reconstructed ancestors. The latter has been linked to high environmental temperatures in the Precambrian era, the era relating to most reconstructed proteins. We found that inferred ancestors of the serum paraoxonase (PON) enzyme family, including the mammalian ancestor, exhibit dramatically increased thermostabilities compared with the extant, human enzyme (up to 30 °C higher melting temperature). However, the environmental temperature at the time of emergence of mammals is presumed to be similar to the present one. Additionally, the mammalian PON ancestor has superior folding properties (kinetic stability)-unlike the extant mammalian PONs, it expresses in E. coli in a soluble and functional form, and at a high yield. We discuss two potential origins of this unexpectedly high stability. First, ancestral stability may be overestimated by a "consensus effect," whereby replacing amino acids that are rare in contemporary sequences with the amino acid most common in the family increases protein stability. Comparison to other reconstructed ancestors indicates that the consensus effect may bias some but not all reconstructions. Second, we note that high stability may relate to factors other than high environmental temperature such as oxidative stress or high radiation levels. Foremost, intrinsic factors such as high rates of genetic mutations and/or of transcriptional and translational errors, and less efficient protein quality control systems, may underlie the high kinetic and thermodynamic stability of past proteins.

Fitness landscapes1,2, depictions of how genotypes manifest at the phenotypic level, form the basis for our understanding of many areas of biology2–7 yet their properties remain elusive. Studies addressing this issue often consider specific genes and their function as proxy for fitness2,4, experimentally assessing the impact on function of single mutations and their combinations in a specific sequence2,5,8–15 or in different sequences2,3,5,16–18. However, systematic high-throughput studies of the local fitness landscape of an entire protein have not yet been reported. Here, we chart an extensive region of the local fitness landscape of the green fluorescent protein from Aequorea victoria (avGFP) by measuring the native function, fluorescence, of tens of thousands of derivative genotypes of avGFP. We find that its fitness landscape is narrow, with half of genotypes with two mutations showing reduced fluorescence and half of genotypes with five mutations being completely non-fluorescent. The narrowness is enhanced by epistasis, which was detected in up to 30% of genotypes with multiple mutations arising mostly through the cumulative impact of slightly deleterious mutations causing a threshold-like decrease of protein stability and concomitant loss of fluorescence. A model of orthologous sequence divergence spanning hundreds of millions of years predicted the extent of epistasis in our data, indicating congruence between the fitness landscape properties at the local and global scales. The characterization of the local fitness landscape of avGFP has important implications for a number of fields including molecular evolution, population genetics and protein design.

The last decade has seen a growing number of experiments aimed at systematically mapping the effects of mutations in different proteins, and of attempting to correlate their biophysical and biochemical effects with organismal fitness. While insightful, systematic laboratory measurements of fitness effects present challenges and difficulties. Here, we discuss the limitations associated with such measurements, and in particular the challenge of correlating the effects of mutations at the single protein level ("protein fitness") with their effects on organismal fitness. A variety of experimental setups are used, with some measuring the direct effects on protein function and others monitoring the growth rate of a model organism carrying the protein mutants. The manners by which fitness effects are calculated and presented also vary, and the conclusions, including the derived distributions of fitness effects of mutations, vary accordingly. The comparison of the effects of mutations in the laboratory to the natural protein diversity, namely to amino acid changes that have fixed in the course of millions of years of evolution, is also debatable. The results of laboratory experiments may, therefore, be less relevant to understanding long-term inter-species variations yet insightful with regard to short-term polymorphism, for example, in the study of the effects of human SNPs.

How does environmental complexity affect the evolution of single genes? Here, we measured the effects of a set of mutants ofBacillus subtilisglutamate dehydrogenase across 19 different environments - from homogenous single cell populations in liquid media to heterogeneous biofilms, plant roots and soil communities. The effects of individual gene mutations on organismal fitness were highly reproducible in liquid cultures. Strikingly, however, 84% of the tested alleles showed opposing fitness effects under different growth conditions (environmental pleiotropy). In biofilms and soil samples, different alleles dominated in parallel replica experiments. Accordingly, we found that in these heterogeneous cell communities the fate of mutations was dictated by a combination of selection and drift. The latter relates to programmed prophage excisions that occurred along biofilm development. Overall, per individual condition, by the combined action of selection, pleiotropy and chance, a wide range of glutamate dehydrogenase mutations persisted and sometimes fixated. However, across longer periods and multiple environments, nearly all this diversity would be lost - indeed, considering all environments and conditions we have tested, wild-type is the fittest allele.

The emergence of catalysis in a noncatalytic protein scaffold is a rare, unexplored event. Chalcone isomerase (CHI), a key enzyme in plant flavonoid biosynthesis, is presumed to have evolved from a nonenzymatic ancestor related to the widely distributed fatty-acid binding proteins (FAPs) and a plant protein family with no isomerase activity (CHILs). Ancestral inference supported the evolution of CHI from a protein lacking isomerase activity. Further, we identified four alternative founder mutations, i.e., mutations that individually instated activity, including a mutation that is not phylogenetically traceable. Despite strong epistasis in other cases of protein evolution, CHI's laboratory reconstructed mutational trajectory shows weak epistasis. Thus, enantioselective CHI activity could readily emerge despite a catalytically inactive starting point. Accordingly, X-ray crystallography, NMR, and molecular dynamics simulations reveal reshaping of the active site toward a productive substrate-binding mode and repositioning of the catalytic arginine that was inherited from the ancestral fatty-acid binding proteins.

