Deep protein representations enable recombinant protein expression prediction

A crucial process in the production of industrial enzymes is recombinant gene expression, which aims to induce enzyme overexpression of the genes in a host microbe. Current approaches for securing overexpression rely on molecular tools such as adjusting the recombinant expression vector, adjusting cultivation conditions, or performing codon optimizations. However, such strategies are time-consuming, and an alternative strategy would be to select genes for better compatibility with the recombinant host. Several methods for predicting soluble expression are available; however, they are all optimized for the expression host Escherichia coli and do not consider the possibility of an expressed protein not being soluble. We show that these tools are not suited for predicting expression potential in the industrially important host Bacillus subtilis. Instead, we build a B. subtilis-specific machine learning model for expressibility prediction. Given millions of unlabelled proteins and a small labeled dataset, we can successfully train such a predictive model. The unlabeled proteins provide a performance boost relative to using amino acid frequencies of the labeled proteins as input. On average, we obtain a modest performance of 0.64 area-under-the-curve (AUC) and 0.2 Matthews correlation coefficient (MCC). However, we find that this is sufficient for the prioritization of expression candidates for high-throughput studies. Moreover, the predicted class probabilities are correlated with expression levels. A number of features related to protein expression, including base frequencies and solubility, are captured by the model.     

Role of cooperativity in protein folding and protein mosaic assemblage relevance for protein conformational diseases

Biological systems are organized in intricate and highly structured networks with hierarchies and multiple scales. Cells can be considered as "meso-scale level" systems placed between the "macro-scale level" (systems of cellular networks) and the "micro-scale level" (systems of molecular networks). In fact, cells represent complex biochemical machineries made by networks of molecules connected by biochemical reactions. Thus, the brain should be studied as a system of "networks of networks". Recently, the existence of a Global Molecular Network (GMN) enmeshing the entire CNS was proposed. This proposal is based on the evidence that the extra-cellular matrix is a dynamic molecular structure capable of storing and releasing signals and of interacting with receptors and proteins on the cell membranes. Proteins have a special role in molecular networks since they can be assembled into high-order molecular complexes, which have been defined as Protein Mosaics (PM). Protein monomers in a PM (the "tesserae" of the mosaic) can interact via classical and non-classical cooperativity behaviour involving allosteric interactions. In the present paper, new features of allostery and cooperativity for protein folding, assemblage and topological features of PM will be discussed. Against this background, alterations in PM via allosteric modulations and non-classical cooperativity mechanisms may lead to protein aggregates like beta amyloid fibrils. Such aggregates cause pathological changes in the GMN structure and function leading to neurodegenerative diseases such as Alzheimer's disease. Thus, a novel view of the so called Protein Conformational Diseases (PCD) is proposed.     

Analysis of leukocyte membrane protein interactions using protein microarrays

Background  

Protein microarrays represent an emerging class of proteomic tools to investigate multiple protein-protein interactions in parallel. A sufficient proportion of immobilized proteins must maintain an active conformation and an orientation that allows for the sensitive and specific detection of antibody and ligand binding. In order to establish protein array technology for the characterization of the weak interactions between leukocyte membrane proteins, we selected the human leukocyte membrane protein CD200 (OX2) and its cell surface receptor (hCD200R) as a model system. As antibody-antigen reactions are generally of higher affinity than receptor-ligand binding, we first analyzed the reactivity of monoclonal antibodies (mAb) to normal and mutant forms of immobilized CD200R.     

Repeated two-hybrid screening detects transient protein–protein interactions

Yeast two-hybrid (Y2H) screening is a powerful method to detect protein–protein interactions (PPI) at the genomic-scale. A recently proposed framework for binary interactome mapping recommends the repeated screening approach to improve the quality of PPI data. Such repeated screening reveals Y2H interactions ranging from highly sampled to singleton interactions. The quality and the biological significance of interactions from distinguished sampling classes remain unknown. In order to systematically characterize such interactions, we have chosen a dataset of 1,262 interactions that were screened repeatedly four-times. The interactions were classified as highly sampled, weakly sampled, and singleton interactions. We assessed the quality of interactions in different sampling classes using features such as protein structural properties, conservation in yeast and presence of known domain–domain interactions that are previously associated with false positive rates. Our analysis reveals that the quality of singleton interactions is comparable to that of highly sampled interactions. Interestingly, singletons encompass a higher fraction of known domain–domain interactions than highly sampled ones. Furthermore, we observed that the singleton interactions are transient in nature, while the highly sampled interactions are predominantly part of stable complexes. Hence, the repeated Y2H screening method is ideal for detecting transient PPIs that are crucial in cellular signaling pathways.     

