The genetic landscape of a metabolic interaction.

While much prior work has explored the constraints on protein sequence and evolution induced by physical protein-protein interactions, the sequence-level constraints emerging from non-binding functional interactions in metabolism remain unclear. To quantify how variation in the activity of one enzyme constrains the biochemical parameters and sequence of another, we focus on dihydrofolate reductase (DHFR) and thymidylate synthase (TYMS), a pair of enzymes catalyzing consecutive reactions in folate metabolism. We use deep mutational scanning to quantify the growth rate effect of 2696 DHFR single mutations in 3 TYMS backgrounds under conditions selected to emphasize biochemical epistasis. Our data are well-described by a relatively simple enzyme velocity to growth rate model that quantifies how metabolic context tunes enzyme mutational tolerance. Together our results reveal the structural distribution of epistasis in a metabolic enzyme and establish a foundation for the design of multi-enzyme systems.

A continuous epistasis model for predicting growth rate given combinatorial variation in gene expression and environment.

Quantifying and predicting growth rate phenotype given variation in gene expression and environment is complicated by epistatic interactions and the vast combinatorial space of possible perturbations. We developed an approach for mapping expression-growth rate landscapes that integrates sparsely sampled experimental measurements with an interpretable machine learning model. We used mismatch CRISPRi across pairs and triples of genes to create over 8,000 titrated changes in E. coli gene expression under varied environmental contexts, exploring epistasis in up to 22 distinct environments. Our results show that a pairwise model previously used to describe drug interactions well-described these data. The model yielded interpretable parameters related to pathway architecture and generalized to predict the combined effect of up to four perturbations when trained solely on pairwise perturbation data. We anticipate this approach will be broadly applicable in optimizing bacterial growth conditions, generating pharmacogenomic models, and understanding the fundamental constraints on bacterial gene expression. A record of this paper's transparent peer review process is included in the supplemental information.

Engineering Proteins Using Statistical Models of Coevolutionary Sequence Information.

Homologous protein sequences are wonderfully diverse, indicating many possible evolutionary "solutions" to the encoding of function. Consequently, one can construct statistical models of protein sequence by analyzing amino acid frequency across a large multiple sequence alignment. A central premise is that covariance between amino acid positions reflects coevolution due to a shared functional or biophysical constraint. In this review, we describe the implementation and discuss the advantages, limitations, and recent progress on two coevolution-based modeling approaches: (1) Potts models of protein sequence (direct coupling analysis [DCA]-like), and (2) the statistical coupling analysis (SCA). Each approach detects interesting features of protein sequence and structure-the former emphasizes local physical contacts throughout the structure, while the latter identifies larger evolutionarily coupled networks of residues. Recent advances in large-scale gene synthesis and high-throughput functional selection now motivate additional work to benchmark model performance across quantitative function prediction and de novo design tasks.

The Genetic Landscape of a Metabolic Interaction.

Enzyme abundance, catalytic activity, and ultimately sequence are all shaped by the need of growing cells to maintain metabolic flux while minimizing accumulation of deleterious intermediates. While much prior work has explored the constraints on protein sequence and evolution induced by physical protein-protein interactions, the sequence-level constraints emerging from non-binding functional interactions in metabolism remain unclear. To quantify how variation in the activity of one enzyme constrains the biochemical parameters and sequence of another, we focused on dihydrofolate reductase (DHFR) and thymidylate synthase (TYMS), a pair of enzymes catalyzing consecutive reactions in folate metabolism. We used deep mutational scanning to quantify the growth rate effect of 2,696 DHFR single mutations in 3 TYMS backgrounds under conditions selected to emphasize biochemical epistasis. Our data are well-described by a relatively simple enzyme velocity to growth rate model that quantifies how metabolic context tunes enzyme mutational tolerance. Together our results reveal the structural distribution of epistasis in a metabolic enzyme and establish a foundation for the design of multi-enzyme systems.

Accelerating Biological Insight for Understudied Genes.