Ancestral reconstruction is a powerful tool for studying protein evolution as well as for protein design and engineering. However, in many positions alternative predictions with relatively high marginal probabilities exist, and thus the prediction comprises an ensemble of near-ancestor sequences that relate to the historical ancestor. The ancestral phenotype should therefore be explored for the entire ensemble, rather than for the sequence comprising the most probable amino acid at all positions [the most probable ancestor (mpa)]. To this end, we constructed libraries that sample ensembles of near-ancestor sequences. Specifically, we identified positions where alternatively predicted amino acids are likely to affect the ancestor's structure and/or function. Using the serum paraoxonases (PONs) enzyme family as a test case, we constructed libraries that combinatorially sample these alternatives. We next characterized these libraries, reflecting the vertebrate and mammalian PON ancestors. We found that the mpa of vertebrate PONs represented only one out of many different enzymatic phenotypes displayed by its ensemble. The mammalian ancestral library, however, exhibited a homogeneous phenotype that was well represented by the mpa. Our library design strategy that samples near-ancestor ensembles at potentially critical positions therefore provides a systematic way of examining the robustness of inferred ancestral phenotypes.

To carry out their activities, biological macromolecules balance different physical traits, such as stability, interaction affinity, and selectivity. How such often opposing traits are encoded in a macromolecular system is critical to our understanding of evolutionary processes and ability to design new molecules with desired functions. We present a framework for constraining design simulations to balance different physical characteristics. Each trait is represented by the equilibrium fractional occupancy of the desired state relative to its alternatives, ranging from none to full occupancy, and the different traits are combined using Boolean operators to effect a “fuzzy”-logic language for encoding any combination of traits. In another paper, we presented a new combinatorial backbone design algorithm AbDesign where the fuzzy-logic framework was used to optimize protein backbones and sequences for both stability and binding affinity in antibody-design simulation. We now extend this framework and find that fuzzy-logic design simulations reproduce sequence and structure design principles seen in nature to underlie exquisite specificity on the one hand and multispecificity on the other hand. The fuzzy-logic language is broadly applicable and could help define the space of tolerated and beneficial mutations in natural biomolecular systems and design artificial molecules that encode complex characteristics., Graphical abstract, Highlights • Complex combinations of physical constraints shape the evolution of biomolecules. • An analytical system based on equilibrium thermodynamics and fuzzy logic is proposed to embody these constraints. • Computational design simulations subject to fuzzy-logic design recapitulate elements observed in natural proteins. • This design language can be used to analyze natural biomolecular systems and design new ones exhibiting complex traits.

The emergence of catalysis in a non-catalytic protein scaffold is a rare, unexplored event. Chalcone isomerase (CHI), a key enzyme in plant flavonoid biosynthesis, is presumed to have evolved from a non-enzymatic ancestor related to the widely-distributed fatty-acid binding proteins (FAPs) and a plant protein family with no isomerase activity (CHILs for “CHI-like”). Ancestral inference confirmed that CHI evolved from a protein lacking isomerase activity. We also identified four alternative founder mutations, i.e. mutations that individually instated activity, including a mutation that is not phylogenetically traceable. Despite strong epistasis in other cases of protein evolution, CHI’s laboratory reconstructed mutational trajectory shows weak epistasis. Thus, enantioselective CHI activity can readily emerge despite a catalytically inactive starting point. X-ray crystallography, NMR, and MD simulations reveal reshaping of the active site toward a productive substrate-binding mode and repositioning of the catalytic arginine that was inherited from the ancestral fatty-acid binding proteins.

Serum paraoxonases (PONs) are detoxifying lactonases that were first identified in mammals. Three mammalian families are known, PON1, 2, and 3 that reside primarily in the liver. They catalyze essentially the same reaction, lactone hydrolysis, but differ in their substrate specificity. Although some members are highly specific, others have a broad specificity profile. The evolutionary origins and substrate specificities of PONs therefore remain poorly understood. Here, we report a newly identified family of bacterial PONs, and the reconstruction of the ancestor of the three families of mammalian PONs. Both the mammalian ancestor and the characterized bacterial PONX_OCCAL were found to efficiently hydrolyze N-acyl homoserine lactones that mediate quorum sensing in many bacteria, including pathogenic ones. The mammalian PONs may therefore relate to a newly identified family of bacterial, PON-like “quorum-quenching” lactonases. The appearance of PONs in metazoa is likely to relate to innate immunity rather than detoxification. Unlike the bacterial PON, the mammalian ancestor also hydrolyzes, with low efficiency, lactones other than homoserine lactones, thus preceding the detoxifying functions that diverged later in two of the three mammalian families. The bifunctionality of the mammalian ancestor and the trade-off between the quorum-quenching and detoxifying lactonase activities explain the broad and overlapping specificities of some mammalian PONs versus the singular specificity of others.

