Illuminating protein space with a programmable generative model.

Three billion years of evolution has produced a tremendous diversity of protein molecules1, but the full potential of proteins is likely to be much greater. Accessing this potential has been challenging for both computation and experiments because the space of possible protein molecules is much larger than the space of those likely to have functions. Here we introduce Chroma, a generative model for proteins and protein complexes that can directly sample novel protein structures and sequences, and that can be conditioned to steer the generative process towards desired properties and functions. To enable this, we introduce a diffusion process that respects the conformational statistics of polymer ensembles, an efficient neural architecture for molecular systems that enables long-range reasoning with sub-quadratic scaling, layers for efficiently synthesizing three-dimensional structures of proteins from predicted inter-residue geometries and a general low-temperature sampling algorithm for diffusion models. Chroma achieves protein design as Bayesian inference under external constraints, which can involve symmetries, substructure, shape, semantics and even natural-language prompts. The experimental characterization of 310 proteins shows that sampling from Chroma results in proteins that are highly expressed, fold and have favourable biophysical properties. The crystal structures of two designed proteins exhibit atomistic agreement with Chroma samples (a backbone root-mean-square deviation of around 1.0 Å). With this unified approach to protein design, we hope to accelerate the programming of protein matter to benefit human health, materials science and synthetic biology.

Empowering drug off-target discovery with metabolic and structural analysis.

Elucidating intracellular drug targets is a difficult problem. While machine learning analysis of omics data has been a promising approach, going from large-scale trends to specific targets remains a challenge. Here, we develop a hierarchic workflow to focus on specific targets based on analysis of metabolomics data and growth rescue experiments. We deploy this framework to understand the intracellular molecular interactions of the multi-valent dihydrofolate reductase-targeting antibiotic compound CD15-3. We analyse global metabolomics data utilizing machine learning, metabolic modelling, and protein structural similarity to prioritize candidate drug targets. Overexpression and in vitro activity assays confirm one of the predicted candidates, HPPK (folK), as a CD15-3 off-target. This study demonstrates how established machine learning methods can be combined with mechanistic analyses to improve the resolution of drug target finding workflows for discovering off-targets of a metabolic inhibitor.

Switching an active site helix in dihydrofolate reductase reveals limits to subdomain modularity.

To what degree are individual structural elements within proteins modular such that similar structures from unrelated proteins can be interchanged? We study subdomain modularity by creating 20 chimeras of an enzyme, Escherichia coli dihydrofolate reductase (DHFR), in which a catalytically important, 10-residue α-helical sequence is replaced by α-helical sequences from a diverse set of proteins. The chimeras stably fold but have a range of diminished thermal stabilities and catalytic activities. Evolutionary coupling analysis indicates that the residues of this α-helix are under selection pressure to maintain catalytic activity in DHFR. Reversion to phenylalanine at key position 31 was found to partially restore catalytic activity, which could be explained by evolutionary coupling values. We performed molecular dynamics simulations using replica exchange with solute tempering. Chimeras with low catalytic activity exhibit nonhelical conformations that block the binding site and disrupt the positioning of the catalytically essential residue D27. Simulation observables and in vitro measurements of thermal stability and substrate-binding affinity are strongly correlated. Several E. coli strains with chromosomally integrated chimeric DHFRs can grow, with growth rates that follow predictions from a kinetic flux model that depends on the intracellular abundance and catalytic activity of DHFR. Our findings show that although α-helices are not universally substitutable, the molecular and fitness effects of modular segments can be predicted by the biophysical compatibility of the replacement segment.

Development of antibacterial compounds that constrain evolutionary pathways to resistance.

Antibiotic resistance is a worldwide challenge. A potential approach to block resistance is to simultaneously inhibit WT and known escape variants of the target bacterial protein. Here, we applied an integrated computational and experimental approach to discover compounds that inhibit both WT and trimethoprim (TMP) resistant mutants of E. coli dihydrofolate reductase (DHFR). We identified a novel compound (CD15-3) that inhibits WT DHFR and its TMP resistant variants L28R, P21L and A26T with IC50 50-75 µM against WT and TMP-resistant strains. Resistance to CD15-3 was dramatically delayed compared to TMP in in vitro evolution. Whole genome sequencing of CD15-3-resistant strains showed no mutations in the target folA locus. Rather, gene duplication of several efflux pumps gave rise to weak (about twofold increase in IC50) resistance against CD15-3. Altogether, our results demonstrate the promise of strategy to develop evolution drugs - compounds which constrain evolutionary escape routes in pathogens.

