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.

A central role for regulated protein stability in the control of TFE3 and MITF by nutrients.

The TFE3 and MITF master transcription factors maintain metabolic homeostasis by regulating lysosomal, melanocytic, and autophagy genes. Previous studies posited that their cytosolic retention by 14-3-3, mediated by the Rag GTPases-mTORC1, was key for suppressing transcriptional activity in the presence of nutrients. Here, we demonstrate using mammalian cells that regulated protein stability plays a fundamental role in their control. Amino acids promote the recruitment of TFE3 and MITF to the lysosomal surface via the Rag GTPases, activating an evolutionarily conserved phospho-degron and leading to ubiquitination by CUL1ß-TrCP and degradation. Elucidation of the minimal functional degron revealed a conserved alpha-helix required for interaction with RagA, illuminating the molecular basis for a severe neurodevelopmental syndrome caused by missense mutations in TFE3 within the RagA-TFE3 interface. Additionally, the phospho-degron is recurrently lost in TFE3 genomic translocations that cause kidney cancer. Therefore, two divergent pathologies converge on the loss of protein stability regulation by nutrients.

CasPlay provides a gRNA-barcoded CRISPR-based display platform for antibody repertoire profiling.

Protein display technologies link proteins to distinct nucleic acid sequences (barcodes), enabling multiplexed protein assays via DNA sequencing. Here, we develop Cas9 display (CasPlay) to interrogate customized peptide libraries fused to catalytically inactive Cas9 (dCas9) by sequencing the guide RNA (gRNA) barcodes associated with each peptide. We first confirm the ability of CasPlay to characterize antibody epitopes by recovering a known binding motif for a monoclonal anti-FLAG antibody. We then use a CasPlay library tiling the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) proteome to evaluate vaccine-induced antibody reactivities. Using a peptide library representing the human virome, we demonstrate the ability of CasPlay to identify epitopes across many viruses from microliters of patient serum. Our results suggest that CasPlay is a viable strategy for customized protein interaction studies from highly complex libraries and could provide an alternative to phage display technologies.

CRISPR-based peptide library display and programmable microarray self-assembly for rapid quantitative protein binding assays.

CRISPR-inspired systems have been extensively developed for applications in genome editing and nucleic acid detection. Here, we introduce a CRISPR-based peptide display technology to facilitate customized, high-throughput in vitro protein interaction studies. We show that bespoke peptide libraries fused to catalytically inactive Cas9 (dCas9) and barcoded with unique single guide RNA (sgRNA) molecules self-assemble from a single mixed pool to programmable positions on a DNA microarray surface for rapid, multiplexed binding assays. We develop dCas9-displayed saturation mutagenesis libraries to characterize antibody-epitope binding for a commercial anti-FLAG monoclonal antibody and human serum antibodies. We also show that our platform can be used for viral epitope mapping and exhibits promise as a multiplexed diagnostics tool. Our CRISPR-based peptide display platform and the principles of complex library self-assembly using dCas9 could be adapted for rapid interrogation of varied customized protein libraries or biological materials assembly using DNA scaffolding.

Metabolic stress promotes stop-codon readthrough and phenotypic heterogeneity.

Accurate protein synthesis is a tightly controlled biological process with multiple quality control steps safeguarded by aminoacyl-transfer RNA (tRNA) synthetases and the ribosome. Reduced translational accuracy leads to various physiological changes in both prokaryotes and eukaryotes. Termination of translation is signaled by stop codons and catalyzed by release factors. Occasionally, stop codons can be suppressed by near-cognate aminoacyl-tRNAs, resulting in protein variants with extended C termini. We have recently shown that stop-codon readthrough is heterogeneous among single bacterial cells. However, little is known about how environmental factors affect the level and heterogeneity of stop-codon readthrough. In this study, we have combined dual-fluorescence reporters, mass spectrometry, mathematical modeling, and single-cell approaches to demonstrate that a metabolic stress caused by excess carbon substantially increases both the level and heterogeneity of stop-codon readthrough. Excess carbon leads to accumulation of acid metabolites, which lower the pH and the activity of release factors to promote readthrough. Furthermore, our time-lapse microscopy experiments show that single cells with high readthrough levels are more adapted to severe acid stress conditions and are more sensitive to an aminoglycoside antibiotic. Our work thus reveals a metabolic stress that promotes translational heterogeneity and phenotypic diversity.