Semantically predicting protein functions based on protein functional connectivity

BackgroundThe current availability of public protein–protein interaction (PPI) databases which are usually modelled as PPI networks has led to the rapid development of protein function prediction approaches. The existing network-based prediction approaches mainly focus on the topological similarities between immediately interacting proteins, neglecting the protein functional connectivity which is the functional tightness between proteins. In this paper, we attempt to predict the functions of unannotated proteins based on PPI networks by incorporating the protein functional connectivity, as well as the similarity of protein functions, into the prediction procedure.ResultsAn approach named Semantic protein function Prediction based on protein Functional Connectivity (SPFC) is proposed to achieve a higher accuracy in predicting functions of unannotated protein. We define the functional connectivity and function addition for each protein, and incorporate them into the prediction. We evaluated the SPFC on real PPI datasets and the experiment results show that the SPFC method is more effective in function prediction than other network-based approaches.ConclusionIncorporating the functional connectivity of each protein into the function prediction can significantly improve the accuracy of protein prediction.     

Designing sequence to control protein function in an EF-hand protein

The extent of conformational change that calcium binding induces in EF-hand proteins is a key biochemical property specifying Ca(2+) sensor versus signal modulator function. To understand how differences in amino acid sequence lead to differences in the response to Ca(2+) binding, comparative analyses of sequence and structures, combined with model building, were used to develop hypotheses about which amino acid residues control Ca(2+)-induced conformational changes. These results were used to generate a first design of calbindomodulin (CBM-1), a calbindin D(9k) re-engineered with 15 mutations to respond to Ca(2+) binding with a conformational change similar to that of calmodulin. The gene for CBM-1 was synthesized, and the protein was expressed and purified. Remarkably, this protein did not exhibit any non-native-like molten globule properties despite the large number of mutations and the nonconservative nature of some of them. Ca(2+)-induced changes in CD intensity and in the binding of the hydrophobic probe, ANS, implied that CBM-1 does undergo Ca(2+) sensorlike conformational changes. The X-ray crystal structure of Ca(2+)-CBM-1 determined at 1.44 A resolution reveals the anticipated increase in hydrophobic surface area relative to the wild-type protein. A nascent calmodulin-like hydrophobic docking surface was also found, though it is occluded by the inter-EF-hand loop. The results from this first calbindomodulin design are discussed in terms of progress toward understanding the relationships between amino acid sequence, protein structure, and protein function for EF-hand CaBPs, as well as the additional mutations for the next CBM design.     

Frustration and hydrophobicity interplay in protein folding and protein evolution

A lattice model is used to study mutations and compacting effects on protein folding rates and folding temperature. In the context of protein evolution, we address the question regarding the best scenario for a polypeptide chain to fold: either a fast nonspecific collapse followed by a slow rearrangement to form the native structure or a specific collapse from the unfolded state with the simultaneous formation of the native state. This question is investigated for optimized sequences, whose native state has no frustrated contacts between monomers, and also for mutated sequences, whose native state has some degree of frustration. It is found that the best scenario for folding may depend on the amount of frustration of the native structure. The implication of this result on protein evolution is discussed.     

Sortase-mediated protein ligation: a new method for protein engineering

Sortase (SrtA), a transpeptidase from Staphylococcus aureus, catalyzes a cell-wall sorting reaction at an LPXTG motif by cleaving between threonine and glycine and subsequently joining the carboxyl group of threonine to an amino group of pentaglycine on the cell wall peptidoglycan. We have applied this transpeptidyl activity of sortase to in vitro protein ligation. We found that in the presence of sortase, protein/peptide with an LPXTG motif can be specifically ligated to an aminoglycine protein/peptide via an amide bond. Additionally, sortase can even conjugate substrates such as (d)-peptides, synthetic branched peptides, and aminoglycine-derivatized small molecules to the C terminus of a recombinant protein. The sortase-mediate protein ligation is robust, specific, and easy to perform, and can be widely applied to specific protein conjugation with polypeptides or molecules of unique biochemical and biophysical properties.     