Adaptation to mutational inactivation of an essential gene converges to an accessible suboptimal fitness peak.

The mechanisms of adaptation to inactivation of essential genes remain unknown. Here we inactivate E. coli dihydrofolate reductase (DHFR) by introducing D27G,N,F chromosomal mutations in a key catalytic residue with subsequent adaptation by an automated serial transfer protocol. The partial reversal G27- > C occurred in three evolutionary trajectories. Conversely, in one trajectory for D27G and in all trajectories for D27F,N strains adapted to grow at very low metabolic supplement (folAmix) concentrations but did not escape entirely from supplement auxotrophy. Major global shifts in metabolome and proteome occurred upon DHFR inactivation, which were partially reversed in adapted strains. Loss-of-function mutations in two genes, thyA and deoB, ensured adaptation to low folAmix by rerouting the 2-Deoxy-D-ribose-phosphate metabolism from glycolysis towards synthesis of dTMP. Multiple evolutionary pathways of adaptation converged to a suboptimal solution due to the high accessibility to loss-of-function mutations that block the path to the highest, yet least accessible, fitness peak.

Chimeric dihydrofolate reductases display properties of modularity and biophysical diversity.

While reverse genetics and functional genomics have long affirmed the role of individual mutations in determining protein function, there have been fewer studies addressing how large-scale changes in protein sequences, such as in entire modular segments, influence protein function and evolution. Given how recombination can reassort protein sequences, these types of changes may play an underappreciated role in how novel protein functions evolve in nature. Such studies could aid our understanding of whether certain organismal phenotypes related to protein function-such as growth in the presence or absence of an antibiotic-are robust with respect to the identity of certain modular segments. In this study, we combine molecular genetics with biochemical and biophysical methods to gain a better understanding of protein modularity in dihydrofolate reductase (DHFR), an enzyme target of antibiotics also widely used as a model for protein evolution. We replace an integral α-helical segment of Escherichia coli DHFR with segments from a number of different organisms (many nonmicrobial) and examine how these chimeric enzymes affect organismal phenotypes (e.g., resistance to an antibiotic) as well as biophysical properties of the enzyme (e.g., thermostability). We find that organismal phenotypes and enzyme properties are highly sensitive to the identity of DHFR modules, and that this chimeric approach can create enzymes with diverse biophysical characteristics.

Proteostasis Environment Shapes Higher-Order Epistasis Operating on Antibiotic Resistance.

Recent studies have affirmed that higher-order epistasis is ubiquitous and can have large effects on complex traits. Yet, we lack frameworks for understanding how epistatic interactions are influenced by central features of cell physiology. In this study, we assess how protein quality control machinery-a critical component of cell physiology-affects epistasis for different traits related to bacterial resistance to antibiotics. Specifically, we disentangle the interactions between different protein quality control genetic backgrounds and two sets of mutations: (i) SNPs associated with resistance to antibiotics in an essential bacterial enzyme (dihydrofolate reductase, or DHFR) and (ii) differing DHFR bacterial species-specific amino acid background sequences (Escherichia coli, Listeria grayi, and Chlamydia muridarum). In doing so, we improve on generic observations that epistasis is widespread by discussing how patterns of epistasis can be partly explained by specific interactions between mutations in an essential enzyme and genes associated with the proteostasis environment. These findings speak to the role of environmental and genotypic context in modulating higher-order epistasis, with direct implications for evolutionary theory, genetic modification technology, and efforts to manage antimicrobial resistance.

Evolution on the Biophysical Fitness Landscape of an RNA Virus.