Distinct Hepatic PKA and CDK Signaling Pathways Control Activity-Independent Pyruvate Kinase Phosphorylation and Hepatic Glucose Production.

Pyruvate kinase is an important enzyme in glycolysis and a key metabolic control point. We recently observed a pyruvate kinase liver isoform (PKL) phosphorylation site at S113 that correlates with insulin resistance in rats on a 3 day high-fat diet (HFD) and suggests additional control points for PKL activity. However, in contrast to the classical model of PKL regulation, neither authentically phosphorylated PKL at S12 nor S113 alone is sufficient to alter enzyme kinetics or structure. Instead, we show that cyclin-dependent kinases (CDKs) are activated by the HFD and responsible for PKL phosphorylation at position S113 in addition to other targets. These CDKs control PKL nuclear retention, alter cytosolic PKL activity, and ultimately influence glucose production. These results change our view of PKL regulation and highlight a previously unrecognized pathway of hepatic CDK activity and metabolic control points that may be important in insulin resistance and type 2 diabetes.

Sequencer Hacking Unlocks Quantitative Protein Studies.

In this issue of Molecular Cell, Layton et al. (2019) repurpose a common next-generation DNA sequencer to enable high-throughput protein biochemical studies, identifying improved sequence variants for stronger protein-protein interactions and dissecting the contributions of specific amino acids to enzymatic activity.

Expression of TorsinA in a heterologous yeast system reveals interactions with lumenal domains of LINC and nuclear pore complex components.

DYT1 dystonia is caused by an in-frame deletion of a glutamic acid codon in the gene encoding the AAA+ ATPase TorsinA (TorA). TorA localizes within the lumen of the nuclear envelope/endoplasmic reticulum and binds to a membrane-spanning cofactor, lamina associated polypeptide 1 (LAP1) or lumenal domain like LAP1 (LULL1), to form an ATPase; the substrate(s) of TorA remains ill-defined. Here we use budding yeast, which lack Torsins, to interrogate TorA function. We show that TorA accumulates at nuclear envelope-embedded spindle pole bodies (SPBs) in a way that requires its oligomerization and the SUN (Sad1 and UNc-84)-domain protein, Mps3. We further show that TorA physically interacts with human SUN1/2 within this system, supporting the physiological relevance of these interactions. Consistent with the idea that TorA acts on a SPB substrate, its binding to SPBs is modulated by the ATPase-stimulating activity of LAP1. TorA and TorA-ΔE reduce the fitness of cells expressing mps3 alleles, whereas TorA alone inhibits growth of cells lacking Pom152, a component of the nuclear pore complex. This genetic specificity is mirrored biochemically as TorA, but not TorA-ΔE, binds Pom152. Thus, TorA-nucleoporin interactions might be abrogated by TorA-ΔE, suggesting new experimental avenues to interrogate the molecular basis behind nuclear envelope herniations seen in mammalian cells lacking TorA function.

Organisms with alternative genetic codes resolve unassigned codons via mistranslation and ribosomal rescue.

Organisms possessing genetic codes with unassigned codons raise the question of how cellular machinery resolves such codons and how this could impact horizontal gene transfer. Here, we use a genomically recoded Escherichia coli to examine how organisms address translation at unassigned UAG codons, which obstruct propagation of UAG-containing viruses and plasmids. Using mass spectrometry, we show that recoded organisms resolve translation at unassigned UAG codons via near-cognate suppression, dramatic frameshifting from at least -3 to +19 nucleotides, and rescue by ssrA-encoded tmRNA, ArfA, and ArfB. We then demonstrate that deleting tmRNA restores expression of UAG-ending proteins and propagation of UAG-containing viruses and plasmids in the recoded strain, indicating that tmRNA rescue and nascent peptide degradation is the cause of impaired virus and plasmid propagation. The ubiquity of tmRNA homologs suggests that genomic recoding is a promising path for impairing horizontal gene transfer and conferring genetic isolation in diverse organisms.

PKCε contributes to lipid-induced insulin resistance through cross talk with p70S6K and through previously unknown regulators of insulin signaling.