Determinants for intrinsically disordered protein recruitment into phase-separated protein condensates

Multivalent interactions between amino acid residues of intrinsically disordered proteins (IDPs) drive phase separation of these proteins into liquid condensates, forming various membrane-less organelles in cells. These interactions between often biased residues of IDPs are also likely involved in selective recruitment of many other IDPs into condensates. However, determining factors for this IDP recruitment into protein condensates are not understood yet. Here, we quantitatively examined recruitment tendencies of various IDPs with different sequence compositions into IDP-clustered condensates both in vitro as well as in cells. Condensate-forming IDP scaffolds, recruited IDP clients, and phase separation conditions were carefully varied to find key factors for selective IDP partitioning in protein condensates. Regardless of scaffold sequences, charged residues in client IDPs assured potent IDP recruitment, likely via strong electrostatic interactions, where positive residues could further enhance recruitment, possibly with cation–pi interactions. Notably, poly-ethylene glycol, a widely used crowding reagent for in vitro phase separation, abnormally increased IDP recruitment, indicating the need for careful use of crowding conditions. Tyrosines of IDP clients also strongly participated in recruitment both in vitro and in cells. Lastly, we measured recruitment degrees by more conventional interactions between folded proteins instead of disordered proteins. Surprisingly, recruitment forces by an even moderate protein interaction (K_d ∼ 5 μM) were substantially stronger than those by natural IDP–IDP interactions. The present data offer valuable information on how cells might organize protein partitioning on various protein condensates.

Diverse interactions between folded and disordered proteins collectively dictate selective protein recruitment into bimolecular condensates.     

Membrane protein microarrays

This paper describes the fabrication of microarrays consisting of G protein-coupled receptors (GPCRs) on surfaces coated with gamma-aminopropylsilane (GAPS). Microspots of model membranes on GAPS-coated surfaces were observed to have several desired properties-high mechanical stability, long range lateral fluidity, and a thickness corresponding to a lipid bilayer in the bulk of the microspot. GPCR arrays were obtained by printing membrane preparations containing GPCRs using a quill-pin printer. To demonstrate specific binding of ligands, arrays presenting neurotensin (NTR1), adrenergic (beta1), and dopamine (D1) receptors were treated with fluorescently labeled neurotensin (BT-NT). Fluorescence images revealed binding only to microspots corresponding to the neurotensin receptor; this specificity was further demonstrated by the inhibition of binding in the presence of excess unlabeled neurotensin. The ability of GPCR arrays to enable selectivity studies between the different subtypes of a receptor was examined by printing arrays consisting of three subtypes of the adrenergic receptor: beta1, beta2, and alpha2A. When treated with fluorescently labeled CGP 12177, a cognate antagonist analogue specific to beta-adrenergic receptors, binding was only observed to microspots of the beta1 and beta2 receptors. Furthermore, binding of labeled CGP 12177 was inhibited when the arrays were incubated with solutions also containing ICI 118551, and in a manner consistent with the higher affinity of ICI 118551 for the beta2 receptor relative to that for the beta1 receptor. The ability to estimate binding affinities of compounds using GPCR arrays was examined using a competitive binding assay with BT-NT and unlabeled neurotensin on NTR1 arrays. The estimated IC(50) value (2 nM) for neurotensin is in agreement with the literature; this agreement suggests that the receptor -G protein complex is preserved in the microspot. This first ever demonstration of direct pin-printing of membrane proteins and ligand-binding assays thereof fills a significant void in protein microchip technology--the lack of practical microarray-based methods for membrane proteins.     

Temperature-responsive protein pores

We describe temperature-responsive protein pores containing single elastin-like polypeptide (ELP) loops. The ELP loops were placed within the cavity of the lumen of the alpha-hemolysin (alphaHL) pore, a heptamer of known crystal structure. The cavity is roughly spherical with a molecular surface volume of about 39,500 A3. In an applied potential, the wild-type alphaHL pore remained open for long periods. In contrast, the ELP loop-containing alphaHL pores exhibited transient current blockades, the nature of which depended on the length and sequence of the inserted loop. Together with similar results obtained with poly(ethylene glycols) covalently attached within the cavity, the data suggest that the transient current blockades are caused by excursions of ELP into the transmembrane beta-barrel domain of the pore. Below its transition temperature, the ELP loop is fully expanded and blocks the pore completely, but reversibly. Above its transition temperature, the ELP is dehydrated and the structure collapses, enabling a substantial flow of ions. Potential applications of temperature-responsive protein pores in medical biotechnology are discussed.     