Viral evolutionary pathways are determined by the fitness landscape, which maps viral genotype to fitness. However, a quantitative description of the landscape and the evolutionary forces on it remain elusive. Here, we apply a biophysical fitness model based on capsid folding stability and antibody binding affinity to predict the evolutionary pathway of norovirus escaping a neutralizing antibody. The model is validated by experimental evolution in bulk culture and in a drop-based microfluidics that propagates millions of independent small viral subpopulations. We demonstrate that along the axis of binding affinity, selection for escape variants and drift due to random mutations have the same direction, an atypical case in evolution. However, along folding stability, selection and drift are opposing forces whose balance is tuned by viral population size. Our results demonstrate that predictable epistatic tradeoffs between molecular traits of viral proteins shape viral evolution.

Differential Enzyme Flexibility Probed Using Solid-State Nanopores.

Enzymes and motor proteins are dynamic macromolecules that coexist in a number of conformations of similar energies. Protein function is usually accompanied by a change in structure and flexibility, often induced upon binding to ligands. However, while measuring protein flexibility changes between active and resting states is of therapeutic significance, it remains a challenge. Recently, our group has demonstrated that breadth of signal amplitudes in measured electrical signatures as an ensemble of individual protein molecules is driven through solid-state nanopores and correlates with protein conformational dynamics. Here, we extend our study to resolve subtle flexibility variation in dihydrofolate reductase mutants from unlabeled single molecules in solution. We first demonstrate using a canonical protein system, adenylate kinase, that both size and flexibility changes can be observed upon binding to a substrate that locks the protein in a closed conformation. Next, we investigate the influence of voltage bias and pore geometry on the measured electrical pulse statistics during protein transport. Finally, using the optimal experimental conditions, we systematically study a series of wild-type and mutant dihydrofolate reductase proteins, finding a good correlation between nanopore-measured protein conformational dynamics and equilibrium bulk fluorescence probe measurements. Our results unequivocally demonstrate that nanopore-based measurements reliably probe conformational diversity in native protein ensembles.

Stability of the Influenza Virus Hemagglutinin Protein Correlates with Evolutionary Dynamics.

Protein thermodynamics are an integral determinant of viral fitness and one of the major drivers of protein evolution. Mutations in the influenza A virus (IAV) hemagglutinin (HA) protein can eliminate neutralizing antibody binding to mediate escape from preexisting antiviral immunity. Prior research on the IAV nucleoprotein suggests that protein stability may constrain seasonal IAV evolution; however, the role of stability in shaping the evolutionary dynamics of the HA protein has not been explored. We used the full coding sequence of 9,797 H1N1pdm09 HA sequences and 16,716 human seasonal H3N2 HA sequences to computationally estimate relative changes in the thermal stability of the HA protein between 2009 and 2016. Phylogenetic methods were used to characterize how stability differences impacted the evolutionary dynamics of the virus. We found that pandemic H1N1 IAV strains split into two lineages that had different relative HA protein stabilities and that later variants were descended from the higher-stability lineage. Analysis of the mutations associated with the selective sweep of the higher-stability lineage found that they were characterized by the early appearance of highly stabilizing mutations, the earliest of which was not located in a known antigenic site. Experimental evidence further suggested that H1N1 HA stability may be correlated with in vitro virus production and infection. A similar analysis of H3N2 strains found that surviving lineages were also largely descended from viruses predicted to encode more-stable HA proteins. Our results suggest that HA protein stability likely plays a significant role in the persistence of different IAV lineages. IMPORTANCE One of the constraints on fast-evolving viruses, such as influenza virus, is protein stability, or how strongly the folded protein holds together. Despite the importance of this protein property, there has been limited investigation of the impact of the stability of the influenza virus hemagglutinin protein-the primary antibody target of the immune system-on its evolution. Using a combination of computational estimates of stability and experiments, our analysis found that viruses with more-stable hemagglutinin proteins were associated with long-term persistence in the population. There are two potential reasons for the observed persistence. One is that more-stable proteins tolerate destabilizing mutations that less-stable proteins could not, thus increasing opportunities for immune escape. The second is that greater stability increases the fitness of the virus through increased production of infectious particles. Further research on the relative importance of these mechanisms could help inform the annual influenza vaccine composition decision process.

Rational Design of Novel Allosteric Dihydrofolate Reductase Inhibitors Showing Antibacterial Effects on Drug-Resistant Escherichia coli Escape Variants.