Modifications of the bacterial ribosome regulate the function of the ribosome and modulate its susceptibility to antibiotics. By modifying a highly conserved adenosine A2503 in 23S rRNA, methylating enzyme Cfr confers resistance to a range of ribosome-targeting antibiotics. The same adenosine is also methylated by RlmN, an enzyme widely distributed among bacteria. While RlmN modifies C2, Cfr modifies the C8 position of A2503. Shared nucleotide substrate and phylogenetic relationship between RlmN and Cfr prompted us to investigate evolutionary origin of antibiotic resistance in this enzyme family. Using directed evolution of RlmN under antibiotic selection, we obtained RlmN variants that mediate low-level resistance. Surprisingly, these variants confer resistance not through the Cfr-like C8 methylation, but via inhibition of the endogenous RlmN C2 methylation of A2503. Detection of RlmN inactivating mutations in clinical resistance isolates suggests that the mechanism used by the in vitro evolved variants is also relevant in a clinical setting. Additionally, as indicated by a phylogenetic analysis, it appears that Cfr did not diverge from the RlmN family but from another distinct family of predicted radical SAM methylating enzymes whose function remains unknown.

DNA-binding and modifying proteins show high specificity but also exhibit a certain level of promiscuity. Such latent promiscuous activities comprise the starting points for new protein functions, but this hypothesis presents a paradox: a new activity can only evolve if it already exists. How then, do novel activities evolve? DNA methyltransferases, for example, are highly divergent in their target sites, but how transitions toward novel sites occur remains unknown. We performed laboratory evolution of the DNA methyltransferase M.HaeIII. We found that new target sites emerged primarily through expansion of the original site, GGCC, and the subsequent shrinkage of evolved expanded sites. Variants evolved for sites that are promiscuously methylated by M.HaeIII [GG((A)/(T))CC and GGCGCC] carried mutations in 'gate-keeper' residues. They could thereby methylate novel target sites such as GCGC and GGATCC that were neither selected for nor present in M.HaeIII. These 'generalist' intermediates were further evolved to obtain variants with novel target specificities. Our results demonstrate the ease by which new DNA-binding and modifying specificities evolve and the mechanism by which they occur at both the protein and DNA levels.

We discuss the basic features of divergent versus convergent evolution and of the common scenario of parallel evolution. The example of quorum-quenching lactonases is subsequently described. Three different quorum-quenching lactonase families are known, and they belong to three different superfamilies. Their key active-site architectures have converged and are strikingly similar. Curiously, a promiscuous organophosphate hydrolase activity is observed in all three families. We describe the structural and mechanistic features that underline this converged promiscuity and how this promiscuity drove the parallel divergence of organophosphate hydrolases within these lactonase families by either natural or laboratory evolution.

Why do certain proteins evolve much slower than others? We compared not only rates per protein, but also rates per position within individual proteins. For ∼90% of proteins, the distribution of positional rates exhibits three peaks: a peak of slow evolving residues, with average log 2 [normalized rate], log 2 μ, of ca. −2, corresponding primarily to core residues; a peak of fast evolving residues (log 2 μ ∼ 0.5) largely corresponding to surface residues; and a very fast peak (log 2 μ ∼ 2) associated with disordered segments. However, a unique fraction of proteins that evolve very slowly exhibit not only a negligible fast peak, but also a peak with a log 2 μ ∼ −4, rather than the standard core peak of −2. Thus, a “freeze” of a protein's surface seems to stop core evolution as well. We also observed a much higher fraction of substitutions in potentially interacting residues than expected by chance, including substitutions in pairs of contacting surface-core residues. Overall, the data suggest that accumulation of surface substitutions enables the acceptance of substitutions in core positions. The underlying reason for slow evolution might therefore be a highly constrained surface due to protein–protein interactions or the need to prevent misfolding or aggregation. If the surface is inaccessible to substitutions, so becomes the core, thus resulting in very slow overall rates.

Many, if not most, enzymes can promiscuously catalyze reactions, or act on substrates, other than those for which they evolved. Here, we discuss the structural, mechanistic, and evolutionary implications of this manifestation of infidelity of molecular recognition. We define promiscuity and related phenomena and also address their generality and physiological implications. We discuss the mechanistic enzymology of promiscuity—how enzymes, which generally exert exquisite specificity, catalyze other, and sometimes barely related, reactions. Finally, we address the hypothesis that promiscuous enzymatic activities serve as evolutionary starting points and highlight the unique evolutionary features of promiscuous enzyme functions.

How do intricate multi-residue features such as protein-protein interfaces evolve? To address this question, we evolved a new colicin-immunity binding interaction. We started with Im9, which inhibits its cognate DNase ColE9 at 10(-14) M affinity, and evolved it toward ColE7, which it inhibits promiscuously (Kd > 10(-8) M). Iterative rounds of random mutagenesis and selection toward higher affinity for ColE7, and selectivity (against ColE9 inhibition), led to an approximately 10(5)-fold increase in affinity and a 10(8)-fold increase in selectivity. Analysis of intermediates along the evolved variants revealed that changes in the binding configuration of the Im protein uncovered a latent set of interactions, thus providing the key to the rapid divergence of new Im7 variants. Overall, protein-protein interfaces seem to share the evolvability features of enzymes, that is, the exploitation of promiscuous interactions and alternative binding configurations via 'generalist' intermediates, and the key role of compensatory stabilizing mutations in facilitating the divergence of new functions.