The rapid expansion of genome sequence data is increasing the discovery of protein-coding genes across all domains of life. Annotating these genes with reliable functional information is necessary to understand evolution, to define the full biochemical space accessed by nature, and to identify target genes for biotechnology improvements. The majority of proteins are annotated based on sequence conservation with no specific biological, biochemical, genetic, or cellular function identified. Recent technical advances throughout the biological sciences enable experimental research on these understudied protein-coding genes in a broader collection of species. However, scientists have incentives and biases to continue focusing on well documented genes within their preferred model organism. This perspective suggests a research model that seeks to break historic silos of research bias by enabling interdisciplinary teams to accelerate biological functional annotation. We propose an initiative to develop coordinated projects of collaborating evolutionary biologists, cell biologists, geneticists, and biochemists that will focus on subsets of target genes in multiple model organisms. Concurrent analysis in multiple organisms takes advantage of evolutionary divergence and selection, which causes individual species to be better suited as experimental models for specific genes. Most importantly, multisystem approaches would encourage transdisciplinary critical thinking and hypothesis testing that is inherently slow in current biological research.

Structurally distributed surface sites tune allosteric regulation.

Our ability to rationally optimize allosteric regulation is limited by incomplete knowledge of the mutations that tune allostery. Are these mutations few or abundant, structurally localized or distributed? To examine this, we conducted saturation mutagenesis of a synthetic allosteric switch in which Dihydrofolate reductase (DHFR) is regulated by a blue-light sensitive LOV2 domain. Using a high-throughput assay wherein DHFR catalytic activity is coupled to E. coli growth, we assessed the impact of 1548 viable DHFR single mutations on allostery. Despite most mutations being deleterious to activity, fewer than 5% of mutations had a statistically significant influence on allostery. Most allostery disrupting mutations were proximal to the LOV2 insertion site. In contrast, allostery enhancing mutations were structurally distributed and enriched on the protein surface. Combining several allostery enhancing mutations yielded near-additive improvements to dynamic range. Our results indicate a path toward optimizing allosteric function through variation at surface sites.

Many proteins exhibit a property called 'allostery'. In allostery, an input signal at a specific site of a protein ­ such as a molecule binding, or the protein absorbing a photon of light ­ leads to a change in output at another site far away. For example, the protein might catalyze a chemical reaction faster or bind to another molecule more tightly in the presence of the input signal. This protein 'remote control' allows cells to sense and respond to changes in their environment. An ability to rapidly engineer new allosteric mechanisms into proteins is much sought after because this would provide an approach for building biosensors and other useful tools. One common approach to engineering new allosteric regulation is to combine a 'sensor' or input region from one protein with an 'output' region or domain from another. When researchers engineer allostery using this approach of combining input and output domains from different proteins, the difference in the output when the input is 'on' versus 'off' is often small, a situation called 'modest allostery'. McCormick et al. wanted to know how to optimize this domain combination approach to increase the difference in output between the 'on' and 'off' states. More specifically, McCormick et al. wanted to find out whether swapping out or mutating specific amino acids (each of the individual building blocks that make up a protein) enhances or disrupts allostery. They also wanted to know if there are many possible mutations that change the effectiveness of allostery, or if this property is controlled by just a few amino acids. Finally, McCormick et al. questioned where in a protein most of these allostery-tuning mutations were located. To answer these questions, McCormick et al. engineered a new allosteric protein by inserting a light-sensing domain (input) into a protein involved in metabolism (a metabolic enzyme that produces a biomolecule called a tetrahydrofolate) to yield a light-controlled enzyme. Next, they introduced mutations into both the 'input' and 'output' domains to see where they had a greater effect on allostery. After filtering out mutations that destroyed the function of the output domain, McCormick et al. found that only about 5% of mutations to the 'output' domain altered the allosteric response of their engineered enzyme. In fact, most mutations that disrupted allostery were found near the site where the 'input' domain was inserted, while mutations that enhanced allostery were sprinkled throughout the enzyme, often on its protein surface. This was surprising in light of the commonly-held assumption that mutations on protein surfaces have little impact on the activity of the 'output' domain. Overall, the effect of individual mutations on allostery was small, but McCormick et al. found that these mutations can sometimes be combined to yield larger effects. McCormick et al.'s results suggest a new approach for optimizing engineered allosteric proteins: by introducing mutations on the protein surface. It also opens up new questions: mechanically, how do surface sites affect allostery? In the future, it will be important to characterize how combinations of mutations can optimize allosteric regulation, and to determine what evolutionary trajectories to high performance allosteric 'switches' look like.

A simplified strategy for titrating gene expression reveals new relationships between genotype, environment, and bacterial growth.