In drug discovery, systematic variations of substituents on a common scaffold and bioisosteric replacements are often used to generate diversity and obtain molecules with better biological effects. However, this could saturate the small-molecule diversity pool resulting in drug resistance. On the other hand, conventional drug discovery relies on targeting known pockets on protein surfaces leading to drug resistance by mutations of critical pocket residues. Here, we present a two-pronged strategy of designing novel drugs that target unique pockets on a protein's surface to overcome the above problems. Dihydrofolate reductase, DHFR, is a critical enzyme involved in thymidine and purine nucleotide biosynthesis. Several classes of compounds that are structural analogues of the substrate dihydrofolate have been explored for their antifolate activity. Here, we describe 10 novel small-molecule inhibitors of Escherichia coli DHFR, EcDHFR, belonging to the stilbenoid, deoxybenzoin, and chalcone family of compounds discovered by a combination of pocket-based virtual ligand screening and systematic scaffold hopping. These inhibitors show a unique uncompetitive or noncompetitive inhibition mechanism, distinct from those reported for all known inhibitors of DHFR, indicative of binding to a unique pocket distinct from either substrate or cofactor-binding pockets. Furthermore, we demonstrate that rescue mutants of EcDHFR, with reduced affinity to all known classes of DHFR inhibitors, are inhibited at the same concentration as the wild-type. These compounds also exhibit antibacterial activity against E. coli harboring the drug-resistant variant of DHFR. This discovery is the first report on a novel class of inhibitors targeting a unique pocket on EcDHFR.

Biophysical principles predict fitness landscapes of drug resistance.

Fitness landscapes of drug resistance constitute powerful tools to elucidate mutational pathways of antibiotic escape. Here, we developed a predictive biophysics-based fitness landscape of trimethoprim (TMP) resistance for Escherichia coli dihydrofolate reductase (DHFR). We investigated the activity, binding, folding stability, and intracellular abundance for a complete set of combinatorial DHFR mutants made out of three key resistance mutations and extended this analysis to DHFR originated from Chlamydia muridarum and Listeria grayi We found that the acquisition of TMP resistance via decreased drug affinity is limited by a trade-off in catalytic efficiency. Protein stability is concurrently affected by the resistant mutants, which precludes a precise description of fitness from a single molecular trait. Application of the kinetic flux theory provided an accurate model to predict resistance phenotypes (IC50) quantitatively from a unique combination of the in vitro protein molecular properties. Further, we found that a controlled modulation of the GroEL/ES chaperonins and Lon protease levels affects the intracellular steady-state concentration of DHFR in a mutation-specific manner, whereas IC50 is changed proportionally, as indeed predicted by the model. This unveils a molecular rationale for the pleiotropic role of the protein quality control machinery on the evolution of antibiotic resistance, which, as we illustrate here, may drastically confound the evolutionary outcome. These results provide a comprehensive quantitative genotype-phenotype map for the essential enzyme that serves as an important target of antibiotic and anticancer therapies.

Free Superoxide is an Intermediate in the Production of H2O2 by Copper(I)-Aß Peptide and O2.

Oxidative stress is considered as an important factor and an early event in the etiology of Alzheimer's disease (AD). Cu bound to the peptide amyloid-ß (Aß) is found in AD brains, and Cu-Aß could contribute to this oxidative stress, as it is able to produce in vitro H2O2 and HO˙ in the presence of oxygen and biological reducing agents such as ascorbate. The mechanism of Cu-Aß-catalyzed H2O2 production is however not known, although it was proposed that H2O2 is directly formed from O2 via a 2-electron process. Here, we implement an electrochemical setup and use the specificity of superoxide dismutase-1 (SOD1) to show, for the first time, that H2O2 production by Cu-Aß in the presence of ascorbate occurs mainly via a free O2˙(-) intermediate. This finding radically changes the view on the catalytic mechanism of H2O2 production by Cu-Aß, and opens the possibility that Cu-Aß-catalyzed O2˙(-) contributes to oxidative stress in AD, and hence may be of interest.

Resistance to dual-gene Bt maize in Spodoptera frugiperda: selection, inheritance, and cross-resistance to other transgenic events.