Serum paraoxonases (PONs) are calcium-dependent lactonases with anti-atherogenic and detoxification functions. Here we describe the directed evolution and characterization of recombinant variants of serum paraoxonase PON3 that express in an active and soluble manner in Escherichia coli. These variants were obtained by combining family shuffling and phylogeny-based mutagenesis: the limited diversity of accessible, cloned PON3 genes was complemented by spiking the shuffling reaction with ancestor/ consensus mutations, mutations to residues that comprise the consensus or appear in the predicted ancestors of the PON family. We screened the resulting libraries for PON3's lactonase activity while ensuring that the selected variants retained the substrate specificity of wild-type mammalian PON3s. The availability of highly stable, recombinant PON3 that is free of all other serum components enabled us to explore unknown biochemical features of PON3, including its binding to HDL particles, the effect of HDL on PON3's stability and enzymatic activity, and ex vivo tests of its anti-atherogenic properties. Overall, it appears that PON3 possesses properties very similar to those of PON1: the enzyme's lactonase activity is selectively stimulated by binding to apoAI-HDL, with a concomitant increase in its stability. PON3 also exhibits potentially anti- atherogenic functions, although at levels lower than those of PON1.

Most protein mutations, and mutations that alter protein functions in particular, undermine stability and are therefore deleterious. Chaperones, or heat-shock proteins, are often implicated in buffering mutations, and could thus facilitate the acquisition of neutral genetic diversity and the rate of adaptation. We examined the ability of the Escherichia coli GroEL/GroES chaperonins to buffer destabilizing and adaptive mutations. Here we show that mutational drifts performed in vitro with four different enzymes indicated that GroEL/GroES overexpression doubled the number of accumulating mutations, and promoted the folding of enzyme variants carrying mutations in the protein core and/or mutations with higher destabilizing effects (destabilization energies of >3.5 kcal mol -1, on average, versus ∼1 kcal mol -1 in the absence of GroEL/GroES). The divergence of modified enzymatic specificity occurred much faster under GroEL/GroES overexpression, in terms of the number of adapted variants (≥2-fold) and their improved specificity and activity (≥10-fold). These results indicate that protein stability is a major constraint in protein evolution, and buffering mechanisms such as chaperonins are key in alleviating this constraint. Amino acid mutations that alter a protein's function can affect the stability of the protein. Protein chaperones are thought to 'buffer' these mutations, so they could theoretically facilitate the acquisition of hidden genetic diversity and the increase rate of adaptation. Nobuhiko Tokuriki and Dan Tawfik examined the ability of bacterial GroEL/GroES chaperonins to buffer destabilizing and adaptive mutations. In vitro mutational drift experiments on four enzymes revealed that GroEL/GroES overexpression doubles the number of accumulating mutations and promotes the folding of enzymes carrying mutations in the core of the protein. As mutation rates in bacteria increase under stress, these results suggest that a combination of increased mutation rates and chaperone overexpression can dramatically accelerate the rate of acquisition of new protein function. This work has implications for natural evolution, and the experimental system used provides a useful way of producing designer enzymes via in vitro directed evolution. Amino acid mutations that alter a protein's function can affect the stability of the protein, but these mutations are believed to be 'buffered' by chaperones, or heat-shock proteins-potentially facilitating the acquisition of genetic diversity and the rate of adaptation. Here, the overexpression of bacterial GroEL/GroES chaperonins is found to double the number of accumulating mutations in four different enzymes in vitro.

Phenotypic mutations (errors occurring during protein synthesis) are orders of magnitude more frequent than genetic mutations. Consequently, the sequences of individual protein molecules transcribed and translated from the same gene can differ. To test the effects of such mutations, we established a bacterial system in which an antibiotic resistance gene (TEM-1 β-lactamase) was transcribed by either a high-fidelity RNA polymerase or its error-prone mutant. This setup enabled the analysis of individual mRNA transcripts that were synthesized under normal or error-prone conditions. We found that an increase of ≈20-fold in the frequency of transcription errors promoted the evolution of higher TEM-1 expression levels and of more stable enzyme variants. The stabilized variants exhibited a distinct advantage under error-prone transcription, although under normal transcription they conferred resistance similar to wild-type TEM-1. They did so, primarily, by increasing TEM-1's tolerance to destabilizing deleterious mutations that arise from transcriptional errors. The stabilized TEM-1 variants also showed increased tolerance to genetic mutations. Thus, although phenotypic mutations are not individually subjected to inheritance and natural selection, as are genetic mutations, they collectively exert a direct and immediate effect on protein fitness. They may therefore play a role in shaping protein traits such as expression levels, stability, and tolerance to genetic mutations.

The directed evolution of proteins has benefited greatly from site-specific methods of diversification such as saturation mutagenesis. These techniques target diversity to a number of chosen positions that are usually non-contiguous in the protein's primary structure. However, the number of targeted positions can be large, thus leading to impractically large library size, wherein almost all library variants are inactive and the likelihood of selecting desirable properties is extremely small. We describe a versatile combinatorial method for the partial diversification of large sets of residues. Our library oligonucleotides comprise randomized codons that are flanked by wild-type sequences. Adding these oligonucleotides to an assembly PCR of wild-type gene fragments incorporates the randomized cassettes, at their target sites, into the reassembled gene. Varying the oligonucleotides concentration resulted in library variants that carry a different average number of mutated positions that comprise a random subset of the entire set of diversified codons. This method, dubbed Incorporating Synthetic Oligos via Gene Reassembly (ISOR), was used to create libraries of a cytosine-C5 methyltransferase wherein 45 individual positions were randomized. One library, containing an average of 5.6 mutated residues per gene, was selected, and mutants with wild-type-like activities isolated. We also created libraries of serum paraoxonase PON1 harboring insertions and deletions (indels) in various areas surrounding the active site. Screening these libraries yielded a range of mutants with altered substrate specificities and indicated that certain regions of this enzyme have a surprisingly high tolerance to indels.