A lack of high-throughput techniques for making titrated, gene-specific changes in expression limits our understanding of the relationship between gene expression and cell phenotype. Here, we present a generalizable approach for quantifying growth rate as a function of titrated changes in gene expression level. The approach works by performing CRISPRi with a series of mutated single guide RNAs (sgRNAs) that modulate gene expression. To evaluate sgRNA mutation strategies, we constructed a library of 5927 sgRNAs targeting 88 genes in Escherichia coli MG1655 and measured the effects on growth rate. We found that a compounding mutational strategy, through which mutations are incrementally added to the sgRNA, presented a straightforward way to generate a monotonic and gradated relationship between mutation number and growth rate effect. We also implemented molecular barcoding to detect and correct for mutations that 'escape' the CRISPRi targeting machinery; this strategy unmasked deleterious growth rate effects obscured by the standard approach of ignoring escapers. Finally, we performed controlled environmental variations and observed that many gene-by-environment interactions go completely undetected at the limit of maximum knockdown, but instead manifest at intermediate expression perturbation strengths. Overall, our work provides an experimental platform for quantifying the phenotypic response to gene expression variation.

Altered expression of a quality control protease in E. coli reshapes the in vivo mutational landscape of a model enzyme.

Protein mutational landscapes are shaped by the cellular environment, but key factors and their quantitative effects are often unknown. Here we show that Lon, a quality control protease naturally absent in common E. coli expression strains, drastically reshapes the mutational landscape of the metabolic enzyme dihydrofolate reductase (DHFR). Selection under conditions that resolve highly active mutants reveals that 23.3% of all single point mutations in DHFR are advantageous in the absence of Lon, but advantageous mutations are largely suppressed when Lon is reintroduced. Protein stability measurements demonstrate extensive activity-stability tradeoffs for the advantageous mutants and provide a mechanistic explanation for Lon's widespread impact. Our findings suggest possibilities for tuning mutational landscapes by modulating the cellular environment, with implications for protein design and combatting antibiotic resistance.

Strategies for Engineering and Rewiring Kinase Regulation.

Eukaryotic protein kinases (EPKs) catalyze the transfer of a phosphate group onto another protein in response to appropriate regulatory cues. In doing so, they provide a primary means for cellular information transfer. Consequently, EPKs play crucial roles in cell differentiation and cell-cycle progression, and kinase dysregulation is associated with numerous disease phenotypes including cancer. Nonnative cues for synthetically regulating kinases are thus much sought after, both for dissecting cell signaling pathways and for pharmaceutical development. In recent years advances in protein engineering and sequence analysis have led to new approaches for manipulating kinase activity, localization, and in some instances specificity. These tools have revealed fundamental principles of intracellular signaling and suggest paths forward for the design of therapeutic allosteric kinase regulators.

A Two-Enzyme Adaptive Unit within Bacterial Folate Metabolism.

Enzyme function and evolution are influenced by the larger context of a metabolic pathway. Deleterious mutations or perturbations in one enzyme can often be compensated by mutations to others. We used comparative genomics and experiments to examine evolutionary interactions with the essential metabolic enzyme dihydrofolate reductase (DHFR). Analyses of synteny and co-occurrence across bacterial species indicate that DHFR is coupled to thymidylate synthase (TYMS) but relatively independent from the rest of folate metabolism. Using quantitative growth rate measurements and forward evolution in Escherichia coli, we demonstrate that the two enzymes adapt as a relatively independent unit in response to antibiotic stress. Metabolomic profiling revealed that TYMS activity must not exceed DHFR activity to prevent the depletion of reduced folates and the accumulation of the intermediate dihydrofolate. Comparative genomics analyses identified >200 gene pairs with similar statistical signatures of modular co-evolution, suggesting that cellular pathways may be decomposable into small adaptive units.

High-Order Epistasis in Catalytic Power of Dihydrofolate Reductase Gives Rise to a Rugged Fitness Landscape in the Presence of Trimethoprim Selection.