Transgenic crop "pyramids" producing two or more Bacillus thuringiensis (Bt) toxins active against the same pest are used to delay evolution of resistance in insect pest populations. Laboratory and greenhouse experiments were performed with fall armyworm, Spodoptera frugiperda, to characterize resistance to Bt maize producing Cry1A.105 and Cry2Ab and test some assumptions of the "pyramid" resistance management strategy. Selection of a field-derived strain of S. frugiperda already resistant to Cry1F maize with Cry1A.105 + Cry2Ab maize for ten generations produced resistance that allowed the larvae to colonize and complete the life cycle on these Bt maize plants. Greenhouse experiments revealed that the resistance was completely recessive (Dx = 0), incomplete, autosomal, and without maternal effects or cross-resistance to the Vip3Aa20 toxin produced in other Bt maize events. This profile of resistance supports some of the assumptions of the pyramid strategy for resistance management. However, laboratory experiments with purified Bt toxin and plant leaf tissue showed that resistance to Cry1A.105 + Cry2Ab2 maize further increased resistance to Cry1Fa, which indicates that populations of fall armyworm have high potential for developing resistance to some currently available pyramided maize used against this pest, especially where resistance to Cry1Fa was reported in the field.

Ethylmalonic encephalopathy ETHE1 R163W/R163Q mutations alter protein stability and redox properties of the iron centre.

ETHE1 is an iron-containing protein from the metallo ß-lactamase family involved in the mitochondrial sulfide oxidation pathway. Mutations in ETHE1 causing loss of function result in sulfide toxicity and in the rare fatal disease Ethylmalonic Encephalopathy (EE). Frequently mutations resulting in depletion of ETHE1 in patient cells are due to severe structural and folding defects. However, some ETHE1 mutations yield nearly normal protein levels and in these cases disease mechanism was suspected to lie in compromised catalytic activity. To address this issue and to elicit how ETHE1 dysfunction results in EE, we have investigated two such pathological mutations, ETHE1-p.Arg163Gln and p.Arg163Trp. In addition, we report a number of benchmark properties of wild type human ETHE1, including for the first time the redox properties of the mononuclear iron centre. We show that loss of function in these variants results from a combination of decreased protein stability and activity. Although structural assessment revealed that the protein fold is not perturbed by mutations, both variants have decreased thermal stabilities and higher proteolytic susceptibilities. ETHE1 wild type and variants bind 1 ± 0.2 mol iron/protein and no zinc; however, the variants exhibited only ≈ 10% of wild-type catalytically activity. Analysis of the redox properties of ETHE1 mononuclear iron centre revealed that the variants have lowered reduction potentials with respect to that of the wild type. This illustrates how point mutation-induced loss of function may arise via very discrete subtle conformational effects on the protein fold and active site chemistry, without extensive disruption of the protein structure or protein-cofactor association.

Simplified 2,4-dinitrophenylhydrazine spectrophotometric assay for quantification of carbonyls in oxidized proteins.

This work proposes a modification of the 2,4-dinitrophenylhydrazine (DNPH) spectrophotometric assay commonly used to evaluate the concentration of carbonyl groups in oxidized proteins. In this approach NaOH is added to the protein solution after the addition of DNPH, shifting the maximum absorbance wavelength of the derivatized protein from 370 to 450nm. This reduces the interference of DNPH and allows the direct quantification in the sample solution without the need for the precipitation, washing, and resuspension steps that are carried out in the traditional DNPH method. The two methods were compared under various conditions and are statistically equivalent.

Mechanism of superoxide and hydrogen peroxide generation by human electron-transfer flavoprotein and pathological variants.