Small libraries for directed evolution can be obtained by neutral drifts that maintain the protein's original function, yielding highly polymorphic, stable and evolvable variants. We describe methods for preparing such libraries, using serum paraoxonase (PON1). An optimized GFP variant fused to PON1 reported levels of soluble, functional enzyme, enabling selection by flow cytometry and identification of enzyme variants exhibiting improved specific and total activities toward several substrates, including toxic organophosphates.

This article describes a set of standard control experiments for the authentication of new protein variants isolated through library selection and site-directed mutagenesis. These controls are specifically designed to rule out artifacts derived from 'double transformants' -- i.e. cells transformed with, or infected by, two different plasmids simultaneously. These seem to have been the source of past artifacts and, as demonstrated here, are far more common than generally recognized. By following standard protocols for cloning, plasmid isolation, subcloning, in combination with functional assays, the presence of such artifacts can be ruled out. This protocol needs to be applied for any new variant isolated from heterogeneous gene repertoires, and in particular for variants isolated by selection for either enzymatic activity, or binding.

Derived from the yeast whole-genome duplication, Saccharomyces cerevisiae GAL1 and GAL3 encode the catabolic enzyme galactokinase (Gal1) and its transcriptional coinducer (Gal3), whereas the ancestral, preduplicated GAL1 gene performed both functions. Previous studies indicated that divergence was primarily driven by changes in upstream promoter elements, and changes in GAL3's coding region are assumed to be the result of drift. We show that replacement of GAL3's open-reading-frame with GAL1's results in an extended lag phase upon switching to growth on galactose with up to 2.5-fold differences in the initial cell masses. Accordingly, the binding affinity of Gal3 to Gal80 was found to be greater than 10-folds higher than that of Gal1, with both a higher association rate (ka) and lower dissociation (kd) rate. Thus, while changes in the noncoding, regulatory regions were the initial driving force for GAL3's subfunctionalization as a coinducer, adaptive changes in the protein sequence seem to have followed.

Systematic mappings of the effects of protein mutations are becoming increasingly popular. Unexpectedly, these experiments often find that proteins are tolerant to most amino acid substitutions, including substitutions in positions that are highly conserved in nature. To obtain a more realistic distribution of the effects of protein mutations, we applied a laboratory drift comprising 17 rounds of random mutagenesis and selection of M.HaeIII, a DNA methyltransferase. During this drift, multiple mutations gradually accumulated. Deep sequencing of the drifted gene ensembles allowed determination of the relative effects of all possible single nucleotide mutations. Despite being averaged across many different genetic backgrounds, about 67% of all nonsynonymous, missense mutations were evidently deleterious, and an additional 16% were likely to be deleterious. In the early generations, the frequency of most deleterious mutations remained high. However, by the 17th generation, their frequency was consistently reduced, and those remaining were accepted alongside compensatory mutations. The tolerance to mutations measured in this laboratory drift correlated with sequence exchanges seen in M.HaeIII’s natural orthologs. The biophysical constraints dictating purging in nature and in this laboratory drift also seemed to overlap. Our experiment therefore provides an improved method for measuring the effects of protein mutations that more closely replicates the natural evolutionary forces, and thereby a more realistic view of the mutational space of proteins., Author Summary Understanding and predicting the effects of single nucleotide polymorphisms (SNPs) is of fundamental importance in many fields. Systematic experimental mappings of the effects of such mutations within a given gene/protein comprise an essential experimental tool for determining protein function and for refining models of protein evolution, as well as an important resource for improving prediction algorithms. Here, we present the results of a laboratory system that mimics the manner by which protein sequences diverge in nature: a prolonged process of gradually accumulating random mutations that retain the protein’s structure and function. The change in frequencies of mutations over generations, as obtained by deep sequencing, enabled us to assess the relative effects of all possible SNPs at the background of an accumulating number of mutations. Compared to previous reports, we found that > 80% of all possible amino acid exchanges have potential deleterious effects, with 67% being clearly deleterious. Tolerance vs. purging of mutations in our prolonged drift also showed better correlation with natural diversity. Overall, our experimental setup provides a better understanding of how protein sequences diverge in nature, plus a new basis for improving the prediction accuracy of the effects of protein mutations, and specifically of SNPs.

Errors in protein synthesis, so-called phenotypic mutations, are orders-of-magnitude more frequent than genetic mutations. Here, we provide direct evidence that alternative protein forms and phenotypic variability derived from translational errors paved the path to genetic, evolutionary adaptations via gene duplication. We explored the evolutionary origins of Saccharomyces cerevisiae IDP3 - an NADP-dependent isocitrate dehydrogenase mediating fatty acids s-oxidation in the peroxisome. Following the yeast whole genome duplication, IDP3 diverged from a cytosolic ancestral gene by acquisition of a C-terminal peroxisomal targeting signal. We discovered that the pre-duplicated cytosolic IDPs are partially localized to the peroxisome owing to +1 translational frameshifts that bypass the stop codon and unveil cryptic peroxisomal targeting signals within the 3'-UTR. Exploring putative cryptic signals in all 3'-UTRs of yeast genomes, we found that other enzymes related to NADPH production such as pyruvate carboxylase 1 (PYC1) might be prone to peroxisomal localization via cryptic signals. Using laboratory evolution we found that these translational frameshifts are rapidly imprinted via genetic single base deletions occurring within the very same gene location. Further, as exemplified here, the sequences that promote translational frameshifts are also more prone to genetic deletions. Thus, genotypes conferring higher phenotypic variability not only meet immediate challenges by unveiling cryptic 3'-UTR sequences, but also boost the potential for future genetic adaptations.