Evolutionary fitness landscapes of several antibiotic target proteins have been comprehensively mapped showing strong high-order epistasis between mutations, but understanding these effects at the biochemical and structural levels remained open. Here, we carried out an extensive experimental and computational study to quantitatively understand the evolutionary dynamics of Escherichia coli dihydrofolate reductase (DHFR) enzyme in the presence of trimethoprim-induced selection. To facilitate this, we developed a new in vitro assay for rapidly characterizing DHFR steady-state kinetics. Biochemical and structural characterization of resistance-conferring mutations targeting a total of ten residues spanning the substrate binding pocket of DHFR revealed distinct changes in the catalytic efficiencies of mutated DHFR enzymes. Next, we measured biochemical parameters (Km, Ki, and kcat) for a mutant library carrying all possible combinations of six resistance-conferring DHFR mutations and quantified epistatic interactions between them. We found that the high-order epistasis in catalytic power of DHFR (kcat and Km) creates a rugged fitness landscape under trimethoprim selection. Taken together, our data provide a concrete illustration of how epistatic coupling at the level of biochemical parameters can give rise to complex fitness landscapes, and suggest new strategies for developing mutant specific inhibitors.

Engineering allosteric regulation in protein kinases.

Phosphoregulation, in which the addition of a negatively charged phosphate group modulates protein activity, enables dynamic cellular responses. To understand how new phosphoregulation might be acquired, we mutationally scanned the surface of a prototypical yeast kinase (Kss1) to identify potential regulatory sites. The data revealed a set of spatially distributed "hotspots" that might have coevolved with the active site and preferentially modulated kinase activity. By engineering simple consensus phosphorylation sites at these hotspots, we rewired cell signaling in yeast. Using the same approach with a homolog yeast mitogen-activated protein kinase, Hog1, we introduced new phosphoregulation that modified its localization and signaling dynamics. Beyond revealing potential use in synthetic biology, our findings suggest that the identified hotspots contribute to the diversity of natural allosteric regulatory mechanisms in the eukaryotic kinome and, given that some are mutated in cancers, understanding these hotspots may have clinical relevance to human disease.

An evolutionary hotspot defines functional differences between CRYPTOCHROMES.

Mammalian circadian clocks are driven by a transcription/translation feedback loop composed of positive regulators (CLOCK/BMAL1) and repressors (CRYPTOCHROME 1/2 (CRY1/2) and PER1/2). To understand the structural principles of regulation, we used evolutionary sequence analysis to identify co-evolving residues within the CRY/PHL protein family. Here we report the identification of an ancestral secondary cofactor-binding pocket as an interface in repressive CRYs, mediating regulation through direct interaction with CLOCK and BMAL1. Mutations weakening binding between CLOCK/BMAL1 and CRY1 lead to acceleration of the clock, suggesting that subtle sequence divergences at this site can modulate clock function. Divergence between CRY1 and CRY2 at this site results in distinct periodic output. Weaker interactions between CRY2 and CLOCK/BMAL1 at this pocket are strengthened by co-expression of PER2, suggesting that PER expression limits the length of the repressive phase in CRY2-driven rhythms. Overall, this work provides a model for the mechanism and evolutionary variation of clock regulatory mechanisms.

Conserved amino acid networks modulate discrete functional properties in an enzyme superfamily.

In this work, we applied the sequence-based statistical coupling analysis approach to characterize conserved amino acid networks important for biochemical function in the pancreatic-type ribonuclease (ptRNase) superfamily. This superfamily-wide analysis indicates a decomposition of the RNase tertiary structure into spatially distributed yet physically connected networks of co-evolving amino acids, termed sectors. Comparison of this statistics-based description with new NMR experiments data shows that discrete amino acid networks, termed sectors, control the tuning of distinct functional properties in different enzyme homologs. Further, experimental characterization of evolutionarily distant sequences reveals that sequence variation at sector positions can distinguish homologs with a conserved dynamic pattern and optimal catalytic activity from those with altered dynamics and diminished catalytic activities. Taken together, these results provide important insights into the mechanistic design of the ptRNase superfamily, and presents a structural basis for evolutionary tuning of function in functionally diverse enzyme homologs.

An evolution-based strategy for engineering allosteric regulation.