Insulin resistance drives the development of type 2 diabetes (T2D). In liver, diacylglycerol (DAG) is a key mediator of lipid-induced insulin resistance. DAG activates protein kinase C ε (PKCε), which phosphorylates and inhibits the insulin receptor. In rats, a 3-day high-fat diet produces hepatic insulin resistance through this mechanism, and knockdown of hepatic PKCε protects against high-fat diet-induced hepatic insulin resistance. Here, we employed a systems-level approach to uncover additional signaling pathways involved in high-fat diet-induced hepatic insulin resistance. We used quantitative phosphoproteomics to map global in vivo changes in hepatic protein phosphorylation in chow-fed, high-fat-fed, and high-fat-fed with PKCε knockdown rats to distinguish the impact of lipid- and PKCε-induced protein phosphorylation. This was followed by a functional siRNA-based screen to determine which dynamically regulated phosphoproteins may be involved in canonical insulin signaling. Direct PKCε substrates were identified by motif analysis of phosphoproteomics data and validated using a large-scale in vitro kinase assay. These substrates included the p70S6K substrates RPS6 and IRS1, which suggested cross talk between PKCε and p70S6K in high-fat diet-induced hepatic insulin resistance. These results identify an expanded set of proteins through which PKCε may drive high-fat diet-induced hepatic insulin resistance that may direct new therapeutic approaches for T2D.

Kinase Substrate Profiling Using a Proteome-wide Serine-Oriented Human Peptide Library.

The human proteome encodes >500 protein kinases and hundreds of thousands of potential phosphorylation sites. However, the identification of kinase-substrate pairs remains an active area of research because the relationships between individual kinases and these phosphorylation sites remain largely unknown. Many techniques have been established to discover kinase substrates but are often technically challenging to perform. Moreover, these methods frequently rely on substrate reagent pools that do not reflect human protein sequences or are biased by human cell line protein expression profiles. Here, we describe a new approach called SERIOHL-KILR (serine-oriented human library-kinase library reactions) to profile kinase substrate specificity and to identify candidate substrates for serine kinases. Using a purified library of >100000 serine-oriented human peptides expressed heterologously in Escherichia coli, we perform in vitro kinase reactions to identify phosphorylated human peptide sequences by liquid chromatography and tandem mass spectrometry. We compare our results for protein kinase A to those of a well-established positional scanning peptide library method, certifying that SERIOHL-KILR can identify the same predominant motif elements as traditional techniques. We then interrogate a small panel of cancer-associated PKCß mutants using our profiling protocol and observe a shift in substrate specificity likely attributable to the loss of key polar contacts between the kinase and its substrates. Overall, we demonstrate that SERIOHL-KILR can rapidly identify candidate kinase substrates that can be directly mapped to human sequences for pathway analysis. Because this technique can be adapted for various kinase studies, we believe that SERIOHL-KILR will have many new victims in the future.

Encoding human serine phosphopeptides in bacteria for proteome-wide identification of phosphorylation-dependent interactions.

Post-translational phosphorylation is essential to human cellular processes, but the transient, heterogeneous nature of this modification complicates its study in native systems. We developed an approach to interrogate phosphorylation and its role in protein-protein interactions on a proteome-wide scale. We genetically encoded phosphoserine in recoded E. coli and generated a peptide-based heterologous representation of the human serine phosphoproteome. We designed a single-plasmid library encoding >100,000 human phosphopeptides and confirmed the site-specific incorporation of phosphoserine in >36,000 of these peptides. We then integrated our phosphopeptide library into an approach known as Hi-P to enable proteome-level screens for serine-phosphorylation-dependent human protein interactions. Using Hi-P, we found hundreds of known and potentially new phosphoserine-dependent interactors with 14-3-3 proteins and WW domains. These phosphosites retained important binding characteristics of the native human phosphoproteome, as determined by motif analysis and pull-downs using full-length phosphoproteins. This technology can be used to interrogate user-defined phosphoproteomes in any organism, tissue, or disease of interest.

Comparative Proteomics Enables Identification of Nonannotated Cold Shock Proteins in E. coli.