© 2015 Elsevier Ltd. All rights reserved. Epistasis is a key factor in evolution since it determines which combinations of mutations provide adaptive solutions and which mutational pathways toward these solutions are accessible by natural selection. There is growing evidence for the pervasiveness of sign epistasis - a complete reversion of mutational effects, particularly in protein evolution - yet its molecular basis remains poorly understood. We describe the structural basis of sign epistasis between G238S and R164S, two adaptive mutations in TEM-1 β-lactamase - an enzyme that endows antibiotics resistance. Separated by 10 Å, these mutations initiate two separate trajectories toward increased hydrolysis rates and resistance toward second and third-generation cephalosporins antibiotics. Both mutations allow the enzyme's active site to adopt alternative conformations and accommodate the new antibiotics. By solving the corresponding set of crystal structures, we found that R164S causes local disorder whereas G238S induces discrete conformations. When combined, the mutations in 238 and 164 induce local disorder whereby nonproductive conformations that perturb the enzyme's catalytic preorganization dominate. Specifically, Asn170 that coordinates the deacylating water molecule is misaligned, in both the free form and the inhibitor-bound double mutant. This local disorder is not restored by stabilizing global suppressor mutations and thus leads to an evolutionary cul-de-sac. Conformational dynamism therefore underlines the reshaping potential of protein's structures and functions but also limits protein evolvability because of the fragility of the interactions networks that maintain protein structures.

New protein folds have emerged throughout evolution, but it remains unclear how a protein fold can evolve while maintaining its function, particularly when fold changes require several sequential gene rearrangements. Here, we explored hypothetical evolutionary pathways linking different topological families of the DNA-methyltransferase superfamily. These pathways entail successive gene rearrangements through a series of intermediates, all of which should be sufficiently active to maintain the organism's fitness. By means of directed evolution, and starting from HaeIII methyltransferase (M.HaeIII), we selected all the required intermediates along these paths (a duplicated fused gene and duplicates partially truncated at their 5' or 3' coding regions) that maintained function in vivo. These intermediates led to new functional genes that resembled natural methyltransferases from three known classes or that belonged to a new class first seen in our evolution experiments and subsequently identified in natural genomes. Our findings show that new protein topologies can evolve gradually through multistep gene rearrangements and provide new insights regarding these processes.

The amidohydrolase superfamily comprises hundreds of hydrolytic enzymes of the (beta/alpha)8 barrel fold with mono- or binuclear active-site metal centers, and a diverse spectrum of substrates and reactions. Promiscuous activities, or cross-reactivities, between different members of the same superfamily may provide important hints regarding evolutionary and mechanistic relationships. We examined three members: dihydroorotase (DHO), phosphotriesterase (PTE), and PTE-homology protein (PHP). Of particular interest are PTE, which is thought to have evolved within the last several decades, and PHP, an amidohydrolase superfamily member of unknown function, and the closest known homologue of PTE. We found a diverse and partially overlapping pattern of promiscuous activities in these enzymes, including a significant lactonase activity in PTE, esterase activities in both PTE and PHP, and a weak PTE activity in DHO. Directed evolution was applied to improve the promiscuous esterase activities of PTE and PHP. Remarkably, the most recurrent mutation increasing esterase activity in PTE, or PHP, maps to the same location in their superposed 3D structures. The evolved variants also exhibit newly acquired promiscuous activities that were not selected for, including very weak, yet measurable, paraoxonase activity in PHP. Our results illustrate the mechanistic, structural, and evolutionary links between these enzymes, and highlight the importance of studying laboratory evolution intermediates that might resemble node intermediates along the evolutionary pathways leading to the divergence of enzyme superfamilies.

The availability of vast gene repertoires from both natural sources (genomic and cDNA libraries) and artificial sources (gene libraries) demands the development and application of novel technologies that enable the screening or selection of large libraries for a variety of enzymatic activities. We describe recent developments in the selection of enzyme-coding genes for directed evolution and functional genomics. We focus on HTS approaches that enable selection from large libraries (10(6) gene variants) with relatively humble means (i.e. non-robotic systems), and on in vitro compartmentalization in particular.

How proteins with new functions (e.g., drug or antibiotic resistance or degradation of man-made chemicals) evolve in a matter of months or years is still unclear. This ability is dependent on the induction of new phenotypic traits by a small number of mutations (plasticity). But mutations often have deleterious effects on functions that are essential for survival. How are these seemingly conflicting demands met at the single-protein level? Results from directed laboratory evolution experiments indicate that the evolution of a new function is driven by mutations that have little effect on the native function but large effects on the promiscuous functions that serve as starting point. Thus, an evolving protein can initially acquire increased fitness for a new function without losing its original function. Gene duplication and the divergence of a completely new protein may then follow.

Members of the serum paraoxonase (PON) family have been identified in mammals and other vertebrates, and in invertebrates. PONs exhibit a wide range of physiologically important hydrolytic activities, including drug metabolism and detoxification of nerve agents. PON1 and PON3 reside on high-density lipoprotein (HDL, 'good cholesterol') and are involved in the prevention of atherosclerosis. We describe the first crystal structure of a PON family member, a variant of PON1 obtained by directed evolution, at a resolution of 2.2 A. PON1 is a six-bladed beta-propeller with a unique active site lid that is also involved in HDL binding. The three-dimensional structure and directed evolution studies permit a detailed description of PON1's active site and catalytic mechanism, which are reminiscent of secreted phospholipase A2, and of the routes by which PON family members diverged toward different substrate and reaction selectivities.