Reactive oxygen species production by mitochondrial enzymes plays a fundamental role both in cellular signaling and in the progression of dysfunctional states. However, sources of reactive oxygen species and the mechanisms by which enzymes produce these reactive species still remain elusive. We characterized the generation of reactive oxygen species by purified human electron-transfer flavoprotein (ETF), a mitochondrial enzyme that has a central role in the metabolism of lipids, amino acids, and choline. The results showed that ETF produces significant amounts of both superoxide and hydrogen peroxide in the presence of its partner enzyme medium-chain acyl-CoA dehydrogenase (MCAD). ETF-mediated production of reactive oxygen species is partially inhibited at high MCAD/ETF ratios, whereas it is enhanced at high ionic strength. Determination of the reduction potentials of ETF showed that thermodynamic properties of the FAD cofactor are changed upon formation of a complex between ETF and MCAD, supporting the notion that protein:protein interactions modulate the reactivity of the protein with dioxygen. Two pathogenic ETF variants were also studied to determine which factors modulate the reactivity toward molecular oxygen and promote reactive oxygen species production. The results obtained show that destabilized conformations and defective protein:protein interactions increase the ability of ETF to generate reactive oxygen species. A possible role for these processes in mitochondrial dysfunction in metabolic disorders of fatty acid ß-oxidation is discussed.

Mutations at the flavin binding site of ETF:QO yield a MADD-like severe phenotype in Drosophila.

Following a screening on EMS-induced Drosophila mutants defective for formation and morphogenesis of epithelial cells, we have identified three lethal mutants defective for the production of embryonic cuticle. The mutants are allelic to the CG12140 gene, the fly homologue of electron transfer flavoprotein:ubiquinone oxidoreductase (ETF:QO). In humans, inherited defects in this inner membrane protein account for multiple acyl-CoA dehydrogenase deficiency (MADD), a metabolic disease of ß-oxidation, with a broad range of clinical phenotypes, varying from embryonic lethal to mild forms. The three mutant alleles carried distinct missense mutations in ETF:QO (G65E, A68V and S104F) and maternal mutant embryos for ETF:QO showed lethal morphogenetic defects and a significant induction of apoptosis following germ-band elongation. This phenotype is accompanied by an embryonic accumulation of short- and medium-chain acylcarnitines (C4, C8 and C12) as well as long-chain acylcarnitines (C14 and C16:1), whose elevation is also found in severe MADD forms in humans under intense metabolic decompensation. In agreement the ETF:QO activity in the mutant embryos is markedly decreased in relation to wild type activity. Amino acid sequence analysis and structural mapping into a molecular model of ETF:QO show that all mutations map at FAD interacting residues, two of which at the nucleotide-binding Rossmann fold. This structural domain is composed by a ß-strand connected by a short loop to an α-helix, and its perturbation results in impaired cofactor association via structural destabilisation and consequently enzymatic inactivation. This work thus pinpoints the molecular origins of a severe MADD-like phenotype in the fruit fly and establishes the proof of concept concerning the suitability of this organism as a potential model organism for MADD.

Cofactors and metabolites as protein folding helpers in metabolic diseases.

In the past few decades, improved early diagnosis methods, technological developments and an increasing crosstalk between clinicians and researchers has led to the identification of an increasing number of inborn metabolic diseases. In these disorders, missense mutations are the most frequent type of genetic defects, frequently resulting in defective protein folding. A better understanding at the molecular level of protein misfolding and its role in disease has prompted the emergence of therapies based in the use of small molecules that have the ability to correct protein folding defects. Well-known cases are reported for phenylketonuria and Gaucher's disease. Most of these compounds have a specific mechanism of action interacting directly with a particular protein, the so called pharmacological chaperones. Among such small molecules are protein ligands, either natural substrates or synthetic derivatives, cofactors, competitive inhibitors, and agonist/antagonists. In this review we will start by briefly overviewing the mechanisms through which such ligands exert a stabilizing action, and then move on to an extended discussion on therapeutic approaches and use of vitamins and substrates to correct protein misfolding in metabolic disorders. Examples of vitamins that have been successfully prescribed to rescue some cases of inborn errors of metabolism will be presented. In particular, the role of riboflavin supplementation in the treatment of fatty acid ß-oxidation disorders will be thoroughly analyzed, focusing on recent reports that shed light on the molecular basis of vitamin responsiveness. Moreover, we will highlight the latest studies that point to a synergistic effect of cofactors and metabolites in the rescue of defective fatty acid ß-oxidation enzymes. The synergism of multiple small molecules may underlie a promising general pharmacological strategy for the treatment of metabolic diseases in general.