Recent advances in mass spectrometry-based proteomics have revealed translation of previously nonannotated microproteins from thousands of small open reading frames (smORFs) in prokaryotic and eukaryotic genomes. Facile methods to determine cellular functions of these newly discovered microproteins are now needed. Here, we couple semiquantitative comparative proteomics with whole-genome database searching to identify two nonannotated, homologous cold shock-regulated microproteins in Escherichia coli K12 substr. MG1655, as well as two additional constitutively expressed microproteins. We apply molecular genetic approaches to confirm expression of these cold shock proteins (YmcF and YnfQ) at reduced temperatures and identify the noncanonical ATT start codons that initiate their translation. These proteins are conserved in related Gram-negative bacteria and are predicted to be structured, which, in combination with their cold shock upregulation, suggests that they are likely to have biological roles in the cell. These results reveal that previously unknown factors are involved in the response of E. coli to lowered temperatures and suggest that further nonannotated, stress-regulated E. coli microproteins may remain to be found. More broadly, comparative proteomics may enable discovery of regulated, and therefore potentially functional, products of smORF translation across many different organisms and conditions.

Heterogeneity of Stop Codon Readthrough in Single Bacterial Cells and Implications for Population Fitness.

Gene expression noise (heterogeneity) leads to phenotypic diversity among isogenic individual cells. Our current understanding of gene expression noise is mostly limited to transcription, as separating translational noise from transcriptional noise has been challenging. It also remains unclear how translational heterogeneity originates. Using a transcription-normalized reporter system, we discovered that stop codon readthrough is heterogeneous among single cells, and individual cells with higher UGA readthrough grow faster from stationary phase. Our work also revealed that individual cells with lower protein synthesis levels exhibited higher UGA readthrough, which was confirmed with ribosome-targeting antibiotics (e.g., chloramphenicol). Further experiments and mathematical modeling suggest that varied competition between ternary complexes and release factors perturbs the UGA readthrough level. Our results indicate that fluctuations in the concentrations of translational components lead to UGA readthrough heterogeneity among single cells, which enhances phenotypic diversity of the genetically identical population and facilitates its adaptation to changing environments.

Expression of Recombinant Phosphoproteins for Signal Transduction Studies.

Complex signaling cascades are difficult to study in vitro without phosphorylated proteins. Here, we describe a technique for the routine production of recombinant phosphoproteins by directly incorporating phosphoserine as a nonstandard amino acid. This protocol utilizes an optimized phosphoserine orthogonal translation system and an engineered strain of E. coli containing no genomic amber codons. This approach has been used to generate a variety of phosphorylated proteins to understand the role of protein phosphorylation in cell signaling.

Genome mining unearths a hybrid nonribosomal peptide synthetase-like-pteridine synthase biosynthetic gene cluster.

Nonribosomal peptides represent a large class of metabolites with pharmaceutical relevance. Pteridines, such as pterins, folates, and flavins, are heterocyclic metabolites that often serve as redox-active cofactors. The biosynthetic machineries for construction of these distinct classes of small molecules operate independently in the cell. Here, we discovered an unprecedented nonribosomal peptide synthetase-like-pteridine synthase hybrid biosynthetic gene cluster in Photorhabdus luminescens using genome synteny analysis. P. luminescens is a Gammaproteobacterium that undergoes phenotypic variation and can have both pathogenic and mutualistic roles. Through extensive gene deletion, pathway-targeted molecular networking, quantitative proteomic analysis, and NMR, we show that the genetic locus affects the regulation of quorum sensing and secondary metabolic enzymes and encodes new pteridine metabolites functionalized with cis-amide acyl-side chains, termed pepteridine A (1) and B (2). The pepteridines are produced in the pathogenic phenotypic variant and represent the first reported metabolites to be synthesized by a hybrid NRPS-pteridine pathway. These studies expand our view of the combinatorial biosynthetic potential available in bacteria.

SPAK and OSR1 play essential roles in potassium homeostasis through actions on the distal convoluted tubule.