Allosteric regulation provides a way to control protein activity at the time scale of milliseconds to seconds inside the cell. An ability to engineer synthetic allosteric systems would be of practical utility for the development of novel biosensors, creation of synthetic cell signaling pathways, and design of small molecule pharmaceuticals with regulatory impact. To this end, we outline a general approach-termed rational engineering of allostery at conserved hotspots (REACH)-to introduce novel regulation into a protein of interest by exploiting latent allostery that has been hard-wired by evolution into its structure. REACH entails the use of statistical coupling analysis (SCA) to identify 'allosteric hotspots' on protein surfaces, the development and implementation of experimental assays to test hotspots for functionality, and a toolkit of allosteric modulators to impinge on endogenous cellular circuitry. REACH can be broadly applied to rewire cellular processes to respond to novel inputs.

Evolution-Based Functional Decomposition of Proteins.

The essential biological properties of proteins-folding, biochemical activities, and the capacity to adapt-arise from the global pattern of interactions between amino acid residues. The statistical coupling analysis (SCA) is an approach to defining this pattern that involves the study of amino acid coevolution in an ensemble of sequences comprising a protein family. This approach indicates a functional architecture within proteins in which the basic units are coupled networks of amino acids termed sectors. This evolution-based decomposition has potential for new understandings of the structural basis for protein function. To facilitate its usage, we present here the principles and practice of the SCA and introduce new methods for sector analysis in a python-based software package (pySCA). We show that the pattern of amino acid interactions within sectors is linked to the divergence of functional lineages in a multiple sequence alignment-a model for how sector properties might be differentially tuned in members of a protein family. This work provides new tools for studying proteins and for generally testing the concept of sectors as the principal units of function and adaptive variation.

Melting of Duplex DNA in the Absence of ATP by the NS3 Helicase Domain through Specific Interaction with a Single-Strand/Double-Strand Junction.

Helicases unwind double-stranded nucleic acids, remove secondary structures from single-stranded nucleic acids, and remove proteins bound to nucleic acids. For many helicases, the mechanisms for these different functions share the ability to translocate with a directional bias as a result of ATP binding and hydrolysis. Nonstructural protein 3 (NS3) is an essential enzyme expressed by the hepatitis C virus (HCV) and is known to catalyze the unwinding of both DNA and RNA substrates in a 3'-to-5' direction. We investigated the role of nucleic acid binding in the unwinding mechanism by examining ATP-independent unwinding. We observed that even in the absence of ATP, the NS3 helicase domain (NS3h) unwound duplexes only when they contained a 3'-tail (i.e., 3'-to-5' directionality). Blunt-ended duplexes and 5'-tailed duplexes were not melted even in the presence of a large excess concentration of the protein. NS3h was found to diffuse rapidly along single-stranded DNA at a rate of 30 nucleotides(2) s(-1). Upon encountering an appropriate single-strand/double-strand (ss/ds) junction, NS3h slowly melted the duplex under conditions with an excess protein concentration relative to DNA concentration. When a biotin-streptavidin block was placed into the ssDNA region, no melting of DNA was observed, suggesting that NS3h must diffuse along the ssDNA, and that the streptavidin blocked the diffusion. We conclude that the specific interaction between NS3h and the ss/dsDNA junction, coupled with diffusion, allows binding energy to melt duplex DNA with a directional bias. Alternatively, we found that the full-length NS3 protein did not exhibit strict directionality and was dependent on duplex DNA length. NS3 was able to unwind the duplex even in the presence of the biotin-streptavidin block. We propose a noncanonical model of unwinding for NS3 in which the enzyme binds directly to the duplex via protein-protein interactions to melt the substrate.

A new test of computational protein design: predicting posttranslational modification specificity for the enzyme SMYD2.

In this issue of Structure, Lanouette and colleagues use a combination of computation and experiment to define a specificity motif for the lysine methyltransferase SMYD2. Using this motif, they predict and experimentally verify four new SMYD2 substrates.

Probing RNA translocases with DNA.

For some helicases, it is possible to investigate RNA translocase activity on DNA substrates because the enzyme acts on both substrates. Potassium permanganate (KMnO4) footprinting is a method used to chemically probe the conformation of DNA as well as the binding of proteins. Combining footprinting methods with rapid mixing methods that utilize a chemical quench-flow instrument can enable tracking of the translocase with nucleotide resolution.

Finding a common path: predicting gene function using inferred evolutionary trees.

Reporting in Cell, Li and colleagues (2014) describe an innovative method to functionally classify genes using evolutionary information. This approach demonstrates broad utility for eukaryotic gene annotation and suggests an intriguing new decomposition of pathways and complexes into evolutionarily conserved modules.