The well-tempered protein

Globular proteins are the most functionally versatile class of molecules in the biosphere. They play leading roles in practically every aspect of cell physiology, including gene expression, developmental and metabolic regulation, transport, and catalysis. Essential to a protein's function is its characteristic and geometrically well-defined three-dimensional shape, or native state. Proteins, however, are synthesized by the cell as biologically inactive linear chains; it is only upon folding to their native states that globular proteins come to life. The general principles underlying the behavior of proteins as amphiphilic heteropolymer molecules, as well as those unique properties of proteins as biopolymers that are evolutionarily selected for folding and function are important for understanding globular proteins. Recently, there have been many successes in recent studies of lattice and off-lattice coarse-grained models of proteins, as well as in the main current challenges facing the field. © 1999 Elsevier Science Ltd.     

Cotton-plant protein isolates

The seeds of the cotton plant are an important source of food protein. The presence of gossypol in the seeds limits the use of protein isolates as additives to food products. The use of salt and alkaline solutions for extracting the proteins leads to the formation of isolates with relatively high gossypol constants. Extraction in an acid medium permits the presence of the toxin to be excluded. In this process, a number of problems arise which are connected with the presence of a large amount of phytin (about 5%) in the seeds of the cotton plant. The latter affects both the yield of food protein on extraction and functional properties. A method is proposed for eliminating traces of phytin from an acid isolate. A substantial influence of phytin on the properties of the proteins has been observed in those cases where the latter is strongly bound to the proteins at acid pH values.Institute of the Chemistry of Plant Substances, Academy of Sciences of the Uzbek SSR, Tashkent. Translated from Khimiya Prirodnykh Soedinenii, No. 3, pp. 366–369, May–June, 1983.     

Polarizable protein packing

To incorporate protein polarization effects within a protein combinatorial optimization framework, we decompose the polarizable force field AMOEBA into low order terms. Including terms up to the third-order provides a fair approximation to the full energy while maintaining tractability. We represent the polarizable packing problem for protein G as a hypergraph and solve for optimal rotamers with the FASTER combinatorial optimization algorithm. These approximate energy models can be improved to high accuracy [root mean square deviation (rmsd) < 1 kJ mol(-1)] via ridge regression. The resulting trained approximations are used to efficiently identify new, low-energy solutions. The approach is general and should allow combinatorial optimization of other many-body problems.     

Globular protein gelation

In this paper, significant advances in the structure and rheological properties of globular protein gels have been described. The achievements of scattering methods, which have provided information over a range of distances from the monomer to the micron scale, are reviewed. The kinetics of gelation measured by rheological methods have also been discussed, including currently applied models for the gelation time, the gel elastic modulus and the critical gel concentration.     

Surface-tethered protein switches

Protein switches are engineered fusion proteins with an input domain that recognizes and responds to an input signal and an output domain whose function is regulated by the state of the input domain. Here we demonstrate a fully functional surface tethered protein switch that offers a potential route to a universal biosensing platform.     

Quantitative interrogation of protein co-aggregation using multi-color fluorogenic protein aggregation sensors

Co-aggregation of multiple pathogenic proteins is common in neurodegenerative diseases but deconvolution of such biochemical process is challenging. Herein, we developed a dual-color fluorogenic thermal shift assay to simultaneously report on the aggregation of two different proteins and quantitatively study their thermodynamic stability during co-aggregation. Expansion of spectral coverage was first achieved by developing multi-color fluorogenic protein aggregation sensors. Orthogonal detection was enabled by conjugating sensors of minimal fluorescence crosstalk to two different proteins via sortase-tag technology. Using this assay, we quantified shifts in melting temperatures in a heterozygous model protein system, revealing that the thermodynamic stability of wild-type proteins was significantly compromised by the mutant ones but not vice versa. We also examined how small molecule ligands selectively and differentially interfere with such interplay. Finally, we demonstrated these sensors are suited to visualize how different proteins exert influence on each other upon their co-aggregation in live cells.

A little leak will sink a great ship! We prepared a series of multi-color protein aggregation sensors and developed a dual-color thermal shift assay to simultaneously and quantitatively report on protein co-aggregation of two different proteins.     

Real-time measurements of protein dynamics using fluorescence activation-coupled protein labeling method

We present a fluorescence activation-coupled protein labeling (FAPL) method, which employs small-molecular probes that exhibit almost no basal fluorescence but acquire strong fluorescence upon covalent binding to tag-proteins. This method enables real-time imaging of protein labeling without any washout process and is uniquely suitable for real-time imaging of protein dynamics on the cell surface. We applied this method to address the spatiotemporal dynamics of the EGF receptor during cell migration.