KEY POINTS: STE20 (Sterile 20)/SPS-1 related proline/alanine-rich kinase (SPAK) and oxidative stress-response kinase-1 (OSR1) phosphorylate and activate the renal Na(+) -K(+) -2Cl(-) cotransporter 2 (NKCC2) and Na(+) Cl(-) cotransporter (NCC). Mouse models suggest that OSR1 mainly activates NKCC2-mediated sodium transport along the thick ascending limb, while SPAK mainly activates NCC along the distal convoluted tubule, but the kinases may compensate for each other. We hypothesized that disruption of both kinases would lead to polyuria and severe salt-wasting, and generated SPAK/OSR1 double knockout mice to test this. Despite a lack of SPAK and OSR1, phosphorylated NKCC2 abundance was still high, suggesting the existence of an alternative activating kinase. Compensatory changes in SPAK/OSR1-independent phosphorylation sites on both NKCC2 and NCC and changes in sodium transport along the collecting duct were also observed. Potassium restriction revealed that SPAK and OSR1 play essential roles in the emerging model that NCC activation is central to sensing changes in plasma [K(+) ]. ABSTRACT: STE20 (Sterile 20)/SPS-1 related proline/alanine-rich kinase (SPAK) and oxidative stress-response kinase-1 (OSR1) activate the renal cation cotransporters Na(+) -K(+) -2Cl(-) cotransporter (NKCC2) and Na(+) -Cl(-) cotransporter (NCC) via phosphorylation. Knockout mouse models suggest that OSR1 mainly activates NKCC2, while SPAK mainly activates NCC, with possible cross-compensation. We tested the hypothesis that disrupting both kinases causes severe polyuria and salt-wasting by generating SPAK/OSR1 double knockout (DKO) mice. DKO mice displayed lower systolic blood pressure compared with SPAK knockout (SPAK-KO) mice, but displayed no severe phenotype even after dietary salt restriction. Phosphorylation of NKCC2 at SPAK/OSR1-dependent sites was lower than in SPAK-KO mice, but still significantly greater than in wild type mice. In the renal medulla, there was significant phosphorylation of NKCC2 at SPAK/OSR1-dependent sites despite a complete absence of SPAK and OSR1, suggesting the existence of an alternative activating kinase. The distal convoluted tubule has been proposed to sense plasma [K(+) ], with NCC activation serving as the primary effector pathway that modulates K(+) secretion, by metering sodium delivery to the collecting duct. Abundance of phosphorylated NCC (pNCC) is dramatically lower in SPAK-KO mice than in wild type mice, and the additional disruption of OSR1 further reduced pNCC. SPAK-KO and kidney-specific OSR1 single knockout mice maintained plasma [K(+) ] following dietary potassium restriction, but DKO mice developed severe hypokalaemia. Unlike mice lacking SPAK or OSR1 alone, DKO mice displayed an inability to phosphorylate NCC under these conditions. These data suggest that SPAK and OSR1 are essential components of the effector pathway that maintains plasma [K(+) ].

A flexible codon in genomically recoded Escherichia coli permits programmable protein phosphorylation.

Biochemical investigation of protein phosphorylation events is limited by inefficient production of the phosphorylated and non-phosphorylated forms of full-length proteins. Here using a genomically recoded strain of E. coli with a flexible UAG codon we produce site-specific serine- or phosphoserine-containing proteins, with purities approaching 90%, from a single recombinant DNA. Specifically, we synthesize human MEK1 kinase with two serines or two phosphoserines, from one DNA template, and demonstrate programmable kinase activity. Programmable protein phosphorylation is poised to help reveal the structural and functional information encoded in the phosphoproteome.

Robust production of recombinant phosphoproteins using cell-free protein synthesis.

Understanding the functional and structural consequences of site-specific protein phosphorylation has remained limited by our inability to produce phosphoproteins at high yields. Here we address this limitation by developing a cell-free protein synthesis (CFPS) platform that employs crude extracts from a genomically recoded strain of Escherichia coli for site-specific, co-translational incorporation of phosphoserine into proteins. We apply this system to the robust production of up to milligram quantities of human MEK1 kinase. Then, we recapitulate a physiological signalling cascade in vitro to evaluate the contributions of site-specific phosphorylation of mono- and doubly phosphorylated forms on MEK1 activity. We discover that only one phosphorylation event is necessary and sufficient for MEK1 activity. Our work sets the stage for using CFPS as a rapid high-throughput technology platform for direct expression of programmable phosphoproteins containing multiple phosphorylated residues. This work will facilitate study of phosphorylation-dependent structure-function relationships, kinase signalling networks and kinase inhibitor drugs.