In vitro compartmentalisation (IVC), a technique for selecting genes encoding enzymes based on compartmentalising gene translation and enzymatic reactions in emulsions, was used to investigate the interaction of the DNA cytosine-5 methyltransferase M.HhaI with its target DNA (5′-GCGC-3′). Crystallog raphy shows that the active site loop from the large domain of M.HhaI interacts with a flipped-out cytosine (the target for methylation) and two target recognition loops (loops I and II) from the small domain make almost all the other base-specific interactions. A library of M.HhaI genes was created by randomising all the loop II residues thought to make base-specific interactions and directly determine target specificity. The library was selected for 5′-GCGC-3′ methylation. Interestingly, in 11 selected active clones, 10 different sequences were found and none were wild-type. At two of the positions mutated (Ser252 and Tyr254) a number of different amino acids could be tolerated. At the third position, however, all active mutants had a glycine, as in wild-type M.HhaI, suggesting that Gly257 is crucial for DNA recognition and enzyme activity. Our results suggest that recognition of base pairs 3 and 4 of the target site either relies entirely on main chain interactions or that different residues from those identified in the crystal structure contribute to DNA recognition.

The cytosine C5 methyltransferase M.HaeIII recognises and methylates the central cytosine of its canonical site GGCC. Here we report that M.HaeIII can also, with lower efficiency, methylate cytosines located in a wide range of non-canonical sequences. Using bisulphite sequencing we mapped the methyl- cytosine residues in DNA methylated in vitro and in vivo by M.HaeIII. Methyl-cytosine residues were observed in multiple sequence contexts, most commonly, but not exclusively, at star sites (sites differing by a single base from the canonical sequence). The most frequently used star sites had changes at positions 1 and 4, but there is little or no methylation at star sites changed at position 2. The rate of methylation of non-canonical sites can be quite significant: a DNA substrate lacking a canonical site was methylated by M.HaeIII in vitro at a rate only an order of magnitude slower than an otherwise identical substrate containing the canonical site. In vivo methylation of non-canonical sites may therefore be significant and may have provided the starting point for the evolution of restriction-modification systems with novel sequence specificities.

Cellular compartmentalization is vital for the evolution of all living organisms. Cells keep together the genes, the RNAs and proteins that they encode, and the products of their activities, thus linking genotype to phenotype. We have reproduced this linkage in the test tube by transcribing and translating single genes in the aqueous compartments of water-in-oil emulsions. These compartments, with volumes close to those of bacteria, can be recruited to select genes encoding catalysts. A protein or RNA with a desired catalytic activity converts a substrate attached to the gene that encodes it to product. In other compartments, substrates attached to genes that do not encode catalysts remain unmodified. Subsequently, genes encoding catalysts are selectively enriched by virtue of their linkage to the product. We demonstrate the linkage of genotype to phenotype in man-made compartments using a model system. A selection for target-specific DNA methylation was based on the resistance of the product (methylated DNA) to restriction digestion. Genes encoding HaeIII methyltransferase were selected from a 10 7 -fold excess of genes encoding another enzyme.

Backbone modifications via insertions and deletions (InDels) may exert dramatic effects, for better (mediating new functions) and for worse (causing loss of structure and/or function). However, contrary to point mutations (substitutions), our knowledge of the evolution and structural‐functional effects of InDels is limited and so is our capability to engineer them. We sought to assess how deleterious InDels are relative to point mutations and understand the mechanisms that mediate their acceptance. Analysis of the evolution of InDels in orthologous protein phylogenies indicated that their rate of purging is 9- to 100-fold higher than for point mutations. In yeast, for example, the substitutions-to-InDels ratio is approximately 14-fold higher in protein coding than in noncoding regions. The incorporation of InDels relative to substitutions is not only slow but also nonlinear. On average, � 50 substitutions accumulate before the appearance of the first InDel. We also found enriched substitutions in sequential and spatial proximity to InDels, suggesting that certain substitutions are correlated with InDels. As indicated by the lag in InDels accumulation, some of these correlated substitutions may have occurred first, as apparently neutral mutations, and later enabled the accumulation of InDels that would be otherwise purged. Thus, compensatory substitutions may follow InDels in an “adaptive walk” as traditionally assumed, but might also accumulate first, by “neutral roaming.” The dynamics of InDels accumulation also depends on their genomic frequencies—InDels in flies are 4-fold more frequent than in yeast and tend to be compensated rather than enabled.

This review outlines the strategies we apply for directed enzyme evolution using targeted libraries, namely, libraries that diversify specific residues with predefined mutational compositions. The theoretical grounds underlining the design of such libraries are described, including the mutational load, the ratio of beneficial versus deleterious mutations, and screening capacity. We point out the advantage of using mutational spiking strategies for "hedging the bets," exploring a large number of potentially beneficial mutations, and tuning the library's mutational load. Also highlighted are the merits of low-throughput screens that measure multiple parameters at high accuracy, and of using the desired substrate and reaction conditions rather than surrogates. We subsequently describe library construction strategies (rational and analytical) based on structure and sequence analyses, including ancestral libraries, which are particularly suitable for low-throughput screens. We also discuss the critical role of including compensatory, stabilizing mutations during library construction. Finally, the design efficiency and the optimal mutational loads of libraries are assessed by comparing targeted mutational libraries versus libraries of random mutations.

Optimization processes, such as evolution, are constrained by diminishing returns-the closer the optimum, the smaller the benefit per mutation, and by tradeoffs-improvement of one property at the cost of others. However, the magnitude and molecular basis of these parameters, and their effect on evolutionary transitions, remain unknown. Here we pursue a complete functional transition of an enzyme with a >10(9)-fold change in the enzyme's selectivity using laboratory evolution. We observed strong diminishing returns, with the initial mutations conferring >25-fold higher improvements than later ones, and asymmetric tradeoffs whereby the gain/loss ratio of the new/old activity decreased 400-fold from the beginning of the trajectory to its end. We describe the molecular basis for these phenomena and suggest they have an important role in shaping natural proteins. These findings also suggest that the catalytic efficiency and specificity of many natural enzymes may be far from their optimum.

Biological processes are often executed by a small number of molecules per individual cell, leading to significant cell-to-cell variability (“noise”) in gene expression (Paulsson 2004; Maheshri and O'Shea 2007; Raj and van Oudenaarden 2008; Tawfik 2010). When analyzing gene expression noise, it is convenient to distinguish between intrinsic variations, resulting from stochastic production, and extrinsic variations propagating from global (e.g., ribosomes, polymerases, metabolites, etc.) or pathway-specific factors (Elowitz et al. 2002). Intrinsic noise is of particular interest, as it reflects on the transcription process itself (Paulsson 2004, 2005; Raser and O'Shea 2004; Raj and van Oudenaarden 2008; Rinott et al. 2011; So et al. 2011). The prevailing model of gene expression noise assumes that genes transit stochastically between states that are permissive or nonpermissive to transcription (Paulsson 2004, 2005; Raser and O'Shea 2004; Friedman et al. 2006; Raj et al. 2006; Zenklusen et al. 2008; So et al. 2011). This two-state model predicts a scaling relationship between mean expression m and the coefficient of variation (noise, SD/mean) η: η2 = b/m + ηext2 where b is the typical number of protein molecules made during a single “on” state (“burst size”) and ηext denotes the extrinsic noise (Paulsson 2004, 2005; Raser and O'Shea 2004; Bar-Even et al. 2006; Raj et al. 2006; Pedraza and Paulsson 2008; Tan and van Oudenaarden 2010; Taniguchi et al. 2010). Note that burst size, b, accounts for all the transcription-translation processes following the main stochastic event (burst initiation), integrating the number of mRNA molecules produced per burst and the number of protein molecules made per each mRNA molecule. Upon a perturbation, the noise–mean relationship may change, depending on whether burst size or burst frequency were modulated (Pedraza and Paulsson 2008; Zenklusen et al. 2008; Tan and van Oudenaarden 2010). Genome-wide analysis of the noise–mean relationship in yeast (Bar-Even et al. 2006; Newman et al. 2006) or Escherichia coli (Taniguchi et al. 2010; So et al. 2011) genes reported a general dependency that was well defined by the scaling relation η2 = b/m + ηext2, suggesting a similar burst size for many genes. The expression of genes deviating from the scaling curve, displaying higher-than-expected noise (Bar-Even et al. 2006; Newman et al. 2006), was more responsive to changing conditions and also diverged more between related species (Tirosh and Barkai 2008; Choi and Kim 2009; Lehner 2010). Notably, high noise, responsiveness, and divergence were all correlated with the organization of gene promoters: All three measures were low in promoters lacking a TATA box and containing a nucleosome free region (NFR) proximal to the transcription start site (referred to as DPN promoters—depleted proximal nucleosome), and were high in TATA-containing promoters that lack NFR (OPN, occupied proximal nucleosome) (Field et al. 2008; Tirosh and Barkai 2008; Choi and Kim 2009). TATA box was also shown to increase noise in Pho5 expression (Raser and O'Shea 2004) and in synthetic promoters (Blake et al. 2006; Murphy et al. 2010). The observation that genes with a characteristic promoter structure had a high noise (relative to that expected given their mean expression) is consistent with the idea that promoter sequence influences not only burst frequency but also burst size. Still, the principles by which promoter sequences regulate those two processes are not understood, primarily because most studies analyzing the interplay between promoter sequence variations and gene expression consider mean expression only (e.g., Yun et al. 2012). To distinguish between the effects of promoter sequence on burst size and burst frequency, we generated large libraries of sequence-mutated promoters. Specifically, we chose 22 yeast promoters that span a range of expression and noise levels. Using mutagenic PCR, we generated hundreds of sequence variants of each promoter. Each variant was fused to a fluorescent reporter, and the associated mean expression and noise (coefficient of variation) in a population of identical cells were measured. We found that sequence variants in each of the libraries defined a scaling curve η2 = b/m + ηext2, with a constant estimated burst size b, that was largely promoter-specific and was particularly large for OPN promoters containing a TATA box. A small fraction of sequence mutations leading to a large change in burst size was identified in the OPN-type promoters containing a TATA box. These changes were biased toward reducing burst size, and were almost fully explained by elimination of a TATA box or insertion of a new out-of-frame translation start site. Interestingly, mutations that deleted a TATA box in low-noise DPN-type promoters did not reduce burst size. Our results suggest that burst size is a promoter-specific property that is insensitive to most sequence mutations but is largely influenced by the interaction between TATA box and promoter nucleosomes.